Mine Water Inrush Sources Online Discrimination Model Using Fluorescence Spectrum and CNN

Mine water inrush disasters served as severe accidents in China cause severe economic losses and threaten the safety of coal mine production. The existing mine water sources discrimination methods have to let miners collect water samples of different place in situ, which make dynamic online analysis virtually impossible. This paper proposes a novel inrush water source discrimination model using laser-induced fluorescence (LIF) technology and convolution neural network (CNN) to achieve mines inrush water source online discrimination which can reduce humankind involvement. Experiment collected 161 items water samples of four different water sources of Xinji No. 2 coal mine. The LIF auto launched 405-nm lasers into water samples to calculate reflected fluorescence spectra. An improved smoothing method is proposed to reduce high-frequency random fluctuations of fluorescence spectra and further to compute auto-correlation fluorescence spectra features. Based on CNN frame and spectra features, mine waters source online discrimination model is constructed. Experiment randomly selected 80 percent of samples of all for training CNN model, the remaining for testing the proposed model. Theoretical analysis and experimental results demonstrate that the recognition rate of the proposed method achieves 98%.This method is an effective assessment method to discriminate inrush water source types of mines. It provides a new train of thought to solve online discriminant inrush water source types of mines.