Considering　the　difficulty　to　built　mechanisms　models　for　complex　industrial　processes,　this　paper　proposes　a　data-driven　modelling　method.　This　method　includes　a　process　for　processing　real　industrial　data　and　an　improved　deep　learning　algorithm.　The　data　processing　flow　completes　the　basic　preparation　work　for　the　raw　data,　and　the　improved　algorithm　calculates　the　model　based　on　this.　The　improved　algorithm　is　Arima-ACNN-LSTM　coupling　algorithm,　which　can　enhance　the　original　LSTM　algorithm　by　adopting　CNN　and　Attention　mechanisms.　The　experimental　results　show　that,　the　accuracy　of　the　model　using　the　proposed　improved　algorithm　is　better　than　that　using　the　ARIMA　algorithm,　the　LSTM　algorithm　and　the　ARIMA-LSTM　coupling　algorithm.　So　this　paper　provides　an　inspiring　approach　for　industrial　processes　modelling　with　complex　characteristics.　©　Published　under　licence　by　IOP　Publishing　Ltd.