Implementation of Genetic-Algorithm-Based Forecasting Model to Power System Problems

Implementation of Genetic-Algorithm-Based Forecasting Model to Power System Problems

Sajad Madadi (University of Tabriz, Iran), Morteza Nazari-Heris (University of Tabriz, Iran), Behnam Mohammadi-Ivatloo (University of Tabriz, Iran) and Sajjad Tohidi (University of Tabriz, Iran)
DOI: 10.4018/978-1-5225-4766-2.ch007
OnDemand PDF Download:
Available
$37.50
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Power system includes many types of markets. Such markets are generally cleared at certain times, whereas market participators have to determine their operational plans before meeting the actual conditions. Therefore, forecasting methods can assist market players. Forecasting methods are applied to forecast electricity demand. The unknown conditions in the power system are increased by integration of renewable generation units. Forecasting methods, which are used for the load forecasting, are updated because the output power of renewable generation units such as wind farms and photovoltaic (PV) panels have more deviation than power demand. The pool market can be introduced as other parameter that is forecasted by market players. In this chapter, the authors investigate a mathematical model for forecasting of wind. Then, the forecasting model is proposed. Genetic algorithm is applied as an optimization method to handle delay associated with wind forecasting.
Chapter Preview
Top

1. Introduction

The restructured power network includes different electricity markets. The electrical cost of industrial unit can be reduced by participating in electrical markets. Such markets are a motivation for small industrial units to manage electrical demand and store electrical energy. An essential step for scheduling electrical demand is load forecasting. Results of the load forecasting are used to reach optimal industrial planning. Load forecasting methods proposed in the recent research studies can be classified to three types. First type is long-term forecasting with horizon of prediction of generally 20 to 50 years. Results of this type of load forecasting are applied for expansion planning. A model is proposed in (T. Hong, Wilson, & Xie, 2014) for long-term load forecasting, which predict load with implementing hourly information of loads. Second type is mid-term forecasting that include forecasting of few months. Such forecasting model are utilized for scheduling aims, risk management, and valuation of the exchange traded futures and bilateral contracts (Maciejowska & Weron, 2016). Third type of forecasting is allocated to short-term forecasting, for which the forecasting horizon is two weeks or less (Sun, Xiaorong, Luh, Peter, Cheung, Kwok, Guan, Wei, Michel, Laurent, Venkata, Miller, & Melanie, 2016).

Different modeling methods are implemented to predict short-term forecasting problem such as load, wind speed, market price, and solar radiation. Since artificial neural networks have a good performance to model non-linear behaviors, such networks are more popular to create a forecasting model. The most common neural network model is the multi-layer Perceptron (MLP). This type of neural network is known as a supervised network because it requires a desired output in order to train. The goal of such type of network is to create a model that correctly maps the input to the output using historical data so that the model can be used to produce the output when the desired output is unknown. A model based on MLP for load forecasting is presented in (Hippert, Pedreira, & Souza, 2001). In (Sheikhan & Mohammadi, 2013), MLP neural network is combined with particle swarm optimization (PSO) algorithm, which is applied to predict hourly load. The optimization process includes setting the number of neurons in the hidden layers of MLP. The application of MLP neural network is improved by selecting the weights and bias in training the MLP with optimization approaches such as PSO. In (Yeh, Yeh, Chang, Ke, & Chung, 2014), a wind power forecasting is proposed, which is based on the MLP neural network and is improved by using simplified swarm optimization. Radial basis function (RBF) is other type of artificial neural networks, which is generally used to accomplish nonlinear forecasting projects. The RBF network has a hidden layer of basis function or neurons. The output of each neuron calculation is based on distance between the neuron center and the input vector. Forecasting models based on RBF neural networks are presented in (Li, Gong, Shi, & Jing, 2010; W. Li, Wei, Sun, Wan, & Miao, 2009; Yona, Atsushi, Senjyu, Tomonobu, Saber, Ahmed Yousuf, Funabashi, Toshihisa, Sekine, Hideomi, Kim, & Chul-Hwan, 2007). A RBF neural networks is created in (Chang, 2013) by combining orthogonal least squares (OLS) algorithm and genetic algorithm (GA). These networks are used to forecast the wind speed.

Complete Chapter List

Search this Book:
Reset