Accurate　quantitative　simulation　of　forest　fires　is　of　great　significance　to　the　risk　control　of　forest　fires.　Cellular　automata　model　is　often　used　to　simulate　forest　fires.　However,　the　traditional　cellular　automata　model＇s　cell　state　setting　is　too　simple,　and　the　conversion　rules　are　not　considered　comprehensively　enough,　resulting　in　unsatisfactory　simulation　results.　In　this　paper,　it　is　proposed　to　learn　the　relationship　between　the　burning　state　of　cells　and　their　own　geographical　conditions　from　the　historical　data　of　forest　fires,　and　to　convert　the　burning　probability　into　the　time　required　for　complete　burning　of　cells,　so　as　to　simulate　different　spreading　speeds　under　different　geographical　conditions.　Experiments　show　that　this　is　a　potentially　effective　method　for　forest　fire　simulation.　©　2022　IEEE.