It is very important to accurately predict the network flow queuing delay to improve the network performance of city public services. City public services are the offspring of the paradigm “smart + connected communities” and aim to overcome the problems of isolated and fragile data collection because of administrative divisions. Quality of service is one of the important evaluation indexes in service-level agreements, in which low delay is a basic requirement for measurement. To improve city public services and predict the network flow queuing delay of city public services in advance, we propose a framework based on long short-term memory (LSTM) that will allow the government to enhance service efficiencies and offer better service experiences for every citizen. The results obtained by using the investigated framework show that the proposed algorithm has superior aggregated prediction accuracy and real-time performance to other methods.