Adaptive neuro-fuzzy modeling for real-time solar irradiance prediction using PV module operating parameters

• Real-time solar irradiance prediction using only V mp , I mp , and temperature—no pyranometer or module disconnection required. • Introduces an ANFIS model trained on synthetic SDM data across multiple irradiance and temperature conditions. • Achieves near-perfect prediction accuracy with RMSE of 1.05689 W/m² and R² of 0.99999 using Gaussian membership functions. • Validated experimentally via IoT-based PV monitoring system, achieving RMSE of just 0.0328 W/m². • Offers a scalable, low-cost, sensorless solution for embedded and on-grid PV system monitoring and MPPT. Precise estimation of solar irradiance is fundamental for the effective operation, monitoring, and forecasting of photovoltaic (PV) systems. Pyranometers are the standard for solar irradiance measurement, but their cost, sensitivity, and frequent recalibration make them less practical for large-scale use. This study proposes a novel data-driven approach for real-time irradiance prediction based on an Adaptive Neuro-Fuzzy Inference System (ANFIS). The proposed model uniquely leverages directly measurable electrical parameters of PV modules—namely, the voltage at the maximum power point (V mp ), current at the maximum power point (I mp ), and cell temperature (T)—to predict solar irradiance (G) without requiring module disconnection. This enables continuous, non-invasive monitoring, suitable for embedded and grid-connected PV systems. A synthetic dataset comprising 4,000 samples was generated using a MATLAB-based Single-Diode Model (SDM) for a SunPower SPR-X20-250-BLK PV module. Simulations were conducted under four temperature conditions (15°C, 25°C, 45°C, and 65°C) and ten irradiance levels (100 to 1000 W/m²), resulting in 40 (I–V) and (P–V) curves representing diverse environmental scenarios. The ANFIS model was trained and evaluated using eight different membership function (MFs) types and varying MFs counts. The optimal configuration—Gaussian membership function (gaussmf) with 10 MFs—achieved outstanding predictive performance with a Root Mean Square Error (RMSE) of 1.05689 W/m², Mean Absolute Error (MAE) of 0.41864 W/m². Further, an experimental validation was conducted using a custom-built Internet of Things (IoT)-based PV monitoring system comprising three 8 W PV modules (total power: 24 W), an ESP32-based data acquisition unit, and a solar panel WS400A multimeter. The system recorded V mp , I mp , V oc , I sc , and T under real outdoor conditions. The trained ANFIS model, when tested on this experimental data, yielded a predicted irradiance value of 806.30 W/m² with an RMSE of 0.0328 W/m², affirming the model’s capability to maintain high accuracy under minimal input variation. This research demonstrates the efficacy of ANFIS for solar irradiance prediction using operational PV data, offering a viable alternative to traditional measurement systems.