Research On The Methods Of Long-term Gas-load Forecasting Of Shanghai

It has become the emphasis of shanghai’s energy policy that increasing theproportion of the gas, particularly natural gas, in energy consumption structure in thecontext of the energy conservation,emission reduction and building smart and greencity. The main gas of Shanghai is composed of coal gas, liquefied petroleum gas andnatural gas. With rapid urbanization, the gas industry in Shanghai is in the stage ofrapid development. As a basis for decision making gas load forecasting is particularlyimportant for ensuring the gas supply scientific, efficient and meeting the needs ofpeople’s daily life and the sustainable development of economic and society. MoreIntelligent and accurate of gas load forecasting must be developed as soon, and it isthe emphases of this paper. This article studies the popular variety of forecastingtechniques currently: regression analysis, genetic algorithms, artificial neuralnetworks, adaptive neural fuzzy inference and pattern search methods, and thelong-term gas load in Shanghai deeply, meanwhile explore three combinedforecasting models. The main contents are as follows: Correlation analysis of theinfluential factors of gas load. In load forecasting, the regularity of load variationsneeds to be extracted from historical data and related factors. Using the method ofspearman rank correlation analysis screens out the main factors that have greaterinfluence.The first model integrates an improved genetic algorithms and an artificialneural network with switches, which uses the improved genetic algorithm tooptimize and adjust connection weights and the switch parameters of the artificialneural network. The experimental results show that the prediction results of themodel proposed are more accurate than BP neural network, or standard geneticalgorithm and an artificial neural network with switches.The second model is acombination of the model search algorithm and the adaptive neural fuzzy inferencesystem, of which target is to predict the annual load of coal gas. For the trainingparameters and model structure of the adaptive neural-fuzzy inference system needto be artificially adjusted, although, after many times of test, the accuracy ofprediction may also fails to meet the requirement of the problem, Using pattern search algorithm to optimize the parameters of the fuzzy inference system to find theoptimal configuration. Experimental results show that the proposed model has moregreatly improved in terms of training time and prediction accuracy than the adaptiveneural-fuzzy inference system lone.