A Nonlinear Homotopy-LM Algorithm and Its Application in Soft Sensor Modeling

In order to novel nonlinear homotopy improve the learning effic LM learning algorithm is iency of existing neural network proposed by taking advantage of learning algorithms, a both Levenberg-Marquardt algorithm and homotopy method. The influence of different transition functions on the generalization ability of neural network is analyzed. Research results show that the proposed algorithm has good properties of stability and converging ability. According to the demands of practical industry, the improved algorithm is applied to construct a soft sensor model for real-time measuring the acrylonitrile yield. Its performance is compared with existing soft sensor modeling methods. Application results show that this model has higher measuring precision as well as better generalization ability, and can satisfy the need of spot measurement.