Research On Soft Sensor Of Grain Size Based On The Improved Least Square Support Vector Machine

The grain size measurement methods and soft sensor are introduced, and aiming at the series of problems of grain size online measurement, this paper put forward soft sensor techniques based on improved least squares support vector machine to predict grain size. The main contents include:1) Basic particle swarm optimization algorithm and standard particle swarm optimization algorithm are introduced, focusing on the improvement of the standard particle swarm optimization. PSO is improved from two aspects, on one hand, the mean best position is introduced to change particle’s velocity update, on the other hand, wavelet mutation mechanism is introduced into the mean best position. The experimental results show that the proposed algorithm has powerful optimizing ability and higher optimizing precision. In addition, analyze the quantum-behaved PSO and put forward the improved quantum-behaved PSO based on wavelet mutation. The experimental results show that the improved quantum-behaved PSO based on wavelet mutation has high performance.2) Support vector machine regression algorithm is introduced, focusing on LS-SVR, hybrid kernel function is introduced into LS-SVR. The impact of the SVR model parameters on the SVR performance is researched, focusing on the parameters selected based on the improved QPSO algorithm. The simulation on the nonlinear function and two groups of data set show that the method based on WQPSO-LSSVR is good.3) The improved support vector machine regression algorithm is applied in the soft sensor model of aluminum billet grain size from actual production process. The result shows that the estimate of the soft sensor of particle size has a high performance based on WQPSO-LSSVR, the method is feasible and superior.