Data-driven Based Transient Stability Assessment Of Power Systems

Given the background of massive renewable generation integration,power systems digitization and intelligentization,and long-distance large-capacity ultra-high-voltage transmission interconnection,power systems are facing a set of challenges such as strong coupling of all grid components,fast fluctuation of generators and loads,and high dynamic complexity.These factors have a great influence on power systems transient stability.This paper focuses on power system transient stability assessment,using data-driven techniques,and aims to improve the model accuracy and computational efficiency of stability assessment.In general,this paper is highlighted with the following four parts:Firstly,a fast transient stability batch assessment algorithm is proposed in this paper.To achieve this goal,cascaded convolutional neural networks are designed to “learn” from existing time-domain simulation(TDS)results in the batch and “infer” stability conclusions using current available TDS output.In other words,this data-driven model is employed to capture data from different TDS time intervals,extract features,predict stability probability,and determine stability conclusions before the simulator reaches the end of simulating time windows.Therefore,TDS can be terminated early so as to reduce the average simulation time for batch assessment without losses of accuracy.While accumulating more knowledge in batch processing,early termination criterion is refreshed continuously via feedback learning to terminate TDS increasingly earlier,with the increase of existing TDS results in the batch.Besides,an entropy based prioritization strategy is designed to distinguish informative samples,dynamically schedule TSBA task queue and timely update model,for further simulation time reduction.Overall,this work reduces the computational burden of transient stability batch assessment without losses of accuracy,making it possible to generate enough transient data samples not only for online assessment model training but also for “N-k” security check of grid planning and day-ahead dispatch in a limited time.Secondly,all kinds of new grid components(e.g.,distributed generators)are widely integrated into the distributed network,which has a significant impact on the steady-state and transient characteristics of both transmission and distributed networks.Therefore,traditional load model is no longer suitable for the analysis of the transient stability of transmission network,and transmission and distributed system co-analysis is required.To address this challenge,we propose a novel data-driven framework to generate synthetic unbalanced distribution networks that include node connectivity,time-series nodal consumption data,and locations and capacity of electrical components.Specifically,our unbalanced graph learning-based generative adversarial network(UGL-GAN)generates a synthetic network that can mimic a single real-world three-phase unbalanced distribution system.The synthetic distributed networks generated by the proposed framework are able to mimic the characteristics of the original real data,while does not preserve any of the information inherent to real-world systems,thus protecting utility privacy.Thirdly,in order to provide real-time contingency information in response to the requirement of online transient stability assessment,this paper proposes a novel framework summarized as“train in the cloud and infer on the edge”,deploying the single-ended fault-location task on the edge of data source for distributed inference instead of sending measured data to a centralized cloud.To archive this goal,a modified long short-term memory(LSTM)network is trained in the cloud,covering various fault scenarios.Meanwhile,the trained model is imported on a small embedded artificial intelligence(e-AI)module equipped in fault recorder of the substation,in order to infer fault information using local single-ended measurements in real time.Finally,we propose a data-driven transient stability boundary generation algorithm for online security monitoring.By introducing a transient stability index with its adjoint sensitivity,this paper develops a search strategy to obtain more data samples near the stability boundary,while traverse the rest part with fewer samples.Although it significantly reduces the computational burden,it still faces the challenge of “curse of dimensionality” and “combination explosion”.To relieve such challenges,critical scenarios selection mechanism is proposed,in order to find out the most representative scenarios and periodically update TSB for transient stability online monitoring and assessment.