Power Load Forecasting Based On Convolution Neural Network Considering Multi Factors

With the rapid development of energy technology,people pay more and more attention to workload prediction.Good load forecasting results can not only make the power department more reasonable,avoid resource waste,but also guide the start and stop of the unit,increase and decrease the output.With the intelligent grid,the accuracy of load forecasting is also required.Some factors such as meteorology,economy and date type will have a great impact on load forecasting.This thesis briefly discusses the situation and necessity of load forecasting,and analyzes the relevant classification of load forecasting and the theoretical methods of power load forecasting.Aiming at the problems of large error and large amount of calculation in power load forecasting,a power load forecasting model based on convolution neural network is established.Taking the load situation of a certain area as an example,the long-term load forecasting and short-term load forecasting are carried out.Compared with the BP neural network,the simulation results show that the convolution neural network fitting is better,but there is still room for improvement.Considering that the prediction results are affected by many factors,aiming at the problem of large error in the prediction results,Pearson analysis method is used to quantify,normalize and analyze the extracted 18 influencing factors one by one.Finally,10 most relevant influencing factors are extracted and the weight is allocated.In order to solve the problem that the convolution neural network method does not consider the influence of time series and is not ideal for the forecasting results,a hybrid model of convolution neural network and long-term memory network is proposed.The long-term load forecasting and short-term load forecasting are carried out with the same data as before.The simulation results with Python show that the hybrid model of convolution neural network and long-term memory network has significantly improved the effect of long-term load forecasting and short-term load forecasting compared with the BP neural network and convolution neural network,and the fitting degree is quite excellent,which shows that the long-term memory network has a good performance The time series feature extraction is very effective,and the improved method is more superior.