140

Journal of Power Electronics, Vol. 7, No. 2, April 2007

JPE 7-2-7

Development of a Neuro Controller for a Negative Output Elementary Luo Converter

Ramanujam Kayalvizhi , Sirukarumbur Pandurangan Natarajan* and Padmaloshani Palanisamy **

Dept. of Instrumentation Engg., Annamalai University, Chidambaram, India

**

Krishnaswamy Engineering College, Cuddalore, India

ABSTRACT The negative output elementary Luo converter is a newly developed DC-DC converter. Due to the time-varying and switching nature of the above converter, its dynamic behavior becomes highly non-linear. Conventional controllers are incapable of providing good dynamic performance for such a converter and, hence, a neural network is utilized as a controller in this work. The performance of the chosen Luo converter using PI versus neuro controls is compared under load and line disturbances using MATLAB and TMS320F2407 DSP. The results validate the superiority of the developed neuro controller. Keywords: Neuro control, negative output elementary Luo converter

1. Introduction
The negative output elementary Luo converter [1] is a recently developed sub-set of DC-DC converters. A neuro controller is developed in this work to regulate the output voltage of this converter under line and load disturbances since such a controller works well for complex, non-linear and time-variant systems [2-5]. The performance of the converter using PI versus neuro controllers is evaluated using MATLAB as well as TMS320F2407 DSP. The results are presented and analyzed.
Manuscript received October. 2, 2006; revised March. 5, 2007 Corresponding Author: [email protected] Tel: +91-4144-239736, Fax: +91-4144-238275, Annamalai Univ. * Dept. of Instrumentation Engg., Annamalai Univ.,India ** Krishnaswamy Engineering College, India

2. Modeling of a Negative Output Elementary Luo Converter

A negative output elementary Luo converter (Fig.1) performs step-up/step-down conversions from positive input DC voltage to negative output DC voltage. The voltage transfer ratio of the above converter is (k/(1-k)) where k is the duty ratio [1]. The circuit parameters of the chosen converter are listed in Table 1. The circuits (Figs.1(a) and 1(b)) for the switch-on and switch-off modes of the chosen converter are modeled using a state-space approach. At this point, these two models are averaged over a single switching period T using a state-space averaging technique. The state variables are: x1 = iL U= VI x2 = vC Y = VO x3 = iLo x4 = vCO

Development of a Neuro Controller for a Negative Output Elementary Luo Converter

141

Using the equivalent circuits (Figs.1(a) and 1(b)) for the switch-on period (mode 1) and switch-off period (mode 2) of the chosen circuit, the respective state matrices A1,B1,C1 and A2,B2,C2 are:

0 0 0 0 A1 = 0 1 / LO 0 0

0 1/ C 0 1 / CO

1 / LO 1 / RC O 0 0

Fig. 1b

Negative output elementary Luo converter - mode 2

3. Neuro Controller
The input-output data necessary for the off-line training of the neural network have been obtained in the present work using the voltage transfer ratio of the chosen Luo converter. The data set is made sufficiently rich to ensure stable operation since no additional learning will take place after training. A back-propagation algorithm [2-5] is used for training of the created network. The LEARNGDM function which has a gradient descent with momentum weight / bias learning is used in this work. Learning occurs according to the learning parameters: learning rate=0.01 and momentum constant t = 0.9. The neural network weights are: After load disturbances: w1= -4.3396 After supply disturbances 1.0328 -0.9796 -4.1411 w2= 0.0259 -0.5370 b2 = 0.2595 After supply disturbances: b1= 4.5387 -0.6714 -0.7780 -4.1621 0.0682

B1 = [1 / L 0 0 0] T
C1 = [0 0 0 1]
0 A2 = 1 / C 0 0 1/ L 0 1 / LO 0 0 1/ C 0 1 / CO 0 0 1 / LO 1 / RC O

B1 = [0 0 0 0]T
C 2 = [0 0 0 1]

1.0723

Fig. 1

Negative output elementary Luo converter

w1=

4.2698 -1.0492 0.8594 4.1234 0.4589

b1=

-4.3006 0.7756 0.6985 4.0341 -0.0568

w2= -0.0172 b2 = 0.3386

-1.2668

Fig. 1a

Negative output elementary Luo converter - mode 1

MSE is the performance criteria used in this work that evaluates the network according to the mean of the square of the error between the target and computed output. The minimum MSE that can be achieved in this work is 1e-7.

142

Journal of Power Electronics, Vol. 7, No. 2, April 2007

For a back-propagation training algorithm, the derivative of the activation function is needed. Therefore, the activation function selected must be differentiable. The sigmoid function satisfies this requirement and it is the commonly used soft-limiting activation function. It is also quite common to use linear output nodes to make learning easier and using a linear activation function in the output layer does not squash (compress) the range of output. Hence, a bipolar sigmoid activation function and a linear activation function are used for the hidden and output layers, respectively. Trials have been carried out to obtain maximum accuracy with a minimum number of neurons per layer. The feed forward neural network developed consists of one neuron in the input layer, four neurons in the hidden layer and one neuron in the output layer. The optimum number of neurons for the hidden layer is chosen as four (Table 2) since the number of epochs for training the neural network is reduced considerably. The tansig function is found to be better than the logsig activation function for the hidden layer since the logsig function takes (Table 3) approximately 200 more epochs than the tansig function. The input to the neuro controller (Fig.2) is voltage error(e). The output of the controller is the corrected duty ratio(dk).

4. Results and Conclusion

A PI controller with settings Kp=0.4205 and Ti=0.22ms obtained by the Ziegler-Nichols tuning technique has been used for comparison with the developed neuro controller. Fig. 3 shows the closed-loop responses using the above two controllers with k = 0.666 and for set point voltage 20V under sudden changes of 25% of the rated supply (10V) at t = 0.04sec and t = 0.06sec. Fig.4 shows the

corresponding responses under sudden changes of 20% of the rated load. With 50 KHz switching frequency (i.e. 20sec cycle time) of the converter, the neuro controller takes approximately 100 cycles to reject the supply disturbances (Fig.3) since the controller implemented regulates the output voltage within 2msec which is acceptable for power supply applications. Likewise, load disturbances are also effectively rejected (Fig.4). Performance indices listed in Table 4 establish the effectiveness of the simulated neuro controller except for the high overshoots under supply disturbances. The above controllers are also implemented using TMS320F2407 DSP and Figs.5 and 6 show the corresponding responses. Fig.7 shows the DSP based closed loop control scheme for the chosen converter. The converter output voltage is initially scaled down (Fig. 7) suitably by a resistance divider network in the signal conditioning circuit. The output voltage of the divider circuit is fed to the on-chip ADC of the DSP through a high impedance differential amplifier to compute the digital equivalent of the output voltage. This is compared with the reference voltage to compute the error which is processed by the DSP based PI/neuro control algorithms to suitably adjust the duty cycle of the PWM signal. This PWM signal provided by the event manager module of DSP is applied to the MOSFET through the optocoupler and the MOSFET driver. Optocoupler HCPL-4506 provides isolation between the event manager module of the DSP and the gate of the MOSFET. In order to strengthen the pulses, an IR2110 driver is used. Table 4 also portrays the experimental performance evaluation of the controllers. These results also validate the superiority of the neuro controller developed.
Table 1 Circuit parameters of the negative output elementary Luo converter

PI/NEURO CONTROLLER

Fig. 2

Block diagram of PI/neuro controls for negative output elementary Luo converter

Parameters Inductors L, LO Capacitors C, CO Load resistance R Input voltage VI Switching frequency fs Duty ratio k MOSFET Diode

Value 100H 5F 10 10V 50kHz 0.1 to 0.9 IRFP9240 UF5042

Development of a Neuro Controller for a Negative Output Elementary Luo Converter

143

Table 2

Choice of hidden neurons

Number of hidden neurons 1 2 3 4 5 6 7 8

Number of epochs for training the network Performance goal not met Performance goal not met 400 200 300 400 450 480

Table 3

Choice of bipolar sigmoid transfer function as activation function for hidden layer

Number of hidden neurons 1 2 3 4 5 6 7 8

Number of epochs with tansig activation function Performance goal not met Performance goal not met 400 200 300 400 450 480

Number of epochs with logsig activation function Performance goal not met Performance goal not met 500 400 480 510 550 480

Table 4

Performance evaluation of controllers for negative output elementary Luo converter Supply disturbance
Supply increase (25%) Supply decrease (25%) Peak over shoot (% ) 17.5 Settling time (msecs) 3.4 Load increase (20%) Peak over shoot (% ) 25 Settling time (msecs) 1

Load disturbance
Load decrease (20%) Peak over shoot (% ) 20 Settling time (msecs) 1

RESULTS

CONTROLLERS

Peak over shoot (% ) 20

Settling time (msecs) 3.9

PI SIMULATION NEURAL PI EXPERIMENTAL NEURAL

24.75 20 15

2 10 5

19.8 15 2.5

2 10 1

1.5 20 5

0.5 11 6

1.16 15 10

0.5 15 3

144

Journal of Power Electronics, Vol. 7, No. 2, April 2007

0
-5

-5 s t l o v n i e g a t l o V t u p t u O

-10

-15

s t l o v n i e g a t l o v t u p t u O

-10

-15

-20

-20

-25

-25

-30

0.01

0.02

0.03

0.04

Time in secs

0.05 Time in secs

0.06

0.07

0.08

0.09

0.1

0.01

0.02

0.03

0.04

0.05 Time in secs

0.06

0.07

0.08

0.09

0.1

(a) PI controller
Closed loop response of Negative output Elementary Luo Converter under 25% of supply disturbance 0

Time in secs (a)PI controller

0 0 Closed loop response of Negative output Elementary Luo Converter under load disturbance Closed loop response of Negative output Elementary Luo Converter under load disturbance

-5

-5 -5

-10 s t l o v n i e g a t l o V t u p t u O

-15

-20

s t l o s v t l n o i v e g n i a t e l g o a v t l t o u v p t t u u p O t u O

-10 -10

-15 -15

-20
-25

-20

-30 0

-25
0.01 0.02 0.03 0.04 0.05 Time in secs 0.06 0.07 0.08 0.09 0.1

-25

0.01 0.01

0.02 0.02

0.03 0.03

0.04 0.04

Time in secs (b) Neuro controller

Time in secs

0.05 0.06 Time in secs 0.06 0.05

0.07 0.07 0.08

0.08 0.09

0.09

0.1

(b) Neuro controller

Fig. 3

Closed loop responses with sudden disturbances of 25% of rated supply using Matlab

Fig. 4

Closed loop responses with sudden disturbances of 20% of rated load using Matlab

Output voltage in volts

Volts/div=10 Time /div=10

Output voltage in volts

Volts/div=10 Time /div=20

Time in msec (a) PI controller 0 Output voltage in volts Volts/div=5 Time /div=5

Time in msec (a) PI controller Volts/div=10 Time /div=10

Output voltage in volts

Time in msec (b) Neuro controller

Time in msec (b) Neuro controller

Fig. 5

Closed loop responses with sudden disturbances of 25% of rated supply using DSP DSP TMS320F2407

Fig. 6

Closed loop responses with sudden disturbances of 20% of rated load using DSP

Development of a Neuro Controller for a Negative Output Elementary Luo Converter

145

Event

Event manager

To gate of MOSFET Optocoupler and driver circuits

PI / neuro control algorithms

Signal conditioning circuit

Converter output

Dual 10 -bit ADC

Fig. 7

Block diagram of PI / neuro controllers for negative output elementary Luo converter using DSP

References
[1] F.L. Luo, Negative output Luo converters: voltage lift technique, IEE Procs. Electr. Power Appl., vol. 146 (2), pp. 208-224, 1999. [2] H.C.Chan, K.T.Chan and C.C.Chan, A neural network controller for switching power converters, IEEE Power Electronics Special Conference Record:0-7803-1234-0/93, 1993, pp. 887-892. [3] R.Kayalvizhi, S.P.Natarajan, P.Padmaloshani and R.Vijayarajeswaran, Development of neuro controller for negative output re-lift Luo converter, IEEE Conference Record:0-7803-9296-5/05, PEDS 05, Dec. 2005, Malaysia, pp. 1520 1524. [4] R.Kayalvizhi, S.P.Natarajan and P.Padmaloshani, Development of neuro controller for negative output self-lift Luo converter, IEEE Conference Record: 0-7803-9514-X/06, ICIEA 2006, May 2006, Singapore , pp. 517 522. [5] Allan Insleay and Geza Joos, A neural network based approach to the regulation of DC/DC buck converters, IEEE Conference Record: 0-7803-1443-3/93, CCECE/CCGEI 93, 1993, pp. 214-217. S.P.Natarajan was born in 1955 in Chidambaram. He obtained the B.E. (Electrical and Electronics) and M.E. (Power Systems) degrees in 1978 and 1984, respectively, from Annamalai University and then the Ph.D in Power Electronics from Anna University, Chennai in 2003. He is currently a Professor in Instrumentation Engineering at Annamalai University where he has put in 26 years of service. His research interests are in modeling and control of DC-DC converters and multiple connected power electronic converters, control of Permanent Magnet Brushless DC Motors, embedded control for multi level

inverters and matrix converters etc. He is a life member of the Instrument Society of India and Indian Society for Technical Education. He has completed an AICTE R & D project titled Investigations on Controllers for Permanent Magnet Brushless DC Motor. R.Kayalvizhi was born in Chidambaram in 1963. She obtained the B.E. (Electronics and Instrumentation) and M.E (Power Systems) degrees from Annamalai University with distinction in 1984 and 1988, respectively. She is presently an Assistant Professor in the Department of Instrumentation Engineering, Annamalai University where she has put in a total service of 21 years. She completed Ph.D in the same Department on intelligent control strategies for Luo converters. Her research interests are inDC-DC converters: modeling, simulation and implementation of intelligent control strategies. She is a life member of Indian Society for Technical Education. P.Padmaloshani was born in 1972. She obtained the B.E (Electronics and Communication) and M.E (Process Control & Instrumentation) degrees from Bharathiyar & Annamalai Universities in 1993 and 2005, respectively. She is presently a Lecturer in Krishnaswamy Engineering College, Cuddalore, India. Her research interests are in DC-DC converters: modeling, simulation and implementation of neuro control strategies.