In　wastewater　treatment　systems,　extracting　meaningful　features　from　process　data　is　essential　for　effective　monitoring　and　control.　However,　the　multi-time　scale　data　generated　by　different　sampling　frequencies　pose　a　challenge　to　accurately　extract　features.　To　solve　this　issue,　a　multi-timescale　feature　extraction　method　based　on　adaptive　entropy　is　proposed.　Firstly,　the　expert　knowledge　graph　is　constructed　by　analyzing　the　characteristics　of　wastewater　components　and　water　quality　data,　which　can　illustrate　various　water　quality　parameters　and　the　network　of　relationships　among　them.　Secondly,　multiscale　entropy　analysis　is　used　to　investigate　the　inherent　multi-timescale　patterns　of　water　quality　data　in　depth,　which　enables　us　to　minimize　information　loss　while　uniformly　optimizing　the　timescale.　Thirdly,　we　harness　partial　least　squares　for　feature　extraction,　resulting　in　an　enhanced　representation　of　sample　data　and　the　iterative　enhancement　of　our　expert　knowledge　graph.　The　experimental　results　show　that　the　multi-timescale　feature　extraction　algorithm　can　enhance　the　representation　of　water　quality　data　and　improve　monitoring　capabilities.　(c)　2024　The　Chemical　Industry　and　Engineering　Society　of　China,　and　Chemical　Industry　Press　Co.,　Ltd.　All　rights　are　reserved,　including　those　for　text　and　data　mining,　AI　training,　and　similar　technologies.