Research On Neural Network Method For Parameter Identification Of Power Grid Branche

Transmission line parameters play an important role in the safe and stable operation of the power grid.With the increasing demand for energy,the transmission network is becoming more and more complex,and the requirements for the stability of the distribution system are becoming higher and higher.The transmission line parameters play an important role in the stable operation of the distribution system.However,the line parameters will inevitably change due to temperature changes and line aging,so the line parameters must be updated regularly.Therefore,it is of practical significance to study real-time parameter identification methods with high accuracy and robustness.In recent years,some documents have proposed real-time parameter calculation methods based on PMU and SCADA system data,but these methods have some limitations.First of all,the traditional real-time parameter calculation method based on PMU and SCADA system mainly constructs the calculation equation around the multi-dimensional electrical variable data at a single time,which means that the measuring elements used must operate stably during the measurement period,but in reality,the measuring elements may appear unstable operation stage.Therefore,it is necessary to study a real-time parameter identification method that does not rely on a single or a few moments of measured data but utilizes a large number of historical and reliable data stored in the system.Therefore,this paper proposes a parameter identification method based on multi-task spatiotemporal pooled graph neural network,which improves the accuracy and robustness of model identification by fusing branch topology constraint information with a large amount of historical data.Considering the possible measurement noise in the actual scene,this paper proposes a pooling module to effectively mitigate the impact of transient data mixing.Secondly,the previous real-time parameter identification methods based on PMU and SCADA rely heavily on the transmission of integrated data when completing the multi-branch simultaneous identification task.The transmission of OPGW(Optical Fiber Composite Overhead)optical fiber and cable technology used in communication may have data transmission risks due to extreme high and low temperatures,and the use of risky data may cause identification parameter distortion.For this reason,this paper uses the federated learning idea to avoid the high-risk process of transmitting huge data and uses distributed identification instead.Based on the strong fitting and robustness of neural network,the accuracy of parameter identification is improved.Finally,this paper designs a tail weight judgment module,which uses the maximum mean difference distribution algorithm [3] to map the weight and the weight of the previous period to the regenerative Hilbert space at the same time,and returns a warning sign.This not only completes the task of parameter identification,but also avoids high-risk transmission,and can also assist in the feedback of branch parameter anomalies.Finally,the simulation and comparison experiments are carried out through the measured data of the power grid,and the simulated noise is added according to the actual environment.Both methods in this paper have achieved satisfactory results.At the same time,the two methods can be switched according to the needs of the component scene,which has good practicability..