Machine learning applications in estimating transformer loss of life

Abstract

Transformer life assessment and failure diagnostics have always been important problems for electric utility companies. Ambient temperature and load profile are the main factors which affect aging of the transformer insulation, and consequently, the transformer lifetime. The IEEE Std. C57.91-1995 provides a model for calculating the transformer loss of life based on ambient temperature and transformer's loading. In this paper, this standard is used to develop a data-driven static model for hourly estimation of the transformer loss of life. Among various machine learning methods for developing this static model, the Adaptive Network-Based Fuzzy Inference System (ANFIS) is selected. Numerical simulations demonstrate the effectiveness and the accuracy of the proposed ANFIS method compared with other relevant machine learning based methods to solve this problem. Index Terms-Adaptive Network-Based Fuzzy Inference System (ANFIS), transformer asset management, data-driven model, loss of life estimation.

Data Fusion and Machine Learning Integration for Transformer Loss of Life Estimation

“…Various data-driven machine learning methods, including but not limited to ANFIS, RBF and MLP, can be considered as suitable candidates for solving the estimation problems. Without addressing the details of these machine learning methods, and by referring to the companion paper [4], the transformer loss of life is estimated using these three methods, as shown in Fig. 2(a).…”

“…In this regard, the required data is synthesized on the basis of the mentioned IEEE standard. More details of the data synthesis process are not reported in this paper, but available in the companion paper [4]. The following cases are studied to investigate the performance of integration of the machine learning and data fusion techniques for estimating the transformer loss of life.…”

“…Given that a significant amount of data can be collected from sensors installed in transformers, machine learning methods can be of value in estimating transformer lifetime. A machine learning-based study with the goal of estimating transformer loss of life is proposed in [4]. By leveraging the historical data of transformer loading and ambient temperature, various machine learning methods, including Adaptive Network-Based Fuzzy Inference System (ANFIS), Multi-Layer Perceptron (MLP) network and Radial Basis Function (RBF) network are employed to accurately estimate the transformer loss of life.…”

“…The primary objective in this paper is to integrate data fusion and machine learning techniques for providing a more accurate and reliable estimation of transformer loss of life. Various types of machine learning methods to estimate the transformer loss of life, as proposed in the companion paper [4], set the stage for using data fusion techniques, and thus call for additional studies. In general, all tasks that demand any type of estimation from multiple sources can reap the benefit of using data fusion techniques.…”

“…Two types of data fusion techniques, including Ordered Weighted Averaging (OWA) and Kalman filter are presented and studied in this paper. By leveraging these data fusion techniques, two premier outputs estimated by machine learning methods [4], are fused together to improve the transformer loss of life estimation. Comparison between the proposed data fusion techniques, i.e., OWA and Kalman filer, is further provided in this work.…”