Research On Short-term Power Load Forecasting Method Based On Improved DBN

Ensuring the balance of power supply and demand is one of the important conditions to ensure industrial production and social stability,so accurate prediction of power load is an inevitable requirement to achieve social stability and harmonious economic development.Traditional prediction methods cannot effectively overcome the problem of prediction accuracy reduction caused by random fluctuation of power grid.With the development of deep learning,a large number of deep learning technologies have been applied to short-term load prediction research and achieved ideal prediction effects.Therefore,the short-term power load forecasting method is studied based on the improved deep confidence network.The specific research process is as follows:Firstly,the problem that DBN prediction model could easily fall into local optimal solution was analyzed.Considering its good learning and generalization performance in the face of massive power load data,ELM algorithm was introduced to fix the weight and bias of DBN prediction model,and ELM-DBN short-term power load prediction model was proposed.The ELM-DBN short-term power load forecasting model is better than DBN short-term power load forecasting model for short-term power load forecasting research.Secondly,in order to reduce the nonlinear,non-stationary and random fluctuation of power load series and further improve the prediction ability of the model,a Fast ELM-DBN short-term power load prediction model was established by introducing the wavelet decomposition algorithm.Three kinds of residual correction network structures were constructed to modify the prediction results of Fast ELM-DBN short-term power load prediction model respectively.Considering that the power load is easily affected by the day type,the parameter transfer network structure is introduced to predict the ordinary day and Saturday and day respectively.Finally,in order to verify the superiority of the short-term power load prediction model based on the improved deep confidence network,Fast ELM-DBN short-term power load prediction model is used for comparative analysis.The simulation results show that the single linear perceptron residual correction network structure has the best prediction effect,and the prediction accuracy is improved by 1.70% compared with Fast ELM-DBN short-term power load prediction model.After migration network structure parameters are introduced,further the short-term power load forecasting model is optimized,the final accuracy can reach99.14%,significantly improve the prediction accuracy of short-term power load forecasting model,prove the validity of the proposed method,for short-term power load forecasting method research provides important reference basis.