Research On Forecast Methods Of Irrigation Water Use

In recent years, the absence of water resources becomes more and more serious. Saving and reasonable using water resources have been a hot topic. In China, water resources are relatively deficient and the average amount per capita are very few. So China is one of the thirteen poor water countries in the world. The agriculture is a big household of water use and agricultural water use mostly consumes to the irrigation. The accurate forecast of irrigation water use has extremely vital significance to set down reasonable and effective translating schemes and dispatching plans of water resources as well as enable the limited water resources to display the biggest benefit. Meanwhile it can be used for reference to instruct agricultural production of irrigated area.This article summarizes the class of forecast methods and the research progress of irrigation water use forecast home and abroad. On the base of summarizing predecessor experiences, taking Tiejia irrigated area Donggang City Liaoning Province as the research object, the author uses time series, neural network, gray model as well as combination model between gray system and neural network to carry on full-scale forecast analysis to irrigation water use of Tiejia irrigation area. Also the paper proposes several effective improving methods.Time series are a group of figure sequences arranged by observation values of identical phenomenon one after another in different time. The stationary random process is a special form of the stochastic process. This article uses linear extrapolation of stationary time series to inquire into regression forecast model of irrigation water use with 24-year history data. The model's relative error between forecast result and actual value is less, so the precision is higher.Basing on artificial neural network theory, this research makes preliminary discussion to the choice of neural network forecast model, the modeling flow and realization methods of irrigation water use forecast model, and then the paper establishes forecast model of irrigation water use on the base of artificial neural network. This article separately uses three layers' back propagation neural network (BP network) model, radial basic function neural network as well as generalized regression neural network model to forecast irrigation water use. When it is training network, in order to guarantee the data's magnitude is the same, the article makes normalization processing to the neural network input and the output data (finally makes deoxidization processing to the forecast result). The article uses neural network toolbox of MATLAB 7 to carry on network modeling and simulation experiment as well as to train and optimize network performance. It provides new thought and methods for establishing forecast model of irrigation water use. The results indicate that radial basic function neural network exists partial minimum shortcoming and the forecast effect is inferior to BP neural network and generalized regression neural network. Therefore, we suggest people use BP neural network and the generalized regression neural network model in the actual production.The gray system forecast is to carry on quantity size forecast to the system time series. The article introduces the content of gray forecast system by the numbers and elaborates the thought and actual operation process of GM (1, 1) model on emphasis, and then it establishes GM (1, 1) model of irrigation water use. After comparison we discover, the forecast result of straightly using GM (1, 1) model is not very ideal. For this, the paper first carries on residual error modification to the established model, and then it uses three kinds of precision test methods to check up modified model, finally, the results indicate all of the tests satisfies the demand. The paper also attempts to establish the equal-dimension and new-info model to forecast irrigation water use. This method's forecast precision is high and it is effective.Aiming at the respective advantages and disadvantages of neural network and gray model, this article analyzes the two methods' complementation characteristic and proposes a forecast method that combines neural network and gray model. It separately establishes the series, the inlaid and the parallel gray neural network model. The result shows, the forecast precision of the series and the parallel gray neural network model is higher, and the errors are less than 4%. So both methods can be used to forecast middle-long term irrigation water use.