This paper introduces a new application of Deep Belief Network (DBN) as an emerging artificial neural network for recognizing and classifying different partial discharge (PD) patterns. Phase Resolved Partial Discharge (PRPD) technique with different window intervals is used to manipulate three PD types including corona, surface and internal discharges measured in a High Voltage (HV) lab. Four approaches are proposed for extracting features from the raw measured data. In the first approach, the DBN is used as both a feature extractor and a PD classifier. The other three approaches extract discriminatory features using statistical and vector norm based operators to train a DBN classifier. The impact of the various phase windows through the PRPD method on the performance of the trained classifier is evaluated to obtain the best window interval. It is shown that the deep architectures are capable of learning important distinguishable features from PD data without any data preprocessing. This eliminates time-consuming feature extraction processes which produce the hand crafted features. Based on a comparison analysis, when the input data is corrupted by noise levels or no feature extraction technique is used to preprocess the data, the proposed approach outperforms other techniques such as Artificial Neural Networks (ANNs), Fuzzy logic classifiers and Support Vector Machines (SVMs).

This paper considers the problem of distributed adaptive fuzzy event-based finite-time prescribed performance leader-following tracking control for heterogeneous nonlinear multi-agent systems (NMASs) over a directed topology. Each agent is considered in a non-affine nonstrict-feedback form under input saturation and output constraint which contains unknown dynamics and external disturbances. Fuzzy logic systems (FLSs) are exploited as an effective online approximation tool to tackle system uncertainties. By employing the unique property of FLS, the algebraic loop problem is overcome and by designing novel adaptive laws of the FLS weights, the computation burden is decreased significantly. The threshold of the event-triggered condition is improved compared to the conventional relative-threshold mechanism. A modified performance function called finite-time performance function is introduced to constrain the synchronization errors within the prescribed performance bounds in finite time. The dynamic surface control technique is then developed to avoid the issue of the explosion of complexity. Moreover, by developing a new decomposition for the controller gain function resulting from the mean-value theorem and introducing an auxiliary system, the input saturation nonlinearity that affects the non-affine form stability is handled. Through the Lyapunov stability analyses, it is shown that the developed control algorithm ensures the closed-loop NMAS trajectory to be cooperatively semi-globally uniformly ultimately bounded. Additionally, the tracking errors are driven to a predefined region around zero in finite time. Finally, the efficiency of the established theoretical results is validated by the simulation studies.

Differential protection is the most common type of protection in power transformers. However, inrush current due to transformer energization may appear as fault current to the protective relay, causing a false trip of the transformer. Therefore, discrimination of the transformer inrush current from internal faults is necessary for improving the security of the protection scheme. In this paper, a new method for discrimination of the transformer inrush current from internal fault current is proposed. First, nonlinear state-space model of a real single-phase transformer is derived, which incorporates the nonlinear phenomena of hysteresis and magnetic saturation. Based on the derived model, an Extended Kalman Filter (EKF) is used for estimation of the primary winding current. A residual signal is defined as the difference between the measured and estimated currents. When a healthy transformer is energized, the EKF perfectly estimates the primary winding current and hence, the residual signal is almost zero. However, when the transformer is faulty, the EKF cannot effectively estimate the current due to the existing large model mismatches. Consequently, a large residual signal is created. The superiorities of the proposed algorithm are demonstrated using different experimental scenarios.

Monitoring of winding faults is of great significance for the assessment of the transformer maintenance status. In this study, the time series analysis is combined with frequency response analysis (FRA) for better interpretation and analysis of FRA results. Here, healthy baseline residual data are modeled using the time series analysis for faults detection of the transformer winding. Different statistical software, such as R 4.0.2 and ITSM2000, are used to analyze measured data. To detect the transformer’s winding faults, various analyses, such as the autocorrelation and partial autocorrelation functions plot, the Box–Pierce test, the Ljung–Box test, the McLeod–Li test, the turning points test, the Wallis and Moore phase–frequency test, and the Jarque–Bera test, and the order of minimum corrected Akaike’s information criteria are used. The simulation results show that the suggested method is able to classify and discriminate mechanical and electrical faults of the transformer’s winding with good accuracy.