摘要
分析了典型参数优化算法的局限,对LSSVM目标函数含二次损失函数、样本特征空间分布形状不规则情况,提出了混合参数优化算法,用待优化参数重构LSSVM目标函数,通过自适应遗传算法、交叉验证来优化目标函数、选择最优的核和其它参数,依此建立了陀螺漂移误差序列预测模型。实验结果表明,该预测模型有较高的训练、泛化精度;可为陀螺仪动态补偿、可靠性辅助决策提供可靠依据。
The limitations of typical parameter optimization algorithms are discussed, and some issues are analyzed such as least square support vector machine （LSSVM） with squared cost function and samples with irregular distribution in characteristic space. The hybrid parameter optimization algorithm is presented. The object function of LSSVM is reconstructed by the parameters to be optimized, adaptive Genetic algorithm and Cross-Validation are used to optimize the object function and select the optimal parameters, including kernei parameters and others. The drift error forecast model of the gyro was established using the proposed method. Experimental results show that the forecast model based on the hybrid parameter optimization algorithm has high forecast accuracy in training and generalization, and provides strong basis for the gyro dynamic compensation and reliability aid decision.
出处
《电子测量与仪器学报》
CSCD
2007年第5期55-59,共5页
Journal of Electronic Measurement and Instrumentation
