Urban Gas Load Forecasting Based on Multi-Time Scales

为了有效降低燃气管网运营成本，提高燃气管网智能化程度，需要对燃气负荷进行全方位预测，但由于燃气负荷预测不仅与燃气负荷数据、温湿度数据等空间特性有关，还具有趋势性、周期性等时间特性，导致燃气负荷预测机理建模困难且模型预测精度较低。考虑到不同时间尺度的燃气负荷对应的影响参数有较大的差异，因此针对不同时间尺度下的燃气负荷预测模型，包括日负荷、月负荷、季度负荷和年负荷，分别提出了基于改进量子粒子群算法优化的人工神经网(IQPSO-NN)的燃气日负荷和月负荷预测模型、基于BP神经网络算法的季度负荷预测模型、基于灰色GM(1, 1)马尔科夫方法的年负荷预测模型。采用平均预测精度对三种模型预测效果进行评价，对燃气负荷的预测精度均在93%以上，能够精确的预测各时间尺度上的燃气负荷。To effectively reduce the operational costs of gas pipeline networks and enhance their intelligent level, comprehensive prediction of gas load is required. However, gas load prediction not only relates to spatial characteristic such as gas load data and temperature/humidity data, but also exhibits temporal characteristics including trend features and periodicity. This dual nature leads to difficulties in constructing accurate predictive models and results in lower prediction accuracy for gas load forecasting mechanisms. Considering the significant differences in the influencing parameters corresponding to gas loads at different time scales, distinct prediction models have been proposed for various temporal dimensions of gas load forecasting (including daily, monthly, quarterly, and annual loads). Specifically, an Improved Quantum Particle Swarm Optimization-based Artificial Neural Network (IQPSO-NN) was developed for daily and monthly load forecasting, a BP neural network algorithm was employed for quarterly load prediction, and a Grey GM(1, 1) Markov method was implemented for annual load forecasting. The prediction performance of the three models was evaluated using average prediction accuracy. All models demonstrated gas load prediction accuracy exceeding 93%, indicating their capability to precisely forecast gas loads across various time scales.