【摘要】 目的 为克服传统时间序列预测方法在天然气负荷预测中存在的局限性,提出一种基于SVMD-GRUAttention-SVR的天然气负荷组合预测模型.方法 使用斯皮尔曼相关系数法对影响因素进行相关性分析,获取强相关特征.通过逐次变分模态分解（SVMD）将原始负荷序列分解为若干个子信号分量,并将注意力机制引入门控循环神经网络（GRU）,对各子分量分别进行预测,将预测结果叠加融合得到初步预测结果和预测误差,之后利用支持向量回归（SVR）模型对预测误差进行校正,获取最终负荷预测值.结果 对比不同模型的预测结果,该组合预测模型的均方误差、平均绝对误差、均方根误差和决定系数分别为0.002 5、0.038 6、0.049 6和0.981 3,具有更高的预测精度.结论 所提组合模型能够有效提高天然气负荷预测精度,可为天然气负荷预测研究提供理论支持,为天然气公司平稳供气提供决策依据.更多还原

【Abstract】 ObjectiveIn order to overcome the limitations of traditional time series forecasting methods in natural gas load forecasting,a natural gas load combination prediction model based on SVMD-GRU-Attention-SVR is proposed.MethodsThe correlation analysis of the influencing factors is performed using the Spearman correlation coefficient method to obtain strong correlation features.The original load sequence is decomposed into several sub-signal components through successive variational modal decomposition(SVMD),and the attention mechanism is introduced into the gated recurrent neural network(GRU) to predict each sub-component separately,and the predictions are combined to generate an initial forecast and a forecast error,The SVR model is used to correct the prediction error and obtain the ultimate load prediction value.ResultsCompared with the prediction results of different models,the mean square error,mean absolute error,root mean square error and determination coefficient of the combined prediction model are 0.002 5,0.038 6,0.049 6 and 0.981 3,respectively,which has higher prediction accuracy.ConclusionThe proposed combination model can effectively improve the accuracy of natural gas load forecasting,provide theoretical support for natural gas load forecasting research,and provide decision-making basis for natural gas companies to supply gas smoothly.更多还原