Development of a Neural Network Algorithm of Making Decision on Emergency and Pre-Emergency Situations in the Coal Mine for Multicriterial Electro-Optical System

Abstract

Development of a neural network algorithm of making decision on emergency or pre-emergency situation in the coal mine by the multi-criterial electro-optical system is executed. The mathematical formulation of a neural network algorithm which includes the following stages is made: data reduction from the electro-optical sensors (EOS); normalization of signals from EOS; prediction of gases and dust concentration values with use of neural networks; making decision on the arisen situation by comparison of the predicted values with threshold. Influence of suspended coal dust concentration on a methane concentration when performing a pretreatment of EOS signals is considered. Influence of unwanted gases on EOS indications (cross sensitivity) is considered. Normalizing of signals is carried out for range from – 1 up to 1. This range corresponds to the entrance range of the neurons activation function of a hyperbolic tangent. Autoregressive neural networks of the NAR and NARX types were used for prediction of gases and dust concentration values. The research for each of neural networks for the purpose of determination of architecture, which provides the best accuracy of prediction, is executed. This research is executed by change of the following parameters: quantity of neurons on average a layer – from 5 to 11; number of time counting in a vector of the return delay – from 30 to 150. Recommendations about creation of neural networks architecture for realization of situations definition algorithm in the coal mine are as a result formulated: number of neurones in the hidden layer 7 – 11, a delay more than 20% of total number of sample for training.