基于果蝇算法和SVM的天然气日负荷预测

Gas Daily Load Forecasting Based on Fruit Fly Optimization Algorithm and SVM

为了有效提高天然气短期负荷预测的准确度,提出了一种集成果蝇优化算法和SVM(Support Vector Machine)的混合优化策略FOA-SVM。首先,采用K-近邻算法对燃气负荷样本中离群数据进行查找定位,并用特征曲线法对离群数据进行修正。其次,综合考虑节假日、日期类型以及天气等影响因素,建立了基于SVM的天然气日负荷预测模型,并采用果蝇优化算法优化SVM的模型参数。最后,采用宁夏平罗县居民燃气日负荷数据和多种通用的定量误差评价方法,对建立的预测模型的可行性和有效性进行了验证。仿真结果表明,基于果蝇优化算法和SVM的组合优化方法相对于人工神经网络和单纯SVM方法,具有更好的预测精度。

A hybrid process of modelling and optimization, which integrates the fruit fly optimization algorithm（FOA） and support vector machine（SVM）, is proposed to effectively improve the prediction accuracy of short-term gas load. Firstly, the abnormal data in historical gas load by the means of K-nearest neighbour（KNN） is located and examined; Moreover, the history gas data is modified with the characteristic curve method. Secondly, the FOA-SVM forecast model is established, using FOA to optimize model parameters of SVM, and fully considering some influence factors, such as holidays, date type and weather. Finally, the gas daily load predicting model is evaluated according to some assessment methods commonly used, and tested with raw samples from the Pingluo gas company in Ningxia to prove the feasibility and effectiveness. Simulations show that FOA-SVM has higher forecast accuracy for better applicability compared with ANN and SVM methods.