The real-time estimation of hazardous gas dispersion by the integration of gas detectors, neural network and gas dispersion models

Release of hazardous materials in chemical industries is a major threat to surrounding areas. Current gas dispersion models like PHAST and FLACS, use release velocity, release elevation, meteorological parameters, and other related information as model input. In general, such information is not always available during an on-going accident. In this paper, we develop a fast prediction approach which could bypass the input parameters that are difficult to obtain and predict the released gas concentration at certain off-site location using parameters that could be obtained easily. The new approach is an integration of gas detectors, artificial neural network (ANN) and one of the aforementioned gas dispersion models. PHAST is applied to simulate numbers of release scenarios and the results containing the spatial and temporal distributions of released gas concentration are prepared as input and target data samples for training the neural network. The approach was applied to a case study involving a hypothetical chlorine release with varying release rates and atmospheric conditions. The results of the approach that are concentration and dispersion time profiles in the environmental sensitive locations were validated against PHAST. The validation shows highly correlations with PHAST and convincingly demonstrates the effectiveness of the proposed approach.