Photovoltaic array fault diagnosis method based on lstm-dae

By combining LSTM-DAE with SVM-LR, and using unlabeled data to pre-train the model, the temporal features of the photovoltaic array are extracted and the model parameters are optimized. This solves the problem of low accuracy in photovoltaic array fault diagnosis in existing technologies and achieves efficient fault type identification.

CN119557722BActive Publication Date: 2026-06-09HUAIAN OF JIANGSU ELECTRIC POWER CO POWER SUPPLY
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Patent Type
Patents(China)
Current Assignee / Owner
HUAIAN OF JIANGSU ELECTRIC POWER CO POWER SUPPLY
Filing Date
2024-10-09
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing single machine learning algorithms have low accuracy and high learning costs in photovoltaic array fault diagnosis, are difficult to effectively handle nonlinear relationships and noise, and are overly dependent on labeled data.

Method used

A semi-supervised learning method based on LSTM-DAE is adopted. The LSTM-DAE model is pre-trained using unlabeled data, and SVM and LR are combined for feature selection and model optimization. Temporal features are extracted by LSTM, and a hybrid SVM-LR classifier is used for fault classification.

Benefits of technology

It improves the accuracy of photovoltaic array fault diagnosis, reduces the dependence on labeled data, and enhances the model's feature learning ability and fault type identification ability.

โœฆ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119557722B_ABST
    Figure CN119557722B_ABST
Patent Text Reader

Abstract

The application discloses a photovoltaic array fault diagnosis method based on LSTM-DAE. The current and voltage information of the photovoltaic module is collected and pretreated; whether the fault occurs is judged according to whether the maximum power difference between the measured curve and the reference curve exceeds the fault early warning threshold; when the fault is detected, the voltage and current and voltage information are read, and the sample is trained and tested; the sample is introduced into the LSTM-DAE model for feature extraction, wherein the model is an LSTM-DAE model constructed in an offline state by using historical photovoltaic array fault data as a sample; the extracted feature data is transmitted to the SVM-LR hybrid classifier, the output of the SVM is fitted into the LR classifier by a binomial method to optimize the performance of the model parameters, and after k-fold cross validation, the final fault classification model is obtained. Compared with the prior art, the application can reduce the processing of label data, effectively improve the accuracy and efficiency of fault detection, and realize real-time monitoring and fault diagnosis in photovoltaic fault classification.
Need to check novelty before this filing date? Find Prior Art