Elektrik enerji piyasasında yerel marjinal fiyatların ajan bazlı modelleme tekniği ile incelenmesi
Analysis of local marginal prices in the electricity market using agent-based modelling technique
- Tez No: 932728
- Danışmanlar: DR. ÖĞR. ÜYESİ CANAN KARATEKİN
- Tez Türü: Doktora
- Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2025
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Lisansüstü Eğitim Enstitüsü
- Ana Bilim Dalı: Elektrik Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Elektrik Mühendisliği Bilim Dalı
- Sayfa Sayısı: 165
Özet
Bir elektrik piyasası, generatörler, yük servis sağlayıcıları ve sistem operatöründen oluşur. Sistem operatörleri hem üretim ve tüketim seviyelerini denetler hem de elektrik sisteminin fiziksel kısıtlamalarına karşın güvenli bir şekilde yük akışının yapılmasını sağlar. Bu süreçte generatörler üretim parametreleri ve maksimum üretim kapasitelerini sistem operatörüne bildirirken, yük servis sağlayıcıları da talep ettikleri güç miktarını her saat başı olmak üzere sistem operatörüne gönderirler. Alıcı ve satıcının eşleşmesi sonucunda yerel marjinal fiyatları (YMF) ve her bir generatörün ne kadar üretim yapması gerektiği yapılan optimal yük akışı çalışmaları sonucu ortaya çıkar. Yerel marjinal fiyatları belirlenirken, hem üreticiler hem de tüketiciler dinamik bir sürecin üyesidirler. Generatörler, uyguladığı öğrenme süreciyle belirli fiyat dağılımını kendi lehine çevirirken, alıcı tarafı da sistem üzerinde olabilecek değişikliklerle daha düşük fiyat dağılımlarını elde edebilirler. Bu tez çalışması dört kısımdan oluşmaktadır. Birinci kısımda, sabit ve fiyata duyarlı yük durumunda generatör öğrenmesi ile bara marjinal fiyatlar, hat tıkanıklığı, generatör günlük kazanç ve karları değişimi, ajan bazlı modelleme tekniği ile incelenmiştir. Çalışmanın ikinci kısmını oluşturan kapasite stopajı çalışmalarında seçilen generatör ve stopaj türünün generatör kazançlarında nasıl bir farklılık oluşturduğu ve hangi seçimlerin generatör kazançlarını pozitif anlamda artırdığı incelenmiştir. Üçüncü çalışmada, özelleştirilmiş bir elektrik piyasasında generator öğrenmesi ile generatör kâr maksimizasyonunun simülasyonu gerçekleştirilmiştir. Ayrıca, iletim hattı arızası durumunda dağıtık üretim birimlerinin elektrik piyasasına etkileri ve yerel generatörler üzerine etkisi araştırılmıştır. Yine çalışmanın devamında uzun vadeli hat arızası durumunda hat yerleri, dağıtık generatör konumu ve boyutunun üretim şirket kârları üzerine etkisi araştırılmıştır. Dördüncü çalışmada, test sistemindeki dinamik değişimlerin bara marjinal fiyatlarınına etkisi, baralara eklenen dağıtık üretim birimlerinin yer ve boyutları, yük talebi ve arızalı hat etkileri 6 boyutlu optimizasyon problemi olarak ele alınmış ve metaheuristik algoritmalar ile analizler yapılmıştır. Bu tez çalışması sonucunda öğrenme algoritmalarının üretim şirketleri tarafından kullanılması ile bu kuruluşların kârlılığını büyük ölçüde artırdığı görülmüştür. Simülasyonlar, dağıtık üretim (DÜ) birimlerinin piyasa dinamikleri üzerinde önemli bir etkiye sahip olabileceğini, daha büyük ve stratejik olarak konumlandırılmış DÜ birimlerinin generatör şirketlerinin kârlarını daha da artırdığını göstermiştir. Buna ek olarak bu çalışma, öğrenen generatör şirketlerinin gerçek marjinal maliyetlerinden daha yüksek marjinal maliyetler beyan ederek ekonomik stopaj yapmalarının bara marjinal fiyatlarının yükselmesine neden olduğuna işaret etmiştir. Bu davranış generator şirketlerinin kârlılığına fayda sağlasa da, piyasa verimliliği ve ataleti açısından zorluklar ortaya çıkarmakta ve bu tür stratejik manevraları ele almak için güçlü düzenleyici önlemlerin gerekliliğinin altını çizmektedir. Genel olarak bu çalışma, serbestleştirilmiş elektrik piyasalarının işleyişini araştırmak ve optimize etmek için güçlü analitik araçlar olarak ajan-tabanlı modellerin etkinliğini doğrulamaktadır. Ayrıca piyasa performansını ve direncini artırmayı amaçlayan politika ve operasyonel kararlara rehberlik edebilecek öngörüler sunmaktadır. Bu çalışmanın en önemli katkısı, test sistemindeki dinamik değişimlerin bara marjinal fiyatlarına etkisinin, baralara eklenen dağıtık üretim birimlerinin yer ve boyutlarının, yük talebi ve arızalı hat etkilerinin 6 boyutlu optimizasyon problemi olarak tanımlanıp metaheuristik algoritmalar ile çözümünü gerçekleştirmektir. Problemin çözümü için 5 ve 30 baralı test sistemlerinde ajan bazlı modelleme tekniği kullanılmıştır.
Özet (Çeviri)
An electricity market consists of generators, load service providers, and system operators. System operators monitor both production and consumption levels and ensure the safe flow of load despite the physical constraints of the electrical system. In this process, generators report their production parameters and maximum production capacities to the system operator, while load service providers send the amount of power they require to the system operator at the beginning of each hour. As a result of the matching of buyers and sellers, locational marginal prices (LMP) and the amount of production each generator should undertake are determined through optimal load flow studies. When determining locational marginal prices, both producers and consumers are part of a dynamic process. Generators, through their learning process, can turn a certain price distribution to their advantage, while the buyer side can obtain lower price distributions through potential changes in the system. As the load demand from the generators increases, the primary action to meet the current demand is to use generators with low production costs. However, considering the physical constraints (line congestion) and the increasing load, the appropriate action is to use generators with high production capacity. In this case, it is clear that generators that meet load demands can increase profits. The expected production cost parameters and production capacity information from generators under normal conditions should be accurately reported to the system operator; however, as load demand increases, generators have a much greater opportunity to earn more profit. It is clear that generators that increase their profits will hold power in the electricity market. Generators can conduct capacity withholding activities to gain a foothold in the electricity market. This can be done in two ways. Economic capacity stoppage, where the values of production cost parameters are not correctly provided but the maximum production capacity is accurately conveyed, and physical capacity stoppage, where the production cost parameters are correctly provided but the production capacity information is not accurately conveyed. Additionally, a combined capacity withholding tax, which uses both types together, can also be applied. Generators enter a learning process in their stopover operations. This is called generator learning. In this process, production parameters and production capacity values are considered as decision variables, and the optimization problem is solved using a type of reinforcement stochastic learning-based VREV (Version of Erev-Roth Algorithms) algorithm with the aim of increasing the generators' profit. In the economic capacity constraint, higher production cost parameters are obtained, while in the physical capacity constraint, lower production capacity values are obtained. With the additions made to an electrical system, the dynamic state of the system is obtained; for example, when a reserve application is made or when distributed generation units are integrated, different production and bus pricing can be encountered. At the same time, any physical failure in the system (such as a line break) or a change in demand will inevitably lead to changes. Changes made to electrical systems should comply with the endurance standards of the physical equipment present in the system. The changes made to the electrical system should not adversely affect the operation of the devices in the system. Therefore, the electrical system should be treated as an optimization problem, and the desired optimal solution should be found under certain constraints. As electrical systems are becoming increasingly complex, traditional optimization methods are not suitable for finding the optimal solution set for large-scale systems, which is why metaheuristic-based algorithms are being applied. In metaheuristic applications, finding the optimal solution is not always guaranteed; however, they are widely used because they yield results close to the optimal solution within a certain time frame. This thesis report consists of 4 studies. In the first study, the determination of local marginal prices (LMP) at each transmission bus and their market impacts in a restructured wholesale electricity market managed by an ISO operating on an AC transmission network were examined in the AMES wholesale seller simulation environment based on reinforcement learning. The study was conducted under the assumption that there are no generator outages and weather conditions, and the day-ahead market (DAM) environment where the production-consumption balance is not disrupted was simulated using the AMES V2.06 model. The conducted studies have been grouped under two main topics. The subject of the first study is Generator learning. Under fixed load, the normal and learning conditions of the generators were analyzed. The effects of the learning state on LMP determination, line congestions, and generator load distribution have been examined. In the second study, a price-sensitive demand analysis was conducted. By transitioning from fixed buyer loads to a price-based load model, the average LMP, average demand, average generator costs, and average Lerner index of the variable loads under variability were examined in both normal and learning conditions. Additionally, the daily earnings and profits of the generators were examined and analyzed according to their learning states. It has been observed that generator learning under fixed load affects LMP values, line congestion, and the load distribution of the generators. In the learning scenario, although congestion continued on some lines where the LMP value increased significantly, it was observed that the generator load distribution decreased. In the price-sensitive demand analysis, as load sensitivity (load flexibility) increased, there was a regular decrease in the Average Total Demand, Average LMP, Average Cost, and Average LI values under normal conditions. Additionally, this situation has also reduced the daily profits and earnings of the generators to the same extent. Considering the conditions of generator learning, a decrease in daily earnings and profit amounts is observed in all average values, but due to the learning effect, these values are still at high levels. As sensitivity increases, buyers' tendency to pay high prices decreases. The second study was conducted in the AMES environment, and three separate capacity witholding analyses—economic, physical, and combined—were applied for the analysis of the energy market. In the capacity witholding studies, it was examined how the selected generator and witholding type create differences in generator gains and which selections positively increase the generator gains. In the first analysis, under economic capacity withholding, studies were conducted in a day-ahead market environment with a test system having a daily fixed load profile. It was desired to observe which generator or generators in this test system increase their daily profits and whether they gain market power. During the learning process, the production parameters of the generators are updated again, and the obtained values are the values to be reported to the system operator. It is assumed that the maximum production capacity information of the generators is accurately reported to the system operator. In the second analysis, physical capacity witholding analyses were applied to the same test system. By examining the impact of the selected generator or generator groups on all the generators in the system, the factors affecting the daily earnings of the generators have been identified. During the learning process, the generators participating in the withholding study achieve a lower production capacity value, there are no changes in the production cost parameter values, and these values are reported to the system operator. The impact of the participating generators' earnings on the electricity market is then examined. It is assumed that the production cost parameters of the generators are reported accurately. In the third analysis, both economic and physical capacity witholding analyses were conducted on the same test system, and under these conditions, the daily net earnings of the generators were determined. It has been examined which generators, when selected for Economic and Physical Capacity Witholding, increase the profit of the generators. The analyses conducted in this study show that economic capacity withholding is much more advantageous than physical capacity withholding in terms of increasing the average net profits of generators. However, the impact of the system operator in mitigating market power has not been considered in these analyses. To alleviate the market, the reported costs and production capacities of generators should be closely examined by the system operator. Actual operating costs can be estimated from publicly available information such as fuel type and fuel prices, so economic capacity withholding can be monitored and controlled more easily than physical capacity withholding. However, it may be more difficult to verify whether the mandatory outages of production units are accurately reported. As a result, in these studies, it has been observed that the generator with low production cost but limited production capacity is not prominent in price determination due to physical constraints. Therefore, it cannot be said that the capacity withholding studies conducted by this generator have a specific impact on the profits of other generators. Conversely, when used in relatively large generator capacity withholding analyses, the lowest capacity generator's profit may vary. When the Economic Capacity Witholding is applied solely to the highest capacity generator, when the highest and lowest capacity generators are subjected to Economic Capacity Witholding together, or when the highest capacity generator is subjected to Combined Capacity Witholding, the lowest capacity generator cannot transfer load and its gain is 0. When the highest capacity generator enters physical capacity withholding alone, its profit is low. The two highest-capacity generators achieve significant gains when they enter economic capacity witholding, whether or not there is physical capacity witholding. In the third study, the effects of GenCo learning with the VREV algorithm in the AMES wholesale electricity market are analyzed for both error-free and erroneous situations, and the ways to increase generator daily profits and revenues are examined. In the first part of the study, the LMP profile of the buses in the test system, the load distribution of the generators, and the power flows on the lines are analyzed without generator learning and with learning. To demonstrate the accuracy of the results in the learning condition, the results obtained using the random seed (ID=03) applied in the relevant study are shown. The generator's daily net profits and revenues are evaluated based on the study using 20 random seeds, and average values are obtained and compared with the relevant study. In the second part of the study, a fault analysis is applied to the test system with the above-mentioned features. Each of the 6 lines in the test system is subjected to a 24-hour fault, and to examine whether the demanded load is met, a distributed generation unit with a power ranging from 10 MW to 50 MW is integrated into 5 buses, respectively. The impact of each dynamic component is analyzed by applying optimal load flow to the system. The learning processes of how GenCos maximize their profits and contribute to market stability have been examined. As the primary objective, a simulation of generator profit maximization through GenCo learning in a customized electricity market has been conducted. Additionally, the effects of distributed generation units on the electricity market and their impact on locational generators in the event of a transmission line failure have been investigated. Furthermore, in the continuation of the study, the impact of line locations, distributed generator positions, and sizes on the profits of production companies in the case of long-term line failures was investigated. To achieve the specified objectives, analyses were conducted on the functionality and efficiency of free electricity markets using agent-based modeling on the AMES platform. It has been observed that the use of learning algorithms by production companies, with strategic learning offer strategies adapted to real-time market conditions, significantly increased the profitability of these organizations. This situation has resulted in a significant increase in generator profits due to the sharp rise in GenCo's profits through the use of learning scenarios. Additionally, the study highlighted the crucial role of the independent system operator in maintaining market stability by effectively managing supply and demand through locational marginal pricing (LMP). Simulations show that distributed generation (DG) units can have a significant impact on market dynamics, with larger and strategically positioned DG units further increasing the profits of generator companies. Additionally, this study has indicated that learning generator companies declaring marginal costs higher than their actual marginal costs and engaging in economic withholding have led to an increase in LMPs. Although this behavior benefits the profitability of GenCos, it poses challenges in terms of market efficiency and inertia, highlighting the necessity for strong regulatory measures to address such strategic maneuvers. Overall, this study validates the effectiveness of agent-based models as powerful analytical tools for investigating and optimizing the functioning of liberalized electricity markets. It also provides insights that can guide policy and operational decisions aimed at enhancing market performance and resilience. In the fourth study, the locational marginal prices (LMP) at each transmission bus in the restructured wholesale electricity market were examined in the MATPOWER and MOST simulation environments. Various studies have been conducted considering the generator reserve effect, distributed generation effect, desired load demand effect, and faulty line effect. The aim of this study is to examine the extent to which the aforementioned effects impact locational marginal prices and how favorable these results can be for the load side. As a result of the conducted studies, it is aimed for the distribution of bus prices to be close to each other and for the demand to be met at a lower price. In this study, the impact of dynamic changes in the test system on the distribution of marginal prices at the bus; the location and size of distributed generation units added to the buses; load demand and faulty line effects were addressed as a 6-dimensional optimization method, and the aim was to find the optimal solution set using selected metaheuristic algorithms such as Particle Swarm Optimization (PSO), Grey Wolf Optimization (GWO), Whale Optimization Algorithm (WOA), hybrid algorithms PSOGWO and IGWO. The optimization studies were conducted with 30 rounds and 500 iterations. The Wilcoxon Statistical test was applied to compare the algorithms with each other. The obtained results are as follows:When looking at the optimization parameters, it is expected that the objective function will decrease in the case where the load demand is at its minimum. It has been observed that the load demand is at its lowest in the solution set of all algorithms.It has been observed that the location and size of distributed production units are important in finding an appropriate solution. The location of the fault has a significant impact on the LMP, and it has important effects on the separation and line congestion in the LMP distribution. When analysed according to the best results, WOA reached the optimum result by obtaining the lowest value, IGWO, GWO, PSO algorithms were the algorithms that could obtain close results. According to the average results, IGWO reached the most optimum value. The results obtained in GWO are also close to the optimum value. Upon analyzing the box plot on the objective function, it is seen that IGWO, where the dispersion is low, is the most performant algorithm. It is followed by other algorithms, especially GWO and PSOGKO. When the box plot on the YMF distribution is analysed, PSOGKO, where the dispersion is small, is the most performant algorithm, while other algorithms, especially WOA and IGWO, are ranked. When the significance is investigated according to the Wilcoxon statistical test, it is understood that IGWO is the algorithm with the highest performance. GWO algorithm also follows this algorithm. Considering all the criteria, it is seen that IGWO is the most performant algorithm. Other algorithms, especially GWO and PSOGWO, are ranked.
Benzer Tezler
- Yapay sinir ağları kullanılarak Türkiye gün öncesi piyasası elektrik fiyat tahmini
Turkish day ahead market electricity clearing price forecasting using artificial neural network
ASLIHAN DALGIN
Yüksek Lisans
Türkçe
2017
Enerjiİstanbul Teknik ÜniversitesiEnerji Bilim ve Teknoloji Ana Bilim Dalı
DOÇ. DR. GÜLGÜN KAYAKUTLU
- Farklı makine öğrenmesi yöntemleri ile enerji piyasasında yerel marjinal fiyat tahmini
Local marginal price forecasting using different machine learning approaches
BAŞAK ERSÖZ YILDIRIM
Yüksek Lisans
Türkçe
2023
Elektrik ve Elektronik MühendisliğiYıldız Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
PROF. DR. OZAN ERDİNÇ
- Elektrik enerji piyasalarında iletim hat parametrelerinin tıkanıklık üzerine etkileri
The effects of transmission line parameters on congestion in competitive power markets
CANAN ZOBİ
Doktora
Türkçe
2005
Elektrik ve Elektronik Mühendisliğiİstanbul Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
DOÇ.DR. CANBOLAT UÇAK
- Economic comparison of new built internal combustion gas engine power plant in Uzbekistan and Europe based on levelized cost of electricity (LCOE) method
Özbekistan ve Avrupa'da yeni inşa edilen içten yanmalı gaz motorlu elektrik santralının seviyelendirilmiş elektrik maliyeti (LCOE) yöntemine göre ekonomik karşılaştırması
SERKAN TOPLAK
Yüksek Lisans
İngilizce
2024
Enerjiİstanbul Teknik ÜniversitesiEnerji Bilim ve Teknoloji Ana Bilim Dalı
PROF. DR. MEHMET ÖZGÜR KAYALICA
- Tekstil sektöründen enerji sektörüne: Sanko holding örneği
From the textile industry to energy industry: In case of Sanko holding company
TUĞBA BAŞARSLAN ARSLAN