Fast-training feedforward neural network for multi-scale power quality monitoring in power systems with distributed generation sources
TL;DR: A deep learning approach for power quality monitoring in systems with distributed generation sources is presented that combines a signal processing stage using variational mode decomposition (VMD) to obtain the times scales of multi-component signals, and a deep learning stage using a simple feedforward neural network to classify the disturbances.
About: This article is published in Measurement.The article was published on 2021-01-01. It has received 16 citations till now. The article focuses on the topics: Feedforward neural network & Deep learning.
Citations
More filters
TL;DR: In this article , a metaheuristic technique called electrostatic discharge algorithm (ESDA) is coupled with an artificial neural network (ANN) to create the proposed hybrid, which is compared to several conventionally trained ANNs to investigate the effect of hybridization.
56 citations
TL;DR: The proposed hybrid convolutional neural network method is a novel approach that covers the steps of an expert examining a signal and its classification performance is relatively high compared to other methods, the computational complexity is almost the same.
Abstract: As a result of the widespread use of power electronic equipment and the increase in consumption, the importance of effective energy policies and the smart grid begins to increase. Nonlinear loads and other loads in electric power systems are considered as the main reason for power quality disturbance. Distortions in signal quality and shape due to power quality disturbance cause a decrease in total efficiency. The proposed hybrid convolutional neural network method consists of a 1D convolutional neural network structure and a 2D convolutional neural network structure. The features acquired by these two convolutional neural network architectures are classified using the fully connected layer, which is traditionally used as the classifier of convolutional neural network architectures. Power signals are processed using a 1D convolutional neural network in their original form. Then these signals are converted into images and processed using a 2D convolutional neural network. Then, feature vectors generated by 1D and 2D convolutional neural networks are combined. Finally, this combined vector is classified by a fully connected layer. The proposed method is well suited to the nature of signal processing. It is a novel approach that covers the steps of an expert examining a signal. The proposed framework is compared with other state-of-the-art power quality disturbance classification methods in the literature. While the proposed method's classification performance is relatively high compared to other methods, the computational complexity is almost the same.
34 citations
TL;DR: In this article , the authors present the main barriers, research gaps, gains and suggestions for applying deep learning to power quality, including lack of novelty, low transparency of deep learning methods and lack of benchmark databases.
12 citations
TL;DR: The proposed framework produces higher performance than other state-of-the-art methods in the literature and is able to classify composite PQD signals that it has not encountered before.
Abstract: Distributed generation (DG) sources are preferred to meet today's energy needs effectively. The addition of many different types of renewable energy sources to the grid causes various problems in signal quality. Detection and classification of these problems increase efficiency by both the producer and the consumer. In the literature, incredibly singular and some composite power quality disturbance (PQD) detection is performed effectively. However, the multitude of composite PQD variations degrades the performance of existing algorithms. In this study, the classification of all PQD variations that may occur is performed by using singular PQD and some composite PQD signals. A different number of subcomponents representing the signal are created according to each signal characteristic. Instantaneous energies from these subcomponents are used as deep learning (DL) input. Deep learning cycles are created as much as the instantaneous energy number of each signal. Each cycle has specific features of defining a single event. Therefore, the proposed approach is able to classify composite PQD signals that it has not encountered before. The proposed method's performance is first evaluated with the known PQD events and compared with the current state-of-the-art methods in the literature. Then, a dataset containing the combinations of different events not encountered during the training is created, and the performance is evaluated on this dataset. In the experiments performed, it is revealed that the proposed framework produces higher performance than other state-of-the-art methods
11 citations
TL;DR: In this article , a Support Vector Machine (SVM) was used to classify power quality events in a power system. But, the performance of the SVM was not as good as other contemporary optimizers.
Abstract: Power quality has emerged as a sincere denominator in the planning and operation of a power system. Various events affect the quality of power at the distribution end of the system. Detection of these events has been a major thrust area in the last decade. This paper presents the application of Support Vector Machine (SVM) in classifying the power quality events. Well-known signal processing techniques, namely Hilbert transform and Wavelet transform, are employed to extract the potential features from the observation sets of voltages. Supervised architecture consisting of SVM has been constructed by tuning the parameters of SVM by various algorithms. It has been observed that Augmented Crow Search Algorithm (ACSA) yields the best accuracy compared to other contemporary optimizers. Further, Principal Component Analysis (PCA) is employed to choose the most significant features from the available features. On the basis of PCA, three different models of tuned SVMs are constructed. Comparative analysis of these three models, along with recently published approaches, is exhibited. Results are validated by the statistical one-way analysis of variance (ANOVA) method. It is observed that SVM, which contains attributes from both signal-processing techniques, gives satisfactory results.
5 citations
References
More filters
TL;DR: This work proposes an entirely non-recursive variational mode decomposition model, where the modes are extracted concurrently and is a generalization of the classic Wiener filter into multiple, adaptive bands.
Abstract: During the late 1990s, Huang introduced the algorithm called Empirical Mode Decomposition, which is widely used today to recursively decompose a signal into different modes of unknown but separate spectral bands. EMD is known for limitations like sensitivity to noise and sampling. These limitations could only partially be addressed by more mathematical attempts to this decomposition problem, like synchrosqueezing, empirical wavelets or recursive variational decomposition. Here, we propose an entirely non-recursive variational mode decomposition model, where the modes are extracted concurrently. The model looks for an ensemble of modes and their respective center frequencies, such that the modes collectively reproduce the input signal, while each being smooth after demodulation into baseband. In Fourier domain, this corresponds to a narrow-band prior. We show important relations to Wiener filter denoising. Indeed, the proposed method is a generalization of the classic Wiener filter into multiple, adaptive bands. Our model provides a solution to the decomposition problem that is theoretically well founded and still easy to understand. The variational model is efficiently optimized using an alternating direction method of multipliers approach. Preliminary results show attractive performance with respect to existing mode decomposition models. In particular, our proposed model is much more robust to sampling and noise. Finally, we show promising practical decomposition results on a series of artificial and real data.
4,111 citations
TL;DR: In this article, real and reactive power management strategies of EI-DG units in the context of a multiple DG microgrid system were investigated. And the results were used to discuss applications under various microgrid operating conditions.
Abstract: This paper addresses real and reactive power management strategies of electronically interfaced distributed generation (DG) units in the context of a multiple-DG microgrid system. The emphasis is primarily on electronically interfaced DG (EI-DG) units. DG controls and power management strategies are based on locally measured signals without communications. Based on the reactive power controls adopted, three power management strategies are identified and investigated. These strategies are based on 1) voltage-droop characteristic, 2) voltage regulation, and 3) load reactive power compensation. The real power of each DG unit is controlled based on a frequency-droop characteristic and a complimentary frequency restoration strategy. A systematic approach to develop a small-signal dynamic model of a multiple-DG microgrid, including real and reactive power management strategies, is also presented. The microgrid eigen structure, based on the developed model, is used to 1) investigate the microgrid dynamic behavior, 2) select control parameters of DG units, and 3) incorporate power management strategies in the DG controllers. The model is also used to investigate sensitivity of the design to changes of parameters and operating point and to optimize performance of the microgrid system. The results are used to discuss applications of the proposed power management strategies under various microgrid operating conditions
1,531 citations
TL;DR: In this paper, the authors investigated preplanned switching events and fault events that lead to islanding of a distribution subsystem and formation of a micro-grid, and they concluded that an appropriate control strategy for the power electronically interfaced DG unit can ensure stability of the microgrid and maintain voltage quality at designated buses, even during islanding transients.
Abstract: This paper investigates (i) preplanned switching events and (ii) fault events that lead to islanding of a distribution subsystem and formation of a micro-grid. The micro-grid includes two distributed generation (DG) units. One unit is a conventional rotating synchronous machine and the other is interfaced through a power electronic converter. The interface converter of the latter unit is equipped with independent real and reactive power control to minimize islanding transients and maintain both angle stability and voltage quality within the micro-grid. The studies are performed based on a digital computer simulation approach using the PSCAD/EMTDC software package. The studies show that an appropriate control strategy for the power electronically interfaced DG unit can ensure stability of the micro-grid and maintain voltage quality at designated buses, even during islanding transients. This paper concludes that presence of an electronically-interfaced DG unit makes the concept of micro-grid a technically viable option for further investigations.
1,136 citations
TL;DR: This paper reviews several optimization methods to improve the accuracy of the training and to reduce training time, and delve into the math behind training algorithms used in recent deep networks.
Abstract: Deep learning (DL) is playing an increasingly important role in our lives. It has already made a huge impact in areas, such as cancer diagnosis, precision medicine, self-driving cars, predictive forecasting, and speech recognition. The painstakingly handcrafted feature extractors used in traditional learning, classification, and pattern recognition systems are not scalable for large-sized data sets. In many cases, depending on the problem complexity, DL can also overcome the limitations of earlier shallow networks that prevented efficient training and abstractions of hierarchical representations of multi-dimensional training data. Deep neural network (DNN) uses multiple (deep) layers of units with highly optimized algorithms and architectures. This paper reviews several optimization methods to improve the accuracy of the training and to reduce training time. We delve into the math behind training algorithms used in recent deep networks. We describe current shortcomings, enhancements, and implementations. The review also covers different types of deep architectures, such as deep convolution networks, deep residual networks, recurrent neural networks, reinforcement learning, variational autoencoders, and others.
907 citations
TL;DR: Deep learning, reinforcement learning and their combination-deep reinforcement learning are representative methods and relatively mature methods in the family of AI 2.0 and their potential for application in smart grids is summarized and an overview of the research work on their application is provided.
Abstract: Smart grids are the developmental trend of power systems and they have attracted much attention all over the world. Due to their complexities, and the uncertainty of the smart grid and high volume of information being collected, artificial intelligence techniques represent some of the enabling technologies for its future development and success. Owing to the decreasing cost of computing power, the profusion of data, and better algorithms, AI has entered into its new developmental stage and AI 2.0 is developing rapidly. Deep learning (DL), reinforcement learning (RL) and their combination-deep reinforcement learning (DRL) are representative methods and relatively mature methods in the family of AI 2.0. This article introduces the concept and status quo of the above three methods, summarizes their potential for application in smart grids, and provides an overview of the research work on their application in smart grids.
322 citations