基于最小二乘支持向量机的泵性能分析

Pump performance analysis based on least squares support vector machine

泵的性能曲线是泵选型、优化调度和泵站运行的重要依据,通常该曲线均是通过试验或是根据试验数据和性能图表上的数据进行曲线拟合而获得,但这些方法复杂昂贵,而且拟合精度不高。针对以上方法的缺点,提出了一种基于交叉验证最优参数选择的最小二乘支持向量机(LSSVM)泵性能预测方法。通过最小二乘支持向量机(LSSVM)学习算法网络的设计和构建,并应用网络搜索-交叉验证的方法对支持向量机参数进行优化选择,模拟得到复杂和非线性很强的泵的性能曲线,经优化模型输出值和试验值、同多项式拟合值以及径向基神经网络误差的比较,交叉验证最优参数选择的最小二乘支持向量机具有优良的非线性建模能力和泛化能力,在有限学习样本条件下仍获得了很高的精度,平均相对误差为0.02378%,为泵的性能分析提供了一种简便可行的智能方法。

Performance curve of pump is important basis of pump types selecting, optimal operation and pump station running. The curve is usually obtained from experiment or by fitting the experimental data of performance graph, but those methods are complex, high expense and imprecision. Based on optimum parameter selection with cross validation, the least squares support vector machine （LSSVM） method was presented for pump performance forecast in the light of the difficulty of the above two methods. Complicated and strong nonlinear pump performance curve was simulated by network design and conformation of LSSVM learning algorithm and the optimized SVM parameters were selected by the method of network searching and cross validation. Compared the errors with output values of the optimized model, test value and output value from polynomial fitting and RBFNN, LSSVM whose parameter was optimized with cross validation had excellent ability of nonlinear modeling and generalization. It gained high precision under limited learning samples （mean relative error is 0.02378%） and provided a simple and feasible intelligent approach for pump performance analysis.