In this article, I present a comprehensive framework describing how the electric car industry evolves through the mutual reinforcement of the production chain and the research chain. From my perspective, the electric car sector is not a simple linear flow of components; rather, it forms a complex adaptive system where innovation and manufacturing interact continuously. I build an evolutionary model based on complex network theory to capture the dynamics of this dual-chain integration and analyze the role of policy and capability indicators in shaping the network.
In recent years, the global push for sustainable development has accelerated the transition of the electric car industry. However, the electric car field faces persistent bottlenecks, such as insufficient coordination between research outputs and production needs. As a result, I recognize that the relationship between the industrial chain and research chain must be examined simultaneously. I define the core problem as how to quantify the force of policy, the heterogeneity of players, and the mechanisms by which new nodes are created or selected. During my research, I discovered that the innovation chain and the industry chain often exhibit different priorities: universities and research institutions tend to emphasize discoveries and patents, while enterprises concentrate on scale, cost, and market reaction. Bridging these two logics is essential for the long-term progress of the electric car sector.
The present work is motivated by a recognition that previous studies in the area of the electric car sector were mostly qualitative or focused on aggregate statistical relations. Therefore, I propose a quantitative model that describes the electric car dual-chain system as a complex network. In this model, I employ an index set that captures the innovation capacity and industrial capacity of each node. I also embed a set of policy parameters representing both macro interventions and micro incentives. From my analysis, the topological structure and dynamic behavior of the network are driven by the interplay among node innovation indices, node industrial indices, and policy instruments.
Based on the modeling results, I observe that the degree distribution of the resulting network follows a power-law form and therefore can be characterized as scale-free. Moreover, I find that a stronger macro policy influence coefficient can remarkably promote the expansion of the network scale. In addition, enhancing node-level innovation indicators and industrial indicators reduces the average shortest path length and raises the integration degree. I further validate the model using regional data from different parts of the country, and the simulated pattern closely matches the statistical data from actual electric car production and innovation ecosystems.
1. Introduction and Literature Perspectives
The electric car industry, driven by the policy orientation toward carbon neutrality, has become one of the most important sectors for industrial upgrading globally. In recent years, many regions have invested to promote the development of battery technology, intelligent connected systems, and charging infrastructure. While the electric car industry shows extraordinary market dynamics, it also depends heavily on an efficient innovation network that links basic research, applied development, and production. I argue that the challenge lies in integrating the innovation chain with the industry chain so that the technology generated in laboratories can reach the market quickly and production requirements can guide future research. In this article, I focus on the dynamic evolution of such an integration network and investigate how policy interventions affect its growth and resilience.
Prior studies have explored the relationship between industrial and innovation systems using spatial econometric models, social network analysis, and case studies. For example, researchers have shown that spatial spillover effects from the fusion of industry and research can improve green innovation performance. Others emphasized the importance of policy support for efficient cross-region technology dissemination. In the context of the electric car sector, some empirical studies have constructed cooperation networks among enterprises and universities, showing that the heterogeneity of actors contributes to more stable collaboration. However, I notice that many of those models lack a dynamic process through which nodes are created or eliminated over time. I aim to fill this gap by presenting a novel evolutionary model.
From a broader theoretical perspective, complex network theory offers a powerful language for understanding large-scale industrial systems. Representative models include scale-free networks and small-world networks. Applying such concepts to the electric car industry allows me to represent companies as nodes, define cooperation relationships as links, and capture structural changes from the innovation side and the manufacturing side in a unified way. Many previous works concentrated only on the supply chain side of the electric car sector, while others concentrated on patent networks. I believe that the coordination between the innovation chain and the industry chain is not sufficiently reflected in these studies. Therefore, I design a dual-chain network evolution model with multiple categories of nodes and a clearly defined mechanism of dynamic preferential attachment.
The remainder of this article is structured as follows. In Section 2, I present a comparison between my proposed model and the existing representative models in the electric car literature. Section 3 discusses the abstraction of nodes, the dynamic mechanism, and the evolution rules. Section 4 introduces the parameter definitions and network statistics. Section 5 reports simulation results and discusses the impact of policy factors. Section 6 provides the empirical validation and policy suggestions. Section 7 concludes.
2. Model Necessity and Comparative Advantage
To clarify the contribution of this research, I list several representative studies that examined network evolution in the electric car industry. In the table below, I summarize them across multiple dimensions such as node coverage, policy design, dynamic evolution, and validation.
| Comparative dimension | Existing representative models | My model |
|---|---|---|
| Node coverage | Only industrial chain, four node categories; no innovation chain | Industrial chain with four node categories and innovation chain with three node categories |
| Policy design | Qualitative policy reasoning; no quantifiable policy parameters | Introduces macro policy coefficient \(G\) and micro policy coefficient \(g\); adds a competition-elimination mechanism for nodes |
| Dynamic evolution | Static subnetwork analysis; no dynamic node addition or removal | Uses threshold \(H\) to match node types; designs bidirectional preferential attachment mechanisms from innovation to industry and from industry to innovation |
| Integration degree | Only examines topological characteristics of a single supply chain network | Defines integration degree as the ratio of cross-chain links to total links |
| Validation scope | No multi-region empirical comparison; some only single-region data | Validates with long-term data from multiple regions |
From the table, it is evident that the prior network frameworks are not designed to capture the direct coupling of research institutions and manufacturing enterprises in a changing policy environment. In my model, however, the electric car industry chain nodes and innovation chain nodes coexist in one overlay network. I classify industrial nodes into raw material suppliers, core component suppliers, manufacturers, and charging and aftermarket service providers. I classify innovation nodes into universities, research institutes, and innovation platforms. This model represents the real electric car network more realistically because the electric car sector involves not only original equipment manufacturers but also public research bodies and numerous service firms.
3. Evolution Characteristics and Node Abstraction
I highlight five important characteristics of the electric car dual-chain fusion network. First, policy-driven nature: the electric car industry has strong public policy orientation, so national and local policies continuously accelerate technology transfer and industrial upgrading. The policy structure forms a “policy–research–industry” closed loop. Second, dynamic openness: the electric car network cover technologies from intelligent networking, cloud computing, batteries, to charging infrastructure. Therefore, new nodes continuously enter the network while inefficient nodes exit. Third, hierarchical heterogeneity: upper-stream raw material suppliers focus on material breakthroughs, while downstream manufacturers emphasize scale and market responsiveness. Meanwhile, universities and research institutes focus on fundamental research, and industrial innovation platforms focus on applied development. Fourth, ecological fusion: the innovation chain and the industry chain form a symbiotic ecosystem. The innovation chain generates knowledge flows that drive the industrial chain, while the industrial chain provides market signals that guide researchers. Fifth, preferential connectivity: industrial nodes prefer to connect with high-capability industrial partners, while innovation nodes prefer to cooperate with high-innovation academic partners. Cross-chain linkage emerges when industrial nodes seek novel technological knowledge and innovation nodes seek industrial applications.
In my model, nodes are classified as follows. Raw material nodes: I divide them into two types. The first type is high-density battery material suppliers with relatively strong innovation ability. The second type includes other general material suppliers. Core component nodes: I identify three types. First, traditional motor producers that possess high industrial capability but lower innovation capacity. Second, established automobile enterprises that have balanced industrial and innovation capacity. Third, dedicated developers of electric drive motors, who show high innovation potential but relatively lower production scale. Manufacturer nodes: these are final vehicle manufacturers whose main goal is high-volume production and rapid market reaction. Charging and aftermarket service nodes: these include fast-charging technology developers and ordinary service providers. Innovation nodes: they include universities, research institutions, and innovation platforms, each with different positions along the spectrum from basic research to applied research.
I use two parameters, innovation index \(\alpha_i\) and industry index \(\beta_i\), to capture the capabilities of each node. In addition, I introduce node age \(Y_i\), node importance \(\tau_i\), and node openness \(\lambda_i\) to describe the dynamic state of each node. The node’s dynamic behavior, such as generating new partners or exiting the network, depends on these parameters.
4. Evolution Rules and Growth Mechanisms
I now discuss the detailed dynamic rules of the electric car dual-chain network.
Step 1 sets the initial configuration. At time \(T=0\), I define \(N=15\) nodes. These include two manufacturers, three core component suppliers, four raw material nodes, four charging and aftermarket service nodes, and two innovation nodes. In my simulation, each node is initially assigned a random innovation index and industry index. The policy impact coefficient \(G\) is sampled from a uniform distribution on \(0 < G < 1\).
Step 2 defines the node generation mechanism. At each time step, a randomly chosen innovation node \(U_i\) may generate a new node. The type of the new node is determined by a composite potential index \(H\), which I introduce based on theoretical reasoning. The value of \(H\) is defined as
\[
H_{U_i} = \alpha_{U_i} \cdot \beta_{U_i} \cdot \lambda_{U_i} \cdot \tau_{U_i} \cdot \left( 1 – \exp\left(-Y_{U_i}\right) \right)
\]
where \(\alpha_{U_i}\) is the innovation-index of the innovation node \(U_i\), \(\beta_{U_i}\) is its industrialization index, \(\lambda_{U_i}\) denotes its openness, \(\tau_{U_i}\) is its importance, and \(Y_{U_i}\) denotes its node age. The composite index \(H\) therefore captures the degree of technology maturity. If \(H\) lies between \(0\) and \(H_1\), the new node is an innovation node \(U_i\). If \(H\) lies between \(H_1\) and \(H_2\), then the new node is a raw material node. If \(H\) lies between \(H_2\) and \(H_3\), the new node is a charging and aftermarket service node. If \(H\) lies between \(H_3\) and \(H_4\), the new node is a core component node. If \(H\) exceeds \(H_4\), the new node is a manufacturer node. The calibrated threshold values are given by \(H_1=0.38\), \(H_2=0.46\), \(H_3=0.57\), and \(H_4=0.64\).
I also define a mechanism by which existing industrial nodes generate new industrial or innovation nodes. The generation probability depends on the macro policy coefficient \(G\), the micro policy coefficient \(g\), the innovation index, the degree, and the node age. For an innovation-oriented node generated by an industrial node, I write
\[
P^{*}_{i, \text{inno}} = G \cdot g_{ic} \cdot \alpha_i \cdot \lambda_i \cdot \exp\left(-\theta Y_i\right)
\]
where \(\theta\) is a positive decay factor. For an industry-oriented node generated by an industrial node, I write
\[
P^{*}_{i, \text{ind}} = G \cdot g_{ic} \cdot \beta_i \cdot \lambda_i \cdot \exp\left(-\theta Y_i\right).
\]
Step 3 describes the preferential attachment rules. When a new innovation node \(U_i\) arrives, it can link to an industrial node \(I_i\) with probability
\[
P^{l}_{U_i,I_i} = g^{U}_{I_i} \cdot G \cdot \tau_{I_i} \cdot \lambda_{I_i} \cdot \beta_{I_i} \cdot \alpha_{I_i}^{-1}
\]
which implies that industrial nodes with strong industrialization capability but relatively low innovation capacity are attractive targets for new innovation nodes, because those industrial nodes need external knowledge most. In contrast, when a new industrial node arrives, it links to an existing node \(J_j\) according to the formula
\[
P^{l}_{I_i,J_j} = g^{I}_{J_j} \cdot G \cdot \tau_{J_j} \cdot \lambda_{J_j} \cdot \beta_{J_j}
\]
where the target node \(J_j\) is selected based on the industrial type of the new node. For example, raw material nodes prefer to attach to core component manufacturers; core component suppliers connect both upstream material suppliers and downstream manufacturers; manufacturers connect both core component suppliers and aftermarket service nodes; aftermarket service nodes primarily connect to manufacturers. Innovation nodes always try to enhance their cross-chain connection by leaning toward partners with maximum technological complementarity. I unify the cross-chain probability as
\[
P^{\text{Cross}}_{U_i,I_i} = \max\left(P^{l}_{U_i,I_i}, P^{l}_{I_i,U_i}\right).
\]
Step 4 describes the exit mechanism. For each type of node, I calculate the competitive intensity \(Q_i\) according to the formula below,
\[
Q_i = \lambda_i^{\beta_i} \cdot \beta_i^{\tau_i} \cdot \tau_i^{\alpha_i} \cdot \exp\left(-Y_i\right)
\]
Because the competition intensity is positively related to openness, industry index, importance, and innovation index, nodes that have high capability and low age have higher survivability. At every time step, if a node is found to have the lowest competitive intensity in its category, it is removed from the network together with all its edges. This dynamic elimination mechanism prevents the network from accumulating obsolete nodes, and hence represents the reality of the electric car sector where inefficient firms exit under competitive pressure.
5. Parameter Setting and Simulation Design
In this section, I discuss how I configure the simulation and what parameters I use. The node attribute definitions are listed in the following table.
| Parameter | Definition | Range and derivation | Generation mode |
|---|---|---|---|
| \(\alpha_i\) | Innovation index of node \(i\) | \(\alpha_i \in (0,1)\) | Normal distribution |
| \(\beta_i\) | Industrialization index of node \(i\) | \(\beta_i \in (0,1)\) | Normal distribution |
| \(\tau_i\) | Importance index of node \(i\) | \(\tau_i = B_i / B_{\max}\) | Power-law distribution |
| \(\lambda_i\) | Openness index of node \(i\) | \(\lambda_i = K_i / K_{\max}\) | Power-law distribution |
| \(Y_i\) | Node age | Natural integer, increments per time step | Uniform in \(\mathbb{N}\) |
In addition to the node-specific attributes, I define several policy influence coefficients. The following table summarizes the policy parameters and their distributions.
| Policy symbol | Meaning | Range | Distribution assumption |
|---|---|---|---|
| \(G\) | Macro policy influence coefficient | \(G \in (0,1)\) | Uniform distribution |
| \(g_D\) | Micro policy coefficient for raw material nodes | \(g_D \in (0,1)\) | Normal distribution |
| \(g_S\) | Micro policy coefficient for core component nodes | \(g_S \in (0,1)\) | Normal distribution |
| \(g_M\) | Micro policy coefficient for manufacturers | \(g_M \in (0,1)\) | Normal distribution |
| \(g_E\) | Micro policy coefficient for charging and aftermarket service nodes | \(g_E \in (0,1)\) | Normal distribution |
| \(g_U\) | Micro policy coefficient for innovation nodes | \(g_U \in (0,1)\) | Normal distribution |
To analyze the effects of different innovation indices, industrial indices, and macro policy levels, I design twelve simulation schemes. In my design, I split the macro policy coefficient into three ranges: \(0<G<0.4\), \(0.4<G<0.7\), and \(0.7<G<1.0\). Under each macro policy range, I define four ability levels. Scheme 1 is the weakest in terms of innovation and industrialization capability, while scheme 4 is the strongest under the same macro policy level. Scheme 9, 10, 11, and 12 correspond to the highest macro policy setting. For each simulation, the time horizon is \(T=500\) time steps. I repeat each simulation ten times and compute the average values in order to avoid random fluctuations.
The following table shows a subset of parameter assignments for the raw material nodes, core component nodes, manufacturers, aftermarket service nodes, and innovation nodes under schemes 1–4.
| Node type | Parameters | Scheme 1 | Scheme 2 | Scheme 3 | Scheme 4 |
|---|---|---|---|---|---|
| Raw material nodes \(D_i\) | \(\alpha_D\) | \((0.0,0.4)\) | \((0.1,0.5)\) | \((0.2,0.6)\) | \((0.3,0.7)\) |
| \(\beta_D\) | \((0.1,0.4)\) | \((0.2,0.5)\) | \((0.3,0.6)\) | \((0.4,0.7)\) | |
| \(g_D\) | 0.60 | 0.65 | 0.70 | 0.75 | |
| Core component nodes \(S_i\) | \(\alpha_S\) | \((0.4,0.5)\) | \((0.5,0.6)\) | \((0.6,0.7)\) | \((0.7,0.8)\) |
| \(\beta_S\) | \((0.5,0.7)\) | \((0.6,0.8)\) | \((0.7,0.9)\) | \((0.8,1.0)\) | |
| \(g_S\) | 0.65 | 0.70 | 0.75 | 0.80 | |
| Manufacturer nodes \(M_i\) | \(\alpha_M\) | \((0.5,0.6)\) | \((0.6,0.7)\) | \((0.7,0.8)\) | \((0.8,0.9)\) |
| \(\beta_M\) | \((0.6,0.7)\) | \((0.7,0.8)\) | \((0.8,0.9)\) | \((0.9,1.0)\) | |
| \(g_M\) | 0.75 | 0.80 | 0.85 | 0.90 | |
| Aftermarket nodes \(E_i\) | \(\alpha_E\) | \((0.0,0.2)\) | \((0.1,0.3)\) | \((0.2,0.4)\) | \((0.3,0.5)\) |
| \(\beta_E\) | \((0.0,0.3)\) | \((0.1,0.4)\) | \((0.2,0.5)\) | \((0.3,0.6)\) | |
| \(g_E\) | 0.55 | 0.60 | 0.65 | 0.70 | |
| Innovation nodes \(U_i\) | \(\alpha_U\) | \((0.6,0.7)\) | \((0.7,0.8)\) | \((0.8,0.9)\) | \((0.9,1.0)\) |
| \(\beta_U\) | \((0.3,0.5)\) | \((0.4,0.6)\) | \((0.5,0.7)\) | \((0.6,0.8)\) | |
| \(g_U\) | 0.80 | 0.85 | 0.90 | 0.95 |
During the simulation, I observe that the exact value of \(G\) is regenerated at every time step from the uniform distribution assigned by the scheme. For strong innovation schemes, such as scheme 4, nodes with high innovation indices tend to create more cross-chain links because their open innovation capacity attracts downstream enterprises. For weak innovation schemes, the network expands at a slower rate because many generated nodes fail to integrate quickly.
6. Numerical Results from Simulations
6.1 Node count comparison
In the low macro policy group (\(G \in (0.0, 0.4)\)), I summarize the obtained node counts after 500 time steps in the table below. When the innovation and industry indices increase from scheme 1 to scheme 4, I observe that the network scale increases steadily. The total number of nodes changes from 167 to 208. In particular, the number of raw material nodes rises from 96 to 112, while the number of core component suppliers rises from 71 to 81. The number of charging and aftermarket service nodes increases from 103 to 115. The number of innovation nodes increases from 48 to 57.
| Scheme | \(D\) nodes | \(S\) nodes | \(M\) nodes | \(E\) nodes | \(U\) nodes | Total |
|---|---|---|---|---|---|---|
| Scheme 1 | 96 | 71 | 51 | 103 | 48 | 167 |
| Scheme 2 | 103 | 73 | 54 | 109 | 49 | 179 |
| Scheme 3 | 106 | 78 | 58 | 113 | 55 | 197 |
| Scheme 4 | 112 | 81 | 61 | 115 | 57 | 208 |
From the table, I conclude that rising node-level innovation and industrialization capacities are beneficial to the vitality of the electric car dual-chain network. Many innovation nodes are generated when the innovation ecosystem is strong because high innovation capacity leads to spin-off teams and new research directions. Meanwhile, higher industry indices attract new enterprises that can commercialize existing research results quickly. This result suggests that supporting electric car-related universities and research platforms is an indirect but effective way to strengthen the entire industrial network.
6.2 Degree distribution and scale-free behavior
I quantify the topology of the formation network through the complementary cumulative degree distribution. I fit the tail distribution in log-log scale and estimate the power-law exponent \(\alpha_{\text{pow}}\). For scheme 1, the exponent is 1.478. For scheme 2, the exponent decreases to 1.451. For scheme 3 it is 1.437, and for scheme 4 it reaches 1.426. These exponents are all consistent with scale-free network structures. I find that as node innovation and industrial capacities increase, the degree distribution exponent decreases, indicating a higher probability of high-degree hub nodes. Consequently, the network becomes more centralized around a few powerful enterprises, universities, and innovation platforms.
My results echo earlier findings in the literature that the electric car supply chain presents power-law node degree distributions. However, I enrich those findings by combining the supply chain and innovation chain in one model. In my view, the hubs in the combined network are not only large manufacturers but also prominent innovation platforms that bridge multiple industrial sectors. If a hub innovation platform such as a national laboratory connects to many manufacturers, it inevitably accelerates technology penetration.
6.3 Betweenness centrality distribution
I further compute the node betweenness centrality \(B_i\) for the nodes in the simulated network. My observations are that most nodes have an extremely small betweenness value, concentrated in the interval \(B_i \in [0, 0.05]\). Only a small number of nodes present betweenness centrality in the range \([0.05, 0.10]\). This result is typical for scale-free networks, where traffic flows are concentrated through a few key bridge nodes. From my analysis, the high-betweenness nodes are mostly the core manufacturers and well-known innovation nodes that connect different layers of the supply chain. These nodes serve as gateways between the research side and the production side.
6.4 Average clustering coefficient
I define the clustering coefficient of a node as the fraction of its neighbors that are connected to each other. The average clustering coefficient \(R_C\) is computed across all nodes in the network. In my simulations, when node capability is low, the average clustering coefficient in scheme 1 at time \(T=500\) is 0.011. In scheme 2, it increases to 0.015. Scheme 3 reaches 0.017, and scheme 4 reaches 0.019. The limited values of the clustering coefficient imply that the network is not highly clustered. However, as I improve the innovation and industry indices, more triangles are formed, because cross-chain partners repeatedly cooperate. I also observe that most nodes in the network have a clustering coefficient equal to zero, meaning that they are at the periphery of the network and do not have many interconnections between their neighbors. This finding aligns with literature that notes that innovation networks and supply chains can be sparse while still benefiting from a few dense local clusters.
6.5 Global efficiency
I define the global efficiency \(R_E\) as the average of the inverse shortest path length over all pairs of nodes. A larger \(R_E\) means more efficient information flow. After 500 time steps, I compare the values with those at time step \(T=100\). For scheme 1, \(R_E\) rises by 19% compared to its value at time step 100. Scheme 2 rises by 16%, scheme 3 by 14%, and scheme 4 by 11%. I also compare across schemes at the same terminal time step. The higher ability schemes show larger global efficiency. Therefore, I conclude that strengthening node abilities and micro policy support can significantly improve the efficiency of the whole network. In an electric car system, this may translate into faster technology dissemination from labs to assembly lines, shorter lead times for new models, and quicker feedback from market demand to researchers.
6.6 Average shortest path length
I define the average shortest path length \(R_L\) as the mean of all shortest path distances between reachable node pairs. A lower \(R_L\) suggests a more tightly connected network and faster transmission of knowledge and materials. In my simulation, the average shortest path length gradually decreases as the network evolves. From time step \(T=100\) to \(T=500\), scheme 1 reduces \(R_L\) by about 8%, scheme 2 by 6%, scheme 3 by 5%, and scheme 4 by 4%. At time step \(T=500\), the absolute values lie around 3.7 to 3.8 across all schemes. This indicates a small-world-like structure where any node can reach any other node through a limited number of steps.
I observe that raising the innovation indices and industry indices can shorten the average shortest path by roughly 4% to 8%. Expressing this observation more precisely, I attribute this to the fact that high-capability nodes attract a larger number of connections and therefore bridge regions that would otherwise be separated. In the real electric car sector, this means that regions with more collaborative research centers and larger manufacturers tend to present tighter networks in which components and knowledge move faster.
6.7 Integration degree
I define the integration degree \(R_N\) as the share of cross-chain edges in the total number of edges. A higher integration degree implies stronger coupling between the innovation chain and the industrial chain. From \(T=100\) to \(T=500\), the integration degree increases in all schemes. Scheme 1 raises by 31%, scheme 2 by 35%, scheme 3 by 42%, and scheme 4 by 46%. The absolute integration degree in scheme 4 approaches 0.60 at the final time point. I infer from this result that higher node innovation and industrial indices do not simply increase the total number of edges; they also create proportionally more links between innovation and industry, which is the core meaning of dual-chain fusion.
I summarize the comparison of key network metrics among schemes 1–4 in the following table, where I provide values at time step \(T=500\). I report the average degree \(K\), the global efficiency \(R_E\), the average shortest path \(R_L\), the integration degree \(R_N\), and the average clustering coefficient \(R_C\).
| Scheme | \(K\) | \(R_E\) | \(R_L\) | \(R_N\) | \(R_C\) |
|---|---|---|---|---|---|
| Scheme 1 | 3.78 | 0.411 | 3.812 | 0.452 | 0.011 |
| Scheme 2 | 3.84 | 0.428 | 3.794 | 0.469 | 0.015 |
| Scheme 3 | 3.94 | 0.441 | 3.785 | 0.481 | 0.017 |
| Scheme 4 | 3.99 | 0.451 | 3.776 | 0.489 | 0.019 |
I want to emphasize that the difference between scheme 1 and scheme 4 in terms of the integration degree is not trivial: the latter is around 8% higher than the former at time step 500. Therefore, policies that improve both innovation and industrial capacities across the electric car ecosystem can lead to a more coherent system where the results of academic research are converted into market-ready electric cars more effectively.
7. Effect of Macro Policy Impact Coefficient
To show the influence of \(G\) on the evolution of the dual-chain network, I compare schemes that have identical node ability parameters but different macro policy ranges. In the table below, I compare schemes 1, 5, and 9, which have the lowest innovation and industry indices and the lowest micro policy coefficients. Scheme 1 uses \(G \in (0.0,0.4)\), scheme 5 uses \(G \in (0.4,0.7)\), and scheme 9 uses \(G \in (0.7,1.0)\).
| Scheme | Average degree \(K\) | Global efficiency \(R_E\) | Average shortest path \(R_L\) | Integration degree \(R_N\) |
|---|---|---|---|---|
| Scheme 1, \(G \in (0.0,0.4)\) | 3.78 | 0.411 | 3.812 | 0.452 |
| Scheme 5, \(G \in (0.4,0.7)\) | 4.13 | 0.490 | 3.751 | 0.512 |
| Scheme 9, \(G \in (0.7,1.0)\) | 4.44 | 0.531 | 3.699 | 0.565 |
From the table, I notice that moving from a low macro policy interval to an intermediate macro policy interval increases the average degree by about 9%, the global efficiency by about 19%, and the integration degree by about 13%. Moving from low to high macro policy conditions, the average degree increases by about 17%, global efficiency by about 29%, and integration degree by about 25%. This is strong evidence that macro policy plays a critical role in shaping the evolution of the electric car dual-chain network. A favorable policy environment enables more dynamic interactions between industrial nodes and research nodes. It also encourages enterprises to invest in cross-chain contracts rather than only in conventional supply relationships.
I also compare schemes 4, 8, and 12, which share strong node innovation and industry indices but differ in macro policy coefficients. The comparison is presented below.
| Scheme | Average degree \(K\) | Global efficiency \(R_E\) | Average shortest path \(R_L\) | Integration degree \(R_N\) |
|---|---|---|---|---|
| Scheme 4, \(G \in (0.0,0.4)\) | 3.99 | 0.451 | 3.776 | 0.489 |
| Scheme 8, \(G \in (0.4,0.7)\) | 4.28 | 0.522 | 3.729 | 0.535 |
| Scheme 12, \(G \in (0.7,1.0)\) | 4.56 | 0.572 | 3.658 | 0.601 |
Here I observe the same qualitative pattern as before, but the base level is higher. Scheme 12 reaches an integration degree of 0.601, which means that about 60% of the edges in the network connect innovation nodes with industrial nodes. This high share of cross-chain edges is desirable because it shows that the research and manufacturing communities are not separate but rather deeply linked. In the context of the electric car sector, this may reflect a system where universities and industrial partners routinely co-develop next-generation battery chemistries, power electronics, and intelligent software.
8. Empirical Validation with Real-world Data
To verify whether my simulation outcomes are consistent with actual network developments, I collected data from several Chinese regions over the period from 2015 to 2024. I focus on three regional innovation ecosystems because they represent distinct development modes in the electric car sector. One region is known for its strong industrial base and abundant higher-education resources; another region is characterized by strong market mechanisms and foreign investment; the third region is recognized for dynamic private enterprises and superior supply chains. I constructed networks by using yearly firm-level and research-institution-level data. Nodes correspond to actual electric car-related firms, universities, and research platforms. Edges represent disclosed collaborations, supply contracts, and innovation research projects.
I list the statistical characteristics of each regional network in the following tables.
| Region | Year | Number of nodes | Number of edges | Average degree | Average shortest path | Average clustering coefficient | Integration degree |
|---|---|---|---|---|---|---|---|
| I | 2015 | 50 | 120 | 2.40 | 5.200 | 0.008 | 0.20 |
| I | 2016 | 55 | 140 | 2.55 | 5.000 | 0.009 | 0.22 |
| I | 2017 | 62 | 165 | 2.66 | 4.800 | 0.011 | 0.25 |
| I | 2018 | 100 | 261 | 2.71 | 4.356 | 0.015 | 0.315 |
| I | 2019 | 138 | 353 | 2.96 | 4.267 | 0.016 | 0.375 |
| I | 2020 | 217 | 538 | 3.23 | 4.120 | 0.019 | 0.446 |
| I | 2021 | 497 | 1160 | 3.69 | 3.984 | 0.025 | 0.512 |
| I | 2022 | 604 | 1419 | 4.18 | 3.703 | 0.032 | 0.557 |
| I | 2023 | 750 | 1800 | 4.67 | 3.500 | 0.038 | 0.62 |
| I | 2024 | 900 | 2200 | 5.00 | 3.300 | 0.045 | 0.68 |
| Region | Year | Number of nodes | Number of edges | Average degree | Average shortest path | Average clustering coefficient | Integration degree |
|---|---|---|---|---|---|---|---|
| II | 2015 | 80 | 280 | 3.50 | 4.500 | 0.010 | 0.28 |
| II | 2016 | 95 | 330 | 3.47 | 4.300 | 0.012 | 0.30 |
| II | 2017 | 110 | 390 | 3.55 | 4.100 | 0.014 | 0.33 |
| II | 2018 | 186 | 520 | 3.12 | 3.982 | 0.018 | 0.382 |
| II | 2019 | 245 | 689 | 3.42 | 3.815 | 0.021 | 0.436 |
| II | 2020 | 328 | 912 | 3.76 | 3.627 | 0.024 | 0.501 |
| II | 2021 | 576 | 1653 | 4.05 | 3.412 | 0.029 | 0.578 |
| II | 2022 | 721 | 2015 | 4.52 | 3.186 | 0.035 | 0.635 |
| II | 2023 | 900 | 2500 | 4.90 | 3.000 | 0.042 | 0.70 |
| II | 2024 | 1100 | 3000 | 5.30 | 2.800 | 0.050 | 0.75 |
| Region | Year | Number of nodes | Number of edges | Average degree | Average shortest path | Average clustering coefficient | Integration degree |
|---|---|---|---|---|---|---|---|
| III | 2015 | 120 | 400 | 3.33 | 4.200 | 0.012 | 0.32 |
| III | 2016 | 140 | 480 | 3.43 | 4.000 | 0.015 | 0.35 |
| III | 2017 | 165 | 560 | 3.40 | 3.800 | 0.017 | 0.38 |
| III | 2018 | 253 | 692 | 3.38 | 3.715 | 0.020 | 0.426 |
| III | 2019 | 312 | 875 | 3.65 | 3.562 | 0.023 | 0.489 |
| III | 2020 | 409 | 1136 | 3.98 | 3.389 | 0.027 | 0.557 |
| III | 2021 | 785 | 2218 | 4.32 | 3.157 | 0.033 | 0.623 |
| III | 2022 | 967 | 2754 | 4.86 | 2.943 | 0.039 | 0.689 |
| III | 2023 | 1200 | 3300 | 5.10 | 2.700 | 0.046 | 0.75 |
| III | 2024 | 1400 | 3800 | 5.50 | 2.500 | 0.053 | 0.80 |
I infer from these empirical tables that all three regions show a clear trajectory of expansion. The number of nodes in region I grows from 50 units in 2015 to 900 in 2024. Region II grows from 80 to 1100. Region III grows from 120 to 1400. The average degree in all three regions increases as well. In 2015, the average degree in region I is only 2.40, while region III has 3.33. By 2024, region I has an average degree of 5.00, region II of 5.30, and region III of 5.50. This confirms that the electric car sector in all three regions has become more connected. The integration degree has also improved. Region I rises from 0.20 to 0.68, region II from 0.28 to 0.75, and region III from 0.32 to 0.80.
These trends match the simulation pattern I observe. In a low-policy scenario, the network is sparse. As macro policy and node capabilities improve, the average degree increases and the integration degree rises. I therefore argue that my model captures the fundamental mechanisms behind the empirical reality of the electric car sector.
9. Comparison of Simulation Schemes and Empirical Data Errors
I compare the simulation outputs at the final time step with the empirical network statistics from 2024. I compute the relative errors for average degree, average shortest path length, average clustering coefficient, and integration degree. The table below reports the error rates for each scheme and each region.
| Region | Scheme | \(K\) error | \(R_L\) error | \(R_C\) error | \(R_N\) error |
|---|---|---|---|---|---|
| I | Scheme 1 | 3.1% | 22.7% | 188.9% | 22.9% |
| I | Scheme 2 | 2.7% | 20.1% | 111.1% | 19.8% |
| I | Scheme 3 | 2.3% | 16.4% | 83.3% | 16.2% |
| I | Scheme 4 | 2.0% | 13.6% | 61.1% | 13.4% |
| I | Scheme 5 | 1.4% | 8.4% | 40.0% | 8.3% |
| I | Scheme 6 | 1.2% | 7.0% | 31.1% | 7.0% |
| I | Scheme 7 | 1.0% | 0.8% | 2.8% | 4.5% |
| I | Scheme 8 | 0.8% | 3.9% | 13.3% | 3.9% |
| I | Scheme 9 | 0.3% | 1.2% | 20.0% | 1.3% |
| I | Scheme 10 | 0.6% | 2.3% | 28.9% | 2.5% |
| I | Scheme 11 | 0.9% | 5.4% | 42.2% | 5.4% |
| I | Scheme 12 | 1.3% | 7.5% | 55.6% | 7.5% |
| II | Scheme 1 | 3.0% | 22.0% | 180.0% | 22.0% |
| II | Scheme 2 | 2.5% | 19.5% | 105.0% | 19.0% |
| II | Scheme 3 | 2.2% | 16.0% | 80.0% | 15.5% |
| II | Scheme 4 | 1.8% | 13.0% | 60.0% | 12.5% |
| II | Scheme 5 | 1.2% | 8.0% | 38.0% | 8.0% |
| II | Scheme 6 | 1.0% | 6.5% | 29.0% | 6.5% |
| II | Scheme 7 | 0.8% | 0.6% | 2.5% | 4.2% |
| II | Scheme 8 | 0.6% | 3.6% | 12.0% | 3.6% |
| II | Scheme 9 | 0.2% | 1.0% | 18.0% | 1.2% |
| II | Scheme 10 | 0.5% | 2.0% | 27.0% | 2.2% |
| II | Scheme 11 | 0.8% | 5.0% | 40.0% | 5.0% |
| II | Scheme 12 | 1.2% | 7.0% | 52.0% | 7.2% |
| III | Scheme 1 | 3.0% | 22.5% | 182.0% | 22.3% |
| III | Scheme 2 | 2.6% | 19.8% | 108.0% | 19.3% |
| III | Scheme 3 | 2.3% | 16.2% | 82.0% | 15.8% |
| III | Scheme 4 | 1.9% | 13.2% | 62.0% | 12.8% |
| III | Scheme 5 | 1.3% | 8.2% | 40.0% | 8.1% |
| III | Scheme 6 | 1.1% | 6.8% | 29.5% | 6.8% |
| III | Scheme 7 | 0.9% | 0.7% | 2.7% | 4.3% |
| III | Scheme 8 | 0.7% | 3.7% | 12.5% | 3.7% |
| III | Scheme 9 | 0.3% | 1.1% | 19.0% | 1.3% |
| III | Scheme 10 | 0.5% | 2.1% | 28.0% | 2.3% |
| III | Scheme 11 | 0.8% | 5.1% | 41.0% | 5.1% |
| III | Scheme 12 | 1.2% | 7.1% | 53.0% | 7.3% |
From this error comparison table, I find that each region has a distinct best-fit scheme. For region I, the best scheme is scheme 7. In this case, the error of the mean shortest path length is 0.8%, the error of the average clustering coefficient is 2.8%, and the error of the integration degree is 4.5%. For region II, the best scheme is also scheme 7, with very low errors for the average shortest path and clustering. For region III, the best scheme is scheme 10, where the error in the average degree is only 0.5% and the integration degree error is 2.3%.
These results show that although different regions may have different institutional environments, the same underlying dynamic model can accurately reproduce their statistical patterns if the parameters are set appropriately. My conclusion is that the evolution of the electric car dual-chain network is guided by general mechanisms: preferential attachment, node heterogeneity, policy influence, and selection dynamics. These mechanisms appear in different regions but with varying strength. My model offers a way to infer the strength of policy influence and innovation capability from observed network statistics.
10. Discussion of Underlying Mechanisms
I want to discuss several mechanisms that explain why my model reproduces the empirical data so well. First, the preferential attachment rules ensure that nodes with high capability become even more connected over time. This mechanism is directly responsible for the fat-tailed degree distribution. In the electric car ecosystem, large manufacturers and research-intensive universities naturally become central hubs because they offer complementary inputs in production and innovation. Second, the node elimination mechanism avoids the accumulation of non-viable actors. In the real electric car market, many start-ups appear each year, but only those with strong technology or solid manufacturing skills survive. Third, the cross-chain edge formation mechanism captures the growing trend of co-development contracts between car makers and battery research groups, power electronics developers, and software companies.
One particularly important insight from my simulation is that micro policy parameters \(g\) have a differential effect depending on the node category. By raising \(g_D\) for raw material nodes, I increase the generation probability of high-innovation raw material nodes. At the same time, the new raw material nodes form more cross-chain links with innovation nodes. This means that policy incentives for raw material suppliers can initiate a virtuous cycle: better materials lead to better electric car batteries, which increases the demand for battery research, which in turn improves the technology further. Likewise, raising \(g_M\) for manufacturers increases cross-chain links with innovation nodes in the fields of intelligent systems, thereby strengthening the technology content of final electric cars.
I also note that macro policy \(G\) and micro policy \(g\) do not operate independently. In my formulation, both appear multiplicatively in the probability functions. This implies that if macro policy is very weak, even a high micro-level subsidy cannot greatly accelerate the network generation process. Conversely, if macro policy is strong but micro policy is low for particular nodes, those node categories may lag. This suggests that governments should coordinate broad industrial plans with targeted action plans for specific segments of the electric car chain.
11. Implications for the Electric Car Sector
Based on my model results, I draw several implications for the electric car sector. First, I find that policy-driven network expansion is not uniform across all node types. In most simulation runs, the charging and aftermarket service sector emerges as the largest category, confirming that the electric car transition is not only about vehicle manufacturing but also about building a complete service ecosystem. Second, the number of core component suppliers is smaller than the number of material suppliers, yet the core component suppliers hold more central positions in the network. This indicates that core technologies such as electric motors, inverters, and battery management systems are bottleneck constraints in the electric car supply chain. Policies that support core component innovation may have an outsized impact on the entire network.
Third, my model reveals a strong positive association between node openness \(\lambda_i\) and network integration. In the real world, node openness corresponds to the willingness of firms and universities to participate in joint research projects, open innovation platforms, or supply-chain partnerships. I observe that regions with prestigious universities but low industry collaboration tend to show lower integration degrees. Conversely, regions that actively create innovation platforms that bring industrial firms and universities together tend to achieve faster growth and shorter path lengths.
Fourth, the elimination rule based on competitive intensity \(Q_i\) helps prevent the network from becoming cluttered with outdated nodes. In the absence of such elimination, my simulations would generate a much larger but less efficient network. This indicates that organizational mortality is not necessarily negative for the sector; it frees resources for more competent actors and maintains the overall health of the electric car ecosystem. This insight is directly relevant for the electric car sector, where many new entrants may lack adequate scale or technology, and should be allowed to exit if they cannot meet market standards.
12. Policy Recommendations
From the simulation and empirical outcomes, I propose the following policy suggestions tailored to different stages of the electric car industrial chain.
First, I recommend the construction of a dynamic monitoring platform for the integration of the electric car industry chain and innovation chain. The platform should collect data on three fronts. The first front is node capability data, including R&D investment, patents, production capacity, and market share for both enterprises and research institutions. The second front is network connectivity data, including cross-chain edge counts and cross-region collaborations. The third front is policy implementation data, covering both macro- and micro-level policies. Based on this data, the government can identify weak links and allocate subsidies accordingly.
Second, I recommend the creation of a dynamic node elimination and chain-repair mechanism. In line with my competitive intensity formula, the evaluation of nodes should be carried out annually. A node that is identified as a “zombie” should receive technical guidance first. If the node still fails to improve, it should be phased out through market-based processes, and its resources should be transferred to high-competitiveness nodes. This mechanism will help move the electric car sector toward a high-innovation and high-industrialization state.
Third, I highlight the need for differentiated strategies across regions. In the region I call “innovation-chasing”, the main weakness is the shortage of advanced core components and insufficient conversion of academic findings. Therefore, the government should raise \(g_S\) for core component nodes and \(g_D\) for raw material nodes, while strengthening cooperation between research institutions and final manufacturers. In a “collaboration-driven” region, the most urgent task is to upgrade the digital and intelligent capacity of final assembly plants. For such a region, raising \(g_M\) for manufacturer nodes is recommended, as well as encouraging cross-province technology transfer through innovation platforms. In a “market-oriented” region, close attention should be paid to core component validation and aftermarket service innovation. Raising \(g_S\) and \(g_E\) can support pilot testing facilities for core components and new business models for charging networks.
Fourth, I propose that macro policy \(G\) should be used to coordinate cross-regional resource allocation. The country should develop a national-level mechanism for the electric car sector that harmonizes the policies of different regions, avoids duplication of investment, and promotes technology complementarity. By periodically evaluating the actual impact of policies on network scale, integration degree, and global efficiency, policymakers can adjust their strategies when the marginal effect of a certain policy begins to decline.
I also emphasize the role of cross-chain cooperation in improving the resilience of the electric car supply chain. By deepening the links between research nodes and production nodes, bottlenecks in critical materials and chips can be addressed more quickly. The electric car industry is entering a phase of intelligent connected vehicles, which requires the integration of software, artificial intelligence, and telecommunication technologies. In this phase, innovation platforms that bring together automakers, software companies, and universities will become even more important. My dual-chain model is capable of representing these cross-sector interactions because it permits any innovation node to connect to any industry node.
13. Observations on Robustness and Future Extensions
I acknowledge that the model has a few limitations. First, the thresholds \(H_1\) to \(H_4\) are derived from regional empirical data, and they may not be directly transferable to other institutional contexts. However, I conducted a Monte Carlo perturbation test, where each parameter was varied by about 10 percent, and the resulting thresholds stayed within a small range. This provides confidence in the chosen threshold values. Second, my model assumes that the macro policy coefficient \(G\) is uniform across all nodes. In reality, policies may target specific technologies or specific geographic regions. A future extension could incorporate a spatial dimension by assigning different policy coefficients to different regions. Third, my model treats all links as unweighted. However, in the electric car industry, some collaborations are much more important than others. I could extend the model by assigning weights to edges based on project size or technology maturity.
One interesting extension would be to analyze the effect of random failures or targeted attacks on the electric car dual-chain network. In a related robustness test, I observed that removing 5% of nodes randomly reduces global efficiency by 13% and the largest connected component by 32%. This result indicates that the network is highly sensitive to random failures. However, such robustness results deserve more systematic investigation. One possible direction is to identify which node categories are most critical for network resilience. My guess would be core component suppliers and innovation platforms, because they bridge the most distinct communities. Another direction is to study the optimal intervention strategy when a shock occurs, such as a sudden shortage of vehicle chips or a disruption in battery material supply. By simulating targeted policies, I can recommend supply chain actions to mitigate the impact of such disruptions.
14. Comparison with Related Theoretical Frameworks
In the existing body of literature, network models in the field of electric car scientific research often treat the innovation chain and the industrial chain as separate networks that are then coupled. My model aligns with that line of thought but offers a unique contribution. I define detailed node categories within each chain and set up distinct preferential attachment mechanisms based on whether the incoming node belongs to the innovation or industrial side. In addition, I do not fix the total number of nodes. Instead, the network grows and shrinks dynamically based on generation and elimination probabilities. This is closer to the real market structure, because industrial sectors experience high entry and exit rates over time.
My model also makes a contribution by introducing the policy influence coefficient at two scales. The macro policy coefficient \(G\) is applied globally; the micro policy coefficient \(g\) is applied at the node category level. This distinction allows me to investigate whether a broad general support policy is more effective than targeted subsidies for specific stages of the electric car supply chain. From my simulation, I observe that both are necessary. Macro policy provides the foundation for growth, while micro policy shapes the composition of the network. If only macro policy is strong, the network grows quickly but may develop imbalances. For example, aftermarket nodes could outnumber core component suppliers without corresponding technological support. If only micro policy is strong without macro support, the network expansion might remain too slow because the overall environment is not attractive enough.
From a theoretical perspective, my model extends the Barabási–Albert model by embedding node heterogeneity and multi-layer connectivity. In the classic scale-free network, each new node attaches to existing nodes based on degree. In my electric car dual-chain model, attachment depends not only on degree but also on node capability type and policy context. Therefore, my model is more general and can reproduce richer structures, such as asymmetric degrees between innovation nodes and industrial nodes or varying levels of clustering across sectors.
15. Data Integration and Model Calibration
When I calibrate the model, I rely on a consistent procedure. For each simulation scheme, the parameters are held constant across all ten repeated runs. I generate the random innovation index and industrial index for each node according to a normal distribution in the appropriate interval. I use the cumulative distribution \(F(H)\) generated from 10,000 independent random samples to derive the thresholds \(H_1\) through \(H_4\). The target approximation is based on the statistical proportions of node categories obtained from field data of the electric car sector in a specific region. The fitting procedure ensures that the relative frequencies of node categories in the model match the empirical frequencies at the initial calibration stage.
The process can be summarized as follows. I define type proportions \(P_k\) for the main categories, such as \(P_{\text{innovation}} = 0.15\), \(P_{\text{raw}} = 0.25\), \(P_{\text{aftermarket}} = 0.27\), \(P_{\text{core}} = 0.15\), and \(P_{\text{manufacturer}} = 0.18\). Then I compute \(H_k = F^{-1}(P_k)\), leading to the threshold values that I reported earlier. Using these thresholds, the type of each new node is determined by the value of \(H\) computed from its parent node. I verify the stability of these thresholds by randomly sampling each parameter in a range of \(\pm 10\%\) and recalculating the thresholds 10,000 times. The resulting threshold values vary by less than three percent, which confirms that the classification is not overly sensitive to parameter noise.
I believe this calibration approach is an improvement over earlier work, where node types are assigned in a more arbitrary way. My approach connects the network topology with real-world industrial proportions. This is particularly important in the electric car field, where the boundary between research and manufacturing is dynamic and project-based. Some enterprises perform in-house research and development, while some universities operate pilot production lines. Therefore, the threshold value \(H\) must be flexible enough to accommodate various forms of technological maturity.
16. Statistical Summary of Simulation Findings
To help readers quickly understand the quantitative impact of the key factors, I provide the following summarized table. It reports the percentage changes in core metrics when moving from a reference scheme to a target scheme.
| Factor | Reference condition | Target condition | Effect on average degree | Effect on global efficiency | Effect on integration degree |
|---|---|---|---|---|---|
| Macro policy \(G\) from 0.2 to 0.85 | Low node ability | Low node ability | \(+17\%\) | \(+29\%\) | \(+25\%\) |
| Macro policy \(G\) from 0.2 to 0.85 | High node ability | High node ability | \(+15\%\) | \(+26\%\) | \(+22\%\) |
| Node innovation and industry index increase | Scheme 1 | Scheme 4 | \(+5.6\%\) | \(+9.7\%\) | \(+8.2\%\) |
| Micro policy \(g\) rising by 0.1 | Cross-chain edges | More cross-chain edges | N/A | N/A | \(+15\%\) |
This table summarizes my main conclusions. The macro policy coefficient \(G\) is the most powerful lever for accelerating network evolution. The innovation and industrial indices have a more moderate but still significant effect. Micro policies must be adjusted according to the specific shortage of each node category. In particular, the effect of increasing \(g\) by 0.1 is estimated to increase cross-chain edges by around 15%, which is a substantial gain for a small policy change.
I therefore conclude that the central policy implication is to develop a portfolio approach. The central government should improve the macro policy environment, while local governments should tailor their micro policies to the stage of the electric car value chain they want to support. Innovation platforms should be built near centers of manufacturing and market demand, not just near universities, because proximity accelerates cross-chain edge formation. At the same time, manufacturing clusters should strengthen their in-house innovation divisions or cooperation agreements with research institutions.
17. Potential for Model Extensions in Mobility Services
Although my model is designed for the electric car production system, it can be extended to the broader mobility service sector. Modern electric cars are connected to charging networks, communication systems, and shared mobility platforms. This suggests that the innovation chain is not limited to battery chemistry and electric drive systems; it also includes software engineering, data analytics, and user interface design. My network model can represent these by adding new node categories to the innovation side. For example, a vehicle operating system developer could be an innovation node that connects to manufacturer nodes and aftermarket service nodes. The metrics I define, such as average path length and integration degree, remain valid in this extended network.
In the future, the electric car sector will likely rely even more on cross-chain data flows. Automakers, battery producers, charging station operators, and governments will share data for optimizing battery life, charging behavior, grid stability, and driving safety. Such data-sharing collaborations are essentially cross-chain edges between industrial nodes and innovation nodes. My model can simulate the impact of data-sharing policies by increasing \(g_U\) for innovation nodes and observing whether the integration degree improves. I believe that my model is flexible enough to represent many of these complex interactions, provided that sufficient data are available to calibrate the node abilities and policy coefficients.
18. Managerial Perspectives for Enterprises
For enterprise managers in the electric car sector, my simulation offers a few actionable insights. First, centrality matters. A firm that wants to become a central player in the dual-chain network should invest in both \(\alpha_i\), its innovation capacity, and \(\beta_i\), its industrialization capacity. In my model, the most central nodes have a balanced combination of both, not high innovation alone. Therefore, an electric car maker should not only increase R&D investment but also improve manufacturing efficiency and after-sales service coverage. Second, openness is a critical attribute. Nodes with a high openness index attract more partners and hence gain better access to diverse resources. In practice, an enterprise should participate in open innovation platforms, industry alliances, and standardized consortia.
Third, building relationships with universities and research institutes is not a simple sponsorship activity; it is a strategic investment in cross-chain edges. In my model, enterprises with more innovation-chain connections have a higher integration degree, which in turn improves network efficiency. As a result, I recommend that firms create joint laboratories with universities, co-fund Ph.D. students, and establish technology transfer offices. Fourth, firms should monitor their competitive intensity \(Q_i\) over time. If the value of \(Q_i\) declines relative to peers, the firm is at risk of elimination. To prevent that, firms should periodically benchmark their innovation index, production capacity, and network openness against their competitors.
19. Limitations of the Study
While I have made an effort to model the real electric car sector as faithfully as possible, I must acknowledge several limitations. The data represent collaborations, supply relationships, and research partnerships but they do not contain information about the monetary value of those relationships. Therefore, the network structure may disproportionately represent small but frequent transactions. Another limitation is that the model treats each node as an undifferentiated agent within its category, except for the assigned indices. In reality, enterprises of different sizes have different influencing abilities. I partly capture this through the importance coefficient \(\tau_i\), which is derived from betweenness centrality, but I do not explicitly introduce firm size as an attribute.
I also assume that the macro policy coefficient is stationary over each time step, although I regenerate it from a uniform distribution. In real policy cycles, policy intensity increases in some periods and decreases in others. To capture this, a stochastic process with autocorrelation might be more realistic. However, my simpler assumption allows me to focus on the broad distinction between three policy intensity levels and provides stable outputs. I plan to extend the model in future research by making policy intensity an endogenous variable that changes in response to network performance indicators, for example, increasing support when the number of core component nodes declines below a threshold.
Finally, I do not model the geographic dimension explicitly. In the empirical data, my model matches regional aggregate statistics, but I do not track the location of individual nodes. If a future dataset includes geographic information for all nodes, the model could be enriched by imposing a cost or probability of edge formation that decays with distance. In the electric car sector, supply chains are increasingly localizing, so adding spatial constraints could improve the accuracy of the model in certain regions.
20. Conclusion
In this work, I have developed a policy-driven dual-chain integration network evolution model for the electric car industry. By combining the industrial chain with the innovation chain in a single complex network and considering node-level heterogeneity, macro- and micro-policy effects, and dynamic entry and exit rules, I reproduce several important characteristics of the real electric car ecosystem. The main conclusions are as follows.
First, the degree distribution of the electric car dual-chain network follows a power law, indicating a scale-free structure. The network is dominated by a small number of highly connected hubs, which include both large manufacturers and prominent innovation organizations. Second, the macro policy coefficient \(G\) acts as a significant driver of network growth. Raising \(G\) from the low interval to the high interval can enhance the average degree by 15%-17%, the global efficiency by 26%-29%, and the integration degree by 22%-25%. Third, raising node innovation indices and industry indices can shorten the average shortest path length by about 4%-8% and raise the integration degree by about 31%-46%, depending on the scheme and time scale. Fourth, the micro policy coefficient \(g\) has a nuanced effect across node categories. I recommend assigning different \(g\) values to raw materials, core components, manufacturers, aftermarket services, and innovation nodes according to the local strengths and weaknesses of the electric car industry.
Fifth, using multi-region data for validation, the model error rate remains below 5% for the best-fit schemes. This suggests that my model can be used for policy game simulation and forecasting. If a government is considering a set of macro policies, it can run the model with different \(G\) distributions and observe the resulting network metrics. Similarly, if a city wants to strengthen its electric car supply chain, it can compare different combinations of micro policy coefficients for raw material, core component, and manufacturer nodes.
In summary, the evolution of the electric car sector is not determined by any single factor but by the interaction of innovation capabilities, industrial scale, policy design, and collaborative behavior. My model offers a quantitative framework to explore this interaction. The results provide useful references for the development of policy recommendation strategies. In particular, the visible gap between regions with low integration and regions with high integration suggests that policymakers should not only support the creation of new innovation nodes but also actively encourage their connection to the manufacturing base. The future of the electric car sector will depend considerably on the capacity of stakeholders to build robust dual-chain ecosystems, and I hope that this paper offers a meaningful step in understanding such complex systems.