With the rapid global shift towards sustainable transportation, the large-scale adoption of battery EV cars has become a critical component in addressing energy crises and environmental pollution. As a researcher focused on power distribution systems, I recognize that the planning of charging infrastructure for these battery EV cars must be scientifically rigorous and economically viable. The placement of EV charging stations directly impacts operator costs, grid reliability, and the convenience for battery EV car owners. In this article, I propose an optimization-based method for siting EV charging stations in distribution networks, aiming to minimize power losses and maintain stable voltage profiles while considering operational constraints and the integration of vehicle-to-grid (V2G) technologies. The proliferation of battery EV cars necessitates innovative planning approaches to ensure that charging services are accessible without compromising grid performance. This work delves into heuristic search algorithms, specifically Particle Swarm Optimization (PSO) and Symbiotic Organism Search (SOS), to solve the complex optimization problem, with simulations conducted on the IEEE 33-node test system. The goal is to provide a framework that balances technical efficiency with practical serviceability for battery EV car users.
The increasing penetration of battery EV cars introduces new loads that can strain distribution networks if not properly managed. From my perspective, optimizing charging station locations is not merely about reducing costs but also about enhancing grid resilience and user satisfaction. Existing studies often focus on cost-benefit analyses or user proximity, but they may overlook the dynamic interactions between battery EV car charging behaviors and grid stability. Here, I emphasize a holistic model that incorporates V2G capabilities, allowing battery EV cars to serve as distributed energy resources. This approach aligns with global efforts to decarbonize transportation and energy sectors, making it imperative to develop robust planning tools. In the following sections, I will detail the mathematical formulation, heuristic algorithms, simulation results, and insights gained from this research, consistently highlighting the role of battery EV cars in modern power systems.
To establish a foundation, let me define the key components of the distribution network model. Consider a radial distribution system with $N_b$ nodes and $N_{br}$ branches. Each branch $br$ has a resistance $R_{br}$, reactance $X_{br}$, impedance $Z_{br}$, and current $I_{br}$. The impedance is expressed as:
$$Z_{br} = R_{br} + jX_{br}$$
The apparent power loss in branch $br$ is calculated as:
$$S_{br} = I_{br}^2 Z_{br}$$
The total system load, encompassing both conventional loads and those from battery EV car charging stations, is given by:
$$L_{total} = \sum_{i=1}^{N_b} (L_i + L_{cs,i})$$
where $L_i$ is the real-time load at node $i$, and $L_{cs,i}$ is the additional charging load at node $i$ due to EV charging stations. This additional load is significant because each battery EV car connected to a station increases demand, affecting local voltage and power flow. The charging load $L_{cs,i}$ depends on the number of battery EV cars in G2V (grid-to-vehicle) and V2G (vehicle-to-grid) modes:
$$L_{cs,i} = n_{G2V} r_c + n_{V2G} r_d$$
Here, $n_{G2V}$ represents the number of battery EV cars charging from the grid, $n_{V2G}$ denotes those discharging to the grid, $r_c$ is the charging power per battery EV car, and $r_d$ is the discharging power per battery EV car. Typically, only a fraction of battery EV cars support V2G, but this capability can help stabilize the grid by providing ancillary services. The total active power loss across all branches is:
$$P_{loss} = \sum_{br=1}^{N_{br}} P_{loss,br}$$
where $P_{loss,br}$ is the active power loss in branch $br$. My objective is to minimize this loss, leading to the objective function:
$$\min f = \min P_{loss}$$
This minimization is subject to several constraints that ensure feasible grid operation. The power balance constraint requires that the total power supplied equals the total load plus losses:
$$P_S = \sum_{i=1}^{N_b} P_i + P_{loss}$$
where $P_S$ is the active power from the substation, and $P_i$ is the injected power at node $i$. Voltage and current limits must be adhered to, as defined by:
$$
\begin{cases}
U_{min} \leq U_i \leq U_{max} \\
I_{min} \leq I_{br} \leq I_{max}
\end{cases}
$$
with $U_i$ being the voltage magnitude at node $i$, bounded by $U_{min}$ and $U_{max}$, and $I_{br}$ the current in branch $br$, bounded by $I_{min}$ and $I_{max}$. These constraints are critical for maintaining power quality and preventing equipment overloads, especially with the added variability from battery EV car charging.
To encapsulate the parameters and variables, I provide Table 1, which summarizes the notation used in the mathematical model. This table aids in clarifying the relationships between different components, particularly those involving battery EV cars.
| Symbol | Description | Unit |
|---|---|---|
| $N_b$ | Number of nodes in the distribution network | – |
| $N_{br}$ | Number of branches in the distribution network | – |
| $R_{br}$ | Resistance of branch $br$ | Ω |
| $X_{br}$ | Reactance of branch $br$ | Ω |
| $Z_{br}$ | Impedance of branch $br$ | Ω |
| $I_{br}$ | Current in branch $br$ | A |
| $S_{br}$ | Apparent power loss in branch $br$ | VA |
| $L_i$ | Base load at node $i$ | kW |
| $L_{cs,i}$ | Additional load from battery EV car charging at node $i$ | kW |
| $n_{G2V}$ | Number of battery EV cars in G2V mode | – |
| $n_{V2G}$ | Number of battery EV cars in V2G mode | – |
| $r_c$ | Charging power per battery EV car | kW |
| $r_d$ | Discharging power per battery EV car | kW |
| $P_{loss}$ | Total active power loss in the system | kW |
| $P_S$ | Total active power from substation | kW |
| $U_i$ | Voltage magnitude at node $i$ | pu |
| $U_{min}, U_{max}$ | Minimum and maximum voltage limits | pu |
| $I_{min}, I_{max}$ | Minimum and maximum current limits | A |
The optimization problem is non-linear and high-dimensional due to the power flow equations and multiple constraints. Traditional gradient-based methods may struggle with such complexities, prompting the use of heuristic search algorithms. In my work, I compare two such algorithms: Particle Swarm Optimization (PSO) and Symbiotic Organism Search (SOS). PSO is a well-established metaheuristic inspired by bird flocking, where particles adjust their positions based on personal and global best solutions. However, I find that SOS, a newer algorithm based on ecological symbiosis, offers superior performance for this application. SOS mimics mutualistic, commensal, and parasitic relationships in nature to explore the solution space effectively.
Let me detail the SOS algorithm as I implemented it. The algorithm begins with initializing a population of $N$ organisms (solutions), each representing a potential set of charging station locations. For a $D$-dimensional problem (where $D$ is the number of charging stations to site), each organism $X_i$ is initialized as:
$$X_i = L + \text{rand}(1, D) \times (U – L)$$
where $L$ and $U$ are the lower and upper bounds of the search space (e.g., node indices), and $\text{rand}(1, D)$ generates a random vector of length $D$ with values uniformly distributed in [0,1]. This initialization ensures diversity across the population. The fitness of each organism is evaluated using the objective function $f$, which computes the power loss based on the charging station placements and the associated battery EV car loads.
The SOS algorithm proceeds through three phases iteratively. In the mutualism phase, two distinct organisms $X_i$ and $X_j$ (with $i \neq j$) interact to mutually benefit. New organisms are generated as follows:
$$
\begin{aligned}
X_{i,new} &= X_i + \text{rand}(0,1) \times (X_{best} – M_v \times BF_1) \\
X_{j,new} &= X_j + \text{rand}(0,1) \times (X_{best} – M_v \times BF_2) \\
M_v &= \frac{X_i + X_j}{2}
\end{aligned}
$$
Here, $X_{best}$ is the current best organism in the population, $M_v$ is the mutual vector representing the average relationship between $X_i$ and $X_j$, and $BF_1$ and $BF_2$ are benefit factors randomly chosen as 1 or 2 to simulate varying degrees of benefit. The new organisms are evaluated, and they replace $X_i$ and $X_j$ only if they yield better fitness (i.e., lower power loss). This phase encourages cooperation, potentially leading to faster convergence towards optimal siting for battery EV car charging stations.
In the commensalism phase, organism $X_i$ benefits from $X_j$ without affecting $X_j$. The update rule is:
$$X_{i,new} = X_i + \text{rand}(-1,1) \times (X_{best} – X_j)$$
The random number $\text{rand}(-1,1)$ introduces both positive and negative adjustments, allowing for exploration. Again, $X_{i,new}$ replaces $X_i$ if it improves fitness. This phase helps organisms escape local optima by leveraging information from others, which is crucial when dealing with the irregular load patterns introduced by battery EV cars.
The parasitism phase introduces a competitive element. A parasite vector $P_v$ is created by copying $X_i$ and modifying some of its dimensions with random values within the bounds. Then, $P_v$ competes with a randomly selected host organism $X_j$. If $P_v$ has better fitness, it replaces $X_j$; otherwise, $X_j$ survives. This phase mimics natural selection, eliminating weak solutions and promoting diversity. The algorithm cycles through these phases until a termination criterion is met, such as a maximum number of iterations or convergence tolerance. The overall process is summarized in Table 2, which compares the characteristics of PSO and SOS for this planning problem.
| Aspect | Particle Swarm Optimization (PSO) | Symbiotic Organism Search (SOS) |
|---|---|---|
| Inspiration | Social behavior of bird flocking | Ecological symbiotic relationships |
| Key Parameters | Inertia weight, cognitive and social coefficients | Benefit factors, random numbers in phases |
| Population Interaction | Global and personal best influence | Mutualism, commensalism, parasitism |
| Exploration vs. Exploitation | Balanced through velocity updates | High exploration via parasitic phase |
| Computational Complexity | Moderate, depends on swarm size | Similar to PSO, but phase-based |
| Suitability for Battery EV Car Planning | Effective but may converge prematurely | Superior in avoiding local optima |
To validate the proposed method, I conducted simulations using the IEEE 33-node distribution test system. This system has a base voltage of 11 kV, base power of 100 MVA, and total load of 3715 kW and 2300 kVar. Power flow is solved via the forward-backward sweep method, which efficiently handles radial networks. I consider two cases: Case 1 involves siting charging stations without network partitioning, while Case 2 divides the network into three regions to ensure that battery EV car owners in all areas have reasonable access to charging services. This division reflects practical urban planning, where charging infrastructure should be distributed to reduce travel distances for battery EV car users. In both cases, I assume each charging station serves a varying number of battery EV cars: 10, 20, and 30 per station. The charging power $r_c$ is set to 20 kW per battery EV car in G2V mode, and discharging power $r_d$ is 9 kW per battery EV car in V2G mode, with 10% of connected battery EV cars supporting V2G. These parameters are based on typical fast-charging scenarios for modern battery EV cars.
The simulation results for Case 1 are presented in Tables 3 and 4, showcasing the outcomes from PSO and SOS algorithms, respectively. Each table includes the optimal charging station locations (node indices), the minimum node voltage in per unit (pu), the total active power loss in kilowatts (kW), and the computation time in seconds (s). The locations are selected to minimize power loss while adhering to constraints. As observed, both algorithms tend to choose nodes 2, 4, and 19 across different numbers of battery EV cars, indicating that these nodes are favorable due to their proximity to the substation and load centers. However, SOS consistently achieves lower power losses and higher minimum voltages than PSO, demonstrating its efficacy. For instance, with 30 battery EV cars per station, SOS reduces power loss by approximately 0.28 kW compared to PSO, while maintaining a slightly better voltage profile. The computation times are comparable, though SOS shows variability depending on the population size and iteration count.
| Number of Battery EV Cars per Station | Charging Station Locations (Nodes) | Minimum Node Voltage (pu) | Power Loss (kW) | Computation Time (s) |
|---|---|---|---|---|
| 10 | 2, 4, 19 | 0.9025 | 212.12 | 467.25 |
| 20 | 2, 4, 19 | 0.9016 | 213.33 | 440.71 |
| 30 | 2, 4, 19 | 0.9005 | 215.20 | 531.52 |
| Number of Battery EV Cars per Station | Charging Station Locations (Nodes) | Minimum Node Voltage (pu) | Power Loss (kW) | Computation Time (s) |
|---|---|---|---|---|
| 10 | 2, 4, 19 | 0.9037 | 212.01 | 581.42 |
| 20 | 2, 4, 19 | 0.9031 | 213.14 | 387.54 |
| 30 | 2, 4, 19 | 0.9022 | 214.92 | 470.05 |
To further analyze the voltage distribution, I computed the voltage profiles across all nodes for Case 1 under SOS optimization. The voltages remain within acceptable limits (0.9 pu to 1.05 pu), but as the number of battery EV cars increases, the minimum voltage decreases slightly due to higher loads. This trend underscores the importance of siting charging stations strategically to mitigate voltage drops, especially in networks with high penetration of battery EV cars. The convergence behavior of both algorithms is illustrated through the power loss over iterations. SOS typically converges faster than PSO, reaching near-optimal solutions within fewer iterations. This efficiency is attributed to the symbiotic phases that enhance exploration and exploitation simultaneously, making SOS well-suited for complex planning problems involving battery EV cars.
For Case 2, where the network is partitioned into three regions, the optimization results are shown in Tables 5 and 6. The division forces the algorithms to select stations from each region, promoting geographical dispersion. Consequently, the optimal locations shift to nodes like 2, 20, and 19, which are more spread out compared to Case 1. This dispersal increases power losses marginally because stations are no longer concentrated near the substation, but it significantly improves accessibility for battery EV car owners across the network. For example, with 30 battery EV cars per station, SOS yields a power loss of 233.74 kW, which is higher than in Case 1 but still within operational bounds. The minimum voltages are slightly lower than in Case 1, reflecting the impact of distributed charging loads. However, the trade-off is justified by the enhanced service coverage, reducing the need for long-distance travel to charge a battery EV car.
| Number of Battery EV Cars per Station | Charging Station Locations (Nodes) | Minimum Node Voltage (pu) | Power Loss (kW) | Computation Time (s) |
|---|---|---|---|---|
| 10 | 2, 20, 19 | 0.9027 | 215.57 | 326.51 |
| 20 | 2, 20, 19 | 0.9019 | 223.38 | 507.69 |
| 30 | 2, 20, 19 | 0.9006 | 236.18 | 956.87 |
| Number of Battery EV Cars per Station | Charging Station Locations (Nodes) | Minimum Node Voltage (pu) | Power Loss (kW) | Computation Time (s) |
|---|---|---|---|---|
| 10 | 2, 20, 19 | 0.9031 | 214.75 | 348.65 |
| 20 | 2, 20, 19 | 0.9023 | 221.39 | 598.24 |
| 30 | 2, 20, 19 | 0.9006 | 233.74 | 1378.42 |
A comparative analysis of the two cases reveals insightful trends. In Table 7, I aggregate the power losses and minimum voltages for both algorithms across different battery EV car counts. This summary highlights that SOS consistently outperforms PSO in minimizing losses and preserving voltage stability, regardless of network partitioning. The performance gap widens as the number of battery EV cars increases, emphasizing SOS’s robustness in handling higher loads. Moreover, the computation times for SOS are generally higher in Case 2 due to the increased complexity from region-based constraints, but the trade-off is acceptable given the improved solution quality. These findings affirm that heuristic algorithms like SOS are vital for optimizing charging infrastructure in the era of battery EV car proliferation.
| Case | Algorithm | Number of Battery EV Cars | Power Loss (kW) | Minimum Voltage (pu) |
|---|---|---|---|---|
| Case 1 | PSO | 10 | 212.12 | 0.9025 |
| 20 | 213.33 | 0.9016 | ||
| 30 | 215.20 | 0.9005 | ||
| Case 1 | SOS | 10 | 212.01 | 0.9037 |
| 20 | 213.14 | 0.9031 | ||
| 30 | 214.92 | 0.9022 | ||
| Case 2 | PSO | 10 | 215.57 | 0.9027 |
| 20 | 223.38 | 0.9019 | ||
| 30 | 236.18 | 0.9006 | ||
| Case 2 | SOS | 10 | 214.75 | 0.9031 |
| 20 | 221.39 | 0.9023 | ||
| 30 | 233.74 | 0.9006 |
To deepen the discussion, I explore the sensitivity of the results to key parameters. For instance, varying the charging power $r_c$ from 20 kW to 30 kW per battery EV car increases power losses proportionally, but the optimal station locations remain relatively stable, indicating the robustness of the model. Similarly, adjusting the percentage of V2G-capable battery EV cars from 10% to 20% reduces losses slightly due to grid support during peak times. These variations underscore the importance of accurate parameter estimation in real-world planning for battery EV car charging networks. Furthermore, I consider the impact of load growth over time; as more battery EV cars are adopted, the planning model can be extended to multi-stage optimization, ensuring that charging infrastructure evolves with demand.
In terms of algorithmic performance, SOS exhibits faster convergence rates than PSO, as evidenced by the power loss reduction over iterations. This efficiency stems from the symbiotic interactions that prevent premature convergence to local optima. For practical applications, this means that SOS can handle larger networks with hundreds of nodes, making it scalable for urban areas with high densities of battery EV cars. However, both algorithms require careful tuning of parameters, such as population size and iteration limits, to balance accuracy and computational effort. I recommend using SOS as the primary solver for similar planning problems, supplemented by PSO for benchmarking.
The integration of V2G technology adds a layer of complexity but also opportunity. By allowing battery EV cars to discharge power, charging stations can act as virtual power plants, alleviating grid stress during high-demand periods. My model incorporates this by including $n_{V2G}$ and $r_d$ in the load calculations. In simulations, V2G contribution helps flatten the voltage profile and reduce losses, especially when battery EV cars are parked during daytime hours. This aligns with smart grid initiatives and enhances the value proposition of battery EV car ownership. Future work could explore dynamic pricing models to incentivize V2G participation, further optimizing grid operations.
From a broader perspective, this planning method contributes to the sustainable integration of battery EV cars into power systems. By minimizing losses and ensuring voltage stability, it supports grid reliability while promoting EV adoption. The regional partitioning in Case 2 demonstrates that user convenience can be prioritized without severely compromising technical metrics. This is crucial for public acceptance, as battery EV car owners need reliable and accessible charging options. Policymakers and utility companies can leverage such models to design incentive programs and infrastructure investments that align with carbon reduction goals.
In conclusion, I have presented a comprehensive planning method for EV charging stations based on heuristic search algorithms. The mathematical model minimizes power loss subject to operational constraints, and the SOS algorithm proves superior to PSO in finding optimal solutions. Simulations on the IEEE 33-node system show that strategic siting of charging stations can significantly reduce losses and maintain voltage levels, even with high penetration of battery EV cars. The division of the network into regions, while increasing losses slightly, greatly enhances service accessibility for battery EV car owners. This trade-off is acceptable in real-world scenarios where user satisfaction is paramount. Future research directions include incorporating traffic flow models, renewable energy sources, and real-time data analytics to further refine the planning process. As battery EV cars become ubiquitous, such advanced planning tools will be indispensable for building resilient and efficient charging infrastructure.