The　heat　exchange　station　is　an　essential　part　of　the　central　heating　system.　In　the　actual　heating　system　operation　control,　the　short-Term　heating　load　prediction　of　the　heat　exchange　station　based　on　historical　operation　data　plays　an　important　role.　In　this　paper,　firstly,　based　on　the　Long　Short-Term　Memory　(LSTM)　neural　network,　a　short-Term　heating　load　prediction　model　of　heat　exchange　station　is　established.　Secondly,　because　of　the　difficulty　of　adjusting　parameters　of　the　traditional　LSTM　neural　network,　a　Dynamic　Auxiliary　Individual　Genetic　Algorithm　(DAIGA)　was　proposed.　Then,　the　primary　hyperparameters　of　the　prediction　model　were　optimized　by　the　proposed　genetic　algorithm　to　make　the　prediction　performance　of　the　model　more　accurate　and　stable.　Finally,　through　comparison　experiments　with　a　variety　of　typical　heating　load　prediction　models,　the　proposed　DAIGA-LSTM　prediction　model　has　strong　applicability　and　good　prediction　performance.　©　2021　IEEE.