Study On Sensor’s Dynamic Modeling And Compensation Using Multiple Kernel Support Vector Machine

Sensor is the window of the measuring system, and also a tool of obtaining informationfor the measuring systems, the performance of the measuring system is determined by its testaccuracies. The performance of the measuring system can be improved by improving thesensor’s dynamic characteristics and static characteristics, and its dynamic characteristics isparticularly critical. At present, the research topic about how to improve the dynamiccharacteristics of the sensor has become more and more important all over the world. Themain purpose of this article is improving the sensor’s dynamic characteristic by applyingsupport vector machine(SVM) to sensor dynamic modeling and compensation. And the thesishas further studied the application of the multiple kernel least squares support vector machine(MK-LSSVM) in sensor dynamic modeling and compensation, through simulations toanalysis the effect of different support vector machine algorithm in dealing with the sensorproblems. The main contents of this thesis are as follows:Firstly, research on the the core theory and its implementation method theory andimplementation of the standard support vector regression(-SVR), Modified support vectorregression(-SVR), least squares support vector regression(LS-SVR) and MK-LSSVM, aswell as the composition of kernel function and kernel function selection of weights.Secondly, several traditional methods of sensor dynamic modeling are analyzed, basedon this, from the dynamic characteristics of the sensor, it has further studied the application ofSVM method in dynamic modeling of sensor. By-SVR,-SVR, LS-SVR and MK-LSSVM,it has conducted the simulation experiment of dynamic modeling for the sensor model that haslinear and nonlinear dynamic performance, and different dynamic modeling effects of kernelfunction are analyzed in the experiment. According to the comparative analysis of thesimulation experiment, the algorithm of MK-LSSVM has certain superiority than other SVMin the dynamic modeling of sensor, relying less on the kernel function and its parameter andhaving high modeling precision and good anti noise resistance performance.Finally, research on the application of SVM in the sensor dynamic compensation.Through the research of the dynamic compensation principle based on reference model, usethe SVM to design the dynamic compensator of sensor. By-SVR,-SVR, LS-SVR andMK-LSSVM, it has conducted the simulation experiment of dynamic compensation for thesensor model that has linear and nonlinear dynamic performance, and compared and analyzedthe effect on the sensor of the dynamic compensators that select different algorithm under thesame model. Results show that the dynamic compensator that is designed by SVM methodhas largely improved the dynamic characteristics of sensors, but in terms of the comparison,the dynamic compensator that is designed MK-LSSVM method has the best effect on sensor, which has effectively improved the dynamic output characteristics of sensor and designed amore effective learning algorithm for the dynamic compensator of sensor.