In recent years, the rapid growth of electric car adoption has posed significant challenges to charging infrastructure planning and operation. As a researcher in this field, I have observed that the integration of electric vehicles into power systems requires sophisticated approaches to demand prediction, facility siting, and scheduling strategies. This review aims to systematically analyze the key technologies involved in electric car charging infrastructure, focusing on demand forecasting, planning methodologies, and operational strategies. The rise of China EV markets has further emphasized the need for efficient solutions to handle the increasing charging demands while ensuring grid stability and user satisfaction.
Electric car charging demand prediction serves as the foundation for all subsequent decisions in infrastructure development. It can be categorized into time distribution and spatial distribution predictions, each with distinct applications. Time distribution prediction involves short-term and mid-to-long-term forecasts. Short-term prediction, covering hours to days, relies on data-driven methods like regression analysis and artificial intelligence techniques. For instance, support vector machines (SVM) and neural networks, such as long short-term memory (LSTM) and gated recurrent units (GRU), have shown high accuracy in forecasting charging loads for specific locations like residential areas or charging stations. However, these methods require extensive datasets, which are often scarce in the early stages of electric car adoption. Mid-to-long-term prediction, spanning months to years, typically uses models like the Bass model to estimate electric car penetration rates and associated charging needs. The Bass model describes the diffusion of new products, such as electric cars, and can be expressed as:
$$ \frac{dN(t)}{dt} = n(t) = a(m – N(t)) + b \frac{N(t)}{m} (m – N(t)) $$
where $N(t)$ is the cumulative number of electric cars at time $t$, $n(t)$ is the adoption rate, $m$ is the market potential, $a$ is the innovation coefficient, and $b$ is the imitation coefficient. This model helps in projecting the long-term evolution of charging demand, particularly in regions like China EV hubs, where policy incentives and infrastructure development play crucial roles.
Spatial distribution prediction, on the other hand, focuses on the geographical allocation of charging demand. Methods include origin-destination (OD) matrix-based approaches and travel chain-based simulations. The OD matrix captures the flow of electric cars between nodes in a transportation network, allowing for dynamic traffic simulations. In contrast, travel chain methods model fixed routes for electric cars, such as those used by private vehicles or buses, by analyzing sequences of activities (e.g., home, work, leisure). A comparison of these methods is summarized in Table 1.
| Method | Applicable Vehicle Types | Input Parameters | Traffic Modeling Detail | Spatial Output |
|---|---|---|---|---|
| OD Matrix-Based | Random demand vehicles (e.g., taxis) | Time-varying OD matrices | High (dynamic traffic flow) | Node-specific demand |
| Travel Chain-Based | Fixed-route vehicles (e.g., private cars) | Chain probabilities, travel times | Low (static routes) | Area-based demand (e.g., residential zones) |
The OD matrix-based approach involves simulating electric car movements based on historical data, as shown in the following process: start by extracting travel destinations from the OD matrix, select optimal paths considering traffic conditions, check if charging is needed based on state of charge (SOC) thresholds, and iterate until the simulation period ends. This method excels in capturing real-time variations but requires comprehensive data. Travel chain-based methods, meanwhile, simulate daily routines by sampling chain types and travel times, then assessing charging needs at each stop. For example, a typical travel chain might include sequences like home-work-home or home-other-home, with charging decisions triggered when SOC falls below a threshold. This approach is simpler but less adaptable to dynamic traffic changes.
Charging facility planning encompasses siting and capacity determination, which must balance multiple factors, including costs for operators, users, and the grid. Key objectives include minimizing construction and maintenance costs, reducing user travel and waiting times, and ensuring grid stability. Constraints often involve charging capacity limits, user satisfaction metrics (e.g., maximum waiting time), existing infrastructure, and grid safety standards like voltage and line capacity. In China EV projects, these factors are critical due to high urbanization and grid integration challenges.
For siting, models such as the p-median, p-center, set covering, and maximum covering models are commonly applied. The p-median model aims to minimize the total weighted distance from demand points to facilities, making it ideal for optimizing overall social benefits. Its formulation can be represented as:
$$ \min \sum_{i \in I} \sum_{j \in J} h_i d_{ij} y_{ij} $$
subject to:
$$ \sum_{j \in J} x_j = p $$
$$ \sum_{j \in J} y_{ij} = 1 \quad \forall i \in I $$
$$ y_{ij} \leq x_j \quad \forall i \in I, j \in J $$
$$ x_j, y_{ij} \in \{0,1\} $$
where $I$ is the set of demand points, $J$ is the set of candidate sites, $h_i$ is the demand at point $i$, $d_{ij}$ is the distance, $y_{ij}$ indicates if demand $i$ is served by facility $j$, and $x_j$ denotes if a facility is built at $j$. This model has been adapted in studies to incorporate traffic and grid constraints, particularly in dense urban areas of China EV networks. In contrast, the p-center model minimizes the maximum distance to any facility, suitable for emergency scenarios, while set covering and maximum covering models focus on coverage under resource limits. Flow-based models, such as intercepting models, are better suited for highway charging stations, where the goal is to maximize the capture of travel flows. A summary of these siting models is provided in Table 2.
| Model | Primary Objective | Key Constraints | Typical Application |
|---|---|---|---|
| p-Median | Minimize total weighted distance | Facility count, user access | Urban areas, social cost optimization |
| p-Center | Minimize maximum distance | Worst-case service range | Emergency planning |
| Set Covering | Cover all demand with minimal facilities | Service radius, demand coverage | Rural or low-density regions |
| Maximum Covering | Maximize demand coverage with limited facilities | Budget, capacity limits | Resource-constrained environments |
| Flow-Based | Maximize intercepted travel flows | Path coverage, capacity | Highways and intercity routes |
Capacity planning involves determining the number and power of charging piles within stations, often using queuing theory to model waiting times. The objective is to minimize the sum of construction costs and user waiting costs, subject to constraints like maximum queue lengths or demand fulfillment. For instance, in a charging station, the capacity can be optimized by solving:
$$ \min C_{\text{con}} + C_{\text{wait}} $$
where $C_{\text{con}}$ is the construction cost and $C_{\text{wait}}$ is the waiting cost, derived from queuing models like M/M/c systems. With the integration of renewable energy, some studies propose hybrid systems where solar or wind power supplements charging stations, enhancing economic and environmental benefits. In China EV demonstrations, this approach has been tested to reduce grid dependency and promote sustainability.
Charging scheduling strategies leverage the flexibility of electric car loads to benefit the grid, operators, and users. These strategies can be categorized based on objectives: peak shaving and valley filling, frequency and voltage regulation, renewable energy integration, and cost reduction. Predicting the schedulable potential of electric car clusters is essential, as it involves estimating available charging times, power, and capacities based on driving patterns and user behavior. Methods include cluster analysis and machine learning techniques like random forests or LSTM networks to bound the schedulable capacity under uncertainties.
For peak shaving, strategies aim to flatten the load profile by shifting charging to off-peak hours. For example, time-of-use pricing encourages users to charge during low-demand periods, but without coordination, this can create new peaks. Advanced methods, such as dual time-segment charging or probabilistic start-time distributions, have been developed to distribute loads more evenly. The optimization model for minimizing load variance can be expressed as:
$$ \min \sum_{t=1}^{T} (L_t + P_t – \bar{L})^2 $$
where $L_t$ is the base load, $P_t$ is the charging power, and $\bar{L}$ is the average load. This reduces grid stress and improves efficiency, particularly in regions with high electric car penetration like China EV hotspots.
Frequency and voltage regulation utilize the rapid response of electric car charging power. For frequency regulation, controllers adjust charging rates in real-time to balance supply and demand, often using sliding mode or proportional-integral techniques. In voltage regulation, electric cars can inject or absorb reactive power, aiding in maintaining grid stability. The coordination with other devices, like capacitors, can be modeled as a combined optimization problem to ensure voltage limits are met while fulfilling charging demands.
Renewable energy integration focuses on aligning charging with intermittent sources like wind and solar. Strategies often involve two-stage optimization: day-ahead scheduling based on forecasts and real-time adjustments to handle uncertainties. For instance, a bilevel model might minimize wind curtailment in the upper level while optimizing charging costs in the lower level. This approach has been applied in pilot projects to enhance the utilization of renewables, supporting global sustainability goals.
Cost reduction strategies prioritize economic benefits for users and operators. Direct control methods involve centralized scheduling by aggregators, who offer incentives for charging flexibility. Alternatively, price-based indirect control allows users to self-schedule based on dynamic tariffs. Game theory models have been used to design incentive mechanisms that fairly distribute benefits among stakeholders. In China EV markets, such strategies are crucial for encouraging participation in grid services.
Demonstration applications worldwide highlight the practical implementation of these technologies. In Shanghai’s 2019 demand response pilot, electric car charging stations participated in peak shaving and valley filling, with response rates varying by charger type: private chargers showed low participation in valley filling but high potential due to numbers, while dedicated chargers and swap stations achieved high response rates. In Baoding’s V2G pilot, electric car owners earned rewards for discharging to the grid, demonstrating the viability of vehicle-to-grid technologies. Similarly, Jiangsu’s large-scale V2G center in China has enabled bidirectional power flow, contributing to grid stability. Internationally, Denmark’s Parker project tested V2G for frequency regulation, and Switzerland’s SunnYparc project explores price-based charging to integrate solar power. These cases underscore the progress and challenges in real-world electric car integration.
Despite advancements, several limitations persist in electric car charging research. First, modeling charging behavior remains challenging due to uncertainties in user preferences, travel patterns, and external factors like holidays or weather. Data privacy concerns also hinder the collection of detailed datasets needed for accurate predictions. Second, long-term interactions between charging infrastructure development and electric car adoption are underexplored; most studies focus on static planning without considering dynamic feedback effects. Third, coordinated planning between charging facilities and the power grid is often overlooked, leading to suboptimal solutions under uncertainty. Multi-stage stochastic optimization could address this by incorporating evolving demands and grid upgrades. Fourth, pricing mechanisms for orderly charging lack robust designs that balance user incentives and grid benefits. Cooperative game theory and empirical data from demonstrations, such as those in China EV projects, could inform better tariff structures. Lastly, standardized frameworks for electric car participation in grid services are needed to streamline operations across different charging scenarios, such as private, public, and dedicated stations.
In conclusion, the planning and operation of electric car charging infrastructure involve a multifaceted approach that integrates demand prediction, facility siting, and scheduling strategies. Key technologies like AI-based forecasting, optimization models, and V2G systems have shown promise in enhancing efficiency and grid integration. However, addressing the limitations in behavior modeling, long-term planning, and market design is essential for future advancements. As electric car adoption accelerates, particularly in regions like China EV markets, continued research and real-world testing will be critical to developing sustainable and resilient charging ecosystems. This review provides a foundation for understanding these complexities and guiding further innovations in the field.