Application Research Of Neural Network In Gas Spectrographic Sensing System

The coal mine gas monitoring is the most important for our country current coal mine safety production supervision management. The development of real-time gas disaster monitoring technology and equipment for preventing gas accidents is an important line of defense and safeguard measures. Using laser Spectrographic combined with neural network for early warning of gas is a brand-new method, which suits the early warning requirements of online, real-time, the fast system of coal mine gas monitoring in our country. It makes this study some standing in the world and establishing the feasible early warning model, and then it can alleviate the aspect that our country coal production megaton mortality rates are much higher than the international standard.First, this article discusses coal mine gas early warning model and correlation theories which are based on tunable diode laser absorption spectrum technology. From theoretical analysis of the near-infrared selective absorption of gas, it gives the gas absorption survey theory basis and determined the methane absorption lines. Controlled by circuit, it uses DFB lasers to scan a vibration-rotation absorption line at 1.65μm nearby of methane. Through examining of the absorption of methane 2υ3 band R (3) line, it surveys the methane gas density. With wavelength modulation, it studies the optimum parameter. Detecting the second-harmonic to obtain a higher SNR, it sets up inversion method of spectrum data.Next, the article introduces some correlation theories of neural network and the BP algorithm, and proposes the methods of gas density analysis and forecast modeling using the BP neural network. Neural network can learn from the historical density data and discover the inherent changing laws, and store in the network specific weights and bias values to predict the future data. Using three-layered feed forward neural network, it sets up gas density prediction model and discusses various parameters setting problems of network. Using Matlab to do simulation experiments, it shows that BP neural network is feasible for early warning of gas density.Finally, from the optimization perspective of hardware structure, with comparative studying hardware realizations of sigmoid function, this article discovers the reasonable design proposal. Using EDA techniques and top-down design methodology, it achieves the BP neural network algorithm in hardware by FPGA technology.