根据元件故障与保护动作和断路器跳闸之间的内在逻辑关系，建立了面向元件的电网故障诊断模型，并采用误差反向传播的梯度下降法修正网络参数。该模型是一种由Noisy-Or、Noisy-And节点组成的特殊的贝叶斯网络，能够处理电网故障诊断中的不确定性，具有语义精确、推理快速、学习效率高等特点，适用于大规模电力系统的多重复杂故障诊断。实际电网故障案例验证了该方法的正确性和有效性。

This paper proposes a cooperative protection system with a multi-agent system, which could adapt protection functions to the complicated changes in the power system automatically and intelligently. In the cooperative protection system, multi-agent system allows the cooperation between plural numbers of equipments. Some agents stay in one place and collect local power system information such as current and voltage data. Other agents move about to utilize data or functions installed in other equipment. The cooperation of multi-agent enables the flexible utilization of all hardware resources by any equipment, so that various adaptive protection functions are realized. Simulation of the proposed method has been applied to some typical protection function, i.e. overcurrent protection function. The results showed that the cooperative protection system could achieve the optimum protection capability. Although many practical aspects need to be thoroughly studied, it is quite promising to realize the adaptive protection using the multi-agent technology.