Solar energy is one of the most important energy sources with increasing penetration into the power supply in the form of heat or photovoltaic systems, providing reduced environmental impacts. The solar energy systems are preferred to their source of energy because the sun is a renewable source; although, their penetration is reduced due to various factors. One of the most important factors is the predictability of solar radiation, which is a complicated task. The weather variation is a continuous, data-intensive, multidimensional, dynamic and chaotic process, so that these properties make weather forecasting a complicated procedure. Additionally, the solar radiation depends on the clouds and the meteorological conditions are non-linear processes, which affect the efficiency of photovoltaic systems.
The empirical and the dynamic approaches are used in weather forecasting and were introduced by Lorenz in 1969 . Specifically, the cloudiness prediction is based on image processing analysis, where the use of threshold for pixel intensity, determines the clear as well as the cloudy sky . The spectral measurements of solar irradiance are another technique for cloudiness estimation. Furthermore, numerical methods have been developed in order to face the weather prediction problem [3-5]. The most weather prediction systems use a combination of empirical and dynamic techniques .
The accurate weather prediction is crucial for the planning of energy production. Specifically, the solar radiation affects the performance of photovoltaic systems. The latitude, the season of the year, the altitude, the reflection, the weather conditions, such as cloud amount, air humidity, air pollution, and the path of radiation in the atmosphere are main parameters for the estimation of solar radiation. The cloudiness is one of the most important factors for the predictability of solar irradiation.
Soft Computing is a set of methodologies implementing computational intelligence systems for solving non-linear real world problems, without prior knowledge and symbolic representation of their rules . Artificial Neural Networks, Fuzzy Logic, Evolutionary Programming, Genetic Algorithms, Mimetic Algorithms and Artificial Immune Systems are subfields of soft computing.
Artificial Neural Networks (ANNs) are powerful alternative – data modeling – tools that are able to correlate complex input and output data. In the recent past, the ANNs have been applied to model large data with large dimensionality [8-10].
The ANN approach has several advantages over conventional phenomenological or semi-empirical models, since it requires known input data set without any assumptions [10, 11]. It exhibits rapid information processing and is able to develop a mapping of the input and output variables. Such a mapping can subsequently be used to predict desired outputs as a function of suitable inputs .
The ANNs are used in ample scope applications, such as weather prediction [12-14], heat transfer prediction , short term load forecasting [16-19], numerical simulation of nonlinear equations [20, 21], financial prediction [22, 23], telecommunications, medicine [24, 25], signal processing and other fields [26-28].
In an ANN design stage, the implementation of the appropriate ANN architecture to solve a real-world problem is relatively complex. A neural network with few neurons implies inadequate lore, while a big one leads to poor generaliz- ation ability, presenting overfitting . Usually, trial and error strategy is the most common way for specification of an ANN architecture .
This study focuses on the development and assessment of ANN architectures based on Multi-Layer Perceptrons (MLPs) and the application of these implementations to the problem of cloudiness forecasting. This problem is important as the clouds influence the photovoltaic cells’ efficiency and therefore additional generating reserves are required to cover the energy requirements. The specified ANN and the obtained results are presented in this paper.
For full text: click here
(Author: Vassiliki H. Mantzari, Dimitrios H. Mantzaris