PDF《第六届“泰迪杯” 数据挖掘挑战赛》

s electricity consumption behavior, which in turn facilitates the electricity users to save electricity and achieve energy conservation and consumption reduction. In this paper, we use MATLAB to carry out wavelet noise reduction processing on the relevant data and remove the abnormity point and other preconditioning steps. The first problem is to analyze the preprocessed data, extract the transient characteristics of a single device, the four characteristics of active power and reactive power, current harmonic, voltage current track (V-I trajectory), and give the calculation and extraction method of characteristic eigenvalues of each load. The formula is used to calculate the real-time power consumption of each single device. The results are detailed in Annex 1. The second problem, we establish a model to identify any single device. The similarity degree based load identification model is established by the four types of load characteristics extracted by the first problem. Firstly, the feature similarity is defined to represent the similarity degree of any two equipment under a certain feature. The entropy method is used to determine the similarity weight coefficient of each characteristic. Finally, the feature similarity is weighted to determine the total characteristics. Choose to match the highest similarity device with the unknown device. Finally, a formula is applied to calculate the real-time power consumption of the unknown equipment. The results are detailed in Annex 2. The third problem is state recognition based on event detection and non-intrusive power load decomposition. First, the time point of the event can be detected. In this paper, the change value of active power is used as the basis of the event detection, and the process and the specific processing method of the algorithm are given. Then the 0-1 programming model is established for multi load running status recognition, and a non-intrusive load identification algorithm based on continuous 0-1 two times programming is proposed. The device group of the known operation records of Annex

3 validates the effectiveness of the above method. Finally, the real-time electricity consumption estimation model under multi load is established, and the real-time power consumption of each device is calculated. The calculation results are detailed in Annex 3. The fourth problem, using the decision tree based non-intrusive power load identification method, proposes to use the decision tree algorithm and the database to realize the load identification from the sampling data of different combined load running. The characteristic parameters of the load identified by the load identification are compared with the characteristic parameters of the known load in the database, and the nearest similar identification results are selected to determine the equipment composition of the equipment group, and then the state of each device is determined by the 0-1 programming model of problem three. Finally, the real-time power consumption estimation model based on multi load is applied to calculate the real-time power consumption of each device in the device group. The results are detailed in Annex 4. Key words: non-intrusive;

load identification;

load decomposition;

event detection;

decision 目录 1.研究背景与意义.4 2.变量说明.4

3 问题分析.5 4.问题一

6 4.1 数据预处理.6 4.1.1 降噪处理.6 4.1.2 数据变换.7 4.2 负荷特征分析.9 4.2.1 暂态特征.9 4.2.2 稳态特征.11 5.问题二

17 5.1 相似度与权系数.17 5.2 模型建立.18 5.3 模型求解.19 6.问题三