Rapid　scanning　of　IPv6　addresses　is　crucial　for　understanding　the　network　profile　of　IPv6.　However,　due　to　the　vast　address　space　of　IPv6,　achieving　complete　traversal　scanning　is　difficult.　Therefore,　the　key　to　scanning　the　IPv6　address　space　lies　in　reducing　the　scanning　area　of　invalid　addresses　and　improving　the　address　generation　capability.　By　analyzing　the　dataset　of　surviving　IPv6　addresses,　we　can　determine　the　address　allocation　mode　and　structured　characteristics.　This　allows　us　to　uncover　the　potential　distribution　pattern　of　active　addresses　and　deduce　the　active　address　region　for　scanning　and　detection.　In　this　paper,　we　introduce　a　novel　algorithm　called　6MCBLM,　which　is　based　on　the　attention　mechanism　of　multi-scale　CNN　fused　with　BiLSTM　structure.　6MCBLM　predicts　the　generation　of　potential　active　IPv6　addresses　by　extracting　spatial　features　of　address　sequences　through　word　vectorization　and　multi-scale　Convolutional　Neural　Networks.　It　also　learns　the　temporal　sequences　of　address　sequences　using　the　BiLSTM　structure　features.　Moreover,　the　algorithm　weights　the　spatial　and　temporal　features　using　self-attention　mechanism　to　capture　important　higher-order　features.　It　combines　different　sampling　strategies　to　sample　the　output.　Experimental　results　demonstrate　that　6MCBLM,　compared　to　other　IPv6　target　generation　algorithms,　effectively　integrates　the　multi-dimensional　features　of　address　structure　and　generates　addresses　based　on　learned　feature　information.　The　generated　addresses　exhibit　higher　survival　and　generation　rates.　©　2024　IEEE.