Dong et al. (2025) Prediction of water consumption and affecting factor analysis using Inception-V4 network and enhanced single candidate optimization: a case study

Identification

• Journal: Applied Water Science
• Year: 2025
• Date: 2025-12-11
• Authors: Junlei Dong, Fei Li, Jiongchen Kou, Zaihui Cao, Mehdi Asadi
• DOI: 10.1007/s13201-025-02697-7

Research Groups

• College of Modern Information Technology, Henan PolyTechnic, Zhengzhou, China
• Henan Engineering Research Center of Fault-Tolerant Server, Zhengzhou, China
• Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China
• University of Chinese Academy of Sciences, Beijing, China
• Academic Department, Shanghai Wantyoung Education, Shanghai, China
• College of Art and Design, Zhengzhou University of Aeronautics, Zhengzhou, China
• College of Technical Engineering, The Islamic University, Najaf, Iraq

Short Summary

This study developed a novel hybrid ESCO-Inception-V4 model for long-term water consumption forecasting in Shanghai, demonstrating superior accuracy (RMSE 0.0519, MAE 0.0407) compared to benchmark models and identifying key influencing factors. The model predicts a steady increase in water demand for Shanghai over the next 15 years.

Objective

• To develop a highly accurate and reliable long-term water consumption forecasting model using a hybrid Inception-V4 deep learning architecture combined with the Enhanced Single Candidate Optimization (ESCO) algorithm.
• To identify and analyze the important socioeconomic and climatic factors influencing regional water consumption.
• To benchmark the proposed ESCO-Inception-V4 model against other state-of-the-art forecasting methods.

Study Configuration

• Spatial Scale: Shanghai, China (case study)
• Temporal Scale:
  • Data collection period: 2001–2023
  • Training period: 2001–2018
  • Testing period: 2019–2023
  • Forecasting period: 2024–2038 (15 years)

Methodology and Data

• Models used:
  • Primary model: Inception-V4 (IV4) deep convolutional neural network.
  • Optimization algorithm: Enhanced Single Candidate Optimization (ESCO).
  • Benchmark optimization algorithms (combined with Inception-V4): Butterfly Optimization Algorithm (BOA), White Shark Optimizer (WSO), Supply-Demand-Based Optimization (SDO), Harris Hawks Optimization (HHO).
• Data sources:
  • Shanghai Municipal Water Authority and the National Bureau of Statistics of China.
  • Data type: Yearly water consumption (dependent variable, in billion cubic meters) and 13 influencing factors (independent variables) from 2001 to 2023.
  • Influencing factors: Population, urbanization rate, gross domestic product (GDP), industrial output, agricultural activity, precipitation, average temperature, water price, per capita income, and other socioeconomic and climatic conditions.
  • Preprocessing: Min-Max scaling for normalization, linear interpolation for missing values.

Main Results

• The ESCO-Inception-V4 model achieved superior predictive performance with a Root Mean Square Error (RMSE) of 0.0519 and a Mean Absolute Error (MAE) of 0.0407.
• This performance was significantly better than benchmark models (WSO-IV4, SDO-IV4, BOA-IV4, HHO-IV4), which had RMSE values ranging from 0.117 to 0.2279 and MAE values from 0.0937 to 0.191.
• The model demonstrated strong generalization ability with an R-squared (R²) value of 0.93 for the training dataset and 0.81 for the testing dataset.
• Key influencing factors on water consumption were identified: Total Industrial Value Added (81% influence), Industrial Value Added (73% influence), Total water resources (48% influence), Effective Integrated Area (46% influence), and Total Food (23% influence).
• The 15-year forecast (2024–2038) for Shanghai indicates a steady increase in annual water consumption, rising from 2.33 billion cubic meters in 2024 to a projected 2.89 billion cubic meters in 2038, representing an approximate 24% increase.

Contributions

• Novel application of the Enhanced Single Candidate Optimization (ESCO) algorithm to dynamically optimize the parameters of a deep Inception-V4 neural network for water consumption forecasting, improving hyperparameter tuning, avoiding local minima, and accelerating convergence.
• Development of a robust ESCO-Inception-V4 hybrid model that demonstrates superior prediction accuracy, stability, and efficiency compared to other metaheuristic-optimized Inception-V4 models.
• Quantification and analysis of the major socioeconomic and environmental drivers of water use in Shanghai, providing actionable insights for regional water resource policy and planning.
• Provision of a validated, data-driven tool for sustainable long-term water resource planning and management, specifically applied to the complex urban environment of Shanghai.

Funding

• Scientific and technological key project in Henan Province, P. R. China (No. 222102240049)
• Alliance of International Science Organizations (ANSO-CRKP-2020-02)
• Key Research Program of the Innovation Academy for Green Manufacture, Chinese Academy of Sciences (IAGM-2019-A16-1)
• Strategic Priority Research Program of the Chinese Academy of Sciences (XDA2003020302)

Citation

@article{Dong2025Prediction,
  author = {Dong, Junlei and Li, Fei and Kou, Jiongchen and Cao, Zaihui and Asadi, Mehdi},
  title = {Prediction of water consumption and affecting factor analysis using Inception-V4 network and enhanced single candidate optimization: a case study},
  journal = {Applied Water Science},
  year = {2025},
  doi = {10.1007/s13201-025-02697-7},
  url = {https://doi.org/10.1007/s13201-025-02697-7}
}