Article Preview
Top2. Motivation
One of the most important tasks performed by artificial neural networks (ANNs) is prediction. Prediction - one of the most popular, but it is one of the most difficult tasks of data mining. The greatest interest in terms of planning and development of solar energy is the prediction of solar radiation, the price of solar CSP systems, the efficiency of solar modules and collectors, the electrical characteristics of Si-crystalline modules.
The most promising areas of application of CI technologies and algorithms in solar energy are:
- 1.
ANN: Prediction of solar radiation, energy consumption of a solar building, characteristics of silicon photovoltaic modules, development of heating controllers for solar buildings, simulation of solar air heaters, photovoltaic systems, development of intelligent tracking systems for the sun.
- 2.
Fuzzy logic: forecasting solar radiation, developing solar building controllers, solar air conditioning controllers.
- 3.
Genetic algorithms: development of intelligent control methods for tracking the maximum power point in order to improve the efficiency of photovoltaic systems in various conditions of temperature and illumination, determining the coefficients of the Angstrom equation, modeling the performance of solar water heating systems, the cost of hybrid solar-wind systems, CSP-diesel hybrid power systems, development of a methodology for optimal sizing of stand-alone photovoltaic / wind turbine systems, flat solar air heaters.
- 4.
Data Mining: Increasing the productivity of solar cells.
Top3. Background
In the article (Benghanem et al., 2009) developed ANN models to estimate and model daily cumulative insolation. In the article (Benghanem et al., 2009) developed six ANN models using various combinations as inputs: air temperature, sunshine duration, relative humidity, day of the year. For each model, the output is the daily total insolation. For each of the developed ANN models, the high correlation coefficient achieved. In work (Chaabene & Ammar, 2008) used a neuro-fuzzy dynamic model to predict insolation and ambient temperature. The medium-term forecast, which allows one to obtain daily meteorological behavior, consists of a neuro-fuzzy assessment based on meteorological data on the behavior of parameters in previous days and on patterns of distribution over time. Short-term forecasting estimates a 5-minute time step forward, the meteorological evolution of parameters. According to the calculation of the normalized root mean square error and normalized bias mean error, the developed model performs a satisfactory estimate of meteorological parameters.