Chemical Characteristics Of Groundwater And Discriminate Models Of Water Bursting Source Based On Elman Neural Network: A Case Study On Xieyi Coalmine

Discriminating the source of coal mine water bursting source quickly and correctly is the precondition of water control. This article chooses Xieyi Coalmine in Huainan as the research area, collects and analyses the hydro-geological data of the past years, studied water chemical types of different aquifers in the mine of ion and water mine sudden rapid identification of dynamic regression Elman neural network model and Elman model with the BP model and the fuzzy comprehensive evaluation model, gray correlation model, Bayesian model comparison, the main conclusions drawn are as follows(1) The main water-type in the coal aquifer is HCO3-K+Na, and those in the Taiyuan limestone aquifers (GroupⅠand GroupⅢ)and the Ordovician limestone aquifer are HCO3-Ca·Mg and HCO3-Ca·Mg. The water samples from the groundwater in mined areas are few, in which the main water-type is HCO3·SO4-(K+Na). Analyzing the overall characteristics of the aquifer water quality of the aquifer, through the contrast map ion concentration, it is found that coal water has low Ca2+,Mg2+ contents, water of Taiyuan limestone aquifer has higher Ca2+,Mg2+,K++Na+,HCO3- contents than water of the Ordovician limestone aquifer, and the highest HCO3- and SO42- contents are observed in the groundwater in mined areas. On the analysis of the distribution of ions in different aquifers by box charts, it is found that the groundwater in mined areas can be distinguished easily from water of other three aquifers. After testing one water sample, if its [TDS]> 1500mg / L and [SO42-] and [K++Na+] and [HCO3-]> 300mg / L, it can be considered as the groundwater in mined areas. The result shows the discrimination accuracy is more than 95%. If its [Ca2+]<50mg/L and [Mg2+]<20mg/L, it must come from the coal aquifer and the discrimination accuracy is more than 60%.(2) Through using the traditional model to discriminating water bursting source with water samples, we find determine effect of Bayesian model is better than others. The determine effect of over standard weighting in fuzzy comprehensive evaluation is poorer than others. The determine effect of over standard weighting is obvious poorer than biasing weighting because the over standard weighting distribute the value which more than the average value large weight. Less than the average value is distributed small weight. That's not appropriate in discriminating water bursting source. On the whole, no one incorrectly judged in samples of the coal aquifer. The water quality of Taiyuan limestone aquifer and Ordovician limestone aquifer is very close. Taiyuan limestone aquifer and Ordovician limestone aquifer have strong hydraulic link, so some water samples are difficult to distinguish because mixed with other layers'water.(3) Establishes the distinguishing model for water bursting by Elman neural networks and BP neural networks with groundwater chemical characteristics, respectively. Experimental results show that, the determine effect of neural network model is better than the traditional model. One of 11 samples is incorrectly judged by Elman network model. The Elman neural model is more precise and faster than BP neural model in discrimination. The Elman neural model could better response characteristics of groundwater systems. It provides an assistant means for decision-making to prevent water-inrush from coal floor.