Hybrid AHP and Artificial Neural Networks for Multi-Criteria Modeling of Senegal’s Energy Transition Pathways toward 2040
Papa Touty Traore *
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
El Hadji Abdoul Aziz Cisse
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
Kharma Gaye
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
Mamadou Lamine Diallo
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
Pape Gueye Ndiaye
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
Mor Ndiaye
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
Issa Diagne
Semiconductor and Solar Energy Laboratory, Department of Physics, Faculty of Science and Technology, Cheikh Anta Diop University, Dakar, Senegal.
*Author to whom correspondence should be addressed.
Abstract
Senegal’s energy transition requires balancing cost, system resilience, environmental performance, employment, energy independence, and resource availability. This study develops a hybrid Analytic Hierarchy Process (AHP)–artificial neural network (ANN) framework to assess four transition pathways towards 2040: Ambitious Renewable Transition (TAR), Gas–Renewables Mix (MGRE), Domestic Gas Dominance (DGDM), and Slow Transition (TL). AHP is first used to structure the six decision criteria and establish the 2035 baseline ranking. Because the available AHP dataset is limited, perturbation-based oversampling is applied to generate 1,000 synthetic samples for ANN training. The ANN uses six input neurons, two hidden layers with 12 and 6 neurons, respectively, and one output neuron. The dataset is divided into 80% training and 20% testing subsets, and sensitivity is examined using perturbation ranges of ±5%, ±10%, and ±15%. For 2035, MGRE records the highest score (0.78), followed by DGDM (0.72), TAR (0.65), and TL (0.50). Under the specified 2040 weighting assumptions, TAR rises to 0.82, while MGRE, DGDM, and TL decline to 0.70, 0.60, and 0.45, respectively. The results indicate that stronger emphasis on environmental impact and energy independence can shift the preferred pathway towards renewables. These projections should be interpreted as conditional scenario outcomes rather than deterministic forecasts.
Keywords: Analytic Hierarchy Process, artificial neural network, Multi-criteria decision analysis, energy transition, renewable energy, natural gas