摘要
单一模型在迭代训练过程中由于模型的自身误差,最终会降低预测精度。为了提高预测的准确性,引入完整集成经验模态分解-双向长短期记忆网络(CEEMDAN-BiLSTM)作为误差修正模型,提出一种栈式自编码器(SAE)与CEEMDAN-BiLSTM相结合的负荷预测模型。通过SAE模型学习气象因素、工作日类型、气温影响下负荷序列的主要特征,预测过程中产生的误差序列则反映了负荷序列的次要特征;使用CEEMDAN算法将误差序列分解为数个分量,针对每一项分量建立BiLSTM模型学习误差序列的时序特征,将各项分量的预测值累加得到误差的预测结果;将两种模型的预测值求和从而达到修正误差的目的。通过比较几种模型的预测结果表明:SAE与CEEMDAN-BiLSTM组合模型应用在短期电力负荷预测具有更好的准确性与稳定性。
In the process of iterative training, the prediction accuracy of a single model will be reduced due to its own errors. In order to improve the accuracy of the prediction, this paper introduced the integrated empirical mode decomposition-bidirectional long short term memory network(CEEMDAN-BiLSTM) as the error correction model, and proposed a load prediction model combining the stack self-encoder(SAE) and CEEMDAN-BiLSTM. The main characteristics of the load sequence under the influence of meteorological factors, working days and air temperature were learned through the SAE model. The error sequences generated in the prediction process reflected the secondary characteristics of the load sequence. CEEMDAN algorithm was used to decompose the error sequence into several components. For each component, the BiLSTM model was established to learn the time sequence characteristics of the error sequence. The predicted values of the two models were summed to correct the errors. By comparing the prediction results of several models, it is shown that the combination model of SAE and CEEMDAN-BiLSTM has better accuracy and stability in the short-term power load prediction.
作者
黄炜
陈田
Huang Wei;Chen Tian(College of Mechanics,Shanghai Dianji University,Shanghai 200120,China)
出处
《计算机应用与软件》
北大核心
2022年第7期52-58,共7页
Computer Applications and Software
基金
上海市高峰高原学科项目(A1-5701-18-007-03)
上海市自然科学基金项目(15ZR1417200)。