CL-MOEEL: Closed-loop Multiobjective Evolutionary Ensemble Learning

This is the code implementation for the paper: Silicon Content Prediction in Blast Furnace Ironmaking Process Based on Closed-loop Multiobjective Evolutionary Ensemble Learning (Accepted by IEEE TIM).

This public GitHub repository contains:

• All source code related to our proposed algorithm;

• The benchmark dataset files (Residential Building Data set): data/Residential Building Data Set/Residential-Building-Data-Set-preprocessed.xlsx;

• The practical industrial dataset files (Silicon Content Prediction Data Set): data/Silicon Content Prediction Data Set/Si_predict.xlsx;

Note: Some code is inspired by jMetalPy. A paper introducing jMetalPy is available at: jMetalPy: A Python framework for multi-objective optimization with metaheuristics.

Locations of Modules for the Proposed CL-MOEEL in the Project Files

• Proposed CL-MOEEL Algorithm: core/algorithm/moeec_nsgaii.py, class name: MOEEC_NSGAII_v5;
• Proposed Variable-Length Encoding Scheme: core/problem/solution.py;
• Proposed Knowledge-Guided Selection (KGS) Method: core/algorithm/moeec_nsgaii.py, lines 205–277;
• Two Proposed Variable-Length Crossover Operators (inside-cluster crossover and outside-cluster crossover): core/algorithm/moeec_nsgaii.py, lines 364–443;
• Precomputation Technique (to reduce the algorithm's time complexity): core/problem/ensemble_model.py, located in the MultiKernelTransformer class.

Test CL-MOEEL on Benchmark Dataset & Practical Industrial Dataset

Step 1. Run algorithm

python "scripts/run_moeec_nsgaii_v5_kfold.py" --dataset=""

• --dataset: the dataset name, e.g., "UCI14_ResidentialBuilding" (benchmark dataset ), or "Si_predict" (practical industrial dataset).

Step 2. Process results

python "scripts/processing_results_kfold.py" --dataset="" --path=""

• --dataset: the dataset name, e.g., "UCI14_ResidentialBuilding" (benchmark dataset ), or "Si_predict" (practical industrial dataset).

• --path: the result path, e.g., "MOEEC_NSGAII_v5_[year]-[month]-[day]_kfold", where "[year]-[month]-[day]" is the time stamp of the folder. Note that this name will be automatically generated after running the script "run_moeec_nsgaii_v5_kfold.py".

After running the above command, two result files, "results.csv" and "inds.csv" , will be generated in the directory specified by the --path parameter. The "results.csv" file contains detailed records of all metrics (e.g., RMSE, $R^2$, HR, Runtime) for the five-fold cross-validation across each repeat experiment, including results for each fold. The "inds.csv" file records the average results of the five-fold cross-validation across all repeat experiments.

Test CL-MOEEL on Your Own Dataset

Step 1. Creat a custom dataset class

Create a new script in the "core/dataset/" directory, defining your own dataset class (which should inherit from the MyDatasetKFold class), and import the newly defined class name in the __init__.py file.

Step 2. Run CL-MOEEL and process results

Execute the following command to run the algorithm and process the experimental results.

# run algorithm
python "scripts/run_moeec_nsgaii_v5_kfold.py" --dataset=""

# process results
python "scripts/processing_results_kfold.py" --dataset="" --path=""

Note that the dataset parameter (--dataset) should be changed to the name of your defined dataset class, and the algorithm results will be saved in the "models/" directory. After processing the results, the evaluation metrics (RMSE, $R^2$, HR, Runtime) will be computed and saved in "inds.csv" in the same directory.

Citation

If you find our work useful, please cite our paper:

@ARTICLE{10925555,
  author={Zhang, Jingchuan and Zhao, Guodong and Wang, Xianpeng},
  journal={IEEE Transactions on Instrumentation and Measurement}, 
  title={Silicon Content Prediction in Blast Furnace Ironmaking Process Based on Closed-Loop Multiobjective Evolutionary Ensemble Learning}, 
  year={2025},
  volume={74},
  number={},
  pages={1-15},
  doi={10.1109/TIM.2025.3551036}}