Abstract
A reliable energy supply in distribution systems is crucial for end users because modern-day activities rely on high-quality electricity. Nevertheless, these systems have several obstacles to fulfilling the requisite demands due to the substantial losses. Higher losses in the system have led to energy waste, making the system less efficient. Also, changes in load cause the system to experience instability in the voltage profile. Hence, there is a need for network reconfiguration and reactive power dispatch with metaheuristic algorithms to minimize power loss and voltage deviation and enhance the voltage stability index. Also, it is important to have proper energy management by forecasting energy consumption. This thesis aims to develop an artificial intelligence-based metaheuristic algorithm to optimize power loss and voltage deviation, enhance the voltage stability index into single and multi-objective functions, and forecast energy consumption. Different metaheuristic algorithms were used in this research, such as the hybrid Archimedes optimization algorithm (HAOA), chaotic sinusoidal map-based AOA (CAOA), enhanced particle swarm optimization (EPSO), improved pathfinder algorithm (IPFA), and reinforcement learning PSO (RPSO) combined with adaptive network-based fuzzy inference systems (ANFIS) called (RPSO-ANFIS). The innovation behind those methods is improving AOA; Levy flight was hybridized with AOA to improve sluggish performance (HAOA). In CAOA, a chaotic sinusoidal map was used to improve the initialization of AOA. In EPSO, a new acceleration constant was used to enhance the basic PSO performance. The IPFA uses the inertia weight to strengthen basic PFA fluctuation and vibration coefficients. The RPSO-ANFIS uses reinforcement learning to adaptively change the parameters of PSO based on the time-varying acceleration coefficient (RPSO) combined with ANFIS called (RPSO-ANFIS).
The findings showed that the HAOA tested on IEEE 33 and 69 bus systems revealed reduced power losses of 139.55 kW and 98.46 kW for both systems, respectively. Also, the VD was reduced to 0.00134 p.u, and VSI was enhanced to 0.785 p.u for the 33-bus system, whereas on the 69-bus system, 0.000306 p.u. and 0.8126 p.u. were obtained for VD and VSI, respectively. The respective power loss and VD obtained by the proposed CAOA are 139.365 kW and 0.00134 p.u., and the result shows that CAOA is superior to other methods previously reported values in the literature, showing a 31.24% reduction in power loss for the 33-bus system. The other Chapters, five, six, and seven, were based on the ORPD problem to minimize power loss using different optimization algorithms, and the performance of these optimization algorithms was validated in IEEE 9, 14, 30,
vi
39, 57, and 118 bus systems. The power loss obtained for the 9 and 14 bus systems (Chapter Five) was 7.543 MW and 12.253 MW, respectively, for EPSO. The power loss obtained for the 30 and 118 bus systems (Chapter Six) was 16.035 and 115.048 MW, respectively, for IPFA. Also, the power loss of 37.109 MW and 22.304 MW were obtained for the 39 and 57 bus systems (Chapter Seven) for the Bat algorithm. The results obtained from each technique demonstrated the capability of metaheuristic algorithms to lower power loss while maintaining handling constraints. The application of machine learning to forecast energy consumption was provided in chapter eight. The findings demonstrated that PSO-ANFIS exceeds the capabilities of standalone ANFIS; however, RPSO-ANFIS markedly outperforms all four performance metrics. Using different performance matrices to validate the performance of the RPSO-ANFIS, the results show that RPSO-ANFIS gives good results in all the performance metrics with the value of 1.6232 for RMSE, 1.2846 for MAD, 39.509 for MAPE, and 1.2922 for MAE, compared to both ANFIS and PSO-ANFIS. This result highlights the efficacy of combining PSO with reinforcement learning to transform the ANFIS model for precise forecasting of the electricity consumption among student residents at the University of Johannesburg over a medium-term perspective.