Fast Analysis Of Cascading Outages Based On Model-driven And Data-driven Methods

In the recent worldwide power system blackouts,cascading outages of transmission lines should be responsible,and its study is of great significance for prevention.As most cascading outages are caused by the over-limit of the static security constraints,the study mainly focuses on static security analysis(SSA).Due to the combinatorial explosion of faults,the SSA study of cascading outages suffers great computation burdens.The available methods include model-driven and data-driven methods.The model-driven method possesses a good robustness property,while its accuracy and speed are difficult to coordinate.The data-driven method accelerates the analysis but with generalization capability problem when adapted to large-scale systems.As to the calculation amount reduction,it is promising to integrate the advantages of the two methods.To improve the study efficiency with reliable results,we propose a model-driven and data-driven method for fast analysis of cascading outages,and the main contributions summarized as follows advance the SSA security assessment and control:Firstly,addressed to a large amount of active power flow calculation in security assessment,we propose a fast active power flow calculation method for cascading outages.The proposed method coordinates the fast distribution factor(DF)model and the accurate Newton-Raphson(NR)model,and the appropriate one is selected adaptively by a classifier.The classifier classifies outages on the potential errors of the DF model,and its design is driven by model and data jointly.Specifically,it extracts features from the analytical error of the DF model and applies the support vector machines as the classification learning tool.The calculation efficiency of the proposed method is accelerated by one order of magnitude,compared with that of the NR model.Secondly,as to large amounts of reactive power flow and voltage calculations in security assessment,we introduce the first iteration of the Newton-Raphson(1J)model to the active power flow calculation method,and a fast active and reactive power flow calculation method for cascading outages is proposed by coordinating three models.The proposed method prefers the fast DF and 1J model for all outages,but outages with large potential errors are applied with the accurate NR model.Identifying large error outages depends on the classifier,which is designed by integrating model-driven and data-driven methods.The classifier takes the 1J error features as inputs and applies support vector machines for training.With a satisfying accuracy,the proposed method shortens the calculation time,which only accounts for 1/5?1/3 of the NR model.Thirdly,considering optimal corrective transmission switching(CTS)with high-dimensional control variables in security control,we propose a fast optimization method.An optimization model is constructed and aims to minimize the CTS strategy cost,which is quantified by the change in the transmission loss charges pre-and post-control.For its equality constraint,the model-driven and data-driven power flow calculation method is applied.The optimization model is solved by a heuristic rule based on a cost-performance ratio.The proposed method accelerates the efficiency by 2?5 times compared with that of the alternative current optimal power flow(AC-OPF).Fourthly,to handle the possible high cost and no solution problem for overload elimination by CTS only,we propose a fast optimization method of comprehensive correction for active power security.Based on the CTS optimization method,we add the generation rescheduling(GR)and load shedding(LS)and construct an optimized model for the comprehensive controls.The model aims to minimize the cost for the comprehensive strategy,and the GR and LS power flow calculations are considered in the model-driven and data-driven equality constraint.When solving the comprehensive optimization model,a heuristic rule is established,and a 10x speed-up of the CTS optimization part guarantees the efficiency for the whole model solving.As GR and LS strategies can be adjusted continuously,the proposed method further reduces the strategy cost and avoids the no solution problem caused by the discreteness of CTS.