【作者】 李婷；

【导师】 陈峦；

【作者基本信息】 电子科技大学 ， 电气工程， 2021， 硕士

【摘要】 随着经济发展,人民生活水平增长,国内用电量近年来不断增加,产生的电力消耗可能导致计划外的停电事故,且电力能源有大规模储存困难的缺点。此外,电动汽车的快速发展,分布式可再生能源的大量接入都给电网负荷增加了波动性,短期电力负荷预测面临的影响因素逐渐增多。传统预测方法不再适用于复杂多变的电力负荷,挖掘新型的短期负荷预测方法迫在眉睫。本文以基于EABLT-RIO的短期负荷预测研究为课题。在四川省杰出青年科技人才项目(2020JDJQ0037)的支持和引导下,研究了国内外预测方法,总结了短期负荷预测的分类方法及相关方法的优缺点,并分析了新形势下的人工智能预测方法的重要作用。本文提出的EABLT集成模型短期负荷预测方法研究中,首先建立了差分移动自回归模型ARIMA,BP神经网络和LSTM长短期记忆网络、时序卷积网络TCN等单一模型。然后针对当前负荷复杂多变,且规律性较强的特点,基于集成算法,将上述网络集成组合,建立了EABLT预测方法,有效的提升了预测精度。本文提出了基于EABLT-RIO的短期负荷预测的研究模型。该模型通过RIO优化方法,对训练后的EABLT模型的预测数据的残差进行估算,矫正了EABLT模型的输出,提高了模型预测精度。并研究了EABLT-RIO模型的不确定性估计的质量,验证了模型的准确性与有效性。结合我国某地区的真实电力负荷数据集,通过算例分析,验证了所构建的EABLT模型的有效性和集成组合预测的优越性。在RIO模型中直接引进预测模型进行残差估计,并对比分析,证明RIO模型具有良好的泛化能力。最后基于输入输出高斯核函数,进行预训练模型的误差校正,并最终证明EABLT-RIO模型可以有效提高预测精度。更多还原

【Abstract】 With the development of economy and the growth of people’s living standards,domestic electricity consumption has been increasing in recent years.A large amount of electricity consumption will lead to unexpected power outages.Moreover,it is difficult to store electric energy on a large scale.In addition,the rapid development of electric vehicles and the access of a large number of distributed renewable energy sources have increased the volatility of grid load,and the influencing factors of short-term power load prediction are gradually increasing.Traditional forecasting methods are no longer suitable for complex and changeable loads,so it is urgent to excavate new short-term forecasting methods.This paper takes short-term load forecasting research based on EABLT-RIO as the subject.With the support and guidance of the Sichuan Outstanding Young Science and Technology Talent Project(2020JDJQ0037),this paper studies the domestic and foreign forecasting methods,summarizes the classification methods of short-term load forecasting and the advantages and disadvantages of related methods,and analyzes the important role of artificial intelligence forecasting methods under the new situation.In the short-term load prediction method of EABLT integrated model proposed in this paper,the Autoregressive Integrated Moving Average mode(ARIMA),Back Propagation(BP)neural network and Long Short-Term Memory network(LSTM),and Temporal Convolutional Network(TCN)and other sub-models are first established.Then,combined with the complexity,variability and strong regularity of the current load,based on the integration algorithm,the above networks are integrated and combined to establish the EABLT prediction method,which effectively improves the prediction accuracy.This paper presents a research model of short-term load forecasting based on EABLT-RIO.In this model,the residual of the EABLT model after training was estimated by RIO optimization method,which further corrected the output of EABLT model and improved the prediction accuracy of the model.The accuracy and effectiveness of the EABLT-RIO model were verified by studying the quality of uncertainty estimation.Combined with the real power load data set of a certain region in China,the validity of the EABLT model and the superiority of the integrated combination prediction are verified through the analysis of a numerical example.The prediction model was directly introduced into the RIO model to carry out residual estimation and comparative analysis,which proved that the RIO model had good generalization ability.Finally,based on the input and output Gaussian kernel functions,the error correction of the pre-training model is carried out,and finally it is proved that the EABLT-RIO model can effectively improve the prediction accuracy.更多还原