With　the　rapid　development　of　economy　and　the　emission　of　a　lot　of　polluted　gases,　air　pollution　is　increasingly　serious.　Air　quality　prediction　is　an　effective　way　to　provide　early　warning　of　harmful　air　pollutants,　which　can　protect　public　health.　A　hybrid　model　of　air　quality　prediction　which　uses　K-Means　clustering　and　deep　neural　network　is　proposed　in　this　paper.　The　deep　neural　network　with　capacity　of　regressive　computation　consists　of　bidirectional　LSTM　(Long　Short-Term　Memory)　and　fully　connected　neural　network.　First　of　all,　the　historical　meteorological　monitoring　data　of　Qingdao　City　is　taken　as　the　research　target,　and　the　meteorological　data　is　divided　into　four　categories　according　to　the　quarter　by　k-Means　clustering　algorithm.　Then　the　classified　meteorological　data　and　the　data　of　historical　concentration　of　air　pollutants　are　used　to　train　neural　network.　A　better　hyperparameter　combination　is　selected　by　lots　of　trial.　Next,　the　hybrid　model　is　applied　on　the　test　set,　and　the　mean　square　error　between　predicted　value　and　true　value　is　used　as　the　evaluation　criterion　of　predictive　property.　Last,　through　comparing　with　other　algorithm　models,　it　is　proved　that　the　proposed　hybrid　model　can　achieve　higher　precision　for　air　quality　prediction.　©　2019　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.