Energy consumption estimation of Diyarbakir maternity hospital with LSTM and gru

Predicting energy consumption in hospitals, which are critical institutions, is vitally important because accurate forecasting is strategically necessary for both cost optimization and ensuring energy supply security. Numerous studies in the literature have addressed energy consumption forecasting, showing varying degrees of success across different sectors. However, significant fluctuations observed in the accuracy rates of existing forecasting studies clearly indicate that research in this area has not yet reached the desired standard. Consequently, there is a clear need for models capable of predicting energy consumption more accurately. Motivated by this need, this study aims to forecast the energy consumption of Diyarbakır Maternity Hospital using half-hourly energy consumption data recorded between June 1, 2017, and June 30, 2024, employing Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. These models were trained using different time windows of 10, 20, 30, 60, 120, and 240 steps, and their performances were comparatively evaluated across these varying window sizes. The analyses revealed that the LSTM model performed best with a 60-step window, while the GRU model achieved its optimal performance with a 120-step window. When examining performance metrics, the LSTM model resulted in an RMSE of 25.07, an MAE of 13.87, and an R² of 0.98. Similarly, the GRU model achieved an RMSE of 25.11, an MAE of 13.73, and an R² of 0.98. The fact that both models reached an R² value of approximately 0.98 indicates that the predictions closely reflect actual energy consumption values. Consequently, LSTM and GRU-based models can be considered effective approaches for energy management in complex structures such as hospitals.