Data-Driven Hybrid Model for Forecasting Wastewater Influent Loads Based on Multimodal and Ensemble Deep Learning

With the advent of the Industry 4.0 and the introduction of smart technologies in wastewater treatment plants (WWTP); forecasting influent loads is essential to regulate exaggerated operational strategies for WWTP. However, due to various water usage and sources, it is challenging to forecast the fluctuating influent loads. To deal with highly nonlinear and temporal-correlated characteristics of influent loads, in this article, we proposed hybrid influent forecasting model based on multimodal and ensemble-based deep learning (ME-DeepL). The proposed ME-DeepL forecasting model combines strength of multiple deep-learning algorithms in ensemble-learning architecture to handle propagated intrinsic sublayers by empirical mode decomposition of influent loads. Then, the proposed model was assessed to predict the loads on long-term (daily), and short-term (hourly) with multisteps forecast horizons. The experimental results revealed that ME-DeepL model exhibited superior forecasting performance in comparison to five recurrent neural network-based reference models, by capturing the informative features and temporal patterns from fluctuating influent loads.

Automated summary

This article introduces a hybrid influent forecasting model based on multimodal and ensemble-based deep learning, showing superior performance in predicting WWTP influent loads compared to five recurrent neural network-based reference models.