Research On Short-term Power Load Forecasting Method Based On Neural Network Combination Model

Power load forecasting is the basis of smart grid construction,and the accuracy of its forecast is of great significance to the construction of smart grid.Short-term power load forecasting is an important reference value for grid dispatching.In order to solve the problem that the traditional load forecasting methods cannot make full use of the correlation of data and thus lead to low accuracy,a short-term power load forecasting model based on Double Attention mechanism(Double Attention)and Bi LSTM(Bi-directional Long and Short Term Memory Neural Network)is proposed.The short-term electric load forecasting model combining the optimized particle swarm algorithm and convolutional neural network(CNN)and bi-directional long and short-term memory neural network(Bi LSTM)is proposed for the parameter tuning problem.The main work and research results of this paper are as follows:(1)This paper analyzes the theoretical basis of power load forecasting and introduces in detail the change law of power load data and possible external influencing factors.In the power load forecasting,in addition to the historical load also the relationship between weather factors,holiday factors,electricity prices and other factors and power load is fully analyzed and utilized.The correlation analysis between these influencing factors and electricity load data is performed through the feature attention mechanism,and then the influencing factors with larger correlation coefficients are input into the model together with the electricity load data to improve the accuracy of the model.(2)By comparing and analyzing the accuracy of different neural network models in power load prediction,Long Short-Term Memory(LSTM)is more advantageous for analyzing and fitting data with long-term dependencies.In this paper,a short-term power load prediction model is trained using a dual attention mechanism and a two-way long and short-term memory neural network.The attention mechanism is used to analyze each influence factor and assign weights to each feature influence,and the factors with significant influence are selected as input together with the electricity load data and input to the model for training.Two datasets,the 2019 electricity load dataset from Belgium and the 2020 electricity load dataset from a region in China,are used for validation,proving that the model is more accurate in short-term electricity load forecasting(3)A CNN-Bi LSTM with optimized particle swarm for short-term electricity load forecasting model is proposed to address the problems of influence size of influence factors,model parameter tuning and low forecasting accuracy.Firstly,the internal features of the power load data are analyzed using CNN,and the factors with greater influence are retained using the maximum pooling method and input into the model together as inputs.In the parameter tuning process,the optimized PSO algorithm is chosen to be used,and finally the output of the prediction results is performed through the fully connected layer,which leads to more accurate prediction results.A combined model combining the particle swarm optimization algorithm with convolutional neural network and bidirectional long-and short-term memory network is used,and the advantages of the model are demonstrated using a 2019 dataset from a region in China.(4)To verify the scientificity and generalization of the model combination,the model effect is validated by multiple datasets.An ablation experiment is also specifically set up,and the experimental results of the ablation experiment also demonstrate the advantages of the combined model in short-term electricity load forecasting.