论文部分内容阅读
针对目前板形模式识别模型泛化能力不高、训练速度慢等缺陷,以1次、2次、3次、4次勒让德正交多项式为板形缺陷基本模式,提出了由支持向量回归机(SVR)构建的模式识别模型;为了提高该模型的精确度,引入万有引力算法(GSA)优化SVR的参数,由此构成GSA-SVR预测模型。仿真试验结果表明:GSA-SVR模型不仅识别结果精度高,而且与PSO-BP神经网络模型相比泛化能力更强,训练速度更快,其识别结果可以为板形控制提供有效的依据。
Aiming at the defect that the generalized pattern recognition model of the plate-shaped pattern recognition is not generalized and the training speed is slow, the quadratic quadratic polynomials of Legendre orthogonal first, second, third and fourth quadratic form are given. (SVR); in order to improve the accuracy of the model, the GSA-SVR prediction model is constructed by introducing the universal gravitational algorithm (GSA) to optimize the SVR parameters. Simulation results show that GSA-SVR not only has high recognition accuracy, but also has more generalization ability and faster training speed than PSO-BP neural network model. The recognition results of GSA-SVR model can provide an effective basis for plate-shaped control.