googletag. display(\"div-ad-articleLeader\");
The biological aspect of stimulating the development of artificial neural networks can be said that the study of artificial neural networks began in 1943 with McCulloch and Pitts.
Their work is based on the desire to understand the function of the biological brain, and they propose a \"computational element\" model called McCulloch-
Pitts neurons, which weights and sums the inputs of these elements, and then performs threshold logic operations.
The features of biological neurons that inspire artificial neurons are summarized as follows: cell body (soma)
: Perform the central function of the neuron.
Equivalents in artificial neurons are the total and threshold parts of the model.
Branch Crystal: Branch Crystal
Like the structure of receiving input signals processed by soma.
These are represented by weighted inputs from artificial neurons.
Axon: carries the output signal from the cell.
In artificial neurons, this is the output of the threshold function.
Compared to electronic digital computers, the biological brain is by far the best \"computer\" we know \".
Nevertheless, both have their pros and cons.
For example, the brain is not very good at doing arithmetic, and the digital computer is good at this task.
On the other hand, the brain is very good at tasks such as face recognition, and the digital computer did not successfully complete this task until 1993.
More and more people are asking digital computers to solve complex problems that their biological peers can easily solve.
This trend has always been the driving force for the development of neural networks.
It is important to understand that unlike biological networks, artificial neural networks are not physical entities, but one or more algorithms implemented in computer programs.
In order to realize the visualization of artificial neural network, some graphical conventions are adopted.
The following figure shows a single artificial neuron with the above basic elements, I . E. e.
, Weighted input, Total and threshold parts, and output.
Credit: Chris de villierscre: Chris de villierstring of the artificial neural network compared to so-
Known as an expert system containing knowledge (data)
There is no database for artificial neural networks.
They need to be trained with data suitable for a given problem in order to best adjust the weight and achieve the desired performance.
A kind of training is called supervision training.
In supervised training, the data set in the question considered is used to train the network.
These data contain certain parameters or variables (inputs)
Causing certain conditions or States (outputs).
A subset of available data, usually 70% for training the network.
This is an iterative process in which each input vector and its associated output or target (
It\'s usually a single number, but it can be a vector)
, Presented to the network.
Then adjust the input weight according to the appropriate algorithm (e. g.
Rules of perceptual learning
To minimize mistakes, I. e.
, The difference between network output and target.
The process is repeated before an acceptable error level is obtained.
After that, the remaining 30% of the data is used to validate the network.
If the verification is successful, the network is applied to the problem;
If not, re-train it with a different data set.
In this case, different network structures may also need to be selected, such as more or less nodes, hidden layers, etc.
Practical application examples of artificial neural network has been applied in a variety of problem-solving applications, and with the discovery of new applications, this field is expanding.
The following examples illustrate various practical or potential neural network applications.
Stereo vision has been shown in the background of stereo vision, although the operating principles of biological vision and vision are very different.
The \"vision\" of the neural network, the behavior of the two methods is similar.
In this study, the neural network consists of several sub-networks, each consisting of three input neurons, each connected to a continuous two layers of nine hidden nodes, and finally to an output neuron.
It is important that the neural network is trained to detect the features of stereo vision rather than mimic its biological features.
The results show that both the neural network and its biological network have a strong response to the edges of vertical strips and light.
Fault Detection of power lines in transmission lines
Distance protection is to protect the power system from transmission line faults by isolating fault lines.
The fault condition is the line-to-line and line-to-ground faults.
It has been proved that the neural network solution is better than the traditional method of isolating faulty lines using sensing relays, but it takes a long time of training, and the software implementation is inefficient in calculation, resulting in slow response.
An efficient fault detection system requires a decision to be made within a power cycle, which is 20 ms in the case of a 50 hz system.
Another hardware implementation of the neural network allows fast, real-time
Time Distance protection.
Quality management in the different inspection techniques used in the quality inspection of finished products, is a way to capture the image of the item, enlarge the image and visually inspect it.
In the
PCB assembly plant, the embedded automatic detection system based on artificial neural network is found to be superior to the traditional method.
1974 The Ministry of Environment of Malaysia has adopted a Water Quality Index (WQI)
The level of river pollution in Malaysia is graded.
WQI is based on the weighted values of dissolved oxygen, biochemical oxygen demand, chemical oxygen demand, pH, ammonia nitrogen and suspended solids.
Getting WQI requires converting raw data to subdata
The pollutant index before WQI manual calculation.
A neural network solution was then implemented that uses raw data from the past and present to predict WQI.
The accuracy of this method was found to be between 92.
4% and 99. 96 percent.
Conclusion artificial neural networks are more and more applied to solve practical problems. life problems.
As more and more powerful computers become readily available on the one hand, and on the other, powerful microprocessor devices enter embedded applications, applications of artificial neural networks may become common if we do not realize them.
Function articleAdMiddleLoad (){if (typeof jQuery ! = \"undefined\"){
Last position of Var = 0;
Minimum distance between Ads var = 1200;
Minimum distance from var at the bottom = 300;
Var adsToPlace = 20; for(var i = 1;
I /element is located after the ad & in the last position! $(this). hasClass(
\"Article \")
/Element is not advertising &&($(this). position(). top -
Enough space between Ads & $ (\". articleText\"). height()-$(this). position().
Enough space from the bottom of the article | $ (this). prev(). hasClass(
\"Article \")
/The previous element is a slide)){$(this). before($(
\"ArticleAdMiddle \")); googletag. display(\"div-ad-
\"I \");
$ (this). position(). top;
SpotFound = true; return false; }});
/If no location is found, stop looping and remove any remaining banner slots. if(! spotFound){for(var j = i;
J
