I have an NN with 192 inputs and 48 outputs where I use it for electricity forecasting. It has only one hidden layer and neurone. Previously I used this with back-propagation. Now I want to have better results, so I train it with GA. But results with GA are worse than with BP (rarely get better results with GA). I have tried with different parameter arrangements (code is attached). But still, I cannot find the reason. I checked with different amount of training sets (10, 15, 20, 30) and different amount of hidden neurones. But when I increase them, results get even worse. Please, someone, help me for this.
Regards,
Dara
----------------------------------code------------------------------------------
for i = 1:17;
p = xlsread('Set.xlsx')';
t = xlsread('Target.xlsx')';
IN = xlsread('input.xlsx')';
c = xlsread('Compare.xlsx')';
inputs = p(:,i+20:i+27);
targets = t(:,i+20:i+27);
in = IN(:,i);
C = c(:,i);
[I N ] = size(inputs)
[O N ] = size(targets)
H = 1;
Nw = (I+1)*H+(H+1)*O;
net = feedforwardnet(H);
net = configure(net, inputs, targets);
h = @(x) mse_test(x, net, inputs, targets);
ga_opts=gaoptimset('TolFun',1e(-20),'display','iter','Generations',2500,'PopulationSize',200,'MutationFcn',@mutationgaussian,'CrossoverFcn',@crossoverscattered,'UseParallel', true);
[x_ga_opt, err_ga] = ga(h, Nw,[],[],[],[],[],[],[], ga_opts);
net = setwb(net, x_ga_opt');
out = net(in)
Sheet = 1;
filename = 'Results.xlsx';
xlRange =['A',num2str(i)];
xlswrite(filename,x_ga_opt,Sheet,xlRange);
i = i + 1;
end
-------------------------------Objective Function---------------------------------
function mse_calc = mse_test(x, net, inputs, targets)
net = setwb(net, x');
y = net(inputs);
e = targets - y;
mse_calc = mse(e);
end
http://www.mathworks.com/matlabcentral/answers/306379-why-it-gives-worse-results-when-i-use-genetic-algorithm-for-training-nns-than-when-i-use-back-propag
Dear Researcher
GA is one of teh evalutionary optimization techniques and it cannot be utulized as approach to train Neural Networks. You can train your Neural networks with Back-Propagation method and then optimize you NN with GA or other methods. I have attached some articles to this answer I hope this will help you.
regards
Aydin
1.https://www.researchgate.net/publication/306048088_Optimizing_radio_frequency_identification_network_planning_through_ring_probabilistic_logic_neurons?ev=prf_pub
2.https://www.researchgate.net/publication/303402995_Modeling_Mechanical_Properties_of_FSW_Thick_Pure_Copper_Plates_and_Optimizing_It_Utilizing_Artificial_Intelligence_Techniques?ev=prf_pub
3.https://www.researchgate.net/publication/269367058_Introducing_Genetic_Algorithm_as_an_Intelligent_Optimization_Technique?ev=prf_pub
4.https://www.researchgate.net/publication/272572353_Introducing_Neural_Networks_as_a_Computational_Intelligent_Technique?ev=prf_pub
5.https://www.researchgate.net/publication/224132638_Modeling_of_forced_dermal_wound_healing_using_intelligent_techniques?ev=prf_pub
6.https://www.researchgate.net/publication/224132561_Optimizing_fuzzy_logic_controller_for_diabetes_type_I_by_genetic_algorithm?ev=prf_pub
Article Optimizing radio frequency identification network planning t...
Article Modeling Mechanical Properties of FSW Thick Pure Copper Plat...
Article Introducing Genetic Algorithm as an Intelligent Optimization Technique
Article Introducing Neural Networks as a Computational Intelligent Technique
Conference Paper Modeling of forced dermal wound healing using intelligent techniques
Conference Paper Optimizing fuzzy logic controller for diabetes type I by gen...
First, you can use genatic to train ANN. I tried it before and the results were perfect. In your code, I think you did not normalized your data. Second, you used GA only ones. It does not work in this way. Try to write a function with the number of inputs you need in your network in each layer. Write your ANN manually. Then use GA tool to optimize your function. If you need help I can help you.
The natural way of training a NN is using backpropagation. GAs are good to optimize parameter values not to train a NN
Hi all.Thank you very much for your answers. Please check your inboxes. I need your help to solve this problem. seems I am in a wrong track. Thank you very much again for your answers
Yes, you can train your NN using a GA. I've done it, and it works; however, I would suggest that you implement your own GA instead of using the Matlab tool because you can optimize your GA using as many different operators as you want.
Yes, you can use genetic algorithms (or other EAs such as evolution strategies, estimation of distribution algorithms, etc.) to train your ANN.
Furthermore, there is no need to choose between EA and BP: you can use both of them in an integrated cooperative way (forming in this case a memetic algorithm).
This paper is a few years old, but tackles the basic issues:
https://www.researchgate.net/publication/268318798_Adjusting_Weights_in_Artificial_Neural_Networks_using_Evolutionary_Algorithms
There is a ton of references in the literature to this kind of hybridization (I think Lawrence Davis already tried this model in the late 1980s and early 1990s), which is a very good starting point for such a training problem.
Chapter Adjusting Weights in Artificial Neural Networks using Evolut...
- Badges
- Science topic