Novel Machine Learning-Based Energy Consumption Model of Wastewater Treatment Plants

Abstract

Wastewater treatment plants (WWTPs) can account for up to 1% of a country’s energy consumption. Meanwhile, WWTPs have high energy-saving potential. To achieve this, it is necessary to establish appropriate energy consumption models for WWTPs. Several recent models have been developed using logarithmic, exponential, or linear functions. However, the behavior of WWTPs is non-linear and difficult to fit with simple functions, particularly for non-numerical variables. Thus, traditional modeling methods cannot effectively describe the relationship between water and energy in WWTPs. Therefore, a machine learning method was adopted in this study to investigate the energy consumption in WWTPs; a novel energy consumption model with a non-numerical variable (discharge standard) for WWTPs was developed using the random forest algorithm. The model can also predict the energy consumption of WWTPs after upgrading discharge standards. We found that the unit electricity consumption of WWTPs exhibited an average increase of 17% after the effluent discharge standard was increased from class I B to class I A (as per China’s classification). The correlation coefficient of the model was 0.702. Thus, the developed model can provide a better understanding of energy efficiency in WWTPs.

Application of Machine Learning-based Energy Use Forecasting for Inter-basin Water Transfer Project

“…We also found that the DNN model, an ANN model with multiple hidden layers between the input and output layers, was the most effective machine learning algorithm in predicting the energy use of the water distribution system. Other studies found that the RF model effectively predicted energy consumption for wastewater treatment plants (Zhang et al 2021). We concluded that DNN and SVR models performed better than the RF model for the treatment system (in the inter-basin water transfer project), which was not the same as the wastewater treatment plant.…”

“…The California Energy Commission, the primary energy planning agency in California, has been using regression analysis to forecast energy use in agricultural and municipal water pumping (California Energy Commission 2005). Machine learning-based models have also proven to be useful in simulating the energy use for a wastewater treatment plant (Bagherzadeh et al 2021;Das, Kumawat, and Chaturvedi 2021;Li and Tang 2021;Zhang et al 2021) and a distribution system (Salvino, Gomes, and Bezerra 2022). Various studies have applied machine learning algorithms in forecasting water-related energy demand but were limited to forecasting the energy use for a single water facility.…”

AI and Big Data in Water Environments

“…Finally, data sparsity is a common issue in water research, as demonstrated in almost all articles in this virtual issue. For example, there are only from <100 to 2387 experimental or literature data points in the training data sets. ,, Techniques that can handle sparse data and efforts to significantly increase the sample size would greatly help improve the model quality …”

“…1 In this virtual issue, three articles reported ML models for water and wastewater treatment processes. Zhang et al 2 modeled the energy consumption (unit electricity consumption) in wastewater treatment plants (WWTPs) using the random forest algorithm. The input variables included design treatment capacity, annual average load rate, and removal ratios between the influent and effluent (BODi and BODe, CODi and CODe, and NH 3 -Ni and NH 3 -Ne).…”

“…For example, there are only from <100 to 2387 experimental or literature data points in the training data sets. 2,4,5 Techniques that can handle sparse data and efforts to significantly increase the sample size would greatly help improve the model quality. 1 Overall, this Virtual Issue reflects only the beginning of the applications of AI and big data in water-related research.…”