基于组合核函数OSVR算法的起重机减速齿轮箱磨损趋势预测

Wear Trend Prediction of Crane Gearbox Based on OSVR Method with Combined Kernel Functions

针对起重机减速齿轮箱的磨损过程具有非线性与时变性,传统磨损趋势预测方法无法有效兼顾预测精度与执行效率的问题,提出了一种基于组合核函数的在线支持向量机回归(online support vector regression,OSVR)预测算法。OSVR的在线学习算法能够适应时间序列的时变性并提高执行效率,同时可利用不同的核函数性能,通过组合模型提高预测精度。采用实际齿轮箱铁谱数据对预测算法进行验证,结果表明,基于组合核函数的OSVR预测算法具有很好的预测精度和适应性,能有效预测起重机齿轮箱的磨损故障,且相比于单一OSVR算法和灰色神经网络组合算法有更高的效率和预测精度。

For the nonlinear and nonstationary crane gearbox wear process,prediction precision and efficiency cannot be effectively balanced by using traditional prediction methods.An OSVR prediction method based on combined kernel functions was proposed.The OSVR algorithm can be adapted to time-varying time series and the efficiency can be improved.Simultaneously,the kernel-combined model would be able to improve the prediction accuracy.Experimental results show that the trend of the crane gearbox wear process can be predicted effectively by using the new method,and the prediction is more accurate than the results of the OSVR method with single kernel function and the gray-neural network method.