Demand Response Based Zonal Load Forecasting For Electric Vehicle Charging Stations

In recent years,the large-scale access of new energy vehicles to the distribution network has affected the safe operation of the traditional distribution network,while electric vehicles,as highly flexible mobile energy storage units,and the demand response of electric vehicle users under the influence of various incentives in the electricity market,will change the load characteristics and affect load forecasting.To address the problem that traditional forecasting methods do not consider specific scenarios,this paper divides the area where EV charging posts are located into residential areas and public areas based on demand-side incentives and EV charging behaviour,and proposes a zonal forecasting method for charging loads that takes into account demand response.(1)In order to research the charging load characteristics in the context of electric vehicle resources incorporated into the electric demand response,a pilot case in Shanghai was selected to analyse charging behaviour and demand response participation,and the results show that the degree of flexibility of electric vehicle charging varies from region to region.Based on the pilot results,two real datasets from different charging regions,CAISO and Elaad NL,were selected and statistical methods such as autocorrelation plots and Pearson correlation coefficients were used to explore the relationship between charging load and influencing factors in the two datasets.Finally,the data is pre-processed to obtain a new feature called the similarity of predicted day,by fusing the relevant features.(2)A Bi-LSTM multi-step fusion model based on real-time tariff decomposition is proposed for residential area customers with high participation in price-based demand response.Considering the impact of real-time electricity prices on charging loads in residential areas,the wavelet transform is used to decompose real-time electricity prices and extract the detailed features.To solve the problem of long training time associated with wavelet decomposition,K-means and decision trees are used to extract similarity days to avoid the possible impact of large amounts of invalid data on the training phase.Considering the influence of future information on the load,a backward transfer mechanism is added to the traditional LSTM,thus enabling the Bi-LSTM model to consider both past and future influences.To address the cumulative error problem of a single Bi-LSTM model,a multi-step fusion model based on Bi-LSTM is established by fusing recursive multi-step forecasting and multi-output forecasting,combining the accuracy of rolling forecasts and the wholeness of multi-output forecasting.(3)A combined CNN-AM-Bi LSTM prediction model based on response feature decomposition is proposed for the characteristics of high participation of user demand response and high volatility of charging load in public areas.According to the predictability of the demand response signal and the independence of the seasonal signal,wavelet decomposition and other methods are used to decompose the charging load in the public area at different levels to form the seasonal base load and the demand response dominant load.The seasonal base load is analysed for smoothness and ARMA is selected for prediction based on the analysis results.To address the problem that the demand response dominant load has many influencing factors and is difficult to obtain,CNN is used to extract deep features and introduce an attention mechanism to make the Bi-LSTM model focus more on the features most relevant to the load information,thus improving the prediction effect.Finally,the two parts of the prediction results are superimposed to obtain the public area EV charging load prediction results.