Abstract:

Based on the principles of collaborative optimization, this paper presented an intelligent Multi-Agent model, which employed ANN to build response surface of subspace agent, and used subtask-programming agent to perform the traditional optimization processing. By the approximate satisfactory degree calculated by result analysis agent, task assignment agent was able to select different executive routes autonomously. The response surface agent acquired gradually the accurate response surface of subspace objects and constrainted in accordance with updating criterion. When subspace agent received the target vector from system level, it can use the response surface to fastly produce the optimal solution and sent it to global blackboard structure, by which system level agent coordinated the inconsistent information of different subspace agents. Finally, an engineering example was applied to prove the proposed method can improve the efficiency and accuracy of collaborative optimization for complex mechanical products.

Key words: Multi-Agent, ANN, response surface, collaborative optimization

摘要：

根据协同优化算法的思想，提出了一种Multi-Agent模型，利用神经网络建立子系统优化Agent的近似响应面。子任务规划Agent进行传统的优化进程；任务调度Agent根据结果分析Agent计算的样本集近似满意度自主选择不同的执行路径；近似响应面Agent通过使用更新准则，逐步获取满足子系统级约束和目标的精确的响应面。当子系统获取指标变量后，通过子系统优化Agent内部的响应面快速获取优化解向量，并将该向量返回到全局黑板数据结构中，系统级优化Agent可以利用该结构，协调各个子Agent不一致的信息，从而有效地提高了复杂机械产品协同优化中的效率和精度。

关键词: Multi-Agent, 人工神经网络, 响应面, 协同优化