To　address　the　operation　optimization　problem　of　integrated　energy　system,　an　electric-thermal-gas　integrated　energy　system　model　consisting　of　gas　turbine,　waste　heat　boiler,　organic　Rankine　cycle,　air　source　heat　pump　and　integrated　demand　response　model　is　established.　Then,　an　operation　optimization　method　based　on　soft　actor-critic　algorithm　is　proposed.　Firstly,　the　integrated　energy　system　framework　and　equipment　model　are　built.　For　the　problem　that　the　traditional　integrated　demand　response　model　is　not　accurate,　a　neural　network　model　which　can　represent　the　real　response　capability　of　users　is　established　by　combining　historical　data　and　the　gate　recurrent　unit.　The　neural　network　model　is　applied　to　energy　pricing　scenarios.　Secondly,　an　integrated　energy　system　economic　dispatch　model　is　established　with　the　objective　of　minimizing　system　energy　purchase　cost　and　wind　and　photovoltaic　power　curtailment　cost.　A　deep　reinforcement　learning　framework　is　used　to　formulate　the　optimization　problem.　The　action　space,　state　space,　and　reward　function　of　the　soft　actor-critic　agent　interaction　process　with　the　environment　are　set　up.　The　trained　model　can　be　directly　used　for　real-time　decision　making　without　further　iterative　computation.　The　simulation　results　show　that　the　proposed　method　can　effectively　perform　energy　management　and　energy　pricing　optimization　to　reduce　the　overall　operation　cost　of　the　system.　©　2022　Science　Press.　All　rights　reserved.