Residential Power Load Short-term Forecasting Method Based On Cyclic Neural Network Algorithm

With the reform of the electric power market and the advancement of various works,the electric power production and dispatching depends very much on the forecast result of the electric power load.Load forecasting refers to the prediction of the specific value of electricity consumption,which can not only guarantee the normal operation of power system,but also greatly reduce the consumption of resources.With the replacement of all kinds of power load forecasting models,the prediction model based on neural network has been widely recognized.Firstly,this paper summarizes the importance of power load forecasting in practical application,and explores the research status of power load forecasting at home and abroad.Because the power load data has the characteristics of time series,this paper focuses on the research of the cyclic neural network model,which is applied to the power load prediction,and introduces the related algorithms.The randomness of climate,date and other characteristics also increases the complexity of short-term load forecasting.In this paper,based on the obtained data of residential electricity load in a certain place,the load data are preprocessed and the noise error is reduced on the basis of full consideration of the influencing factors such as temperature,humidity,wind speed and holidays,so as to prepare the data for the forecasting work.Secondly,RNN load prediction model and LSTM load prediction model will be built respectively,and the trained model will be used to forecast the 24 hours load on July 1,2019,and the results will be verified and analyzed.Focus is on the general circulation neural network input sequence,the influence degree of the input sequence to load consideration is not enough,and characteristics of neural network based on Attention mechanism(Attention)is the key for the input sequence information into more Attention,thus overcome the traditional recursive neural network as the input sequence of growth and the problem that the performance degradation,And made great breakthroughs in natural language processing.In order to make the prediction model mine the law of load data better,this paper will introduce the Attention mechanism to further learn the output data of LSTM,give different weights to the input sequence,and put forward an improved self-attention-LSTM model prediction method based on the self-attention mechanism.Validation analysis was performed again.Finally,the experimental results of RNN load prediction model,LSTM load prediction model and self-attention-LSTM load prediction model with self-attention mechanism are comprehensively analyzed.Are introduced by comparison,found that the concentration mechanism of circulation neural network load forecasting model in prediction accuracy and so on various aspects are better than normal neural network load forecasting model,and mean absolute percentage error evaluation indexes have very obvious drop,reflects the good mapping ability,in the actual power load forecasting scenario also have good application prospects.