In　this　paper,　a　method　based　on　least　square　support　vector　machine　(LSSVM)　and　genetic　algorithm　(GA)　is　applied　for　the　coefficient　of　performance　(COP)　prediction　and　the　load　regulation　of　each　chiller　in　the　water　chiller　system.　In　order　to　show　the　generalizability　of　this　method,　two　twin-screw　water　chiller　systems　with　different　nominal　cooling　capacities　applied　in　different　situations　are　studied.　The　proposed　model　uses　two　compressors’　load　rate,　inlet　temperature　of　cooling　water,　outlet　temperature　of　cooling　water,　inlet　temperature　of　condensing　water　and　outlet　temperature　of　condensing　water　as　input　parameters.　COP　is　used　as　the　output　parameter.　To　increase　the　accuracy　of　the　model,　more　than　10,000　on-site　testing　data　points　are　randomly　divided　into　the　training　set　and　the　testing　set　for　each　case.　The　results　show　that　this　GA-LSSVM-based　model　is　accurate　enough　for　COP　prediction.　For　the　first　case,　98.05%　of　total　points　locate　within　the　±　5%　lines　and　determination　coefficient　is　0.9835.　For　the　second　case,　99.66%　of　total　points　locate　within　the　±　5%　lines　and　determination　coefficient　is　0.9907.　Based　on　the　proposed　model　with　high　precision,　two　different　typical　working　conditions　are　used　for　two　cases　to　develop　the　control　strategy　of　each　chiller＇s　load　regulation,　which　is　significantly　helpful　to　improve　the　performance　of　the　water　chiller　system.　©　2020　Elsevier　B.V.