Research On Short-term Power Load Forecasting Based On Optimized CNN-LSTM Combination Model

In recent years,with the integration of a large number of electric vehicles and distributed power sources into the power grid,the complexity and uncertainty of load increase greatly,which has brought challenges to the safe and stable operation of the power system.Accurate short-term load forecasting is one of the key technologies to deal with the above problems,and it can also provide the necessary basis for power generation enterprises to reduce the cost of power generation enterprises.Based on the characteristics of Short-Term power load,this paper deeply analyzes the main factors that influence short-term power load prediction,and builds Convolutional Neural Networks(CNN)and Long short-term Memory neural networks(LSTM)from deep learning theories.combined with short-term power load optimization combination forecast model.The main work of this paper is as follows:1.Firstly,the characteristics of power load and the classification of power load prediction are briefly analyzed,and the potential rules of power load are studied to lay a theoretical foundation for short-term power load prediction;Then analyze the main factors that affect the fluctuation of power load and the causes of prediction error;In order to improve the accuracy of short-term power load prediction,the collected power load data are preprocessed.The preprocessing methods mainly include the removal of abnormal data,the supplement of missing data,and the normalization of data.Then,the factors with strong correlation with short-term power load are selected as the input features of the prediction model.Finally,the evaluation index of short-term load forecasting error and the basic flow of short-term load forecasting are introduced.2.The long-term and short-term memory neural network,support vector machine,extreme learning machine and other models are selected as the basic comparison models of this paper.On the basis of the combined prediction model of CNN neural network and LSTM neural network,in order to further improve the prediction accuracy of the CNN-LSTM combined model,the Attention Mechanism is introduced to establish the CNN-LSTM-attention combined model.Finally,two sets of experiments show that the prediction accuracy of the combined model is greatly improved compared with the CNN-LSTM model.3.In order to overcome the limitations and subjectivity of setting neural network parameters according to experience,better play the performance of neural network and further improve the prediction accuracy,Sparrow Search Algorithm is introduced in this paper.SSA)to optimize the parameters of the CNN-LSTM-Attention combination model integrating attention mechanism.Because the CNN-LSTM-Attention combination model integrating attention mechanism has a large number of parameters and difficulty in optimization,and considering that the traditional sparrow search algorithm is easy to fall into local optimization although it has strong adaptability and search ability,In this paper,chaotic initialization,multi-population mechanism and cross operation are used to optimize the traditional sparrow search algorithm,so as to improve the performance of the sparrow search algorithm.Then,the Improved Sparrow Search Algorithm was used to optimize the parameters of the CNN-LSTM-Attention combination model.Finally,ISSA-CNN-LSTM-Attention short-term power load prediction model was established.In order to verify the validity of the model proposed in this paper,the real load data collected from a place in Hunan were used for verification,and several groups of comparative experiments were designed.By analyzing the evaluation indexes of short-term power load prediction,Compared with SSA-CNN-LSTM-Attention model and CNNLSTM-Attention model,ISSA-CNN-LSTM-Attention model has the smallest error and the highest prediction accuracy.Experiments show that the ISSA-CNN-LSTM-Attention model proposed in this paper achieves the best prediction effect,which provides a new idea for short-term power load prediction.