【摘要】 为了在煤矿瓦斯涌出量相关影响因素的作用发生改变时,还能够准确预测瓦斯涌出量,提出一种基于虚拟状态变量的卡尔曼滤波预测方法。将相关影响因素通过能够识别瓦斯涌出量模型的非线性网络进行映射,用所得到的输出向量作为虚拟状态变量,提出预测残差方差比检验方法,计算虚拟状态变量的最佳维数,确定能够反映当前瓦斯涌出量的最小样本个数的储量样本。采用基于储量样本计算得到具有最佳维数的虚拟状态变量,建立卡尔曼滤波瓦斯涌出量预测模型。结果表明:对于瓦斯涌出量相关因素作用发生变化的情形,采用固定的训练样本和网络结构建立的基于人工神经网络的预测方法,预测结果的平均误差为5.82%,最大误差为16.56%,采用动态调整的虚拟状态变量建立的卡尔曼滤波预测方法具有较好的跟踪能力和反应速度,预测性能明显改善,其平均误差为0.94%,最大误差为2.08%,表明所建议的方法是可行和有效的。更多还原

【Abstract】 In order to accurately predict the coal gas emission quantity even when the related factors changed,a novel approach based on virtual state variables and Kalman filter was proposed.The factors were mapped by a nonlinear network with the capability of recognizing the model of gas emission quantity,and the obtained output vectors were used as the virtual state variables.The forecasting variance ratio testing method was introduced to calculate the best dimension of virtual state variables and determine the reserved samples which can identify the current gas emission quantity with the minimum number of samples.The Kalman filter based gas emission quantity forecasting model was established by using the virtual state variables with the best dimension calculated on the basis of the reserved samples.The results show that,in the situation that the effect of the related factors was changed,the averaged prediction biases is 5.82% and the maximum deviation is 16.56% by using the ANN with the fixed training samples and network structure.The tracking ability and the dynamic behavior are remarkably improved to the averaged biases of 0.94% and the maximum error of 2.08% by using the Kalman filter based forecaster based on the proposed dynamic adjusting virtual state variables.It is indicated that the suggested approach is feasible and effective.更多还原