基于长短期记忆网络的电网动态轨迹趋势预测方法

Trend Prediction Method of Power Network Dynamic Trajectory Based on Long Short Term Memory Neural Networks

为满足大电网主动防御对算法速度和精度的要求,提出基于长短期记忆(long short term memory,LSTM)网络的电网动态轨迹趋势预测方法。首先,针对电压时序相轨迹的几何特征,提取节点状态的时序演进规律,快速辨识系统发电机运动的同趋性;其次,基于LSTM快速预测等值机系统的受扰轨迹;最后依据扩展等面积准则计算切机量,实现暂态功角稳定的紧急控制。IEEE 39系统算例验证了方法的有效性,该方法无需复杂计算、耗时短,具有较好的工程应用价值。

In order to meet the requirement of speed and precision for active defense of large power grid,a prediction method based on long short term memory for dynamic trajectory of power network was proposed in this paper.Firstly,aiming at the geometric characteristics of the track of the voltage sequential phasor,the time sequence evolution rule of the node state was extracted,and the convergence of the generator motion was identified quickly.Secondly,based on the long short term memory,the disturbed trajectory of the equivalent two-machine system was predicted rapidly.Finally,the emergency control of transient power angle stability was realized by calculating the cutting capacity according to the extended equal area criterion(EEAC).An example of IEEE 39 system shows the effectiveness of the proposed method.This method does not require complex computation and is time-consuming and has a good application value in engineering.