WO2022121744A1 - 基于v‐i轨迹图和神经网络的非侵入式负荷识别方法 - Google Patents

基于v‐i轨迹图和神经网络的非侵入式负荷识别方法 Download PDF

Info

Publication number
WO2022121744A1
WO2022121744A1 PCT/CN2021/134659 CN2021134659W WO2022121744A1 WO 2022121744 A1 WO2022121744 A1 WO 2022121744A1 CN 2021134659 W CN2021134659 W CN 2021134659W WO 2022121744 A1 WO2022121744 A1 WO 2022121744A1
Authority
WO
WIPO (PCT)
Prior art keywords
load
value
neural network
power
trajectory
Prior art date
Application number
PCT/CN2021/134659
Other languages
English (en)
French (fr)
Inventor
于淼
强柱成
陆玲霞
王丙楠
包哲静
Original Assignee
浙江大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江大学 filed Critical 浙江大学
Publication of WO2022121744A1 publication Critical patent/WO2022121744A1/zh
Priority to US18/322,571 priority Critical patent/US20230296654A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R21/00—Arrangements for measuring electric power or power factor
    • G01R21/133—Arrangements for measuring electric power or power factor by using digital technique
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R21/00—Arrangements for measuring electric power or power factor
    • G01R21/001—Measuring real or reactive component; Measuring apparent energy
    • G01R21/003—Measuring reactive component
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06311—Scheduling, planning or task assignment for a person or group
    • G06Q10/063114—Status monitoring or status determination for a person or group
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R22/00—Arrangements for measuring time integral of electric power or current, e.g. electricity meters
    • G01R22/06—Arrangements for measuring time integral of electric power or current, e.g. electricity meters by electronic methods
    • G01R22/10—Arrangements for measuring time integral of electric power or current, e.g. electricity meters by electronic methods using digital techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S20/00—Management or operation of end-user stationary applications or the last stages of power distribution; Controlling, monitoring or operating thereof
    • Y04S20/20—End-user application control systems
    • Y04S20/242—Home appliances

Definitions

  • the invention relates to the field of non-intrusive load monitoring (NILM) technology, in particular to a non-intrusive load identification method based on a V-I trajectory graph and a neural network.
  • NILM non-intrusive load monitoring
  • NILM non-intrusive load monitoring
  • the non-invasive load detection device has a wide application prospect.
  • Some load identification methods use the V-I trajectory characteristics of the load at steady state for load identification, but do not fully utilize the power characteristics of the load. Some methods only use some current harmonic components and the power characteristics of the load at steady state, and do not fully utilize the V-I trajectory characteristics.
  • the present invention proposes a load identification method that can fully utilize both the V-I trajectory feature and the power feature.
  • the technical solutions adopted are as follows:
  • a non-invasive load identification method based on V-I trajectory graph and neural network including the following steps:
  • Step 1 collect the voltage, current and power data of the electricity entering the user terminal in real time, and perform filtering processing
  • Step 2 judge whether a switching event occurs through the bilateral sliding window algorithm, and return to step 1 if no switching event occurs;
  • Step 3 if a switching event is detected, after the load operating state reaches a stable state, the steady-state voltage, current and power data of the load are obtained according to the steady-state data before and after the event;
  • step 4 the V-I track is obtained from the steady-state voltage and current data obtained in step 3, and then the V-I track is converted into an RGB image with a size of 2N*2N; wherein, the power is expressed as the pixel value of the RGB image.
  • Step 5 normalize the RGB image obtained in step 4, and use the recognition network to obtain the load recognition result.
  • the recognition network is composed of a convolutional neural network, and the historical operation data of the electrical equipment and the RGB color map constructed based on the V-I trajectory feature are used as the ground truth for training.
  • the identification network can be set according to the actual situation. If you can directly run on STM32F7 or above MCU to improve the real-time performance of the system, you can construct a simple convolutional neural network model according to the actual situation. For example, the recognition network structure is shown in Figure 2, including two layers of convolution layers and two layers of pooling. layer and three fully connected layers. Or use a computer or server to improve the recognition effect, you can slightly modify the existing neural network model such as the Alexnet model.
  • the specific method for judging the load switching event is:
  • Step 2.1 set two sliding windows, and remove the maximum and minimum values in each window.
  • Step 2.2 Calculate the difference between the average values of the two windows. If the difference is greater than the set threshold, it is considered that a switching event occurs.
  • the method for converting the V-I track into an RGB image with a size of 2N*2N is:
  • Step 4.1 first set the initial value of each pixel to (0,0,0);
  • Step 4.2 according to the obtained steady-state voltage and current of the load, obtain the maximum value Umax and Imax of the absolute value of the voltage and current.
  • Step 4.5 Set the size of the corresponding pixel value according to the size of the active power of the load.
  • step 4.2 for the high-power load, Umax and Imax are directly set to fixed values, and the fixed value is greater than the value of Umax or Imax in the high-power load. In this way, the V-I trace can contain most of the current data.
  • the V-I trajectory is divided into three stages, and then the color information of the pixels in each stage is set differently.
  • the V-I trajectory characteristic map obtained in this way can largely reflect the phase difference, impedance characteristics and power level of voltage and current. Because the Umax settings of high power and low power are different when forming the RGB feature map, it is easy to identify high power and low power loads. For low-power loads, the shape and brightness of the RGB feature map are used for identification (brightness includes power information). different) can be identified.
  • the present invention can make full use of both the V-I trajectory feature and the power feature by constructing an RGB map, and then perform load identification. This method can fully identify the small power load and the high power load. For small power loads, loads with similar power are separated according to the shape of the load's trajectory, and loads with similar shapes are separated according to the power value of the load. For high-power loads, it is mainly based on the shape of the load's trajectory. The overall recognition effect is better.
  • Fig. 1 is the flow chart of the method of the present invention
  • FIG. 2 is a schematic structural diagram of a convolutional neural network model in an embodiment of the present invention.
  • Fig. 3 is the flow chart of bilateral sliding window algorithm
  • FIG. 4 is a V-I trajectory-based feature diagram of some loads (left: air conditioner, middle: refrigerator, right: electric light) in an embodiment of the present invention.
  • Figure 5 is a gray image of three channels R, G, and B in the air conditioner feature map (left: R channel, middle: G channel, right: B channel)
  • the present invention provides a non-invasive load identification method based on V-I trajectory graph and neural network, as shown in Figure 1, the implementation steps include:
  • S1 First extract 5 kinds of household electrical equipment from the BLUED dataset, then construct an RGB color map based on V-I trajectory features and train a convolutional neural network model as a recognition network, such as the Alexnet model, etc.
  • the recognition network model in this embodiment is shown in the figure.
  • the built recognition network model is not complicated, including two convolution layers, two pooling layers and three full layers.
  • the connection layer, the specific structure is shown in Figure 2.
  • S2 Collect the voltage, current and power data of the power input terminal in real time, and filter the obtained voltage and current data; the voltage and current sampling frequency of the BLUED public data set in this embodiment is 12KHz, and the power value frequency is 60Hz. A cycle contains 200 sampling points.
  • S3 Determine whether there is a switching event through the bilateral sliding window algorithm.
  • the specific parameters in this embodiment are as follows: two sliding windows with a window size of 5 and 5 are set, and the maximum value and the minimum value are removed from each window. Calculate the mean for each window, and calculate the difference between the mean. It is then compared with a pre-set threshold. If the difference between the average values is greater than the set threshold, it is considered that a switching event occurs. The judgment process is shown in FIG. 3 .
  • S5 Obtain the V-I track from the steady-state voltage and current data obtained in S4, and then convert the V-I track into an RGB image with a size of 2N*2N.
  • N is 32; specifically, the following steps are included:
  • the value of the (Xj, Yj) pixel is set to (color_value, 0, 0)
  • the value of the (Xj, Yj) pixel is set to (0, color_value, 0)
  • the value of the (Xj, Yj) pixel is set to (0, 0, color_value)
  • the V-I trajectory characteristic map obtained in this way can reflect the load characteristics such as the phase difference of voltage and current, impedance characteristics and power size.
  • Figure 4 below is a graph of the trajectory characteristics of some loads in this embodiment.
  • the high-power load and low-power load in Figure 4 can also be seen directly by the naked eye, in which the power of the electric lamp is small, so the corresponding brightness is small.
  • the Umax of the high-power load is directly set to 400V
  • the trajectory feature map is concentrated in the middle, and the feature map of the low-power load covers the entire area.
  • each load trajectory feature map is composed of three colors of red, green and blue (Fig. 5 is a schematic diagram of the three channels of R, G, B in the trajectory of the air conditioning feature map being separated), and has a direction, the brightness of the low-power load feature map Also according to the size of the power is not the same.
  • S6 Normalize the RGB image obtained in S5, input it into the pre-trained convolutional neural network, and obtain the recognition result. Because the input end of the neural network is a picture, the normalization processing in this embodiment is very simple, and the value of each pixel point can be directly divided by 255.
  • the RGB image in the present invention already contains information such as V-I track feature, voltage and current phase difference, active power, etc., so the recognition effect is much better than the method of using V-I track or power information alone.
  • the convolutional neural network used is not complicated, so it can run directly on the embedded device, so it can improve the real-time performance and does not depend on the computing support of the server.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Health & Medical Sciences (AREA)
  • Power Engineering (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Water Supply & Treatment (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Public Health (AREA)
  • Image Analysis (AREA)
  • Remote Monitoring And Control Of Power-Distribution Networks (AREA)

Abstract

本发明申请了一种基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，所述方法包括：实时采集入户的电压和电流数据以及有功功率数据；通过有功功率的变化来判断有无负荷投切事件以及负荷运行状态有没有达到稳定状态；根据事件前后的稳态电压电流数据获取负荷的电压电流数据以及功率数据；采用简单的图像处理技术把V‐I轨迹转换成包含电压电流相位差、功率等信息的RGB彩色图像。得到RGB彩色图像之后，进行归一化处理；通过事先训练好的卷积神经网络进行负荷识别。与现有技术相比，本发明通过卷积神经网络充分提取负荷的稳态特征，而且神经网络模型能够在嵌入式设备上直接运行，不需要依赖于服务器的运算支持。

Description

基于V‐I轨迹图和神经网络的非侵入式负荷识别方法 技术领域
本发明涉及非侵入式负荷识别技术(non‐intrusive load monitoring,NILM)领域，尤其涉及一种基于V‐I轨迹图和神经网络的非侵入式负荷识别方法。
背景技术
负荷识别方法主要有侵入式负荷识别和非侵入式负荷识别两大类。虽然侵入式负荷识别方法的识别结果较为准确，但是成本高等原因不太受欢迎。而非侵入式负荷识别方法(non‐intrusive load monitoring,NILM)成本低而且实用性强，所以NILM成为了当今电力系统智能计量领域的热点。通过在入户电表上安装嵌入式非侵入电力识别模块，然后通过负荷识别算法来检测建筑内的负荷工作情况。结合有效的电源管理，可以在不影响用户体验的情况下实现省电节能。
研究表明如果给消费者提供建筑内能耗的实际能耗，可以激发消费者节能的动力，据统计可以有效节能10％‐20％。因此非侵入式负荷检测装置拥有广泛应用前景。
现阶段的NILM大部分方法都没有充分利用电器负荷的稳态特征，而且都把云服务器作为数据处理中心，很多识别运算都依赖于服务器。有些负荷识别方法利用负荷稳态时的V‐I轨迹特征来进行负荷识别，没有充分利用到负荷的功率特征。还有些方法只利用稳态时的一些电流谐波分量以及负荷的功率特征，没有充分利用V‐I轨迹特征。
发明内容
本发明针对现有技术的不足，提出了一种能够把V‐I轨迹特征和功率特征都充分利用的负荷识别方法，采用的技术方案具体如下：
一种基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，包括如下步骤：
步骤1，实时采集用电入户端的电压和电流以及功率数据，并进行滤波处理；
步骤2，通过双边滑动窗口算法判断是否发生投切事件，若无投切事件发生则返回步骤1；
步骤3，若检测到发生投切事件，则待负荷运行状态达到稳定后，根据事件前后的稳态数据获取负荷的稳态电压、电流以及功率数据；
步骤4，从步骤3得到的稳态电压、电流数据获得V-I轨迹，然后把V-I轨迹转换成大 小为2N*2N的RGB图像；其中，功率表示为RGB图像的像素值。
步骤5，将步骤4得到的RGB图像进行归一化处理，并利用识别网络得到负荷识别结果。其中，所述识别网络由卷积神经网络构成，采用电器设备的历史运行数据及其构造的基于V-I轨迹特征的RGB彩色图作为真值进行训练。
进一步地，识别网络可以根据实际情况来设置。在STM32F7等以上的MCU上直接运行而提高系统实时性的话，可以根据实际情况自己构造简单的卷积神经网络模型，例如识别网络结构为图2所示，包括两层卷积层、两层池化层以及三层全连接层。或者利用电脑或服务器来提高识别效果，则可以稍微改动现有的神经网络模型比如Alexnet模型等。
进一步地，所述步骤2中，判断负荷投切事件的具体方法为：
步骤2.1，设置两个滑动窗口，在每个窗口中去掉最大值和最小值。
步骤2.2，计算两个窗口平均值之差，若差值大于设定的阈值则认为发生投切事件。
进一步地，所述步骤4中，把V-I轨迹转换成大小为2N*2N的RGB图像的方法为：
步骤4.1，首先设置每个像素点的初始值为(0,0,0)；
步骤4.2，根据获取的负荷的稳态电压、电流，获得电压电流绝对值的最大值Umax和Imax。
步骤4.3，计算Δu＝Umax/N和Δi＝Imax/N
步骤4.4，对于每个采样点的(Uj，Ij)(0<j≤sample，sample为每个周期里的采样点数)，计算
Figure PCTCN2021134659-appb-000001
Yj＝N+int(Ij/Δi)作为要具体设置的RGB像素点坐标，不需要轨迹的连续化处理。
步骤4.5根据负荷有功功率大小，设置对应像素值的大小。
进一步地，所述步骤4.2中，对于大功率负荷，其Umax和Imax直接设置成固定值，所述固定值为大于大功率负荷中Umax或Imax的值。这样V-I轨迹能够把大部分的电流数据都能包含进去。
进一步地，为了在RGB图像中充分反映负荷的稳态特征，V-I轨迹分成三个阶段，然后每阶段里像素点的颜色信息设置得不一样。这样得到的V-I轨迹特征图很大程度上能够反映电压和电流的相位差、阻抗特征以及功率大小等特征。因为构成RGB特征图的时候，大功率和小功率的Umax设置不一样，所以很容易识别出大功率和小功率负荷。对于小功率负荷根据RGB特征图的形状和亮度(亮度包含功率信息)进行识别，对于大功率的负荷根据RGB特征图的形状(功率不一样的话，因为电流大小不一样，所以特征图的形状也不一样)可以进行识别。
本发明的有益效果是：本发明通过构建RGB图能够把V‐I轨迹特征和功率特征都充分利用起来，进而进行负荷识别。该方法能够充分的识别出小功率负荷和大功率负荷。对于小功率负荷，功率类似的负荷根据负荷的轨迹形状来分别，形状类似的负荷根据负荷的功率值来分别。对于大功率负荷主要是根据负荷的轨迹形状来分别。整体的识别效果更好。
附图说明
图1为本发明方法流程图；
图2为本发明实施例中的卷积神经网络模型结构示意图；
图3为双边滑动窗口算法流程图；
图4为本发明实施例中一些负荷(左：空调，中：冰箱，右：电灯)的基于V-I轨迹的特征图。
图5为空调特征图中把R,G,B三个通道分别表示的灰色图(左：R通道，中：G通道，右：B通道)
具体实施方式
结合附图以及利用BLUED公共数据集的实施方式来解释本发明，具体实施步骤如下：
本发明提供了一种基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，如图1所示，其实施步骤包括：
S1：先从BLUED数据集提取5种家用电气设备，然后构造基于V-I轨迹特征的RGB彩色图并训练卷积神经网络模型作为识别网络，例如Alexnet模型等。在本实施例中的识别网络模型如图所示，为了能够在STM32F7以上的MCU上直接运行，搭建的识别网络模型并不复杂，包括两层卷积层、两层池化层以及三层全连接层，具体的结构如图2所示。
S2：实时采集用电入户端的电压和电流以及功率数据，对获得的电压电流数据进行滤波处理；本实施例中的BLUED公共数据集的电压电流采样频率为12KHz,功率值频率为60Hz，每个周期包含200个采样点。
S3：通过双边滑动窗口算法判断有没有发生投切事件。本实施例中的具体参数如下：设置窗口大小为5，5的两个滑动窗口，在每个窗口中去掉最大值和最小值。计算每个窗口的平均值，并计算平均值之差。然后跟事先设定好的阈值进行比较。若平均值之差大于设定的阈值则认为发生投切事件。该判断过程如图3所示。
S4：负荷连续3次达到稳态状态以后，获取负荷的稳态电压电流数据以及功率数据；
S5：从S4得到的稳态电压电流数据获得V-I轨迹，然后把V-I轨迹转换成大小为2N*2N的RGB图像，本实施例中，N为32；具体包括如下步骤：
(1)首先每个像素点的初始值设置为(0,0,0)。
(2)对小功率负荷求得电压电流绝对值的最大值Umax和Imax，对大功率负荷Imax直接设置固定值，使其能够让V-I轨迹包含所有电流信息。本实施例中将有功功率值小于510W的负荷看作小功率负荷，其他看作大功率负荷。对于大功率负荷设置Umax的固定值为400V，Imax的固定值为20A。这样V-I轨迹能够把大部分的电流数据都能包含进去。
(3)计算Δu＝Umax/N和Δi＝Imax/N。
(4)对于每个采样点(Uj，Ij)(0＜j≤200)计算
Figure PCTCN2021134659-appb-000002
Yj＝N+int(Ij/Δi)，不需要轨迹的连续化处理。
(5)根据负荷有功功率大小，设置对应像素值的大小。有功功率P大于510W的时候，因为功率值比较大的电器设备特征比较明显，利用一般的V-I轨迹特征也可以正确识别出来，所以每个像素点的值直接设置color_value＝255。有功功率P小于510W的时候，设置color_value＝int(P/2)。
(6)为了在RGB图像中充分反映负荷的稳态特征，设置(Xj,Yj)像素点值的具体过程如下；
If 0＜j＜200/3:
(Xj,Yj)像素点的值设置为(color_value，0，0)
If 200/3＜j＜2*200/3:
(Xj,Yj)像素点的值设置为(0，color_value，0)
else：
(Xj,Yj)像素点的值设置为(0，0，color_value)
这样得到的V-I轨迹特征图就能够反映电压和电流的相位差、阻抗特征以及功率大小等负荷特征。下面的图4是本实施例中的一些负荷的轨迹特征图。从图4中大功率负荷和小功率负荷直接通过肉眼也可看出来，其中电灯的功率较小，因而对应的亮度较小。因为大功率负荷的Umax直接设置成400V，所以轨迹特征图比较集中于中间，小功率负荷的特征图覆盖整个区域。然后每个负荷轨迹特征图由红色，绿色和蓝色三种颜色组成(图5空调特征图中轨迹的R，G，B三个通道分开示意图)，而且具有方向，小功率负荷特征图的亮度也是按照功率大小不一样的。
S6：将S5得到的RGB图像进行归一化处理，输入到事先训练好的卷积神经网络，并得到识别结果。因为神经网络的输入端是一张图片，所以本实施例中的归一化处理很简单，直接每个像素点的值除以255就可以。本发明中的RGB图像已经包含了V-I轨迹特征、电 压电流相位差、有功功率等信息，所以识别效果比单独使用V-I轨迹或者功率信息的方法好很多。而且所用到的卷积神经网络并不复杂，所以能够在嵌入式设备上直接运行，所以能提高实时性，不依赖于服务器的运算支持。
显然，上述实施例仅仅是为清楚地说明所作的举例，而并非对实施方式的限定。对于所属领域的普通技术人员来说，在上述说明的基础上还可以做出其他不同形式的变化或变动。这里无需也无法把所有的实施方式予以穷举。而由此所引申出的显而易见的变化或变动仍处于本发明的保护范围。

Claims (4)

  1. 一种基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，包括以下的步骤：
    步骤1，实时采集用电入户端的电压和电流以及功率数据，并进行滤波处理；
    步骤2，通过双边滑动窗口算法判断是否发生投切事件，若无投切事件发生则返回步骤1；
    步骤3，若检测到发生投切事件，则待负荷运行状态达到稳定后，根据事件前后的稳态数据获取负荷的稳态电压、电流以及功率数据；
    步骤4，从步骤3得到的稳态电压、电流数据获得V-I轨迹，然后把V-I轨迹转换成大小为2N*2N的RGB图像；其中，功率表示为RGB图像的像素值；其中，把V-I轨迹转换成大小为2N*2N的RGB图像的方法为：
    步骤4.1，首先设置每个像素点的初始值为(0,0,0)；
    步骤4.2，根据获取的负荷的稳态电压、电流，获得电压电流绝对值的最大值Umax和Imax；
    步骤4.3，计算Δu＝Umax/N和Δi＝Imax/N
    步骤4.4，对于每个采样点的(Uj，Ij)，0<j≤sample，sample为每个周期里的采样点数，计算
    Figure PCTCN2021134659-appb-100001
    Yj＝N+int(Ij/Δi)作为要具体设置的RGB像素点坐标；
    步骤4.5，根据负荷有功功率大小，设置对应像素值的大小；将V-I轨迹分成三个阶段，然后将每阶段里像素点设置成三种不同的颜色，设置的过程具体为：
    If 0＜j＜sample/3:
    (Xj,Yj)像素点的值设置为(color_value，0，0)
    If sample/3＜j＜2*sample/3:
    (Xj,Yj)像素点的值设置为(0，color_value，0)
    else：
    (Xj,Yj)像素点的值设置为(0，0，color_value)
    其中，color_value表示设置的像素点的值；
    步骤5，将步骤4得到的RGB图像进行归一化处理，并利用识别网络得到负荷识别结果；其中，所述识别网络由卷积神经网络构成，采用电器设备的历史运行数据及其构造的基于V-I轨迹特征的RGB彩色图作为真值进行训练。
  2. 根据权利要求1所述的基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，其特征在于，所述识别网络结构由两层卷积层、两层池化层以及三层全连接层组成，在STM32F7 的MCU上直接运行；或者采用Alexnet模型，在电脑或服务器中运行。
  3. 根据权利要求1所述的基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，其特征在于，所述步骤2中，判断负荷投切事件的具体方法为：
    步骤2.1，设置两个滑动窗口，在每个窗口中去掉最大值和最小值；
    步骤2.2，计算两个窗口平均值之差，若差值大于设定的阈值则认为发生投切事件。
  4. 根据权利要求1所述的基于V‐I轨迹图和神经网络的非侵入式负荷识别方法，其特征在于，所述步骤4.2中，对于大功率负荷，其Umax和Imax直接设置成固定值，所述固定值为大于大功率负荷中Umax或Imax的值。
PCT/CN2021/134659 2020-12-08 2021-11-30 基于v‐i轨迹图和神经网络的非侵入式负荷识别方法 WO2022121744A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/322,571 US20230296654A1 (en) 2020-12-08 2023-05-23 Non-intrusive load monitoring method based on v-i trajectory and neural network

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011443609.8 2020-12-08
CN202011443609.8A CN112418722B (zh) 2020-12-08 2020-12-08 基于v-i轨迹图和神经网络的非侵入式负荷识别方法

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/322,571 Continuation US20230296654A1 (en) 2020-12-08 2023-05-23 Non-intrusive load monitoring method based on v-i trajectory and neural network

Publications (1)

Publication Number Publication Date
WO2022121744A1 true WO2022121744A1 (zh) 2022-06-16

Family

ID=74775510

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/134659 WO2022121744A1 (zh) 2020-12-08 2021-11-30 基于v‐i轨迹图和神经网络的非侵入式负荷识别方法

Country Status (3)

Country Link
US (1) US20230296654A1 (zh)
CN (1) CN112418722B (zh)
WO (1) WO2022121744A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116359602A (zh) * 2023-03-07 2023-06-30 北京智芯微电子科技有限公司 非侵入式电动车充电识别方法、装置、介质和智能电表
CN117349639A (zh) * 2023-12-04 2024-01-05 国网江西省电力有限公司供电服务管理中心 电动自行车充电行为的辨识方法及装置

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112418722B (zh) * 2020-12-08 2021-09-10 浙江大学 基于v-i轨迹图和神经网络的非侵入式负荷识别方法
CN113036759B (zh) * 2021-03-31 2023-09-26 上海电力大学 一种电力用户负荷细粒度识别方法及识别系统
CN113191253A (zh) * 2021-04-28 2021-07-30 浙江大学 一种边缘机器学习下基于特征融合的非侵入式负荷识别方法
CN113393121A (zh) * 2021-06-15 2021-09-14 贵州电网有限责任公司 一种基于负荷电力指纹特征的非侵入式负荷识别方法
CN113537327A (zh) * 2021-07-06 2021-10-22 江阴长仪集团有限公司 基于Alexnet神经网络和颜色编码的非侵入式负荷识别方法及系统
CN113255236B (zh) * 2021-07-07 2021-10-08 浙江大学 基于孪生网络的非侵入式负荷自适应识别方法
CN113723479A (zh) * 2021-08-18 2021-11-30 南京工程学院 一种基于grnn与均值漂移算法的非侵入式负荷识别方法
CN113919578B (zh) * 2021-10-18 2022-10-04 四川大学 一种基于图神经网络的个人用户短期负载预测方法
CN114325081A (zh) * 2021-12-29 2022-04-12 润建股份有限公司 一种基于多模态特征的非侵入式负荷识别方法
CN114330461B (zh) * 2022-03-16 2022-06-24 北京智芯微电子科技有限公司 用于非侵入负荷识别的v-i轨迹生成方法、装置及神经网络
CN114662624B (zh) * 2022-05-25 2022-09-13 浙江大学 基于可再训练孪生网络的非侵入式负荷识别方法及系统
CN115601603B (zh) * 2022-11-29 2023-04-07 北京志翔科技股份有限公司 模型训练和用电器类别识别方法、设备及存储介质
CN117169657B (zh) * 2023-11-03 2024-01-12 山东和兑智能科技有限公司 一种基于人工智能的高压电缆状态监测方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130123882A1 (en) * 2009-01-26 2013-05-16 Arizona Board of Regents, a Body Corporate of the State of Arizona Acling for and on Behalf of Arizo Dipolar antenna system and related methods
CN110188771A (zh) * 2019-05-17 2019-08-30 湖南大学 基于图像处理的非侵入式电力负荷特征提取、识别方法、系统及介质
CN112180193A (zh) * 2020-09-28 2021-01-05 华中科技大学 一种基于轨迹图像识别的非侵入式负荷识别系统及方法
CN112418722A (zh) * 2020-12-08 2021-02-26 浙江大学 基于v-i轨迹图和神经网络的非侵入式负荷识别方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11195057B2 (en) * 2014-03-18 2021-12-07 Z Advanced Computing, Inc. System and method for extremely efficient image and pattern recognition and artificial intelligence platform
CN111864896B (zh) * 2019-04-29 2022-04-15 清华大学 一种电力负荷监测方法及系统
CN111766462B (zh) * 2020-05-14 2023-03-10 中国计量大学 一种基于v-i轨迹的非侵入式负荷识别方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130123882A1 (en) * 2009-01-26 2013-05-16 Arizona Board of Regents, a Body Corporate of the State of Arizona Acling for and on Behalf of Arizo Dipolar antenna system and related methods
CN110188771A (zh) * 2019-05-17 2019-08-30 湖南大学 基于图像处理的非侵入式电力负荷特征提取、识别方法、系统及介质
CN112180193A (zh) * 2020-09-28 2021-01-05 华中科技大学 一种基于轨迹图像识别的非侵入式负荷识别系统及方法
CN112418722A (zh) * 2020-12-08 2021-02-26 浙江大学 基于v-i轨迹图和神经网络的非侵入式负荷识别方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116359602A (zh) * 2023-03-07 2023-06-30 北京智芯微电子科技有限公司 非侵入式电动车充电识别方法、装置、介质和智能电表
CN116359602B (zh) * 2023-03-07 2024-05-03 北京智芯微电子科技有限公司 非侵入式电动车充电识别方法、装置、介质和智能电表
CN117349639A (zh) * 2023-12-04 2024-01-05 国网江西省电力有限公司供电服务管理中心 电动自行车充电行为的辨识方法及装置
CN117349639B (zh) * 2023-12-04 2024-03-08 国网江西省电力有限公司供电服务管理中心 电动自行车充电行为的辨识方法及装置

Also Published As

Publication number Publication date
CN112418722A (zh) 2021-02-26
CN112418722B (zh) 2021-09-10
US20230296654A1 (en) 2023-09-21

Similar Documents

Publication Publication Date Title
WO2022121744A1 (zh) 基于v‐i轨迹图和神经网络的非侵入式负荷识别方法
CN111864896B (zh) 一种电力负荷监测方法及系统
Wang et al. Recognition and drop-off detection of insulator based on aerial image
CN107730477B (zh) 变电站设备红外智能监测系统的监测图像特征提取方法
CN104374052A (zh) 基于神经网络的中央空调节能控制方法及系统
CN113036759B (zh) 一种电力用户负荷细粒度识别方法及识别系统
CN108154492B (zh) 一种基于非局部均值滤波的图像去雾霾方法
CN103606134A (zh) 一种弱光照视频图像的增强方法
CN102592144A (zh) 一种基于多摄像机无重叠视域行人匹配方法
CN113191253A (zh) 一种边缘机器学习下基于特征融合的非侵入式负荷识别方法
CN104796582A (zh) 基于随机喷射retinex的视频图像去噪与增强方法及装置
CN109215042A (zh) 一种基于计算机视觉的光伏电池板热斑效应检测系统及其计算方法
CN104616026A (zh) 一种面向智能视频监控的监控场景类型辨识方法
CN112216061A (zh) 一种雨水情监测预警方法及系统
CN110427868A (zh) 一种行人重识别中的特征提取方法
CN101702234A (zh) 多目摄像机视频人脸最佳视角的选择方法
CN115601603B (zh) 模型训练和用电器类别识别方法、设备及存储介质
CN107895365B (zh) 一种输电通道外破防护的图像匹配方法及监测系统
CN116466164A (zh) 基于微型机器学习的非侵入式用电负荷分解方法及系统
CN115759664A (zh) 一种基于电能使用行为的用户碳画像方法
CN110502967A (zh) 基于人员大数据的目标场景人工智能匹配方法和装置
CN114461662A (zh) 居民用户需求响应高潜力用户筛选方法及系统
CN113537330B (zh) 一种非侵入式负荷识别方法
CN102855843A (zh) 面向oled显示器的用于三维绘制的颜色集节能优化方法
CN107918941B (zh) 一种用于输电通道外破防护的可视化监测系统及方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21902451

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21902451

Country of ref document: EP

Kind code of ref document: A1