Research Of Urban Gas Short-term Load Forecasting Based On ANN

With the development of economy and technology, especially the improvement of requirement of people's life and survival environment, people pay more and more attention on gas industry. Because of the relationship between energy structure and environment, and energy pushs the life quality to a further step.Gas as an ideal, clean, high quality energy source, gas industry can develop healthily and rapidly. Natural gas will be the main gas fuels with the project"west-to-east". It is the trend to enhance the development of natural gas industry all over the world. Our country begins to change the structure of energy source greatly, to increase the proportion of clean energy in townsman and industry step by step. At the same time, as both the gas supply unit and the gas using unit must obey the agreement"take or pay", it's needed to know the gas load effectively. Thus, urban gas load forecasting is an important research. It is a very important research task to forecast city gas load, for one hand, it will affect the planning of city gas pipe network, on the other hand, it is connected to the investment benefits and security of entire natural gas pipelines, and it's meaningful for urban gas optimizing attemperation and gas pipeline optimizing operation .Gas load forecasting include : long-term,middle-term,short-term,very short-term load forecasting. This dissertation emphasizes on short-term forecasting. There are many methods to gas load forecasting, including: regression analytical method, time serial method, elasticity coefficient forecasting, index analytical method, grey method, fuzzy logic forecasting, artificial neural network forecasting model, experts system forecasting model, optimizing combination forecasting model, etc..This paper choose ANN model, through comparing the advantages and disadvantages of each methods, and the practical data analysis. This dissertation introduces the rationale and algorithm and approaches about ANN. And then validate the correctness of theories by calculating instance with optimizing L-M algorithm. The outcomes are satisfied, by adopting different model, which include normal daily load forecasting, similar daily load forecasting,hour load forecasting.This dissertation also takes the method of using RBF. The effect is much better than pure ANN. So, if the net combines ANN and RBF, the outcome will be improved.With the rapidly development of GA and wavelet theory recent years, precision will be improved by add these two algorithms into ANN, which could amend the...