The power transformer is one of the most critical and expensive components in a power grid. Today, the monitoring and condition assessments of power transformers have become essential tasks in many power utilities. Within the condition-based asset management of power transformers, priority attention should be directed to transformer insulation monitoring, diagnosis and data interpretation. The conditions of the transformer insulation systems are influenced by multiple factors, including mechanical, electrical and environmental stresses. This implies that the health condition of a transformer insulation system should be evaluated in a comprehensive manner.
Over the past decade, various diagnostic methods have been developed to assess the condition of transformer insulation systems [1-3]. Among these diagnostic techniques, the dissolved gas analysis (DGA) and oil characteristics tests are of great importance for evaluating the quality of the transformer oil-paper insulation systems. The DGA technique is used to detect faults that occur in the transformers, such as thermal faults, arcing and partial discharge. The oil characteristics tests evaluate the ageing condition of the transformer insulation. However, the assessments provided by any individual diagnostic technique normally deal with one aspect of the transformer’s health status and different diagnostic methods may have varying degrees of significance in determining the overall health condition of the transformer. Therefore, to improve the quality of the condition assessments of the transformer insulation systems, the Health Index (HI) approach is proposed and is widely accepted across utility companies. The Health Index is an objective and quantitative approach that combines the results from various chemical and electrical tests, onsite inspections, and information regarding the transformer’s operation and loading history. One of the key challenges of the HI approach lies in integrating the diagnostic results of all the tests by assigning an appropriate weighting factor for each individual test.
This thesis aims to develop an HI for insulation systems using liquid-filled power transformers. The HI approach combines the diagnosis results obtained from the DGA and oil characteristics tests and provides an overall health condition assessment on the transformer oil-paper insulation system. In this thesis, a suite of intelligent algorithms for HI computation are developed, based on machine learning techniques. These techniques use the historical test results for which the corresponding transformers conditions are known to construct a mathematical model; this can then be used in determining the HI for the transformer of interest. The procedure for constructing the training database for the intelligent algorithms is described in this thesis. In particular, this thesis applies the support vector machine (SVM) and fuzzy support vector machine (FSVM) techniques for automatically computing the health index of the insulation system for an in-service transformer from its oil characteristic tests results. Additionally, several data pre-processing techniques are implemented to deal with noises, outliers and class imbalance problems in the training dataset and thus improving the overall accuracy of the HI computation for the transformers under investigation.
The growing concern for environmental protection and fire safety has led to the adoption of biodegradable oil-filled transformers in the power industry. Thus, this thesis also investigates the properties of several types of biodegradable oil and compares the properties of conventional mineral oils. Such investigations will help to extend the intelligent HI algorithm for evaluating the health condition of the insulation systems in biodegradable oil-filled transformers.