Research On Short-term Electric Load Forecasting Based On Machine Learning

Power load forecasting is the basis for ensuring the safe and stable operation of power system,realizing reasonable power dispatch and improving the operation efficiency of power grid system.Therefore,a more accurate power load forecasting method is needed.The limitations of traditional power load forecasting methods are highlighted by their low prediction accuracy and resource wastage,especially in the case of drastic power load fluctuations.However,the machine learning method has strong nonlinear fitting ability and can fully explore the intrinsic characteristics of load data,which is more effective in improving the accuracy of electric load forecasting.To address this situation,machine learning methods are applied to short-term electric load forecasting.The research contents are as follows.1.firstly,the original power load data sample set is pre-processed to complete the processing of abnormal data and the normalization of load data;then,through the analysis of load characteristics,it is found that the power load has certain periodicity and regularity,meanwhile,the power load is also affected by many external factors and shows certain randomness;finally,this paper fully takes into account the intrinsic change law of power load Finally,this paper determines the main influencing factors of electric load according to the Pearson correlation coefficient method,taking into full consideration the internal variation law and external influencing factors of electric load.2.Taking machine learning theory and shallow neural network prediction as an example,firstly,the concept and classification of machine learning are briefly introduced;then,three commonly used models of RF,CNN and LSTM are listed,and their basic principles and internal structures are described;finally,the prediction results of three shallow neural network models of BP,ELM and LSTM are compared and analyzed.After several experimental simulations,the prediction results show that the shallow neural network LSTM model has a better prediction effect,which provides theoretical guidance for the subsequent combined model prediction.3.In order to improve the prediction accuracy of the shallow neural network model,a combined model load prediction method based on PSO-LSSVM is constructed,and the parameters of the LSSVM model are automatically searched for using the PSO algorithm,and the prediction effects of the PSO-LSSVM model are compared and analyzed with those of the PSO-ELM and PSO-SVM models.After several experimental simulations,the prediction results show that the combined PSO-LSSVM model has high prediction accuracy and good convergence.4.In order to further improve the prediction ability of the model,a short-term electric load forecasting method based on CEEMD-GA-DBN model is constructed.First,CEEMD is used to decompose the original load data sample set sequence into multiple IMFs and one res to reduce the non-smoothness and complexity of the load sample set sequence;then,the GA-DBN model is used to forecast the multiple components obtained from the decomposition separately;finally,the results of different forecast components are superimposed to obtain the final electric load forecast value.The prediction effects of this model are compared and analyzed with those of each model of PSO-LSSVM,DBN,GA-DBN,EMD-GA-DBN,and EEMD-GA-DBN.After several experimental simulations,the prediction results show that the CEEMD-GA-DBN model has the best prediction effect and the prediction accuracy is as high as 96.88%.Fig.[59]table[18]Ref.[80]...