Abstract
Сontent
- Introduction
- 1. Aims
- 2. Scientific significance
- 3. The content of the work by stages
- 4. Aprobation
- Conclusion
- References
Introduction
Ambient air is the Earth is in constant motion. His movement we feel as wind, which transfers heat from the equator to the poles and the moisture from the sea to the land where it falls in the form of life-giving rain. The only source of energy that causes the motion of the atmosphere, is the sun. Uneven heating of the earth's surface, which in turn heats the air creates a difference in atmospheric pressure. Cold air is denser, so it drops down and creates a high pressure area. Wind - this displacement of air from the high pressure areas to low pressure areas. Thus, constantly forms a certain state of the atmosphere, called weather. In turn, a person tries to keep an eye on weather processes using the Meteorology and monitor global climate change affecting the local weather conditions in the regions. Thus, having information about the weather conditions, meteorologists usually can predict the probability of occurrence and duration of certain natural phenomena: the presence or absence of precipitation, cloudiness, wind direction and intensity. Through the use of computer technology in recent decades has increased and the accuracy of weather forecasts. But in some cases still occur mismatch predicted and actual weather conditions, when the weather forecasters, they say, "get a finger to the sky." Therefore, the task of analyzing and forecasting of meteorological parameters from the time series remains relevant today. As a first approximation it is to build a mathematical model of a finite or differential, and the subsequent construction of the forecast based on it.
1. Aims
The purpose of master's work is to develop the final mathematical model to predict the dynamics of meteorological parameters in the form of a software product that includes a number of modern methods for pretreatment and forecasting of meteorological data.
The object of the study are time series of observed meteorological parameters: temperature, humidity, pressure, wind speed.
Subjects of research are the methods of preliminary preparation and forecasting data, quality and range of the forecast results
So, master's work aims are:
1. Examine existing methods of pretreatment and forecasting time series, and introduce them into existing prognostic complex;
2. Perform a review of existing methods of forecasting;
3. Develop the finite mathematical model of the meteorological parameters dynamics;
4. To test the mathematical models;
5. Implement the proposed program model;
6. Provide an information protection program of the mathematical model.
2. Scientific significance
Scientific novelty of the research lies in the fact that for the prediction of meteorological parameters used are not the final mathematical model, and the differential mathematical model. In addition to determining the minimum dimension of the model we used the method of Grassberger- Procaccia, which is quite simple and has a small error. Finally, I implement prediction using neural networks, because this method is most efficient for short-term predictions.
3. The content of the work by stages
In this paper, the calculation of prediction will be performed by implementation of the following steps:
Figure 1 – The scheme of the developed mathematical model(animation:volume - 25.2 KB, size - 617x263, the number of frames - 5, the number of repetitions - 7, delay between frames 50 ms, delay between the last and the first frame of 150 ms)
1. obtaining time series from a database (DB) Vantage Proweather station 2;
2. data analysis. Leveling of the time series is carried out in order to reduce the error of the input data. It’s supposed to achieve that using a moving average, since the experiments showed that this method had the best result.
3. definition of model dimension;
Grassberger-Procaccia method is to restore the attractor, "similar" to the original, sequential shift by the value of t. To estimate the embedding dimension is consistently getting new dimension and measure certain characteristics of the resulting multi-dimensional series. After a certain value, this value is usually ceases to increase, indicating that the achievement of the embedding dimension.
4. forecasting;
Neural networks may be approximated by continuous functions. We prove a generalized approximation theorem: using linear operations and cascading can be of arbitrary non-linear element to get a device that calculates every continuous function with a given accuracy. This means that the nonlinear response of the neuron can be arbitrary: from sigmoidal to an arbitrary wave packet or wavelet sinus or polynomial. From the choice of the nonlinear function may depend on the complexity of a particular network, but with any non-linear network is a universal approximator, and the correct choice of the structure can be accurately approximated by the continuous operation of any machine.