| There are three characteristics about complexity of warfare system including co-evolution, uncertainty and avalanche. This is why it's difficult to create and optimize warfare scenario. Data farming can be a good idea to solve the problem. This paper applies the idea of Data Farming as basis, and adopts the Multi-Agent simulation technology to implement scenario creating and optimizing, therefore, explains why multi-Agent simulation is introduced to Data Farming and how to apply Data Farming with Multi-Agent under an implementation framework. It also analyses the relationship between exploration and learning in Data Farming.This paper explains the importance in building a various resolution model and how to build an Agent model reflecting the characteristics of attributes and actions in the process of creating scenario. It also describes the process of exploring attributes and actions. Game theory is introduced to multi-Agent cooperation and the relationship between cooperation and learning is analyzed in the part of scenario optimizing. After analyzing Q learning algorithm, an algorithm based on long-time payoff matrix is proposed, in which the matrix will gradually converge at a stable value by constant interaction with environment and payments from it. It also can drive the Agent to take the optimal action.In the end, under the military application background of the counterwork between detection and interference in the boundary air-duty task, an application case is presented. With the multi-Agent carrying out Data Farming, method in solving the creation and optimization of scenario are effectively validated. |
