Research On Power Load Forecasting Based On Improved Long Short Term Memory Network

Power load forecasting is used to predict the trend of power load change in a certain region in the future by mining the past power load change rules.Power load forecasting plays a great role in the formulation of power grid dispatching plan,power plant generation planning and the economic and stable operation of the power system.Power load has the characteristics of volatility and randomness,and there are many factors that affect the change of power load.So how to use these factors to accurately model and predict power load is the most important problem at present.The complex network structure is composed of Convolutional Neural Network(CNN)and Deep Bidirectional Long Short Term Memory Network(DBiLSTM)to establish the power load forecasting model.The main research contents are as follows:(1)A combined model of convolutional neural network and deep bidirectional long short-term memory network for power load prediction is proposed to address the issues of poor prediction accuracy and poor performance in feature extraction using a simple single algorithm.Using a CNN network to extract effective feature vectors from historical load sequences as inputs to the DBiLSTM network,and modeling the dynamic changes of the proposed time series features.Using publicly available power load data from a certain region and establishing it as a simulation dataset for experimental simulation,the simulation shows that using CNN for feature extraction of time series has better fitting performance.The results of this model were compared with the prediction models of traditional classical BP network,LSTM network,CNN-LSTM network,and BiLSTM network,indicating that the proposed CNN-DBiLSTM network has good prediction accuracy.(2)Aiming at the problem that the hyperparameter of deep neural network will affect the performance of the model,while the traditional hyperparameter search method needs a lot of computing resources,a hunter prey optimization algorithm(HPO)with more efficient search performance is proposed to optimize the LSTM hyperparameter to find the optimal solution.The experimental results show that the best combination of model hyperparameter can be found more efficiently after hyperparameter optimization of the model,and the prediction accuracy of time series can be improved.(3)Aiming at the problem that single-step forecasting can not meet the requirement of forecasting the trend of power load in the future in practical engineering,a multi-step forecasting method for power load is proposed.In view of the problem that the model ignores the correlation between the series in multi-step prediction and the prediction accuracy is not high due to the accumulated error,the iterative multi-step prediction is adopted on the basis of combining the spatial characteristics of multiple points,and the spatial relationship between different observation points is fully considered.The experimental results show that the power load iterative multi-step forecasting model considering the multi-point spatial characteristics greatly improves the forecasting accuracy,and is of great significance for forecasting the trend of the future power load in the region.