A Hybrid Grey-Artificial Neural Network Model for High-Accuracy Demand Forecasting of Petroleum Products in Chad's Road Transport Sector

This study aims to develop a high-accuracy forecasting model for petroleum product demand in Chad's road transport sector. The core objective is to overcome the limitations of traditional models when applied to the short, volatile, and growth-oriented time series typical of developing economies, thereby providing a reliable tool for national energy planning and policy formulation. We propose a novel integrated hybrid grey-artificial neural network model (abbreviated as GM-ANN) featuring an end-to-end co-optimization framework. The GM(1,1) component captures the exponential trend, while the ANN models non-linear residuals, with all parameters trained simultaneously via backpropagation within a single differentiable graph. Using annual diesel and gasoline consumption data (spanning from 2011 to 2023), the model is benchmarked against six alternatives (including standalone GM(1,1), advanced grey variants, and ARIMA). The proposed model achieves superior performance. With diesel data, the models yields a MAPE of 3.45% and R² of 0.992; and with gasoline consumption data, the model yields a MAPE of 5.12% and R² of 0.986. This represents the first fully differentiable, co-optimized GM-ANN hybrid, moving beyond sequential approaches. The study offers a novel, practical solution for fuel demand forecasting in Chad under data-scarce, high-uncertainty conditions.