Research On Data-driven Grid Transient Voltage Stability Assessment Method

With the goal of "carbon neutral",China has accelerated the pace of r enewable energy development,and the installed capacity of renewable energy is growing rapidly.Due to the great volatility and randomness of wind power,phot ovoltaic and other forms of renewable energy,the system operation state gradually tends to diversify,and voltage instability problems in the grid occur frequently,which seriously affects the safety and stability of the system.In order to ensure the safe and stable operation of the system,it is necessary to establish an evaluation method that can judge the transient voltage stability state of the system in real time.Traditional transient voltage stability assessment methods require solving complex differential-algebraic equations,which can not balance accuracy and speed in practical applications.To address the above problems,this paper proposes a data-driven transient voltage stability assessment method based on the following aspects.First,deep learning’ basic theory and model are briefly explained by taking CNN and Res Net as examples;taking IEEE 39-bus system as an example,the simulation of different load levels,motor proportions,new energy permeability,fault clearing time,and fault location is carried out.The influence of the above five factors on transient voltage stability is analyzed,which lays a foundation for the subsequent data-driven voltage stability evaluation network sample composition.Secondly,an evaluation model based on an improved deep residual network is proposed.Using the data-driven method,historical data or simulation data of the system under large disturbances under different operating conditions are used as model input,and the evaluation model learns the impli cit relationship between the system state and transient voltage dynamic process through a large amount o f dynamic information in the input data.In order to capture the key information in the fault process,a convolutional attention module is latched into the residual net work to explore the potentially spatiotemporal relationships in the dynamic trajectory of the power system through attention to the temporal and spatial channels.To address the problem that the model tends to learn form the majority class of samples in the training process,a loss function based on a gradient harmonizing mechanism is introduced to reduce the impact of unbalanced samples on the evaluation results.In order to enhance the model’s ability to extract data features,this paper replaces the traditional convolution kernel with an asymmetric convolution module to improve the stability of the model during the training process.The excellent performance of this paper’s method in transient stability evaluation is further validated by simulations on a modified IEEE 39-bus system.Finally,to improve the evaluation capability of the model when the number of samples is insufficient and to enhance the generalizability performance of the model,this paper introduces a domain transfer learning approach,embedding a coordinate attentive mechanism in the residual network during the feature extraction stage to capture the long-range correlation between features,and using adversarial approach to narrow the gap between the source and target domains to train the network in the target domain with the source domain data to enhance the evaluation performance of the network under insufficient data,which is trained on the target domain with the source domain data.