Short Term Electricity Price Forecasting Based On Deep Learning In Electricity Market

The power industry continues to move from monopoly to competition,and price changes in the electricity market have an adverse impact on utility revenues and user costs.The change in electricity price has both brought challenges to the supply side of the power generation and also affected the use of the user side.As a lever for market regulation,accurate electricity price forecasting is particularly important.Under the premise of ensuring power reliability,high efficiency and safe supply,it is necessary to ensure that the power sales company accurately grasps the market orientation and provides competitive prices and services.The current electricity price forecasting method has a low utilization rate of its periodic variation law and a short forecasting step length,which makes the electricity price forecast have large errors.Through the data in the US PJM electricity market electricity price database,analyze the impact of the establishment of electricity sales companies under the electricity market reform on electricity prices,and learn the relevant theories of deep learning.A two-way LSTM model based on ELU activation function is proposed to predict the short-term electricity price change on the supply side of the electricity market.Mainly for the shortcomings and shortcomings of the circulating neural network,by analyzing the factors affecting the price of electricity,using the LSTM model,the E-BLSTM model is designed and analyzed after the optimization and improvement of the activation function.The accuracy of the model is proved when the finite iteration reaches convergence.The specific work is as follows:1)For the problem that the electricity price data is affected by many factors,the fuzzy similarity principle is used to preprocess the data,and the neural network algorithm is introduced into the electricity price prediction model.According to the correlation between time series,the sample data is learned and trained to reduce the experimental error.2)According to the gradient disappearance problem in the back propagation calculation process,according to the sensitivity of the electricity market supply side to the electricity price,the linear and nonlinear characteristics of the electricity price are captured,and the E-BLSTM model for the supply side electricity price forecast is designed.Using LSTM to maintain long-term memory characteristics,improve sigmoid function and tanh function,add three types of valves to LSTM model,introduce ELU activation function into bidirectional LSTM electricity price prediction model,improve step size and solve gradient disappearance problem.Usingre the optimized ADAM gradient descent algorithm,the weight of the neural network is iteratively updated according to the training data,and the optimal loss function is selected to improve the accuracy of the electricity price prediction.3)Comparing the designed E-BLSTM model with the common LSTM model,in order to ensure the accuracy of the experiment,and comparing with the ARIMA model and the ARMA model,the experiment proves that the algorithm can converge to a lower loss rate and can supply the electricity market.Accurate prediction of the electricity price with large side fluctuations proves the validity and convergence of the model.