DEEP NEURAL NETWORK-CONTROLLED ISOLATED THREE-PORT INTERLEAVED FLYBACK BOOST CONVERTER FOR SMART LOCAL ENERGY GOVERNANCE IN GRID-CONNECTED PHOTOVOLTAIC SYSTEMS

Povzetek

The energy management architectures for grid-connected PV energy systems must be high-efficiency, multi-port power conversion topologies that can manage energy simultaneously from PV energy resources, battery storage and the utility grid; but conventional isolated PV power conversion suffers from poor dynamic response, suboptimal switching coordination, and limited adaptability to variable irradiance and load conditions. The conventional control methods such as LSTM networks, GA-tuned controllers and traditional ML classifiers suffer from various drawbacks, including slow convergence, high computational requirements, and lack of transferability from nonlinear operation of three-port topologies. For isolated three-port interleaved flyback boost converter (ITIFBC), this work proposes a novel DNN controlled ITIFBC based on a multilayer perceptron (MLP), which is trained using 12,000 pre-processed operating samples with feature normalization, SMOTE oversampling and wrapper-based RFE. The DNN, optimized by Adam, controls the real-time duty-cycle modulation and interleaving phase-shifting of the three ports, maximizing power transfer efficiency on all three ports concurrently. Evaluated using MAE, RMSE, η, THD, VRA, and PF, the proposed system achieves η = 97.4%, THD = 1.37% (−62.3% over PI), RMSE = 0.0031, R² = 0.9981, dynamic response improvement of 41.7% over LSTM, VRA = 99.2%, and PF = 0.991.Overall, DNN-driven control holds significant promise for improving the reliability, scalability, and energy output of PV systems in grid-connected applications and smart microgrids, making it a valuable solution for future renewable energy integration and smart grid applications.