code for Advancing Sustainable Development Goals with Machine Learning and Optimization for Wet Waste Biomass to Renewable Energy Conversion paper submitted to the journal of cleaner produciton
Zhu, S., Preuss, N. and You, F., 2023. Advancing sustainable development goals with machine learning and optimization for wet waste biomass to renewable energy conversion. Journal of Cleaner Production, 422, p.138606.
https://doi.org/10.1016/j.jclepro.2023.138606
This repository contains models and results for the Advancing Sustainable Development Goals with Machine Learning and Optimization for Wet Waste Biomass to Renewable Energy Conversion paper submitted to the Journal of Cleaner Production.
HTC_data.csv and Pyrolysis_data.csv contain data for the models.
HTC_data.xlsx and Pyrolysis_data.xlsx contain summary statistics in addition to the dataset, while SI_Dataset(1).docx provides citation support for the data.
sz_master_071323.ipynb contains all necessary model codes, of which further overview is given in code comments and the remainder of the readme. The sz_master_071323.html file contains sample output from the model.
The code was developed on Python 3.9.7 using optuna version 3.0.2, xgboost 1.6.1, and sklearn 1.1.1. Performance and results may differ if the most recent versions of those packages and Python are used. Additional packages as listed in the first cell of the .ipynb are necessary, though a specific version is not needed.
Use pip install optuna==3.0.2 and similar commands to install packages with specific versions.
Data used for the model can be found in the HTC_data.csv and Pyrolysis_data.csv files. This data, with citation support, can be found in SI_Dataset(1).doxc
To run the code, use the jupyter notebook sz_master_071323.ipynb.
To adjust which dataset, pyrolysis or HTC, is being used, set the variable p in the Technology Type section to true for pyrolysis and false for HTC.
To adjust which optimal solution, CSI or REI, is being used to generate uncertainty analysis points, set the variable c in the Technology Type section to true for CSI and false for REI.
Because Optuna trials are time-consuming, running the block "Running Optuna Trials - S3-S4", users may wish to only run specific trials. This can be done by uncommenting any of the following lines of code to run that specific scenario.
#HTC
#XGB_opt(h_X_train, h_y_train, h_X_test, h_y_test)
#RF_opt(h_X_train, h_y_train, h_X_test, h_y_test)
#DNN_opt(h_X_train, h_y_train, h_X_test, h_y_test)
# pyrolysis
#XGB_opt(p_X_train, p_y_train, p_X_test, p_y_test)
#RF_opt(p_X_train, p_y_train, p_X_test, p_y_test)
#DNN_opt(p_X_train, p_y_train, p_X_test, p_y_test)
Optuna is non-deterministic, so optimal hyperparameters are subject to change.
Likewise, DNN optimization is non-deterministic, and so R2 and RSME values for the optimal hyperparameters are subject to change.
Likewise, generating the lognormal distributions is non-deterministic, so the outputs for CSI and REI values are subject to change.
Likewise, scipy-optimize as used in the paper is non-deterministic, though a seed can be set.
Data for Table S1 and Table S2 can be found in the .xlsx versions of the dataset at the bottom of the file.
Figure 2 was created using Origin 2023b using the "Correlation plot, v 1.31" app with the pyrolysis and HTC data.csv files.