克隆规划-交叉验证参数优化的LSSVM及惯性器件预测

Least square support vector machine based on parameters optimization of clone programming-cross validation and inertial component forecasting

为改善最小二乘支持向量机的泛化性能,将克隆规划、交叉验证相结合的优化算法用于最小二乘支持向量机的参数优化.克隆规划算法是具有局部、全局搜索能力的优化算法,能有效避免陷入局部极值;交叉验证算法的无偏估计性抑制了训练过程中“过拟合”和“欠拟合”.在该优化算法中,用交叉验证误差构造抗体抗原亲合度,用克隆规划算法寻找最小二乘支持向量机的最优参数.用优化的最小二乘支持向量机回归模型建立了惯性器件时间序列预测模型.实验结果验证了算法的有效性及预测模型的泛化性能.预测模型为动态补偿、故障预测提供了依据.

For improving the generalization ability of the least square support vector machine （LSSVM）, the parameter optimization algorithm of clone programming-cross validation is employed to select optimal parameters of LSSVM. The clone programming algorithm has the superior capability in local and global search, and local minimums are refrained efficiently; cross validation has the unbiased estimator property, and therefore, the problems such as over training or insufficient training are avoided. In the optimization algorithm, the avidity function is constructed by the cross validation error, and moreover, optimal parameters of LSSVM are chosen by the clone programming algorithm. The time series forecasting model of the inertial component is built with LSSVM. Experimental results prove the effectiveness of the optimization algorithm and generalization ability of the forecasting model, and the forecasting model provides a support on dynamic compensation and fault forecasting of the inertial component.