针对矿井涌水量预测研究中存在的相关影响因素考虑较少、模型预测精度不高和适用性不强的问题，建立了混沌理论与广义回归神经网络耦合的新的预测模型（Chaos-GRNN模型）。从理论上分析了矿井水文系统产生混沌现象的机理；由混沌理论得到涌水量序列相空间重构后的嵌入维数、时间延迟和最大Lyapunov指数，以此确定GRNN的输入层神经元个数、取值和预测时长；采用交叉验证法获得GRNN的光滑因子，建立Chaos-GRNN模型；对平煤十二矿涌水量（2014年1月至2015年12月）进行模型验证。结果表明：矿井水文系统演化过程的循环迭代是产生混沌的根本原因，其表象特征为演化过程的不可逆性、非平稳性和演化结果的多样性；平煤十二矿涌水量时间序列具有混沌特征，其嵌入维数m=7，即涌水量的影响因素为7个，GRNN输入层神经元个数为7；时间延迟τ为13个月，由此确定了GRNN输入层神经元的取值；最大Lyapunov指数为0.053 0，确定了GRNN预测时长为19个月；Chaos-GRNN模型预测精度达到了94.98%。该预测模型利用混沌理论量化了广义回归神经网络的输入层和预测时长，充分考虑了矿井涌水量的影响因素，提高了预测精度和适用性。

A new prediction model (Chaos-GRNN model) coupled with chaos theory and generalized regression neural network was established to solve the problems of less consideration of relevant influencing factors,low model prediction accuracy and poor applicability in mine water inflow prediction research. The mechanism of chaotic phenomenon in mine hydrological system was theoretically analyzed; the embedding dimension,time delay and maximum Lyapunov exponent of the water inflow time series were calculated by chaos theory,so as to determine the number of neurons in the input layer of GRNN. The number and the value of input layer neurons,and the prediction duration of the model were determined by the m dimensions’sequence.The smoothing factor of GRNN was obtained by cross-validation method. Finally,the Chaos-GRNN model was established. As an example,the water inflow from January 2014 to December 2015 of Pingdingshan Coal Mine No.12 was predicted. The results show that： the cyclic iteration of mine hydrological system evolution process was the fundamental cause of chaos,and its representation features are irreversibility,nonstationarity and diversity of evolution results;  the water inflow time series of Pingdingshan Coal Mine No.12 has chaotic characteristics,its embedding dimension M=7,that is,the influencing factors of water inflow are 7,and the number of neurons in the input layer of GRNN is 7; the time delay τ=13 months,which determines the value of neurons in the input layer of GRNN; the maximum Lyapunov index is 0.053 0,and the prediction time of GRNN is 19 months; the prediction accuracy of the Chaos-GRNN model was 94.98%. Chaos-GRNN model used chaos theory to quantify the number and the value of input layer neurons,and prediction duration of the model. This model fully considered the influencing factors of mine water inflow,improved both prediction accuracy and the applicability of model.