A Neural Network Trained by Multi-Tracker Optimization Algorithm Applied to Energy Performance Estimation of Residential Buildings

Energy performance analysis in buildings is becoming more and highlighted, due to the increasing trend of energy consumption building sector. Many studies have declared great potential soft computing for this analysis. A particular methodology sense employing hybrid machine learning that copes with drawbacks single methods. In work, an optimized version a popular model, namely feed-forward neural network (FFNN) used simultaneously predicting annual thermal demand (ATED) weighted average discomfort degree-hours (WADDH) by analyzing eleven input factors represent circumstances. The optimization task carried out multi-tracker algorithm (MTOA) which powerful metaheuristic algorithm. Moreover, three benchmark algorithms including slime mould (SMA), seeker (SOA), vortex search (VSA) perform same comparison purposes. accuracy models assessed using error correlation indicators. Based on results, MTOA (with root mean square errors 2.48 5.88, along Pearson coefficients 0.995 0.998 ATED WADHH, respectively) outperformed techniques behavior building. This could optimize 100 internal variables FFNN acquire WADHH excellent accuracy. Despite different rankings four prediction phase, 9.84 95.96, 0.972 0.997 was still among best, altogether, FFNN-MTOA recommended promising applications real-world projects.