Short-Term Power Load Forecasting Based On WT-ALCSO-BP

In the current era,artificial intelligence technology is developing rapidly,and the application of machine learning in various fields is becoming more and more extensive.With the rapid development of the economy,electricity has become the main energy source in people’s life and production,which has indirectly contributed to the coupling between machine learning and the field of electricity and energy.By using an effective forecasting model to make early perception and forecasting of power load,etc.,to provide scientific guidance for the power generation department,so as to avoid waste and achieve effective load balance,which is the trend of the modernization of power and energy.However,traditional prediction models often have problems such as unstable convergence and low accuracy.On the other hand,for application scenarios with strong regularity such as electricity,it is necessary to consider many correlation factors including the load itself,that is the accuracy and applicability of raw data are required to be high.How to accurately forecast power load is a major challenge faced by current forecasting models.Aiming at this problem,the method of constructing load forecasting model based on multi-model combination provides a new solution.However,due to the different organizational structures of various models,how to effectively fuse and how to ensure the validity of their models after fusion is a new challenge.Therefore,this paper conducts in-depth research on the above problems,and proposes a short-term power load forecasting model based on wavelet analysis and improved cat swarm algorithm to optimize BP neural network.The main research work and results are as follows:（1）A data preprocessing model based on DB4 wavelet is constructed.Starting from the load itself,the characteristics and laws of the power load data are deeply studied,and a data preprocessing model based on DB4 wavelet is constructed.It is proposed to use the DB4 wavelet in wavelet analysis to decompose,denoise and reconstruct the load historical data,and eliminate the noise caused by various factors in the original data,so as to ensure the validity of the data in later use.（2）A strong correlation factor analysis model of power based on FP-Growth for power load is constructed.Aimed at the problems existing in the current analysis of the correlation factors of power load data,most of them do not use or use the prediction model of human experience,a power load correlation analysis method based on FP-Growth is proposed.Through the analysis of various factors and load data,the largest impact factors are scientifically selected to provide reliability guarantees for the input parameters of the later forecasting model.（3）An optimized cat swarm algorithm（ALCSO）based on Logistic function selfadaptation and Cauchy multi-policy perturbation is proposed.Aiming at the problems of low convergence and low precision in the traditional cat swarm algorithm,the logic function in machine learning is introduced as an adaptive weight to improve the speed in the tracking mode,and the optimization imbalance is adjusted to improve the convergence of the algorithm;Combined with the Cauchy distribution function in probability theory,the multi-strategy perturbation mechanism of the algorithm is constructed to ensure the accuracy of the algorithm,and its improved effectiveness is verified through various test functions.The experimental results show that the convergence of the optimized algorithm is improved by 15%-48.15%,and the accuracy is improved by 14.18%-45.21%.（4）A short-term forecasting model of power load（WT-ALCSO-BP）based on the optimized BP neural network by the improved cat swarm algorithm was constructed.In order to avoid the problems of slow network convergence,low accuracy and sensitive input parameters,the improved cat swarm algorithm was used to optimize the BP neural network,and a short-term prediction model of power load based on the optimized BP neural network was constructed.Moreover,we use the power load data and influencing factor data of Guizhou Province in 2020 to conduct model performance inspection and testing.The experimental results show that the model in this paper has improved the accuracy by 2.16%-3.04% and the convergence by 21%-54%,compared with the traditional prediction model,and has a better prediction effect.In this paper,a multi-combination prediction model for power load is constructed,aimed at the characteristics of the power energy field.By introducing the wavelet analysis model,the original load data cleaning and noise reduction are realized,the cat swarm algorithm is further optimized,and the BP neural network is improved by combining the optimized cat swarm algorithm,and finally a short-term prediction model of power load based on the optimized BP neural network is constructed,to achieve more real-time and more accurate prediction of power load.This research provides an effective reference scheme for scientific and effective power load balancing and optimal configuration in the field of power energy,which has high research significance and value.