计算机工程与应用 ›› 2015, Vol. 51 ›› Issue (24): 260-265.

• 工程与应用 • 上一篇    下一篇

基于动态聚类的瓦斯浓度高斯预测模型

魏  林1,张  冲2,刘永超2,付  华2,尹玉萍2   

  1. 1.辽宁工程技术大学 基础教学部,辽宁 葫芦岛 125105
    2.辽宁工程技术大学 电气与控制工程学院,辽宁 葫芦岛 125105
  • 出版日期:2015-12-15 发布日期:2015-12-30

Gas concentration Gaussian prediction model based on dynamic clustering

WEI Lin1, ZHANG Chong2, LIU Yongchao2, FU Hua2, YIN Yuping2   

  1. 1.Department of Basic Education, Liaoning Technical University, Huludao, Liaoning 125105, China
    2.School of Electrical and Control Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2015-12-15 Published:2015-12-30

摘要: 为了实现准确可靠的瓦斯浓度预测,利用总歪指标取最大值来确定动态聚类的最佳聚类,以此减少不确定、随机因素干扰所引起的预测误差。由于高斯过程回归参数少、易实现,能输出具有较高置信度的置信区间,故利用高斯过程回归进行瓦斯浓度区间预测,并利用微分进化算法来确定高斯过程超参数。将动态聚类算法和高斯回归模型结合,实现了瓦斯浓度的区间预测模型。通过实验对比分析,结果表明该方法能够有效地预测出瓦斯浓度变化趋势,较高斯过程预测模型提高了瓦斯浓度的预测精度。

关键词: 动态聚类, 微分进化, 高斯过程, 时间序列, 置信区间

Abstract: In order to realize accurate and reliable gas concentration prediction, the maximum total slanting indicators are used to determine the best dynamic clustering to reduce the prediction error caused by the uncertainty and random interference factors. Due to low Gaussian process regression parameters and easy implementation, it can get the confidence interval with high degree of confidence, and predict the gas concentration range with gaussian process regression.Differential evolution algorithm is used to determine the Gaussian process parameters. This paper combines dynamic clustering algorithm and Gaussian regression model,and achieves interval prediction model of gas concentration. By comparison with experimental analysis, the results show that the method can effectively predict the gas concentration change trend, and improve the prediction precision of the gas concentration than Gaussian process.

Key words: dynamic clustering, differential evolution, Gaussian process, time series, confidence interval