A Hybrid Model with Applying Machine Learning Algorithms and Optimization Model to Forecast Greenhouse Gas Emissions with Energy Market Data

• Published in: Sustainable cities and society 2022-07, Vol.82, p.103886, Article 103886
• Main Authors: Emami Javanmard, Majid, Ghaderi, S.F.
• Format: Article
• Language: English
• Subjects: Air Pollution; Energy Market Data; Forecasting; Greenhouse Gas Emissions; Machine-Learning Algorithms; Optimization Model
• ISSN: 2210-6707; 2210-6715
• DOI: 10.1016/j.scs.2022.103886

•Forecasting Greenhouse Gas Emission Applying 9 Machine Learning Algorithms•Presented an Optimization Model for Increase Prediction Accuracy in GHG Emission•Metaheuristic Algorithms Results Show Optimization Model Can Increase Prediction Accuracy.•Presented 10-Years Forecasting for GHG Emission High Prediction Accuracy•Evaluation Relationship Between Energy Consumption and GHG Emission by Stepwise Regression In recent decades, many countries have encountered air pollution and environmental problems caused by greenhouse gas (GHG) emissions. One of the essential approaches to managing and planning for GHG emissions reduction is an accurate prediction. This study has presented a hybrid machine learning and mathematical programming approach that enjoys a high prediction accuracy and is implemented with small data to predict the emission of greenhouse gases (CO2, N2O, CH4, and Fluorinated-Gas). Data in the energy sector affecting GHG emissions and also related data to the rates of GHG emissions from 1990 to 2018 in Iran have been collected. Then, the forecast of emission of each gas has been performed by nine algorithms of ANN, AR, ARIMA, SARIMA, SARIMAX, RF, SVR, KNN, and LSTM. The forecast accuracy of each algorithm has been evaluated with five indicators. The results of GHG emissions till 2028 have been forecasted and presented with algorithms with higher forecast accuracy. In the next step, the outcomes of the machine learning algorithms have been regarded as the input data to the mathematical model, which has been then implemented by the PSO and GWO metaheuristic algorithms, and the emission of greenhouse gases has been predicted until the year 2028. The prediction accuracy of the proposed approach has been evaluated by five indices and compared with the results of machine learning algorithms. Finally, the Stepwise Regression algorithm has been applied to evaluate the relationships between the data. In general, the results obtained from this study indicated an increase in forecast accuracy using an optimization model. This optimization model has increased the forecast accuracy compared to the machine learning algorithms applied in this study. The improvements were at least 31.7% with the PSO algorithm and 12.8% with the GWO algorithm. Based on the high-accurate prediction of the proposed approach, the emission of greenhouses gases in Iran will exceed 1096 Mt/year in 2028. [Display omitted]