Nemangwele, F.Ratshitanga, M.Nemakonde, Pfano2026-06-172026-06-172026-05-19Nemakonde, P. 2026. Machine Learning Applications in Blockchain for Renewable Energy Systems. . .https://univendspace.univen.ac.za/handle/11602/3200M.Sc. in PhysicsDepartment of PhysicsThe transition towards decentralized renewable energy systems offers a critical solution to the "Energy Trilemma," yet its practical implementation in emerging economies such as South Africa is obstructed by grid instability, inaccurate demand planning, and the lack of secure local market mechanisms. This thesis addresses the "Deployment Feasibility Gap" in Peer-to-Peer (P2P) energy trading by establishing a synergistic framework that integrates advanced Machine Learning (ML) forecasting with Distributed Ledger Technology (DLT). The research first investigates the limits of predictive accuracy for community microgrids. A novel hybrid deep learning model, Bidirectional Long-Short-Term-Memory with Gated Recurrent Unit (BiLSTM-GRU), is developed for regional solar irradiance forecasting, while a rigorous comparative analysis of ensemble methods such as eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Random Forest is conducted for residual demand. To optimize these models, the study contrasts bio-inspired Swarm Intelligence, Honey Badger Algorithm (HBA), Particle Swarm Optimization (PSO) with probabilistic Gaussian Process Bayesian Optimization, and Heteroscedastic Evolutionary Bayesian Optimization (GP-BO, HEBO). Results demonstrate that the HBA-optimized XGBoost model, when coupled with robust feature scaling, achieves superior predictive fidelity, significantly reducing Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) in volatile grid conditions. To operationalize these forecasts, the study proposes "GreenGrids," a P2P trading architecture built on the Hedera Hashgraph network. This Proof-of-Concept validates the technical and economic viability of using high-throughput, low-latency DLT for micro-energy transactions, overcoming the scalability limitations of traditional blockchains. Synthesizing these technical findings with a critique of the South African regulatory landscape, the study culminates in the Deployment Feasibility Framework (DFF). This four-pillared framework offers a comprehensive blueprint for implementing sustainable, community-level energy markets, bridging the gap between theoretical computational models and real-world socioeconomic applications.1 online resource (xii, 93 leaves): color illustrations, color mapsenUniversity of VendaBlockchainUCTDCommunity energyDemand responseEnergy predictionP2P energy tradingDissertationNemakonde P. Machine Learning Applications in Blockchain for Renewable Energy Systems. []. , 2026 [cited yyyy month dd]. Available from:Nemakonde, P. (2026). <i>Machine Learning Applications in Blockchain for Renewable Energy Systems</i>. (). . Retrieved fromNemakonde, Pfano. <i>"Machine Learning Applications in Blockchain for Renewable Energy Systems."</i> ., , 2026.TY - Dissertation AU - Nemakonde, Pfano AB - The transition towards decentralized renewable energy systems offers a critical solution to the "Energy Trilemma," yet its practical implementation in emerging economies such as South Africa is obstructed by grid instability, inaccurate demand planning, and the lack of secure local market mechanisms. This thesis addresses the "Deployment Feasibility Gap" in Peer-to-Peer (P2P) energy trading by establishing a synergistic framework that integrates advanced Machine Learning (ML) forecasting with Distributed Ledger Technology (DLT). The research first investigates the limits of predictive accuracy for community microgrids. A novel hybrid deep learning model, Bidirectional Long-Short-Term-Memory with Gated Recurrent Unit (BiLSTM-GRU), is developed for regional solar irradiance forecasting, while a rigorous comparative analysis of ensemble methods such as eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Random Forest is conducted for residual demand. To optimize these models, the study contrasts bio-inspired Swarm Intelligence, Honey Badger Algorithm (HBA), Particle Swarm Optimization (PSO) with probabilistic Gaussian Process Bayesian Optimization, and Heteroscedastic Evolutionary Bayesian Optimization (GP-BO, HEBO). Results demonstrate that the HBA-optimized XGBoost model, when coupled with robust feature scaling, achieves superior predictive fidelity, significantly reducing Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) in volatile grid conditions. To operationalize these forecasts, the study proposes "GreenGrids," a P2P trading architecture built on the Hedera Hashgraph network. This Proof-of-Concept validates the technical and economic viability of using high-throughput, low-latency DLT for micro-energy transactions, overcoming the scalability limitations of traditional blockchains. Synthesizing these technical findings with a critique of the South African regulatory landscape, the study culminates in the Deployment Feasibility Framework (DFF). This four-pillared framework offers a comprehensive blueprint for implementing sustainable, community-level energy markets, bridging the gap between theoretical computational models and real-world socioeconomic applications. DA - 2026-05-19 DB - ResearchSpace DP - Univen KW - Blockchain KW - Community energy KW - Demand response KW - Energy prediction KW - P2P energy trading LK - https://univendspace.univen.ac.za PY - 2026 T1 - Machine Learning Applications in Blockchain for Renewable Energy Systems TI - Machine Learning Applications in Blockchain for Renewable Energy Systems UR - ER -