Compressed　air　energy　storage　(CAES)　possesses　the　advantages　of　high　reliability,　good　economic　performance,　longer　discharge　time,　extended　service　life,　and　comprehensive　utilization　of　heat,　cold,　and　electricity.　In　this　study,　a　CAES　test　bench　based　on　a　pneumatic　motor　(PM)　is　built,　and　the　output　performance　of　the　CAES　system　is　investigated　under　the　variable　working　conditions　represented　by　the　change　of　the　compressed　air　pressure　and　the　fluctuation　of　the　load　demand.　The　effects　of　major　parameters　like　torque,　voltage,　current,　rotation　speed,　and　regulated　pressure　on　the　power　output　and　energy　efficiency　of　the　PM　and　generator　are　analyzed.　By　the　evaluation　of　learning　rate　difference,　number　of　hidden　layer　neural　networks,　and　training　function,　a　prediction　model　of　CAES　based　on　artificial　neural　network　(ANN)　is　developed.　By　checking　the　mean-square　error　and　related　coefficient,　the　ANN　model　is　validated　by　subsequent　experimental　results.　Finally,　to　predict　the　maximum　power　output　and　energy　efficiency　of　the　PM　and　generator,　genetic　algorithm　is　supplemented　for　the　parameters　optimization.　From　the　result,　it　can　be　found　that　the　high　power　output　range　of　the　PM　is　concentrated　in　the　low　voltage,　high　current　and　medium-to-high　rotation　speed　regions.　Likewise,　the　low　compressed　air　consumption　rate　of　PM　is　concentrated　mainly　in　the　regions　of　low　voltage,　low　rotation　speed,　high　current　and　high　torque.　When　the　regulated　pressure　reaches　10.5　bar　and　the　current　value　sync　reaches　13　A,　PM　can　generate　a　maximum　output　power　of　approximate　658　W.　©　2023　Elsevier　Ltd