Benchmarking reinforcement learning and accurate modeling of ground source heat pump systems: Intelligent strategy using spiking recurrent neural network combined with spider WASP inspired optimization algorithm

• The proposed Spider Wasp Optimization with the Spiking Recurrent Neural Network (SWO-SRNN) model boosts Ground source heat pump (GSHP) efficiency with enhanced predictive accuracy. • Emperor Penguins Colony (EPC) optimization reduces data complexity, improving GSHP performance prediction. • Real-time GSHP thermal imbalance mitigation using intelligent optimization. • Proposed model outperforms conventional methods (R=0.92, RMSE=0.14). Ground source heat pump (GSHP) has recently gained a great attention because of its efficient utilization of geothermal energy for building cooling and heating. However, GSHP systems face significant challenges in real-time applications because of thermal imbalances, and fluctuating cooling loads, demanding a long-term performance prediction mechanism. This study proposed an innovative predictive hybrid strategy leveraging Spider Wasp Optimization with the Spiking Recurrent Neural Network (SWO-SRNN). The SWO is utilized to refine the parameters of SRNN, reducing the model’s loss and training complications. The developed model begins with the collection of datasets representing the parametric modeling of GSHP. Consequently, Emperor Penguins Colony (EPC) optimization algorithm was also employed for selecting the essential features, which reduces the data dimensionality and assists the predictive algorithm to focus on important features in its training phase. Furthermore, the proposed SWO-SRNN was trained using the selected features to predict the ground temperature and Coefficient of Performance (COP), which enables to make appropriate actions to optimize the functioning of GSHP. Finally, statistical analysis was used to evaluate the robustness of the developed SWO-SRNN models. The statistical results prove the effectiveness and superiority of the proposed SWO-SRNN method compared the standalone SRNN model for performance prediction of the GSHP. The simulated results revealed that the deterministic coefficient (R 2 ) and RMSE of the predicted ground temperature were 0.89 and 0.14 for SWO-SRNN, compared to 0.82 and 0.151 for the classical SRNN, respectively. Therefore, SWO-SRNN demonstrated superior predictive accuracy, establishing itself as a highly effective optimization tool for forecasting the energetic performance of GSHPs. These findings highlight the potential of the proposed method to be further explored and extended for real-world applications and future research in intelligent energy systems.