Research On Electricity Price Forecasting Of Electricity Market Based On Deep Learning

With the power industry moving from monopoly to competitive,electricity market has become an essential field affecting the development of the energy system gradually.Price changes in the electricity market not only brought challenges to the supply side of the power generator,but also has an impact on the flow of funds between consumers and producers,which makes accurate electricity price forecasting become particularly important.Due to the characteristic of the randomness,volatility,and obvious peak-to-valley differences in electricity prices,it is more difficult for electricity market participants to predict future electricity prices,which brings greater risks to electricity trading.By analyzing three years of real-time electricity price database in the US PJM power market,to determine different electricity classification standards based on deep learning methods.Insufficiency of MATLAB data fitting for electricity price prediction,the two deep neural network models were established,which based on Long Short Term Memory(LSTM)and MLP-LSTM based on the combination of Multi-layer Perceptron(MLP)and LSTM to carry out changes in electricity market electricity prices prediction.The main work in this thesis is as follows:Firstly,analyzing the influencing factors and classification criteria of electricity consumption and price determination,and the real-time electricity price database of the PJM electricity market were analyzed.Using the principle of least squares fitting algorithm based on MATLAB,establishing the fitting curve and prediction model of real-time electricity prices in different consumption areas.Secondly,the original data of the real-time electricity price data of the PJM power market from 2017 to 2020 is subjected to the minimum-maximum normalization process.While maintaining the correlation and analysis accuracy of the original data,the sampled data are sampled at equal intervals of 5min,10min and 20min respectively,this method has reduced the time density of traversing the database.At different time scales,the data loss function after sampling and the loss function of the original data were calculated and compared.Thirdly,utilizing three advantages of LSTM networks,including contextual information correlation,extracting the characteristics of changing trends,and introducing a controlled self-loop that can track information over a long time,proposing a model of electricity price prediction based on LSTM network to realize short-term and medium-term forecast of regional electricity prices.Finally,the normalized electricity price data was normalized,a new MLP-LSTM deep network electricity price prediction model that MLP prediction mechanism combined with the LSTM deep structure was established.The window period of the model uses the day,tenth,month,etc.It uses a MLP to perform one-step or short-term prediction to extract short-term or local change characteristics of the signal.Through the LSTM depth model,the short-term or local features are synthesized in time series and the long short term memory of the signal characteristics to realize the sliding prediction of the electricity price time series signal.Under different time scale windows,it reflects the seasonal and periodic characteristics of electricity price fluctuations.On the basis of the above research that based on the real-time electricity price data of the PJM power market,obtained good experimental results that through the LSTM model and the MLP-LSTM model for prediction.Comparing with the MATLAB data fitting results,the model proposed in research has improved the prediction accuracy greatly.