Research On Short-term Electricity Price Forecasting Algorithm For Smart Grid Based On Machine Learning

With the continuous deepening of the global electricity market reform,the pattern of the electricity market has gradually changed from a centralized purchase to a liberalized transaction model.As the most critical factor in the electricity market,electricity price is closely related to people’s lives.Especially in the market competition environment,electricity price forecasting is an inevitable task for electricity production companies and electricity demand customers.However,the electricity market is easily affected by factors such as geographic location,weather conditions,holiday activities,and current social events,making electricity prices,which have the characteristics of non-stationarity,randomness,seasonality,and high volatility.Therefore,how to establish an accurate and stable electricity price forecasting model to achieve a win-win situation for both parties in market transactions has become the focus of widespread attention worldwide.In recent years,artificial intelligence has continued to appear in the public’s field of vision,and the machine learning algorithms which involves having penetrated all aspects of people’s lives.In view of this,the majority of researchers have applied machine learning algorithms to the field of electricity price forecasting,and there have been many breakthroughs,and many classic forecasting models have also been proposed.Traditional electricity price forecasting models,such as physical models and statistical models,have a simple structure and a single algorithm,whose forecasting results are often difficult to meet the required expectations,the forecasting deviations are also particularly obvious.Relatively speaking,artificial intelligence models and hybrid models perform better in terms of prediction effects,but the existing proposed machine learning model still has problems such as low prediction accuracy,slow prediction rate,and difficulty in determining parameters when dealing with short-term electricity prices with relatively high complexity.Therefore,to obtain a more effective electricity price forecasting model,this paper proposes two hybrid models based on machine learning algorithms.Through multiple sets of simulation experiments on the electricity price data of the German and Australian electricity markets,the significant advantages of the model proposed in this paper are verified.Specifically,the main research work and innovations of this article are as follows:(1)The traditional single model cannot achieve satisfactory forecasting results,and also meet the forecasting needs of complex grids.Therefore,this paper is dedicated to exploring a novel hybrid electricity price forecasting model,and improving the performance of the three stages of the hybrid model.In the preprocessing stage of the time series,two novel decomposition algorithms are proposed to denoise the original electricity price series.Besides,based on the long and short term memory neural network,two extended models are proposed as the subject prediction model in the hybrid model.At the same time,the hyperparameter optimization algorithm is introduced to improve the prediction performance of the subject prediction model.Finally,the proposed hybrid model performs better in forecasting performance.(2)Based on the idea in(1),a novel hybrid model based on hybrid mode decomposition(HMD)algorithm,convolutional long short term memory network(CNNLSTM)and Elman neural network is proposed to forecast the electricity price.In the proposed model,HMD is used to deeply decompose the electricity price data into several subsequences,CNNLSTM and Elman are adopted to forecast the electricity price subsequences.Besides,the Bayesian optimization(BO)algorithm is introduced to optimize the parameters of CNNLSTM.Through three sets of simulation experiments on the electricity price sequence sampled every 1 hour interval in the German area,it fully verified that the HMD-CNNLSTM-Elman proposed in this paper performs best in forecasting performance.(3)Aiming at the low prediction rate and high prediction cost of the model in(2),a novel hybrid model based on improved variational mode decomposition(IVMD),Quantile Regression Bi-directional long short-term memory network(QRBi LSTM)and Coyote optimization algorithm(COA)is proposed herein.In the ensemble IVMDQRBi LSTM-COA model,IVMD is used to deeply decompose the electricity price data into several components,QRBi LSTM is adopted to forecast the components obtained by IVMD,meanwhile,COA is introduced to optimize the parameters of the model,which can significantly enhance forecasting performance.The effectiveness of the proposed model is intended to be demonstrated by using electricity price data for 30 minutes intervals from NSW and QLD in Australia.The final experimental results show the predominant stability and accuracy of the proposed IVMD-QRBi LSTM-COA model.