Moving-Average Transient Model for Predicting the Back-surface Temperature of Photovoltaic Modules
Document
Description
The operating temperature of photovoltaic (PV) modules has a strong impact on the expected performance of said modules in photovoltaic arrays. As the install capacity of PV arrays grows throughout the world, improved accuracy in modeling of the expected module temperature, particularly at finer time scales, requires improvements in the existing photovoltaic temperature models. This thesis work details the investigation, motivation, development, validation, and implementation of a transient photovoltaic module temperature model based on a weighted moving-average of steady-state temperature predictions.
This thesis work first details the literature review of steady-state and transient models that are commonly used by PV investigators in performance modeling. Attempts to develop models capable of accounting for the inherent transient thermal behavior of PV modules are shown to improve on the accuracy of the steady-state models while also significantly increasing the computational complexity and the number of input parameters needed to perform the model calculations.
The transient thermal model development presented in this thesis begins with an investigation of module thermal behavior performed through finite-element analysis (FEA) in a computer-aided design (CAD) software package. This FEA was used to discover trends in transient thermal behavior for a representative PV module in a timely manner. The FEA simulations were based on heat transfer principles and were validated against steady-state temperature model predictions. The dynamic thermal behavior of PV modules was determined to be exponential, with the shape of the exponential being dependent on the wind speed and mass per unit area of the module.
The results and subsequent discussion provided in this thesis link the thermal behavior observed in the FEA simulations to existing steady-state temperature models in order to create an exponential weighting function. This function can perform a weighted average of steady-state temperature predictions within 20 minutes of the time in question to generate a module temperature prediction that accounts for the inherent thermal mass of the module while requiring only simple input parameters. Validation of the modeling method presented here shows performance modeling accuracy improvement of 0.58%, or 1.45°C, over performance models relying on steady-state models at narrow data intervals.
This thesis work first details the literature review of steady-state and transient models that are commonly used by PV investigators in performance modeling. Attempts to develop models capable of accounting for the inherent transient thermal behavior of PV modules are shown to improve on the accuracy of the steady-state models while also significantly increasing the computational complexity and the number of input parameters needed to perform the model calculations.
The transient thermal model development presented in this thesis begins with an investigation of module thermal behavior performed through finite-element analysis (FEA) in a computer-aided design (CAD) software package. This FEA was used to discover trends in transient thermal behavior for a representative PV module in a timely manner. The FEA simulations were based on heat transfer principles and were validated against steady-state temperature model predictions. The dynamic thermal behavior of PV modules was determined to be exponential, with the shape of the exponential being dependent on the wind speed and mass per unit area of the module.
The results and subsequent discussion provided in this thesis link the thermal behavior observed in the FEA simulations to existing steady-state temperature models in order to create an exponential weighting function. This function can perform a weighted average of steady-state temperature predictions within 20 minutes of the time in question to generate a module temperature prediction that accounts for the inherent thermal mass of the module while requiring only simple input parameters. Validation of the modeling method presented here shows performance modeling accuracy improvement of 0.58%, or 1.45°C, over performance models relying on steady-state models at narrow data intervals.