A data-driven framework for seasonal PV forecasting and coordinated EV charging to improve hosting capacity in unbalanced distribution grids

• Seasonal PV forecasting is performed using a hybrid CNN–BiLSTM model. • Coordinated EV charging is optimized using a genetic algorithm. • Voltage unbalance factor is reduced from 4.83% to below 2%. • Transformer loading and feeder losses are significantly reduced. • Hosting capacity is increased to support up to 80% EV penetration. The rapid growth of photovoltaic (PV) generation and electric vehicles (EVs) presents operational challenges for distribution networks, including voltage deviations, increased losses, phase unbalance, and transformer overloading. These effects constrain the network’s hosting capacity (HC) and limit the integration of clean energy resources. Existing studies often simplify PV and EV profiles deterministically, neglect seasonal variability, and focus narrowly on voltage or thermal limits, thereby overlooking critical indicators such as the voltage unbalance factor (VUF). This paper proposes a holistic framework to enhance hosting capacity in unbalanced distribution systems with high PV–EV penetration. A hybrid CNN-BiLSTM model is developed to forecast seasonal PV generation with a high coefficient of determination (R2 = 0.9707), while stochastic EV charging profiles capture user-driven variability. A Genetic Algorithm (GA)-based coordinated optimization aligns EV charging with PV availability and network constraints. Case studies on the modified IEEE 13-bus feeder demonstrate that the proposed method reduces maximum VUF from 4.83% to below 2%, maintains bus voltages within 0.999–1.049 pu, minimizes losses, and prevents transformer overloading, enabling secure operation up to 80% EV penetration. The results confirm that coordinated PV–EV management provides a scalable and practical strategy for distribution system operators to support high renewable adoption without major infrastructure upgrades.