基于LSTM的天然气用气量智能预测方法

An Intelligent Prediction Method of Natural Gas Consumption Based on LSTM and Historical Data

用户侧用气量的准确预测是天然气生产及管网运行调度的前提。为弥补现有预测方法未考虑数据本身误差对预测结果的影响,本文提出了一种基于历史数据的天然气用气量智能预测方法。该方法通过数据清洗和异常值筛选对原始数据进行预处理,降低原始数据误差对预测结果的影响;通过三次样条插值解决用气量数据丢失和用气量非等时间间隔的问题,采用小波降噪降低原始数据中的噪声;最后,通过实测数据构建了居民用气和工业用气两种类型的数据集,并通过构建的LSTM网络预测用气量。结果表明,该方法可以有效地预测天然气的用气量,与未处理的数据相比,预测误差分别降低了19.1%和27.9%。

An accurate prediction of natural gas consumption is crucial for gas companies to develop gas supply plans and avoid resource wastage.Existing prediction methods often overlook the impact of data errors on prediction results.This paper proposes an intelligent prediction method for natural gas consumption based on historical data.The original data is preprocessed through data cleaning and outlier screening to reduce the impact of original data on prediction results.Cubic spline interpolation is employed to address issues related to data loss and irregular time intervals in gas consumption records.Wavelet denoising techniques are applied to reduce the noise in the original data.Moreover,two distinct data sets of residential gas and industrial gas are constructed based on the measured data,and gas consumption is predicted through the constructed LSTM network.The experimental findings demonstrate the effectiveness of the proposed method,with a respective reduction in prediction errors of 19.1%and 27.9%compared to the unprocessed data.