Hydrogen　is　a　promising　way　to　achieve　high　efficiency　and　low　emissions　for　Wankel　rotary　engines.　In　this　paper,　the　intake　and　exhaust　phases　and　excess　air　ratios　(lambda)　were　optimized　using　machine　learning　(ML)　and　genetic　algorithm　(GA).　Firstly,　a　one-dimensional　model　was　built　and　verified　under　various　lambda.　Secondly,　the　variables　were　determined　using　sensitivity　analysis　method,　and　the　sample　for　training　models　was　generated　using　the　Latin　hypercube　sampling.　Finally,　a　prediction　model　for　performance　and　emissions　was　built　using　ML　and　combined　with　GA　for　multi-objective　optimization.　The　results　show　that　the　timing　of　intake　port　full　closing　(IPFC)　and　exhaust　port　start　opening　(EPSO)　exhibits　the　most　significant　influence　on　performance　and　emissions,　while　the　other　phases　are　less　influential.　Both　indicated　mean　effective　pressure　(IMEP)　and　indi-cated　specific　nitrogen　oxides　(ISNOx)　increase　as　the　IPFC　timing　is　advanced,　while　indicated　specific　fuel　consumption　(ISFC)　decreases　as　EPSO　timing　is　delayed.　Compared　with　the　original　engine,　the　optimized　IMEP　is　improved　by　0.18%,　ISFC　is　reduced　by　2.39%,　and　ISNOx　is　reduced　by　up　to　65.43%.　It　is　an　efficient　way　to　use　ML　combined　with　GA　to　improve　performance　and　reduce　emissions　simultaneously.