基于改进的SSA−BP神经网络的矿井突水水源识别模型研究

Research on the recognition model of mine water inrush source based on improved SSA-BP neural network

  • 摘要: 机器学习与寻优算法的结合在矿井突水水源识别上得到广泛应用,但突水水样数据具有随机性且寻优算法易陷入局部最优,提高模型泛化能力和跳出局部最优需进一步研究。针对上述问题,提出了一种改进的麻雀搜索算法(SSA)优化BP神经网络模型,用于对矿井突水水源进行定量辨识。以鲁能煤电股份有限公司阳城煤矿为研究对象,通过常规离子浓度分析、Piper三线图对该煤矿水样的水化学特征进行分析,初步判断矿井水来源于奥灰含水层和三灰含水层,并确定Na++K+浓度、Ca2+浓度、Mg2+浓度、\mathrmHCO_3^- 浓度、\mathrmSO_4^2- 浓度、Cl浓度、矿化度、总硬度、pH值作为突水水源识别指标;建立基于改进SSA−BP神经网络的矿井突水水源识别模型:首先进行SSA参数设置,引入Sine混沌映射使麻雀种群均匀分布,然后通过计算适应度值进行麻雀种群的更新,引入随机游走策略扰动当前最优个体,如果满足终止条件,则获得最优BP神经网络权重和阈值,最后基于构建的BP神经网络,输出识别结果。研究结果表明:① 改进的SSA−BP模型在训练集上的识别准确率达95.6%,在测试集上的识别准确率达100%。② 改进的SSA−BP神经网络模型与BP神经网络模型、SSA−BP神经网络模型对比结果:BP神经网络模型误判率为5/18,SSA−BP神经网络模型的误判率为2/18,改进的SSA−BP神经网络模型误判率为0,迭代10次后趋于稳定,且与设定的目标误差相差最小,初始适应度值最优,识别结果可信度高。③ 将阳城煤矿5组矿井水水样数据作为输入层数据输入到训练好的模型中,矿井水水样的主要来源为奥灰含水层、三灰含水层和山西组含水层,模型识别结果与水化学特征分析的结论相互印证,实现了精准区分。

     

    Abstract: The combination of machine learning and optimization algorithms has been widely applied in the recognition of mine water inrush sources. However, the data of water inrush samples is stochastic and the optimization algorithm is prone to getting stuck in local optima. Further research is needed to improve the model's generalization capability and jump out of local optima. In order to solve the above problems, an improved sparrow search algorithm (SSA) is proposed to optimize the BP neural network model for quantitative recognition of mine water inrush sources. Taking Yangcheng Coal Mine of Luneng Coal and Electricity Co., Ltd. as the research object, the hydrochemical characteristics of the coal mine water sample are analyzed through conventional ion concentration analysis and Piper three line diagram. It is preliminarily determined that the mine water comes from the Ordovician limestone aquifer and the three limestone aquifers. The Na++K+ concentration, Ca2+ concentration, Mg2+ concentration, \mathrmHCO_3^- concentration, \mathrmSO^2-_4 concentration, Cl concentration, mineralization degree, total hardness, and pH value are determined as the recognition indicators for water inrush source. The mine water inrush source recognition model is established based on an improved SSA-BP neural network. Firstly, the SSA parameters are set. Sine chaotic mapping is introduced to evenly distribute the sparrow population. Secondly, the sparrow population is updated by calculating fitness values, and a random walk strategy is introduced to perturb the current optimal individual. If the termination condition is met, the optimal BP neural network weight and threshold are obtained. Finally, based on the constructed BP neural network, the recognition results are output. The research results indicate the following points. ① The improved SSA-BP model has an recognition accuracy of 95.6% in the training set and 100% in the testing set. ② The comparison results of the improved SSA-BP neural network model with the BP neural network model and SSA-BP neural network model show that the BP neural network model has a misjudgment rate of 5/18, the SSA-BP neural network model has a misjudgment rate of 2/18, and the improved SSA-BP neural network model has a misjudgment rate of 0. After 10 iterations, it tends to stabilize and has the smallest error difference from the set target. The initial fitness value is the best, and the recognition results have high credibility. ③ Five sets of mine water samples from Yangcheng Coal Mine are inputted into the trained model as input layer data. The main sources of mine water samples are the Ordovician limestone aquifer, the three limestone aquifers, and the Shanxi formation aquifer. The results of model recognition are mutually confirmed with the conclusions of hydrochemical characteristic analysis, and precise segmentation is achieved.

     

/

返回文章
返回