Research On Methods For Transient Stability Assessment And Preventive Control Based On Deep Learning And Reinforcement Learning

Transient stability analysis is a critical issue in the operation of power system.With the ongoing transformation of the power market,the use of newly developed energy technologies,and the furthering of inter-regional connectivity,the incredibly challenging power grid structure and operation conditions introduce new risks.To ensure the security of the power system,research on transient stability assessment and preventive control is extremely important.For the past few years,the rapid development of AI technology and the rapid popularization of phasor measurement unit provide a basis for the research.This article mainly conducts in-depth research on transient stability assessment and preventive control using deep learning and reinforcement learning technology.The main research contents are as follows:For the transient stability assessment under the concept of security region,the traditional evaluation method focuses on the stability of the power grid at the current time.When the power grid status is slowly deteriorating,it may miss the best pre-control time.Therefore,a method to predict the transient stability margin based on the long short-term memory(LSTM)network and attention mechanism(AM)is proposed.The LSTM layer reduces the dimension of the historical steady-state power flow data,and the temporal characteristics are extracted from the data.Subsequently,the AM is introduced to differentiate the characteristics and historical transient stability margin data for the models to identify the information associated with stability.Finally,the LSTM and fully connected layers are used to predict the transient stability margin,ensuring rapid and accurate situational awareness of operators in terms of transient stability.For the transient stability assessment under the concept of stability region,the transient data after failure show obvious spatial-temporal characteristics.However,the existing methods do not fully exploit the characteristics,which limits the evaluation performance.Therefore,a transient stability assessment method based on temporal convolution network(TCN)and graph attention network(GAT)is proposed.This method only takes the measured bus voltage amplitude and phase-angle data as the input.With the advantages that GAT can process the diagram data and the unique causal hole convolution operation characteristics of TCN,the spatial and temporal characteristics are automatically extracted.So the method can realize the accurate evaluation of transient stability.In addition,the improved focus loss function is used as the model training target,which can dynamically adapt to the discriminant boundary of difficult and easy samples in the training process and adaptively deal with the problem of sample imbalance.Therefore,the mode can reduce the phenomenon of misclassification of unstable samples and improve the global accuracy.To address the complex and time-consuming problems of power system transient stability preventive control calculations and to address the lack of generalization ability of traditional methods in the case of topology change,a new preventive control method based on pooling graph attention(PGAT)embedded deep deterministic policy gradient(DDPG)is proposed.First,a transient stability evaluator is proposed to establish the mapping relationship between system operational data and stability margins.It utilizes graph convolution for spatial feature extraction and graph pooling to improve robustness.Then,the evaluator is embedded in the transient stability constrained optimal power flow(TSCOPF)to replace the complex dynamic part of the model,thus creating a fast interactive environment.Finally,the feature extraction capability of the evaluator is also embedded into the DDPG agent as a priori knowledge.It can enhance the attention to topology information.In turn,the agent can better capture spatial information and make fast and accurate decisions.