Variable Selection Method Based on Partial Mutual Information and Its Application to NOx Emission Prediction
- Resource Type
- Conference
- Authors
- Tianmu, QIN; Jinzhe, ZHANG; Mo, YOU; Tingting, YANG
- Source
- 2020 39th Chinese Control Conference (CCC) Chinese Control Conference (CCC), 2020 39th. :1017-1021 Jul, 2020
- Subject
- Computing and Processing
Robotics and Control Systems
Signal Processing and Analysis
Input variables
Data models
Coal
Mutual information
Benchmark testing
Atmospheric modeling
Kernel
NOx emission
coal-fired boiler
partial mutual information (PMI)
data-driven modeling
- Language
- ISSN
- 1934-1768
Data-driven modeling methods are widely used in industrial processes as the foundation of control and optimization. The selection of optimal variable set plays an important role in model performance. In order to enhance the model prediction accuracy, a partial mutual information (PMI) method was proposed to select the optimal variable set. Benchmarks were used to validate the effectiveness of PMI method. Then, PMI method was applied to select main influencing factors of NOx emission of coal-fired boiler and the selection results were used as inputs of three different data-driven models. The comparison between the models with or without variable selection was made. The results showed that the PMI method enhanced the model prediction accuracy and avoided the over-fitting problem.