Heat-exchange　station　is　the　key　mid-link　between　heat　source　and　heat　users　in　district　heating　system,　playing　a　role　in　distributing　heat　and　regulating　supply　and　demand.　Accurate　load　prediction　at　the　level　of　heatexchange　stations　and　then　regulating　the　heat　supply　is　the　important　step　to　optimize　district　heating　system.　Based　on　real　and　limited　data,　the　framework　of　heat-exchange　station　load　prediction　method　is　established　in　this　study,　which　includes　5　procedures　＇data　preprocessing　method,　input　variable　selection,　solar　radiation　substitution,　prediction　model　establishment　and　model　transplantation　application＇.　The　sliding　box　plot　method　was　proposed　for　data　preprocessing,　and　the　effects　of　data　preprocessing　under　the　widths　of　24　h,　48　h,　72　h,　96　h　and　120　h　were　compared,　with　48　h　as　the　optimal　sliding　window　width.　Support　Vector　Regression　(SVR)　algorithm　was　used　to　establish　a　heating　load　prediction　model　for　heat-exchange　station　in　residential　buildings.　And　parameter　substitution　was　made　for　solar　radiation　data　that　cannot　be　obtained　in　weather　forecasts.　High　accuracy　was　obtained　with　a　MAPE　of　5.80　%　and　RMSE　of　0.18　GJ.　The　above　model　was　transplanted　and　applied　in　three　other　heat-exchange　stations　of　different　building　types.　The　MAPEs　were　5.51　%,　7.79　%　and　6.54　%,　with　corresponding　RMSEs　of　0.017　GJ,　0.04　GJ　and　0.04　GJ,　respectively,　which　all　achieved　good　prediction　results.　The　research　results　provide　certain　technical　support　for　the　operation　and　regulation　of　district　heating　systems.