Application Research Of A Nonlinear Combination Model Based On Multi-objective Optimization Algorithm And Error Sequence

Time series prediction is a statistical method that studies the historical evolution of things over time and their inherent development laws,and builds a system model to predict its future development trend.Because time series prediction can accurately predict the future state of things by analyzing the existing historical data,it has been widely used.With the continuous development of society and the advancement of science and technology,especially with the continuous development and improvement of the main technologies such as prediction methods and application algorithms involved in the establishment of prediction models,the applicability of time series prediction has become more prominent.Nowadays,it has a wide range of applications in pattern recognition,finance,economics and other branches of applied science that involve time recording or measurement.However,traditional time series prediction models have relatively strict requirements for the stability and continuity of data,which leads to a narrow application,in order to make up for this deficiency and capture more data information,scholars proposed the concept of a combination forecasting model that combines different models with an appropriate criterion and an effective way.Combined models have developed rapidly in the past few decades,but in these combined models,the effects of each model are mixed,and some of them still have some shortcomings.Therefore,it is extremely necessary to build a stable combined forecasting model with excellent forecasting performance and stronger applicability.In previous studies,most combined prediction models were limited to using different prediction models to predict the original data and then making a simple linear combination.However,due to the randomness,volatility and nonlinearity of the real data,the prediction accuracy that a simple neural network model can achieve is relatively limited,and the error sequence after model fitting is not idealized white noise,which means the error sequence contains the data information that cannot be fully extracted by simple one-time modeling.Thus,owing to the reasons above,this paper proposes a nonlinear combination forecasting model based on multi-objective grey wolf optimization algorithm and variational mode decomposition combined with error sequence.It mainly consists of two parts:First,the combined prediction model is applied to the original data set to obtain the error sequence based on the preliminary prediction results;then the error sequence is combined with the preliminary prediction result to obtain the final prediction value.In the preprocessing part of removing data noise,the variational mode decomposition(VMD)is used to decompose and reconstruct the original data set.In the neural network optimization part,the multi-objective grey wolf optimization algorithm(MOGWO)is used to construct the structural parameters of Elman neural network.MOGWO is a new swarm intelligence optimization algorithm proposed in recent years,which has obtained good prediction performance in practical applications.In this paper,to verify the effectiveness of the proposed model and its applicability in the social and economic fields,the new nonlinear combination forecasting model combined with error series is applied to non-stationary time data in real life,forecast the 10-min and 30-min wind speed data in Penglai City,Shandong Province,China,and Shanghai Composite Index data,Shenzhen component Index data,Shanghai-Shenzhen 300 Index data and small and medium-sized Board Index data from 2015 to 2018.In addition,this paper adopts a relatively complete evaluation index system,which includes nine different evaluation indicators(AE,MAPE,NMSE,RMSE,MAE,IA,FB,U1,U2)and two statistical testing methods(DM test,FE test)to make a more comprehensive and scientific evaluation of the proposed new combination forecasting model.At the same time,this paper also compares the new combination forecasting model of nonlinear combined error series with other nine prediction models.The results show that compared with other comparative models,the proposed model has excellent performance and stability for forecast performance.The innovations of this paper mainly include the following four aspects:Firstly,the novel combination of error sequences is used for prediction,and the optimized artificial neural network is used to nonlinearly combine the original time series with the error sequence to obtain a higher prediction effect.Second,the VMD method is used to extract and reconstruct data features to reduce the noise of complex data.In addition,a multi-objective optimization algorithm is used to optimize the model structure to further improve the stability and accuracy of the prediction.Thirdly,this paper makes a scientific and comprehensive evaluation of the proposed method and the comparison models,and introduces the FE test and DM test on the basis of the use of classical indicators,so as to make the experimental results more reliable.Fourth,in order to reflect the universality of the method,this paper applies the model to predict the wind speed and the stock price index.The results show that the proposed model has good prediction ability and generalization ability,which further enriches and improves the prediction model system.The shortcomings of this paper mainly include the following three aspects:Firstly,the physical factors such as air pressure and temperature are not considered in the prediction of wind speed data,which may affect the accuracy of the results.Second,there are many excellent algorithms and models,and the failure to consider other comparative prediction models is also a potential deficiency of this paper.Thirdly,this paper only selects two types of data for simulation prediction and does not apply to the prediction of nonlinear data in other fields.In the future research,this model can be extended to more fields.