In the age of digital technology, artificial intelligence (AI) emerges as a key player in demand prediction. In a recent publication in the journal Sustainability, researchers used AI techniques for oil demand forecasting in both exporting and importing nations. Further, researchers compared the performance of AI techniques with existing statistical models.
Study: Revolutionizing Energy Planning: AI-Powered Oil Demand Forecasting. Image credit: Bself/Shutterstock
Background
Oil demand forecasting is crucial for energy planning. It aids governments, energy firms, and stakeholders make informed decisions regarding production, consumption, and pricing. Traditional methods have limitations, but AI and machine learning offer advanced tools for precise forecasting. However, there is limited research on AI-based analysis considering both endogenous and exogenous factors in oil demand forecasting.
The current study aims to fill these gaps by analyzing oil demand in importing and exporting countries and incorporating comprehensive factors. This research empowers policymakers and energy stakeholders with improved forecasting tools for navigating the evolving global oil market landscape.
Machine learning models for energy demand forecasting
Machine learning models have gained popularity in energy demand forecasting due to their superior performance over traditional statistical methods such as time series and regression analysis. Researchers have employed a range of machine learning algorithms to attain exceptional predictive accuracy and robustness.
For instance, Duan et al. introduced the prediction model termed KELM-GMCC, which outperformed existing models such as Support Vector Regression (SVR), Back Propagation (BP), and Extreme Learning Machine (ELM) in terms of accuracy. Zhang and Liang applied Bayesian neural networks to predict refined oil product demand and demonstrated their accuracy. Nia et al. emphasized the cost reduction and increased profitability achieved through smart forecasting methods.
Furthermore, Abbasimehr et al. introduced a multi-layer Long Short-Term Memory (LSTM) forecasting method for the furniture industry, outperforming other models in efficiency. Integrating AI models with econometric models, Huang et al. improved the accuracy of energy demand forecasts. Al-Fatah and Aramco used a hybrid approach to AI techniques to predict oil demand in different regions.
The current study explored beyond the energy sector, including blood demand prediction in Ghana, retail supply chain demand, and steel sector demand in Malaysia. It emphasizes the potential of machine learning models in various industries and regions. Additionally, the study highlights findings on factors influencing energy consumption, such as economic growth, energy source preferences, and environmental pressures. It underscores the importance of accurate demand forecasting for efficient supply chain management and informed decision-making.
A proposed method for forecasting oil demand
The study draws its primary value data from the electronic sources of gas and oil companies connecting Iran and China. The training data in the SVR model consists of independent variables and their corresponding dependent values. The SVR equation is constructed in the feature space, emphasizing the weight vector, a constant value, and the feature function. Optimization aims to minimize the experimental error while balancing it with model complexity.
The study highlights that the input variables for demand forecasting encompass various factors, including carbon emissions, income level, energy price, gross domestic product (GDP), population growth, urbanization, trade liberalization, inflation, foreign direct investment (FDI), financial development, energy sanctions, and pandemics. Data analysis and model fitting were conducted using R software and specific libraries.
Results and analysis
Results show that carbon emissions have declined recently, possibly due to increased renewable energy use globally. FDI growth has been variable. The financial crisis (2007–2009) impacted GDP and per capita income growth. Oil consumption has dropped since 2015, attributed to oil sanctions and the COVID-19 pandemic. Reduced demand led to lower oil prices.
The SVM model predicts oil demand, with actual and predicted values presented. Weight parameters from model fitting are also observed. Factors such as per capita income, FDI rate, inflation, oil price, sanctions, and the pandemic influence oil demand. Urbanization has the least impact, while trade liberalization, carbon emissions, and the pandemic are significant.
The SVM model exhibits the lowest prediction error for demand forecasting, followed by an autoregressive integrated moving average (ARIMA) model. Mann-Whitney tests reveal higher demand uncertainty in oil-exporting countries. The ARIMA and general linear models (GLMs) are less stable in such uncertainty compared to the SVM model.
Conclusion
In summary, researchers explored forecasting oil demand in both exporting and importing nations using AI methods. The results were compared with statistical models. The SVR model outperformed other models in terms of prediction accuracy and stability. It was found that AI models, incorporating input variables for forecasting instead of relying solely on historical data, were more successful in demand forecasting, aligning with prior research.
In the context of oil-exporting and importing countries facing varying levels of demand uncertainty, the SVR model demonstrated higher stability. This suggests that AI models can enhance supply chain management. Furthermore, the study highlights the economic challenges posed by sanctions and the pandemic for oil-exporting countries, emphasizing the need for diversification and income generation from non-oil sources.
Optimizing Wastewater Treatment with Machine Learning