Algorithm Research Based On ICE-FTS-MOBP Hybrid Model

The efficient and stable power system is an important foundation for the country's economic development.However,since the production,transportation,distribution and consumption of electric energy are almost simultaneously carried out,the electric power cannot be stored in a large amount,and thus the power generation capacity of the electric power system must maintain a dynamic balance with the load demand.Therefore,it is of great practical significance to improve the accuracy of short-term power load forecasting.However,the short-term load raw data itself has characteristics such as nonlinearity and high noise.The single prediction model usually ignores data preprocessing and model parameter optimization,which often leads to poor prediction results and is difficult to meet the actual needs of the power sector.In this context,hybrid models with good predictive effects are becoming more widely used.In order to provide valuable information to decision makers and iaprove the sta-bility and accuracy of prediction,this paper proposes a new hybrid model—ICE-FTS-MOBP with data preprocessing,parameter optimization,forecasting and evalua-tion.The model combines the adaptive white noise complete set of empirical mode decomposition techniques ICEEMDAN and FTS-MOBP modes.Among them,ICEEMDAN preprocesses the original data sequence and eliminates various random uncertain factors that affect the prediction accuracy.The FTS-MOBP model combines the advantages of fuzzy time series and BP neural network to make the two comple-ment each other to exert greater predictive performance.The main steps of the hybrid model are as follows.In the first step,based on the modal decomposition technique,the half-hour load raw data of Victoria,Australia is preprocessed,and the improved data set is decom-posed and decomposed by the improved adaptive noise fully integrated empirical mode decomposition technique(ICEEMADAN).Noise and reconstruction effectively eliminate modal mixing and obtain ideal load data,which lays a foundation for im-proving prediction accuracy.In the second step,the point prediction experiment is car-ried out by combining fuzzy time series and BP neural network model,and BP network weights and thresholds are optimized by multi-objective algorithm.Seven evaluation indicators are selected,and the obtained prediction results are compared with different types of benchmark models.The experimental results show that the proposed model can effectively improve the accuracy and stability of prediction.In addition,in order to prove the superiority of the selected multi-objective algorithm,four test functions were selected to test the performance of the multi-objective dragonfly algorithm,and the simulation experiment of interval prediction was carried out to quantify the change of prediction results caused by the uncertainty factor.A prediction interval with both reliability and clarity is constructed.The experimental results show that when the sig-nificance level is chosen to be 0.3,the prediction interval is the most effective,and it has the characteristics of wide application range and strong robustness.In the third step,considering the problem of prediction accuracy distribution,two methods of DM test and prediction validity are used to comprehensively consider the significance of the prediction results.The results show that the hybrid prediction model has the best prediction ability.The first-order prediction effect is greater than 0.93,and the second-order value is greater than 0.85,and the validity of the first-order and second-order predictions is higher than other power load prediction models in most cases.In addi-tion,the paper also discusses the contribution of each part of the model to the predic-tion.The conclusions show that the modal decomposition technique provides stronger performance optimization for the final result than other parts.Finally,the post-work outlook and improvement suggestions are put forward to facilitate the orderly follow?up work.The innovation of this paper is that it is difficult to extract the fault characteristic frequency caused by noulinear and non-stationary characteristics of short-term electric load data.The method of adaptive white noise is used to pre-process the data by the improved complete ensemble empirical mode decomposition with adaptive noise(ICEEMDAN).The main internal features are extracted and the noise points are elim-inated by filtering layer-by-layer decomposition and integration.This method has not yet been applied in the field of domestic power load forecasting.It can be effectively verified by this paper to improve the effect of load forecasting.In addition,most of the previous studies on load forecasting only focus on the establishment of single target-improvement of prediction accuracy.This paper uses multi-objective dragonfly opti-mization(MODA)algorithm to optimize the initial weight and threshold of BPNN model to achieve high precision and high stability at the same time.There are two shortcomings in this paper.One is to ignore the influence of some factors on the time series,and only make a preliminary analysis and research on the prediction work.How to effectively consider multiple factors and design predictive models and algorithms with multiple variables will be a direction worthy of further study in the future.Second,the comparison of sequence denoising is relatively insufficient.The improved com-plete ensemble empirical mode decomposition with adaptive noise(ICEEMDAN)of adaptive white noise used in this paper is ouly one kind of denoising algorithm in the EMD family.Although the denoising effect is more significant,other types of de-noising still need to be used in the next step.The algorithm performs a comprehensive comparison of the denoising effects,such as Kalman filtering,wavelet decomposition,and so on.