As the complexity, breadth of expertise and number of agents involved in megaprojects grow, collaborative innovation models become invaluable for helping to achieve sustainable project development. On this basis, the purpose of this study is to explore the innovation output mechanisms used for collaborative innovation in megaproject (CIMP) systems by the agent-based modelling (ABM) approach, and to promote the efficiency and effectiveness of organisational collaborative innovation through variable controls. A multi-agent simulation CIMP model was developed using the NetLogo tool. The model encompasses the behavioural factors and interaction rules that affect organisational CIMP. Four simulations were conducted, and the results showed that (1) the innovation environment, including policy environment, cultural climate, and engineering demand, has a positive effect on the output rate of CIMP; (2) a larger scale of innovative network organisation accelerates innovation output; (3) innovative organisations must avoid enforcing high standards for cooperation, communication, and recognition abilities when selecting partner organisations; (4) innovative organisations’ ability to absorb technology, information, and knowledge is positively related to output, while an increase in behavioural costs reduces the scale of innovative network organisations, thereby affecting their output. This study developed its CIMP theory from the perspective of organisational behaviour. The findings are expected to provide methodological and practical guidance for the selection of innovation agents, behavioural patterns, and for guaranteeing efficient innovation in collaborative megaproject organisations.