Data-driven modeling using deep neural networks for power systems demand and locational marginal price forecasting

Download

Honest Jimu_MSc_Thesis.pdf

Date

2022-9

Author

Jimu, Honest

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

386
views

250
downloads

Forecasting electricity demand and locational marginal prices (LMPs) have become critical components for power system security and management. Electricity Demand Forecasting (EDF) aids the utility in maximizing the use of power-generation plants and scheduling them for both reliability and cost-effectiveness. In this thesis, a novel Deep Neural Network Long Short-Term Memory (DNN-LSTM) forecasting model is suggested to improve accuracy and robustness for predicting hourly day ahead power system load and LMPs in two distinct markets, North Pool (NP), and New England-ISO (NE-ISO). Historical load, weather, statistical features derived from historical data, and system outage information (known as Line Outage Distribution Factors (LODFs)) will be used as input features in the proposed model. Two distinct demand-forecasting models will be modeled using two case studies that present different market patterns from different geographical locations. The deep neural network model will be compared with the state-of-the-art Lasso Estimated Autoregressive (LEAR) model using a variety of performance metrics, including Symmetric Mean Average Percentage Error (sMAPE), Root Mean Square Error (RMSE), Mean Average Percentage Error (MAPE), Relative Mean Average Error (rMAE), Mean Average Error (MAE). The results acquired from the two experimental case studies on the markets revealed that the proposed DNN model showed significant improvement in hourly demand and LMP forecasts and therefore outperformed contemporary statistical forecasting techniques in accuracy, computational time, and reliability.

Subject Keywords

Deep Neural Networks, Electricity Load Forecasting, Locational Marginal Price Forecasting, Line Outage Distribution Factors (LODFs)

URI

https://hdl.handle.net/11511/99444

Collections

Northern Cyprus Campus, Thesis

Suggestions

OpenMETU
Core

Application of a Hybrid Machine Learning model on short term electricty demand prediction Assar, Ahmed Khaled Ahmed Farouk; Fahrioğlu, Murat; Sustainable Environment and Energy Systems (2022-2) Electricity demand forecasting is an important procedure in the electricity market and plays a great role in assuring a sustainable and efficient operation chain. By accurately forecasting the demand, one can see a considerable reduction in production costs as well as saving energy resources. Therefore, optimizing the demand forecasting techniques became an inseparable goal of power economics, leading to the introduction of machine learning to this sector that proved to be superior to other pre-defined alte...
Hydro-Optimization-Based Medium-Term Price Forecasting Considering Demand and Supply Uncertainty İLSEVEN, Engin; Göl, Murat (2018-07-01) This paper proposes an electricity market model of Turkish electricity market for monthly and yearly electricity price forecasting in medium-term by means of supply and demand dynamics formed via a theoretical approach. The electricity market model created within this scope consists of three main components related to electricity demand, supply, and price segments along with hydro optimization submodel, which takes into account the nonlinear relation between supply and price. Electricity price is determined...
An empirical evidence for generalized shrinkage methods: application of bagging in day-ahead electricity price forecasting and factor augmentation . Özen, Kadir; Yıldırım Kasap, Dilem; Department of Economics (2020) Fundamental dynamics behind electricity prices are multi-dimensional and elaborate. A popular approach to forecasting electricity price is to utilize large number of predictors. In this study, using the day-ahead electricity price data from commonly studied markets of five major series and GEFCom2014 data, a variant of shrinkage method, Bootstrap Aggregation (bagging) is proposed to incorporate information from available predictors. Bagging manifests itself as a computationally simpler alternative to common...
Dynamic correlations between oil prices and the stock prices of clean energy and technology firms: The role of reserve currency (US dollar) KOCAARSLAN, BARIŞ; Soytaş, Uğur (2019-10-01) There is increased interest in the dynamic relationships between the stock prices of clean energy and technology firms and oil prices in the literature. Existing works suggest a time-dependent link between them, but there is a gap of knowledge regarding the drivers of this time-dependent relationship. To contribute to this literature, we first identify dynamic conditional correlations (DCCs) between the prices of clean energy and technology stocks and oil prices to investigate the nature of these dynamic co...
A novel methodology for medıum and long-term electricity market modeling İlseven, Engin; Göl, Murat; Department of Electrical and Electronics Engineering (2020-11-15) In the electricity market, there is a considerable degree of uncertainty in electricity demand, supply, and price due to the uncertainty in parameters such as economic growth, weather conditions, fuel prices, and timing of new investments, etc. These factors in return affect the predictability of the electricity market. This thesis aims to increase the predictability and observability of the electricity market by means of a suitable and validated electricity market modeling methodology designed for medium a...

Citation Formats

H. Jimu, “Data-driven modeling using deep neural networks for power systems demand and locational marginal price forecasting,” M.S. - Master of Science, Middle East Technical University, 2022.