Prediction of Sewage Treatment Cost in Rural Regions with Multivariate Adaptive Regression Splines

Abstract

In this paper, to interpret the cost structure of decentralized wastewater treatment plants (DWWTPs) in rural regions, a simple nonparametric regression algorithm known as multivariate adaptive regression spline (MARS) was proposed and applied to simulate the construction cost (CC), operation and maintenance cost (OMC), and total cost (TC). The effects of design treatment capacity (DTC), removal efficiency of chemical oxygen demand (RCOD), and removal efficiency of ammonia nitrogen (RNH3-N) on the cost functions of CC, OMC, and TC were analyzed in detail. The results indicated that: (1) DTC is the most important parameter to determine cost structure with relative importance of 100%, followed by RCOD and RNH3-N with relative importance of 16.55%, and 9.75%, respectively; (2) when DTC is less than 5 m3/d, the slopes of CC and TC on DTC are constants of 1.923 and 1.809, respectively, with no relationship with RCOD and RNH3-N; (3) when DTC is less than 20 m3/d, the OMC is a constant of 435 RMB/year; and (4) in other cases, CC, OMC, and TC are related to RCOD and RNH3-N besides DTC. Compared with widely used support vector machine (SVM) models and multiple linear regression (MLR) models, the MARS model has better statistical significance with greater R values and smaller RMSE and MAPE values, which indicated that the MARS model is a better way to approximate the cost for DWWTPs.