Elsevier

Applied Soft Computing

Volume 40, March 2016, Pages 58-63
Applied Soft Computing

Multicores and GPU utilization in parallel swarm algorithm for parameter estimation of photovoltaic cell model

https://doi.org/10.1016/j.asoc.2015.10.054Get rights and content

Highlights

  • Better estimation of parameters, on two models.

  • In this work here, we successfully identified the relevant parameters of two photovoltaic models. To prove the efficacy of the proposed method, we included a comparison study.

  • Utilization of multicores and GPU facilities.

  • We have implemented the parallel swarm algorithm utilizing the multicores and GPU computing capabilities of a computer.

Abstract

Bio-inspired metaheuristic algorithms have been widely applied in estimating the extrinsic parameters of a photovoltaic (PV) model. These methods are capable of handling the nonlinearity of objective functions whose derivatives are often not defined as well. However, these algorithms normally utilize multiple agents in the search process, and thus the solution process is extremely time-consuming. In this regard, it takes much time to search the possible solutions in the whole search domain by sequential computing devices. To overcome the limitation of sequential computing devices, parallel swarm algorithm (PSA) is proposed in this work with the aim of extracting and estimating the parameters of the PV cell model by utilizing the power of multicore central processing unit (CPU) and graphical processing unit (GPU). We implement this PSA in the OpenCL platform with the execution on Nvidia multi-core GPUs. Simulation results demonstrate that the proposed method significantly increases the computational speed in comparison to the sequential algorithm, which means that given a time requirement, the accuracy of a solution from the PSA can be improved compared to that from the sequential one by using a larger swarm size.

Introduction

Solar energy is a clean and renewable energy source that has been increasingly popular due to its ease of maintenance and cost-saving in the long term usage. It is widely known that the initial cost of photovoltaic (PV) system is normally high, an accurate investigation and analysis for the electrical characteristic of the applied PV component is of critical importance in the design stage. From previous experiences on working with maximum power point tracking (MPPT) algorithm in [1], [2], we found that an accurate model is crucial to obtain desirable results.

Over the past years, a variety of PV cell models has been developed to predict the electrical characteristics of PV cells. They are normally analytical equations predicting the values of the operating current I and voltage V of the PV cell with a series of model parameters, which are constant quantities that characterize the PV cell model. For instance, the single-diode model [3] is widely applicable in simulation tools such as PSIM [4] and PVsyst [5] due to its simplicity. PV cell model cannot be explicitly applied due to the lack of proper model parameters characterizing the PV cells. The parameter estimation is a discipline that provides tools for the efficient use of data assisting the mathematical modeling and estimation of parameter values existing in the model. In general, the methods of parameter estimation can be classified into two groups [6]:

  • i.

    Analytical method [7], [8]: it is a traditional approach that predicts parameters by mathematical expressions;

  • ii.

    Numerical extraction [9], [10]: it fits the operating points on the IV characteristic curves via a numerical method.

Even though the methods above are easy to develop, they both have inevitable defects. The accuracy of the former method may depend heavily on the key components that produce the IV curve (e.g. the maximum power point (MPP), short-circuit current Isc, and open circuit voltage Voc, etc.), and consequently the aggregated error can be significant if these components’ parameters are incorrectly specified. The numerical parameter extraction method is capable of obtaining accurate solution since all the measured data are utilized to extract these parameters. However, numerical methods are computationally expensive, and their outcome depends heavily on the relevant fitting algorithm, cost function, and initial values of the parameters to be extracted.

More recently, the computational intelligence methods, such as cuckoo search (CS) [11], particle swarm optimization (PSO) [12], genetic algorithm (GA) [13], simulated annealing (SA) [14], bacterial foraging algorithm (BFA) [15], and differential evolution (DE) [16] are applied to determine the values of the model parameters. Albeit accurate and feasible output, these methods applied multiple agents or particles in a random search process, which consumed excessive time in a sequential computing platform. Thus, the approaches were impractical for online applications.

Today's computing environments are becoming more multifaceted, exploiting the capabilities of a range of multi-core microprocessors, central processing units (CPUs), digital signal processors (DSPs), reconfigurable hardware: field programmable gate arrays (FPGAs), and graphic processing units (GPUs) [17]. To distribute the workload of PSO appropriately to the available computing device, this paper proposes a parallel swarm algorithm (PSA) based parameter estimation method for PV cell models. Error and statistical analysis are carried out to illustrate the accuracy and efficiency of the proposed method. Henceforth, we organize the remaining part of this paper as follows. Section 2 briefly introduces the single-diode PV cell model. The development environment for parallel algorithms in general is introduced in Section 3. This section provides useful tips in adopting parallel tools in an optimization work. Further, in Section 4, the PSA – based parameter estimation is described. This is followed by Section 5, which discusses the simulation results, and finally is the conclusion, presented as Section 6.

Section snippets

PV cell model

When a PV cell that is connected to an external circuit, is exposed to incident light, the reverse current is generated across the pn junction, this current is the photocurrent Ipv, shown in Fig. 1. By eliminating the effect of photocurrent, a PV cell behaves like a normal diode. Its IV characteristics can be simply modeled as a linear independent current source in parallel to the diode. The single-diode model improves the simple model by recognizing the series resistance Rs and parallel

Development environment for parallel algorithms

This section aims to provide general guidelines for anyone who would like to convert their sequential works to parallel platform. The parallel platform in this context does not mean linking a cluster of computers in parallel to perform a task. In fact, the context of parallelism here covers the utilization of multi-core and graphic processing units (GPUs) computation capabilities. The motivation to apply the existing work in [22], [23] to parallel platform is based on this simple ideology: the

Parallel swarm algorithm for PV cell model's parameter estimation

In this section, the concept of the particle swarm optimization (PSO) algorithm is introduced, which is then followed by the parallel implementation of the PSO algorithm.

Results and discussions

Table 3 records the value of parameters of the single-diode model extracted by the PSA. By careful analysis of the values of the root-mean-square error frms, it is observed that frms decreases when the swarm size increases. This simply indicates that a good estimation of these parameters requires a large number of particles and iterations. This is in fact a computationally expensive task. Also, from this table, one can observe that some parameter values are highly deviated from their true

Conclusions

In this paper, the parallel swarm algorithm (PSA) has been applied to estimate the parameters of a single-diode model. The fitness evaluation functions are written in OpenCL kernel, and they are executed on the multicore central processing unit (CPU) and graphical processing units (GPUs). The feasibility and efficiency of the proposed method have been validated by parameter estimation results on a commercial PV cell. Simulation results prove the significant speedup and acceleration when

Acknowledgements

The work is supported by National Basic Research Program of China under grant no. 2012CB316301 and National Natural Science Foundation of China (NSFC) under grant no. 61473236

References (40)

  • Powersim Inc., PSIM Users’ Guide...
  • PVsyst SA, User's Guide PVsyst Contextual Help...
  • D.S.H. Chan et al.

    Analytical methods for the extraction of solar-cell single- and double-diode model parameters from IV characteristics

    IEEE Trans. Electron. Dev.

    (1987)
  • J.-P. Charles et al.

    Consistency of the double exponential model with physical mechanisms of conduction for a solar cell under illumination

    J. Phys. D: Appl. Phys.

    (1985)
  • R. Gottschalg et al.

    The influence of the measurement environment on the accuracy of the extraction of the physical parameters of solar cells

    Meas. Sci. Technol.

    (1999)
  • S. Jing Jun et al.

    Photovoltaic model identification using particle swarm optimization with inverse barrier constraint

    IEEE Trans. Power Electron.

    (2012)
  • A.J. Joseph et al.

    Solar cell parameter extraction using genetic algorithms

    Meas. Sci. Technol.

    (2001)
  • W.T. da Costa et al.

    Identification of photovoltaic model parameters by differential evolution

  • B. Gaster et al.

    Heterogeneous Computing With OpenCL: Revised OpenCL

    (2012)
  • J.V. Beck et al.

    Parameter Estimation in Engineering and Science Parameter

    (1977)
  • Cited by (24)

    • A landscape-aware particle swarm optimization for parameter identification of photovoltaic models

      2022, Applied Soft Computing
      Citation Excerpt :

      Heuristic algorithms have been widely used for parameter extraction of solar PV models because they exhibit some fundamental advantages over traditional mathematical planning methods, such as resilience to dynamic changes, self-organizing capabilities, no need to meet specific mathematical characteristics, and the ability to evaluate multiple solutions in parallel [3,4]. Up to now, many heuristic algorithms have been developed, such as particle swarm optimization (PSO) [5], backtracking search algorithm (BSA) [6], differential evolution (DE) [7], JAYA algorithm [8], parallel swarm algorithm (PSA) [9], cuckoo search optimization (CSO) [10], QUATRE algorithm [11], and so on. Although the above heuristic algorithms can extract parameters for PV models, it is still hard to find more stable and accurate global solutions [12].

    • A least square support vector machine approach based on bvRNA-GA for modeling photovoltaic systems

      2022, Applied Soft Computing
      Citation Excerpt :

      The mathematical models use the physical processes and the related variables to represent the characteristics of PV systems, and researchers have applied many different algorithms to extract the unknown parameters in the mathematical models. For example, the parallel swarm algorithm (PSA) and the ant lion optimizer are respectively used in [5] and [6] to estimate the parameters of a simple PV system which is called the single diode model (SDM). Then, the double diode model (DDM) is studied, and some optimization algorithms are applied to detect the parameters for both SDM and DDM [4,7–9].

    • A GPU-accel erated parallel Jaya algorithm for efficiently estimating Li-ion battery model parameters

      2018, Applied Soft Computing Journal
      Citation Excerpt :

      In addition, comparing the Jaya algorithm with classical CI algorithms on benchmark problems has been studied and better results produced by the Jaya algorithm have been reported [25]. Moreover, recent applications of the general-purpose parallel computing device, the utilization of the Graphic Processing Unit (GPU), has enabled the faster execution of data-intensive algorithms [30–32]. Due to the fast float-point computing capability and parallel computing mechanism, great advantages of the GPU in the scientific computing have been demonstrated [33].

    • Metaheuristic algorithms for PV parameter identification: A comprehensive review with an application to threshold setting for fault detection in PV systems

      2018, Renewable and Sustainable Energy Reviews
      Citation Excerpt :

      The evolution of metaheuristic algorithms started with Genetic Algorithm (GA) followed by Differential Evolution (DE) and Particle Swarm Optimization (PSO). Inspired by these basic algorithms, several new and hybrid metaheuristic algorithms were developed in recent years [45–81]. Some prominent objective functions utilized by various metaheuristic algorithms for PV parameter optimization are: 1) Root Mean Squared Error (RMSE) [48,51–57,60,62–71,73,74], 2) Mean Squared Error (MSE) [45,59,72], 3) Absolute Error (A.E) [46,58,75,76] and 4) Derivative at maximum power point (MPP) [60,67].

    • A parallel double-level multiobjective evolutionary algorithm for robust optimization

      2017, Applied Soft Computing Journal
      Citation Excerpt :

      In [29], the authors presented a multithreading PSO computation model. Ting et al. [30] implemented a parallel GPU-PSO for parameter estimation of photovoltaic cell model. In [31] and [32], parallel multiobjective optimization particle swarm optimization (MOPSO) algorithms based on GPUs were proposed.

    View all citing articles on Scopus
    View full text