Input variable selection for the statistical prediction model on energy consumption of products pipelines

摘要

Selecting an appropriate input vector is a critical issue in forecasting.For predicting the daily energy consumption of multiproduct pipelines,the input vector design is usually based on engineering experience,and the number of the traditional input variables is large.However,such a forecast model based on the traditional input variables needs more computational resources and longer training time.In order to prune the number of input variables,the correlation coefficient and partial correlation coefficient are introduced to measure the correlation between input variables and the output variable.In a case study involving a Chinese products pipeline,a new input vector,which contains 5 input variables,are identified from the traditional 15 input variables through the correlation analysis.To verify the rationality of the new vector,an ANN model based on the new vector is developed to forecast the daily energy consumption,and its forecasted values are compared to the values of the ANN model based on the traditional input vector and actual values.The results show that the ANN model based on the new vector has higher prediction accuracy,indicating that more parsimonious set of input variables can be used in daily energy consumption forecast of products pipeline without sacrificing the accuracy of the forecast.