This paper presents an online data-driven diagnosis method for multiple IGBTs open-circuit faults and current sensor faults in the three-phase PWM inverter. The Fast Fourier Transform (FFT) algorithm is used to extract the fault frequency spectrum features of the three-phase currents. Then, a feature selection method named ReliefF is introduced to select the most critical features by removing redundant and irrelevant features. In addition, as a novel fast learning technology, Random Vector Functional Link (RVFL) network is applied to learn the faulty knowledge from the historical dataset. Based on the well-learned model, the fault type and location of the converter can be accurately identified as long as the three-phase current signals are measured. Offline test results verify that the proposed method can identify the IGBT and sensor faults with an accuracy of 98.83% and outperforms state-of-the-art learning algorithms. Moreover, the real-time hardware-in-the-loop test results show that the proposed method can successfully identify the IGBT faults and current sensor faults within 22 milliseconds. It is robust to the DC-link voltage fluctuations, model parameters, and speed or load variations. The extensibility of the proposed method is also validated based on the test results in terms of other fault modes and drive systems.

Three-phase PWM inverter fed induction motor drive system is widely applied in high power drive applications. Sensor faults are very common in the drive system which, once occur, might result in degraded system performance or even system shutdown. In order to rapidly and accurately diagnose the sensor faults, this paper proposes an intelligent time-adaptive data-driven method to identify the fault location and fault type of sensors in the drive system. An emerging machine learning technology named extreme learning machine (ELM) is applied to learn the sensor fault dataset, an ensemble ELM classifier is then designed to improve diagnostic accuracy, based on which a time-adaptive fault diagnosis process is proposed to achieve a high and balanced diagnostic accuracy and speed. As a data-driven method, the proposed method only employs the phase current, DC-link voltage, and speed signals as the inputs to the ensemble ELM classifiers, and requires no additional sensors and other hardware. Simulated and experimental tests show that the proposed method can rapidly and accurately detect the fault sensor location and identify offset fault, stuck fault, and noise faults with an average diagnostic accuracy of 98% and average decision time of 10ms after the fault occurs.

Machine learning (ML) based data-driven methods have shown promising performance in power converter fault diagnosis. However, the existing ML model trained by one fault database can only work for the corresponding converter system, but cannot work accurately for a different system with the same topology but different parameters. In this article, a novel transferrable data-driven fault diagnosis is proposed for insulated gate bipolar transistor open-circuit fault diagnosis in three-phase inverters. First, three-phase current signals are sampled as original inputs and fault features are generated by manifold feature learning. Then, an extreme learning machine model is trained by the data from the source system to form an initial diagnostic model. After that, a model adaption process is designed to adjust the model's parameter by minimizing the distribution divergence between the data from the source and target systems. As the biggest advantage, one trained diagnostic model can be applied for different systems without the need for multiple training processes. Finally, offline tests and real-time experiments are conducted to verify the performance of the proposed transferrable method.

Multiple individual microgrids can be integrated as a networked microgrid system for enhanced technical and economic performance. In this paper, a two-stage data-driven method is proposed to hierarchically coordinate individual microgrids towards decentralized operation in a networked microgrid (NMG) system. The first stage schedules active power outputs of micro-turbines and energy storage systems (ESSs) on an hourly basis for energy balancing and cost minimization, where ESSs are controlled by a local P/SoC droop scheme. In the second stage, the reactive power outputs of PV inverters are dispatched every three minutes based on a Q/V droop controller, aiming to reduce network power losses and regulate the voltage under real-time uncertainties. At offline training stage, a multi-agent deep reinforcement learning model is trained to learn an optimal coordination policy, enhanced by a safety model framework. For online application, the trained agent can work locally in a decentralized manner without information exchanges, and the safety model can also be applied to monitor and guide online actions for safety compliance. Numerical test results validate the effectiveness and advantages of the proposed method.

Multiple microgrids can be interconnected to form a networked-microgrid (NMG) system. In this paper, a data-driven decentralized economic frequency control method is proposed for isolated NMG systems. Based on multi-agent deep reinforcement learning (MA-DRL) framework, each DRL agent controls the generator and energy storage system (ESS) in each microgrid of the NMG. For offline training stage, the optimal control strategy is learned by soft actor-critic (SAC) algorithm with a global reward, which aims to restore system frequency while considering economy. Besides, to satisfy system constraints and avoid high learning costs, a safety model scheme is designed and trained to support each DRL agent. Since the agents are trained in the centralized learning process at offline stage, they are able to coordinate in a decentralized manner for online application, which only requires local information to generate optimal control actions. Also, the trained safety model can be applied for online stage to monitor and guide online actions. Finally, numerical tests are conducted to demonstrate the feasibility and effectiveness of the proposed method under the variation of renewable generation and load demand.