This paper presents a state-of-the-art review on machine learning (ML) based intelligent diagnostics that have been applied for partial discharge (PD) detection, localization, and pattern recognition. ML techniques, particularly those developed in the last five years, are examined and classified as conventional ML or deep learning (DL). Important features of each method, such as types of input signal, sampling rate, core methodology, and accuracy, are summarized and compared in detail. Advantages and disadvantages of different ML algorithms are discussed. Moreover, technical roadblocks preventing intelligent PD diagnostics from being applied to industry are identified, such as insufficient/imbalanced dataset, data inconsistency, and difficulties in cost-effective real-time deployment. Finally, potential solutions are proposed, and future research directions are suggested.

Condition Monitoring Based on Partial Discharge Diagnostics Using Machine Learning Methods: A Comprehensive State-of-the-Art Review

Power transformers are of vital importance for the stable operation of power system. Hence, to grasp transformers’ health condition is of imminent importance. However, the evaluation of transformers’ health condition is bound with fuzzy, uncertainty, and even conflict information. To address these problems, a novel evaluation model for transformer, by integrating the merits of fuzzy set theory, fuzzy analytical hierarchical process (AHP), and modified weighted averaging combination, is proposed in this study. The evaluation model contains four factors: dissolved gas analysis, electrical testing, oil testing, and miscellaneous factors, as well as 20 indices. The evaluation process consists of three steps. First, the fuzzy set theory is used to obtain assessment grades for each quantified indices. Second, the fuzzy AHP using fuzzy preference programming, which extends classical AHP and can better tackle the uncertainty existed in the comparison matrix given by experts, is introduced to calculate the weights of indices. Finally, the assessment is processed through the modified weighted averaging combination, which is able to handle the conflicts among evidences. The proposed method is verified by evaluating a realistic transformer, and the results of field examples indicate that the method can evaluate the transformer health condition effectively.

Research on multi-attribute decision-making in condition evaluation for power transformer using fuzzy AHP and modified weighted averaging combination

The use of SF6 gas decomposition components for the fault diagnosis of gas-insulated equipment (GIE) has received widespread attention and been extensively adopted by the power industry. However, the present focus is mainly on the use of a component type or absolute content for generating qualitative judgments on the existence of a fault, and the practical diagnosis method for field diagnosis is extremely lacking. In this paper, ten-year fault data were collected and analyzed. Decomposition components were extracted and combined, based on the decomposition characteristics of SF6, and enhancement factors were given to three component characteristic quantities using fuzzy C-means clustering algorithm. The component characteristic quantities that characterized the high-energy discharge fault, partial discharge, and partial over-thermal fault of the device were extracted: % (SOF2+SO2), % (9SO2F2), and % (5CO2). With adoption of these characteristic quantities as the coordinate axes of the triangle coordinate system, a triangle fault diagnosis method for field diagnosis of SF6 GIE was constructed. The diagnosis method was validated by a large number of experimental data and field failure data, and 128 of the 150 experimental samples were correctly recognized. Moreover, the test conditions did not considerably affect the diagnostic results of the method. This technique can provide a simple and effective field diagnostics technology for the power industry.

Triangle Fault Diagnosis Method for SF 6 Gas-Insulated Equipment

We propose a high sensitivity optical fiber sensor for detecting the acoustic emission generated by the partial discharge in gas-insulated-switchgear (GIS). A model is established to investigate the effects of the cylinder coil on the sensitivity, the bandwidth, and the detection limit. After that, the proposed sensor was modified based on the effects. The average sensitivity of the optimized acoustic emission (AE) optical fiber sensor is 32 dB higher than that of the conventional lead zirconate titanate (PZT) in the frequency range from 20 kHz to 100 kHz. Moreover, the experimental results show that the detectable partial discharge initial voltage for the proposed optical system is 17.1% lower than that for the PZT system, and the amplitude of the signal detected by the proposed optical system is 525% higher than that detected by the PZT system.

A High Sensitivity Optical Fiber Sensor for GIS Partial Discharge Detection

Scientific partial discharge (PD) severity evaluation is highly important to the safe operation of gas-insulated switchgear. However, describing PD severity with only a few statistical features such as discharge time and discharge amplitude is unreliable. Hence, a deep-learning neural network model called stacked sparse auto-encoder (SSAE) is proposed to realise feature extraction from the middle layer with a small number of nodes. The output feature that is almost similar to the input PD information is produced in the model. The features extracted from PD data are then fed into a soft-max classifier to be classified into one of four defined PD severity states. In addition, unsupervised greedy layer-wise pre-training and supervised fine-tuning are utilised to train the SSAE network during evaluation. Results of testing and simulation analysis show that the features extracted by the SSAE model effectively characterise PD severity. The performance of the SSAE model, which possesses an average assessment accuracy of up to 92.2%, is better than that of the support vector machine algorithm based on statistical features. According to the tested number of SSAE layers and features and the training sample size, the SSAE model possesses good expansibility and can be useful in practical applications.

https://ietresearch.onlinelibrary.wiley.com/doi/epdf/10.1049/iet-smt.2016.0326

Assessment of PD severity in gas-insulated switchgear with an SSAE

Partial discharge (PD) is an early manifestation of insulation fault that occurs in electrical equipment, which holds rich feature information used to insulation assessment. In this study, step voltage method was applied to investigate the PD development process of typical protrusion defects in gas-insulated switchgear. Nine features were extracted to represent the deterioration degree of internal insulation. Fuzzy C means clustering was applied to solve the state centres. PD severity was divided into four states based on the PD developing process, and the harmfulness of each state was illustrated. Finally, adaptive objective weight technology based on the theory of maximising deviation was adopted, and two-level fuzzy comprehensive evaluation (TL-FCE) model was established to address the problem of the weight allocation of different features on the evaluation result. The experimental test has been conducted to test the validity and effectiveness of the model. It is proved that TL-FCE owns good performance to assess PD severity, and it is more suitable to handle fuzzy boundary problems than support vector machine.

Feature extraction and severity assessment of partial discharge under protrusion defect based on fuzzy comprehensive evaluation

In order to study the development of partial discharge (PD) under typical protrusion defects in gas-insulated switchgear, we applied step voltages on the defect and obtained the φ-u and φ-n spectrograms of ultra-high frequency (UHF) PD signals in various PD stages. Furthermore, we extracted seven kinds of features to characterize the degree of deterioration of insulation and analyzed their values, variation trends, and change rates. These characteristics were inconsistent with the development of PD. Hence, the differences of these features could describe the severity of PD. In addition, these characteristics could provide integrated characteristics regarding PD development and improve the reliability of PD severity assessment because these characteristics were extracted from different angles. To explain the variation laws of these seven kinds of parameters, we analyzed the relevant physical mechanism by considering the microphysical process of PD formation and development as well as the distortion effect generated by the space charges on the initial field. The relevant physical mechanism effectively allocated PD severity among these features for assessment, and the effectiveness and reliability of using these features to assess PD severity were proved by testing a large number of PD samples.

Yu-lin Dong

Papers

Feature extraction and severity assessment of partial discharge under protrusion defect based on fuzzy comprehensive evaluation

Features Extraction and Mechanism Analysis of Partial Discharge Development under Protrusion Defect