Integrated deep learning and image processing method for modeling energy loss due to shadows in solar arrays

• Shading loss analysis reveals peak losses of 17.67% in April. • Developed methodology demonstrates 2.35% error margin in shading loss calculation. • Deep learning method improves PV array detection by 32% compared to image processing. • Strategic placement of PV arrays is important to reduce the shading losses. Shading poses a serious challenge to photovoltaic (PV) systems power generation, causing energy losses of up to 40 %, leading to power mismatches, hotspot formation, and accelerated module degradation. Accurate modelling and simulation of shading are critical for improving photovoltaic (PV) performance. This study develops two shadow‑detection pipelines: (i) the Classic Hough Transform (CHT) combined with K‑means segmentation and (ii) a new Deep Hough Transform (DHT) that learns semantic line features without the need for PV‑specific training data. A 1-kilowatt capacity solar array with planned shading devices was developed and used to perform the experimental analysis. The proposed methodology achieved an accuracy of 0.85, indicating a 32.81 % improvement in solar array detection compared to CHT methods. Computed energy losses due to shading were within 0.5 % to 1.9 % of System Advisor Model (SAM’s) simulated losses. Evaluation with transient shadows from a pedestrian, vehicle, and cloud showed an average mIoU of 81.8 % highlighting the methods advantage over existing 3D modeling-based simulation software. Statistical analysis confirmed the method’s consistency, yielding Dice = 0.857 (95 % CI 0.728–0.943) and mIoU = 0.771 (CI 0.595–0.893). The parametric analysis highlighted the time of day and number of obstructions as key factors influencing shading on solar arrays, with mornings and evenings experiencing over 6 % shading loss variations. Overall, this integrated approach develops robust, real-time modelling and simulation for optimizing large solar energy systems.