Research On Power Load Forecasting Method Based On Outlier Detection And Fused Neural Network

Since the age of electrical industry,electrical energy has been increasingly used in human’s daily life,becoming one of the essential and important energy sources for human beings.Power load forecasting is to analyze people’s production and life as well as the use of electricity to provide a reliable judgment and decision-making basis for the normal operation of the power system.With the rapid development of smart grid and energy Internet,the amount of data collected from the Internet is increasing and the data structure has become more complex,which can easily be mixed with abnormal data,which can distort the analysis results of load data and eventually lead to the reduction of prediction accuracy of the model.The paper addresses the shortcomings of traditional anomaly data identification methods,and proposes a power load forecasting method based on outlier detection and fused neural networks.Firstly,an outlier detection method based on Local Outlier Factor(LOF)optimized isolated forest is introduced to improve the accuracy of detecting outliers in power load data and to obtain good power load data.The algorithm organically combines LOF with isolated forest,takes LOF as the root node of isolated tree,builds trees on this basis,and replaces the isolated tree branching discrimination operation in the isolated forest algorithm with LOF algorithm,uses the local outlier factor result of each tree as the criterion for judging the high branching of the tree,and enhances the detection ability of the algorithm for local anomaly data.By using multiple sets of data for experiments and comparing with other outlier detection algorithms,it is verified that the algorithm can effectively overcome the K-value sensitivity and enhance the global as well as local anomaly detection performance of the algorithm,and this combined outlier detection effect is significantly better than the detection effect of a single algorithm.Secondly,by analyzing the main influencing factors of power load and studying the characteristics of periodic changes of power load data,we propose an power load forecasting model based on fusing deep belief network and Echo State Network(DBN-ESN).Firstly,the Echo State Network(ESN)is used to roughly forecast the power load to obtain the forecast model of the load variation,and then the forecast data and the related weather and other influencing factors are used as the input of the model,and the load data to be predicted is used as the output of the Deep Belief Networks(DBN)model,and ESN is used again as its error compensation module to achieve high accuracy forecast of power load by this method.Then,for the slow convergence of the forecast model and the instability of the detection results caused by artificially setting the number of implied layers in the DBN,the Beetle Antennae Search algorithm is introduced to further optimize the parameters of the forecast model and enhance the stability of the algorithm.Through various comparison experiments,it is proved that the fusion method can better improve the forecast accuracy of the model.Finally,this paper uses two sets of power plant load data from different regions for experimental validation,and the results show that the method proposed in this paper can better detect anomalies in the data preprocessing,effectively improving the forecast accuracy of the model and having better applicability.