Prediction model of sparse autoencoder-based bidirectional LSTM for wastewater flow rate

Area of Science:

• Environmental Engineering
• Water Resource Management
• Machine Learning Applications

Background:

• Sanitary sewer overflows (SSOs) due to excessive rainfall infiltration and inflow pose significant challenges for municipal administrations.
• Accurate prediction of wastewater flow rates is essential for effective management and mitigation of SSOs.

Purpose of the Study:

• To introduce and evaluate a novel Sparse Autoencoder-based Bidirectional long short-term memory (SAE-BLSTM) network model for predicting wastewater flow rates in sanitary sewer systems.
• To demonstrate the model's superiority over existing machine learning techniques in wastewater flow prediction.

Main Methods:

• The proposed SAE-BLSTM model integrates Sparse Autoencoder (SAE) for feature extraction and dimensionality reduction with Bidirectional long short-term memory (BLSTM) for time series prediction.
• SAE extracts sparse potential features from high-dimensional input data, which are then combined with historical flow rate data to form an augmented feature vector.
• This augmented vector is fed into the BLSTM network for accurate future wastewater flow rate prediction.

Main Results:

• The SAE-BLSTM model demonstrated superior performance compared to Support Vector Machine (SVM), Fully Convolutional Network (FCN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and BLSTM models.
• The model achieved the lowest Root Mean Square Error (RMSE) of 242.55 and Mean Absolute Error (MAE) of 179.05.
• The highest coefficient of determination (R²) of 0.99626 was obtained, indicating a highly accurate prediction of wastewater flow rates.

Conclusions:

• The developed SAE-BLSTM model effectively predicts wastewater flow rates by combining SAE's feature representation capabilities with BLSTM's time series forecasting power.
• This approach offers a significant advancement in managing sanitary sewage systems and mitigating challenges associated with infiltration and inflow.
• The model's high accuracy provides a valuable tool for municipal administrations to improve operational efficiency and environmental protection.