Progresses in the development of an integrated forecasting model of solar radiation and photovoltaic power output without using onsite measurements

Abstract

Renewable resources, and consequently the generated energy, are especially variable, which makes finding an accurate balance between electricity generation and consumption at any moment challenging in the absence of reliable large capacity energy storage systems. Thus, having an accurate forecasting of the generated energy allows for a more efficient management of the electric grid comprising various energy sources. This work presents the study of all fundamental models necessary for the forecasting of photovoltaic power output when there is no measuring instrumentation on site, namely: weather forecasting model, direct normal irradiance forecast improvement model, transposition model, photovoltaic module temperature and power output model and inverter model. The weather forecasting model used in this work is the numerical weather prediction model of the European Centre for Medium-range Weather Forecasts which produces forecasts twice a day with temporal resolution of 1 hour and 0.125° of horizontal resolution in a global grid. Methods for temporal and spatial downscaling are applied to obtain 10-minute values of the forecasted variables for the desired location. The forecasts of direct normal irradiance (DNI) show higher errors (157.16 W/m2 of mean absolute error - MAE - for forecast day 1) than global horizontal irradiance (GHI, MAE of 63.63 W/m2) and thus a corrective algorithm based on artificial neural networks (ANN) was developed to improve these forecasts achieving an MAE of 130.94 W/m2 for forecast day 1. The transposition model converts DNI and diffuse horizontal irradiance (DIF) into irradiance on the tilted plane (GTI). This is done by using transposition coefficients on the direct, diffuse and reflected component of solar irradiance. In this work some of the most employed analytic models for the determination of the diffuse transposition coefficient are compared, being the modified Bugler model selected. In the case of photovoltaic power plants, which are composed of various rows of panels, there is sometimes obscuring of the sun by the front rows over the second and subsequent rows affecting the beam radiation received by these. There is also obscuring of the sky dome affecting the diffuse radiation and obscuring of the reflected radiation from the ground between rows. In this work a transposition model for rows other than the first based on the works of Varga and Mayer (2021) and Tschopp et al. (2022) was developed and is now being evaluated. The photovoltaic power output is very dependent not only on the irradiance on the solar panels but also on their temperature. Thus, a model to determine the temperature of the panel is essential when there are no measurements available. Most models used in the literature are steady-state and empirical which means they can be biased towards a specific technology or location. Besides comparing the most commonly used empirical models, a physical transient model for the determination of the photovoltaic panel temperature was developed. The integration of these models with various non-empirical power output models was evaluated. Finally, the efficiency of the power inverter is considered to obtain the power output supplied to the electric grid.