Research On Personalized Recommendation Algorithm For Charging Piles Based On Deep Learning Framework

In recent years,electric vehicles are being vigorously developed in various countries,due to the increase of people's awareness of environmental protection.More and more people choose electric vehicles to travel.The corresponding charging infrastructure,that is,the number of public charging piles,is also increasing with each passing day.When a large number of electric vehicles are connected to charging piles in a disorderly manner during the peak period of residential electricity consumption,the local grid load will be overloaded.Existing companies have begun to guide the data platform to establish an Internet of things system for charging piles based on grid demand and time-of-use pricing of the power supply time of charging piles encourages users to avoid peak periods of electricity.But how to guide electric-vehicle-users to avoid peak periods of electricity without simultaneously weakening the utilization of charging piles has become an urgent problem.Therefore,designing a suitable and high-precision algorithm for the charging pile recommendation model is the key to solving this problem.The paper combines deep learning with traditional collaborative filtering methods,and uses feature fusion to improve existing algorithms.This algorithm effectively alleviates the problem of sparseness and hidden features of data that are difficult to capture,guides electric vehicle users to avoid peak electricity consumption,and improves the utilization rate of charging piles.The main research work of this paper is as follows:(1)Aiming at the incomplete discovery of the potential information of the embedding vector by the existing neural collaborative filtering methods,this paper proposes a top-N recommendation algorithm for charging piles based on a neural collaborative filtering framework combining multiple feature fusion methods.By using different feature fusion methods for the user and item embedding vectors,the information discovered by the different methods complement each other,so that the model can better find the interaction between the user and the item.The experimental results show that compared with the single-feature fusion method,the combination of multiple feature fusion methods can effectively improve the model's ability to discover the interaction between the user and the item,and greatly improve the performance of the recommendation system.(2)The neural network learning models used for most neural collaborative filtering methods are relatively simple and cannot fully learn complex high-order non-linear recessive features from user-item interaction data.This paper proposes a top-N recommendation algorithm for charging piles incorporating a neural collaborative filtering model of MLP-DRN(Multiple layer perceptron-Deep residual network).This method modifies the neural network learning model based on the neural collaborative filtering framework combining multiple feature fusion methods.The neural collaborative filtering model fused with MLP-DRN deepens the depth of the network.The experimental results show that this model greatly improves the overall fitting ability of the hidden features of users and items,thereby better improving the accuracy of Top-N charging pile recommendations.