An Algorithm Study of Personalized Recommendation for Electric Vehicle Charging Piles Based on Deep Learning Framework

China is vigorously developing electric vehicles, and the penetration rate of new energy vehicles in China has exceeded 10%. The penetration rates in the United States, Europe, and other regions are also growing, leading to a high-speed increase in the demand for charging piles. Although the number of public charging piles has been increasing year by year, big data shows that the utilization rate of public charging piles is less than 15%. Therefore, how to guide users to avoid peak electricity consumption periods without compromising the convenience of charging for electric vehicle users is key to solving the problem. Finding an appropriate scheduling strategy is the core of solving this dilemma. For personalized recommendation algorithms for charging piles, we propose a Top-N recommendation algorithm for charging piles based on a neural collaborative filtering framework that combines multiple feature fusion methods. This algorithm uses concatenation and outer product operations to construct interaction relationships between users and items, respectively. Then, two learners are used to learn their interactive features, and finally, the learned prediction vectors are merged through concatenation. This approach not only fully excavates the potential information of embedded vectors but also has nonlinear feature fitting capabilities, thereby better improving the model's recommendation performance.