Smart Charging Guidance System Optimizes EV Travel and Charging Efficiency
In the rapidly evolving landscape of electric mobility, a groundbreaking study has introduced a novel approach to optimizing the charging experience for electric vehicle (EV) owners. The research, led by Yuan Xiaodong from the Electric Power Research Institute of State Grid Jiangsu Electric Power Co., Ltd., in collaboration with Gan Haiqing, Wang Mingshen, Teng Xinyuan, Ruan Wenjun, and Long Huan, presents an innovative active charging guidance model that leverages the Internet of Vehicles (IoV) to enhance both travel and charging efficiency. Published in the prestigious journal Automation of Electric Power Systems on April 10, 2024, this work addresses the growing challenge of managing the increasing number of EVs and their associated charging demands.
The core of the proposed model lies in its integration of advanced path planning algorithms and queuing theory, specifically tailored to the dynamic nature of urban traffic and the unpredictable patterns of EV charging. By incorporating real-time traffic data, including traffic light waiting times and avoiding unnecessary detours, the researchers have developed an improved A* path planning algorithm. This enhancement allows for the continuous updating of the road network’s spatiotemporal state matrix, ensuring that the recommended routes are always optimized for the current conditions. The result is a significant reduction in travel time, as demonstrated in a case study conducted in the central area of Nanjing, China.
One of the key innovations in this study is the use of a deep belief network (DBN) to predict the short-term arrival rates of EVs at charging stations. This predictive capability is crucial for managing the often unpredictable demand for charging services. By accurately forecasting the number of vehicles expected to arrive at a given station, the system can better allocate resources and minimize wait times. The DBN model was trained on historical data, including vehicle arrival patterns, time of day, weather conditions, environmental temperature, day type (weekday or holiday), and traffic status. The input data was processed through vectorization and normalization, and the model was designed with multiple restricted Boltzmann machines (RBMs) and a fully connected layer, allowing for both unsupervised pre-training and supervised fine-tuning.
The effectiveness of the DBN model was evaluated against other popular machine learning algorithms, such as random forest (RF), support vector machine (SVM), and decision tree (DT). The results showed that the DBN model outperformed these alternatives in terms of mean absolute error (MAE) and root mean square error (RMSE), indicating its superior ability to capture the complex relationships between the input features and the vehicle arrival rates. This high accuracy in prediction is essential for the subsequent steps in the active charging guidance process.
Once the arrival rates are predicted, the next step involves estimating the waiting time at the charging station. To achieve this, the researchers employed the M/G/k queuing model, which is particularly well-suited for scenarios where the service times are not exponentially distributed. In the context of EV charging, the service time corresponds to the duration of the charging process, which can vary significantly depending on the battery state, the type of charger, and the specific vehicle. The M/G/k model takes into account the number of charging points (k), the arrival rate (λ), and the distribution of service times, providing a more realistic representation of the queuing system.
To validate the M/G/k model, the team analyzed the charging duration data from five different charging stations in the central area of Nanjing. The data, collected over a period from 6:00 AM to 9:00 PM, was segmented into 20-minute intervals, resulting in 45 data points per day for each station. Using the Kolmogorov-Smirnov (K-S) test, the researchers confirmed that the arrival of vehicles at the charging stations follows a Poisson distribution, a common assumption in queuing theory. This finding supports the use of the M/G/k model, as it assumes that the inter-arrival times are exponentially distributed, which is a characteristic of the Poisson process.
The charging durations, however, were found to be more complex. Instead of following a simple exponential, normal, or gamma distribution, the data was best described by a mixture of Gaussian distributions. This mixture model, represented by the equation ( T{ch} = sum{i=1}^{N_G} a_i N(mu_i, sigmai^2) ), where ( T{ch} ) is the charging duration, ( N_G ) is the number of Gaussian components, ( a_i ) is the weight of the i-th component, and ( N(mu_i, sigma_i^2) ) is a standard Gaussian distribution with mean ( mu_i ) and variance ( sigma_i^2 ), provides a more accurate fit to the observed data. The optimal number of Gaussian components was determined using the Bayesian information criterion (BIC), with the model having the lowest BIC value being selected. This detailed characterization of the charging duration distribution is critical for the accurate prediction of waiting times.
With the arrival rates and service time distributions in place, the researchers used Monte Carlo (MC) sampling to simulate the queuing process at the charging stations. The simulation involved creating k queues, one for each charging point, and initializing them based on the current status of the chargers (e.g., remaining charging time). Future vehicle arrivals and their corresponding charging durations were then sampled from the predicted distributions, and the queuing process was simulated over the estimated travel time. The average waiting time, calculated after multiple iterations, was used as the predicted waiting time for the target charging station.
The active charging guidance model was tested in a comprehensive case study involving 900 EVs with charging needs, randomly generated within the central area of Nanjing. The performance of the proposed model was compared against a static path planning algorithm, which simply directs vehicles to the nearest charging station. The results were striking: the active guidance model reduced the total travel time by 18.49%, the total waiting time by 31.67%, and the overall charging time (the sum of travel and waiting times) by 27.27%. These improvements are not only significant but also highlight the practical benefits of the model in real-world scenarios.
Further analysis revealed that the active guidance model also led to a more balanced utilization of the charging infrastructure. In the static scenario, certain charging stations, particularly those in more accessible locations, became heavily congested, leading to long wait times and underutilization of other stations. In contrast, the active guidance model effectively distributed the charging demand across the network, reducing the peak load on any single station and improving the overall efficiency of the system. At 3:00 PM, for example, the average utilization rate of the charging stations increased from 69.50% to 89.47%, a substantial improvement that underscores the model’s ability to optimize resource allocation.
The success of the active charging guidance model is a testament to the power of integrating advanced computational techniques with real-time data. The improved A* algorithm, by accounting for dynamic traffic conditions and traffic light delays, ensures that the recommended routes are always the most efficient. The DBN model, with its ability to predict vehicle arrival rates, provides a robust foundation for managing the charging demand. And the M/G/k queuing model, enhanced by the use of a mixture of Gaussian distributions, offers a realistic and accurate prediction of waiting times.
The implications of this research extend beyond the immediate benefits to individual EV owners. By reducing travel and waiting times, the model can help to alleviate traffic congestion and improve the overall efficiency of the urban transportation system. It can also contribute to the stability and reliability of the power grid, as more predictable and evenly distributed charging patterns reduce the risk of overloading. Furthermore, the model’s ability to balance the load across the charging network can lead to cost savings for both operators and users, as it reduces the need for expensive infrastructure upgrades and minimizes the time spent waiting for a charging spot.
The study also highlights the importance of considering the user’s perspective in the design of smart mobility solutions. By focusing on minimizing the total time cost, which includes both travel and waiting times, the model aligns with the primary concern of EV owners: convenience and efficiency. This user-centric approach is a key factor in the model’s success and sets a precedent for future research in the field.
Despite the significant achievements, the researchers acknowledge that there are still areas for improvement. For instance, the current model does not fully account for the impact of pricing on the user’s choice of charging station. In the future, incorporating price factors and using deep reinforcement learning algorithms could further enhance the model’s performance. Additionally, the model could be extended to consider the optimal allocation of charging resources, balancing the needs of the system with the preferences of the users.
The work of Yuan Xiaodong, Gan Haiqing, Wang Mingshen, Teng Xinyuan, Ruan Wenjun, and Long Huan represents a significant step forward in the development of smart charging solutions. Their active charging guidance model, published in Automation of Electric Power Systems, demonstrates the potential of integrating advanced computational methods with real-time data to create more efficient and user-friendly EV charging experiences. As the adoption of electric vehicles continues to grow, such innovations will be crucial for ensuring that the transition to a sustainable transportation system is both smooth and effective.
The research was supported by the National Key R&D Program of China (No. 2021YFB2501600), highlighting the importance of government support in driving technological advancements. The study not only contributes to the academic literature but also has practical applications that can benefit a wide range of stakeholders, from EV owners to city planners and utility companies. As the world moves towards a more sustainable and connected future, the insights and methodologies presented in this paper will undoubtedly play a vital role in shaping the next generation of smart mobility solutions.
Yuan Xiaodong, Gan Haiqing, Wang Mingshen, Teng Xinyuan, Ruan Wenjun, Long Huan, Automation of Electric Power Systems, DOI: 10.7500/AEPS20230730001