Studies On Security Risk Situation Early Warning For Large Scale Hybrid AC/DC Grids

With the application of long-distance and high-capacity HVDC transmission technology,the power grids in China have become large scale hybrid AC/DC grids.The construction of large scale hybrid AC/DC grids improves the optimizing configuration capability of large scale resources,but there also exists the risk that local AC faults can influence the secure operation of the whole system.Security risk situation early warning identifies the possible insecure operating states in the future,which can provide guide information and set aside decision time for subsequent security risk control decision-making.It is an effective technique of assisting power systems in resisting the future security risk.With the large scale integration of renewable generation and the increase of flexible loads,the uncertainties of both source and load sides are increasingly evident due to the random fluctuation of renewable generation and the random response of flexible loads.Because of the dramatic increase of the operating condition(OC)number caused by the uncertaintiesand the huge time cost of the time-domain simulation of hybrid AC/DC grids,the computing burden of security risk situation early warning is increased significantly.Therefore,comprehensive studies on the fast and effective security risk situation early warning method for large scale hybrid AC/DC grids have important theoretical significance and practical value to the secure and stable operation of modern power systems.The critical interface is regarded as the important object of security risk early warning.Total transfer capability(TTC)considering transient angle stability,transient voltage security and transient frequency security constraints is regarded as the key system security indicator.Insecure levels are distinguished according to the type of control actions that is needed to ensure the system security.The multi-time-scale rolling early warning is performed by considering the time sequence difference of typical control actions and utilizing the latest forecasts of source and load uncertainties.On the basis of current research,this dissertation utilizes artificial intelligence techniques such as deep learning,semi-supervised learning,multi-attribute decision tree(DT)and incremental learning to deeply study the security risk situation early warning of large scale hybrid AC/DC grids.The main contributions and innovations of this dissertation are summarized as follows:(1)A fast transfer capability estimation method of critical interface is proposed,which considers the influence of the uncertainties of both source and load sides,and multi-attribute dynamic security constraints.A semi-supervised deep learning model is constructed based on stacked denoising autoencoder(SDAE)and semi-supervised extreme learning machine(ELM),which is deployed to estimate the TTC value considering multi-attribute dynamic security constraints according to operating features.After the probability forecast information of the uncertainties of both source and load sides is acquired,the cumulative distribution function(CDF)of TTC is approximated through the TTC results of a small number of OCs based on point estimation method and Gram-Charlier series expansion.The limit violation probability is defined as the probability that the actual power transfer of the interface is more than TTC,and the calculation formula of available transfer capability under certain limit violation probability is derived.Simulation results of Jiangsu-Shanghai interconnected grid demonstrate that the proposed semi-supervised deep learning model has higher TTC estimation accuracy than shallow learning models,and can effectively utilize unlabeled samples to improve the regression performance.The CDF approximation of TTC based on point estimation method and Gram-Charlier series expansion has high accuracy,and can effectively reduce the number of OCs whose TTC values need to be calculated.(2)A fast graded dynamic security risk early warning method based on control cost is proposed.With N-1 contingencies considered,OCs are ranked into different insecure levels according to the type of preventive control actions that is needed to ensure the system security.If the type of preventive control actions that is needed has higher control cost,the corresponding OC will be ranked into a higher insecure level.With the influence of the source and load uncertainties considered,multi-attribute DT is deployed to identify classification boundaries among different insecure levels in the power injection space of the source and load uncertainties.The OC sets corresponding to different insecure levels are constructed based on these classification boundaries.In order to decrease the missing alarm rate of multi-attribute DT,class weights are introduced in the training process of multi-attribute DT.Simulation results of Jiangsu-Shanghai interconnected grid demonstrate that the graded early warning results can provide scientific basis for the dispatching operation and control decision-making of power systems.Multi-attribute DT can accurately identify the classification boundaries among different insecure levels.The mathematical formulation of insecure OC sets is concise and clear,which is beneficial to the subsequent security risk control optimization decision-making.(3)A multi-time-scale security risk situation rolling early warning method is proposed,which considers the fact that the forecast accuracy of the source and load uncertainties gradually improves as the forecast time approaches,and combines the time sequence characteristics of typical control actions.Multiple contingencies that can be caused by extreme weather process are considered,and typical emergency control actions are added into the early warning ranking strategy.The 6-hour,1-hour and 15-minute rolling early warning framework is constructed to coordinate different types of control actions to resist possible dynamic security risk.The TTC estimation model is updated by incremental learning based on the new samples generated in the process of rolling early warning.Bootstrap method is utilized to estimate the probability distribution information of the actual TTC value,and the probability that the estimation error is smaller than a given error tolerance is calculated,which is taken as the confidence level of early warning results.Simulation results of Xibei-Huadong hybrid AC/DC grid demonstrate that the multi-time-scale rolling early warning can make full use of the latest forecast information of the source and load uncertainties,and effectively coordinate the control actions with different time sequence to resist future security risk together.Incremental learning can significantly speed up the model update,and continuously improve the accuracy of TTC estimation and the confidence level of early warning results.