This　paper　proposes　a　federated　learning-based　intrusion　detection　framework　that　incorporates　the　TCN-GRU　model　for　security　monitoring　in　industrial　control　systems.　The　TCN　module　effectively　extracts　long-term　dependencies　in　network　traffic　through　convolution　operations,　while　the　GRU　module　captures　short-term　dynamic　variations,　enhancing　the　model＇s　ability　to　represent　temporal　trends.　To　ensure　the　convergence　speed　and　stability　of　the　deep　learning　model,　a　combination　of　the　Lookahead　optimizer　and　AdamW　optimizer　is　adopted.　Experimental　results　show　that　the　TCN-GRU　model　exhibits　significant　advantages　over　the　CNN-GRU　model　in　terms　of　accuracy　under　different　communication　rounds　and　numbers　of　clients.　The　Lookahead　optimizer　demonstrates　higher　precision　and　more　stable　performance　throughout　the　entire　training　process.　Overall,　the　proposed　model　framework　performs　excellently　in　addressing　network　security　issues　in　IIoT　scenarios,　providing　a　reliable　solution　for　enhancing　industrial　network　security.　©　2024　Copyright　held　by　the　owner/author(s).　Publication　rights　licensed　to　ACM.