Sliding window neural network based sensing of bacteria in wastewater treatment plants

Ensuring the performance of wastewater treatment processes is important to guarantee that final treated quality safe for reuse. However, bacterial concentration present along different stages process not easily measured routinely plant operators. In this paper, a moving horizon sensing approach based on neural networks proposed estimate in sampled plant. Due difficulties measure bacteria and lack sufficiently large dataset, Wasserstein generative adversarial network (WGAN) designed generate synthetic data. The critic loss computed held-out validation set evaluate WGAN. Then, generated data used train long short term memory (LSTM) developed predict biomass update LSTM weights by sliding window learning approach. Two datasets WWTP are test method: first, effluent concentrations simulated using benchmark simulation model no.1 (BSM) membrane bioreactor (MBR), where three weather profiles influent were considered then, from MBR at King Abdullah University Science Technology (KAUST). Finally, prediction results indicate WGAN successfully generates realistic samples network. addition, estimation method compared with multilayer perceptron (MLP-NN). Results showed improves MLP-NN.