Feature Dimension Reduction And Clustering Method For Photovoltaic System Fault Diagnosis

With the rapid depletion of fossil energy,new energy has also become more and more concerned by other countries.Photovoltaic power generation has gradually entered people's vision due to its advantages of less pollution and zero emissions.Due to the variability of the external environment of the photovoltaic power generation system and the complexity of the internal model,a variety of faults are prone to occur,and modeling is difficult.Monitoring methods for photovoltaic systems are extremely important.Simple monitoring and manual inspection are not only time-consuming and inefficient,but it is also difficult to realize the intelligence and informationization of photovoltaic systems.With the large-scale application of photovoltaic power plants,the data scale has gradually increased,and traditional fault diagnosis methods cannot ideally solve the problem under this situation.Therefore,the requirements for fault diagnosis methods are getting higher and higher,and the processing methods using artificial intelligence are more adapted to the development of the times.By summarizing domestic and foreign references,a PCA + hierarchical clustering method based on data feature dimension reduction with less prior knowledge and large amount of data is proposed to intelligently diagnose the multi-dimensional data of the photovoltaic power generation system..Most of the papers in the literature use simulation software to simulate and do not reflect the photovoltaic power generation system under actual operating conditions.The team uses the independently developed photovoltaic power generation system to set up real experiments and verify the method outdoors.The main contents and innovations of this article are as follows:Firstly,introduce the research background and history of photovoltaic power generation system,summarize the diagnostic methods and clustering research of photovoltaic system failures in photovoltaic field,analyze the operating mechanism of photovoltaic power generation system components under multiple operating conditions,and explain the real faults The data characteristics of the data.Under the real operating environment,the amount of data obtained is huge,the prior knowledge is small,and the lack of real data labels,so traditional supervised learning fault diagnosis methods cannot build accurate models,and they cannot get good results in real scenarios.Be applicable.Secondly,multi-dimensional raw data can be obtained during the operation of a photovoltaic power generation system.After the data is obtained,the data needs to be pre-processed.It is also extremely important to use a reasonable pre-processing method.However,most of the methods in the literature ignore the characteristics of the data itself.In this paper,a suitable normalization method is selected according to the characteristics of the data to retain the characteristics of the original data information to the maximum.Then,PCA is used to reduce the multidimensional data.At the same time,based on the characteristics of the data and little prior knowledge,a hierarchical clusteringmethod is used to distinguish faults.A variety of external indicators have been introduced for the problem that clustering results cannot be evaluated,and the evaluation of clustering results has certain objectivity.Finally,because the methods in the literature use simulation software to simulate,it is impossible to verify the complexity of the photovoltaic power plant during operation and the effectiveness of the method used in the real environment.Therefore,in this experiment,a self-made photovoltaic power plant microsystem was used,and outdoor real experiments were set up to verify the method.Comparing the proposed PCA + hierarchical clustering method based on feature dimensionality reduction with PCA + k-means method,direct hierarchical clustering method,and direct k-means method,it is found that the PCA + hierarchical clustering method based on feature dimensionality reduction has more Good results.