Fault Detection And Diagnosis Of PV System Based On Artificial Intelligent Methods

Recently,the usage of solar energy in electrical power systems,as one of the renewable energy sources,has been gradually raised compared with the other sources.This usage not only provides significant economic and environmental benefits to the world but also offers less operating complexity.However,under unpredictable environmental conditions,different faults in photovoltaic(PV)systems have been strongly affecting their modules’ performance,and hence increase the overall energy loss.In addition,it’s essential not only to sense the system’s anomaly but also to diagnose the various classes of PV faults which saves the maintenance time and cost during the post-fault conditions.Therefore,the need for efficient fault detection and diagnosis methods for the common PV faults are increased,which accordingly enhances the generation efficiency of PV systems.In addition,neglecting detection and diagnosis of the PV faults may result in exceeding the power interrupts time of the PV systems which hence decreases the system availability and reliability.Forecasting the PV system working conditions,considering different PV configurations to build the PV system,under the different faults helps not only to identify the correct fault class but also choose the suitable configuration to enhance the system generation capability.In this research,three various configurations of PV systems are constructed and modeled,i.e.Series-Parallel(SP),Total cross-tied(TCT),and hybrid SP-TCT topology.These PV configurations are proposed to forecast the electrical features under four PV fault types with various classes using the MATLAB/Simulink platform.In the end,the simulation results can help the system operators to select the best and the worst configuration under typical occurred PV faults,entitled Partial Shading(PS),Open Circuit(OC),Line to Line(LL),and Short Circuit(SC)faults.Most of the research studies implemented in the fault detection task suffered from some drawbacks,such as,(1)the case of single fault condition are only considered while the condition of simultaneous multiple-faults isn’t studied which isn’t in line with the current practical applications;(2)they couldn’t distinguish between the permanent and temporary fault conditions on the DC side of the PV system which may results in wrong power interruptions due to the natural temporary conditions such as instant PS;(3)only a few research studies could accurately localize the various PV faults under changed environmental conditions,however,these studies bring additional implementation cost and time.Therefore,this research work revisits the Fault Detection and Diagnosis(FDD)task considering the usage of some Artificial Intelligent(AI)methods for detecting and classifying the abnormal conditions that occurred in the DC side of the PV systems.Using the AI methods has been increased,as an efficient tool,due to their fast and accurate pattern recognition capability.In this section,the thesis work can be presented as follow:1)Fuzzy Logic Control(FLC)is proposed to classify several temporary as well as permanent faults,while the simulated and experimental validations are executed on a real PV array under normal and abnormal conditions,taking into account the variation of solar irradiance and temperature.The proposed FLC-based method is built using predetermined three electrical indicators with associated threshold limits to detect the exact fault class.2)Several single-and multi-faults that occurred in the PV systems are examined to be classified using a Three Sequential Probabilistic Neural Network(TS-PNN)model.The implementation of TS-PNN gives a precise classification of the data inputs.In addition,the proposed method is successfully verified by theoretical and experimental ways using a real simulated PV array,to check its capability to diagnose the occurred faults.3)A novel Mixed Integer Linear Programming(MILP)optimization method is proposed due to its high execution accuracy and burden to localize the OC and SC faults on the PV modules.The proposed MILP methodology is built based on comparing the calculated power with the measured one to find exactly the faulty module location under the variations of environmental conditions.4)A deep-learning Convolutional Neural Network(CNN)classification method is proposed for the advantage of automatic feature extraction.To do so,three electrical indicators are predetermined and analyzed under three fault types.The proposed CNN method is successfully validated using several simulations and experimental tests using the SP configuration of the PV array.