基于特性分析的改进BP短期天然气预测研究

Research on Improved BP Short Term Natural Gas Prediction Based on Characteristic Analysis

天然气负荷具有随机性和波动性,且其预测是一个非线性问题,传统的BP神经网络模型精确度一般,收敛速度较慢。为了提高BP网络对天然气负荷短期预测的精度和收敛速度,深入分析影响天然气负荷的因素特性和管输量特点,将影响因素划分成长、短期因素和基准量、敏感量因素;其次,对因素进行包括量化和归一化在内的数据预处理操作,并用主成分分析法确定主要影响因素;最后,构建基于特性分析的BP神经网络天然气负荷预测模型,并采取改进的遗传算法对模型的初始值进行优化。通过对中部某省近三年天然气负荷数据进行建模分析,仿真结果表明,在天然气短期负荷预测的应用场景下,基于特性分析并利用GA算法改进后的BP网络模型收敛速度快且精度能够达到要求,验证了方法的有效性。

The natural gas load is stochastic and fluctuant,and its prediction is a nonlinear problem.The traditional BP neural network model has low prediction accuracy and slow convergence speed.To improve the prediction accuracy and convergence speed of BP network for short-term prediction of natural gas load,thoroughly analyze the characteristics of factors affecting natural gas load and pipeline transportation volume,and divide the influencing factors into growth,short-term factors,benchmark volume,and sensitive volume factors.Secondly,the influencing factors are quantified and normalized,and the main influencing factors are determined by principal component analysis.Finally,a BP neural network natural gas load forecasting model based on characteristic analysis is constructed,and an improved genetic algorithm is adopted to optimize the initial value of the model.Modeling and analyzing the natural gas load data of a central province in the past three years,the simulation experiments results suggest that the improved BP short-term natural gas prediction model based on characteristic analysis has fast convergence speed and the prediction accuracy can meet the requirements,which tests the effectiveness of the method constructed in this article.