The large-scale integration of clean energy into the power grid presents significant challenges to system stability due to its inherent intermittency and volatility. In this context, the aggregated regulation of electric vehicle (EV) fleets has emerged as a crucial mechanism for maintaining supply-demand balance. The aggregation process generates massive information exchange between the grid, aggregators, and numerous individual EVs. This exchange imposes stringent requirements for real-time and accurate communication, posing a severe challenge to the performance of existing network infrastructures. Consequently, efficient and reliable communication is a foundational prerequisite for effective vehicle-to-grid (V2G) interaction. This paper addresses this challenge by proposing a dedicated communication network resource allocation framework tailored for the hierarchical control of electric vehicle aggregation.
We propose a three-tier communication network architecture: “Central Cloud Platform – Edge Server – Electric Vehicle User.” This structure aligns with the hierarchical “Grid – Aggregator – EV” control model commonly employed for managing distributed electric vehicle resources. The core of our work involves a detailed analysis of the distinct business functionalities and heterogeneous information exchange requirements at each tier. Based on this analysis, we establish tier-specific utility function models that quantify the needs for both information transmission timeliness and accuracy during the demand response process.
The communication resource allocation problem is then formulated with the objective of maximizing the overall system utility derived from these functions. The resources considered are wireless channel allocation and transmit power allocation. To solve this complex mixed-integer nonlinear problem efficiently, we decompose it into sub-problems. Channel allocation is tackled using a hybrid approach combining a Dung Beetle Optimization (DBO) algorithm, enhanced with a sinusoidal strategy to improve population diversity and global search capability, and a bilateral matching theory. Power allocation for the assigned channels is solved using an improved water-filling algorithm based on Lagrangian multipliers for faster computation. Our simulation results demonstrate that the proposed algorithm effectively converges and significantly enhances the overall information transmission utility for the diverse services involved in the hierarchical aggregation control of electric vehicles, outperforming benchmark methods.
1. Communication Network Model for Electric Vehicle Aggregation Control
1.1 Network Architecture
The bi-directional interaction between electric vehicles and the power grid, known as aggregated demand response, is a potent tool for mitigating fluctuations caused by large-scale renewable energy integration. A hierarchical “Grid – Aggregator – Electric Vehicle User” control structure is widely adopted to manage regional EV clusters cooperatively while overcoming the inefficiency of direct cloud-based control. This structure necessitates reliable and timely data exchange. Recognizing the distinct functionalities and communication requirements at each level of the control hierarchy, we establish a corresponding “Central Cloud Platform – Edge Server – Electric Vehicle User” layered communication network architecture, as illustrated below.
This architecture supports four key data transmission services essential for EV aggregation control:
- Service 1: EV Load Data Upload. Distributed EV users upload their physical and electrical parameters (e.g., State of Charge, plug-in time) to their associated Edge Server. This service involves a large volume of data from many sources. While accuracy is important, the system has a degree of fault tolerance as an error from a single electric vehicle has minimal impact on the aggregated capacity calculation. The primary challenge is the overall low latency required due to varying transmission distances.
- Service 2: Regional EV Cluster Adjustable Capacity Upload. To ensure real-time control, the computation of the aggregate adjustable capacity is offloaded to the Edge Servers, avoiding the delay of transmitting massive raw data to the cloud. Each Edge Server calculates the total controllable capacity of its region and uploads this critical parameter to the Central Cloud Platform. This service involves small data packets but has extremely high accuracy requirements, as this parameter directly determines the control strategy.
- Service 3: Regional EV Cluster Absorption Task Dispatch. The Central Cloud Platform, based on the real-time grid absorption target and the received capacity parameters, generates a power allocation strategy for each cluster and dispatches it to the corresponding Edge Servers. This is also a small-packet service with very high accuracy demands.
- Service 4: EV User Charging Control Command Dispatch. Upon receiving its allocated power, the Edge Server generates individualized charging control commands (e.g., power setpoints) and dispatches them to each electric vehicle under its management. This is the final execution step. It involves dispatching a large number of commands and has stringent requirements for both timeliness (to track real-time signals) and accuracy (to ensure safe and correct operation).
1.2 Utility Function Model
Timely and accurate information transmission is paramount. Increased delay leads to control decisions based on outdated information, while transmission errors cause deviations between the actual EV charging power and the grid’s absorption target. Utility functions are employed to quantify the degree of satisfaction for these heterogeneous requirements across different services.
First, a unified timeliness utility function $$U(\tau)$$ is defined for the end-to-end interaction process, as the overall demand response must be completed within a strict time window to track renewable generation.
$$U(\tau) =
\begin{cases}
1, & \tau < t \\
\alpha – \beta \tau, & t \leq \tau < T \\
0, & \tau \geq T
\end{cases}$$
where $\tau$ is the total communication delay, $t$ is the minimum delay for optimal tracking, $T$ is the maximum allowable delay for effective control, and $\alpha$, $\beta$ are model parameters. All four services share this functional form but with different parameter values $(t, T, \alpha, \beta)$.
Second, distinct accuracy utility functions are defined for the upload and dispatch services, reflecting their different data characteristics and criticality.
For upload services:
- EV to Edge Server (Service 1): $$U_{u,EI}(\eta) =
\begin{cases}
1, & \eta \geq Q_{u,EI} \\
1 – \exp(-\alpha_{u,EI} (\eta + \beta_{u,EI})), & q_{u,EI} \leq \eta < Q_{u,EI} \\
0, & \eta < q_{u,EI}
\end{cases}$$
This function shows tolerance for moderate inaccuracies, as it plateaus quickly after the minimum requirement $q_{u,EI}$ is met. - Edge Server to Cloud (Service 2): $$U_{u,IS}(\eta) =
\begin{cases}
1, & \eta \geq Q_{u,IS} \\
(\eta – q_{u,IS})^{\alpha_{u,IS}}, & q_{u,IS} \leq \eta < Q_{u,IS} \\
0, & \eta < q_{u,IS}
\end{cases}$$
This convex-increasing function reflects the high and increasing marginal value of accuracy for the critical capacity parameter.
For dispatch services (Services 3 & 4), which require high precision, a similar convex function is used:
$$U_{d,SI/IE}(\eta) =
\begin{cases}
1, & \eta \geq Q_{d,SI/IE} \\
1 – (Q_{d,SI/IE} – \eta)^{\alpha_{d,SI/IE}}, & q_{d,SI/IE} \leq \eta < Q_{d,SI/IE} \\
0, & \eta < q_{d,SI/IE}
\end{cases}$$
where $d,SI$ denotes Cloud-to-Edge dispatch (Service 3) and $d,IE$ denotes Edge-to-EV dispatch (Service 4). The parameter $\alpha_{d,IE} > \alpha_{d,SI}$ indicates the even higher criticality of the final command sent to the electric vehicle.
1.3 Communication Resource Allocation Model
We consider an Orthogonal Frequency-Division Multiple Access (OFDMA) based network. The core optimization variables are the number of sub-channels allocated to each communication link and the transmit power on each sub-channel. The model is formulated for the EV-to-Edge Server uplink, with analogous models for the other links.
Let $N$ be the number of Edge Servers, and $M_n$ be the number of EVs under server $n$. The data rate $R_{nm}$ for EV $m$ connected to server $n$ is:
$$R_{nm} = \sum_{j=1}^{K_{nm}^{EI}} b^{EI} \log_2 \left(1 + \frac{p_{nm,j}^{EI} g_{nm,j}^{EI}}{\sigma_{nm,j}^{EI}} \right)$$
where $K_{nm}^{EI}$ is the number of sub-channels allocated, $b^{EI}$ is the bandwidth per sub-channel, $p_{nm,j}^{EI}$ is the transmit power, $g_{nm,j}^{EI}$ is the channel gain, and $\sigma_{nm,j}^{EI}$ is the noise power on sub-channel $j$.
The total delay $\tau_{nm}^{EI}$ includes transmission delay and propagation delay:
$$\tau_{nm}^{EI} = \frac{l_{nm}^{EI}}{R_{nm}} + \frac{d_{nm}^{EI}}{v}$$
where $l_{nm}^{EI}$ is the data size and $d_{nm}^{EI}$ is the distance. The delay for the EV cluster upload (Service 1) is the maximum among all EVs: $\tau^{EI} = \max_{n,m}(\tau_{nm}^{EI})$.
The transmission accuracy (packet success rate) $\eta^{EI}$ for the EV cluster upload is derived from the bit error rate (BER):
$$e_{nm,j}^{EI} = 1 – \exp\left(-\frac{p_{nm,j}^{EI} g_{nm,j}^{EI}}{\sigma_{nm,j}^{EI}}\right)$$
$$\eta^{EI} = 1 – \frac{ \sum_{n=1}^{N} \sum_{m=1}^{M_n} \sum_{j=1}^{K_{nm}^{EI}} e_{nm,j}^{EI} \cdot (b^{EI} \log_2(1+\frac{p_{nm,j}^{EI} g_{nm,j}^{EI}}{\sigma_{nm,j}^{EI}})) \cdot \tau_{nm}^{c,EI} }{ \sum_{n=1}^{N} \sum_{m=1}^{M_n} l_{nm}^{EI} }$$
where $\tau_{nm}^{c,EI}$ is the transmission time for the data from EV $m$ to server $n$. Similar expressions define the rate, delay, and accuracy for the Edge-to-Cloud ($IS$) links.
1.4 Problem Formulation
The overall system utility $U_{total}$ is a weighted sum of the utilities for the four services:
$$U_{total} = \omega_1 U_1 + \omega_2 U_2 + \omega_3 U_3 + \omega_4 U_4$$
where $U_1 = \omega_c U(\tau^{EI}) + \omega_e U_{u,EI}(\eta^{EI})$ for Service 1, $U_2 = \omega_c U(\tau^{IS}_{up}) + \omega_e U_{u,IS}(\eta^{IS}_{up})$ for Service 2, $U_3 = \omega_c U(\tau^{SI}_{down}) + \omega_e U_{d,SI}(\eta^{SI}_{down})$ for Service 3, and $U_4 = \omega_c U(\tau^{IE}_{down}) + \omega_e U_{d,IE}(\eta^{IE}_{down})$ for Service 4. The weights satisfy $\omega_1+\omega_2+\omega_3+\omega_4=1$ and $\omega_c+\omega_e=1$.
The optimization problem P1 is formulated as:
$$\textbf{P1: } \max U_{total}$$
subject to:
$$\text{(C1) } U(\tau^{EI}), U(\tau^{IS}), U(\tau^{SI}), U(\tau^{IE}) > 0; \quad U_{u,EI}(\eta^{EI}), U_{u,IS}(\eta^{IS}), U_{d,SI}(\eta^{SI}), U_{d,IE}(\eta^{IE}) > 0$$
$$\text{(C2) } \sum_{m=1}^{M_n} K_{nm}^{EI} \leq K^{EI}, \forall n; \quad \sum_{n=1}^{N} K_{n}^{IS} \leq K^{IS}$$
$$\text{(C3) } \sum_{j=1}^{K_{nm}^{EI}} p_{nm,j}^{EI} \leq P^{E}, \forall n,m; \quad \sum_{i=1}^{K_{n}^{IS}} p_{n,i}^{IS} + \sum_{m=1}^{M_n}\sum_{j=1}^{K_{nm}^{EI}} p_{nm,j}^{IE} \leq P^{I}, \forall n; \quad \sum_{i=1}^{K_{n}^{SI}} p_{n,i}^{SI} \leq P^{S}, \forall n$$
where $K^{EI}$ and $K^{IS}$ are the total available sub-channels for EV-Edge and Edge-Cloud links, respectively. $P^E$, $P^I$, and $P^S$ are the maximum transmit power for EVs, Edge Servers, and the Cloud Platform. Constraint (C1) ensures minimum performance for all services. (C2) is the sub-channel limit. (C3) is the power budget constraint, noting that the Edge Server power $P^I$ is shared for its uplink ($IS$) and downlink ($IE$) to EVs.
2. Resource Allocation Algorithm for the EV Aggregation Control Network
2.1 Solution Framework
Problem P1 is a Mixed-Integer Nonlinear Programming (MINLP) problem. We decompose it for efficient solution: 1) Determine the number of sub-channels $K_{nm}^{EI}$ for each link (integer variable). 2) For a given $K_{nm}^{EI}$, assign specific sub-channels to EVs (discrete matching). 3) Allocate power $p_{nm,j}^{EI}$ across the assigned sub-channels (continuous variable). We propose a hybrid algorithm combining a metaheuristic, matching theory, and convex optimization.
2.2 Sub-channel Assignment via Bilateral Matching
Given the number of sub-channels $K_{nm}^{EI}$ allocated to each EV $m$, the problem of assigning specific channel indices is a one-to-many matching problem between EVs and sub-channels. Each EV ranks sub-channels based on channel gain $g_{nm,j}^{EI}$ (preferring higher gains), and each sub-channel ranks EVs based on the same metric. The matching process proceeds as follows until all EVs are assigned their quota $K_{nm}^{EI}$: 1) Each EV proposes to its top-$K_{nm}^{EI}$ preferred sub-channels. 2) A sub-channel tentatively accepts the proposal from its most preferred EV among the proposers and rejects others. 3) Rejected EVs remove that sub-channel from their list and propose to their next preferred available sub-channel. This process yields a stable matching.
2.3 Power Allocation via Improved Water-Filling
For a given set of sub-channels assigned to an EV $m$, the optimal power allocation to maximize its data rate under a power constraint is given by the water-filling principle. Instead of iterative water-level searching, we use a Lagrangian method for faster computation. The solution for power on sub-channel $j$ is:
$$p_{nm,j}^{EI} = \max \left(0, \frac{1}{\lambda} – \frac{\sigma_{nm,j}^{EI}}{g_{nm,j}^{EI}} \right)$$
where $\lambda$ is the Lagrange multiplier (water-level). The relationship between powers on two sub-channels $j$ and $j’$ is:
$$\frac{1}{g_{nm,j}^{EI}/\sigma_{nm,j}^{EI}} + p_{nm,j}^{EI} = \frac{1}{g_{nm,j’}^{EI}/\sigma_{nm,j’}^{EI}} + p_{nm,j’}^{EI}$$
This allows for direct calculation once the power on one channel is known. The algorithm sorts sub-channels by $\sigma_{nm,j}^{EI}/g_{nm,j}^{EI}$, iteratively allocates power starting from the best channel, and stops when the power budget $P^E$ is met or all channels are active.
2.4 Sub-channel Number Allocation via Enhanced Dung Beetle Optimization
The master problem of finding the optimal vector of $K_{nm}^{EI}$ (the number of channels per EV link) is solved using an enhanced Dung Beetle Optimization (DBO) algorithm. DBO mimics the rolling, breeding, foraging, and stealing behaviors of dung beetles. We enhance the basic DBO by incorporating a sinusoidal strategy during the position update of the “rolling” beetles to increase randomness and avoid premature convergence.
The position of a beetle represents a potential solution vector $\mathbf{K} = [K_{11}^{EI}, K_{12}^{EI}, …, K_{NM_N}^{EI}]$. The fitness function is the total utility $U_{total}$ computed after performing bilateral matching and power allocation for that specific $\mathbf{K}$.
The update for a rolling beetle $i$ at iteration $T+1$ is modified as:
$$\mathbf{K}_i^{T+1} = \mathbf{K}_i^{T} + \sin(\phi_1) \cdot | \sin(\phi_2) \cdot \mathbf{K}_{worst}^T – \epsilon_1 \cdot \mathbf{K}_i^{T} |$$
where $\phi_1, \phi_2$ are random angles, $\mathbf{K}_{worst}^T$ is the worst solution, and $\epsilon_1$ is a golden-section coefficient that adapts during iteration. This sinusoidal disturbance enhances exploration. After update, the continuous position is normalized and discretized to an integer channel number.
The complete algorithm workflow is as follows:
- Initialize the DBO population (positions are channel number vectors).
- For each beetle’s position (solution $\mathbf{K}$):
- For the EV-Edge uplink: Use bilateral matching to assign specific channels based on $K_{nm}^{EI}$.
- For all links: Use the improved water-filling algorithm to allocate power.
- Calculate the total utility $U_{total}$ as the fitness.
- Update the DBO population positions based on rolling, breeding, foraging, and stealing rules (with sinusoidal enhancement for rollers).
- Repeat steps 2-3 until convergence or the maximum iteration count is reached.
- Output the best-found solution $\mathbf{K}^*$ and the corresponding channel assignment and power allocation.
3. Simulation Experiments and Analysis
3.1 Algorithm Convergence Analysis
We construct a simulation scenario with parameters summarized in Table 1.
| Parameter | Value |
|---|---|
| Number of EV Users | 20 |
| Number of Edge Servers ($N$) | 2 |
| Total Sub-channels ($K^{EI}$, $K^{IS}$) | 64, 4 |
| Data Size ($l_{nm}^{EI}$, $l_{n}^{IS}$) | 2 kbit, 1 kbit |
| Bandwidth per Sub-channel ($b^{EI}$, $b^{IS}$) | 0.25 kHz |
| Max Transmit Power ($P^E$, $P^I$, $P^S$) | 3 W, 13 W, 3 W |
| Utility Weights ($\omega_c$, $\omega_e$) | 0.5, 0.5 |
| Service Weights ($\omega_1$, $\omega_2$, $\omega_3$, $\omega_4$) | 0.25, 0.2, 0.35, 0.2 |
We compare three methods: Method I (Proposed: Enhanced DBO + Bilateral Matching), Method II (DBO only for both number and specific channel assignment), and Method III (Adaptive Particle Swarm Optimization + Bilateral Matching). The convergence curves for the total utility $U_{total}$ are shown below.
Method I converges rapidly to a utility of 0.963 within 10 iterations. In contrast, Method II and Method III require 66 and 97 iterations, respectively, to converge to similar or lower values. This demonstrates the superior convergence speed of the proposed hybrid approach. The integration of bilateral matching allows Method I to find the optimal channel assignment for a given channel number vector in a single step, whereas Method II must search for this assignment through iterative DBO updates, significantly increasing complexity.
3.2 Algorithm Complexity Analysis
The proposed algorithm reduces complexity in two key areas. First, the combination of DBO and bilateral matching shrinks the search space for channel assignment from approximately $2^{MK}$ (for exhaustive search over specific channels) to searching over the integer vector $\mathbf{K}$ of length equal to the number of EV links, which is much smaller. Second, the improved water-filling based on Lagrangian multipliers eliminates the iterative water-level search. Figure 7 compares the average execution time for power allocation per user between the classical iterative water-filling and our improved method. The improved method reduces the execution time for EV users, Edge Servers, and the Cloud Platform to 0.47, 0.75, and 0.33 times that of the classical method, respectively. The Edge Server allocation takes longer as it handles power distribution for both its uplink (4 channels) and downlink to EVs (64 channels).
3.3 System Communication Performance Analysis
We compare the proposed hierarchical joint optimization method against two benchmarks: 1) Channel Averaging: Sub-channels are divided equally among users, followed by water-filling power allocation. 2) Layer-wise Optimization: Each layer’s resources are optimized independently to maximize its own local utility, ignoring the global impact.
The overall utility achieved by each method is: Proposed Method (0.963), Channel Averaging (0.9586), Layer-wise Optimization (0.9146). The proposed method provides a balanced improvement, increasing total utility by 0.44% and 5.28% over the two benchmarks, respectively.
A detailed breakdown of the utility components reveals why. The layer-wise method severely degrades the performance of the Edge-to-Cloud uplink (Service 2), as optimizing locally at the Edge does not account for the critical need for highly accurate data at the Cloud. The channel averaging method slightly improves Service 2 but at the cost of reducing performance for the more latency-sensitive and high-volume EV-Edge links (Services 1 and 4). The proposed joint optimization successfully balances the heterogeneous requirements of all four services, maximizing the overall system utility for the electric vehicle aggregation control process.
4. Conclusion
This paper addresses the critical communication needs for scalable electric vehicle aggregation control. We propose a three-tier communication network architecture aligned with the hierarchical control model and analyze the distinct timeliness and accuracy requirements of the data flows at each tier. Utility functions are developed to quantitatively model these heterogeneous needs. The communication resource (sub-channel and power) allocation problem is formulated with the objective of maximizing the overall system utility.
To solve this complex problem efficiently, a novel hybrid algorithm is proposed. It decomposes the problem and employs an enhanced Dung Beetle Optimization algorithm with a sinusoidal strategy for global search over channel numbers, bilateral matching theory for stable specific channel assignment, and an improved water-filling algorithm based on Lagrangian multipliers for fast optimal power allocation.
Simulation results confirm that the proposed algorithm converges significantly faster than benchmark metaheuristic approaches and has lower computational complexity. Most importantly, it effectively balances the communication resources across all services in the electric vehicle aggregation control pipeline, leading to a higher overall information transmission utility compared to equal allocation or independent per-layer optimization strategies. This work provides a practical communication resource management solution to underpin reliable and efficient large-scale vehicle-to-grid integration.