Contaminant dispersion prediction and source estimation with integrated Gaussian-machine learning network model for...

Area of Science:

• Environmental Science
• Computational Chemistry
• Chemical Engineering

Background:

• Accurate prediction of contaminant gas concentrations during leakage is crucial for safety and environmental monitoring.
• Traditional intelligent network models like RBF, BP, and SVM show limitations in prediction accuracy when using numerous input parameters.
• Existing models struggle to achieve close agreement with experimental data in complex gas dispersion scenarios.

Purpose of the Study:

• To develop and evaluate novel machine learning algorithms (MLA) integrated with the classic Gaussian dispersion model for enhanced gas dispersion prediction.
• To assess the performance of these new Gaussian-MLA models against traditional methods in terms of accuracy and computational efficiency.
• To apply the optimized Gaussian-MLA models for identifying emission source parameters using particle swarm optimization (PSO).

Main Methods:

• Integration of classic Gaussian dispersion model with various machine learning algorithms (MLA), including Support Vector Machine (SVM).
• Comparative analysis of the predictive performance of Gaussian-MLA models against traditional network models and the classic Gaussian model.
• Application of Gaussian-MLA models within a particle swarm optimization (PSO) framework for emission source parameter identification.

Main Results:

• The novel Gaussian-MLA models significantly improved prediction accuracy compared to traditional network models using original monitoring parameters.
• The Gaussian-SVM model demonstrated superior performance, achieving prediction results close to the classic Gaussian dispersion model with comparable computation time.
• The PSO method combined with Gaussian-MLA showed enhanced estimation performance for emission source parameters compared to other dispersion models.

Conclusions:

• Gaussian-MLA models offer a significant advancement in predicting contaminant gas dispersion and identifying emission sources.
• The integration of Gaussian models with machine learning provides a robust and accurate forward model for environmental and safety applications.
• This approach holds potential for improving real-world gas leakage monitoring and source localization strategies.