BASED
PROTECTIVE SCHEMES
A bit of biology . . . Most important functional unit in human brain a class of cells called
NEURON
Dendrites
Cell Body
Axon Synapse
Dendrites Receive information Cell Body Process information Axon Carries processed information to other neurons Synapse Junction between Axon end and Dendrites of other Neurons
History
The first step toward artificial neural networks came in 1943 when Warre McCulloch, a neurophysiologist, and a young mathematician, Walter Pitts, wrote a paper on how neurons might work. They modeled a simple neural network with electrical circuits.
Cell Body
Axon y1
. . .
Xp wp
V = f(I)
y2 y3 y4
Receives Inputs X1 X2 Xp from environment Inputs fed-in through connections with weights Total Input = Weighted sum of inputs from all sources Transfer function (Activation function) converts the input to output Output goes to other neurons or environment
X2
X3
X4
Input Layer
- Each neuron gets ONLY one input, directly from outside
Hidden Layer
- Connects Input and Output layers
Output Layer
- Output of each neuron directly goes to outside
y1
y2
Output: Y X2 X3
Model: Y = f( X1 , X2 , X3 ) Parameters
# Input Neurons
Example
3
1 2 1 Specified
0.1
0.1
-0.2 Weights
Free parameters
E = ( Yi Vi )2
Back Propagation
Feed forward
Learning of Ann
Supervised learning.
Unsupervised learning.
LINE
CT TRIP COMMAND
RELAY
CVT
FAULT
NO
?
YES
TRIP COMMAND TO BREAKER
Preparation of suitable training data Selection of a suitable ANN structure Training of the ANN Pattern recognition and classification Evaluation of the trained network using test pattern
The target output of the network is 1 in case a high impedance fault , otherwise it is set to 0. The input layer, hidden layer, output layer has 33,6 and 1 nodes respectively The trained network was tested by the data that was never seen by the n/w. Fig.1 shows the current and output of the neural network for an arcing fault.
Fig.2 shows the current and neural network response for an arc Welder.
Conventional distance relaying doesn't produce satisfactory result, when fault occurs at very close distance to the protected zone. Pre-processed voltages and currents are used as inputs to the ANN, which determines the fault location. Finally, a logic unit issues the trip order based on the output of the ANN.
A simulation is done for a transmission line under different
faulted conditions assuming. The 100 km, 220kV transmission line used to train and test the proposed ANN as shown in figure .
Contd
It is assumed that the relay is to protect 80% of the line, i.e., 80 km. Fault data are generated at different distances for various fault types, fault resistances.
Three phase voltages and currents taken and sampled at the rate 1 kHz. These values are scaled to have values between 1 and -1. Training is done by providing the historical data and the simulated data as obtained above.
For faults inside the protection zone, a trip signal will be sent to the circuit breaker through the logic unit as ANN output is 1.
This biologically inspired computation technique , if trained properly it will produce more accurate and reliable outcomes.