基于系列智能算法的某省天然气用户用气规律分类与短期调峰需求预测

Gas Consumption Classification and Short-term Peak Demand Forecasting of Natural Gas Users in a Certain Province Based on a Series of Intelligent Algorithm

天然气管网系统庞大、用户繁多,在用户用气习惯不明确和调峰需求量模糊的情况下,管网的运行调度面临巨大挑战。本文基于某省天然气管网2018年至2020年用户用气数据,利用k-means++算法对不同用户的用气习惯进行初步类别划分,而后使用BP神经网络对典型用户的特征向量进行学习、分类,结合监督学习和无监督学习两种方式,获取三种典型用户。基于上述分类结果,利用典型用户三年的调峰需求量数据,分别通过七种智能算法模型对用户调峰需求数据进行了预测,旨在得到可靠的用户调峰需求量短期预测模型。模型计算结果表明,七种算法对短期调峰预测均有可靠的性能,其中GDB模型拥有更快的计算速度和更高的准确性,并且对于长期调峰预测具有一定的可行性。

In China,natural gas pipeline network is quite complex due to the large number of users,so the operation and scheduling of the pipeline network is faced with great challenges under the condition of unclear gas consumption habits and fuzzy peak regulating demand.In this paper,K-means++algorithm was used to preliminarily classify the gas consumption habits of different users based on the gas consumption data of the users in a natural gas pipeline network of a certain province from 2018 to 2020.Then,BP neural network was used to learn and classify the feature vectors of typical users,and three types of users were obtained by combining supervised learning and unsupervised learning.Based on the above classification results,the peak regulating demand of users was predicted by seven intelligent algorithm models by using the peak regulating demand data of the typical users for three years,aiming to obtain a reliable short-term prediction model.The model calculation results show that all the seven algorithms have reliable performance for short-term peak regulating prediction,among which the GDB model has the fastest calculation speed and highest accuracy,and has certain feasibility for long-term peak regulating prediction.