Neural Networks in Solar Cell Data Analysis

Harnessing machine learning for better hybrid perovskite solar cells Hybrid organic–inorganic perovskites have become a leading choice for solar-cell materials, boasting efficiencies above 25%. Yet they still fall short of the theoretical performance limit, sparking a surge of research into multicomponent formulations. By investigating different cations, anions, and structural variations, researchers aim to pinpoint compositions that push solar-cell efficiencies closer to their fundamental threshold. Zhang et al. built a data set of 1,346 hybrid perovskites, each with its band gap calculated using the HSE06 functional. The authors converted structural information into four key descriptors—sine matrix, Ewald sum matrix, atom-centered symmetry functions (ACSF), and many-body tensor representation (MBTR). They then applied 11 machine learning models, including gradient-boosted decision trees (LightGBM, CatBoost, XGBoost), convolutional neural networks (CustomCNN, VGG16, Xception, EfficientNetV2B0), and graph neural networks (GATConv, GCNConv, GCSConv). During training, a 50–50 train–test split was used, with 5-fold cross-validation in the training subset to guard against overfitting. Among these models, CustomCNN combined with MBTR offered a top test set R² of about 0.94, underscoring how detailed structural descriptors plus deep neural architectures can capture subtle structure–property relationships. In follow-up density functional theory calculations, the researchers singled out three candidate perovskite structures—C3H8NSnI3, CH3C(NH2)2SnI3, and (CH3)2NH2SnI3—and found that CH3C(NH2)2SnI3 in particular shows fewer deep-level defects. This lower defect activity can reduce nonradiative carrier losses and improve device efficiency. Overall, the study reveals how targeted machine learning, armed with specialized descriptors, accelerates perovskite screening and identifies formulations that are more likely to yield stable, high-performing solar cells. Paper: https://lnkd.in/dxk7nyKd #MachineLearning #PerovskiteSolarCells #MaterialsScience #BigData #EnergyInnovation #BandGapPrediction #DefectEngineering #GreenTech #CNN #GNN #Photovoltaics #RenewableEnergy #AIResearch #AIforScience #DataDriven #Innovation

[new publication][open access until april 10 2025] Leveraging large-scale aerial data for accurate urban rooftop solar potential estimation via multitask learning, Elsevier Solar Energy, Alessia Boccalatte, Ankit Jha, Jocelyn Chanussot The pdf of the paper can be freely downloaded using this link: https://lnkd.in/dQFpSCcd Convolutional Neural Networks (CNNs) have shown remarkable success in remote sensing tasks. In urban contexts, recent research has utilized CNNs to generate rooftop segmentation masks and determine rooftop section orientation from aerial images. This cost-effective approach is especially valuable for large-scale rooftop solar potential estimations when detailed three-dimensional data is unavailable. This research introduces SolarMTNet, a novel multitask dense-prediction network designed for rooftop solar potential prediction using only aerial images. Unlike previous studies that focus on small manually labeled datasets (approximately 2000 scenes) and only segment rooftop orientations while typically assuming constant slopes, SolarMTNet simultaneously segments both orientations and slopes, enhancing the accuracy of solar potential estimations by 40%. SolarMTNet leverages a large, automatically labeled dataset (up to 280000 scenes) created from open-source Swiss geospatial and aerial data, significantly improving generalization. The model is trained on rooftop data from the Zurich and Geneva cantons and cross-validated on the Canton of Vaud, Switzerland. The results show a mean Intersection over Union (mIoU) of 0.67 for orientation segmentation and 0.40 for slope segmentation. The estimated irradiance exhibits an absolute mean percentage difference of only 5% compared to real solar cadaster data derived from detailed model-based calculations, primarily due to shading issues. Finally, SolarMTNet has also been tested in different geographical areas outside Switzerland (France and Germany), demonstrating consistent performance across diverse regions and pixel resolutions.

Discover how machine vision unlocks new insights into perovskite solar cells in this insightful Viewpoint by Yalan Zhang and Alvin Yuanyuan Zhou of Hong Kong Baptist University, published in Accounts of Materials Research. The authors share their perspectives on using convolutional neural networks to automatically analyze critical grain characteristics in perovskite thin films, specifically grain surface area, grain-boundary grooves, and intra-grain surface fluctuations. These often-overlooked features influence solar cell efficiency and stability. Their proposed machine-learning workflow could accelerate the optimization of perovskite photovoltaics by enabling precise microstructure-performance mapping. Learn how AI-driven image analysis bridges materials characterization and device engineering: https://lnkd.in/girN9qJy #PerovskiteSolarCells #MachineLearning #MaterialsScience #AI #RenewableEnergy #AMR

🔋⚡ Artificial Neural Network (ANN)-Based MPPT for PV Systems with Boost Converter ⚡🔋 The integration of Artificial Neural Networks (ANNs) in solar photovoltaic (PV) systems is transforming how we achieve maximum power extraction under dynamic weather conditions. The diagram above showcases a PV system with ANN-based MPPT control that enhances efficiency and ensures reliable charging of battery storage. 🌞 System Overview 1️⃣ PV Array – Captures solar irradiance and generates electrical power. 2️⃣ Boost Converter – Steps up the PV voltage to the required level for charging the battery. 3️⃣ Battery Load – Stores energy for later use, ensuring system stability. 4️⃣ Dual ANN Controllers (ANN1 & ANN2) – ANN1 takes PV voltage (Vpv) and current (Ipv) as inputs to predict temperature (T) and irradiance (G). ANN2 uses these environmental predictions (T & G) to estimate the optimal reference PV voltage (Vpv*). ⚙️ Control Strategy The predicted Vpv* is compared with the actual Vpv. The error is processed through a PI controller. The signal is then scaled and converted into PWM signals to drive the Boost Converter. This ensures the PV system continuously operates at its Maximum Power Point (MPP), even under varying irradiance and temperature. 🤖 ANN Architecture ANN1: Inputs → Vpv & Ipv; Outputs → T & G. ANN2: Inputs → T & G; Output → Vpv*. Both ANNs employ hidden layers with 10 neurons each, enabling efficient pattern recognition and adaptive control. 🌍 Why It Matters? ✅ Higher energy extraction compared to conventional MPPT algorithms (P&O, INC). ✅ Adaptive learning capability for unpredictable solar conditions. ✅ Enhanced charging efficiency for batteries in standalone and hybrid systems. This approach demonstrates how AI-driven control methods are reshaping the future of renewable energy systems by combining advanced algorithms with traditional power electronics. #renewableenergy #solarpower #artificialintelligence #machinelearning #neuralnetworks #pvsolar #greenenergy #sustainabletechnology #smartgrid #energystorage