Dalgacık analizi kullanılarak optik fotoğraflardan bulutluluk oranı tayini

Abstract

Günümüzde fotovoltaik güç sistemlerininin verimlilikleri her ne kadar büyük gelişmeler kaydedilse de hala düşük düzeydedir. Bu sistemlerin sahip olduğu yaklaşık yüzde 25 verim düzeyleri, güneş enerji sistemlerinin büyük ölçüde kullanılması, enerji bağımlılığının azaltılması ve çevreye karşı olan sorumlulukların yerine getirilmesi anlamında engel oluşturmaktadır. Kısa vadede yapılması gereken bu sistemlerin verimlerini düşürebilecek her türlü etkenin etkisini sıfıra indirmek ve bu teknolojilerden mümkün olduğunca fazla bir şekilde yararlanmaktır.Güneş enerjisi sistemlerinin verimleri bu sistemlerin nasıl konumlandırıldığı, bu sistemlerde hangi ekipmanların kullanıldığı ve bunların kaliteleri, panellerin güneş ışığını hangi açıyla aldığı, gölge etkisinin olup olmadığı ve sistemin kurulduğu bölgedeki yıllık radyasyon yoğunluğunun ortalamasının ne olduğu gibi bir çok değişkene bağlıdır. Bu değişkenlerden birisi, özellikle kristal silisyum pillerin performansını önemli ölçüde etkileyen bulutluluk oranı'dır.Eğer bulutluluk oranı fazla ise direkt ışınım oranı azalacak ve difüzif ışınım oranı ise artacaktır. Bu nedenden dolayı difüzif ışınım altında iyi performans veren amorf-silisyum güneş hücresi dışındaki neredeyse her hücrenin performansı büyük ölçüde düşecektir. Bununla birlikte günümüzde kullanılan sistemler eğer şebekeye bağlı ise bu sistemlerin ne düzeyde üretim yapacaklarının bilinmesi, başka bir deyişle sistemlerin güç öngörüsünün yapılabilmesi gerekmektedir. Sağlıklı bir şekilde hesaplanmış bulutluluk oranı verisinin güç öngörüsü için kullanılan veritabanına eklenmesi, güç öngörülerinin daha tutarlı olmasına büyük ölçüde katkıda bulunacaktır.Bu tez çalışmasında bulutluluk oranı, çekilmiş olan yüksek çözünürlüklü fotoğrafların işaret işleme alanında oldukça başarılı olan dalgacık dönüşümü yöntemiyle analiz edilmesiyle hesaplanmıştır. Hangi dalgacığın kullanılması gerektiği konusunda özel bir renk skalası oluşturulmuş ve bu skala her bir dalgacık ile ayrı ayrı dönüşüme sokulmuştur. Bu veriler ışığında farklı hava şartlarında çekilmiş olan fotoğraflar Daubechies-2 dalgacığı kullanılarak 4 seviye boyunca ayrıştırılmış ve elde edilen en iyi yaklaşıklık bileşenine bir eşik değer uygulanarak bulutlu bölgeler (beyaz piksel) ve bulutsuz bölgeleri (siyah pikseller) ifade eden ikilileştirilmiş (binary) fotoğraf elde edilmiştir. Sonrasında bulutlu bölgeleri ifade eden beyaz piksellerin sayısı fotoğraftaki toplam piksel sayısına oranlanarak bulutluluk oranı verisine ulaşılmıştır. Son olarak ise kullanılan yöntemin fotoğraflardaki gürültüyü ne derecede filtreleyebildiğinin anlaşılması anlamında PSNR değerleri hesaplanmış ve yöntemin bu anlamda değerlendirmesi yapılmıştır.Sonuç olarak Dalgacık Dönüşümü'nün fotoğraflardaki gürültü bileşenlerini filtreleme konusunda oldukça başarılı olduğu ve bu başarının bulutluluk oranının yüksek doğrulukla hesaplanabilmesine olanak sağladığı görülmüştür. Özellikle açık, parçalı bulutlu ve parlaklık seviyesinin değişkenlik göstermediği kapalı hava fotoğraflarında Dalgacık Dönüşümü yönteminin bulutluluk oranı hesaplamakta iyi sonuç verdiği ve yüksek bir doğruluk ile bulutluluk oranının hesaplandığı görülmüştür. Kapalı havalarda, özellikle de güneş ışınları sebebiyle parlaklık seviyesinin değişkenlik gösterdiği fotoğraflarda ise Dalgacık Dönüşümü yeterli derecede iyi bir sonuç verememiştir.

Nowadays, the efficiency of solar power systems are at low level although great improvements were made on the subject. According to laboratory tests, monocrystalline solar cells have 27%, polycrystalline solar cells which have been utilized widely across Europe, have 22.5% and among relatively new solar cell technologies perovskite solar cells have 21.5% efficiency when operating maximum capacity. These low efficiency values pose an obstacle to both decreasing energy dependance for all countries which policies regarding to utilizing solar energy technologies ebtablished within and fulfill responsibilities towards environment.What needs to be done in the short term is that to diminish the effects of the factors which possibly may decrease efficiencies of these systems and to benefit from these technologies as much as possible.Efficiencies of solar energy systems depend on many parameters such as placement of modules, equipments used in these systems and their qualities, tilt angle which may have the biggest effect in this regard, shadow effect, average annual solar irradiance at the location of modules and so on. There are many more parameters or factors that effects the power output of solar energy systems. By utilizing solar systems without considering all of the possible factors, low efficiencies tend to decrease even more and thus it becomes almost pointless to use renewable technologies as a reliable energy source.Considering shadow effect and average annual solar irradiance parameters, these factors are also dependant to other circumstances. For instance, if shadow effect takes place in a solar system, it may stem from the dust collected on modules or objects that are in proximity such as trees or mountains but these factors do not effect solar irradiance. The phenomenon that effects solar irradiance and also causes shadow effect is formation of clouds. If clouds are present, solar system may experience shadow effect and direct/diffused radiation ratio is altered in a way that direct radiation ratio decreases and diffused radiation ratio increases. Increased diffused radiation ratio may be advantageous for amophous solar cell systems because these cells tend to give maximum performance under diffused light conditions but apart from this particular case, increased diffused radiation ratio causes a decrease in solar energy systems' power output because of direct radiation dependance of most solar energy systems.Moreover, being obliged to forecast power output of solar systems that are grid integrated requires accurate forecasting of power output which will only be possible if accurate data is collected. Measuring direct and diffused radiation does not contain any data regarding to whether shadow effect takes place or not. By knowing cloudiness ratio, its location based forecasting and possibly integrating these data to power output forecasting algorithms will greatly increase the accuracy of these predictions. In order to eliminate the lack of information based on cloudiness ratio forecasting and to provide useful data for power output prediction, cloudiness ratio is calculated in this thesis using widely known signal analysis technique `Wavelet Transform`.Wavelet Transform technique is developed to be an alternative for Fourier Transform in which electrical signals is analyzed by decomposing them into complex exponentials. In FT, all the time related data is transformed into frequency domain. Although this process is very succesful providing data regarding to what frequency components are present in the signal and at what proportion, this transform does not give any information related to when these components occur in the signal. All the time related data is lost when FT is applied to any signal. In order to have an extensive information without losing time variant components, Short-Time FT was developed which gives data for both frequency and time components. Despite of being able to give information about frequency and time components, STFT is not able to present these datas with same resolution in both domains. Hence low resolution becomes apperent when STFT is applied. That is why `Wavelet Decomposition` is developed to have an extensive information about the signal without any resolution loss.Wavelet Transform decomposes a signal into a family of wavelets. Unlike sinusiods, wavelets can be either symmetric/asymmetric, regular/irregular or sharp/smooth. Also wavelets are short-timed and have an average value of zero. There are different kind of wavelets such as Haar, Daubechies, Morlet, Coiflet,Symlets and etc. When a signal is decomposed using wavelets for N levels, signal is divided into two parts at every level, high and low-pass filters are applied to the signal after dividing process and it goes on until maximum level of decomposition (N) is reached. Section obtained after application high-pass filter is called `detail` and section obtained after application low-pass filter is called `approximation`. These filters are usually called `filter bank` and are fundamental elements of Wavelet Decomposition. High-pass filters are called wavelet functions and low-pass filters are called scaling functions.In this thesis, cloudiness ratio is calculated by analysing high-resolution photos using Wavelet Analyzer Toolbox in MATLAB.8 different high-resolution (10 MP) cloudy sky photos are taken from Istanbul Technical University Energy Institute. It is taken into account that photos have different cloud types and are taken during different weather conditions. The initial file format for the photos were JPEG. Just because JPEG is a compressive file format, it is not suitable for decomposition since a lot of non-visible components are erased during the process of compressing. In image processing the format Bitmap is mostly preferred. Since they have much higher resolutions and thus it is easy to perform changes on them and also have more information than any other format at pixel scale, image format is converted into BMP.In order to determine which wavelet should be used for this study, a simple color scale that has a range of colors with lightest one being white and the darkest one being cyan, is created and decomposed using different wavelets. It has been established that Daubechies-2 wavelet should be used for this study and maximum decomposition level should be 4. Also 2-dimension discrete wavelet decomposition is used for decomposing photographs since they are 2-dimensional.After decomposition, the approximation that gives best trade-off value between noise-free and high-resolution is chosen by visual inspection. Then, chosen approximation is binarized with black pixels indicating cloud-free areas whereas white pixels indicate cloudy areas. Next step of the study was to calculate the numbers of white pixels and dividing it to the total number of pixels in the photograph. By multiplying the result by 100, final cloudiness ratio was obtained as percentage. Lastly, in order to determine Wavelet Transform's talent on filtering out noise components from a given photograph, PSNR was calculated using approximations and denoised approximations. Average of 54 dB PSNR value was calculated which means Wavelet Decomposition method filters out noise components almost completely.Results of this study revealed that 2-D Discrete Wavelet Transform is very successful at filtering out noise components in photographs. Also this decomposing method is good at calculating cloud ratio for partial cloudy, clear weather and cloudy weather conditions in which brightness level is equally dispersed. Analysis was able to distinguish blue colored pixels (sky) from white colored pixels (cloud) with great success. But DWT was not able to give good results on cloudy weather conditions where average pixel value of all 3 layers were about the same and where brightness in photographs caused by sunlight was nonhomogenous.