Short term wind power forecasting using hybrid models

Short term wind power forecasting using hybrid models

Masilamani, S.

Muñoz San Roque, A. (mentor)

Technology, Policy and Management

Engineering, Systems and Services

Erasmus Mundus Joint Master in Economics and Management of Network Industries (EMIN)

2017-08-24

This study improves the Short term wind power forecasting to help bid the wind power in the electricity market. Supplying power lesser/greater than the expected power creates imbalance in the Electricity system. Hence electricity markets impose penalty for supplying power lesser/greater than expected power. Bidding right amount of power is an important issue for the electricity power producers. This issue is very relevant for a wind power producer due to the inherent nature of wind. Wind is characterised by uncertainty and volatility. This study proposes hybrid approaches that use the meteorological forecast of wind power and statistical models to improve the accuracy of the wind power forecast over meteorological forecast. The statistical methods used in the study are linear regression model, Artificial Neural Network (ANN), Support Vector Machine (SVM) and K-Nearest Neighbour. The production data of ten wind farms in a portfolio, meteorological forecasts of the ten wind farms, total production of the portfolio and meteorological forecast of total production were collected for 532 days for every hour. These data were used to train and test the hybrid models. These hybrid models are then compared empirically with the meteorological forecasts. It is found that, for the data used for the study, hybrid model using artificial neural network performs the best but only slightly over the linear regression model. Followed by artificial neural network and linear regression model is support vector machine. Followed by support vector machine is K-Nearest Neighbour model. But all the hybrid models outperform the meteorological forecast of wind power.

Forecasting
Wind Power
Multi-layer Perceptron
Support Vector Machine
K-NN Algorithm
Linear Regression Model

http://resolver.tudelft.nl/uuid:02fb40f9-079e-44e3-9a84-0961686221f4

Student theses

master thesis

(c) 2017 Masilamani, S.