Fault Detection Method For Hot-spot Of Photovoltaic Systems Based On Manifold Learning

In recent years,photovoltaic(PV)energy has received widespread attention in the industry and is one of the important new energy sources.PV energy systems are devices that generate PV energy,utilizing the PV effect to directly convert solar energy into electrical energy,which can be directly fed into the power grid or stored.However,when abnormal faults occur in the PV system,it not only reduces power generation efficiency but also affects the safety and reliability of the system.Hot spots are one of the causes of abnormal failures in PV systems,which not only damage PV modules but also bring fire hazards with great harm.Usually,the generation of hot spot effects can be detected using sensor data from PV systems.However,when facing a large amount of complex high-dimensional nonlinear sensing data,how to effectively preprocess and extract features to improve the accuracy of hot spot effect detection is a current research hotspot.In response to this,this article proposes a PV array hot spot detection method based on popular learning.Manifold learning can effectively characterize the internal structure of high-dimensional data,map high-dimensional data to low-dimensional data,and combine data-driven methods to process and analyze the data,effectively detecting the hot spot effect of PV systems,and ensuring the reliability and feasibility of the algorithm.The main work of this article is as follows:(1)Introduced the basic structure,applicable environment,and operating mechanism of PV systems,and summarized the main types of faults that exist in PV systems;The importance and practical significance of hot spot effect detection were elaborated by combining the internal and external factors generated by the hot spot problem;Finally,a detailed introduction was given to manifold learning and several nonlinear algorithms for manifold learning.(2)A method combining neighborhood preserving embedding(NPE)and canonical correlation analysis(CCA)was proposed to analyze and study the PV component data generated by the hot spot problem;And proposed a grouping technique that utilizes the correlation between various components to simply partition component parameters and improve the correlation between data.By using NPE to analyze and reduce the dimensionality of data,a judgment model for abnormal phenomena is constructed.This method can diagnose and study the PV system without relying on the physical knowledge of the system.Finally,the effectiveness of the method was verified through case analysis and simulation experiments.(3)An early hot spot(EHs)detection method based on an incremental-correlation local tangent space alignment(Ic LTSA)algorithm is proposed.When there are slight anomalies in the PV panel,this method can process the data through incremental LTSA method,extract potential manifolds in system measurements,find the optimal local arrangement matrix to minimize reconstruction errors,obtain nonlinear information hidden in monitoring data,and determine its trend change information to detect EHs.This method reduces computational load,improves detection accuracy,and increases the feasibility of application in PV systems.