Service-Oriented　Computing　provides　a　technical　support　to　AmI　vision.　However,　the　operation　and　management　of　semantic　Web　service　which　is　on　the　basis　of　ontology　reasoning　is　low-efficiency　and　time-consuming,　can　hardly　adapt　to　real-time　applications　in　AmI　system.　So　this　paper　proposed　the　pretreatment-based　SMART　policy,　and　presented　the　scenario　division　as　well　as　its　corresponding　composite　models　of　the　multi-granularity　semantic　service　network.　The　storage　management　and　the　access　mechanism　of　the　multi-granularity　semantic　Web　service　network　were　also　stated　in　detail.　Last,　results　of　experiments　show　that　the　SMART　policy,　deployed　in　the　AmI　system,　performed　a　satisfying　real-time　access　performance,　improved　resource　utilization　and　enhanced　the　user　experience.