Geri Dön

Rüzgar hızı yük tahmin modelleri ve Yalova bölgesinde bir uygulama

Wind speed load forecasting models and an application in Yalova

PDF İndir

  1. Tez No: 683169
  2. Yazar: ZELİHA NUR KİRİŞ
  3. Danışmanlar: DR. ÖĞR. ÜYESİ ÖMER FARUK BEYCA
  4. Tez Türü: Yüksek Lisans
  5. Konular: Endüstri ve Endüstri Mühendisliği, Industrial and Industrial Engineering
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2021
  8. Dil: Türkçe
  9. Üniversite: İstanbul Teknik Üniversitesi
  10. Enstitü: Lisansüstü Eğitim Enstitüsü
  11. Ana Bilim Dalı: Endüstri Mühendisliği Ana Bilim Dalı
  12. Bilim Dalı: Endüstri Mühendisliği Bilim Dalı
  13. Sayfa Sayısı: 90

Özet

Elektrik enerjisi çeşitli kaynaklardan elde edilebilen, santraller yardımıyla üretilen, iletim hatları ile yerleşim ve sanayi bölgelerine aktarılabilen fakat depolanamayan bir enerji kaynağıdır. Enerji daha çok fosil kaynaklardan elde edilmekteydi, bu kaynaklara sahip ülkeler ile kriz yaşanması sonucu, enerji kaynaklarının tükenebilir olması ve ülkelerin gelişmişlik seviyelerine göre enerji ihtiyaçlarının artması sebebiyle yeni enerji kaynakları araştırılmış ve yenilenebilir enerji önem kazanmıştır. Türkiye'nin sahip olduğu rüzgar enerjisi potansiyeli düşünüldüğünde enerji üretiminde rüzgar enerjisi önemli bir kaynak haline gelmiştir. Rüzgar enerjisini etkileyen birçok faktör ve rüzgarın değişken bir kaynak olması sebebiyle üretilen enerjinin verimli kullanılıp dağıtılabilmesi ve tasarrufla planlama yapılabilmesi için rüzgar enerjisi üretim tahminlerine ihtiyaç duyulmaktadır. Her bölgenin coğrafi koşulları (yükseklik, bulunduğu bölge vb.) farklı olduğu için bölgeye özel matematiksel modeller oluşturulmalıdır. Çalışmada Türkiye'nin Yalova ilinde bulunan bir istasyondan elde edilen veriler ile YSA yöntemleri kullanılarak kısa süreli rüzgar hızı tahmini yapılmıştır. Analizde 1 saat sonrasını tahmin ederek oluşacak ani arıza ve bakım planlamalarına müdahale edilmesi planlanmıştır. Öncelikle İstasyondan alınan veriler incelenmiş, veri analizleri yapılmış, var olan verilerden yeni veriler üretilmiş ve veri setleri modeller için uygun hale getirilmiştir. Modellerden elde edilen performans sonuçları kabul edilebilir aralıkta olduğunu göstermektedir. Modellerin analizdeki geçmiş veri büyüklüğünden etkilendiği ve farklı aktivasyon fonksiyonlarında farklı sonuçlar verdiği gözlemlenmiştir. Hata oranı düşük olabilecek kısa dönemli rüzgar hızı tahmin modeli sayesinde planlama daha doğru yapılacak ve enerji piyasaları daha doğru düzenlenebilecektir. Tahmini gerçekleştirmek için 4 farklı YSA yönteminden yararlanılan model kurulmuştur. Çalışma aralığı olarak 2015 (1 Haziran) - 2020 (31 Mayıs) tarihleri ve saatlik veri belirlenmiştir. Modeller arasından en iyi sonuçlar Model 2 (LSTM) ve Model 4 (İki yönlü GRU) ile elde edilmişti. En düşük MSE ve MAE değerini 7 günlük geçmiş veri ile oluşturulan analizde, Model 2 (LSTM) sırası ile 0.00054 ve 0.253 değeri ile elde etmiştir. En düşük MAPE değeri 30 günlük geçmiş veri ile oluşturulan analizde, Model 2 (LSTM) 18.5316 ile elde etmiştir.

Özet (Çeviri)

Electricity is an energy source that can be obtained from various sources, produced with the help of power plants, and transferred to residential and industrial areas via transmission lines, however cannot be stored. Recently, energy has been mostly obtained from fossil sources. New energy resources have been researched and renewable energy has gained importance because of the crisis with the countries that have these resources, the exhaustion of energy resources and the increase in energy needs in accordance with the development level of the countries. Considering the wind energy potential of Turkey, wind energy is an important source in energy production. There are plenty of factors that affect wind energy. In addition, due to the fact that wind is a variable resource, wind energy production forecasts are needed in order to use and distribute the produced energy efficiently and to make economic planning. Since the geographical conditions of each region (elevation, location, etc.) are different, region-specific mathematical models should be created. Load forecasting is to estimate the energy to be produced by using the characteristics of historical data and to minimise the risk of uncertainty for the company that will produce. Many studies have been put forward by companies and researchers on load forecasting, consumption forecasting, electricity price forecasting so far. In the energy industry, load forecasting - forecasting analysis is the most crucial component of the forecaster's decision process in production planning and price forecasting. These estimates, which are in the entire energy sector, are used by both power systems, firms and businesses. In recent years, energy forecasting models have been widely studied. Looking at the 10-year period (2010-2019), half of the energy forecasting studies are load forecasting-based studies. Other studies are price-based, wind-based and solar-based forecasts. For companies engaged in electricity generation and distribution, electricity load forecasting is an important process to increase efficiency and revenues, and the most important reason for making electricity load forecasting is to avoid over- and under-production. Different forecast time intervals are used for distinct purposes. For instance, real-time control of power generation management systems is provided by short-term load forecasting. This forecast period is used for load monitoring, network operations and regulation operations. The most important stage of the electrical power system planning and operational processes of the firm is the mid-term forecasts. Major conditions such as maintenance needs of many customers, demands of large facilities, seasonal changes, new demands affect mid-term forecasts. According to the forecasting results, operational management, energy reinforcement plans and also emergency plans can be made. These forecasts are similar to short-term forecasting, except that these analyses require less accuracy. Namely, their sensitivity is lower. The first step in the establishment, planning and development of generation, transmission and distribution in the future facilities is long-term demand forecasting. There are many factors that affect the loading. Different factors were used for different load forecasting times in the studies. Meteorological data is the most important independent variable in respect of load forecasting, especially in short-term forecasts. In this study, meteorological data were used as a variable, since short-term forecast was carried out. The areas where households and agricultural consumers are most affected by weather conditions can also change the loading profiles of industrial clients. Weather factors such as temperature, humidity, pressure exist in a certain place for a certain period of time. Forecasting the weather is a difficult and complex task owing to their unstable nature, even if they exist in a short time frame. Electric load forecast is a substantial milestone affecting the planning, generation and distribution processes in the electricity market. In this thesis, the speed of the wind after 1 hour is forecasted using short-term forecasting method, and it is aimed to find the potential of the wind energy that can be produced. Wind energy load forecasting can be calculated using the wind speed with the help of the theoretical formula. Energy forecasting provides predicting especially for renewable energy forecasts and electricity markets. Furthermore, it provides clues about how production areas can work efficiently and how production and distribution processes should be planned. Wind speed forecasts were made using hourly data between 2015 (First of June) and 2020 (Thirty-first of May). Meteorological data such as temperature, pressure, humidity, wind direction affecting wind speed were used as well. In this study, short-term wind speed forecasting was made using ANN methods from the data obtained from a station located in Yalova, Turkey. In the analysis, it is planned to intervene in the sudden breakdown and maintenance planning by predicting 1 hour ahead. First, the data received from the station were scrutinised carefully, data analyses were made, new data were produced from the existing data, and the data sets were made suitable for the models. If the wind direction data is used directly, it cannot reflect what is desired. Therefore, various operations were performed on the data in order to better understand the wind direction data and to use it in analysis. Angular data are not decent model inputs, 360° and 0° should be close together and neatly wrapped around. The wind vector was created by splitting into X and Y coordinates and used in the analysis. In the analysis, 4 different models were established. The performance of the models in processing different historical data dimensions was checked. When the results obtained with different activation functions are examined, the error rates of the prediction results are close to each other, while the Adam activation function provides the best result. The performance results obtained from the models are in the acceptable range. It has been observed that the models are affected by the historical data size in the analysis and give different results in terms of different activation functions. Although all four models achieved their goal, Model 3, using CNN and RNN, performed the worst performance. This model has the worst performance with regard to MSE, MAPE and MAE criteria. Both the GRU and LSTM units used to deal with the gradient problem were used in the model and it was investigated which one was more effective by keeping all other variables constant. The performance of Model 2 using LSTM unit gave better results than Model 1 using GRU unit. RNN trained with reverse sequencing learns different representations from the model trained with normal sequencing, and Model 4, set up in two directions, gave almost the same results as Model 2. Among the models, Model 2 (LSTM) and Model 4 (Bidirectional GRU) gave the best results. In the analysis of 7-days historical data, Model 2 (LSTM) showed the lowest MSE and MAE values, and these values were 0.00054 and 0.253, respectively. In the analysis of 30-days historical data, the lowest MAPE value is Model 2 (LSTM) and this value is 18,5316.

Benzer Tezler

  1. Tez No

    688423

    Deep learning for wind energy systems using the hurst exponent and statistical parameters

    Hurst üslü ve istatistiksel parametreleri kullanarak rüzgar enerjisi sistemleri için derin öğrenme

    BEHNAZ ALAFI

    Yüksek Lisans

    İngilizce

    İngilizce

    2021

    Elektrik ve Elektronik Mühendisliğiİstanbul Teknik Üniversitesi

    Elektrik Mühendisliği Ana Bilim Dalı

    PROF. DR. ŞAHİN SERHAT ŞEKER

  2. Tez No

    878622

    Implementation and comparison of different methods for the estimation of aircraft take-off weight based on ADS-B data

    ADS-B verilerine dayalı uçak kalkış ağırlığı tahmini için farklı yöntemlerin uygulanması ve karşılaştırılması

    KÜBRA ÇOLAK BİRGÜL

    Yüksek Lisans

    İngilizce

    İngilizce

    2024

    Uçak Mühendisliğiİstanbul Teknik Üniversitesi

    Uçak ve Uzay Mühendisliği Ana Bilim Dalı

    PROF. DR. METİN ORHAN KAYA

  3. Tez No

    197766

    ANN-artificial neural networks and fuzzy logic models for cooling load prediction

    Soğutucu yük tahmini için YSA-yapay sinir ağları ve bulanık mantık modelleri

    GÖKHAN BOZOKALFA

    Yüksek Lisans

    İngilizce

    İngilizce

    2005

    Makine Mühendisliğiİzmir Yüksek Teknoloji Enstitüsü

    Makine Mühendisliği Ana Bilim Dalı

    DOÇ.DR. SEDAT AKKURT

    Y.DOÇ.DR. GÜLDEN GÖKÇEN

  4. Tez No

    438153

    Impacts of distributed generations on distribution system reliability

    Dağıtık güç üretiminin elektrik daĞıtım sistemlerinin güvenilirliği üzerinde etkisi

    SORUSH SAEEDI

    Yüksek Lisans

    İngilizce

    İngilizce

    2016

    Elektrik ve Elektronik Mühendisliğiİstanbul Teknik Üniversitesi

    Elektrik Mühendisliği Ana Bilim Dalı

    PROF. DR. AYDOGAN OZDEMİR

  5. Tez No

    367632

    Vacuum infusion (VI) process modeling and material characterization with viscoelastic compaction models

    Viskoelastik sıkıştırma modelleri ile malzeme karakterizasyonu ve vakum infüzyon (VI) işlemi modellemesi

    BEKİR YENİLMEZ

    Doktora

    İngilizce

    İngilizce

    2014

    Makine MühendisliğiKoç Üniversitesi

    Makine Mühendisliği Ana Bilim Dalı

    DOÇ. DR. ERCÜMENT MURAT SÖZER

Geri Dön