Research On Power Load Prediction Algorithm Based On Deep Learning

Short-term Load Forecasting(STLF)refers to the prediction of the power Load value of the whole day or several days in the future.It is an important means to ensure the stable operation of the power grid and realize energy conservation and emission reduction.With the rapid development of future power grid technologies such as smart power grid and energy Internet,load prediction theory and technology have developed greatly.Based on the research results of many scholars at home and abroad,this paper tries to propose an accurate and effective short-term power load prediction algorithm.The main work of this paper is as follows:1.Based on the traditional BP neural network model,the short-term power load prediction algorithm based on BP neural network is established.Firstly,the load data is normalized.Then the traditional single hidden layer BP neural network is used to train the BP neural network using the historical load data as the training data of the neural network.Finally,BP neural network is used to predict the load curve.The experiment shows that it is reasonable and effective to use BP neural network to predict the power load,which provides the basic direction and thinking for the following experiments.2.The short-term power load prediction algorithm of cyclic neural network with density space clustering is proposed.Firstly,the 8-dimensional power load characteristic is used to represent the original high-dimensional power load curve.Then the DBSCAN algorithm is used to cluster the load curves.Furthermore,the load characteristics of the predicted days and the Euclidean distance of the clustering center are used to determine the similar daily category of the predicted days.Finally,the same kind of load data is used as the training data of RNN to predict the load curve.The experiment shows that the idea of using the combination model of clustering algorithm and neural network to predict the power load is effective,and the proposed algorithm has a better prediction effect on the short-term power load prediction,which indicates that the algorithm has a certain practicability.3.A short term power load prediction algorithm based on the maximum deviation similarity criterion is proposed.Firstly,the applicability of MDSC clustering algorithm ismodified to make it more suitable for the clustering of power load data.Then the load curve is represented by the load characteristics of four dimensions.Furthermore,the load characteristics of the predicted date and the Euclidean distance of the clustering center are used to determine the similar category of the predicted date.Finally,the same type of load data is used as the training data of LSTM neural network to predict the load curve of the predicted day.Experimental results show that the proposed algorithm is effective in short-term power load forecasting,and it is practical.4.Proposed a short-term power load prediction algorithm based on the combination model of LSTM and GRU based on k-mdsc integrated clustering.Firstly,the parameter setting formula of MDSC clustering algorithm is proposed to fill in the blank where there is no guidance for parameter setting of MDSC clustering algorithm.Then,based on k-means algorithm and MDSC algorithm,k-means integrated clustering algorithm is proposed to cluster the power load data,and load curve is represented by load characteristics of four dimensions.Furthermore,the load characteristics of the predicted date and the Euclidean distance of the clustering center are used to determine the similar category of the predicted date.Finally,the same type of load data is used as the training data of LSTM and GRU combined model to predict the load curve of the predicted day.Experimental results show that the proposed algorithm is effective in short-term power load forecasting,and it is practical.