Intelligent MPPT in PV Systems using metaheuristic-optimised deep neural networks

Solar photovoltaic (PV) systems face problems with multiple local maxima, poor tracking efficiencies and unstable maximum power point tracking (MPPT) when the irradiance changes rapidly or when there is partial shading on the panels. Traditional MPPT methods including Perturb and Observe (P&O) and Incremental Conductance display slow convergence and oscillation around the steady state in the changing environment. This paper proposes a Digital Twin-based Grey Wolf Optimization (GWO-Deep Neural Network) intelligent MPPT system for solar PV systems to overcome these two issues.This paper proposes a hybrid Grey wolf optimization –Deep neural Network (GWO_DNN) based MPPT technique for photovoltaic systems. The comparison between hybrid deep neural network (DNN)-based controllers for maximally tracking MPPs through Grey Wolf Optimisation (GWO), Golden Eagle Optimisation (GO) and Moth-Flame Algorithm (MFA) are evaluated by comparing relative effectiveness based upon converging performance, prediction accuracy, computational efficiency and stability. The proposed method improves tracking efficiency, reduces steady-state oscillations and achieves faster convergence under varying irradiance conditions. Computational models show GWO-DNN controllers achieving lower test mean absolute distance from the true MPP compared to the other controllers with mean absolute mean square errors (MAE) reported as 0.6138 V (GWO-DNN), 0.3506 V (MFA-DNN) and 0.4978 V (GO-DNN). The GWO-DNN controllers also have a lower degree of oscillation in steady-state conditions, showing faster convergence than all other controllers as well as significantly lower prediction errors. In contrast, Moses-DNN performs more reliably with respect to optimal positioning, but its predictive capabilities will limit its effectiveness for practical real-time implementation. Although GO-DNN provides lower inference time than both the GWO app and MFA model as is a whole, the standard deviation of how far down the MPP would be estimated will increase for the time taken to make the prediction. A traditional CNN-LSTM model has demonstrated to consistently underperform under dynamic operating conditions. Thus, the results demonstrate that GWO-DNN represents a superior balance of MPPT controller characteristics across dynamic operating ranges, as such, GWO-DNN will provide the most robust and predictable MPP capability for use in smart solar energy applications. This initiative helps meet the UN Sustainable Development Goals through the past improvements made in renewable energy extraction efficiency, the creation of more intelligent energy infrastructure and ultimately the contribution to greater levels of solar energy production, while also reducing wasted energy through better designs. Specifically, it contributes to SDG7 - Affordable and Clean Energy; SDG9 - Industry, Innovation and Infrastructure; and SDG13 - Climate Action by facilitating the creation of sustainable, resilient and low-carbon energy systems.