Research On IGBT Fault Diagnosis Technology Of PV Inverter System Based On End-to-end Data Driven

With the development of science and technology,the strong support of national policies and the transformation of residents’ lifestyles and life concepts,our country’s energy consumption pattern is also quietly changing.Solar energy is widely used in distributed photovoltaic poverty alleviation power stations,smart photovoltaic home systems and large-scale centralized photovoltaic power stations due to its energy-saving,environmentally-friendly and renewable characteristics.As an important device for realizing DC to AC power conversion,inverter failure rate is greatly increased due to its harsh outdoor working environment,disturbance of AC and DC power supply,equipment assembly process and internal electrical and thermal stress.Inverter failure may cause the collapse of the photovoltaic power generation system,huge economic losses and even personnel safety risks.At the same time,with the arrival of the 5G era and the development of big data and artificial intelligence,the high visibility and interactivity of "smart operation and maintenance" provide users and enterprises with more convenience than traditional manual fault diagnosis.Therefore,it is necessary to study the inverter fault diagnosis from the perspective of artificial intelligence and data driven and find a more scientific and intelligent fault diagnosis method for inverter.This paper takes the centralized single-stage three-phase two-level full-bridge photovoltaic(PV)inverter as the research object.The main research contents are as follows:(1)The topology,working principle and main fault types of the three-phase fullbridge inverter are introduced and the fault characteristics of IGBTs open-circuit faults and short-circuit faults are comprehensively introduced in this paper.At the same time,the types of IGBT faults are coded to facilitate the algorithm to identify different faults.Simulink simulation model and semi-physical simulation model based on RT-box are built to verify the fault characteristic analysis and construct the database.(2)According to the fault database and the fault code table,an inverse fault diagnosis method based on traditional machine learning algorithms is proposed.In order to solve the problem of feature quantity susceptible to noise generated by the descriptive statistics method,the problem of feature quantity dimension being too high and the problem of feature quantity redundancy,this paper first introduces empirical mode decomposition to achieve adaptive noise reductionand secondly introduces the supervised learning-based Generalized discriminant analysis for feature reduction and finally uses BP neural network for fault location.(3)For the problem that traditional diagnosis methods based on machine learning rely excessively on manual feature extraction,this paper introduces long and short-term memory neural networks to achieve end-to-end fault diagnosis.At the same time,in order to solve the problem of long and short-term memory neural network hidden layer parameter selection,this paper adopts the neural network architecture search technology based on genetic algorithm to optimize the neural network structure parameters to realize the complete end-to-end operation of the entire fault diagnosis system.Finally,the accuracy and practicability of the two methods are verified by the existing data in the database.