Research On Branch Parameter Identification Considering Spatial Structure Constraints Of Power Gri

Transmission line parameter identification has always been the basis of real-time analysis and dynamic early warning of the power grid system.The realization of accurate and real-time power grid branch parameter identification provides great help to the dispatching analysis of the power grid system,and is of great significance to the stable operation of the power transmission system.Most of the existing parameter identification algorithms use machine learning methods,but these methods have some shortcomings.Firstly,the machine learning algorithm does not consider the constraints of topology and considers each node as an isolated node separately.It does not pay enough attention to the features of global nodes,which leads to a sharp decline in prediction accuracy under the influence of noise.In order to solve this problem,this paper proposes a power grid branch parameter identification algorithm based on multi task graph neural network,which integrates the topology information and global feature information of the power grid,and effectively enhances the robustness of model prediction.Secondly,if the graph neural network is too deep,there will be a problem that the node features tend to be similar.In order to solve this problem,this paper focuses on the information of important branches and important features through multi head attention mechanism and fully linear connection,and the prediction results are fused and output in multiple subspaces.Experiments show the effectiveness of the above structure.Finally,aiming at the problem that multi task parameter identification is easy to ignore the identification targets with small orders of magnitude,combined with the attention mechanism,a graph neural network composed of U-shaped module and nested loss module is proposed,which makes the model automatically balance the losses of different orders of magnitude through the weight adaptation of loss function.Based of ensuring the accuracy of the model,the problem of identification of the power grid branch parameters is solved.Finally,through the comparative experiment on the data set of the power grid,and adding different types of noise to simulate the actual situation,the experimental results have achieved satisfactory results,which proves the effectiveness of the graph neural network model that proposed in this paper.