Prediction and Detection of Sewage Treatment Process Using N-BEATS Autoencoder Network

Abstract

Effective processing of the massive amounts of information generated by a sewage treatment plant's purification process helps reduce the operating costs of sewage purification while enhancing both control over and the reliability of the purification process. To predict multivariate time series data at sewage treatment plants, we propose a method based on neural expansion analysis for time series forecasting. In addition, we offer a method based on an N-BEATS autoencoder network that combines seasonality analysis with a class of support vector machine algorithms to detect data anomalies in sewage treatment. We also validate the proposed method and compare it with other mainstream machine learning and statistical methods. The results show that the proposed prediction and anomaly detection methods outperform other methods. The prediction results are improved by the highest to 22% compared with the other methods, while the accuracy of anomaly detection, 98%, is also highest among all methods tested. Moreover, the model is more scientific and flexible, with systematic potential and significance.