Open-Circuit Fault Diagnosis Method For Sub-modules Of Modular Five-level Converter Based On Deep Learning

The Modular Multi-Level Converter(MMC)is the core unit of the MMC-HVDC system.It contains a large number of sub-modules(SMs)and each sub-module is a potential failure point.Therefore,in the event of a sub-module failure,the fault must be diagnosed in a timely manner.Since MMC has a large number of structurally symmetrical and identical sub-modules,it is very challenging to diagnose which sub-module has failed.In order to protect the SM circuit to reduce the impact of SM faults,some drive protection circuits have been integrated into sub-module controllers,such as over-temperature,over-current,and overload relays.However,these protection circuits are complex in design and have weak diagnostic capabilities,failing to identify certain faults and making the protection vulnerable to failure.Deep learning,as an emerging method in the field of machine learning,has achieved remarkable results in image recognition,speech recognition and other fields with its powerful automatic feature extraction capabilities.Therefore,based on the theory of deep learning,the open-circuit fault diagnosis method of modular five-level inverter sub-modules is studied.The main research work of this article is as follows:First of all,the topology structure and working principle of the modular five-level converter(MFLC)is studied in this paper.Based on this,the failure characteristics of the open circuit fault of the MFLC sub-module are analyzed in depth,and the capacitance voltage of the 24 sub-modules for MFLC is used as the characteristic variable for fault diagnosis.Since the data input into the deep learning model is an image or a two-dimensional matrix,the capacitance voltage of 24 sub-modules are combined into a 24-channel sequence signal.In order to carry out sequence fault diagnosis experiments based on deep learning,MFLC fault simulation model was built in PSCAD/EMTDC.By changing the parameters of the bridge arm reactance and sub-module capacitance values,a total of 864 24-channel sequence were collected.Secondly,for the sub-modules,when open-circuit fault occurs,the capacitance voltage will only change slightly,which makes the fault features not obvious enough,which makes it difficult to extract features through traditional machine learning algorithms.Therefore,this paper proposes a fault diagnosis method for MFLC based on deep convolutional neural network.The method moves the size along the 24-channel sequence to the sliding window to obtain the gray-scale image.Then the gray-scale image is input into the deep convolutional neural network to perform layer-by-layer supervised feature learning,and the original faolure is automatically extracted.The deep features of the barrier dataset,and finally the concise expression of the deep features is connected to the Softmax classifier to output fault diagnosis results.Finally,in view of the fact that the data collected by the MFLC online monitoring equipment is mostly unlabeled data,this paper proposes an SM open-circuit fault diagnosis method for MFLC based on stacked sparse auto-encoder(SSAE),this method converts SM open-circuit fault detection and location problem for the MFLC into a classification problem.Firstly,along the 24-channel sequence,the sliding window is used to obtain the "data zone"sample,followed by the gray-scale map.Convert the vector input to SSAE for layer-by-layer unsupervised feature learning,construct a concise expression of the underlying features of the original fault dataset,and finally connect the deep feature concise expression to the Softmax classifier to output fault diagnosis results.