The Short-term Prediction Of Earth Rotation Parameters Based On BP Neural Network

With the rapid development of science and technology, the launches of all kinds of spacecraft, and the precision guidance and strike of the military weapon have also made great achievements. As a matter of fact, the development of all these technologies can be summed up in a few words, that is, the development of human’s study of the earth.When it comes to the study of the earth, what people think of at first is the study of the earth rotation parameters. The study of the earth rotation parameters plays an important role in the analysis of the reasons for the natural phenomena happening on the earth. At the same time, through these studies, it can also easily help us to get the highly accurate parameters of coordinate conversion, which can provide irreplaceable assistance to the launch of various types of satellites, spacecraft and missile. At the present stage, the technologies which can accurately measure the earth rotation parameters mainly include the Very Long Baseline Interferometry(VLBI), Satellite Laser Ranging(SLR) and Global Positioning System(GPS) and so on. VLBI and SLR are relatively slow in terms of accessing the real-time data, while GPS is faster, which need two or three hours to access the data. Although these technologies can get highly accurate parameters of the earth rotation, it also has some defects on the efficiency of the information. As a result, how to quickly get more accurate parameters of the earth rotation has become a very important work to us.This article put the parameters, which are provided by International Earth Rotation Service Organization(IERS) and downloaded from line, as the data source. At the same time, it uses the method of Backing Propagation(BP) neural network to forecast the polar motion and the changes of the length of the day. The main steps of the predicting process is that – firstly, the data should be divided into two parts, and then one part is used for training function, and the other part is used for the predicted result. And according to the results of the prediction, we can generally filter the best form and the number of the days of the prediction. Then on the basis of days of the prediction and by adjusting various kinds of parameters and the forms of the prediction, we can get the data. After collecting the data that BP neural network has predicted, we use the commonly used prediction model in reality to predict the same data. Then we use images to display the two data we get before. We should first carry out comparison of the average error. As a result of the consistency of error made by the data which neural network has predicted, there will be a general situation where the predicted error may be collectively good or bad. The analysis of the test results shows that if the parameter has adjusted to the best condition, the predicted data of BP neural network, which based on MATLAB, will get a significantly lower error than that of any other method.