| Short-term power load forecasting is of great significance to economic dispatch,safety inspection,and the stability of social development.The focus of research on short-term power load forecasting methods is how to improve the forecasting accuracy.The improvement of load forecasting accuracy can provide the related power departments with a more reliable basis for formulating production and development plans,and also provide the power system with a reliable basis for economic operation.Therefore,this paper proposes a LSTM based on seasonal index and Adam optimization forecasting method,which aims to improve the accuracy of short-term power load forecasting.The seasonal index can process time series data with seasonality or periodicity to reduce the influence of seasonality or periodicity of the data.Long Short-Term Memory Network(LSTM)has excellent training results on time series data,and it can process data with long-term trends well.After analysis,the power load data is a time series with periodicity and long-term trends.Therefore,considering the advantages of both,this paper proposes LSTM based on seasonal index and Adam optimization which is a new forecasting method.The idea of this method is to calculate the seasonal index from the historical load data,then divide the load data by the seasonal index to obtain the revised data,and then use the LSTM model optimized by Adam to make predictions,finally multiply the forecasting results by the seasonal index to restore.The Adam optimization algorithm in this method is one of the most popular optimization algorithms at present.It performs excellent in the optimization of LSTM models.This paper also shows that the Adam optimization algorithm is more suitable for LSTM models than other popular optimization algorithms through experiments.Taking the power load data of summer weekdays and rest days in a certain area in southern China as an example,modeling and forecasting are implemented through Python and Keras neural network framework.In order to illustrate the forecasting effects of this method clearly,this paper also establishes a LSTM model based on Adam optimization without adding the seasonal index to forecast the power load.And then compare the forecasting results of the two forecasting methods in terms of the mean relative error,the accuracy of load forecasting,and the error rate of load forecasting.The experimental results show that the LSTM method based on the seasonal index and Adam optimization can improve the forecasting effect,whether it is the forecast of the power load on weekdays or on the rest days.For weekdays,the mean relative error of the forecasting results of the method has been reduced by 0.08%,and the accuracy of load forecasting results has been increased by 0.20%.For rest days,the mean relative error of the forecasting results of the method has been reduced by 0.27%,and the accuracy of load forecasting results has been increased by 0.67%.The improvement of the forecasting effect on rest days is more significant than that on weekdays.The validity of the LSTM based on seasonal index and Adam optimization forecasting method is verified. |
