Public transportation in large cities is a typical complex giant system. The use of complex network methods to analyze large city public transport network systems is of great significance for urban transport development. Most of current studies take public transport stations as nodes and routes as connecting edges to construct an abstract network adjacency matrix, so to analyze the complexity characteristics of public transport networks according to indicators such as average (shortest) path length, clustering coefficient, degree distribution, node proximity or median centrality. However, the complexity of urban public transport network is not a static topology network composed of bus stops, routes, and their interconnections, but also the dynamic traffic information, which is seldom considered in existing studies. Distinguishing the characteristics of the passenger flow network in different time periods is of great value for formulating time-sharing public transport management policies. Meanwhile, the widely used big data recently provide high-precision traffic information for the study of dynamic traffic network structures. In this paper, we constructed the bidirectional adjacency matrix of passenger flows in different time periods based on swipe card data of Beijing buses, then compared and analyzed the within-day variation of the bus passenger flow network through complex network indexes. In terms of the structural characteristics, the bus passenger flow network in each time period had a small average shortest path and a large clustering coefficient, meaning a small-world network; the distribution of accumulation degree was fitted as an exponential distribution, which indicates that the bus passenger flow network did not have scale-free characteristics. The bus passenger flow distribution in each time period had an obvious distance attenuation rule, and it was mainly for short-distance travels below 10km, which suggests the bus line and operation management should focus on the distance range below 10 km. Degrees centricity and weighted degree centrality of the spatial pattern in different times presented an obvious core-edge character but changing over time; the weighted degree centrality in the top 10 nodes changed a lot, according to which dynamic public transport hubs should be considered in a precious and accurate traffic planning and management. Our findings provide a reference for public transport planning and management policies. In the future, more comprehensive passenger flow data should be used to explore the structural characteristics from a multi-scale perspective.