中文摘要：

针对瓦斯浓度时间序列的混沌性,提出一种回声状态网络算法（ESN）和无迹卡尔曼滤波器（UKF）、强跟踪滤波器（STF）耦合的混沌时间序列预测模型。对于一维瓦斯浓度混沌时间序列,采用平均轨道周期的C-C算法在时间域确定重构空间的最佳时间延迟和嵌入维数,在相空间通过非线性回归预测模型拟合瓦斯涌出动力演化轨迹,提出带有渐消因子的非线性STUKF滤波器对ESN联合参数进行最优状态估计。试验结果表明：基于STUKF的ESN瓦斯涌出模型预测方法有效,在STUKF滤波器作用下增强了ESN算法的学习效率、提高了模型的跟踪能力,能达到精度高、鲁棒性好等优点。

英文摘要：

Considering the chaotic property of gas concentration time series, a method was put forward by a coupled algorithm which consisted of echo state network, unscented kalman filter and strong tracking filter. The optimal em- bedded dimension and delay time was determined synchronously based on algorithm of C-C and theory of the aver- age orbital period in the time domain, in the phase space the gas dynamic evolution track was matched through non- linear regression forecasting model, the fading factors was introduced to nonlinear kalman filter STUKF which was applied to realize optimal state estimation of parameters of ESN. The simulation test results show that gas emission prediction model based on algorithm of ESN and STUKF is effective, and greatly helpful for improving ESN learning efficiency and traceability, meanwhile can achieve high precision and good robustness, etc.