An enhanced support vector regression -African Buffalo optimisation algorithm for electricity time series forecasting

• المؤلف الرئيسي: Maijama'a, Inusa Sani
• التنسيق: أطروحة
• اللغة: الإنجليزية; الإنجليزية
• منشور في: 2023
• الموضوعات: HB Economic Theory
• الوصول للمادة أونلاين: https://etd.uum.edu.my/11729/1/depositpermission.pdf; https://etd.uum.edu.my/11729/2/s902030_01.pdf; https://etd.uum.edu.my/11729/

Time series forecasting enables informed decision-making and stakeholder benefit. Electricity production-consumption balance is vital in modern economies. Load forecasting predicts electricity consumption, influenced by factors like population, weather, policies, and socio-economic activities. Support Vector Regression (SVR) is a widely used regression technique, but its efficacy depends on optimal tuning of parameters, which is challenging. This study proposes a hybrid approach combining SVR and the African Buffalo Optimization (ABO) algorithm. The classical ABO algorithm faces limitations in population initialization, exploration, and exploitation. Therefore, enhancements have been made to these stages to improve performance. The study presents a series of hybrid algorithms that leverage ABO to optimize SVR hyperparameters. SVR-ABO uses the classical ABO approach. SVR-popABO enhances population diversity using a chaotic function. SVR-explrABO includes Lévy flight to improve exploration and overcome local optima. SVR-expltABO modifies the exploitation mechanism to prevent premature convergence. These four hybrids represent a progressive refinement of the classical ABO for optimizing SVR hyperparameters. Combining the enhanced algorithms results in SVR-eABO, whose forecasting ability has been assessed using MAE, MAPE, RMSE, PA and R2. Evaluated using benchmark datasets, SVR-eABO achieves high accuracy, surpassing standard SVR and other optimization-based SVR variants like SVR-PSO, SVR-ABC, SVR-CS, and SVR-GA. For instance, SVR-eABO achieved 98.51% accuracy on the Household dataset, 98.15% accuracy on the Turkey dataset, 91.17% accuracy on the Appliances dataset, and 96.52% accuracy on the Panama dataset. The proposed SVReABO algorithm holds significant implications for improving load forecasting accuracy, enabling more efficient electricity grid management, and facilitating informed decision-making for energy providers and consumers