Noisy Signals in Wastewater Treatment Plants data-driven control: Spectral Analysis approach for the design of ANN-IMC controllers

摘要

Wastewater Treatment Plants (WWTP) are facilities where different control strategies have been deployed to assure that pollutant concentrations accomplish the established regulations. Among these strategies, Internal Model Controllers (IMCs) have been adopted due to their low complexity and easy implementation. Recently, they have been implemented considering Artificial Neural Networks (ANNs) to avoid their dependence on direct and inverse highly complex and nonlinear mathematical models. Besides, their adoption allow the use of the IMC controller in cloud-based systems to decouple the models from the process under control. Here, an ANN-based IMC structure is proposed as a new WWTP control strategy to manage the dissolved oxygen. This solution is able to offer significant improvements w.r.t. the WWTP default controllers when ideal signals are considered. However, in real environments signals are noise-corrupted producing a significant drop in the IMC performance. For that reason, a new methodology based on spectral analyses is proposed to determine certain parameters of the prediction architectures. Results show an improvement in terms of the prediction errors, i.e., the Root Mean Squared Error (RMSE), between a 62% and a 70% when Long Short Term Memory (LSTM) cells implemented with the new methodology are adopted instead of Multilayer Perceptron (MLP) nets.