The Study of Adaptive Tracking Real-time Rain Series, 2009 Mar – 2010 Oct (Chinese National Funds Supported)

Evolutionary Computing & AI Project,

This is the second project I have done, and is the most biggest project. The objection is to forecast rainfall series for saving the cost caused in the transmission process. I tried to designed an adaptive switch algorithm based on GA, MS-ARIMA, and Island model. The main work I have done is designing the genetic operations for this problem, the first person combine these 3 models into a forecast system, the first team to try to use adaptive algorithm to tracking rain series (Related Work can be seen in the follow)

Main Objection

1.To tracking the future 5 times rainfall series.
2.To construct a rainfall forecasting system

Related Work

There are many types of nonstationary time-series. Like rainfall, wind, power, stokes. They often too complex to forecast, but tracking these series still have very large benefits. For tracking the rain series, many researchers put forward many useful models like using ARIMA model [1], ARCH model [2], Grey Model[3] and forth regression models[4]. In other sides, many researchers also put forward some models based on like Two-Sample Model[5], Bayesian Model[6]. Moreover, ANN[7], GAs[8], PSO[9] also can solve this problem. Some researchers also try to combine some models into one for tracking like [10] using Tree model combined with GAs to track the time-series, or [11] using Markov Model combined with regression model to track the time-series.

Novel Point

In these projects, we mainly put forward a method combined with all three kinds of model to get a good performance. Mainly using MS-AR[12] model combined with a kind of genetic operations designed by me. The operator makes algorithm suits for large database learning to adaptive tracking the rainfall series. These methods have good points like follow

1.The method is designed according to the nature process of rainfall. So it is more efficiency than other methods.

2. The method designed by three different kinds of model, which show the hidden connection between all three kinds of tracking model.

Partly Result

Some Rainfall Forecast Outcomes:

Finally GUI surface:

Download Reference

Return to Research Page