MLP neural networks for accelerated pollution prediction in environmental monitoring systems

Abstract Environmental monitoring and security systems rely on efficient and accurate prediction of contaminant source locations from sensor data. However, modeling atmospheric dispersion often requires significant computational resources. To address this problem, utilizing a Multi-Layer Perceptron (MLP) Neural Network for faster prediction of sensor readings based on source coordinates is a possible alternative approach. Hyperparameters for the MLP are optimized using a Grid Search method, aiming to identify the most suitable configuration for improved performance. We generated synthetic data via the 2D advection–diffusion equation solution with the Finite Differences Method. Verification of the synthetic data integrity is conducted using the Finite Elements Method. The use of an MLP Neural Network model provides precise predictions in a fraction of the time taken by usual approaches. By enabling quick and accurate contaminant dispersion simulations, this approach holds promise for enhancing environmental monitoring and security systems, thereby safeguarding public health.