Research on electric vehicle load forecasting considering regional special event characteristics

Abstract

Listen

Translate article

With the rise of electric vehicles and fast charging technology, electric vehicle load forecasting has become a concern for electric vehicle charging station planners and operators. Due to the non-stationary nature of traffic flow and the instability of the charging process, it is difficult to accurately predict the charging load of electric vehicles, especially in sudden major events. In this article, We proposes a high-precision EV charging load forecasting model based on mRMR and IPSO-LSTM, which can quickly respond to the epidemic (or similar emergencies). Firstly, the missing data in the original EV charging load data are supplemented, and the abnormal data are corrected. Based on this, a factor set is established, which included five epidemic factors, including new confirmed cases, the number of moderate risk areas, the number of high risk areas, epidemic situation and epidemic prevention policies of the city, and other factors such as temperature. Secondly, the processed load data and other data in the influencing factor set are normalized, and the typical characteristic curve is established for personalized processing of the relevant data of epidemic factors, so as to improve the sensitivity of load response to epidemic changes and the ability to capture special data (peak and valley values and turning points of load). Then the maximum relevant minimum redundancy (mRMR) is used to select the optimal feature set from the set of influencing factors. Then, the processed load data and its corresponding optimal selection are input into the IPSO-LSTM model to obtain the final prediction result. Finally, taking the relevant data of EV charging load in a city in China from November 2021 to April 2022 (the city experienced two local epidemics in December 2021 and March 2022 respectively) as an example, the model is evaluated and compared with other models under the forecast period of 1 h. Meanwhile, the performance of the model under different foresight periods (2 h, 4 h, 6 h) is compared and analyzed. The results show that the model has good stability and representativeness, and can be used for EV charging load prediction under the COVID-19 pandemic.