Research On Intelligent Location Algorithm Of IGBT Open-circuit Fault Of MMC Sub-module

As a kind of voltage source converter,Modular Multilevel Converter(MMC)adopts modular design,which has many advantages such as more output levels,low harmonic content,small loss of switch tube,etc.,so it can be widely used in the field of flexible DC transmission.The converter plays an important role in the transmission system.If there is a fault inside the converter,it may lead to the increase of voltage harmonics,waveform distortion,power quality decline and even the failure of the converter.Due to the large number of SM in MMC,the problem of locating fault SM becomes an urgent problem to be solved.Most of the faults in MMC are located based on the variation of the capacitance voltage or current of the sub-module.Some algorithms can only locate the fault bridge arm,but cannot locate the sub-module.However,the research on the precise location of IGBT is rarely reported.In this research,the MMC topology of the half-bridge is taken as the research object,and in-depth analysis and research are conducted focusing on the open-circuit fault and fault location of its internal SMs.The whole research mainly includes the following contents:1)Discuss and simulate the working principle of MMC and two common control strategies,namely,the CPS-SPWM modulation strategy based on the average-balance control and the NLM modulation strategy based on the capacitor voltage sequencing of SMs,analyze their working status during normal operation,and lay a foundation for subsequent fault analysis.2)Analyze the IGBT open circuit and short circuit faults that may occur in the MMC sub-module,emphatically analyze the IGBT open circuit faults,summarize the variation rules of different parameters when faults occur,and select three-phase output voltage,capacitor voltage of sub-module and terminal voltage of sub-module group as fault characteristic parameters according to the rules.3).According to the data selected above,two common data processing methods,namely FFT and WPD,were compared and analyzed.The results showed that FFT analysis could not better indicate the change of fault signal in the time domain,while WPD with time-frequency characteristics was more suitable as a signal processing method.After normalized treatment,fault and non-fault data were differentiated significantly.And it is found that the noise resistance of the positioning algorithm can be improved by using WPD.4)PCA is adopted for data dimension reduction processing,and the processed data dimension can be reduced by 95.8% on the original basis,which greatly reducing the data dimension and effectively avoiding the disaster of dimension.5)Intelligent algorithm is adopted to study the fault location problem of sub-modules,and the fault location problem is transformed into the problem of classification and identification,without the need for accurate system modeling.At first,this reasearch presents a localization algorithm based on GA-BP.The input data of this algorithm are three-phase output voltage and capacitance voltage of sub-module respectively.By comparing this algorithm with BP,GA-PNN,SSAE-SOFTMAX and other algorithms,it can be found that the proposed algorithm has many advantages,such as simple network structure,fewer input neurons,fewer iterations,short training time,etc.,and its positioning accuracy rate can reach 100%.6)To solve the problem of high level MMC fault location difficulty and MMC field fault data collection difficulty,this research further proposes a three-group fault location model based on FOA-LSSVM model.The input data of this model are three-phase output voltage,capacitance voltage of sub-module and voltage of the sub-module group respectively.Based on SVM,FOA-LSSVM,PSO-LSSVM and other algorithms,the simulation comparison of single group of fault location model and three groups of fault location model is conducted.It can be known that the proposed three-group fault location model has many advantages,such as short training time,high accuracy of test set,and small amount of SVM required.Moreover,the three-group fault location model is superior to the single-group fault location model with only submodule capacitance voltage as input data.