An Innovative NOx Emissions Prediction Model Based on Random Forest Feature Selection and Evolutionary Reformer

Developing more precise NOx emission prediction models is pivotal for effectively controlling NOx emissions from gas turbines. In this paper, a Reformer is combined with random forest (RF) feature selection and the chaos game optimization (CGO) algorithm to predict NOx in gas turbines. Firstly, RF evaluates the importance of data features and reduces the dimensionality of multidimensional data to improve the predictive performance of the model. Secondly, the Reformer model extracts the inherent pattern of different data and explores the intrinsic connection between gas turbine variables to establish a more accurate NOx emission prediction model. Thirdly, the CGO algorithm is a parameter-free meta-heuristic optimization algorithm used to find the best parameters for the prediction model. The CGO algorithm was improved using Chebyshev Chaos Mapping to improve the initial population quality of the CGO algorithm. To evaluate the efficiency of the proposed model, a dataset of gas turbines in north-western Turkey is studied, and the results obtained are compared with seven benchmark models. The final results of this paper show that RF can select appropriate input variables, and the Reformer can extract the intrinsic links of the data and build a more accurate NOx prediction model. At the same time, ICGO can optimize the parameters of the Reformer effectively.