CONTROL OF A HEAT EXCHANGER USING NEURAL NETWORK PREDICTIVE CONTROLLER COMBINED WITH AUXILIARY FUZZY CONTROLLER

International Journal of Instrumentation and Control Systems (IJICS) Vol.8, No.2, April 2018
DOI : 10.5121/ijics.2018.8202 11
CONTROL OF A HEAT EXCHANGER USING
NEURAL NETWORK PREDICTIVE
CONTROLLER COMBINED WITH AUXILIARY
FUZZY CONTROLLER
Neethu T R1
, Kalaichelvi P2
, Vetriselvi V3
1,2
Department of Chemical Engineering,
National Institute of Technology Tiruchirappalli,
Tiruchirappalli, 620015, Tamilnadu, India,
3
Department of ICE, National Institute of Technology Tiruchirappalli,
Tiruchirappalli, 620015 Tamilnadu, India.
ABSTRACT
The paper presents an advanced control strategy that uses the neural network predictive controller and the
fuzzy controller in the complex control structure with an auxiliary manipulated variable. The controlled
tubular heat exchanger is used for pre-heating of petroleum by hot water. The heat exchanger is modelled
as a nonlinear system with the interval parametric uncertainty. The set point tracking and the disturbance
rejection using intelligent control strategies are investigated. The control objective is to keep the outlet
temperature of the pre-heated petroleum at a reference value. Simulations of control of the tubular heat
exchanger are done in the Matlab/Stimulant environment. The complex control structure with two
controllers is compared with the conventional PID control, fuzzy control and NNPC. Simulation results
confirm the effectiveness and superiority of the complex control structure combining the NNPC with the
auxiliary fuzzy controller.
KEYWORDS
MPC, NNPC, NNMPC, System identification, Control of plate heat exchanger
1. INTRODUCTION
Model Predictive Control (MPC), a control algorithm which uses an optimizer to solve for the
optimal control moves over a future time horizon based upon a model of the process, has become
a standard control technique in the process industries over the past two decades. In most industrial
applications, a linear dynamic model developed using empirical data is used even though the
process itself is often nonlinear. Linear models have been used because of the difficulty in
developing a generic nonlinear model from empirical data and the computational expense often
involved in using nonlinear models. In this paper, we present a generic neural network based
technique for developing nonlinear dynamic models from empirical data and show that these
models can be efficiently used in a model predictive control framework. This nonlinear MPC

12
based approach has been successfully implemented in a number of industrial applications in the
refining, petrochemical, paper and food industries. Performance of the controller on a nonlinear
industrial process, a polyethylene reactor, is presented.
Model Predictive Control (MPC), a control calculation which utilizes an optimizer to unravel for
the optimal control moves over a future time horizon based upon a model of the process, has
turned into a standard control strategy in the process commercial ventures in the course of recent
decades.
NN have been shown to have good approximation capability for non-linear systems. A large
number of predictive control schemes have been developed based on Multi Layer Perception
(MLP) neural network models since 1990.The key to the successful application of non-linear
predictive controller based on a neural network model is an accurate nonlinear model and an
efficient optimization algorithm. The back propagation learning algorithm, commonly used in
MLP, is essentially a non-linear steepest descent algorithm. The aim of controller design is to
construct a controller that generates control signals that in turn generate the desired plant output
subject to given constraints. Predictive control tries to predict, what would happen to the plant
output for a given control signal. In this way, we know in advance, what effect the control will
have, and by this knowledge the best possible control signal is chosen.
2. SYSTEM IDENTIFICATION
The first stage of model predictive control is to train a neural network to represent the forward
dynamics of the plant. The prediction error between the plant output and the neural network
output is used as the neural network training signal. The process is represented by Fig.1.
Fig.1 Training of Neural Network
The neural network plant model uses previous inputs and previous plant outputs to predict future
values of the plant output. The structure of the neural network plant model is given in Fig.2.
This network can be trained offline in batch mode, using data collected from the operation of the
plant. We can use any of the training algorithms for network training. This process is discussed in
more detail in following sections.

13
Fig.2 Structure of Neural Network Plant Model
3. PREDICTIVE CONTROL
The model predictive control method is based on the receding horizon technique. The neural
network model predicts the plant response over a specified time horizon. The predictions are used
by a numerical optimization program to determine the control signal that minimizes the following
performance criterion over the specified horizon.
= ∑ ( ( + − ( + + ∑ ( ′ (+ − 1 − ′
( + − 2
(1)
where N1, N2, and Nu define the horizons over which the tracking error and the control increments
are evaluated. The u′ variable is the tentative control signal, yr is the desired response, and ym is
the network model response. The ρ value determines the contribution that the sum of the squares
of the control increments has on the performance index.
Fig.3 illustrates the neural network predictive control process. The controller consists of the
neural network plant model and the optimization block.
Fig.3 Illustration of Neural Network Predictive Control Process
The optimization block determines the values of u′ that minimize J, and then the optimal u is
input to the plant. The program generates training data by applying a series of random step inputs
to the Simulink plant model as shown in Fig.4.

14
Fig. 4 Plant Input Output Data’s for Training
4. CONTROL OF PLATE HEAT EXCHANGER USING NEURAL NETWORK
The neural network predictive controller that is implemented in the Neural Network Toolbox
software uses a neural network model of a nonlinear plant to predict future plant performance.
The controller then calculates the control input that will optimize plant performance over a
specified future time horizon. The objective of the controller is to maintain the outlet temperature
of cold fluid by adjusting the mass flow rate of hot fluid mh. Also keep the mass flow rate of cold
fluid as mc = 0.0112kg/s.
The first step in neural network plant model predictive control is to determine the neural network
plant model. For that first we have to find out the mathematical plant model of plate heat
exchanger, based on the energy balance equation of plate heat exchanger. Unsteady-state energy
balances have been used as the basis for the derivation of the mathematical model for the plate
heat exchanger. Assuming U to be constant, the unsteady-state energy balance around the cold
plate is given by:
mcCp Tci-Tco(t + mh(t Cp Thi-Tho(t =McCp
dTco(t)
dt
(2)
And the unsteady-state energy balance around the hot plate is given by:
mhCp Thi-Tho(t + mc(t Cp Tci-Tco(t =MhCp
dTho(t)
dt (3)
Simulink diagram of plate heat exchanger has drawn in Simulink editor based on the energy
balance equation of plate heat exchanger is shown in Fig. 5.
The neural model has been trained using data set obtained from dynamic equations of plate heat
exchanger. The feed forward network with sigmoidal activation function was chosen based on the
trials with different structures of multilayer perception. The lowest error corresponds to 8 neurons
in the hidden layer. Hence it was selected as optimal architecture of ANN. The ANN selected
here consists of 4 neurons in the input layer, 7 neurons in the hidden layer and one neuron in the
output layer. The training algorithm used in this modelling is multi-layer perception algorithm.

15
Fig. 5 Simulink Diagram of Energy Balance Equation of PHE
Fig. 6 Response of Plant Model after Training
The response of plant model after training is complete is shown in Fig.6. The error shown in Fig.
6 is the Prediction error which is the difference between the plant output and the output of the
neural network plant model. This prediction error is used for the training of the neural network.
Neural Network plant model generates the control signal which actually is one step ahead the
prediction of the controller which is depicted as NN output in Fig.6. Neural network predictive
controller has designed by varying controller horizons N2 and Nu, control weighting factor ρ,
search parameter α. The weighting parameter ρ, it multiplies the sum of squared control
increments in the performance function. The parameter α is used to control the optimization. It
determines how much reduction in performance is required for a successful optimization step. We
can select which linear minimization routine is used by the optimization algorithm, and we can
decide how many iterations of the optimization algorithm are performed at each sample time. The
values of controller parameters have chosen for neural network predictive control is shown in
Figure 1.1.

16
Table 1.1 Neural Network Predictive Control Parameters
Controller parameters Values
Cost horizon 8
Control horizon 4
Control weighing factor 0.05
Search parameter 0.001
The designed controller uses a neural network model to predict future plate heat exchanger
responses to potential control signals.
Fig. 1.6 Plant Output and Reference Signal
Fig. 1.6 shows the plant output and reference signal. The result obtained for the random reference
signal proved the tracking ability of controller. Also almost offset free and very close set point
tracking was obtained using NN predictive control strategy. So from the graph, it can be seen that
the neural network predictive controller strategy successfully tracks the random reference signal.
5. CONCLUSION
The result obtained for the random reference signal illustrates and proves the tracking ability of
controller. Also almost offset free and very close set point tracking is obtained using NNMPC
strategy.
In this paper the consideration of dynamic neural models in predictive control for a benchmark
nonlinear process, plate heat exchanger is presented. Neural network controller was used to
maintain the outlet temperature of cold solution (Xanthum gum solution) by adjusting the mass
flow rate of hot water. Non linear auto regressive with exogenous input was recognized utilizing
MLP, and approved on the data produced from the simulation of plate heat exchanger dynamic
equations. This model represents the dynamics of the nonlinear plate heat exchanger and is
utilized as nonlinear predictor in the neural network predictive controller. On analysis of the
response graph (Figure 7.6), it can be seen that the neural network predictive controller strategy
successfully tracks the random reference signal. The outcome got for the random reference signal
outlines and demonstrates the tracking ability of controller. Using the NNMPC strategy almost
offset free and close set point tracking is obtained.

17
REFERENCES
[1] R.W. Serth, T.G. Lestina, 3 e Heat Exchangers, Process Heat Transfer, in: Principles, Applications
and Rules of Thumb, 2014, pp.
[2] F. Peng, G. Cui, Efficient simultaneous synthesis for heat exchanger network with simulated
annealing algorithm, Appl. Therm. Eng. 78 (2015) .
[3] J.J. Klemes, P.S. Varbanov, Heat integration including heat exchangers, combined heat and power,
heat pumps, separation processes and process control, Appl. Therm. Eng. 43 (2012).
[4] T.G. Walmsley, M.R.W. Walmsley, A.S. Morrison, M.J. Atkins, J.R. Neale, A derivative based
method for cost optimal area allocation in heat exchanger networks, Appl. Therm. Eng. 70 (2) (2014).
[5] A. Preglej, J. Rehrl, D. Schwingshackl, I. Steiner, M. Horn, I. Skrjanc, Energyefficient fuzzy model-
based multivariable predictive control of a HVAC system, Energy Build. 82 (2014).
[6] R.F. Garcia, Improving heat exchanger supervision using neural networks and rule based techniques,
Expert Syst. Appl. 39 (3) (2012).
[7] T.A. Tahseen, M. Ishak, M.M. Rahman, Performance predictions of laminar heat transfer and
pressure drop in an in-line flat tube bundle using an adaptive neuro-fuzzy inference system (ANFIS)
model, Int. Commun. Heat Mass Transf. 50 (2014).
[8] J.A. Hernandeza, D. Coloradoa, O. Cort es-Aburtob, Y. El Hamzaouia, V. Velazqueza, B. Alonsoa,
Inverse neural network for optimal performance in polygeneration systems, Appl. Therm. Eng. 50 (2)
(2013).
[9] R.E. Precup, H. Hellendoorn, A survey on industrial applications of fuzzy control, Comput. Ind. 62
(3) (2011).
[10] I. Babuska, R.S. Silva, Dealing with uncertainties in engineering problems using only available data,
Comput. Methods Appl. Mech. Eng. 270.

CONTROL OF A HEAT EXCHANGER USING NEURAL NETWORK PREDICTIVE CONTROLLER COMBINED WITH AUXILIARY FUZZY CONTROLLER

More Related Content

What's hot (18)

Similar to CONTROL OF A HEAT EXCHANGER USING NEURAL NETWORK PREDICTIVE CONTROLLER COMBINED WITH AUXILIARY FUZZY CONTROLLER (20)

More from ijics (20)

Recently uploaded (20)

CONTROL OF A HEAT EXCHANGER USING NEURAL NETWORK PREDICTIVE CONTROLLER COMBINED WITH AUXILIARY FUZZY CONTROLLER