Source-linked AI summary
A radial basis function neural network based approach for the electrical characteristics estimation of a photovoltaic module
Francesco Bonanno, Giacomo Capizzi, Christian Napoli, Giorgio Graditi, Giuseppe Marco Tina
TL;DR
Accurate PV-module models are needed to predict electrical output across changing operating conditions. This paper trains an RBFNN using experimental data and solar irradiance and temperature inputs to estimate I–V and P–V characteristics, achieving close agreement with experimental data, including critical curve regions.
Problem
Accurate PV-module models are needed to predict I–V and P–V output behavior while accounting for parameter changes across operating conditions.
Method
The paper trains an RBFNN with experimental data, using solar irradiance and temperature to model PV-module electrical behavior and estimate I–V and P–V characteristics.
Results
The computed I–V and P–V characteristics match experimental data closely, with improved evaluation in the critical final portions of the curves.
Takeaways & Limitations
RBFNNs provide a low-complexity approach for modeling nonlinear solar-cell characteristics and estimating circuital parameters.
Abstract
from arXiv · showhide
The design process of photovoltaic (PV) modules can be greatly enhanced by using advanced and accurate models in order to predict accurately their electrical output behavior. The main aim of this paper is to investigate the application of an advanced neural network based model of a module to improve the accuracy of the predicted output I--V and P--V curves and to keep in account the change of all the parameters at different operating conditions. Radial basis function neural networks (RBFNN) are here utilized to predict the output characteristic of a commercial PV module, by reading only the data of solar irradiation and temperature. A lot of available experimental data were used for the training of the RBFNN, and a backpropagation algorithm was employed. Simulation and experimental validation is reported.
UPLOADED UNDER SELF-ARCHIVING POLICIES
The paper presents an RBFNN-based approach for estimating photovoltaic-module electrical characteristics.
- The paper is titled “A radial basis function neural network based approach for the electrical characteristics estimation of a photovoltaic module.”
- The authors are affiliated with the University of Catania and ENEA in Italy.
- The paper concerns solar energy, photovoltaic modules, circuital models, radial basis functions, and neural networks.
I. INTRODUCTION
The introduction motivates dynamic photovoltaic-module modeling and presents an RBFNN approach that uses solar irradiance and module temperature to predict electrical behavior across operating conditions.
- Dynamic modeling and characterization are motivated by advances in commercial PV-cell manufacturing and diverse application requirements.
- The proposed RBFNN uses only solar irradiance and PV-module temperature as input data.
- The approach is intended to generalize across different operating conditions while using limited manufacturer-provided data.
- The method relates PV-module I–V characteristics to selected RBFNN centroids and permits calculation of several circuit-model parameters.
II. CIRCUITAL MODELS FOR PV MODULES
This section describes equivalent-circuit models for photovoltaic modules, including the five-parameter model and an additional recombination-current sink for aging amorphous-silicon cells.
- The common five-parameter PV-module model derives from an equivalent circuit containing a current source, diode, and series and parallel resistors.
- The current source represents photocarrier generation, while the diode and resistors represent recombination, leakage, and series-resistance effects.
- The model introduces nonlinear series-resistance temperature dependence δ and diode reverse-saturation-current radiation dependence m.
- For amorphous silicon, an additional current sink represents recombination losses associated with degradation in the intrinsic layer.
RESISTOR AND THE FIVE PARAMETERS MODEL
The section reviews double-diode and practical equivalent-circuit models for representing photovoltaic current–voltage behavior under varying illumination and parasitic effects.
- The double-diode model represents photovoltaic current-density–voltage characteristics more accurately when incident light intensity varies.
- The two diode terms use reverse saturation currents I01 and I02 with quality factors η1 and η2, typically 1 and 2, respectively.
- The photogenerated current Iph depends on irradiation, and the model can represent crystalline-silicon and thin-film photovoltaic modules.
- Series resistance models voltage drop, while parallel leakage resistance represents parasitic current paths in practical cells.
IV. PV MODULE MODELING USING RBFNN
The proposed PV-array RBFNN maps solar radiation, ambient temperature, and load voltage to load current through radial basis neurons. Its Gaussian radial basis functions use centers and a spread width to represent the input mapping.
- Network structure: The RBFNN has input, hidden, and output layers, with a three-dimensional input vector and a single load-current output.The inputs are solar radiation, ambient temperature, and load voltage; the hidden layer contains L radial basis neurons.
- Network structure: Each hidden radial basis neuron connects directly to all input elements and contributes to the PV-array mapping.The proposed network follows the general principle of RBF network mapping.
- Radial basis functions: The Gaussian basis function is centered at µ_j and controlled by scalar σ, which defines the RBF spread width.The center vector has the same dimension as the input vector.
- Radial basis functions: RBF networks may require more neurons than standard feedforward networks but can often be designed and trained faster when many training vectors are available.The passage describes this as a general property of radial basis networks.
V. THE PROPOSED RBFNN BASED MODEL
The proposed model trains an RBFNN by determining its radial basis functions, centers, weights, and biases while minimizing squared error. The implemented network uses 16 hidden radial basis neurons and one linear output neuron trained on PV current data.
- Training procedure: RBFNN training determines the number of radial basis functions and the optimal centers, weights, and biases by minimizing squared error.The network is built iteratively until the mean squared error reaches the expected limits.
- Implemented model: The implemented model is a 16-by-one RBFNN with 16 hidden radial basis neurons and one output linear neuron.The architecture is reported alongside the network weights and centroids in Table I.
- Training data: Training used a random set of 5600 bidimensional [G, V] inputs with related target outputs [I(V, G)].The inputs represent solar irradiation G and voltage V.
VI. SIMULATION RESULTS AND PV MODEL PARAMETERS ESTIMATION
The model converts the RBFNN representation into a PV-module current formulation and evaluates network performance for I–V and P–V characteristics. The spread width must balance smooth approximation against numerical stability.
- PV model formulation: The generated PV-module current is expressed in polynomial form and evaluated through a 16-term summation after expanding the exponential in a power series.The procedure uses numerical approximation and nonlinear circuit theory to model the PV-module components.
- Simulation results: The reported figures present network performance and simulation results for the I–V and P–V characteristics.Figures 3 and 4 include I–V-related performance and RBFNN-based prediction results.
- Simulation results: A larger RBFNN spread width produces a smoother function approximation, but an excessively large width causes numerical problems.The training algorithm seeks a trade-off between smoothness and numerical stability, with typical oscillations resulting in RBF approximations.
VII. CONCLUSION AND DISCUSSION
The paper proposes an RBFNN-based model for predicting photovoltaic module I–V and P–V characteristics. Simulations indicate close agreement with experimental data and improved evaluation of the final portions of the curves.
- The proposed RBFNN models predict photovoltaic module I–V and P–V characteristics.
- At high solar irradiance, the estimated I–V characteristics match the experimental characteristics with superior precision.
- The reported simulations include P–V characteristic performance and prediction results.
- The RBFNN model improves estimation along the final I–V and P–V curve portions, where five- and seven-parameter circuit models can fail.
- The computed I–V and P–V numerical values match closely with experimental data.