Volume 39: Energy Transitions toward Carbon Neutrality: Part II

EV Charging Load Forecasting Based on Model Driven and Data Driven Jiasen Tai, Fenglong Chen, Chunpeng Wang, Yijie Li, Heng Song



Based on the integration of dynamic traffic information, environmental temperature, real-time traffic flow, queuing theory, and other methods, a novel deep learning architecture for predicting Origin-Destination traffic flow in urban transportation systems has been developed to forecast the spatial and temporal distribution of electric vehicle charging loads.This method begins by analyzing the impact of various factors, such as urban traffic network data, daily patterns, and weather, on the driving patterns of electric vehicles. It employs a Graph Convolutional Recursive Neural Network algorithm to separately identify the starting and ending points of private cars and taxis.Next, it introduces impedance models for road segments and nodes, which take into account dynamic traffic information, intersection flow, and an air conditioning energy consumption model that considers environmental temperature and real-time vehicle speed. The method utilizes Graph Convolutional Network to extract spatial features of traffic nodes and their neighboring nodes. Time-related features are extracted using P-Prophet, creating a traffic intersection traffic flow prediction model.To optimize the minimum cost travel routes for electric vehicles, it improves the Floyd dynamic algorithm using a sparse graph optimization strategy, thereby simulating the driving behavior of electric vehicle users. Additionally, it uses K-means clustering to analyze potential charging preferences of electric vehicle users, providing insights into characteristic charging behaviors among typical urban electric vehicle users.

Keywords Electric vehicle, Dynamic traffic information, Charging load, Neural network, Path planning

Copyright ©
Energy Proceedings