基于支持向量机的离心泵磨损特性研究

摘要: 针对离心泵磨损机理错综复杂和高度非线性的特点，提出了基于最小二乘支持向量机（LSSVM）的离心泵磨损特性分析方法，通过对算法的实现，建立了离心泵的磨损特性分析和几何参数优化的智能模型，模拟得到复杂和非线性很强的离心泵的磨损特性关系，分析了磨损随轮叶片几何参数的变化规律。与神经网络和普通支持向量机进行了性能比较，参数优化选择的LSSVM磨损预测的平均相对误差只有0.005%，学习速度为12步，训练时间为1.1秒，学习速度和预测精度得到了很大的改善。为离心泵的磨损特性分析和离心泵的抗磨可靠性设计提供了一种可行的借鉴方法。

The wear characteristic research of centrifugal pump based on support vector machines

Abstract: It is presented that intelligent method is applied to wear characteristic analysis of centrifugal pump basis on the least squares support vector machine aiming at the complex and strong non-linear specialty of centrifugal pump wear mechanism. The intelligent model of centrifugal pump wear characteristic and geometry parameters optimum is built by algorithm realization, complicated and strong nonlinear wear characteristic of centrifugal pump is simulated and wear variational law with vane geometry parameter is analyzed. LSSVM of optimized parameters is compared with neural network (NN) and common SVM in performance, mean relative error of wear forecast is 0.005%, learning speed is 12 steps, training time is 1.1 s, learning speed and forecast precision are improved greatly. A feasible method is provided for wear characteristic analysis and resisting wear reliability design of centrifugal pump.