(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No.8, 2010
There are different ways to define network conditions for point-to-point (from the central control to a specific remoteunit). Two of the most popular network measures are the point-to-point network throughput and maximal delay boundof the largest data. One factor of interest is the sampling time.To keep the illustration simple, the remote unit receives thedata sent from the central controller as u
(t), which can bemathematically expressed as
t ut u
is the time delay to transmit the control signal u
(t)from the central controller to the remote unit. The remote unitalso sends the sensors signals y
(t) of the remote system back to the central controller y
(t), and these two signals are relatedas
t yt y
is the time delay to transmit the measured signalfrom the remote unit to the central controller. There are also processing delays as τ
, at the central and remote unit,respectively which could be approximate small constants or even neglected because these delays are usually smallcompared to τ
.The functions of network variables such as the network throughput, the network management/policy used, the typeand number of signals to be transmitted, the network protocolused, and the controller processing time, and the network traffic congestion condition are taken as the current network conditions n(t) and let z
be a time delay operator whichdefines the signals as
t n z ut u
t c R
t n z yt y
(8)In this paper, we have chosen sampling time as 0.5 ms andsimulations are done.III.
NCSIn this session the proposed Neural Network Controller andFuzzy Logic Controller as the central controller is describedand the results are compared with the PID controller.
Neural Network Controller
The proposed scheme utilizes the neural-network NARMA-L2 Controller. The Neural Network Controller isdesigned to take the error as the input and computes the outputstabilizing signal depending on the input error signal. The block diagram of Neural Network Controller for NCS isshown in Fig. 4.The NARMA-L2 controller, a multilayer neural network has two steps involved as system identification stage andcontrol design stage. In system identification stage, a neuralnetwork model of the plant which has to be controlled isdeveloped and in later stage the neural network plant modelhas been designed to train the controller. The ANN plantspecification has been shown in Table 2.
Figure 4. Neural Network Controller for NCS.TABLE
No. of Inputs 3 No. of Outputs 2 No. of Hidden Layers 2 No. of Training Samples 1000 No. of Training Epochs 200
The error signals are trained for number of epochs by usingthe NARMA-L2 controller and the control signal are generatedfor any challenges in the network.
Fuzzy Logic controller
In general, fuzzy logic control is used for the control of a plant where the plant modeling is difficult. For such systemsthat are difficult to model, fuzzy logic controller has beensuccessful by Mamdani. The basic principle of fuzzy logic liesin the definition of a set where any element can belong to a setwith a certain degree of membership. Using this idea, theknowledge of an expert can be expressed in a relatively simpleform and the inference for given inputs can be implementedvery efficiently. Due to these advantages, fuzzy logic controlis an attractive method for NCS whose modeling is verydifficult because of the stochastic and discrete nature of thenetwork. Fig. 5 shows the structure of FLC for a single inputsingle output plant. In Fig. 5 r(t) is the reference input, y(t) isthe plant output, e(t) is the error signal between the referenceinput and plant output and u
(t) is the control signal.The FLC consists of three parts as 1) Fuzzifier thatconverts the error signal into linguistic values, 2) Inferenceengine that creates the fuzzy output using fuzzy control rulesgenerated from expert experience and 3) Defuzzifier thatcalculate the control input to the plant from the inferredresults. The input and output signals to the FLC are error signal e(t) and control signal u
(t) respectively. In this paper,the trapezoidal fuzzy members are selected for membershipfunctions. Three fuzzy linguistic variables, i.e., Small,Medium and Large are defined. The coefficients of themembership function depend upon the set point and aredetermined by several trial and error experiments with the plant without the network. In order for faster execution of thefuzzy logic controller, the Mamdani’s min-max inferencemethod and the central average defuzzifier are used.The rules used in this paper are asIf e(t) is small then u
(t) is small