Identification Method of Source Term Parameters of Nuclear Explosion Based on GA and PSO for Lagrange-Gaussian Puff Model

Yang Zheng, Yuyang Wang, Longteng Wang, Xiaolei Chen, Lingzhong Huang, Wei Liu, Xiaoqiang Li, Ming–Hsuan Yang, Peng Li, Shanyi Jiang, Hao Yin, Xinliang Pang, Yunhui WuAtmosphere 2023 · DOI

TL;DR

This paper introduces a novel method for identifying the crucial "source term parameters" of a nuclear explosion, which are essential for predicting radioactive fallout. By combining a Lagrange-Gaussian puff model with optimization algorithms like Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) and using historical data, the researchers achieved over 98% accuracy in identifying these parameters, offering a significant improvement for real-time nuclear threat assessment.

What Problem Does This Paper Address?

Predicting the dispersion of radioactive particles after a nuclear explosion is critical for emergency response and environmental protection. However, the accuracy of these predictions heavily relies on precise "source term parameters" – information like the explosion's yield, location, and meteorological conditions at the time of detonation. In real-world nuclear strikes, unlike controlled tests, this information is often unknown or uncertain. This lack of accurate input data severely hampers the effectiveness of dispersion models. Existing methods for identifying these parameters are complex and can be prone to errors, especially in the chaotic aftermath of an actual event. This paper aims to bridge this gap by developing a more robust and accurate method for identifying these vital source term parameters.

Methods

The core of this research lies in combining a Lagrange-Gaussian puff model with two powerful swarm intelligence optimization algorithms: the Genetic Algorithm (GA) and the Particle Swarm Optimization (PSO).

  • Lagrange-Gaussian Puff Model: This model simulates the atmospheric dispersion of radioactive particles by tracking individual "puffs" of material as they move and spread with the wind. It's a standard approach for this type of modeling.
  • Genetic Algorithm (GA): Inspired by natural selection, GA iteratively refines a population of potential solutions (source term parameters) by applying operations like crossover, mutation, and selection to evolve towards better solutions.
  • Particle Swarm Optimization (PSO): This algorithm mimics the social behavior of bird flocking or fish schooling. Particles (potential solutions) move through the search space, influenced by their own best-found position and the best-found position of the entire swarm.

The study uses historical nuclear test data to train and validate these algorithms. The process involves:

  1. Inputting initial, uncertain source term parameters into the Lagrange-Gaussian puff model.
  2. Comparing the model's predicted radiation dose with actual measured data.
  3. Using GA and PSO to iteratively adjust the source term parameters to minimize the difference between predicted and measured doses, thereby optimizing the parameters.

The paper also details the selection of specific genetic operators (tournament selection was found to be most effective) and the optimization of parameters for both GA and PSO (like crossover rate, mutation rate, inertia weight, and learning factors) through extensive experiments.

Datasets

The study utilizes publicly available historical U.S. nuclear weapons experimental data from 1945–1962. Specifically, it focuses on the "KOON" nuclear test conducted on April 6, 1954. This dataset includes:

  • Meteorological parameters (time of explosion, explosion equivalent, explosion center coordinates, explosion height, average wind direction and speed).
  • Radiation contour maps one hour after the explosion.

The "true values" of the source term parameters were derived from this historical data, and Gaussian random errors were added to simulate the uncertainty encountered in real-world scenarios. This approach creates a realistic benchmark for evaluating the identification algorithms.

Key Insights and Ideas

  • The critical role of accurate source terms: The paper strongly reinforces that even sophisticated dispersion models are only as good as their input parameters. Inaccurate source terms lead to unreliable predictions, directly impacting emergency response.
  • Swarm intelligence for a complex problem: Applying GA and PSO to the inverse problem of identifying source terms from dispersion data is a powerful approach. It effectively navigates the complex, multi-dimensional search space of potential parameters.
  • PSO's edge in speed and stability: While both GA and PSO performed well, PSO demonstrated a slight advantage in terms of optimization accuracy and, notably, computational speed when compared under similar conditions. This suggests PSO might be more practical for time-sensitive applications.
  • The "KOON" test as a valuable benchmark: Using a well-documented historical nuclear test provides a robust and realistic dataset for validating the proposed identification method, offering a valuable reference for future research.

Key Results

The study achieved highly promising results:

  • High Accuracy: Both GA and PSO successfully identified the nuclear explosion source term parameters with a prediction accuracy of over 98% for the main parameters.
  • PSO's Performance: When run with the same number of iterations and population size, PSO exhibited better optimization accuracy and faster running times compared to GA.
  • Model Agreement: The optimized source term parameters, when fed back into the Lagrange-Gaussian puff model, produced predicted dose values that showed excellent agreement with the observed historical data. For instance, the statistical mean absolute error (MAE) for the PSO-optimized results was approximately $4.042 \times 10^{-8}$, and the mean variance (MSE) was $4.9332 \times 10^{-13}$.

Limitations and Future Work

The authors acknowledge several limitations and propose avenues for future research:

  • Neglecting Model Errors: The current study assumes the Lagrange-Gaussian puff model perfectly represents the physical process. Future work should investigate the impact of model errors on source term identification and explore using multiple dispersion models (e.g., DELFIC, WSEG) for a more comprehensive analysis.
  • Limited Algorithm Scope: Only GA and PSO were explored. The authors suggest investigating a wider range of swarm intelligence algorithms, as each has unique strengths and weaknesses.
  • Algorithm Optimization: Further optimization of the GA and PSO algorithms themselves, beyond parameter tuning, is recommended. This could involve combining them with other techniques like machine learning or neural networks.
  • Real-time Data Integration: While the study uses historical data, a crucial next step would be to test and validate this method with real-time monitoring data from actual or simulated events to assess its practical applicability in emergency scenarios.

Summary updated

Abstract

Many well-established models exist for predicting the dispersion of radioactive particles that will be generated in the surrounding environment after a nuclear weapon explosion. However, without exception, almost all models rely on accurate source term parameters, such as DELFIC, DNAF-1, and so on. Unlike nuclear experiments, accurate source term parameters are often not available once a nuclear weapon is used in a real nuclear strike. To address the problems of unclear source term parameters and meteorological conditions during nuclear weapon explosions and the complexity of the identification process, this article proposes a nuclear weapon source term parameter identification method based on a genetic algorithm (GA) and a particle swarm optimization algorithm (PSO) by combining real-time monitoring data. The results show that both the PSO and the GA are able to identify the source term parameters satisfactorily after optimization, and the prediction accuracy of their main source term parameters is above 98%. When the maximum number of iterations and population size of the PSO and GA were the same, the running time and optimization accuracy of the PSO were better than those of the GA. This study enriches the theory and method of radioactive particle dispersion prediction after a nuclear weapon explosion and is of great significance to the study of environmental radioactive particles.
BibTeX
@inproceedings{zheng2023identification,
  author = {Yang Zheng and Yuyang Wang and Longteng Wang and Xiaolei Chen and Lingzhong Huang and Wei Liu and Xiaoqiang Li and Ming–Hsuan Yang and Peng Li and Shanyi Jiang and Hao Yin and Xinliang Pang and Yunhui Wu},
  title = {{Identification Method of Source Term Parameters of Nuclear Explosion Based on GA and PSO for Lagrange-Gaussian Puff Model}},
  booktitle = {Atmosphere},
  year = {2023},
  doi = {10.3390/atmos14050877},
}
Figure