An Improved Adaptive PSO-LSTM Algorithm for the Gas Leakage Source Localization

Gas leakage source localization is a matter of utmost security concern. However, the conventional mechanism model fails to accurately depict the actual diffusion environment during a leak, leading to diminished accuracy and substantial deviations in source estimation. To overcome this limitation and bolster the precision and reliability of gas leakage source localization, we introduce an improved adaptive PSO-LSTM algorithm. By leveraging field detection data obtained from ABB's Ability™ high-precision gas leakage detection system, we enhance the conventional particle swarm optimization algorithm (PSO) by fine-tuning learning factors and other critical parameters. Moreover, to harness the profound interpretability of the mechanism model and the potent learning capabilities of the LSTM neural network (data model), we substitute the fitness function of the traditional Gaussian plume diffusion model with training outcomes from the LSTM neural network, employing a sizable dataset. Empirical findings underscore that our refined algorithm attains superior accuracy and robustness in estimating source locations.