系统工程理论与实践
繫統工程理論與實踐
계통공정이론여실천
Systems Engineering—Theory & Practice
2014年
11期
3001~3008
,共null页
龚承柱 李兰兰 杨娟 诸克军
龔承柱 李蘭蘭 楊娟 諸剋軍
공승주 리란란 양연 제극군
燃气管网 短期负荷预测 经验模态分解 相空间重构 最小二乘支持向量机
燃氣管網 短期負荷預測 經驗模態分解 相空間重構 最小二乘支持嚮量機
연기관망 단기부하예측 경험모태분해 상공간중구 최소이승지지향량궤
urban gas pipeline network; short-term load forecasting; empirical mode decomposition; phasespace reconstruction; least square support vector machine
城市燃气管网短期负荷预测对燃气调度系统的安全与稳定具有重要意义。为了提高城市燃气管网短期负荷预测精度,建立了基于经验模态分解（EMD）-相空间重构（PSR）-最小二乘支持向量机（LSSVM）的组合预测模型,首先,运用EMD算法把原始非线性时间序列分解为互不耦合的模态分量,并采用PSR算法确定LSSVM建模中各个分量的输入输出结构;其次,运用PSO算法对LSSVM建模中的参数进行优化,使用训练好的LSSVM模型对各个IMF分量进行回归预测;最后运用该组合模型对郑州市燃气管网负荷进行短期预测。结果表明：与LSSVM回归预测和BP神经网络预测模型相比,本文提出的组合模型的预测精度更高,是一种更为有效的城市燃气管网短期负荷预测方法。
城市燃氣管網短期負荷預測對燃氣調度繫統的安全與穩定具有重要意義。為瞭提高城市燃氣管網短期負荷預測精度,建立瞭基于經驗模態分解（EMD）-相空間重構（PSR）-最小二乘支持嚮量機（LSSVM）的組閤預測模型,首先,運用EMD算法把原始非線性時間序列分解為互不耦閤的模態分量,併採用PSR算法確定LSSVM建模中各箇分量的輸入輸齣結構;其次,運用PSO算法對LSSVM建模中的參數進行優化,使用訓練好的LSSVM模型對各箇IMF分量進行迴歸預測;最後運用該組閤模型對鄭州市燃氣管網負荷進行短期預測。結果錶明：與LSSVM迴歸預測和BP神經網絡預測模型相比,本文提齣的組閤模型的預測精度更高,是一種更為有效的城市燃氣管網短期負荷預測方法。
성시연기관망단기부하예측대연기조도계통적안전여은정구유중요의의。위료제고성시연기관망단기부하예측정도,건립료기우경험모태분해（EMD）-상공간중구（PSR）-최소이승지지향량궤（LSSVM）적조합예측모형,수선,운용EMD산법파원시비선성시간서렬분해위호불우합적모태분량,병채용PSR산법학정LSSVM건모중각개분량적수입수출결구;기차,운용PSO산법대LSSVM건모중적삼수진행우화,사용훈련호적LSSVM모형대각개IMF분량진행회귀예측;최후운용해조합모형대정주시연기관망부하진행단기예측。결과표명：여LSSVM회귀예측화BP신경망락예측모형상비,본문제출적조합모형적예측정도경고,시일충경위유효적성시연기관망단기부하예측방법。
Urban gas pipeline network short-term load forecasting is important for the security and sta- bility of gas distribution dispatching system. In order to improve the forecast precision, this study adopts an integrated model of empirical mode decomposition, phase space reconstruction and least squares sup- port vector machine, i.e., EMD-PSR-LSSVM, for urban gas pipeline network short-term load forecasting. Firstly, EMD is used to decompose the original nonlinear time series into several uncoupling intrinsic mode functions. Then, PSR is used to make the selection of LSSVM input/output-layer units. Further- more, particle swarm algorithm is used to optimize the model parameters and train LSSVM with temporal sequence samples, the trained LSSVM will be used for regression forecasting in advanced. Finally, the original loading data of Zhengzhou is adopted as example for empirical analysis. Results indicate that the EMD-PSR-LSSVM model has a higher outcome as compared to BP neural network and LSSVM regression, which has demonstrated the proposed integrated model is efficient and consistent.
