Key Parameter Extraction and Condition Assessment for Power Transformer Based on Factor Analysis and D-S Evidence Theory

Abstract

Abstract. The condition Assessment of transformer is an important basis for its operation, inspection and maintenance. But a transformer has a variety of Assessment parameters, many of which are redundant. How to objectively and reasonably select the Assessment parameters remains a challenge. To solve this problem, this article analyzed thoroughly the requirements and research status of transformer condition Assessment. On this basis, a multi-level fuzzy comprehensive Assessment model based on the key Assessment parameter system and Dempster-Shafer (D-S) evidence theory was constructed. Specifically, the key parameters were scientifically extracted using factor analysis, and an indicator system consisting of key parameters was constructed. Then, in combination with the analytic hierarchy process (AHP), the optimal weights of each sub-indicator in the assessment system were acquired, and the Assessment results of each item layer were obtained based on the respective membership function. At last, based on D-S evidence theory, the Assessment results of each item layer were fused for the decision-making in Assessment. Case analysis results indicate that the proposed model can effectively and reasonably assess the transformer's operating condition.

Abandoned terrace recognition based on deep learning and change detection on the Loess Plateau in China

“…However, deep learning obtained satisfactory results in our study using only one data source and one model, which is more convenient than other studies. Change detection has a good identification effect for extracting abandoned land (Yang et al, 2019), but problems arise when identifying abandoned land outside the monitoring period (Yang et al, 2020). However, deep learning can identify current abandoned land using realtime image features.…”

“…We used an image semantic segmentation method based on deep learning to recognize and extract abandoned terraces for the first In terms of data sources, the image resolution used by conventional change detection methods is medium to high (Han, 2019;Song, 2015;Song et al, 2011;Yang et al, 2019;Zheng et al, 2016), with some of the data available free of charge and thus convenient For the identification results of abandoned terraces, semantic segmentation is slightly better than change detection (Table 5), mainly because it can recognize small broken terraces. In contrast, change detection does not easily recognize small abandoned terraces or terraced edges due to the image resolution.…”

“…In addition, feature extraction based on NDVI time series data depends on the temporal and spatial resolution of remote sensing data when applied on a large or national scale (Shi, 2020). Computer technology has enabled remote sensing technology, machine learning (Low et al, 2015;Wang et al, 2020), and change detection (Cheng et al, 2011;Kuemmerle et al, 2008;Witmer, 2008;Yang et al, 2019;Zhang, 2010). The most important part of decision tree classification is the establishment of discriminant rules, using differences in the reflection spectral curves of different objects.…”

“…Han (2019) used change detection and machine learning (decision tree) to extract abandoned land, reporting an overall accuracy of 87% and 80%, respectively. Yang et al (2019) extracted abandoned farmland based on change detection, with an overall extraction accuracy of 97.6%. However, it is difficult for the above methods to identify small fragmented plots in mountainous areas (Shi, 2020), and smooth abandoned land edges cannot be obtained, which leads to inaccurate calculation of abandoned land area.…”

“…abandoned land in Chifeng City of China in 2017 using the change detection based on SENTINEL-2 images, with an overall classification accuracy of 87%, PA of 82.54%, and kappa of 0.73 Yang et al (2019). used change detection to extract the annual distribution of abandoned land in Helinger County of China from 2014 to 2017, mainly using GF-1 and SENTINEL-2 images, with an overall classification accuracy of 94.5% and kappa of 0.88.…”