Browsing by Author "Weeratunge, H"

Filter results by typing the first few letters
Now showing 1 - 3 of 3
  • Thumbnail Image
    PublicationOpen Access
    Forecasting renewable energy for microgrids using machine learning
    (Springer Nature, 2025-05-03) Sudasinghe, P; Herath, D; Karunarathne, I; Weeratunge, H; Jayasuriya, L
    Microgrids, comprised of interconnected loads and distributed energy resources, function as single controllable entities with respect to the main grid. However, the inherent variability of distributed wind and solar generation within microgrids presents operational stability challenges concerning voltage regulation and frequency stability. Accurate forecasting of renewable generation is crucial for mitigating these challenges. This work proposes a one-dimensional Convolutional Neural Network (1-D CNN) based approach to forecast photovoltaic (PV) generation and wind energy, using data from the University of California, San Diego microgrid and San Diego Airport weather records. The proposed method is evaluated against various forecasting methods using key metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared value. Results show that the 1-D CNN model achieves an improvement of up to 229.8 times in MSE and a 24.47 fold improvement in MAE compared to baseline models that use traditional statistical methods in forecasting. This demonstrates the potential of machine learning for enhancing microgrid management, particularly in short-term forecasting of renewable generation. © The Author(s) 2025. Evaluated ML-based renewable energy forecasting models by implementing 1-D CNN and LSTM models using real-world data. Proposed 1-D CNN performs better than LSTM and baseline models, achieving higher accuracy and computational efficiency. Accurate forecasting of PV and wind energy generation enhances grid stability, reduces backup power dependency, and supports sustainable energy integration.
  • Thumbnail Image
    PublicationOpen Access
    Interpretable SHAP-bounded Bayesian optimization for underwater acoustic metamaterial coating design
    (Springer Science and Business Media Deutschland GmbH, 2025-09-08) Weeratunge, H; Robe, D. M; Hajizadeh, E
    We present an interpretability-informed Bayesian optimization framework for the inverse design of underwater acoustic coatings composed of polyurethane elastomers with embedded metamaterial features. A data-driven model was used to capture the relationship between acoustic performance, specifically, sound absorption and the corresponding geometrical design variables. To interpret these relationships, we applied SHapley Additive exPlanations (SHAP), enabling the identification of key parameters influencing the objective function and providing both global and local insights into their effects. The insights from the SHAP analysis were used to automatically refine the bounds of the design space, guiding the optimization process toward more promising regions. This approach was tested on two polyurethane materials with different hardness levels and yielded improved optimal designs compared to standard Bayesian optimization without increasing the number of simulations. This work underscores the effectiveness of combining interpretability techniques with optimization for the efficient and cost-effective design of underwater acoustic metamaterials under strict computational constraints and can be generalized towards other materials and engineering optimization problems
  • Thumbnail Image
    PublicationOpen Access
    Topology and size optimization of trusses using graph neural networks: Towards efficient surrogate modeling
    (Elsevier Ltd, 2026-06) Ariyasinghe, N; Wickremasinghe, T; Weeratunge, H; Mallikarachchi, C; Herath, S
    Real-time structural optimization of trusses using machine learning techniques, incorporating both topology and size optimization, is more effective in discrete domains than in continuous ones, with Graph Neural Networks (GNNs) showing strong potential. However, the impact of convolutions in GNNs is not yet fully understood, limiting their full applicability. This paper presents a GNN-based surrogate model for real-time structural optimization and identifies the most suitable convolution type for this task. The study assesses the predictive performance of models trained on a dataset of optimized structures spanning a range of loads, boundary conditions, and design domain sizes. The resulting model effectively predicts both optimal topology and member sizes once the design parameters are provided. Eight graph convolution types are investigated to identify the most suitable method, alongside an evaluation of optimal network architectures. Among the tested approaches, Generalised Graph Convolution achieves the highest accuracy, followed by Topology Adaptive Graph Convolution, producing near-ideal topology predictions for most test cases and maintaining section size prediction errors within ±2% across all test data points. This framework demonstrates strong potential for broader applications.