Artificial intelligence model for water resources management

Hongwu Tang, Yu Lei, Binliang Lin, Yan Zhou, Zonghua GuProceedings of the Institution of Civil Engineers - Water Management 2010 · DOI

TL;DR

This paper presents an integrated artificial intelligence (AI) system that combines hydrodynamic simulation with an artificial neural network (ANN) to optimize sluice gate operations in complex channel networks like Pudong New District, Shanghai. The system aims to improve water resource management by providing intelligent decision support for water delivery, flood prevention, and drainage, overcoming the limitations of manual operations.

What Problem Does This Paper Address?

Managing water resources in complex channel networks, particularly those with numerous interconnected channels and sluice gates, presents significant challenges. Pudong New District in Shanghai exemplifies this, with its intricate network of 17 main channels and 13 sluice gates. These gates control water flow for various purposes, including flood prevention, drainage, water supply, and navigation, while also impacting water quality and preventing saltwater intrusion. Manual operation of these gates, often relying on experience, is inefficient and prone to suboptimal outcomes. The paper addresses the need for a more intelligent, systematic, and efficient approach to manage these complex water systems, moving beyond empirical methods and complicated calculations.

Methods

The core of the proposed solution is an Artificial Intelligence Modelling System that integrates two key components:

  1. Hydrodynamic Model: This model simulates water flow and levels within the channel network. It is based on the Saint-Venant equations, which describe the conservation of mass and momentum for unsteady, one-dimensional flows. The paper mentions using the four-point linear implicit Preissmann finite-difference method for discretizing these equations and the generalized Newton-Raphson method for solving the resulting non-linear algebraic equations. A key aspect is the handling of looped channel networks and the use of a relaxation algorithm to manage lateral discharges from branches.
  2. Artificial Neural Network (ANN) Model for Sluice Gate Operation: This component uses an Artificial Neural Network (ANN), specifically a Back-Propagation Network (BPN) with a three-layer structure (input, hidden, output), to develop optimal sluice gate operation procedures. The ANN takes various water level parameters (outer and inner river levels, expected levels) and physical characteristics of the gates (width, sill elevation) as input and outputs the opening degree and opening time for each sluice gate. The ANN is trained using historical data and then used to generate operational strategies.

The system is structured with a database to store past successful operations, which serves as training data for the ANN. The hydrodynamic model then simulates the consequences of the ANN-generated operations, providing feedback for decision-making.

Datasets

The paper utilizes two primary sources of data:

  • Database of past successful sluice gate operations: This database, obtained by generalizing past operational schemes, is used to train the ANN models. Specific numbers of training samples are mentioned: 230 for water delivery operations and 200 for water drainage operations.
  • Prototype water transferring experiment data: An experiment conducted on 24 September 2003 in the Pudong channel network provided crucial data for calibrating and validating both the hydrodynamic and ANN models. This included measurements of water levels, flow discharges, and sluice gate opening times and degrees from six gauging stations and ten discharge stations.
  • Existing data from database: For validation, additional data sets were used: 99 sets for water delivery and 95 sets for water drainage operations.

Key Insights and Ideas

  • Integration is Key: The paper strongly advocates for the synergy between hydrodynamic simulation and AI. While hydrodynamic models can predict flow, they don't inherently optimize operations. ANNs can learn complex operational patterns, but they need a simulation model to assess the impact of their decisions.
  • ANNs as Intelligent Operators: The study demonstrates that ANNs can effectively learn and replicate complex operational logic for sluice gates, moving beyond simple empirical rules. This is particularly valuable in dynamic environments like tidal channel networks.
  • Two Distinct Operational Models: The authors correctly identify that sluice gate operations can be broadly categorized into "water delivery" and "water drainage" models, each with specific objectives and requiring tailored ANN architectures and training.
  • Data-Driven Optimization: The success of the system hinges on the quality and quantity of historical operational data. The paper highlights the importance of this data for training robust ANN models.
  • Decision Support System: The developed AI system is not just a predictive tool but a decision-support system, providing engineers and managers with actionable insights for integrated water resource management.

Key Results

  • Hydrodynamic Model Accuracy: The hydrodynamic model demonstrated good accuracy, with predicted water delivery and drainage volumes showing relative errors generally less than 10%. Water level predictions at gauging stations had a maximum deviation of 0.1 m.
  • ANN Model Performance:
    • For water delivery operation, the ANN model achieved a Root Mean Square Error (RMSE) of 0.560 h and a coefficient of determination (R²) of 0.982 during testing.
  • For water drainage operation, the ANN model achieved an RMSE of 1.335 h and an R² of 0.987 during testing.
  • The ANN models successfully predicted measured data, with the expected inner river water level for water delivery being closely matched (2.748 m expected vs. 2.735 m predicted).
  • Flood Control Application: In a flood scenario, the system successfully reduced the average inner river water level to a reasonable range (2.68 m), meeting expected demands and minimizing potential losses.
  • Integrated System: The paper successfully integrated the ANN and hydrodynamic models into an information-based system, demonstrating its practical applicability for real-time water resource management.

Limitations and Future Work

  • Data Dependency: The performance of the ANN models is highly dependent on the quality and representativeness of the training data. If the acquired operation database does not adequately cover all operational demands, the model's performance might be compromised.
  • Generalizability: While successful for Pudong New District, the paper doesn't extensively discuss the generalizability of the model to other, potentially different, channel network configurations or hydrological conditions.
  • Real-time Implementation: While an information-based system is presented, the paper focuses on the modeling and validation aspects. Further work could explore the full real-time implementation and its operational challenges.
  • Future Work: The authors suggest that the developed system can be used as a basis for decision-making in managing water resources for large-scale complex channel networks, implying its potential for broader application. Further research could explore incorporating more complex influencing factors or advanced AI techniques.

Summary updated

Abstract

The channel network in Pudong New District, Shanghai, is very complex owing to the large area of its basin, its numerous sluice gates, complex influencing factors and some other management issues involving water delivery, flood prevention, floodwater drainage, navigation and saltwater intrusion. It is generally difficult to achieve efficient water resources management merely through manually operating the sluice gates. Therefore, an artificial intelligence modelling system for managing the water resources in the channel network of Pudong New District has been developed by combining hydrodynamic simulation with an artificial intelligence technique. The artificial neural network model is used to develop sluice gate operation procedures according to the water levels in both the outer and inner rivers. The hydrodynamic model is used to simulate the flow discharges and water levels based on the sluice gate operation procedures. This modelling system has been applied successfully to the water resources management of the Pudong channel network. The results indicate that the modelling system satisfactorily meets the demands for sluice gate operation and water resources optimisation management of the channel network and thus provides decision-making support for integrated management of water resources in this inland channel network.
BibTeX
@inproceedings{tang2010artificial,
  author = {Hongwu Tang and Yu Lei and Binliang Lin and Yan Zhou and Zonghua Gu},
  title = {{Artificial intelligence model for water resources management}},
  booktitle = {Proceedings of the Institution of Civil Engineers - Water Management},
  year = {2010},
  doi = {10.1680/wama.2010.163.4.175},
}
Figure