Islanding detection of distributed generation by using multi-gene genetic programming based classifier

Highlights

• Multi-gene genetic programming (MGP) is used to classify islanding of DG.

• As far as we know, MGP has not been used yet for islanding detection.

• Proposed method has superior performance than an artificial neural ntowrk method.

• The proposed method has high potential for DG islanding detection purposes.

• Proposed method has better performance than traditional islanding detection methods.

Abstract

This paper proposed a new method for detecting islanding of distributed generation (DG), using Multi-gene Genetic Programming (MGP). Islanding has been a serious concern among power distribution utilities and distributed generation owners, because it poses risks to the safety of utilities’ workers and consumers, and can cause damage to power distribution systems’ equipment. Therefore, a DG must be disconnected as soon as an islanding is detected. In addition, an islanding detection method must have high degree of dependability to correctly discriminate islanding from other events, such as load switching, in order to avoid unnecessary disconnection of the distributed generator. In this context, the novelty of the proposed method is that the MGP is capable of obtaining a set of mathematical and logic functions employed to detect and classify islanding correctly. This is a new approach among the computational intelligent methods proposed for DG islanding detection. The main idea was to use local voltage measurements as input of the method, eliminating the need of complex and expensive communication infrastructure. The method has been trained with several islanding and non-islanding cases, by using a power distribution system comprising five concentrated loads, a synchronous distributed generator and a wind power plant. The results showed that the proposed method was successful in differentiating the islanding events from other disturbances, revealing its great potential to be applied in anti-islanding protection schemes for distributed generation.

Introduction

Distributed generation (DG) has been increasing worldwide due to the advantages it can provide to the electrical power systems, such as the possibility of reducing transmission and distribution losses, environmental benefits, the increase in the reliability of the power supply and the deferral of transmission and distribution investments [1]. However, the connection of DG in power distribution systems must be carefully evaluated to minimize or eliminate possible negative impacts on the power quality, protection and operation procedures of the grid [2]. In this context, detecting unintentional islanding is a major concern of power distribution utilities and DG owners.

Unintentional islanding of DG is defined as a condition in which part of a power distribution system becomes electrically isolated from the main supply source (distribution substation) but remains energized by one or more DG. This is a serious concern of distribution utilities and DG owners because of the hazards that an islanded operation poses to the utilities’ workers and consumers; the power quality parameters (voltage and frequency) may not be within acceptable limits; the power system equipment can be damaged due to inadequate protection and poor power quality in the isolated system [1], [3]. A recent survey conducted by the Electric Power Research Institute (EPRI) with 30 entities (cooperatives, investor-owned utilities and municipal utilities) revealed that islanding is the major protection concern regarding the interconnection of DG, as pointed out by 50% of the respondents [2]. Therefore, grid codes recommend that the islanding condition be detected and the DG disconnected as fast as possible. A maximum disconnection time is 2 s, as recommended by [4].

In this context, effective methods for detecting unintentional islanding of DG are necessary. To be effective, the anti-islanding protection must be fast enough to disconnect the DG within the required time interval and, simultaneously, be reliable to correctly distinguish the transients caused by islanding from the ones caused by other events, such as load switching, short-circuits, connection and disconnection of other DG. To reach this goal, several anti-islanding protection schemes have been proposed in the literature, but a general cost-effective solution has not been proposed so far.

Basically, there are remote and local islanding detection methods. Remote methods are efficient but expensive because they are based on the existence of a communication infrastructure among the devices (protection relays, circuit breakers, DG and meters installed along the distribution feeder). On the other hand, local methods are based on measurements acquired at the point where the DG is connected (point of common coupling—PCC), eliminating the need of communication and reducing costs in comparison with the communication-based methods. Local methods can be divided into active, passive and hybrid ones. In general, active methods are based on the DG response after the injection of small disturbances on the DG current or voltage (magnitude and/or frequency): if the DG remains stable after the disturbance inception, there is no islanding. This type of anti-islanding method is efficient, however it may undermine the power quality of the distribution grid and they can be expensive. Passive methods are widely employed, because they do not disturb the power quality and they are cheaper in comparison with the remote and active ones. Nonetheless, they may not detect islanding timely for small variations of DG voltage or frequency, which are caused by a small difference between loads and DG power output after an islanding occurrence. Examples of traditional passive methods are the under/overfrequency, rate of change of frequency (ROCOF), vector surge and voltage protection devices. Finally, anti-islanding hybrid methods combine the advantages of active and passive ones [3], [5], [6], [7].

Remote and active anti-islanding methods are efficient even in the case of small voltage and frequency variation after an islanding takes place. The efficiency of the traditional passive methods can be improved if they are adjusted too sensitive, but they can operate for normal events, such as load and capacitor switching, disconnecting the DG unnecessarily [7]. Thus, they are not reliable to distinguish islanding from other events. To overcome this drawback, computational intelligence has emerged as a potential approach to develop intelligent islanding detection methods [3]. Some relevant researches are discussed below.

An intelligent relay that employs multivariate analysis and data mining was proposed in [8] for detecting islanding of synchronous-based DG. The relay also relies on the selection of the optimal decision tree derived from the data extracted from a training data set. The results showed that the relay was dependable and effective for detecting islanding in a small distribution system, but the used approach is a little bit complex because the optimal decision tree changes if the system operating conditions change. The authors in [9] employed a similar method as in [8], but they considered synchronous and inverter-based DG. The paper can be faced as an improvement of [8], but the complexity of extracting data and selecting the optimal decision tree is still present. With an analogous idea, the authors in [10] proposed a method that used data mining complex correlations to build a classifier for islanding detection purposes. The method was capable of identifying islanding and non-islanding events correctly, and it needs a reduced data-set for training. Despite of the overall good performance, the method needs various features extracted from the voltage waveform to serve as input (ten features were used in the paper). This can be an issue and impact the costs and the performance speed of the method, since additional sensors may be necessary and more processing capability is needed to process the measured data.

In [7], five features were extracted from the voltage measured at the DG location, their standard deviations were calculated and processed by a Support Vector Machine-based classifier (SVM) to differentiate islanding from non-islanding occurrences. The method presented high successful rates in the classification, but like in [10], its cost and performance speed can be a drawback.

To reduce the influence of the number of extracted features, an ANN combined with Particle Swarm Optimization (PSO) was proposed in [11] for DG islanding detection. In the paper, PSO was employed to find optimal values for the following ANN parameters: learning rate, momentum and number of neurons in the hidden layer. The method used only three signals to perform the islanding detection. According to the results, the successful rates were high, but PSO brings some complexity to the method.

A fuzzy decision tree was employed in [12] to detect islanding under a multi-DG scenario, showing comparable performance and complexity as in [8], [9]. Therefore, no relevant gain was observed in comparison with the other approaches.

Notice that the solutions proposed in [7], [8], [9], [10], [11], [12] presented different complexity degree, need different amount of input data and all of them presented high successful rates. However, they do not offer an intuitive or easy-to-understand solution, i.e., the user or protection engineer will not be able to interpret an output, especially if the method provides an unexpected outcome (e.g. a failure to detect islanding, for example).

In this context, this paper investigates the potentiality of employing Multi-gene Genetic Programming (MGP) to build an intelligent classifier for islanding detection of distributed generation. This approach results in an islanding detection technique much simpler than the ones based on decision trees (DT), because there is no need for selecting the optimal DT, which creates complexity. The MGP results in a general equation that separates islanding from non-islanding events. It relies on a representative data training set for achieving good effectiveness. Since the solution is an equation, it provides a more intuitive understanding of the results if compared with the above-mentioned intelligent methods. In addition, the proposed method does not need multiple signals to be used as input; only the voltage waveform is necessary, which was sampled at 64 samples/cycle. It is important to mention that, different from the traditional genetic programming, in the multi-gene genetic programming a solution is described by a linear combination of syntactic trees, producing more powerful and high quality solutions due to its capability of exploring other regions in the solutions space that are not reached by the traditional approach. The results confirmed that the MGP has a great potential to be used as a local-passive, dependable and effective DG islanding detection tool.

This paper is organized as follows: Section 2 presents a review about Genetic Programming and Multi-gene Genetic Programming. In Section 3, the fundamentals and characteristics of the intelligent classifier are presented. Section 4 presents the results and discussion, and Section 5, the conclusions of this research.