The Research And Application Of Single Neuron And Multilayer Feedforward Artificial Neural Network

Artificial neural network(ANN) is a complex network which consists of lots of interlinked neurons. It can simulate some intelligent behaviors of human and possesses abilities in self-learning, self-adaptive and information's distribute storage. Now ANN has been applied in the fields of intelligent control, system identification, intelligent supervision and so on.The development of ANN paves a new way to solve the problem of controlling the nonlinear, long time delay and uncertain complex system. This thesis studies single neuron by constituting a single neuron controller with relatively simple algorithm, which is easy to put into practice, and then carries out a further study by composing an intelligent temperature control system with the single neuron. By applying this system in temperature real-time control of an electric heat furnace with multiple temperature zones which is long time delay, nonlinear and coupling, a better effect can be obtained.For its simple and convenient features, the arithmetic of error back proragation(BP) is the best choice of the multilayer feedforward artifical neural network(MFANN), which is a maturest and most popular network. The study on BP's algorithm and its improvement, structure of several kinds of MFANN and relevant calculation are also involved in this thesis.Load forecasting is an important work in electric power system, because its accuracy affects deeply the running work and productive expenditure of one system, it plays a main role in ensuring that electric power system can run ateconomic, safe and reliable conditions, and has become a significant part in the energy management. Forecasting is considered one of the most potential fields for the application of ANN, and applying ANN forecasting in the electric power system will be fast and accurate. In this thesis, some different types of the MFANN structured in the paper are applied in short term load forecasting for the electric power system, then their forecasting results are comparatively studied. This method of ANN forecasting inherits the idea of traditional time sequence method, and connects with related factor method at the same time. Compared with the traditional method, it has increased the accuracy and speed of forecasting, by which it proves its availability.