۱- Introduction

۲- Problem formulation

۳- Proposed models

۴- Experimental results

۵- Conclusions

References

Abstract

A factor that directly impacts the lifespan of a power transformer is the hot-spot temperature, and its monitoring is vital to prevent faults, reduce costs, keep the safety, and provide a reliable service to consumers. In this paper, we propose two forecasting models to predict the hot-spot temperature of power transformers. The first is the implementation of Set-Membership in the evolving Participatory Learning with Kernel Recursive Least Squares. And the second is a combination of the evolving Participatory Learning with Kernel Recursive Least Squares and the improved version of the Set-Membership concept, named Enhanced Set-Membership. Both Set-Membership and the Enhanced Set-Membership approaches are implemented to update the rate of change of the arousal index, which is a parameter that controls the creation of rules. A data set collected from an experimental transformer is adopted to evaluate the model’s performance. The obtained results are compared with the performance of the original evolving Participatory Learning with Kernel Recursive Least Squares and with the performance of other classical models suggested in the literature. The proposals have lower errors and a competitive number of final rules, suggesting that the models are efficient approaches to modeling complex data with high accuracy.

Introduction

The power transformer is a critical equipment in power distribution [1]. It is responsible for stepped-up the voltage before to be transmitted over long distances to reduce waste, and stepped-down the voltage to provide the energy to consumers safely [2,3]. Due to the composition of a power transformer, it is the most expensive apparatus in energy distribution [4]. In the case of a power transformer’s failure, when the recovering process is possible, it is slow and inefficient [5]. Thereof, monitoring is vital to prevent faults, reduce costs, keep the safety, and provide a reliable service to consumers [6,7]. The annual spent on power transformers’ monitoring hardware will increase more than $ 642 million in eight years until 2020, according to [8,9], indicating the importance of the power transformers in power distribution. Internal failures are about 10% of the total faults, and, among them, winding and bushing defects represent approximately 44% [10]. The bushing is a fragile component constituted of four parts: insulation, conductor, connection clamp, and accessories[11,12]. In the present work, we considered power transformers composed of Resin-bonded paper bushings (RBP) [13]. The principal factor in bushing failures is the hot-spot temperature, representing 32% of the total causes [11]. The hot-spot temperature is the highest temperature near to the top of the power transformers highvoltage (HV)/low-voltage (LV) windings [7,14] and represents the main limiting factor in the load capacity of the transformer [14], since increases in this temperature reduces the lifespan of the insulation and may determine the end life of the power transformer [15]. As the estimation of the hot-spot is a complex task, many models have been proposed in the literature with the purpose of estimating the hot-spot temperature of power transformers. Among them, the most commonly used in practice is the model based on the IEEE Standard C57.91-2011 [16] which is based on transient heating equations and specific thermal characteristics and parameters of power transformers. This deterministic model is imprecise due to assumed simplifications, and consequently, the power transformer must operate below the maximum capacity to prevent damages [7].