ScholarWorks@성균관대학교

초록

Simulating the performance of optoelectronic materials under complex and variable environmental conditions in laboratory settings presents a significant challenge, as laboratory environments often fail to accurately replicate real-world conditions. This limitation hinders the reliability of performance assessments for optoelectronic materials in practical applications. To address this limitation, this research introduces EcoCurrentNet, an innovative model that integrates deep neural networks (DNN) with convolutional layers and residual blocks, combined with a CatBoost regression layer, to effectively capture spatial and nonlinear dependencies. The convolutional component learns the interactions between material features and 12 environmental variables, while residual blocks enhance training stability and gradient flow across deeper layers. A global average pooling and fully connected layer compress the learned features before they are passed to the CatBoost regressor, which iteratively refines the final output. The model is designed based on principles of thermodynamics and material science, particularly the complex relationships between material properties and external factors such as temperature, pressure, and light intensity, which can be described by physical laws governing material behavior. The model achieves a remarkable R2 score of 99.68%, demonstrating its capability to provide accurate assessments of material behavior beyond controlled laboratory conditions. This hybrid architecture illustrates the potential of combining deep residual learning and gradient boosting for modeling complex physical systems, offering a more reliable and efficient approach to material design and paving the way for future innovations in the field.