【作者】 白玮莉；

【导师】 刘志刚；

【作者基本信息】 西南交通大学 ， 轨道交通电气化与自动化， 2010， 硕士

【摘要】 电力系统短期负荷预测是实现电力系统优化运行的基础,对于电力系统运行的安全性、可靠性和经济性都有显著影响。因此,寻求有效的负荷预测方法以提高预测精度具有重要现实意义。迄今为止,研究人员已经提出了许多有效的预测方法,但仍存在大量改进空间。本文在应用HHT(Nilbert-Huang变换)对负荷进分解的基础上,结合目前较为流行的神经网络、支持向量机、粒子群优化等方法对电力系统短期负荷进行预报,主要进行了以下工作：通过对电力系统短期负荷数据进行预处理,采用莱特准则剔除异常值并进行小波去噪。在应用HHT对负荷数据进行分解的过程中采用差分算子和累计求和方法对模态混叠现象进行改善,从而更好的实现各频带的分离。通过对分解后得到的一系列频率从高到低的单分量信号IMF,将低频信号进行重构。然后根据各IMF的频率特性选取合适的预测模型。最后将各分量预测的结果相加就得到了最终预测值。对于高频随机分量IMF1,由于其波动性大,本文考虑温度及节假日因素,并采用神经网络和粒子群优化(PSO)方法对其进行组合预测,仿真结果表明该方案达到较好的预测效果。论文最后以2006年四川省某地电力系统实际负荷数据为样本集进行建模和预测,并以相对误差为性能指标进行了检验。结果表明,该方法达到较高的预测精度。更多还原

【Abstract】 Power system short-term load forecast is the basis of power system optimization running. It can affect safety, reliability and economy of power system operation. Thus, to find effective method has great importance for enhancing the prediction precision. Researchers have proposed many effective methods, but there is a lot of room for improvement. This paper use HHT to decompose the load datafirstly, then combine some popular methods,such as neural network、Support Vector Maehine、PSO to forecast the power system short-term load. The main work included:Pretreatment on the load data, applying wright criteria to remove outliers and wavelet algorithm to denoise; use a methed to improve its inherent flaw modal-mixing. A series of IMFs with frequency from high to low can be obtained after decompose the load data, reconstruct the low-frequency signals, then according to each IMF’s frequency feature we can chose appropriate model to forecast it. At last we can get the final forecast by add the predicted results of each IMF component. Due to IMF1 has large fluctuation, take temperature and weekday into consideration, otherwise neural network and PSO are used to optimize the combination weights. Simulation results indicate that this method has higher accuracy.Finally, the actual data of power system load in a certain place in Sichuan Province in 2006 is used as the sample to set prediction model、test results and verify the accuracy of the model.更多还原