The traceability of sudden water pollution in river canals based on the pollutant diffusion quantification formula

Fei Lin, Honglei Ren, Yuezan Tao, Naifeng Zhang, Yucheng Li, Rujing Wang, Yimin HuFrontiers in Environmental Science 2023 · DOI

TL;DR

This paper introduces a high-speed traceability model designed to pinpoint the source of sudden river pollution by combining a simplified mechanistic diffusion formula with a modified optimization algorithm. By decoupling complex parameters and using a "Beetle Antennae Search" with dynamic step sizes, the researchers achieved high-accuracy results (errors < 2.4%) significantly faster than traditional genetic algorithms.

What Problem Does This Paper Address?

When a sudden pollution event occurs in a river—such as a chemical spill or illegal discharge—emergency responders must immediately identify three things: where it happened, when it happened, and how much was released.

This is known as an "inverse problem." Traditionally, solving it requires complex hydrodynamic models that are computationally expensive and prone to getting stuck in "local optima" (finding a solution that looks good mathematically but isn't the actual source). Furthermore, the parameters of location, time, and mass are usually "coupled," meaning they influence each other in the equations, making the search for a solution a high-dimensional nightmare that is too slow for real-time emergency response.

Methods

The authors propose a two-pronged approach to simplify and speed up the search:

  1. Pollutant Diffusion Quantification Formula: Instead of relying on heavy differential equations for every step, they derived a simplified formula based on the law of pollutant diffusion and normal distribution characteristics. This allows them to "decouple" the parameters, effectively turning a multi-dimensional search into a series of simpler, single-parameter problems.
  2. DP-BAS Algorithm: They utilized the Beetle Antennae Search (BAS), an optimization meta-heuristic inspired by how longhorn beetles find food using their antennae. To improve it, they integrated Dynamic Programming (DP) and a variable step size. This allows the "beetle" to take large leaps initially to explore the search space and smaller, precise steps as it nears the optimal solution, preventing it from getting stuck in local mathematical traps.

Datasets

The model was evaluated using a controlled simulation case rather than a static historical dataset.

  • Scenario: A rectangular river canal (20m wide, 6m deep) with a simulated 1,000 kg pollutant spill.
  • Data Collection: Monitoring data was "collected" 5km downstream every 30 minutes.
  • Noise: To simulate real-world conditions, the researchers added a 10% observation error to the calculated results to account for instrumental monitoring inaccuracies.
  • Benchmarks: The results were compared against two industry-standard methods: Genetic Algorithms (GA) and Differential Evolution-Markov Chain Monte Carlo (DE-MCMC).

Key Insights and Ideas

  • Dimensionality Reduction: The most significant "aha moment" is the decoupling of parameters. By transforming a multi-parameter coupled solution into a single-parameter one, the researchers reduced the computational "friction" that usually slows down these models.
  • The Power of Biology: The BAS algorithm is remarkably efficient for this type of search because it requires only one "beetle" (search agent) rather than a whole "population" (like in GA), which saves memory and processing time.
  • Stability through Iteration: The study found that while the first 1,500 iterations of the algorithm were highly unstable, the results consistently converged and stabilized around the true values after that point, proving the model's reliability even with noisy data.

Key Results

The proposed DP-BAS model outperformed both GA and DE-MCMC in both accuracy and efficiency:

  • Accuracy (Mean Square Error):
    • Pollutant Mass: 2.39%
    • Location: 1.16%
    • Time: 1.19%
  • Speed: The total calculation time for the DP-BAS model was 1,152 seconds, compared to 1,602 seconds for GA and 1,817 seconds for DE-MCMC.
  • Reliability: Even with a 10% artificial error added to the sensor data, the model successfully traced the source, suggesting it is robust enough for real-world sensor noise.

Limitations and Future Work

While the results are promising, the authors acknowledge several areas for improvement:

  • Simplified Geometry: The simulation used a "regular rectangular cross-section." Real-world rivers have complex geometries, bends, and varying depths that affect diffusion.
  • Tributaries: The current model assumes no tributary confluence. Future work needs to address how the model handles the merging of different water bodies.
  • Generalizability: The researchers suggest that more in-depth studies are needed to test the model across different types of river systems and more complex, multi-source pollution scenarios.

Summary updated

Abstract

For the problem that the traceability parameters of sudden water pollution are difficult to determine, a fast traceability model based on a simplified mechanistic model coupled with an optimization algorithm is proposed to improve the accuracy of sudden water pollution traceability. In this paper, according to the diffusion law of pollutants, a quantitative formula of pollutant diffusion is proposed, and the differential calculation process of the pollutant convection equation is optimized. The Dynamic Programming and Beetle Antennae Search algorithm (DP-BAS) with dynamic step size is used in the reverse optimization process, which can avoid the problem of entering the local optimal solution in the calculation process. The DP-BAS is used to inverse solve the quantization equation to realize the decoupling of pollutant traceability parameters, transforming the multi-parameter coupled solution into a single-parameter solution, reducing the solution dimension, and optimizing the difficulty and solution complexity of pollutant traceability. The proposed traceability model is applied to the simulation case, the results show that the mean square errors of pollutant placement mass, location, and time are 2.39, 1.16, and 1.19 percent, respectively. To further verify the model reliability, the Differential Evolution and Markov Chain Monte Carlo simulation method (DE-MCMC) as well as Genetic Algorithms (GA) were introduced to compare with the proposed model to prove that the model has certain reliability and accuracy.
BibTeX
@inproceedings{lin2023traceability,
  author = {Fei Lin and Honglei Ren and Yuezan Tao and Naifeng Zhang and Yucheng Li and Rujing Wang and Yimin Hu},
  title = {{The traceability of sudden water pollution in river canals based on the pollutant diffusion quantification formula}},
  booktitle = {Frontiers in Environmental Science},
  year = {2023},
  doi = {10.3389/fenvs.2023.1134233},
}
Figure