Scholarlink

<p>Accurate performance modelling is essential for designing and operating efficient domestic parabolic dish solar water heaters, particularly in off-grid regions. While artificial neural networks (ANNs) are effective for this nonlinear task, their performance depends heavily on the training algorithm. This study investigates the novel application of two metaheuristic-optimized ANN hybrids—ANN trained with a Genetic Algorithm (ANN-GA) and ANN trained with Grey Wolf Optimization (ANN-GWO)—to predict the thermal efficiency of a parabolic dish system. Using experimental data, both models were developed and compared based on predictive accuracy and computational efficiency. The ANN-GA model achieved superior accuracy with an overall coefficient of determination (R²) of 0.9969. In contrast, the ANN-GWO model converged approximately 478 times faster (in 3&#xa0;s versus 1430&#xa0;s) while maintaining a high R² of 0.9700. Five-fold cross-validation further confirmed the robustness of the models, with ANN-GA achieving mean R², root mean square error (RMSE) and mean absolute percentage error (MAPE) values of 0.9692, 0.4190 and 0.6932%, respectively, while ANN-GWO achieved corresponding values of 0.9495, 0.6734 and 1.2323%, with substantially lower training time. A correlation-based input relevance analysis showed that wind speed had the strongest negative association with collector efficiency, while useful heat gain, mass flow rate, and solar radiation showed positive associations, confirming that the prediction problem is governed by coupled optical, flow, and heat-loss mechanisms rather than by solar radiation alone. The results demonstrate a clear trade-off: ANN-GA is ideal for high-precision design and simulation, whereas ANN-GWO is exceptionally suited for applications requiring rapid convergence, such as real-time system control or monitoring. This work provides a practical framework for selecting hybrid ML models tailored to specific constraints in sustainable solar thermal system development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance prediction of a domestic parabolic dish solar water heater using ANN-GA and ANN-GWO hybrid models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月14日

查看全文
  • Kantu T. Kabengele,
  • Isaac O. Olayode,
  • Lagouge K. Tartibu

摘要

Accurate performance modelling is essential for designing and operating efficient domestic parabolic dish solar water heaters, particularly in off-grid regions. While artificial neural networks (ANNs) are effective for this nonlinear task, their performance depends heavily on the training algorithm. This study investigates the novel application of two metaheuristic-optimized ANN hybrids—ANN trained with a Genetic Algorithm (ANN-GA) and ANN trained with Grey Wolf Optimization (ANN-GWO)—to predict the thermal efficiency of a parabolic dish system. Using experimental data, both models were developed and compared based on predictive accuracy and computational efficiency. The ANN-GA model achieved superior accuracy with an overall coefficient of determination (R²) of 0.9969. In contrast, the ANN-GWO model converged approximately 478 times faster (in 3 s versus 1430 s) while maintaining a high R² of 0.9700. Five-fold cross-validation further confirmed the robustness of the models, with ANN-GA achieving mean R², root mean square error (RMSE) and mean absolute percentage error (MAPE) values of 0.9692, 0.4190 and 0.6932%, respectively, while ANN-GWO achieved corresponding values of 0.9495, 0.6734 and 1.2323%, with substantially lower training time. A correlation-based input relevance analysis showed that wind speed had the strongest negative association with collector efficiency, while useful heat gain, mass flow rate, and solar radiation showed positive associations, confirming that the prediction problem is governed by coupled optical, flow, and heat-loss mechanisms rather than by solar radiation alone. The results demonstrate a clear trade-off: ANN-GA is ideal for high-precision design and simulation, whereas ANN-GWO is exceptionally suited for applications requiring rapid convergence, such as real-time system control or monitoring. This work provides a practical framework for selecting hybrid ML models tailored to specific constraints in sustainable solar thermal system development.