基于PCA-MPA-ANFIS模型的年径流预测研究

为提高径流预测精度,提出一种将海洋捕食者算法(MPA)与自适应神经模糊推理系统(ANFIS)相结合的径流预测方法,选取6个标准测试函数对MPA进行仿真验证,并与PSO算法的仿真结果进行比较;通过主成分分析(PCA)对数据样本进行降维处理,使输入数据简洁且更具代表性;利用MPA优化ANFIS条件参数和结论参数,建立PCA-MPA-ANFIS径流预测模型,并构建PCA-MPA-支持向量机(SVM)、PCA-MPA-BP作对比模型;基于云南省革雷站、新疆伊梨河雅马渡站年径流预测实例对PCA-MPA-ANFIS、PCA-MPASVM、PCA-MPA-BP模型进行验证。结果表明,MPA仿真效果优于PSO算法,具有较好的寻优精度和全局搜索能力;PCA-MPA-ANFIS模型对两个实例年径流预测的平均相对误差分别为1.08%、4.49%,平均相对误差较PCA-MPA-SVM模型分别降低了32.5%、37.1%,较PCA-MPA-BP模型分别降低了58.2%、37.6%,具有较好的预测精度和泛化能力。可见将PCA-MPA-ANFIS模型用于径流预测是可行和有效的。

Annual Runoff Forecasting Based on PCA-MPA-ANFIS Model

In order to improve the accuracy of runoff prediction, a method of runoff prediction combined marine predator algorithm (MPA) with adaptive neural fuzzy inference system (ANFIS) was studied. Six standard test functions were chosen to verify the MPA and compare with the simulation results of PSO. The principal component analysis (PCA) was used to perform dimension reduction processing on the samples data, which make the input data simple and more representative. The MPA was used to optimize condition parameters and conclusion parameters of the ANFIS. The PCA-MPA-ANFIS runoff prediction model was established, and the PCA-MPA-Support Vector Machine (SVM) and PCA-MPA-BP were chosen as comparison models. The annual runoff forecast examples of Gelei Station and Yamato Station validate the PCA-MPA-ANFIS, PCA-MPA-SVM, and PCA-MPA-BP models. The results show that the simulation effect by the MPA is better than that of PSO, and it has better optimization precision and global search ability. The average relative errors of the PCA-MPA-ANFIS model for the annual runoff predictions of the two examples were 1.08% and 4.49%, respectively. The average relative errors were 32.5% and 37.1% lower than those of the PCA-MPA-SVM model, respectively. Compared with PCA-MPA-BP model, the average relative errors has been reduced by 58.2% and 37.6%, respectively, and it has better prediction accuracy and generalization ability. It is feasible and effective to use the PCA-MPA-ANFIS model for runoff prediction.