A Multivariable Probability Density-Based Auto-Reconstruction Bi-LSTM Soft Sensor for Predicting Effluent BOD in Wastewater Treatment Plants

Sensors, Год журнала: 2024, Номер 24(23), С. 7508 - 7508

Опубликована: Ноя. 25, 2024

The precise detection of effluent biological oxygen demand (BOD) is crucial for the stable operation wastewater treatment plants (WWTPs). However, existing methods struggle to meet evolving drainage standards and management requirements. To address this issue, paper proposed a multivariable probability density-based auto-reconstruction bidirectional long short-term memory (MPDAR-Bi-LSTM) soft sensor predicting BOD, enhancing prediction accuracy efficiency. Firstly, selection appropriate auxiliary variables soft-sensor modeling determined through calculation k-nearest-neighbor mutual information (KNN-MI) values between global process BOD. Subsequently, considering existence strong interactions among different reaction tanks, Bi-LSTM neural network model constructed with historical data. Then, multivariate (MPDAR) strategy developed adaptive updating model, thereby its robustness. Finally, effectiveness demonstrated experiments using dataset from Benchmark Simulation Model No.1 (BSM1). experimental results indicate that not only outperforms some traditional models in terms performance but also excels avoiding ineffective reconstructions scenarios involving complex dynamic conditions.

Язык: Английский

Effluent Quality Soft Sensor for Wastewater Treatment Plant with Ensemble Sparse Learning-Based Online Next Generation Reservoir Computing

Water Research X, Год журнала: 2024, Номер 25, С. 100276 - 100276

Опубликована: Ноя. 10, 2024

Real-time monitoring of key quality variables is essential and crucial for stable safe operations wastewater treatment plants (WWTPs). Next generation reservoir computing (NG-RC) has recently garnered significant attention in prediction, such as COD BOD, an effective alternative to traditional (RC), then able act a data-driven soft sensor twin hardware variable measurements. Unlike RC, NG-RC does not require random sampling matrices define the weights recurrent neural networks fewer hyperparameters. However, usually used online but trained offline, thus leading model degradation under dynamic scenarios. This paper proposes sparse approach meet real-time requirements WWTPs mitigate impact measurement noise on model. First, inspired by Woodbury matrix identity, incremental strategy designed, using sequentially arriving data blocks learn output online. Then, ensemble combined alleviate overfitting issues prediction Moreover, based developed perform indicators processes. Finally, two datasets from actual are validate effectiveness proposed

Язык: Английский

Towards sludge bulking diagnosis via dynamic attention graph neural network

Journal of Water Process Engineering, Год журнала: 2024, Номер 69, С. 106774 - 106774

Опубликована: Дек. 12, 2024

Язык: Английский