Photovoltaic Array Fault Diagnosis System Based On Transient Feature Analysis

Developing solar energy is one of the promising methods to cope with fossil shortage and environmental pollution.Photovoltaic(PV)power generation has been the main form of solar energy application.In PV power generation system,PV array is exposed to harsh outdoor environment which may cause several faults,such as shortcircuit,open-circuit,shadow,degradation,etc.These faults may seriously reduce the power generation efficiency,damage PV modules,and even trigger fires.Hence,there is an urgent need of efficient fault diagnosis algorithm and on-line monitoring system for PV arrays.The electrical parameters of PV array show different transient feature under different faults.Accordingly,this paper applies wavelet Multi-Resolution Analysis(MRA)and sparse representation to extract more common fault transient features of PV array,then combines with machine learning algorithms to study photovoltaic array fault diagnosis methods and system.Firstly,a fault diagnosis method for PV array based on MRA and machine learning algorithm is proposed.Four sample signals including current change rate signal,voltage change rate signal,power change rate signal,and conductance rate signal are constructed from the collected PV array current and voltage signals.Then,the four sample signals are slid with the defined sliding windows.The window signals obtained are multi-scale decomposed by wavelet function for obtaining high-frequency signal.Next,Feature vector is exacted by the quantization of the high-frequency signal when fault occurs.Finally,the BP＿Ada Boost strong classifier is applied to train fault diagnosis model to realize the identification of PV working conditions.Secondly,a fault diagnosis method based on sparse representation is studied.The K-SVD algorithm is applied to train the PV array transient current signal sample set for obtaining the overcomplete dictionary.During the detection process,PV array fault types are detected through the correlation of current signal and the signal reconstructed by overcomplete dictionaries.Furthermore,in order to identify multiple faults stably,calculating the root mean square errors of current signal and types of dictionaries reconstruction signal to construct the feature vector.Then,the PV array fault classifier is trained via SVM to diagnose PV array faults.Finally,a PV array fault monitoring system which realizes data acquisition,fault diagnosis and real-time data display is designed.Data acquisition circuit with STM32 and MSP430 as master controller is run to collect PV array voltage,current,temperature and irradiance.The MATLAB host computer receives the data sent from data acquisition terminal over TCP/IP protocol,and runs the proposed fault diagnosis algorithm to detect PV array faults.Finally,PV array working condition,operating data and environmental parameters are displayed on the windows of the monitoring center.