Study On The Signal Reconstruction Of Sensor Based On Ls-svm With Analysis Of The Problem Of Uncertainty

This article forces on the sensor signal reconstruction based on the LS-SVM. Starting with the analyzes of uncertainty, the parametric optimization and the over-fitting problem are considered, looking forward to building a reconstruction method with high precision which can be also used in multifunctional sensor with a multi-input multi-output(MIMO) characteristics.With the deeply analyses of statistical techniques such as jackknife and bootstrap, an appraisal system is built based on the analyses of uncertainty. A resample method which is similar with the bootstrap is proposed. With this method the optimization of the uncertainty estimate can be done and the over-fitting problem will be avoided.In the process of the parametric optimization of the LS-SVM based on GA, considering the calculated amount, based on the study of the principle of GA, a simple two-step GA is proposed. With the two-step GA and the appraisal system, the calculated amount can be reduced without the sacrifice of the precision, which raise the efficiency of the system.simulation experiment using the system models of platinum resistor and a resistance network is proposed with disscutions and analyses of the result. The result shows that the reconstruction system with the analyzes of uncertainty has a good preformence which can raise the precision of the reconstruction while avoilding the over-fitting problem.