The rapid global transition in energy structures is marked by the unprecedented integration of intermittent renewable energy sources and the explosive growth in the adoption of battery electric vehicles (EV cars). This “dual-high” landscape—characterized by high penetration of volatile renewables and high penetration of battery EV cars—fundamentally challenges the traditional operational paradigms of power systems. The inherent intermittency of renewable generation, when coupled with the concentrated charging demand from a large fleet of battery EV cars, can exacerbate grid peak-valley differences by 30% to 50%, exposing critical vulnerabilities in maintaining a rigid supply-demand balance.
Current grid operations under high-penetration battery EV car charging scenarios reveal multiple, often conflicting, issues. A primary challenge is spatiotemporal mismatch: EV charging demand typically peaks during evening load hours, which often coincides with low solar photovoltaic (PV) output, creating a reverse superposition that significantly increases peak-shaving pressure. Secondly, capacity constraints emerge as traditional distribution networks, not designed for large-scale battery EV car integration, see transformer load ratios approaching critical limits, elevating local overload risks. Thirdly, there is an imbalance in economic and environmental benefits: uncoordinated charging leads to prolonged user wait times and low daytime utilization of charging piles, while curtailment of wind and solar power remains a persistent issue in many regions. These contradictions make it difficult to balance grid safety, economic efficiency, and sustainability using conventional approaches.
To address these interconnected challenges, this study proposes a novel data-driven, multi-scale collaborative optimization framework. The core innovation lies in a two-layer optimization architecture that synergizes long-term infrastructure planning with real-time operational control, specifically tailored for grids with high shares of renewables and battery EV cars.
1. Distribution Network Modeling and Load Profile Analysis
To assess the impact of high-penetration battery EV cars, a real-world urban distribution network from the Yangtze River Delta region was selected and modeled using the Open Distribution System Simulator (OpenDSS). The network is formalized as a graph $\mathcal{G} = (\mathcal{N}, \mathcal{E})$, where $\mathcal{N}$ represents the set of nodes (substations, transformers, user connection points) and $\mathcal{E}$ the set of branches (feeders, lines). Each branch $(i, j) \in \mathcal{E}$ is characterized by its complex impedance $Z_{ij} = R_{ij} + jX_{ij}$.
The steady-state operation is governed by the power flow equations, with the core power balance condition at each node $i \in \mathcal{N}$ given by:
$$
S_i = P_i + jQ_i = V_i \sum_{j \in \mathcal{N}} Y_{ij}^* V_j^*
$$
where $S_i$ is the complex power injection, $V_i$ is the node voltage, and $Y_{ij}$ is the admittance between nodes $i$ and $j$.
Customer loads are modeled using typical diurnal load profiles. The load factor $L_i$ for node $i$ is defined as:
$$
L_i = \frac{1}{T \cdot P_{i,\text{max}}} \int_0^T P_i(t) dt
$$
where $T$ is the observation period and $P_{i,\text{max}}$ is the maximum load at node $i$. The total system load must satisfy:
$$
\sum_{i \in \mathcal{N}} P_i(t) + P_{\text{loss}}(t) = P_s(t)
$$
where $P_s(t)$ is the power supplied at the source substation and $P_{\text{loss}}(t)$ is the total system active power loss. Analysis of the network’s load revealed a typical dual-peak profile, as summarized below.
| Peak Period | Load (MW) | Characteristics |
|---|---|---|
| Midday (10:00-12:00) | ~2.1 | Driven by commercial/industrial activity |
| Evening (18:00-21:00) | ~1.9 | Driven by residential demand |
| Minimum (04:00-06:00) | ~0.75 | Base load period |
2. Assessment and Sizing of Renewable Energy Resources
Quantifying the available local renewable resources is a prerequisite for effective integration. Wind and solar were identified as the most viable sources for the target network area.
2.1 Wind Energy Modeling
The wind speed distribution at the site is effectively characterized by the Weibull distribution. The probability density function $f(V)$ and cumulative distribution function $F(V)$ are:
$$
f(V) = \frac{k}{c} \left( \frac{V}{c} \right)^{k-1} e^{-(V/c)^k}
$$
$$
F(V) = 1 – e^{-(V/c)^k}
$$
where $V$ is the wind speed, $k$ is the shape parameter, and $c$ is the scale parameter. Based on seven years of hourly data (average speed $V_m = 7.7$ m/s, standard deviation $\sigma_V = 4.58$ m/s), the parameters were calculated as $k=1.47$ and $c=8.32$ m/s. The site’s annual average wind power density is 574.51 W/m², confirming good potential.
2.2 Solar Energy Modeling
The annual average daily solar irradiation is 4.14 kWh/m²/day. The Beta distribution provided the best fit for the solar resource data. Its PDF and CDF are:
$$
f(x) = \frac{x^{\alpha-1}(1-x)^{\beta-1}}{B(\alpha, \beta)}
$$
$$
F(x) = \frac{B_x(\alpha, \beta)}{B(\alpha, \beta)}
$$
where $B(\cdot)$ is the Beta function, $B_x(\cdot)$ is the incomplete Beta function, $\alpha=0.73$ and $\beta=1.40$ are shape parameters fitted to the data.
2.3 Optimal Renewable System Configuration
Using the Hybrid Optimization of Multiple Energy Resources (HOMER) model, the optimal renewable system was sized considering technical, economic, and environmental constraints. The resulting configuration is a hybrid system designed to serve the base load and a portion of the future battery EV car charging demand.
| Component | Specification | Key Parameters |
|---|---|---|
| Wind Turbine | 1 x 2.3 MW | Cut-in: 3.5 m/s, Rated: 15 m/s, Cut-out: 25 m/s |
| Photovoltaic System | 2.61 MWp | Monocrystalline Si, System Efficiency: 19.4% |
| Power Converter | — | Efficiency: 95% |
3. Data-Driven Renewable Energy System Performance Model
Machine learning (ML) models were developed to predict the performance of the sized wind and PV systems at the specific site, providing critical inputs for the battery EV car charging scheduler.
After data cleaning and anomaly removal, key features were identified. For the wind turbine model, wind speed was the dominant feature (Pearson correlation >0.94). For the PV model, global horizontal irradiance and ambient temperature were key. A k-Nearest Neighbors (kNN) regression algorithm was selected after comparative analysis. Optimal k-values were determined via grid search with 10-fold cross-validation (k=30 for wind, k=27 for PV). The models were evaluated on a separate test dataset using standard metrics.
| Model | RMSE | NRMSE | MAE | $R^2$ |
|---|---|---|---|---|
| Wind Turbine | 71.84 kW | 0.03 | 39.91 kW | 0.99 |
| PV System | 54.03 W | 0.06 | 35.23 W | 0.94 |
The high $R^2$ values indicate the kNN models accurately represent the system performance, enabling reliable prediction of renewable generation $W(t)$ and $S(t)$ for time $t$.
4. Battery EV Car Charging Planning and Cooperative Optimization
The core of this work is a smart charging/discharging algorithm that maximizes the use of local renewables for charging the fleet of battery EV cars and supports the grid during periods of low renewable output. The algorithm operates at the community level (e.g., customers under one distribution transformer). Users can opt into a Vehicle-to-Grid (V2G) agreement. The algorithm flow is as follows:
Step 1: At time $t$, collect multi-source data: predicted wind generation $W(t)$, predicted PV generation $S(t)$, and non-EV load $P(t)$.
Step 2: Generate a list $EVL(t)$ of connected battery EV cars, each with parameters: connection status, V2G agreement status, State of Charge (SOC), and connection node.
Step 3: For each EV in $EVL(t)$:
– If $W(t) + S(t) \geq P(t)$ (renewable surplus) and SOC < 100%, schedule for charging (G2V mode).
– If $W(t) + S(t) < P(t)$ (renewable deficit), an EV is considered for discharging (V2G mode) only if it has a V2G agreement AND its SOC $\geq$ SOC$_{threshold}$ (e.g., 90%, 75%, 60%).
Step 4: Perform a power flow analysis in OpenDSS with the updated EV loads/generation to ensure network constraints are respected.
Step 5: Increment time step $t = t + \Delta t$ and repeat.
The SOC of a battery EV car updates as follows. For charging with efficiency $\eta_{\text{charge}}$:
$$
\text{SOC}(t+\Delta t) = \text{SOC}(t) + \frac{P_{\text{charge}} \cdot \Delta t \cdot \eta_{\text{charge}}}{E_{\text{Battery}}} \times 100\%
$$
For discharging with efficiency $\eta_{\text{discharge}}$:
$$
\text{SOC}(t+\Delta t) = \text{SOC}(t) – \frac{P_{\text{discharge}} \cdot \Delta t}{E_{\text{Battery}} \cdot \eta_{\text{discharge}}} \times 100\%
$$
where $E_{\text{Battery}}$ is assumed to be 36 kWh and $P_{\text{charge/discharge}}$ is 3.6 kW for this study. A typical daily connection pattern for 80 battery EV cars is assumed, as shown below.
| Time Window | Connected EV Cars | Description |
|---|---|---|
| 00:00 – 05:00 | 78 | Night charging period |
| 06:00 – 09:00 | 60 | Morning period, some cars disconnected |
| 09:00 – 17:00 | 5 | Daytime, most cars in use |
| 18:00 – 21:00 | 40 | Evening period |
| 21:00 – 00:00 | 78 | Night charging period |
This operational layer is governed by an upper-layer planning optimization. The placement and capacity of charging infrastructure are optimized using an Improved Grey Wolf Optimization (IGWO) algorithm, which minimizes total cost (investment + grid upgrade) while respecting transformer and line capacity constraints derived from OpenDSS simulations. The combined IGWO-MPC framework ensures that the charging of numerous battery EV cars is both infrastructurally feasible and operationally optimal in real-time.
5. Simulation Results and Analysis
The proposed framework was simulated on the OpenDSS model of the target network (average load 1.51 MW, peak 3.6 MW) integrated with the 2.3 MW wind and 2.61 MW PV systems. Of the 80 simulated battery EV cars, 60 were assumed to have a V2G agreement. The impact of the SOC threshold for V2G eligibility was investigated.
The primary metric is the renewable energy charging ratio: the proportion of total energy delivered to battery EV cars that comes directly from wind and PV generation. With the proposed data-driven scheduling, the system achieved a 63.2% renewable energy charging ratio without any dedicated stationary storage. This significantly reduces the net load drawn from the main grid during charging events.
The V2G capability provides active grid support. The total daily energy exchange for different SOC thresholds demonstrates the algorithm’s flexibility:
| SOC$_{threshold}$ | Total Charging Energy | Total Discharging Energy | Net Energy from Grid |
|---|---|---|---|
| 90% | 2335.4 kWh | 540.0 kWh | 1795.4 kWh |
| 75% | 2392.0 kWh | 727.2 kWh | 1664.8 kWh |
| 60% | 2181.4 kWh | 964.8 kWh | 1216.6 kWh |
A lower SOC threshold allows more frequent and deeper discharge cycles, increasing grid support but also affecting the daily driving range assurance for users. The algorithm dynamically balances this trade-off. Furthermore, by shifting charging to periods of high renewable availability and reducing peak load through smart scheduling, the grid’s peak-to-valley difference is substantially reduced, enhancing overall stability. The decentralized nature of the community-level control also leads to higher charging point utilization and lower average waiting times for users of battery EV cars.
6. Conclusion
This study presents a comprehensive, data-driven solution for the coordinated planning and operation of power grids facing the dual integration of high-penetration renewables and high-penetration battery EV cars. The key contributions are threefold. First, it breaks the dependency on traditional storage by introducing a real-time data-driven scheduling algorithm that dynamically aligns the charging load of battery EV cars with renewable generation peaks, achieving a high renewable charging ratio without stationary storage. Second, it constructs a synergistic optimization system that combines OpenDSS-based network analysis, kNN-based renewable forecasting, and a decentralized MPC-based charging strategy, effectively balancing grid constraints with user autonomy. Third, the framework is highly adaptable, successfully demonstrating performance across different SOC thresholds and fleet sizes.
The simulation results validate the framework’s effectiveness: a 63.2% renewable charging ratio, reduced grid peak-valley difference, improved voltage stability, and enhanced user satisfaction through managed charging. Future work will focus on incorporating explicit uncertainty quantification for renewable forecasts and EV user behavior, integrating stationary storage for even greater flexibility, and refining user behavior models to include more nuanced price and convenience sensitivities. This approach provides a viable, low-cost pathway for the harmonious integration of battery EV cars into the future renewable-dominant power grid.