关键词: 灰色理论, 聚类分析, 关联性分析, 高斯过程回归, 协方差函数

Abstract: Considering the problem of low forecasting accuracy caused by the complexity of the carbon emission sources from the motor and the diversity of its impacts during the mining process of lead-zinc mine, a carbon emission forecasting method for mining process is proposed based on improved Gaussian process regression model combined with the grey theory. Firstly, the sources of carbon emission and their impacts are analyzed and the grey theory is used to cluster and merge the similar impacts. Then, the grey relational analysis is applied to obtain the main impacts. Finally, the forecasting accuracy is not satisfied due to the direct selection of the covariance function based on experience in the traditional Gaussian process regression model. In order to improve the accuracy, a covariance function selection method with prior knowledge is proposed. Based on the parameter estimation result of maximum likelihood estimation and least square method as well as Monte Carlo method, the covariance function that makes the model forecasting accuracy highest and the error minimum is obtained by comparing the forecasting results of several common-used covariance functions modeling. The result shows that improved Gaussian process regression model can be used to forecast the short-term carbon emission of lead-zinc mine mining process with high accuracy and minimum error compared with other forecasting models.

Key words: grey theory, clustering analysis, relational analysis, Gaussian process regression, covariance function