Research On The Optimization Methods Of Partial Load For Weighing System Based On LS-SVM

Weighing system is a very important part in the process of industrial production,and its weighing accuracy affects the success or failure of production.Traditional weighing systems usually use a single-point weighing system.When the weighed material on the production line has a certain quality and the barycenter of the weight is not in the center of the weighing platform,it will bring the deviation of load to the weighing system,then affect the result of the weighing,and ultimately affect the whole production process.Therefore,this paper tries to research the multi-sensor weighing system,which aims to compensate the bias error of weighing system by weighing by multiple sensors.The main content of this paper is to build a support vector machine(SVM)weighing system model.By analyzing the shortcomings of the traditional linear weighing system model,a new method of weighing system modeling using least squares support vector machine(LS-SVM)is proposed.The paper focuses on the modeling of weighing system based on LS-SVM for compensating partial load errors.Firstly,through the research of the status of the weighing system,then determine the design objective of the weighing system.Then,introduce the software and hardware platform of the multi-sensor weighing system in detail.Then the traditional linear multisensor model is analyzed and point out the limitations.Furthermore,the causes of two kinds of bias load in multi-sensor weighing system are analyzed in detail,and the corresponding optimization method is given.Finally,the paper present a multisensor weighing system model based on LS-SVM.Before modeling,this paper discusses the statistical learning theory and gives three steps to solve the problem by using the statistical learning method.Then,based on these three steps,the theoretical content of the LS-SVM is discussed in detail.After establishing the basic theory,then realize the whole process of LSSVM weighing system modeling by using LS-SVM toolbox on Matlab platform,including training data collection,data preprocessing,training model and so on.At the same time,a more optimized model,sparse model,is introduced.Finally,by comparing the residual analysis of the modeling data,the data comparison between the least squares linear model and the LS-SVM nonlinear model,and using the test set for actual prediction,the superiority of LS-SVM for weighing system modeling is verified.The final result show that the nonlinear multisensor weighing system model established by LS-SVM is better than the linear model established by the traditional least square method.Its weighing error is smaller and its stability is stronger.And the bias error of multi-sensor weighing system is compensated effectively.