With　the　development　of　artificial　intelligence,　more　investors　are　applying　machine　learning　and　deep　learning　algorithms　to　financial　time　series,　aiming　to　improve　traditional　quantitative　trading　strategies.　However,　most　of　the　current　research　focus　on　mature　institutional-type　markets　and　strategies　proposed　by　the　current　research　lack　dynamic　adjustment　capability,　especially　in　a　complex　environment　with　high　volatility　and　strong　noise,　are　difficult　to　derive　a　stable　and　profitable　portfolio.　In　response　to　the　above　problems,　we　fully　combine　the　advantages　of　artificial　intelligence　and　traditional　financial　market　theory　techniques　to　construct　a　quantitative　trading　strategy　model　MTL-DDPG　(Multi　Time-scale　LSTM　Deep　Deterministic　Policy　Gradient)　that　can　obtain　excess　returns.　By　conducting　back　testing　and　comparing　with　other　classical　quantitative　trading　strategies,　we　verify　that　our　proposed　quantitative　trading　strategy,　MTL-DDPG,　achieves　good　results　in　the　A-share　market　by　obtaining　excess　returns,　and　the　selected　portfolio　outperforms　the　benchmark　both　in　terms　of　total　return　and　risk　management.　©　2023　ACM.