Research On Gas Emission Dynamic Prediction Based On ASGSO-ENN Algorithm

The traditional nonlinear methods cannot construct a model to reliably and accurately predict the gas emissions for the engineering applications because the gas emission system is highly nonlingar, complex and time-varying. To address this problem, The proposed self-adaptive step glowworm swarm optimization (ASGSO) algorithm is combined with dynamic feedback Elman neural network (ENN) to perform identification of the non-linear gas emission system effectively.Due to its own dynamic feedback and recursion functions, the Elman neural network provides great time-varying data processing ability and network stability. Hence, it can directly better represent the dynamic properties of the essential non-linear system. To solve the problems of slow convergence, inaccuracy and inefficiency that the network training process suffers, the glowworm intelligent algorithm is used to optimize the process. Next, the bionics principles and the optimization procedures of the basic GSO algorithm are analyzed. The analysis shows that the GSO algorithm delivers great local search performance and is easy to operate and implement. To strengthen the GSO algorithm's global optimization performance, the target neighborhood set based on the similarity criterion is redefined. The old rough estimate of the initial perceived radius is replaced with the initially pre-determined accurate threshold of the similarity. After each evolution of the population, the moving step length will be self-adapted according to the density of the distribution between the individuals and the excellent individuals which in the target neighborhood set in order to prevent individuals from generating the oscillation around the extreme points during the search.The online fast and accurate search for the ENN weights and thresholds in the solution space is done due to the great global multi-target search ability of the enhanced ASGSO algorithm. With the predictive control theory, The prediction of gas emission control system is proposed by using the ASGSO-ENN coupling algorithm. The prediction experiments are performed with the monitored history-data of the pits. The results demonstrate that in the context of higher learning efficiency, it greatly outperforms the pure Elman neural network, the GSO-ENN coupling model and the BP neural network commonly used in the engineering in terms of prediction accuracy and generalization. Furthermore, the proposed method is highly robust and thus theoretically helpful to the prevention and relief of the gas disasters.