Nowadays power plants are at risks and uncertainties that are coming from different sectors in a deregulated market environment. Some of these sectors are as follow [17, 85]:
technical limitations
unstable electricity market prices fuel price uncertainty
It is difficult to determine the relationship between the load and the stated factors. Researchers have used ANN systems to define the relationship between the load and selected factors since the ANN technique would be able to encode complex and nonlinear relations. Moreover, research shows that load forecasts which have been produced by ANN techniques are more accurate compared to other factors [69, 70].
This research aims to simulate a combined control scheme for the FLMVPP system with different types of DERs. The main aim of this task is to optimize the DERS’ operation based on some factors such as demand and electricity price broadcast by the FLMVPP system. In doing so, the ANN algorithm is chosen to identify the scheme of this DERs group. In fact, the system needs adequate historical information of each DER
99 to characterize a combined scheme of the DERS’ to different electricity price. In this study, the historical data collected from Australian Electricity Market operator (AEMO) website for different period of time. The AEMO is defined in the following subsection.
5.3.1 Australian National Electricity Trading
A large amount of Australia’s electricity is still produced from the combustion of fossil fuels such as coal, gas, and oil [85].The amount of energy produced in Australia from 1974-2015 has been shown in Figure 5.2.
FIGURE 5.3.ENERGY GENERATION IN AUSTRALIA BY VARIOUS FUEL TYPES [85]
The National Electricity Market (NEM) in Australia is a wholesale market for the supply of electricity to retailers and end-users in the six Australian states [85, 86]. In NEM, the electricity exchange between generators and consumers is facilitated where the total generation output is combined in a pool and scheduled to meet the demand.
This electricity pool is not physical but rather a set of procedures that the Australian Energy Market Operator (AEMO) manages according to statutory provisions [86].
100 Efficient operation of the Australian NEM is retained by AEMO, which also develops the electricity market, improves its efficiency, and coordinates the planning of the interconnected power system. The demand and supply of electricity in Australia is balanced through AEMO’s actions. Hence, sophisticated IT systems are needed to underpin the operation of the NEM including the responsibilities to balance supply with demand, select operational elements at any given time and determine the spot price. A centrally-coordinated dispatch trading process referred to as the spot market is used to match supply and demand instantaneously in real-time [86].
In this research study, to optimize the operation of the FLMVPP system, the ANN tool with three layers is used in MATLAB environment. The input-output data used in this tool are collected from the simulation run in chapter three and AEMO website as well. It is essential to find an appropriate ANN structure as the computational performance of the method is sensitive to the number of layers and neurons. The general structure of ANN technique has been shown in Figure 5.3.
FIGURE 5.4.ARCHITECTURE OF A THREE-LAYER ANN[13]
The question of how many hidden layers and how many hidden nodes should exist in ANN layers have not been precisely solved yet. In this research, the ANN network with three-layer structure; two-hidden and one output layer is considered. To compute an
101 arbitrary function of inputs, neurons of two hidden layers which use sigmoid transfer functions are sufficient and a single hidden layer is enough for classification problems [70]. As choosing the number of neuron in the hidden layer is difficult, the below formula is considered based on theoretical and empirical limits [13, 70].
(5.27)
Where; N is the number of inputs.
The output produced by the BPNN is called estimated output. The difference between target output and the estimated one is called error. Thiserror is usedin the calculation of Least Mean Square Error (LMS) which is shown in the following Equation [13, 70]:
( ) (5.28)
Where E is error, TO is the ANN target and DO is the desired output.
The BPNN operates by propagating errors backward from the output layer and would stop operation when unacceptable error is occurred.
To optimize the operation of the FLMVPP system, some factors needs to be satisfied. These factors are as follows:
Minimizing the cost of the load
Using the VPP resources as much as possible
102 As the storage system is installed in the FLMVPP system, the required numbers of price scenarios become larger since responding to the price signal is not only depending on current demand and prices but also relies on system’s former actions.
5.3.2 Constructing and training the BPNN
To develop ANN model in MATLAB environment a three layer ANN with a set of data and target have been defined.To choose more accurate number of neurons for each hidden layer, the number of neuron is iteratively changed in first and second hidden layer. With each combination, the NN is trained more than one time to find the minimum error for entire data [13, 70]. Details of information of training system in ANN network have been shown in Figure 5.4.
103
FIGURE 5.5.TRAINING SYSTEM IN ANNNETWORK
This figure illustrates the information of training system; the training process will stop when the validation failure reached to its maximum number.