| Low-carbon development has become an important measure for the coordinated development of resources and the environment,and new energy power generation has become the main source of electricity in some developed countries.With the service life of PV arrays increasing,the probability of faults also increases.PV array faults can manifest in various types and have varying degrees of impact on power generation,significantly affecting the efficiency of the PV power station.If not addressed in time,these faults can accelerate the aging of the PV array,shorten its lifespan,and even result in serious safety accidents.In this thesis,a ResNet-based fault diagnosis algorithm using sliding window is proposed,a fault diagnosis model is designed and its effectiveness and superiority is verified.The research content is as follows:(a)Several types of PV array fault diagnosis methods are analyzed,and the fault diagnosis direction was determined based on the primary features of irradiance,temperature,and PV array output current and voltage.Then a PV array simulation model is constructed to study the effects of different irradiance and temperature on the output of PV arrays.(b)Maximum power point tracking for PV arrays is implemented based on P&O method.Using the simulation model,the causes of hot spots and the role of bypass diodes,as well as the output characteristics of the PV array under various fault conditions and the impact of faults of varying degrees on PV array output are analyzed.Based on the summarized simulation results,the dataset of the PV array is cleaned up.(c)A sliding window-based ResNet classification algorithm is innovatively proposed for PV array fault diagnosis.The PV array dataset is preprocessed using normalization,one-hot coding,and sliding window,and a classification model is established to recognize and classify faults.To prove the superiority of this algorithm,classification models based on other algorithms are constructed as a comparison,and a variety of evaluation metrics are used for comparative analysis.To reduce the large number of model parameters,which cause slow model training and slow response of the fault diagnosis interface,the model is optimized by reducing the depth of the neural network without degrading the classification effect.For PV power plants without complete fault datasets,a transfer learning method based on simulation data is proposed,validated and analyzed.(d)Finally,a PV array fault detection platform is implemented based on WebSocket technology.The platform is developed based on vue.js,ECharts and other frameworks,adopting a front-end and back-end separation mode for easy expansion and maintenance.The fault diagnosis model is implemented as a web interface,enabling real-time monitoring of the PV arrays,thus forming a complete system at the software level. |
