Mamodiya Udit, Kishor Indra, Almaiah Mohammed Amin, Hamdi Monia, Shehab Rami, Alkhdour Tayseer
Faculty of Engineering and Technology, Poornima University, Jaipur, Rajasthan, India.
Poornima University, Jaipur, Rajasthan, India.
Sci Rep. 2025 Jul 2;15(1):23492. doi: 10.1038/s41598-025-06830-z.
Maximizing output from renewable solar panels requires higher efficiency. Conventionally, such optimization techniques-MPPT (Maximum Power Point Tracking) along with heuristic algorithms-suffer significantly from slow adaptability and track sub optimality under dynamic environments. This article proposes a numerical modeling framework from hybrid AI models, combining physics-informed neural networks and RL for real-time optimization of orientation in solar panels. The methodology uses numerical modeling for precise energy transformation analysis, and deep learning-based optimization dynamically adjusts the angles of panels to maximize power output. A self-learning adaptive neural network is developed to improve tracking accuracy based on real-time irradiance and temperature variations. Moreover, an Edge AI architecture is introduced to make low-latency decisions with reduced dependency on cloud computation, thus improving the efficiency of the system. Besides, an advanced hybrid model based on CNN-LSTM is applied to solar energy forecasting for predictive control of the maximum energy yield. Experimental validation was performed using UTL 335W and 330W PV modules, where real-time data acquisition was followed by AI-driven optimization. Results show an increase in energy yield by 10-15% compared to traditional MPPT systems, while computations are performed 40-50% faster using AI-based numerical modeling. The proposed approach achieves 25% lower forecasting error (RMSE/MAE) and 30% reduced power consumption through Edge AI implementation. This study sets up a new paradigm for AI-integrated solar optimization, which ensures real-time adaptability and enhanced performance in practical deployment. The findings advance the intelligent solar tracking and set a new benchmark for AI-driven renewable energy management.
要使可再生太阳能板的输出最大化,需要更高的效率。传统上,诸如最大功率点跟踪(MPPT)以及启发式算法等优化技术,在动态环境下存在适应性差和跟踪次优性的显著问题。本文提出了一种基于混合人工智能模型的数值建模框架,将物理信息神经网络和强化学习相结合,用于太阳能板方向的实时优化。该方法使用数值建模进行精确的能量转换分析,基于深度学习的优化动态调整板的角度以最大化功率输出。开发了一种自学习自适应神经网络,以根据实时辐照度和温度变化提高跟踪精度。此外,引入了边缘人工智能架构,以在减少对云计算依赖的情况下做出低延迟决策,从而提高系统效率。此外,基于卷积神经网络-长短期记忆网络(CNN-LSTM)的先进混合模型被应用于太阳能预测,以实现最大能量产量的预测控制。使用UTL 335W和330W光伏模块进行了实验验证,在实时数据采集之后进行人工智能驱动的优化。结果表明,与传统的MPPT系统相比能量产量提高了10%-15%,而使用基于人工智能的数值建模计算速度快40%-50%。通过边缘人工智能实现,所提出的方法预测误差(均方根误差/平均绝对误差)降低了25%,功耗降低了30%。本研究为人工智能集成太阳能优化建立了一个新范式,确保了实际部署中的实时适应性和增强性能。这些发现推动了智能太阳能跟踪的发展,并为人工智能驱动的可再生能源管理设定了新的基准。