Knowledge is the precondition for an agent to compute. In dynamic and non-deterministic multi-agent systems, agent should be able to acquire the knowledge timely and effectively so as to solve the problems. The existing knowledge-acquiring models can't meet the knowledge-acquiring requirements in dynamic and non-deterministic multi-agent system and the agent's capability of acquiring knowledge is limited. The systematic knowledge-acquired cooperation models (KACM) is presented for agent to effectively acquire knowledge in multi-agent systems, including passive model, active terminating model and active non-terminating model. Based on the speech act theory and a formal framework of branch temporal logic, the communication acts in KACM are discussed, how agent responses to the communication acts is investigated, the rigorous semantics of the speech acts and the KACM are defined, and lastly the significance of the research is presented.