Research On Power Load Forecasting Based On XGBoost-LSTM Combined Model

With the rapid development of power industry,there are more and more types of power system.Different power systems are affected by different factors,resulting in a variety of power load data.In different power load application scenarios,the data processing and modeling methods are different.The traditional forecasting methods have been difficult to meet the requirements of better forecasting effect and applicable performance in different power load application scenarios.In order to improve the prediction accuracy and applicability of load forecasting model in a variety of application scenarios,two different data sets are used as data support.The XGBoost algorithm and long-short term memory algorithm in the field of artificial intelligence are studied and improved,and a variety of load forecasting methods are proposed.The main research methods are as follows.Firstly,aiming at the problem of low prediction accuracy in power load application scenarios with few training samples and low feature dimensions,a short-term load forecasting method based on improved PSO optimized XGBoost-Bagging ensemble learning is proposed.The small-scale data set of low dimensional feature vector is selected as the basic training data.The Pearson product moment correlation coefficient is used to extract the main influencing factors,and the similar days are selected by grey correlation analysis.According to the sample set of similar days,a load forecasting model based on XGBoost-Bagging ensemble learning is constructed.XGBoost algorithm based on serial learning mechanism has strong processing ability for nonlinear data.And the parallel integrated learning mechanism of bagging algorithm can improve the generalization ability and stability of the model.At the same time,the improved particle swarm optimization algorithm is used to optimize the model parameters and improve the prediction accuracy.Secondly,for power load application scenarios with large sample size and high feature dimension,a short-term load forecasting method based on Improved PSO algorithm to optimize bidirectional long-short term memory model is proposed.The feature set construction method is adopted for the data set,and the dimension of the feature is added on the basis of the initial feature,so as to obtain the implicit feature information of the data set more comprehensively.Based on the data set constructed by features,a bidirectional long-short term memory model is established for load forecasting.At the same time,the number of hidden layers of neural network is increased,and the deep neural network structure is used to improve the generalization performance of the model to temporal features.In addition,in order to avoid over fitting,the improved PSO algorithm is selected to optimize the neural network parameters,and the optimization performance of the improved PSO algorithm is verified again.Finally,in order to improve the accuracy and applicability of load forecasting models in various power load application scenarios,a short-term load forecasting method based on simulated annealing algorithm optimized XGBoost-Bagging integrated learning and deep bidirectional long-short term memory combined model is proposed.This method combines the above two models.The simulated annealing algorithm is used to optimize the weight coefficients corresponding to the prediction results of each model to obtain the prediction value of the final combined model.According to the two data sets,modeling,prediction and analysis are carried out respectively.The results show that the MAPE values of the prediction errors of the combined model on the two data sets are 0.91% and 0.88% respectively,which is significantly lower than that of each single model.The experiment is repeated 50 times,and the error fluctuation range of the prediction result is only 0.12%,showing high model stability.To sum up,the combined model has high prediction accuracy and applicability in a variety of power load application scenarios.It has certain application value in the field of power system load forecasting.