Research On Electric Vehicle Load Forecasting Based On Deep Learning

When electric vehicles of large quantity are connected to the power grid in the future,the spatial-temporal randomness of charging and discharging will have a great impact on the network loss of the distribution network,power quality,reliability and stability,which will bring huge impact to the power grid dispatching and challenge.This paper proposes an electric vehicle load forecasting model based on deep learning,which provides decision-making basis for power grid intelligent dispatching.Aiming at the problems of low accuracy and idealized model of electric vehicle load forecasting,based on the existing forecasting methods and probabilistic interval prediction,combined with the characteristics of deep learning,such as strong data fitting ability and ability to discover regular pattern of time series.respectively,Two probabilistic prediction methods based on LSTM neural network quantile regression and dilated causal convolution neural network quantile regression.The model uses the Adam stochastic gradient descent method to estimate the parameters of each layer in the network structure to obtain the optimal parameters.Finally,the rolling prediction results in the probability density function of the electric vehicle load at each time point in the next 96 time points.Compared with the simple point prediction result,the interval prediction can obtain more information and fault tolerance,at the same time,it can also obtain an effective prediction of the complete probability distribution of electric vehicle load in a certain period of time in the future.According to the actual data forecast of electric vehicle load in China,the proposed probability density prediction method can not only obtain more accurate point prediction results,but also obtain the result of predicted electric vehicle load of complete probability density function,which is more than the neural network.Compare with neural network their regression accuracy is more accurate,and the range of prediction intervals at the same confidence level is smaller.Aiming at the strong spatial-temporal randomness of electric vehicle charging load,which increases the difficulty of power grid control and the influence of power quality,a spatial-temporal dynamic load forecasting model-two-dimensional dilated causal convolution neural network is proposed.First,a hole factor is added to the temporal dimension of the three-dimensional convolutional convolution kernel to form a two-dimensional dilated convolution layer,so that the model can learn the spatial dimension information.Then through the stacking of such layers to form the entire network,to ensure that the network can accept long-term historical input,so that the model can learn the time dimension information.Experimental simulations show the effectiveness of the proposed prediction model.The research in this paper is not only to solve the problem of forecasting the information collection and operation status of some electric grids containing electric vehicles,but also to further expand the research into the orderly charging and discharging strategies of electric vehicles.