Functional network was proposed by E.Castillo in 1998.It was an extension of neural networks. Unlike of neural networks, it deals with general functional models instead of sigmoid-like ones, and in these networks, there are no weights associated with the links connecting neurons, the internal functions are not fixed but learnable. These functions are not arbitrary, but subject to constraints to satisfy the compatibility conditions imposed by existence of multiple links going form the last input layer to the same output units.In recent years, forecasting work became more and more importent. There are various forecasting objects and methods, but different methods are suitable to different objects. Therefore the study of forecasting itself is one of important contents in forecasting research.Recently the research about linear problems is perfect, because of the world is more complex,the relation among the world is nonlinear, Traditional method is difficult to solve these problems, and to reach accuracy of fitting and forecasting, Functional networks have certain advantages solving non-linear problems.This thesis to be used functional networks solves some problems in forecasting problems, especially some non.linear problems'model and algorithm.Firstly, this thesis summarizes some basic theories and methods and current research status in domestic and abroad about functional networks and forecasting technology. Secondly Non-linear regression forecast model and learning algorithm Based on Functional Networks is proposed, and gave some examples about one arbitiary and more arbitrarys regression model.simulation result shows that the result of this model is more precise than others regression-forecasting methods. Forecasting models and learning algorithms based on Functional Networks about predicting macro-economics problems is proposed in this thesis. Comparing with others models and algorithms, this model is more perior and application. The result compared with others models and algorithms demonstrate that models and algorithms based on functional networks have some value about theory and application.This thesis is an initial attempt about models and learning algorithms based on Functional Networks. It hopes to provide a new reference method to forecaster