Discussion about earthquake to reduce its casualties and damages is very important, especially in the Seismicity area like Iran that the occurrence of this natural phenomenon is seen annually. Iran has an approximate area of 1648000 square kilometers with geographical coordinates 25 to 40 degrees north latitude and 44 to 64 degrees east longitude that located in the middle of Alpine-Himalayan seismic belt. In this erea there are many active faults that their movement continues and the final balance has not been established. The occurrence of severe earthquakes as Buin Zahra earthquake (1962), Tabas (1978), Rudbar (1990), Bojnoord (1997), Bam (2003) and other numerous earthquakes prove this subject. While most natural disasters are out of human control, but it seems that Success in prediction of temporal and local of them can dramatically control damages and casualties. Earthquake occurrence in addition to changes of geometry and physics of the earth crust has many other effects. Some of its effects is in the ionosphere layer that are indicated as changes in the electrons values, ions density and electromagnetic field. Anomalies detection before earthquake is an important role for earthquake prediction. Each geophysical and geochemical parameter of the lithosphere, atmosphere and ionosphere layers that unusually changes before earthquake are known as earthquake precursor. Ionosphere changes that recognition by remote measurements (such as using Global Positioning System (GPS)) are known as earthquake ionospheric precursor.

TEC (Total Electron Content) of the ionosphere can be achieved by GPS data processing. Classic methods such as mean are unable to detect non linear pattern and therefore in complex and nonlinear systems they are not suitable for recognition and prediction of time series. Because of the nonlinear behavior TEC  and land surface changes in order to detect changes, in this paper an attempt is done using an artificial intelligence method including ANN (Artificial Neural Network) and multilayer Perceptron (MLP) for pattern recognition and prediction of TEC variations. Because ionospheric fluctuations usually do not have a normal distribution and do not follow Gaussian curve, in order to detect seismic anomalies, the mean and interquartile range is used to determine the lower and upper bounds. In this study several data sets from the ionospheric  total electron content (TEC) derived from the GPS data processing by Bernese softwares. In this way earthquakes of Ahar located in east Azerbaijan (2012/08/11) and Bushehr (2013/4/9) have been studied and the results were compared with data from global stations. First the stations coordinates were calculated using Bernese software with PPP (Precise Point Positioning) method. Then TEC values were obtained using GIM (Global Ionosphere Model). By analyzing the causes of ionospheric anomalies such as the geomagnetic field and solar activity and remove them from the process, results indicate that some of this anomalies caused by the earthquake and using intelligent algorithms could be useful for the prediction of nonlinear time series and outstanding anomalies ocurr some days before and after earthquake. It can be concluded that ANN algorithm has been able to detect TEC anomalies well. Also TEC values are obtained from ground stations have a high correlation with the results of global standard model.