Research On Reliability Prediction Based On Improved Grey Hybrid Model

With the continuous development of science and technology and the introduction of new materials,the requirements for reliability of products are getting higher and higher,and the life of products is getting longer and longer.For products with high reliability and long life,the actual storage environment must be quickly assessed.For high reliability,accelerated life testing is required for life prediction.In recent years,life prediction technology has developed rapidly.The existing life prediction methods are roughly divided into physical failure models and data-driven methods.For high reliability,for long-life products,data-driven prediction methods do not depend on the failure mechanism of the product.Data-driven methods generally include methods based on statistical models,methods based on reliability functions,and methods based on artificial intelligence,which are life predictions in recent years the main research direction.Based on the defects of traditional statistical models that require large amounts of data,individual independence,and samples obey a certain typical distribution,this paper proposes the use of gray system theory for life prediction.However,in view of the problem of low prediction accuracy of the traditional GM(1,1)model,The background value of GM(1,1)is optimized and improved.In order to improve the prediction accuracy of the GM(1,1)model,the exponential smoothing method is used to optimize the background value and the IGM(1,1)model is established.This improvement method is mainly the geometric method is used as the initial improvement formula,and the integration accuracy of the background value is used as the starting point for this improvement to be analyzed,and then simulated and verified in the traditional accelerated life test.The failure data is small and the acceleration model is difficult to determine based on gray prediction The theory analyzes the constant-life test data that obeys the Weibull distribution,uses stress-related weights to generate background values to supplement the missing data,and establishes an equidistant gray prediction model to modify the parameters of the accelerated life prediction model.The analysis of the calculation examples shows that the gray prediction theory has relatively small errors in the results obtained by the accelerated test data evaluation model,and the prediction accuracy is high.By reanalyzing the traditional GM(1,1)model,it is proposed to use the complex Cotes formula and the Newton interpolation method to reconstruct GM(1,1)Background,while improving the accuracy of model predictions,it is also necessary to prevent the occurrence of data shocks.The sixth-order convergence of the complex Cotes formula is used as the goal to improve the accuracy.The simulation accuracy of the improved model is tested,and the traditional trapezoid Compared with the prediction model of the background value constructed by the formula,the improved model has been optimized from the first-order algebraic accuracy to the sixth-order convergence without obvious oscillation.In further research,it is found that in the case of insufficient sample data,GM(1,1)modeling process of the model also wastes the content contained in the initial value,so in order to maximize the data content of the initial value,the initial value is optimized,and the new initial value is obtained for modeling.The improved IGM(1,1)model is applied to the corrosion life prediction of a certain component,and the theoretical analysis and application examples show that the proposed improvement method is effective.When studying the life prediction of a small sample,I found that time series analysis and neural network models can also be used for statistical analysis.Moreover,in the process of solving actual problems,data prediction is not just a simple problem in one aspect.Combining prediction methods with different aspects and advantages,it is found that the prediction accuracy can be better improved.Therefore,the gray model and the ARMA model found in this content are all related to time,and the gray-ARMA model is combined to evaluate the reliability and improve as much as possible Prediction accuracy.In-depth research on the combination of the two models.Based on the analysis of the existing prediction models,a combined prediction model based on the improved gray GM(1,1)and autoregressive moving average model is proposed.By using the buffer operator to preprocess the data,a fixed-weight and variable-weight GM(1,1)-ARMA(p,q)combined prediction model is established and applied to the reliability data prediction of a certain type of missile.Theory Both analysis and application examples show that the method proposed in this paper is effective.In combination with the fact that most of the data in real life is nonlinear,the BP neural network model is used to correct the residual of the gray model,the gray BP neural network model is established,and the prediction accuracy of the model is verified.To study the Qinghai-Tibet Railway South Tanggula-Ando section Based on the settlement and deformation monitoring data of frozen soil roadbed,a roadbed settlement prediction model based on gray neural network is proposed,and the residual of GM(1,1)fitting data is trained through BP neural network to obtain the residual sequence after training.,And then obtain the new prediction value.Compared with the previous results,it is found that the improved gray BP neural network hybrid model has higher prediction accuracy and can effectively predict the settlement of railway frozen soil roadbeds.According to the prediction results,engineering maintenance suggestions for settlement dangerous points are given.