关于遗传算法优化的最小二乘支持向量机在MBR仿真预测中的研究

摘要

提出了一种基于遗传算法(GA)优化的最小二乘支持向量机(LSSVM)的MBR膜通量预测算法。为了准确的选择LSSVM的参数，该算法采用GA对LSSVM模型的惩罚因子和核函数参数进行优化。针对MBR膜污染因子较为复杂且各因子之间相互交叉，首先对影响MBR膜通量的各因子进行主成分分析(PCA)，提炼出重要因子作为LSSVM的输入层，膜通量作为输出层，然后建立GA-LSSVM仿真预测模型，并用该预测模型运算得出预测结果。通过对比预测结果和实验数据，得出该算法对膜通量有较高的预测精度，并将其与BP神经网络模型进行了比较，结果表明该预测模型具有更高的预测精度。%This paper proposes a prediction algorithm of MBR membrane flux based on GA and LSSVM. The algo-rithm optimizes the penalty factor and kernel parameters of LSSVM model by genetic algorithm. Because of the diver-sity of the factors that affect MBR membrane fouling, we apply principal component analysis (PCA) to extracting the pivotal factors, then take these factors as the LSSVM input, MBR membrane flux as output, and construct GA-LSSVM model in the end. We get predictive results through the model in the end. By comparing the predicted value with expe-rimental value, the model can forecast MBR membrane flux well. We also use BP neural network model to forecast Membrane flux, and get that the algorithm of GA-LSSVM has higher prediction accuracy.