what is back-propagation neural network. how to solve using matlab

Step 1 The network synaptic weights are initialized to small random values. Step 2 From the set of training input/output pairs, an input pattern is presented and the network response is calculated. Step 3 The desired network response is compared with the actual output of the network, and all the local errors to be computed. Step 4 The weights preceding each output node are updated according to the following update formula: Δwij (t)= ¼ η (δ)j Oi+ α Δwij (t-1)

Step 5 The cycle (step 2 to step 4) is repeated until the calculated outputs have converged sufficiently close to the desired outputs or an iteration limit has been reached.

Transfer function used in the back-propagation network is known as ‘sigmoid function’, which is shown below.

f ðsÞ ¼ 1= 1 /(1+e^-s)

where s is the sum of the node inputs. Clearly the node output will be confined to the range 0<f(s)<1.

A practical region for the data is chosen to be [0.1 ....0.9]. In this case each input or output parameter X is normalised as Xn before being applied to the neural network, according to the following equation, shown below. Xn = (0.8 ((X - Xmin)/ (Xmax - Xmin)))+ 0.1 Table 1 Data collected from steel foundry for the present investigation S.No Name of casting Mould properties Chemical composition of charge metal Metal temp (°C) Remarks GCS (N/m2) GSS (N/m2) Permeability Moisture ( % ) C ( % ) Mn ( % ) Si ( % ) S ( % ) P ( % ) Cr ( % ) 1 Side frame 387100 92120 210 3 0.28 0.88 0.36 0.022 0.02 0.021 1596 2 Side frame 363580 72520 210 4 0.26 0.84 0.3 0.034 0.019 0.022 1581 3 Side frame 363580 97020 190 4 0.23 0.83 0.39 0.027 0.025 0.019 1585 Scab 4 Side frame 348880 130340 170 4.5 0.29 0.92 0.42 0.041 0.038 0.017 1610 Crack 5 Side frame 353780 97020 180 4 0.21 0.71 0.47 0.032 0.028 0.018 1623 6 Side frame 392000 97020 200 4 0.28 0.78 0.44 0.029 0.021 0.023 1625 7 Side frame 368480 93100 170 4.5 0.26 0.83 0.31 0.021 0.024 0.021 1613 8 Side frame 383180 77420 210 3.5 0.24 0.86 0.38 0.032 0.021 0.023 1598 9 Side frame 353780 97020 180 4 0.27 0.85 0.38 0.031 0.03 0.028 1624 10 Side frame 348880 97020 190 3.5 0.26 0.83 0.4 0.023 0.022 0.016 1613 Blowhole 11 Side frame 363580 92120 195 4 0.22 0.87 0.39 0.022 0.025 0.019 1615 Scab 12 Side frame 392000 106820 200 3.5 0.27 0.85 0.32 0.027 0.031 0.018 1598 13 Bolster 368480 87220 170 3 0.23 0.97 0.33 0.031 0.021 0.021 1584 14 Bolster 378280 97020 270 4 0.29 0.91 0.47 0.032 0.027 0.019 1595 15 Bolster 353780 92120 180 4 0.24 0.79 0.35 0.036 0.021 0.023 1620 Blowhole 16 Bolster 373380 72520 210 3.5 0.25 0.84 0.38 0.035 0.027 0.022 1615 Scab 17 Bolster 373380 106820 200 4 0.22 0.95 0.35 0.039 0.036 0.021 1613 Crack, air lock 18 Bolster 358680 82320 190 4.5 0.28 0.89 0.43 0.027 0.031 0.019 1582 Misrun, Blowhole 19 Bolster 401800 92120 210 3 0.21 0.78 0.44 0.028 0.017 0.022 1625 Scab 20 Bolster 348880 101920 190 4.5 0.29 0.89 0.47 0.027 0.032 0.018 1617 21 Bolster 368480 97020 180 3.5 0.26 0.81 0.42 0.028 0.021 0.021 1609 22 Bolster 378280 97020 190 4.5 0.24 0.86 0.31 0.034 0.022 0.03 1613 23 Bolster 363580 97020 190 4 0.23 0.85 0.32 0.022 0.017 0.023 1621 24 NTT body 368480 87220 210 4 0.25 0.91 0.42 0.025 0.021 0.028 1593 25 NTT body 358680 97020 180 4 0.27 0.84 0.39 0.035 0.031 0.025 1612 Air lock 26 NTT body 343980 101920 190 4.5 0.26 0.82 0.41 0.025 0.021 0.016 1587 Misrun 27 NTT body 358680 87220 190 4.5 0.24 0.86 0.31 0.036 0.022 0.03 1625 28 NTT body 363580 97020 190 4 0.22 0.83 0.43 0.02 0.02 0.026 1621 Air lock

[MUKESH JAAT]

please help

[Steven Lord]

This sounds an awful lot like a homework problem. If it is, show us what you've tried so far and ask a specific question about where you're having difficulty and we may be able to offer some guidance.

If this is not homework and you simply want to create a neural network and train it using backpropagation, see this page in the documentation and/or see if one of the topics on this page describes the task you want to perform (image classification, function approximation and clustering, or time series and control systems seem like the most likely candidates) and how to use neural networks to accomplish that task.

[Greg Heath]

Were did you get your info?

I'm sure some of our readers can give you a MUCH better introductory source.

The introductory documentation for FITNET should help:

 help fitnet

 doc fitnet

Hope this helps.

Greg

[MUKESH JAAT]

Thanks sir for suggest

Answer 1

[BERGHOUT Tarek]

read this code you will understand : https://www.mathworks.com/matlabcentral/fileexchange/69947-back-propagation-algorithm-for-training-an-mlp?s_tid=prof_contriblnk