Soft Sensor Modeling And Optimization Of Aluminum Strip Grain Size Based On PSO-BP

The aluminum electromagnetic roll-casting system is a nonlinear complex system because of it involves machine, electric, casting and processing and so on. The grain size is always an important evaluation standard of aluminum strip's quality, but it can't be detected on-line in real time because of the limitation of two conditions. One is that the factors' relationship of system is complex, and the other is the limitation of detect technology. To realize the on-line detection of aluminum strip grain size, the soft senor based on intelligent information detection and processing to be an inevitable requirement.To solve this problem, combined with the research and development status of aluminum electromagnetic roll-casting and soft senor technology, the principle and processing of the aluminum electromagnetic roll-casting system was deeply analyzed. Then the influencing factors of grain size including roll-casting factors and electromagnetic factors was determined which's data were collected. After normalization processing and pca, the variables were used as the secondary variable in the soft senor model established based on the BP neural network.To solve the problem of slow convergence speed, poor stability and weak generative ability in the BP neural network model, particle swarm optimization was introduced to optimize the model. To shorten the optimization time and enhance the searching ability in the elementary particle swarm optimization, linear varying inertia weight factor were introduced which was used to optimize the weight and threshold of BP network.The results show that, the soft senor model and prediction estimation of aluminum strip grain size can be obtained based on BP neural network and the structure is simple. After optimized by particle swarm optimization, the real-time, stability and generative ability are significantly enhanced. The deterministic coefficient which is an important evaluation index of model has improved. The research of the paper has an important significance of study on the measurement method in real-time of aluminum strip grain size.