Docstoc

PERFORMANCE EVALUATION OF ANN BASED PLASMA POSITION CONTROLLERS FOR ADITYA TOKAMAK

Document Sample
PERFORMANCE EVALUATION OF ANN BASED PLASMA POSITION CONTROLLERS FOR ADITYA TOKAMAK Powered By Docstoc
					 International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
INTERNATIONAL JOURNAL OF ELECTRICAL ENGINEERING
 6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME
                           & TECHNOLOGY (IJEET)
ISSN 0976 – 6545(Print)
ISSN 0976 – 6553(Online)
Volume 4, Issue 2, March – April (2013), pp. 324-329
© IAEME: www.iaeme.com/ijeet.asp
                                                                           IJEET
Journal Impact Factor (2013): 5.5028 (Calculated by GISI)
www.jifactor.com                                                        ©IAEME



  PERFORMANCE EVALUATION OF ANN BASED PLASMA POSITION
           CONTROLLERS FOR ADITYA TOKAMAK

      J. Femila Roseline1, Jigneshkumar J.Patel2, J.Govindarajan3, N.M.Nandhitha4,
                                     B.Sheela Rani5
       1
         Asst.Professor, Dept. of Electrical and Electronics Engg., Sathyabama University,
                   Jeppiaar Nagar, Old Mahabalipuram Road, Chennai 600 119
                 2
                   Engineer-SD, Electronics Group, Institute of Plasma Research.
                        3
                          Associate Professor-II,Institute of Plasma Research,
      4
        Professor & Head, Dept. of Electronics and Communication Engg., Jeppiaar Nagar,
                             Old Mahabalipuram Road, Chennai 600 119,
                      5
                        Vice Chancellor, Prof. Electronics & Instrumentation,
                              Sathyabama University, Chennai 600 119



  ABSTRACT

          In Aditya tokamaks, electrical energy is generated through plasma confinement in the
  torroidal chamber. The amount of energy generated is directly related to the confinement of
  the plasma within the chamber. Also if the plasma hits the limiters or the walls it leads to
  plasma disruption. Extensive research has been done to develop controllers for confining the
  plasma within the chamber. However these techniques had inherent limitations as they are
  either linear models or fuzzy based controllers. The Fuzzy based controllers are strongly
  dependent on the membership functions. Hence in this paper Artificial Neural Network based
  classifiers are developed to overcome the limitations of the existing system. GRNN, RBN
  based networks were developed and the performance is evaluated with that of the already
  developed BPN based controller. It is found that BPN based controllers provide higher Signal
  To noise ratio than the other controllers.

  Keywords : Tokamaks, Plasma Position, plasma Confinement, radial position, plasma
  current, BPN, voltage, RBN, GRNN;



                                              324
 International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
 6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME

I.     INTRODUCTION

         In Aditya Tokamaks, hot plasma is contained by a magnetic field which keeps it away
 from the machine walls. The combination of two sets of magnetic coils known as toroidal and
 poloidal field coils creates a field in both vertical and horizontal directions, acting as a
 magnetic toroidal chamber. The performance of the machine is dependent on the density of
 the plasma, position of the plasma within the chamber and the time duration for which the
 plasma is stabilized. From the literature, it is found that confinement of plasma within the
 chamber yields better results in Aditya Tokamak. Plasma position control is basically a non-
 linear process. However the initial controllers were linear PID controllers. Performance has
 reduced as the constants can not be fixed accurately. Fuzzy based controller is the first non-
 linear controller used for plasma position controller. However the performance of the system
 is strongly dependent on the membership functions, defuzzification rules and the knowledge
 base. Also certain assumptions made in Fuzzy based controllers are unrealistic in nature.
 Hence it is necessary to develop an intelligent non-linear based controller that adapts to the
 real time conditions and provides the results. General Recurrent Neural Network (GRNN)
 and Radial Basis Function Network (RBFN) have been developed for controlling the plasma
 current in Adithya Tokamak. The network accepts radial position and current as inputs and
 predicts the stabilization voltage. The inputs and output variables for training and testing are
 obtained from Aditya RZIP model.
    The paper is organized as follows: Section II provides the related work. Section III gives an
 overview of the neural networks chosen for developing plasma position controllers. The
 proposed methodology is explained in section IV. Section V is about results and discussion
 and Section VI concludes the work.

     II.   RELATED WORK

         D. Wroblewski et al (1997) trained a neural network which combines signals from a
 large number of plasma diagnostics and estimated the high- beta disruption boundary in the
 DIII-D tokamak. The proposed neural network maps the disruption boundary throughout most
 of the discharge. It can predict the high- beta disruption boundary on a time-scale of the order
 of 100 ms (much longer than the precursor growth time), which makes this approach ideally
 suitable for real time application in a disruption avoidance scheme [1]. J.V. Hernandez et al
 (1996) described the use of neural network algorithms for predicting minor and major
 disruptions in tokamaks by analyzing disruption data from the TEXT tokamak with two
 network architectures. Fluctuating magnetic signal was extrapolated based on L past values of
 the magnetic fluctuation signal measured by a single Mirnov coil [2]. A. Vannucci et al
 (1999) used a neural network is trained with one disruptive plasma discharge and is validated
 using soft X ray signals as input. After training they used the same set of weights to find out
 the disruptions in two other plasma discharges and they observed that neural network is able to
 predict the disruptions more than 3ms in advance when compared to the previously used
 Mirnov coil [3]. Barbara Cannas et al proposed dynamic neural networks to predict the
 plasma disruptions in a nuclear fusion device. Dynamic neural networks act as filters, which
 predict one step ahead the value of diagnostic signals acquired during a plasma pulse [4]. A.
 Sengupta et al (2002) developed two modified neural network techniques which are used for
 indentifying equilibrium plasma parameters of the Superconducting Steady State Tokamak I
 from external magnetic measurements. They used a multi network system which is connected
 in parallel. By using this double neural network the accuracy of the recovered result is better

                                               325
 International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
 6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME

 than the conventional method. They fed the reduced and transformed input set rather than the
 entire set, into the neural network input and called that as the principal component transformation-
 based neural network [5]. A.B. Trunov (2004) developed several neural network approximators
 which were computed on the basis of training data and analyzed their performance. It was found
 that neural networks have better generalization properties than their linear counterparts, and can
 therefore produce reasonably good prediction even with severely reduced input datasets [6].

III. OVERVIEW OF RBFN AND GRNN

         Radial Basis Function Neural network (RBFN) consists of three layers: an input layer, a
 hidden (kernel) layer, and an output layer. The nodes within each layer are fully connected to the
 previous layer. The input variables are each assigned to the nodes in the input layer and they pass
 directly to the hidden layer without weights. The transfer functions of the hidden nodes are RBF.
 An RBF is symmetrical about a given mean or center point in a multidimensional space. A
 Generalized Recursion Neural Network (GRNN) is a variation of the radial basis neural
 networks, which is based on kernel regression networks. A GRNN does not require an iterative
 training procedure as back propagation networks. It approximates any arbitrary function between
 input and output vectors, drawing the function estimate directly from the training data. In
 addition, it is consistent that as the training set size becomes large, the estimation error
 approaches zero, with only mild restrictions on the function.

 IV.   RBFN AND GRNN BASED PLASMA POSITION CONTROLLERS

         RBFN is chosen with two neurons in the input layer and one neuron in the output layer.
 As the architecture of GRNN can not be modified the general four layered GRNN was chosen for
 plasma position control. The exemplars are generated from Aditya RZIP model. Different sets of
 exemplars are used for training and testing the neural network. A set of exemplars used for
 training is shown in Table 1. The input parameters are the radial position and plasma current and
 the output parameter is the plasma stabilization voltage. In order to prevent overflow the values
 are normalized.

                   Table I.Exemplars Used For Trainiga The Neural Network

                   Input Parameters                          Output Voltage
            Plasma current Radial position             Desired output voltage (Va )
               (Ip) in A         (Rp)                           in Volts
               60.0186         0.7498                            0.0095
               60.1467         0.7483                            0.0759
               60.4243         0.7443                            0.2376
               61.0378         0.7303                            0.7360
               61.5420         0.7003                            1.6693
               60.8311         0.6932                            2.3222
               60.2984         0.6891                            3.3453
               60.0457         0.6872                            5.5227
               60.0155         0.6873                            4.5752
               60.0188         0.6875                            2.4668


                                                 326
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME

 V.   RESULTS AND DISCUSSION

        The developed ANN is trained with 75 values. With the adapted weight, the ANN is
tested using another set of 70 values. The relationship between the actual and desired output for
the corresponding input parameters is shown in Table 1. From the last two columns of Table 1
the actual and desired values are nearly same. The relationship between actual and desired
values for different ANN is shown in figure 1. The performance of the ANN bases plasma
position controllers are tabulated in Table 2. Performance metrics are shown in Table 3. The
comparison of Signal to Noise Ratio for GRNN, RBN and BPN is shown in Figure 2.




           Figer 1.Relation between desired and actual outputs for different ANN


          TABLE II :PERFORMANCE OF ANN BASED PLASMA POSITION CONTROLLERS

            Input Parameters                           Output Parameters
                        Radial       Desired output        Actual output voltage (Va)
      Plasma current
                       position      voltage (Va) in
         (Ip) in A                                     GRNN         RBN           BPN
                         (Rp)             Volts
          0.9751       0.99999           0.0007        0.0010       0.0101      0.00075
          0.9773        0.9977           0.0113        0.0113      -0.0086      0.0112
          0.9918        0.9737           0.1092        0.0944       0.0995      0.4468
          0.9752        0.9167           0.2279        0.4791       0.5007      -0.1627
          0.9752        0.9168           0.4188        0.4695       0.4966      0.4209
          0.9752        0.9187           0.4561        0.4558       0.4667      0.4562
          0.9752        0.9194           0.4673        0.4672       0.4682      0.4673
          0.9752        0.9207           0.4892        0.4892       0.4810      0.4893
          0.9752        0.9213           0.5001        0.4999       0.4898      0.5003
          0.9752        0.9220           0.5110        0.5092       0.4982      0.5111
          0.9752        0.9226           0.5218        0.5151       0.5053      0.5053




                                              327
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME

                                 TABLE III.PERFOMANCE METRICS

                              Parameters      GRNN       RBN     BPN
                            Root       Mean
                                              0.0896     0.1002 0.0863
                            Square Error
                            Standard
                                              0.0827     0.0854 0.0816
                            Deviation
                            Signal to Noise
                                              4.8633     4.4095 5.2915
                            Ratio




                        6
                                                                 5.2915

                        5        4.8633
                                                4.4095

                        4
                  SNR




                        3


                        2


                        1


                        0
                               GRNN            RBN              BPN




                        Figer 2.SNR comparison for GRNN, RBN and BPN

 VI.   CONCLUSION AND FUTURE WORK

        GRNN and RBFN based plasma position controllers were developed successfully.
Exemplars were generated using Aditya RZIP model. The performance of these networks is
compared with that of BPN. Though GRNN and RBFN are best suited for predicting the
plasma stabilization voltage from incomplete set of exemplars, BPN based approach provides
better results in terms of Signal to Noise ratio and root mean square. As the exemplars data is
generated from Aditya RZIP model, the data is linear in nature. Hence it is necessary to test
and train the neural network with the plasma discharge shots obtained from Aditya Tokamak.
Also the feasibility of Neuro Fuzzy controller for plasma position control should also be
exploited.

REFERENCES

[1] D. Wroblewski, G.L. Jahns and J.A. Leuer, ‘Tokamak disruption alarm based on a neural
    network model of the high- beta limit ’, Nuclear Fusion, Vol. 37, Number 6, Issue 6 (June
    1997)
[2] J.V. Hernandez, A. Vannucci, T. Tajima, Z. Lin, W. Horton and S.C. Mc Cool, ‘Neural
    network prediction of some classes of tokamak disruptions ’, Nuclear Fusion, Vol. 36,
    Number 8, Issue 8 (August 1996).


                                              328
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 4, Issue 2, March – April (2013), © IAEME

[3]   A. Vannucci*, K.A. Oliveira* and T. Tajima, ‘Forecast of TEXT plasma disruptions using
      soft X rays as input signal in a neural network’, Nuclear Fusion, Vol. 39, Number 2,Issue
      2 (February 1999)
[4]   Barbara Cannas, Alessandra Fanni and Augusto Montisci ‘Dynamic Neural Networks for
      Prediction of Disruptions in Tokamaks’
[5]   A. Sengupta and P. Ranjan, ‘Modified neural networks for rapid recovery of tokamak
      plasma parameters for real time control’, Review of Scientific Instruments , Volume 73,
      Issue 7, American Institute of physics, 2002.
[6]   A. B. Trunov ‘Design of the Real Time neural network control system for the DII-D
      plasma fusion reactor’, Progress in Electromagnetic research symposiun, Pisa, Italy,
      March 28-31, 2004.
[7]   Shaikh Abdul Hannan,R.R.ManzaR.J.Ramteke, ‘Generalised Neural Network and Radial
      Basis function for Heart Disease Diagnosis’, International Journal of Computer
      Applications, Volume 7, Number.13, pp 0975-8887, october 2010.
[8]   Chaitrali S. Dangare and Dr. Sulabha S. Apte, “A Data Mining Approach for Prediction
      of Heart Disease using Neural Networks”, International journal of Computer Engineering
      & Technology (IJCET), Volume 3, Issue 3, 2012, pp. 30 - 40, ISSN Print: 0976 – 6367,
      ISSN Online: 0976 – 6375.
[9]   Dr.Muhanned Alfarras, “Early Detection of Adult Valve Disease–Mitral Stenosis using
      The Elman Artificial Neural Network”, International Journal of Computer Engineering &
      Technology (IJCET), Volume 3, Issue 3, 2012, pp. 255 - 264, ISSN Print: 0976 – 6367,
      ISSN Online: 0976 – 6375.




                                              329

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:4/26/2013
language:
pages:6