基于小波和乘法混合核函数LSSVM的顺风向非高斯空间风压预测

Forecasting non-Gaussian spatial along-wind pressure using wavelet kernel and multiplicative mixed kernel functions based LSSVM

提出了基于Marr小波核函数最小二乘支持向量机(Marr-LSSVM)的顺风向非高斯空间风压预测算法。通过传统高斯核函数(RBF)和多项式核函数(Poly)的乘法运算,提出了Poly*RBF-LSSVM(MK-LSSVM)的空间风压预测算法。运用粒子群优化(PSO)算法,对Marr-LSSVM、传统单核CSK-LSSVM和MK-LSSVM的惩罚参数、核函数参数、权重、尺度因子进行优化,建立基于智能优化的非高斯空间风压预测算法;以30 m和50 m处模拟顺风向风压时程作为输入样本,使用提出的预测算法对40 m处风压时程进行了预测。数值分析表明,Marr-LSSVM、MK-LSSVM比CSK-LSSVM具有明显高的非高斯风压预测性能。

Here, Marr wavelet kernel-based least squares support vector machines (LSSVM) referred to as Marr-LSSVM, was proposed to predict along-wind non-Gaussian spatial wind pressure. Through multiplication operation of the conventional radial basis function (RBF) kernel and polynomial kernel, Poly*RBF-LSSVM was then proposed, it was called the multiplicative mixed kernel MK-LSSVM. By using the particle swarm optimization (PSO) algorithm, optimizations were implemented for penalty parameters, kernel parameters, weights, and scale factors of Marr-LSSVM, conventional single kernel CSK-LSSVM, and MK-LSSVM, thus the non-Gaussian spatial wind pressure forecasting algorithms were built based on intelligent optimization. The simulated along-wind pressures at 30 m and 50 m were taken as input samples, the wind pressure at 40 m was then predicted using the proposed algorithms. The numerical analyses demonstrated that Marr-LSSVM and MK-LSSVM can provide an obvious higher performance to predict the non-Gaussian spatial wind pressure than CSK-LSSVM can. © 2017, Editorial Office of Journal of Vibration and Shock. All right reserved.