DESIGN OF GOVERNOR CONTROLLER USING ARTFICIAL NEURAL NETWORK

Academic Open Internet Journal

ISSN 1311-4360

www.acadjournal.com

Volume 20, 2007

ARTIFICIAL NEURAL NETWORK BASED

DESIGN OF GOVERNOR CONTROLLER

Dr. N. Rengarajan¹ C.S. Ravichandran² Dr. S. Palani³

¹ Principal, Vivekanandha Institute of Engineering&Technology, Tiruchengode, India – 637205

² Asst. Professor / Electrical and Electronics Engineering, Sri Ramakrishna College of Engineering, Coimbatore – 641022

E-mail: eniyanravi@gmail.com / eniyanravi@yahoo.co.in

³ Dean and Professor, Electronics and Communication Engineering, Sona College of Technology, Salem, India - 636005

ABSTRACT

This paper presents the design of Artificial Neural Network (ANN) based PID controller, to realize fast governor action in a power generation plant. The design technique is applied to single area, two area systems, to tune the parameters of the PID controller. Feed forward neural network architecture is chosen for the design of controller, which is trained by a popular back propagation algorithm. Performance of the proposed ANN based controller, is compared with conventional integral and PID controllers, through dynamic simulation. It is observed that ANN based controller provides better performance.

1. INTRODUCTION

Large scale power systems are normally composed of control areas or regions representing coherent groups of generators. Area load changes and abnormal conditions lead to mismatches in frequency and scheduled power interchanges between areas. These mismatches have to be corrected by Governor control, which is defined as the regulation of the power output of generators within a prescribed area [3].

The key assumptions in the classical Governor control problem are [1, 6]:

i. The steady state frequency error following a step load change should vanish. The transient frequency and time errors should be reduced.

ii. The static change in the tie power following a step load in any area should be zero, provided each area can accommodate its own load change.

iii. Any area in need of power during an emergency should be assisted from other areas.

An integral controller provides zero steady state frequency deviation (∆ω) but it exhibits poor dynamic performance [1]. To improve the transient response, various control techniques, such as linear feedback, optimal control and variable structure control have been proposed [5]. Adaptive controllers with self adjusting gain settings have also been proposed for the LFC problem [4, 7].

In this paper, the ANN is applied to self tune the parameters The performances of the PID type controller with fixed gain, Conventional integral controller, ANN based PID controller have been compared through MATLAB Simulation results.

2. SINGLE AND TWO AREA POWER SYSTEM USING PID CONTROLLER:

In Single area system, generation and load demand of one domain is dealt. Any load change within the area has to be met by generators in that area alone through suitable governor action. Thus we can maintain the constant frequency operation irrespective of load change.

Power system parameters taken for the design of the Governor controller are enlisted in Table 1.

Table 1: Power System Parameters

Parameters	Single area system	Two area system
		Area 1	Area 2
Turbine time constant(T_CH) Governor time constant (T_G) Generator angular momentum(M) Governor speed regulation(R) Load change for frequency change of 1 % D = ∆p / ∆ω Turbine rated output (taken as base value) Tie line stiffness (T) Coefficient	0.5 s 0.2 s 10 MJ rad/s 0.05 0.8% 0.8 250 MW -	0.5s 0.2s 10 MJ rad/s 0.05 0.6% 0.6 250 MW	0.6s 0.3s 8 MJ rad/s 0.0625 0.9% 0.9 250 MW
		2 rad/s/Ω

Single area system with governor control using PID controller is shown in figure 1. Modeling of single and two area systems are based on transfer function approach [2].

Figure. 1 Single area system with governor control using PID controller

Two area system with governor control using PID controller is shown in figure 2. Tie line control system must use two pieces of information: the system frequency and the net power flowing in or out over the tie lines.

(i). If frequency decreased and net interchange power leaving the system increased, a load increase has occurred outside the system.

(ii). If frequency decreased and net interchange power leaving the system decreased, a load increase has occurred inside the system

Figure. 2 Two area system with governor control using PID controller

Normal PID controller will work with fixed constants K_P, K_I, K_D. The parameter values are chosen based on the Routh Hurwitz criterion and by comparing the performance of conventional integral controller. Good performance is achieved for all load disturbances ∆P_L compared to conventional integral controller.

2.1. Drawback of PID Controller

The error signal fed to the PID controller is divided into two states: Steady state and Transient state. As the PID controller parameters K_P , K_I , K_D are fixed, same values have to be maintained for both transient and steady states. This poses a restriction in choosing the parameter values of the PID controller, as it is limited by the system specifications. However choosing higher values for transient period and lower values in steady state can solve the problem. This technique is incorporated in generating parameters of the PID controller. Compared to the steady state, transient state exists for a shorter interval.

To improve the performance characteristics (Frequency Deviation ∆ω_, Tie line power flow ∆P₁₂) of the power system the value of the constants K_P, KI, K_Dcan be increased in the transient period and low in steady state.

3. ADAPTATION OF ARTIFICIAL NEURAL NETWORK

In a system, if inputs and the corresponding targets are identified, then we can implement the Artificial Neural Network (ANN) for the input – target pair. ANN is computationally simple, reliable, model free system. One of the main advantages of ANN is, desired output can be obtained for even untrained data within the input range.

In this paper training is carried out using NNTOOL box in MATLAB software version 6.1. NNTOOL method provides the facility to train through one of the methods Say conjugate gradient method, Levenberg-Marquardt method for back propagation. In this paper Levenberg-Marquardt method is employed for it’s superiority in convergence.

In the neural network developed (Figure. 3) TANSIG is employed as transfer function in the hidden layer and PURELIN in the output layer. Then the obtained weights and biases are chosen as the initial weights and biases.

Input hidden output

layer layer layer

Figure. 3 Neural network for the design of ANN based PID controller

3.1. Training Procedure:

Import inputs to the network & corresponding targets either from current workspace or from a file.

Step1: Choose new network icon in the box to create a new neural network.

Step 2: Creation of New Network In this box we can choose the number of layers, number of neurons in each layer and input ranges .

Step 3: Initialization of the network

Step 4: Simulation of the neural network

Step 5: Training the neural network

Step 6: Adaptation of the neural network with trained data

Step 7: Required weights and biases for the neural network

3.2. Design of ANN Controller in Tuning PID Controller Parameters

The range over which error signal is in transient state, is observed. Corresponding values of the proportional, integral and derivative constants are set. This set is kept as target. Range of error signal is taken as the input. This input – target pair is fed and new neural network is formed using “nntool” in the MATLAB Simulink software. Updated weights and biases are given to a fresh neural network. Now the neural network is ready for operation.

The error signal is given as input to the neural network using MATLAB function. Desired target for each input value is obtained. The fresh neural network is written as program and is incorporated in the MATLAB function tool, in simulink diagram. Thus for each error signal fed as input, trained PID controller parameters K_P, K_I, K_Dare given back as output to called MATLAB function tool in simulink diagram.

As the neural network developed is purely dependant on the area control error signal, the network trained can be used for both single and two area systems. Further as the neural network is independent of the time instant, the trained network is more reliable for all disturbances which may occur at different time instances.

For any load change, the required change in generation, called the area control error or ACE, represents the shift in the areas generation required to restore frequency and net interchange to their desired values. Maximum and minimum values of ACE occur in transient state and steady state respectively.

Parameter Comparison of PID controller (figure. 4) and ANN based PID controller (figure.5) for maximum ACE and minimum ACE is carried out and listed in table 2.

Figure. 4 PID controller Figure. 5 ANN based PID controller

Table 2: Parameter comparison of PID controller and ANN based PID controller

Parameters	PID controller		ANN based PID controller
	ACE (min)	ACE (max)	ACE (min)	ACE (max)
K_P	1 1		0.95 1.35
K_I	1.25 1.25		1.2 1.6
K_D	1 1		0.95 1.35

4. DYNAMIC SIMULATION RESULTS

Performance comparison of ANN controller, PID controller, Conventional integral controller for single area system and two area system for different load disturbances (∆P_L) are carried out and the results are shown in figures 6 to 8.

4.1. Single Area System: Frequency Deviation

∆P_L= 1 pu ∆P_L= 0.5 pu

1

2

3

3

∆ω

1

2

3

time time

1. ANN controller; 2. PID controller; 3. Conventional integral controller

Figure. 6 Comparison of ANN controller, PID controller & conventional controller

4.2. Two Area System:

Frequency Deviation:

∆P_L1= 1 pu , ∆P_L2= 0 pu ∆P_L1= 0.3 pu , ∆P_L2= 0.7 pu

        ∆ω₁

2

3

1

1

2

3

1

2

3

   ∆ω₂

1

2

3

time time

1. ANN controller; 2. PID controller; 3. Conventional integral controller

Figure. 7 Comparison of ANN controller, PID controller & conventional controller

∆P_L1= 1 pu, ∆P_L2= 0 pu

Tie Line Power Flow

∆P_L1= 0.3 pu, ∆P_L2= 0.7 pu

∆P₁₂

1

2

3

3

2

1

time time

1. ANN controller; 2. PID controller; 3. Conventional integral controller

Figure. 8 Comparison of ANN controller, PID controller & conventional controller

5. CONCLUSIONS

Comparison of responses with conventional integral controller & PID controller show that the neural-network controller has quite satisfactory generalization capability, feasibility and reliability, as well as accuracy in both single area system & two area system.
Simulations of the networks are carried out for different load changes and simulation results are compared highlighting the performance of ANN controller.
In this paper PID controller parameters are continuously adjusted according to the change in area-control error.
An explicit expression for adjusting the controller parameters ensures that the proposed technique is feasible for implementation in power generation systems.

REFERENCES

[1] Elgerd OI. Electric energy systems theory: An introduction. McGraw-Hill; 1971

[2] Hadi saadat , Power System Analysis Tata McGraw-Hill edition 2002.

[3] Jaleeli N, VanSlyck LS, Ewart DN, Fink LH, Hoffmann AG. Understanding automatic generation control. IEEE Trans Power Systems 1992; 7(3):1106–12.

[4] Pan CT, Liaw CM. An adaptive controller for power system load-frequency control. IEEE Trans Power Systems 1989; 4(1):122–8.

[5] Reddoch P, Julich TT, Tacker E. Model and performance functional for load frequency control in interconnected power systems. In: IEEE Conf Decision and Control, 1971.

[6] Talaq J, Al-Basri F. Adaptive fuzzy gain scheduling for load frequency control. IEEE Trans Power Systems 1999; 14(1):145–50.

[7] Yamashita K, Miyagi H. Two variable self-tuning regulators for load frequency control system n of voltage on load demand. IEEE Proc D 1991; 138(2).

Technical College - Bourgas,