Optimizing ANFIS for heating load forecasting: a hybrid approach for enhanced accuracy in district heating systems

Accurate forecasting of next-day heat load curves is essential for ensuring adequate heat supply and achieving optimal district heat system (DHS) operation. While most previous studies focused on one-step-ahead forecasting using traditional approaches, this study enhances predictive performance by integrating advanced enhancement schemes into the Adaptive Neuro-Fuzzy Inference System (ANFIS). Three hybrid models were developed: ANFIS_ROA (with Rider Optimization Algorithm), ANFIS_DAO (with Dynamic Arithmetic Optimization Algorithm), and ANFIS_EBSCO (with Escaping Bird Search for Constrained Optimization). Experimental results show that the ANFIS_EBSCO model achieved the highest accuracy with an R2 of 0.985 and RMSE of 1.197 during the test phase, outperforming both ANFIS_DAO (R2 = 0.952, RMSE = 1.523) and ANFIS_ROA (R2 = 0.944, RMSE = 1.678). By contrast, the baseline ANFIS model recorded the lowest performance (R2 = 0.891, RMSE = 2.993). These results demonstrate that incorporating metaheuristic optimization significantly improves forecasting accuracy, with ANFIS_EBSCO emerging as a robust tool for efficient DHS operation. Graphical presentation of research process