Study Of The Load Forecasting Of The Liaocheng Grid Based On The Artificial Neural Network

The load forecasting plays an important role in the planning,generator schedule,operation schedule and so on.The reserve capacity of the grid depends on the accuracy of load forecasting.The reserve capacity of the grid can be reduced by the load forecasting with high accuracy,thus the economy and safety of the power system can be enhanced greatly.The operation of the power system has changed greatly in recent years with the development of the renewable energy,electric vehicle,high voltage DC and so on,so that the uncertainty of the power system and the difficulty of load forecasting continue to increase greatly,which has brought higher challenges to the load forecasting of the power.Based on the load of Liaocheng power grid,this thesis carries out in-depth research on the artificial neural network.The main contents are listed as below:(1)The background and research status of the load forecasting is introduced.The features and impact factors of the load forecasting are analyzed according to the characteristics of load.The model and learn methods of neural network are given.The basic principle of the error back-propagation neural network is studied.(2)The features of load in Liaocheng grid are introduced.The historical load data of some lines of Liaocheng grid is obtained so that the features,tendency and correlation of the load in different lines are analyzed.The correlation theory is applied to solve the problem.The equations of correlation index are induced.The load forecasting of Liaocheng grid under different weather is performed based on the artificial neural network.The analysis to the results is done.(3)The research of load forecasting of Liaocheng grid based on the neural network with multi scenarios is performed.The focus of this part is the influence of the number of hidden layer on the accuracy of the load forecasting.The results of the load forecasting with different hidden layers are obtained.The suitable hidden layer is obtained according to the results.The impact of the parameters of artificial neural network on the load forecasting is studied.The load forecasting based on the artificial neural network to the load of different lines is done.The analysis to the results is done.(4)The research on the load forecasting based on the deep learning.The impact factors of load are analyzed.The load forecasting based on the similar day and deep learning is proposed as to the shortage of time series model of the traditional neural network.The load of the similarity day is selected as the input data while considering the day type,season,weather and load growth.The RNN,which is with the self-circulation feedback and all collection,is adopted to handle the time series of load.The hyperbolic tangent function is set as the activation function.The back propagation through time is adopted to train the RNN.Thus the gradient of the whole loss is obtained.The application and the superior of the deep learning on the load forecasting are studied.The comparison of the load forecasting results between the deep learning and traditional learning algorithm is done.And the effectiveness of the deep learning is verified.Furthermore,the stability and the efficiency of the deep learning on the load forecasting are studied.The result is analyzed in detail.