计算机应用 ›› 2015, Vol. 35 ›› Issue (1): 289-293.DOI: 10.11772/j.issn.1001-9081.2015.01.0289

• 行业与领域应用 • 上一篇    下一篇

最小二乘支持向量机与Kalman滤波耦合的瓦斯涌出量动态预测模型

付华, 訾海   

  1. 辽宁工程技术大学 电气与控制工程学院, 辽宁 葫芦岛125105
  • 收稿日期:2014-07-14 修回日期:2014-08-18 出版日期:2015-01-01 发布日期:2015-01-26
  • 通讯作者: 訾海
  • 作者简介:付华(1962-),女,辽宁阜新人,教授,博士生导师,主要研究方向:计算机智能检测、信息融合;訾海(1989-),男,辽宁阜新人,硕士,主要研究方向:智能检测、智能控制.
  • 基金资助:

    国家自然科学基金资助项目(51274118);辽宁省教育厅基金资助项目(L2012119);辽宁省科技攻关项目(2011229011).

Dynamic prediction model for gas emission quantity based on least square support vector machine and Kalman filter

FU Hua, ZI Hai   

  1. College of Electrical and Control Engineering, Liaoning Technical University, Huludao Liaoning 125105, China
  • Received:2014-07-14 Revised:2014-08-18 Online:2015-01-01 Published:2015-01-26

摘要:

针对瓦斯涌出量的多影响因素预测问题,提出一种最小二乘支持向量机(LS-SVM)回归算法与卡尔曼滤波耦合的动态预测方法.该方法依据预测残差方差比检验策略确定自适应的动态训练样本集以取代固定的训练样本集.LS-SVM辨识网络对瓦斯涌出量的相关因素进行非线性映射并提取出最佳维数的状态向量以建立基于卡尔曼滤波最优估计的瓦斯涌出量预测模型.利用矿井监测到的各项历史数据进行实验.结果表明,该模型的预测平均相对误差为2.17%,平均相对变动值ARV为0.008873,相比单一的神经网络或支持向量机预测模型,具有更高的预测精度与更强的泛化能力.

关键词: 非线性, 动态训练样本集, 最小二乘支持向量机, 卡尔曼滤波, 瓦斯涌出量

Abstract:

In order to solve the multifactor problem related to gas emission quantity prediction, a new dynamic prediction method of coupling Least Square Support Vector Machine (LS-SVM) with Kalman filter was proposed. The dynamically adaptive set of training samples were obtained to replace the fixed set of training samples based on the strategy for predicting variance ratio of residual errors. LS-SVM identification network was used to perform nonlinear mapping on relevant factors of gas emission quantity to extract the state vector with the best dimension number. The Kalman filter based gas emission quantity forecasting model was established by using the state vector. Experiments were carried out with the monitoring data of the mine. The experimental results show that the average relative error of results predicted by the model is 2.17% and the average relative variance is 0.008873. The proposed model is superior to other prediction models of neural network and support vector machine in terms of prediction accuracy and generalization ability.

Key words: nonlinear, dynamic training sample set, Least Square Support Vector Machine (LS-SVM), Kalman filter, gas emission quantity

中图分类号: