Because　of　faster　convergence　speed　and　superior　fault　tolerance,　small-world　neural　network　has　attracted　wide　attention　in　recent　years.　However,　in　the　construction　process,　it　may　cause　the　loss　of　important　information　due　to　random　reconnection,　which　may　lead　to　the　decline　of　network　accuracy.　To　solve　this　problem,　derived　from　the　Watts-Strogatz　(WS)　small-world　neural　network,　a　feedforward　small-world　neural　network　based　on　synaptic　consolidation　(FSWNN-SC)　mechanism　is　proposed　in　this　study.　Firstly,　the　regular　feedforward　neural　network　is　pre-trained　by　using　the　network　regularization　method.　Based　on　the　synaptic　consolidation　mechanism,　the　unimportant　connection　weights　of　the　network　are　disconnected　and　the　important　connection　weights　are　retained.　Secondly,　the　rewiring　rules　are　designed　to　construct　a　small-world　neural　network,　which　can　realize　the　sparseness　of　the　network　while　ensuring　the　small-world　properties　of　the　network.　The　gradient　descent　algorithm　is　used　to　train　the　network.　Finally,　four　UCI　benchmark　experiments　and　two　practical　experiments　are　carried　out　to　evaluate　the　model　performance,　and　the　Wilcoxon　signed-ranks　test　is　performed　to　test　the　significant　differences　between　comparative　models.　Experimental　results　show　that　the　FSWNN-SC　model　proposed　in　this　study　not　only　obtains　a　compact　network　structure,　but　also　has　significantly　better　accuracy　than　regular　feedforward　neural　networks　and　other　WS　small-world　neural　networks.　©　2023　Science　Press.　All　rights　reserved.