A distributed problem solving with contract net and case-based reasoning

Contractor net protocol (CNP) and case based reasoning (CBR) are wide-application distributed problem solving methods. The paper presents an interacted learning strategy based CNP and CBR to solve non-structural, nonlinear, and complex urban traffic control decision-making under multiagent system architecture. The manager agent (MA) decomposes a regional control task into several separable subtasks which single traffic agent (TA) could deal with independently and their specifications by knowledge with CBR. MA sends task announcement messages to selected agents and requires them finish bid document in expiration-time. MA analyzes bid messages, eliminates those conflict control schemes, and optimizes their specification and resend to TAs. Through finite cycles of interacted learning, MAS could get a group of best coordinating TAs and optimal control schema. Extended KQML is a tool to realize the distributed decision-making.