基于深度学习的非侵入式住宅用电负荷分解方法

发布时间：2018-02-27 02:29

本文关键词： 负荷分解非侵入式负荷监控深度学习评价指标能源节约　出处：《合肥工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：非侵入式住宅用电负荷分解方法是根据住宅中电力入口处的总用电量分解得到具体负荷的用电量和工作状态的一种方法,其分解结果可用于指导用户用电行为,进而有利于促进电力用户节约用电、实现节能降耗。现有的非侵入式家庭电力负荷分解算法主要根据电力负荷的电气特征建立电力负荷模型,然后利用优化技术或模式识别技术实现电力负荷的分解,方法过于单一,同时大多数研究都没有很好地解决高噪声、大功率非平稳负荷波动干扰下多类负荷混合运行中的负荷分解。针对以上问题,本文提出基于深度学习的非侵入式住宅用电负荷分解方法。首先,采集住宅用电负荷总功率和各个负荷的功率,利用数据扩充技术对住宅中所选择典型的负荷消耗数据进行扩充,以产生充足的数据用于训练,提高负荷数据深度学习网络的泛化性能;其次,采用循环神经网络、卷积神经网络及其变种方法建立非侵入式住宅用电负荷分解模型,利用训练数据和该模型自动提取负荷特征,构建总功率和单个负荷消耗功率之间的复杂对应关系,从而根据住宅中总功耗信息可以分解得到每个负荷的使用及消耗功率信息。最后,本文采用2016年5月最新发布的UK-DALE公开数据集作为实验数据集对参与训练的住宅以及未参与训练的住宅分别进行实验,采用总能量正确分类比例、召回率、精确率、F1-measure、准确率以及均方根误差六种评价指标对所提深度模型结构的性能进行评价。实验结果表明所提出的基于深度学习的非侵入式住宅用电负荷分解方法可以有效地实现用电负荷的分解,并对负荷分解性能都得到了很大的改善,且具有较好的泛化性能。
[Abstract]:The method of non-invasive residential power load decomposition is a method to obtain the specific load consumption and working state according to the total electricity consumption at the entrance of the electric power in the residence. The results of the decomposition can be used to guide the user's electricity consumption behavior. The existing non-invasive power load decomposition algorithms are mainly based on the electrical characteristics of the power load to establish a power load model. Then using optimization technology or pattern recognition technology to achieve power load decomposition, the method is too single, and most of the research does not solve the problem of high noise. In view of the above problems, a method of non-invasive residential power load decomposition based on depth learning is proposed. Collect the total power of the residential power load and the power of each load, use the data expansion technology to expand the typical load consumption data selected in the residence, in order to generate sufficient data for training, Secondly, the cyclic neural network, convolution neural network and its variants are used to establish a non-invasive residential power load decomposition model. The training data and the model are used to extract the load characteristics automatically, and the complex corresponding relationship between the total power and the power consumed by a single load is constructed. Thus according to the total power consumption information in the house can be decomposed to obtain the usage and power consumption information of each load. Finally, In this paper, the UK-DALE open data set, which was released recently in May 2016, was used as the experimental data set to conduct experiments on the residential buildings that participated in the training and those without the training. The correct classification ratio of the total energy and the recall rate were used. The performance of the proposed depth model structure is evaluated by six evaluation indexes: accuracy, accuracy and root mean square error. The experimental results show that the proposed non-invasive power load decomposition method based on depth learning can be used in residential buildings. Effectively realize the decomposing of electric load, The load decomposition performance has been greatly improved and has better generalization performance.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM714

【相似文献】