A Soft Sensor Based on Influent Mode Discrimination Neural Network for A Wastewater Treatment Process

Five-day biochemical oxygen demand (BOD5) is one of the key parameters which is widely used to evaluate the quality of the effluent treated wastewater in activated sludge model number 1 (ASM1). This paper proposed an influent mode discriminator based on nonlinear autoregressive with exogenous inputs (IMD-NARX) neural networks that can predict BOD5 online. This is useful especially in time constraint applications. In IMD-NARX network, a method of constructing IMD was proposed. First, the dimension of the original sample data was reduced by the principal component analysis (PCA) algorithm. Second, the k-means algorithm combined with the dynamic time wrapping (DTW) distance was used to unsupervisedly generate three types of modes from continuous data sampled from three types of weather condition respectively, namely, dry, rainy and stormy. In the mode-discrimination process, based on the three influent modes that were previously generated, the K-Nearest Neighbor (KNN) algorithm was employed to discriminate the influent mode of the each sampled data. Finally, a nonlinear autoregressive with exogenous inputs (NARX) network was designed to train and predict the BOD5 value based on discriminated data. Experiments showed that IMD can reasonably characterize the influent mode of continuous inflow, and IMD-NARX can quickly and accurately predict the value of BOD5 online in real time.