基于深度残差网络的电力系统暂态稳定预测

Power System Transient Stability Prediction Based on Deep Residual Network

电力系统稳定性预测存在映射维度不匹配的问题,导致预测精准性下降,提出基于深度残差网络发热暂态稳定预测方法。将残差单元融入深度残差网络,利用跨层短连接的恒等映射与残差函数,完成线性映射到非线性映射的转换,根据输入特征维度,架构用于暂态稳定预测的深度残差网络模型,通过划分神经网络处理后未知样本,采用交叉熵损失函数,优化深度残差网络模型,基于样本特征与标签数据集,结合样本标注规则,构建出由数据生成、离线训练以及模型应用模块构成的暂态稳定预测方法。仿真数据选取新英格兰10机39节点系统与三组底层量测数据,预测结果具有较为理想的稳定性、精准性以及鲁棒性,且时间复杂度较低。

The mismatch of mapping dimensions causes the decline of power system stability prediction accuracy.Therefore,this paper proposes a method based on deep residual network to predict the transient stability of heating.Firstly,the residual unit was introduced into the deep residual network,and the transformation from linear mapping to nonlinear mapping was achieved by using the identity mapping and residual function of cross layer short connection.Secondly,a deep residual network model for transient stability prediction was constructed by inputting feature dimensions.Then,the cross-entropy loss function was used to optimize the depth residual network model by dividing the unknown samples processed by the neural network.Finally,based on the sample characteristics,labeled data sets and sample labeling rules,a transient stability prediction method composed of data generation,offline training and model application modules was constructed.The simulation data selected New England 10 machine 39 bus system and three groups of bottom measurement data.The results show that the method has ideal stability,accuracy and robustness,and low time complexity.