Fast Assessment Of Transient Voltage Stability Of AC/DC System Based On Deep Learning

The continuous expansion of the scale of AC/DC systems and the increasingly complex operating conditions have led to the noteworthy transient voltage stability problem.Research on the rapid and accurate assessment of transient voltage stability for large scale AC/DC systems is urgently needed to ensure their safe and stable operations.The traditional timedomain simulation method and direct method are challenging to meet the requirements of rapid assessment of large power systems due to time-consuming calculations and poor applicability.The machine learning method,which doesn't depend on a physical model,successfully provides a new solution for the rapid assessment of transient voltage stability of large systems through data mining and improving computing performance.Besides,the deep learning model performs better than the shallow machine learning model when dealing with mass data.This paper performs a rapid assessment of transient voltage stability of AC/DC systems based on deep learning.Two essential issues are thrown into insights during the application of the deep learningbased method to the transient voltage stability assessment.The first is to sharply reduce the cost of excessive simulation time when using the large original systems for batch simulations to obtain samples.It plays a vital role in large systems because the time-consuming process hinders the rapidity of the assessment process to a great extent.The second is to make full use of the regional feature of voltage.Under the circumstances,the transient voltage stability in different regions of the systems needs to be evaluated separately.Therefore,it is effective and practical to reduce time consumption when obtaining samples by using dynamic equivalent technology.Meanwhile,the transient stability assessment should be performed based on the reasonable zoning of the large systems.The primary contents of this paper are listed as follows:1)To ensure the rapid assessment of the transient voltage stability based on the deep learning method,we study the dynamic equivalence issues of AC/DC systems.The vital factors that affect the dynamic equivalent of AC/DC systems are analyzed,which are AC system support capacity and high proportion of renewable energy power generation grid connection.According to these factors,the final equivalent scheme is proposed for China Southern Power Grid.The equivalent system avoids excessive time consumption,which meets a challenge of samples in the deep learning method.2)To assess the transient voltage stability of different zones of the whole system,we propose a fast and reasonable two-stage zoning method based on network structure and transient voltage characteristics.In the first stage,a preliminary result can be easily obtained by the system structure and geographic features.In the second stage,the similarity of transient voltage characteristics among buses is then assessed.With a hierarchical clustering algorithm,the best scheme for zones is obtained.In this process,the WARD distance is chosen as the inter-zone distance.The zoning result of China Southern Power Grid makes it possible to assess the transient voltage stability of different zones based on the deep learning method.3)Based on the above research,a construction method of the input feature set suitable for the transient voltage stability assessment based on deep learning is proposed.From the perspective of power source,grid,and load,we analyze the vital factors affecting transient voltage stability.Based on the mechanism of transient voltage stability,these factors can serve as guidelines for the construction of the input feature set.The input feature set is constructed by taking into account both the steady-state features and the multi-dimensional fault information.The steady-state features mentioned above consider both system-level features and component-level features.The impact of HVDC is also taken into account.4)The transient voltage stability prediction and the classification of risk based on convolutional neural network(CNN)are studied.According to the prediction results,the samples are divided into four different categories by a credibility threshold,including the stable set,the unstable set,the missed-unstable set,and the erroneous one.Furthermore,the CNN regression model is introduced to predict the voltage stability margin.A risk function is constructed according to the predicted transient voltage stability margin and credibility.Thus risks of the transient voltage stability are assessed.5)At last,the effectiveness of the construction method of the input feature set,and the CNN-based transient voltage stability prediction and risk classification method are validated in the equivalent system of China Southern Power Grid.The method proposed in this paper effectively improves the accuracy of prediction by enhancing the construction of the input feature set.Besides,the problem of judging whether it is stable or not close to voltage stability boundary accurately and rapidly is solved.At last,risks of transient voltage stability of a large scale AC/ DC system are assessed.This paper introduces a rapid assessment of transient voltage stability of a large scale AC/DC system based on deep learning.The time-consuming problem of obtaining training samples is solved first.Then an input feature set construction method that considers the key factors of transient voltage stability is proposed.Furthermore,a transient voltage stability assessment method of each zone of the AC/DC system based on deep learning is studied.The research results of this paper meet the requirements of accurate and rapid assessment under the current complex and changing operation situations.Besides,it has important theoretical significance for sensing the overall transient voltage safety risks comprehensively and foresight warning of AC/DC large-scale system.