微型电脑应用

主网变电站继电保护故障通常是突发性的，不会持续一段时间，暂态性质不明显，快速定位效果受限，基于此，提出基于时序Q-learning算法的故障快速定位方法。在时序Q-learning中，使用不同多项式函数参数表示不同主网变电站继电保护动作，采用贪婪策略选择主网变电站继电保护动作，根据继电保护状态反馈结果更新权重，使用时序Q-learning算法进行参数训练。构建故障暂态网络的节点导纳矩阵，计算支路电压、电流，确定故障关联域。按照拓扑图论方式时序Q-learning算法搭建快速定位拓扑结构，通过分析支路电流与故障电流之间距离，计算故障相关度，完成故障快速定位。由实验结果可知，该方法故障相序与实际一致，可以分析主网变电站继电保护暂态性质，适用于复杂多变的继电保护装置。

The relay protection faults of the main network substation are usually sudden and does not last for a period of time. The transient property is not obvious, and the rapid location effect is limited. Based on these factors, a rapid fault location method based on the time series Q-learning algorithm is proposed. In time series Q-learning, different polynomial function parameters are used to represent the relay protection actions of different main network substations. Greedy strategy is used to select the relay protection actions of main network substations. The weights are updated according to the relay protection status feedback results, and time series Q-learning algorithm is used to train parameters. The node admittance matrix of fault transient network is constructed, the branch voltage and current are calculated, and the fault correlation domain is determined. According to the graph theory the rapid location topology structure is built according to the time series Q-learning algorithm. By analyzing the distance between the branch current and the fault current, the fault correlation is calculated to complete the rapid location of the fault. From the experimental results, it can be seen that the fault phase sequence of this method is consistent with the actual situation, which can analyze the transient characteristics of relay protection in the main network substation, and is suitable for complex and changeable relay protection devices.