用于水泥熟料fCaO预测的多核最小二乘支持向量机模型

针对水泥熟料游离氧化钙(fCaO)含量预测模型辨识的问题,考虑到单一核函数无法显著提高模型精度,采用多项式核函数、指数径向基核函数和高斯径向基核函数组合构建等价核的方法,建立了多核最小二乘支持向量机水泥熟料fCaO预测模型。同时,利用改进的粒子群优化算法对多核最小二乘支持向量机模型的6个待确定参数进行迭代寻优,避免了模型参数人工选取的盲目性。最后将基于改进粒子群的多核最小二乘支持向量机模型应用于熟料fCaO含量的实例仿真。结果表明,建立的水泥熟料fCaO含量预测模型精度高、泛化能力强。

Multiple kernel least square support vector machine model for prediction of cement clinker lime content

Aiming at the problem of how to accurately predict the cement clinker fCaO content, the traditional single kernel least squares support vector machine (LSSVM) is difficult to show the complex non-linear relation between the clinker fCaO content and corresponding variables exactly. Thus, the multiple kernel least square support vector machine (MKLSSVM) containing three kernel function is presented based on multiple kernel learning to avoid the influence of the single kernel function on the model accuracy. As a result of artificial selection the parameters of MKLSSVM is blindness and uncertainty. The random perturbation chaos particle swarm optimization (RPCPSO) algorithm is presented to get the best parameters of MKLSSVM. The cement clinker fCaO content model is built by using the RPCPSO algorithm to optimize the parameters of MKLSSVM. Simulation results indicated that the RPCPSO algorithm had a fast convergence speed, and the model had high precision and strong ability of power generalization. Thus, the model was valuable for practical application.