Wind　power　forecasting　is　an　effective　way　to　reasonably　schedule　wind　power　generation　and　ensure　the　stable　operation　of　power　systems.　However,　the　impact　of　physical　attribute　data　related　to　wind　power　on　forecasting　varies,　and　long-term　sequences　of　original　features　often　contain　redundant　information　and　noise,　making　wind　power　forecasting　a　challenging　task.　To　address　this,　a　wind　power　prediction　model　based　on　CNN-BiLSTM　(KG-RCBM)　is　proposed.　This　model　uses　Convolutional　Neural　Network　(CNN)　for　short-term　feature　extraction　to　obtain　local　high-dimensional　features,　which　are　then　analyzed　by　Bidirectional　Long　Short-Term　Memory　Network　(BiLSTM)　to　capture　the　long-term　trends　of　these　local　high-dimensional　features　and　more　comprehensive　sequence　information.　This　approach　effectively　reduces　inaccuracies　caused　by　the　mixing　of　original　data.　The　attention　mechanism　is　utilized　to　dynamically　allocate　weights　to　the　output　data,　addressing　the　issue　of　the　model＇s　inability　to　distinguish　the　importance　of　different　data.　The　RIME　algorithm　is　used　to　optimize　the　hyperparameters　corresponding　to　each　type　of　prediction　model　after　the　Gaussian　Mixture　Mode　(GMM)　classification,　enabling　adaptive　wind　power　forecasting.　Finally,　Wind　power　data　from　Galicia,　Spain,　2019,　as　well　as　hourly　updated　climate　data　from　NASA,　were　utilized　to　validate　our　proposed　wind　power　forecasting　model.　Compared　with　seven　other　models,　the　proposed　model＇s　prediction　accuracy　improves　by　2.32%　to　4.27%,　indicating　a　higher　forecasting　precision.　©　2025　IEEE.