基于PCA-LSSVM的厌氧废水处理系统出水VFA在线预测模型

采用IC厌氧废水处理系统处理人工合成废水,并利用PCA-LSSVM模型对系统出水挥发性脂肪酸(VFA)进行预测.首先利用主成分分析法(PCA)分析影响厌氧废水出水VFA浓度的多个变量的相关性并降低输入变量维数,然后用网格搜索结合10倍交叉验证优化LSSVM模型参数sig2和gam,最后利用建立的模型对实验数据进行仿真预测.仿真结果表明,稳态LSSVM模型对稳态条件下厌氧废水处理系统出水VFA具有很好的仿真预测能力,相对误差在4.72%以内,平均相对百分比误差(MAPE)为1.61%,均方根误差(RMSE)为1.08,相关系数达0.9996;稳态干扰LSSVM模型对厌氧废水处理系统出水VFA的仿真预测精度有所降低但仍然具有较好的预测能力,平均相对百分比误差(MAPE)为15.83%,均方根误差(RMSE)为15.45,相关系数为0.9984,该方法可为厌氧出水VFA在线预测和厌氧废水处理系统的优化控制提供指导.

A PCA-LSSVM model for on-line prediction of the effluent VFA in an anaerobic wastewater treatment system

In this paper, a soft-sensing model based on PCA-LSSVM was employed to predict the effluent volatile fatty acids (VFA) in an IC anaerobic wastewater treatment system treating synthetic wastewater. In order to reduce the dimensions of input variables, principal component analysis (PCA) was used to analyze the correlation of multiple variables. Meanwhile, grid search with 10-fold cross-validation was used to obtain the optimal value of the important parameters (gam and sig2) of the least square support vector machine (LSSVM) model. The results demonstrated that a good forecasting performance was achieved through the steady LSSVM model under steady state. The maximum relative error, mean absolute percentage error (MAPE), root mean square error (RMSE) and correlation coefficient (r) were 4.72%, 1.61%, 1.08 and 0.9996, respectively. Compared with the steady LSSVM model, the non-steady LSSVM model had higher RMSE (15.83%) and MAPE (15.45) but better prediction performance with and r value of 0.9984. Therefore, the method can provide technical guidance for online prediction of VFA and the optimization control of anaerobic wastewater treatment system.