Abstract
The proliferation of electric vehicles (EVs) has necessitated robust charging infrastructure planning and operation to meet growing demand while ensuring grid stability and economic efficiency. This paper provides a comprehensive review of key technologies in EV charging, covering demand forecasting, infrastructure planning, scheduling strategies, and global demonstration projects. By integrating mathematical models, comparative analyses, and real-world applications, this study aims to synthesize current advancements and identify future research directions in this critical domain.
1. Introduction
Electric vehicles have emerged as a pivotal solution to reduce greenhouse gas emissions and mitigate environmental degradation, garnering significant global adoption. In China, the EV fleet grew from 2.11 million in 2018 to 18.13 million by June 2024, underscoring the urgent need for scalable charging infrastructure . The National Development Reform Commission’s goal to support over 20 million EVs by the end of the “14th Five-Year Plan” further highlights the importance of optimizing charging facility planning and operation .
Effective management of EV charging involves three interconnected phases: demand forecasting, infrastructure planning (location and capacity), and charging scheduling. Accurate demand forecasting serves as the foundation for these phases, enabling planners to anticipate spatial and temporal charging needs. Infrastructure planning must then balance user accessibility, operator costs, and grid constraints. Finally, intelligent scheduling strategies optimize power distribution to enhance grid stability and reduce costs for users and operators alike.
2. EV Charging Demand Forecasting
2.1 Time Distribution Forecasting
EV charging demand varies significantly over time, requiring distinct approaches for short-term (hours to days) and long-term (months to years) forecasting.
2.1.1 Short-Term Forecasting
Short-term forecasts guide daily charging dispatch and are influenced by variables like user behavior, weather, and holidays. Common methods include:
- Regression Analysis: Simple and fast, but limited by model assumptions. For example, multivariate linear regression incorporating temperature, humidity, and historical data has been used to predict demand with moderate accuracy .
- Artificial Intelligence (AI) Methods:
- Support Vector Machines (SVM): Effective for nonlinear relationships with small datasets. SVM outperformed linear regression in predicting electric bus charging demand .
- Neural Networks: LSTM and GRU networks excel at capturing temporal patterns. Adding attention mechanisms to LSTM/GRU further improved prediction accuracy for sequential data .
Table 1: Comparison of Short-Term Forecasting Methods
| Method | Advantages | Disadvantages | Applications |
|---|---|---|---|
| Regression | Simple, fast | Low accuracy for complex patterns | Basic demand trend analysis |
| SVM | Handles nonlinearity, small datasets | Struggles with large data | Short-term load forecasting |
| LSTM/GRU | Strong temporal modeling | Requires significant computational power | EV charging load prediction |
2.1.2 Long-Term Forecasting
Long-term forecasts inform infrastructure expansion and grid planning, focusing on EV penetration rates and regional growth. Key methods include:
- Data-Driven Models: Time series and grey models use historical EV population data, but are limited by short data histories .
- Bass Model: A dynamic model for new product adoption, adapted to EV growth by incorporating innovation (e.g., policy incentives) and imitation (e.g., peer influence) factors. The Bass model’s mathematical formulation is:\(n(t) = a(m – N(t)) + b\frac{N(t)}{m}(m – N(t))\) where \(n(t)\) is the adoption rate at time t, \(N(t)\) is cumulative adoptions, m is market potential, and \(a/b\) represent innovation/imitation coefficients . Modified versions accounting for price dynamics have shown improved accuracy .
2.2 Spatial Distribution Forecasting
Spatial forecasting identifies high-demand zones, guiding optimal charger placement. Two primary approaches exist:
- OD Matrix-Based Methods: Use origin-destination matrices to model traffic flow between nodes. OD matrices can be derived via surveys or traffic flow inversion, though the latter may yield non-unique solutions .
- Trip Chain-Based Methods: Simulate individual travel patterns (e.g., home-work-shopping routes) using probability distributions for trip timing, distance, and duration. 典型出行链 (Typical trip chains) include home-work-home and multi-stop itineraries .
Table 2: Spatial Forecasting Method Comparison
| Aspect | OD Matrix-Based | Trip Chain-Based |
|---|---|---|
| EV Type | Random travel (e.g., taxis) | Fixed routes (e.g., private cars) |
| Input Data | Traffic flow, historical OD | Trip chain probabilities, travel metrics |
| Spatial Resolution | Node-level | Zonal (residential/commercial) |
| Computational Cost | High (dynamic traffic modeling) | Moderate (stochastic simulation) |
3. Charging Infrastructure Planning
3.1 Influencing Factors
Planning must balance multiple objectives and constraints:
- Objectives: Minimize infrastructure costs, user travel/waiting time, and grid losses .
- Constraints: Charging capacity, user satisfaction (e.g., maximum waiting time), existing infrastructure, and grid safety (e.g., voltage limits) .
3.2 Site Selection Models
3.2.1 p-median Model
Minimizes the total weighted distance from demand points to p facilities:\(\min \sum_{i \in N} \sum_{j \in X} h_i d_{ij} y_{ij}\) Subject to:\(\sum_{j \in X} x_j = p, \quad \sum_{j \in X} y_{ij} = 1, \quad y_{ij} \leq x_j, \quad x_j, y_{ij} \in \{0,1\}\) Here, \(h_i\) is demand at node i, \(d_{ij}\) is distance to candidate site j, and \(x_j/y_{ij}\) are binary variables for facility location and demand assignment . This model is widely used for its focus on overall system efficiency .
3.2.2 p-center Model
Ensures the maximum distance from any demand point to a facility is minimized, suitable for emergency planning:\(\min d \quad \text{s.t.} \quad d \geq d_{ij} y_{ij} \quad \forall i,j\) However, it prioritizes worst-case scenarios over overall efficiency, limiting practical use .
3.2.3 Coverage Models
- Set Covering: Minimizes facilities needed to cover all demand .
- Maximal Covering: Maximizes demand covered within budget constraints .
Table 3: Site Selection Model Applications
| Model | Objective | Key Constraints | Solving Techniques |
|---|---|---|---|
| p-median | Minimize total travel cost | Facility count p | Particle swarm optimization |
| p-center | Minimize maximum travel distance | Facility count p | Integer programming |
| Set Covering | Minimize facilities for full coverage | Demand coverage threshold | Genetic algorithms |
| Maximal Covering | Maximize demand covered within budget | Budget limits | Simulated annealing |
3.3 Capacity Planning
Capacity planning determines charger numbers and power ratings. Key approaches include:
- Queueing Theory: Models waiting times using \(M/M/c\) queues to balance user delays and infrastructure costs .
- Peak Demand Matching: Sizing capacity to meet maximum observed or predicted demand .
- Renewable Integration: Co-optimizing capacity with solar/wind energy storage to reduce grid dependency .
4. Charging Scheduling Strategies
4.1 Dispatch Potential Forecasting
EVs can act as distributed energy resources, with schedulable potential depending on battery state-of-charge (SOC), user behavior, and grid signals. Clustering and machine learning (e.g., LSTM for SOC prediction, random forests for user participation) are critical for aggregating fleet-level flexibility , .
4.2 Peak Shaving and Valley Filling
Strategies aim to flatten grid load by shifting charging from peak to off-peak hours:
- Time-of-Use (TOU) Pricing: Incentivizes charging during low-price valleys. For example, a dual-order valley strategy split EVs into two groups: those charging at valley start and others at valley end, reducing peak loads by 15-20% in simulations .
- Stochastic Control: Using probability distributions of charging start times to smooth load profiles .
4.3 Frequency and Voltage Regulation
EVs can rapidly adjust charging power to support grid stability:
- Frequency Regulation: Sliding mode PI controllers have been used to coordinate EV fleets for frequency support, reducing system costs by 10-15% .
- Voltage Control: Reactive power adjustment from EV chargers helps maintain voltage levels, often combined with static var compensators .
4.4 Renewable Energy Integration
EVs can absorb surplus solar/wind power, reducing curtailment:
- Real-Time Dispatch: Dynamic pricing linked to renewable generation forecasts guides charging when renewables are abundant .
- Hierarchical Optimization: A two-layer model (top: load smoothing; bottom: user cost minimization) improved wind energy utilization by 25% in case studies .
4.5 Cost Reduction Strategies
- Direct Control: Aggregators optimize charging schedules for fleets, compensating users for flexibility .
- Price Signals: Dynamic TOU pricing that reflects grid conditions can reduce user costs by 10-20% while minimizing grid impact .
5. Global Demonstration Projects
5.1 China’s Pilot Programs
- Shanghai Demand Response (2019): Offered peak shaving (¥0.5/kWh) and valley filling (¥0.3/kWh) incentives. Office chargers showed 75% peak response rates, while private chargers had 5.3% valley response rates, highlighting the need for targeted incentives .
- Baoding V2G Project: 50 bidirectional chargers allowed EVs to sell power back to the grid at ¥0.6/kWh (charging) and ¥0.9/kWh (discharging), with 车主 (vehicle owners) earning up to ¥2,000/month .
- Jiangsu Grid Integration: A 2,000-kW V2G station reduced peak loads by 2.2 MWh annually, integrating with virtual power plants for grid stability .
5.2 International Projects
- Denmark’s Parker Project (2016-2018): 50 EVs provided frequency regulation services via CHAdeMO V2G chargers, demonstrating technical feasibility but facing economic challenges due to volatile market prices .
- Switzerland’s SunnYparc (2022-2025): A 1 GWh solar plant powers 250 chargers, with 50 V2G units storing excess solar energy for peak discharge. Private microgrid regulations enable market-driven pricing .
Table 4: Demonstration Project Summary
| Project | Location | Key Technology | Objectives | Outcomes |
|---|---|---|---|---|
| Shanghai DR | China | TOU pricing | Peak shaving/valley filling | 75% peak response in commercial chargers |
| Baoding V2G | China | Vehicle-to-grid | Grid support, user revenue | ¥2,000/month per EV owner |
| Jiangsu VPP | China | Virtual power plant | Load balancing, renewable integration | 2.2 MWh annual peak reduction |
| Parker Project | Denmark | V2G frequency regulation | Technical validation of V2G | 100% frequency response but unstable profits |
| SunnYparc | Switzerland | Solar+V2G microgrid | Renewable integration, market pricing | 1 GWh solar capacity, 50 V2G chargers |
6. Challenges and Future Directions
6.1 Modeling EV Charging Behavior
Current models underrepresent real-world complexities like user preferences and trip variability. Privacy-preserving data sharing via blockchain could enhance behavioral modeling while protecting user data .
6.2 Long-Term Infrastructure-EV Adoption Dynamics
The mutual influence between charging infrastructure and EV adoption is underexplored. Extending the Bass model to include infrastructure quality as an imitation factor could improve long-term forecasts .
6.3 Grid-Infrastructure Co-Planning
Multi-stage stochastic optimization models are needed to address uncertainty in EV growth and renewable energy deployment. Co-planning with grid upgrades (e.g., smart transformers) can prevent bottlenecks .
6.4 Incentive Design for User Participation
Dynamic pricing and cooperative game theory can align user, operator, and grid interests. For example, profit-sharing mechanisms based on Shapley values have shown promise in fair distribution of V2G benefits .
6.5 Scalable V2G Architectures
Standardizing V2G protocols (e.g., ISO 15118) and developing hierarchical control systems will enable millions of EVs to participate in grid services without overwhelming central controllers .
7. Conclusion
EV charging infrastructure planning and operation require a multidisciplinary approach, integrating demand forecasting, optimization models, and smart grid technologies. While significant progress has been made in modeling and control, challenges remain in behavioral prediction, long-term planning, and user engagement. Future research must prioritize scalable, collaborative solutions that leverage EVs as flexible energy assets, ensuring a sustainable and resilient transportation-energy system.