针对瓦斯浓度时间序列高度的混沌特性，采用微熵率法同步确定最优的嵌入维数与延迟时间，还原瓦斯涌出系统状态空间。以无线传感网络系统采集并经降噪处理后的瓦斯浓度序列作为样本。提出利用带有整定因子的扩展卡尔曼滤波器( EKF)对加权最小二乘支持向量回归机( WLS-SVR)的正则化参数γ与核参数σ进行快速寻优，并依据周期性更新的训练样本建立基于EKF-WLS-SVR耦合算法的动态预测模型以精确预测后续时间点的瓦斯浓度。通过MATLAB进行仿真，结果表明：EKF滤波器对提高WLS-SVR的拟合精度与学习效率方面有很大的帮助。相比于其他模型，该耦合模型具备更高的预测精度与更强的鲁棒特性，有较高的实用价值。%Considering highly chaotic characteristic of gas concentration time series, the differential entropy ratio method was adopted to synchronously determine optimal embedded dimension and delay time so as to restore the system state space of gas emission. The samples with noise elimination came from the gas concentration series which was collected by wireless sensor networks. Extended kalman filter algorithm( EKF) with tuning factor was proposed to rapidly optimize the regularization parameterγand the nuclear parameterσof the weighted least squares support vector regression( WLS-SVR). Periodically updated training samples were used to establish the EKF-WLS-SVR coupling algorithm-based dynamic prediction model to accurately predict gas concentration in the future. The simu-lation of the MATLAB shows that EKF is greatly helpful for improving WLS-SVR fitting precision and learning effi-ciency. The coupling model is practically useful and outperforms other prediction models in terms of prediction ac-curacy and robustness.
展开▼
