Elsevier

Renewable Energy

Volume 117, March 2018, Pages 257-274
Renewable Energy

Comparing Mamdani Sugeno fuzzy logic and RBF ANN network for PV fault detection

https://doi.org/10.1016/j.renene.2017.10.066Get rights and content

Highlights

  • PV fault detection algorithm based on the analysis of the voltage and the power is presented.

  • Two machine learning techniques were developed and compared briefly.

  • Four different Artificial neural networks (ANN) are used for detecting PV faults.

  • Two fuzzy logic systems (Mamdani & Sugeno) are used for examining faults in PV systems.

Abstract

This work proposes a new fault detection algorithm for photovoltaic (PV) systems based on artificial neural networks (ANN) and fuzzy logic system interface. There are few instances of machine learning techniques deployed in fault detection algorithms in PV systems, therefore, the main focus of this paper is to create a system capable to detect possible faults in PV systems using radial basis function (RBF) ANN network and both Mamdani, Sugeno fuzzy logic systems interface.

The obtained results indicate that the fault detection algorithm can detect and locate accurately different types of faults such as, faulty PV module, two faulty PV modules and partial shading conditions affecting the PV system. In order to achieve high rate of detection accuracy, four various ANN networks have been tested. The maximum detection accuracy is equal to 92.1%. Furthermore, both examined fuzzy logic systems show approximately the same output during the experiments. However, there are slightly difference in developing each type of the fuzzy systems such as the output membership functions and the rules applied for detecting the type of the fault occurring in the PV plant.

Introduction

The monitoring and regular performance supervision on the functioning of grid-connected photovoltaic (GCPV) systems is necessary to ensure an optimal energy harvesting and reliable power production. The development of diagnostic methods for fault detection in the PV systems behavior is particularly important due to the expansion degree of GCPV systems nowadays and the need to optimize their reliability and performance.

There are existing techniques which were developed for possible fault detection in grid-connected PV systems. Some of these techniques use meteorological and satellite data for predicting the faults in the GCPV plants [1], [2]. However, some of the PV fault detecting algorithms do not require any climate data (solar irradiance and module temperature) such as the earth capacitance measurements established by Taka-Shima [3].

Other PV fault detection algorithms is based on the comparison of simulated and measured yield by analysing the losses of the DC side of the GCPV plant [4], [5], [6]. Furthermore, a fault detection method based on the ratio of DC side and the AC side of the PV system is proposed by W. Chine et al. [7]. The method can detect five different faults such as faulty modules in a PV string, faulty DC/AC inverter and faulty maximum power point tracking (MPPT) units. On the other hand, S. Silvestre et al. [8] proposed a new procedure for fault detection in GCPV systems based on the evaluation of the current and the voltage indicators. The main advantage of this algorithm is to reduce the number of monitoring sensors in the PV plants and integrating a fault detection algorithm into an inverter without using simulation software or additional external hardware devices.

Further fault detection algorithms focus on faults occurring in the AC-side of GCPV systems, as proposed by M. Dhimish et al. [9]. The approach uses mathematical analysis technique for identifying faulty conditions in the DC/AC inverter units. Moreover, hot-spot detection in PV substrings using the AC parameters characterization was developed by Ref. [10]. The hot-spot detection method can be further used and integrated with DC/DC power converters that operates at the subpanel level. A comprehensive review of the faults, trends and challenges of the grid-connected PV systems is shown in Refs. [11], [12], [13].

Other PV fault detection approaches use statistical analysis techniques for identifying micro cracks and their impact of the PV output power as presented by Ref. [14]. However, T. Zhao at al [15] developed a decision tree (DT) technique for examining two different types of fault using an over-current protection device (OVPD). The first type of fault is the line-to-line that occurs under low irradiance conditions, and the second is line-to-line faults occurring in PV arrays equipped with blocking diodes.

PV systems reliability improvement by real-time field programmable gate array (FPGA) based on switch failures diagnosis and fault tolerant DC-DC converters is presented by Ref. [16]. B. Chong [17] suggested a controller design for integrated PV converter modules under partial shading conditions. The developed approach is based on a novel model-based, two-loop control scheme for a particular MIPC system, where bidirectional Cuk DC-DC converters are used as the bypass converters and a terminal Cuk boost functioning as a while system power conditioner.

Nowadays, fuzzy logic systems widely used with GCPV plants. R. Boukenoui et al. [18] proposed a new intelligent MPPT method for standalone PV system operating under fast transient variations based on fuzzy logic controller (FLC) with scanning and storing algorithm. Furthermore [19], presents an adaptive FLC design technique for PV inverters using differential search algorithm. Furthermore, N. Sa-ngawong & I. Ngamroo [20] proposed an intelligent PV farms for robust frequency stabilization in multi-area interconnected power systems using Sugeno fuzzy logic control, similar approach was developed by Ref. [21] for power optimization in standalone PV systems.

In Refs. [22], [23] authors have used a Mamdani fuzzy logic classification system which consists of two inputs, the voltage and power ratio, and one output membership function. The results can accurately detect several faults in the PV system such as partial shading and short circuited PV modules.

Artificial intelligent networks (ANN) is another machine leaning technique nowadays is used for detecting faults in PV systems. A learning method based on expert systems is developed by Ref. [24] to identify two types of fault (due to the shading effect and to the inverter's failure). Whereas [25] proposed an ANN network that detects faults in the DC side of PV systems which includes faulty bypass diodes and faulty PV modules in a PV string.

A. Millit et al. [26] shows that ANN networks is a possible solution for modelling and estimating the output power of a GCPV systems. However, a failure mode prediction and energy harvesting of PV plants to assist dynamic maintenance tasks using ANN based models is proposed by F. Polo et al. [27]. Further investigation on a very short term load forecasting for a distribution system with high PV penetration is suggested by S. Sepasi [28]. Finally, B. Amrouche & X. Pivert [30] offered an ANN network based daily local forecasting for global solar radiation (GHI). The ANN model is developed to predict the local GHI based on a daily weather forecast provided by the US National Oceanic and Atmospheric Administration (NOAA) for four neighbouring locations.

The main contribution of this work is to present a new algorithm for isolation and identification of the faults accruing in a PV system. The algorithm is capable to detect several faults such as faulty PV module in a PV string, faulty PV string, faulty MPPT, and partial shading conditions effects the PV system. The proposed algorithm is comparing between two different approaches for detecting failure conditions which can be described as the following:

  • 1.

    Artificial Neural Network (ANN) Approach:

Four different ANN networks have been compared using a logged data of several faulty conditions affecting the examined PV plant. The maximum PV fault detection accuracy achieved by the ANN networks is equal to 92.1%.

  • 2.

    Fuzzy Logic Fault Classification Approach:

This approach consists of two types of fuzzy logic interface systems: Mamdani and Sugeno. Both fuzzy interface systems were briefly compared and developed using MATLAB/Simulink software. This approach was tested using a faulty PV data which was logged from the examined 1.1 kWp PV plant installed at the University of Huddersfield.

The overall system design is shown in Fig. 1. The PV plant has a capacity of 1.1 kWp. A computer interface has two options, a PV fault detection algorithms which use MATLAB/Simulink software which contains the ANN and the fuzzy logic interface system. Furthermore, LabVIEW software is used for the real-time long-term data monitoring as well as, data logging software environment.

This paper is organized as follows: Section 2 presents the data acquisition in the PV plant. Section 3 describes the methodology used, Fault detection algorithm and diagnosis rules are presented, while section 4 lists the results and discussion of the work. Finally, section 5 describes the conclusion and future work.

Section snippets

Faults in photovoltaic plants

The faults occurring in a PV system are mainly related to the PV array, MPPT units, DC/AC inverters, the storage system and the electrical grid. This work aims to detecting the faults occurring in the PV array and, with reference to Table 1, eleven different fault are investigated.

It is worthy to mention that PS conditions used in this work corresponds to an irradiance level affects all examined PV modules. Thus, during the experiments, all examined PV modules were tested under the same PS

Methodology

This section reports the PV data acquisition system, PV theoretical modelling, the overall fault detection algorithm, and the detailed design of the proposed artificial neural network and the fuzzy logic interface system.

Results and discussion

This section reports the results of the developed fault detection algorithm. Furthermore, a comparison between the developed machine learning techniques with some ANN and fuzzy logic systems obtained by various researchers is briefly explained in Section 4.4 (discussion section).

Conclusion

This paper presents a new photovoltaic (PV) fault detection algorithm which comprises both artificial neural network (ANN) and fuzzy logic system interface. The algorithm is capable for detecting various fault occurring in the PV system such as faulty PV module, two faulty PV modules and partial shading conditions affecting the PV system. Both machine learning techniques was validated using a 1.1 kWpp PV plant installed at the University of Huddersfield, United Kingdom.

The fault detection

References (37)

Cited by (0)

View full text