Geri Dön

Regresyon ve yapay sinir ağları ile fotovoltaik panel yüzey sıcaklığı tahmini

Photovoltaic panel surface temperature prediction by using regression and ann methods

PDF İndir

  1. Tez No: 639134
  2. Yazar: ÇAĞATAY ÖZTÜRK
  3. Danışmanlar: DR. ÖĞR. ÜYESİ BURAK BARUTÇU
  4. Tez Türü: Yüksek Lisans
  5. Konular: Enerji, Energy
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2020
  8. Dil: Türkçe
  9. Üniversite: İstanbul Teknik Üniversitesi
  10. Enstitü: Enerji Enstitüsü
  11. Ana Bilim Dalı: Enerji Bilim ve Teknoloji Ana Bilim Dalı
  12. Bilim Dalı: Enerji Bilim ve Teknoloji Bilim Dalı
  13. Sayfa Sayısı: 82

Özet

Güneş, tükenmez enerji kaynaklarının en önde gelen türlerinden biridir. Fosil kaynakların tükenmesi ve çevreye verdikleri zararın her geçen gün daha da belirgin hale gelmesiyle birlikte, güneş enerjisine yapılan yatırımlar da artmaktadır. Düşük bakım maliyetleri, kısa zaman içerisinde mega fotovoltaik tarlaların kurulabilmesi ve dünyanın hemen her bölgesinde verim sağlayabilmesi açısından batarya sistemlerinin maliyetlerinin zamanla azalması ile birlikte gelecekte elektrik üretimin en önemli kaynağı olması beklenmektedir. Fotovoltaik elektrik üretimi bugün itibarı ile de dünyada elektrik üretimine, hem ev veya fabrika gibi binaların iç ihtiyaçlarını karşılama hususunda hem de yüksek kurulu güç kapasiteli güneş tarlaları ile katkıda bulunmaktadır. Genellikle silikondan üretilen hücrelerin iç dirençleri artan panel sıcaklığıyla birlikte artmaktadır. Bu direnç elektrik üretimini negatif yönde etkilemektedir. Farklı coğrafi bölgelerde bu sıcaklığı olumlu ya da olumsuz yönde etkileyen meteorolojik değişkenler de değişkenlik gösterir. Dolayısıyla fotovoltaik panel yüzey sıcaklığının tahmin edilmesi bir güneş tarlasının kurulduğu ya da kurulması planlanan lokasyonda ne kadar verimli olduğu ya da olacağı açısından ve bu lokasyonda üreteceği yıllık enerji miktarının öngörülebilmesi açısından oldukça önemlidir. Bu nedenle bu çalışmada aktif bir güneş tarlasındaki fotovoltaik panel sıcaklığının tahmini üzerine yoğunlaşılmıştır. Bu çalışmada toplam dört adet model vasıtası ile fotovoltaik panel yüzey sıcaklığının tahmininin gerçekleştirilmesi ve bu tahminlerin belli hata kriterlerine göre doğruluklarının sınanması amaçlanmıştır. Bu modeller regresyon modeli ve yapay sinir ağı modelleridir. Bu modellerin sınanması ve kıyaslamaları yapılırken R^2, MBE, MAE, RMSE gibi hata kriterlerinden faydalanılmıştır. Gerçekleştirilen modeller arasında tahmin başarımının en yüksek olduğu model FFN modeli olarak göze çarpmaktadır. Başarımın en düşük olduğu model ise Regresyon modeli olduğu saptanmıştır. Ayrıca gerçekleştirlen tüm yapay sinir ağları modelleri regresyon modelinden daha yüksek başarıma sahip olmuştur. Yapay sinir ağları modellerinin başarımının regreyon modeline nazaran başarımlarının yüksekliği göze çarpmaktadır.

Özet (Çeviri)

Solar energy is one of the leading sources of renewable energy sources in electricity generation. The reason for this is that it is one of the most accessible sources, and it can be obtained in most regions of the world. Electricity generation is generally carried out with photovoltaic panels made of silicon and germanium. Their use is increasing and widespread in terms of photovoltaic systems being largely applicable to all fields and producing in a wide spectrum of installed power spectrum from micro scale to macro scale. It is expected that the share of solar energy in electricity generation will increase significantly with the development of battery technologies and the decrease in costs. Photovoltaic systems are a candidate to be the leading actor of electricity generation in the coming years because of their low maintenance costs and the construction of high installed power capacities in a short time. The efficiency of photovoltaic systems varies according to the geography in which it is built. Investors want to know how long the return of that investment will take place, as in investment, before the construction of photovoltaic systems begins.The main gain of a photovoltaic system; It is equal to the total electricity energy produced in kWh, usually multiplied by the unit price in kWh. In addition, institutions managing the grid in many regions require independent electricity producers to declare the amount of electricity they will produce for certain periods. Therefore, it is important to estimate the electrical energy produced by a photovoltaic system. Temperature directly affects the amount of electrical energy produced by photovoltaic panels. With the increase in temperature, an electrical resistance is formed on the material from which the module is formed, resulting in decreases in energy production. Knowing the panel surface temperature is important for determining the return time of the previously disclosed investment and the amount of energy to be reported to the electricity administration. In addition, this surface is valuable in terms of guiding the studies for lowering the temperature. For the reasons explained above, this study focuses on estimating the surface temperature of a photovoltaic panel. Sensor data measuring solar radiation, wind speed, ambient temperature data received from a photovoltaic power plant with an installed capacity of 1 MW within the boundaries of Hatay province and the actual panel surface temperature in the same power plant were collected. These data include the values from the first day of 2018 to the last day. The surface temperature estimation study of the photovoltaic panel was carried out with the help of models installed with four different methods. These models are respectively regression model, artificial neural network models; The FFN model is the GRNN model and the RBFN model. Regression model by establishing multiple regression model, T_m(\hat{Y})=0.0157G-0.6641W+1.0469T_a+8.3302 photovoltaic surface temperature; Equation depending on the radiation, wind speed and ambient temperature was obtained. As expected, the effect of wind speed on panel surface temperature is negative. This results in the wind speed being positive for energy production. As the ambient temperature is high, it directly causes an increase in the temperature of the photovoltaic panels, which has a negative effect on energy production. It is observed that the solar radiation that enables the photovoltaic systems to generate electricity, on the other hand, increases the temperature of the photovolatic panel, thus indirectly causing a decrease in energy production. In the FFN model, panel surface temperature data taken at fifteen minute intervals are divided into groups of five. The first four data of each group were used to train the model and the remaining data was used to test the model. The number of neurons in the hidden layer started with 5 and increased to 20 neurons one by one. Experiments were carried out by increasing the input neurons from 1 to 50. In the GRNN model, as in the FFN model, the panel surface temperature data, taken at intervals of fifteen minutes, is divided into groups of five. The first four data of each group were used to train the model and the remaining data was used to test the model. In this model, Gauss function was chosen as the activation function. Since this network model is fixed as in RBF, no determination has been made regarding the number of layers. The embedding size has been increased one by one from 200 to 200 and the model has been trained. In the RBFN model, as with other artificial neural network models, panel surface temperature data, taken at intervals of fifteen minutes, is divided into groups of five. The first four data of each group were used to train the model and the remaining data was used to test the model. There is one RBF layer in this network structure. Since it is a fixed network model, no adjustment has been made to the number of layers. Gauss function was chosen as the activation function. As explained in this study, it is aimed to estimate the photovoltaic panel surface temperature by means of a total of four models and to test the accuracy of these estimates according to certain error criteria. There is no single metric accepted in the literature in evaluating the model performance. Determination coefficient (R2), Mean-Absolute Error (MAE), Relative Mean-Absolute Error (rMAE), Root Mean Square Error (RMSE), Relative Root Mean Square Error (rRMSE), Mean Bias to evaluate the performance of the method proposed in this thesis study. Criteria such as Error (MBE) and Relative Mean Bias Error (rMBE) were used. Models are evaluated one by one according to the specified error criteria and model achievements are revealed. First, the determination coefficient calculated in the multiple regression model was calculated as R2 = 0.9979. The calculated errors of this model stand out as MAE = 1.257 and MBE = 0.8482. In feed forward neural network model, determination coefficient was calculated as R2 = 0.9992. It has been revealed that the calculated errors of this model are MAE = 0.7039 and MBE = 0.0165. Determination coefficient in generalized regression neural network model; R2 calculated as 0.9988. It has been determined that the calculated errors of this model are MAE = 0.8760 and MBE = 0.0118. Determination coefficient in the radial basis function network model; R2 calculated as 0.9983. The calculated errors of this model were determined as MAE = 0.6867 and MBE = 0.0564. R2 value is between 0 and 1. The large R2 value is a positive factor for the significance of the model. In other words, the significance of the model will increase as R2 approaches one. When the results given by the models described before are examined, the model with the largest R2 value is FFN. (R2 = 0.9992) At the same time, low RMSE value is important for the performance of the model. It can be said that the model with the nearest RMSE value to zero has a better performance. When the errors of the applied models are examined, it is seen that the model with the lowest RMSE value is the FFN model. Therefore, it seems that the best model is the FFN (Feed Forward Neural Network) model, because it has the largest R2 value and the smallest RMSE value. In addition, when determination coefficient and errors are examined, it will be seen that all artificial neural network models give better results than multiple regression model. Thus, the results of this study revealed that artificial neural network methods provide superiority to regression method. In this study light, where the photovoltaic panel surface temperature is estimated using wind speed, solar radiation and ambient temperature variables, a similar study can be carried out using data from different regions. Thus, the relationship between models in different geographical regions and models in this study can be examined. In addition, data such as soil temperature and humidity can be collected and added to the model. This can create a stronger model with more variables.

Benzer Tezler

  1. Tez No

    575318

    Güneş enerji sistemlerinin performans tahmini için yapay sinir ağları ile modellenmesi ve verimliliğin incelenmesi

    Modeling solar systems using artificial neural networks for performance prediction and investigation efficiency

    YASİN İÇEL

    Doktora

    Türkçe

    Türkçe

    2019

    Elektrik ve Elektronik Mühendisliğiİnönü Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    PROF. DR. MEHMET SALİH MAMİŞ

  2. Tez No

    515814

    Fotovoltaik sistemin yapay zeka algoritması ile güç tahmini

    Power estimation of photovoltaic system with the artificial intelligence algorithm

    SULTAN MEHTAP İZMİRLİ AYAN

    Yüksek Lisans

    Türkçe

    Türkçe

    2018

    EnerjiKırklareli Üniversitesi

    Enerji Sistemleri Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ HAYRETTİN TOYLAN

  3. Tez No

    783711

    Comparison of experimental and semi-experimental models for predicting solar thermal power plants with artificial neural network

    Solar termal santralların yapay sinir ağlarıyla öngörüsünde deneysel ve yarı-deneysel modellerin karşılaştırılması

    SHABNAM CHOOPANI

    Yüksek Lisans

    İngilizce

    İngilizce

    2023

    Enerjiİstanbul Teknik Üniversitesi

    Enerji Bilimleri ve Teknoloji Mühendisliği Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ BURAK BARUTÇU

  4. Tez No

    484265

    Güneş panelleri ile üretilen enerjinin ortam bilgileri kullanılarak yapay sinir ağlarıyla tahmini

    Prediction of power produced by solar panels with artificial neural networks by using environmental factors

    İSMAİL KAYRİ

    Doktora

    Türkçe

    Türkçe

    2017

    Elektrik ve Elektronik MühendisliğiFırat Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    PROF. DR. MUHSİN TUNAY GENÇOĞLU

  5. Tez No

    747192

    Güneş paneli enerji üretim tahmininin makine öğrenmesi yöntemleri ile karşılaştırılması

    Comparison of solar panel energy production forecast with machine learning methods

    HAVVA AYYILDIZ KOÇ

    Yüksek Lisans

    Türkçe

    Türkçe

    2022

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolAmasya Üniversitesi

    Yenilenebilir Enerji ve Uygulamaları Ana Bilim Dalı

    DR. ÖĞR. ÜYESİ YAVUZ ÜNAL

Geri Dön