CN114462501A - Non-invasive equipment state identification method and device based on federal learning - Google Patents

Non-invasive equipment state identification method and device based on federal learning Download PDF

Info

Publication number
CN114462501A
CN114462501A CN202210001823.0A CN202210001823A CN114462501A CN 114462501 A CN114462501 A CN 114462501A CN 202210001823 A CN202210001823 A CN 202210001823A CN 114462501 A CN114462501 A CN 114462501A
Authority
CN
China
Prior art keywords
local user
data
network
user node
local
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202210001823.0A
Other languages
Chinese (zh)
Inventor
薛广涛
童侠通
潘昊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
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 Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN202210001823.0A priority Critical patent/CN114462501A/en
Publication of CN114462501A publication Critical patent/CN114462501A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/602—Providing cryptographic facilities or services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/606—Protecting data by securing the transmission between two devices or processes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5072—Grid computing
    • 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
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Security & Cryptography (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioethics (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention relates to a non-invasive equipment state identification method and a non-invasive equipment state identification device based on federal learning, wherein the method comprises the following steps: arranging a non-invasive equipment data acquisition system in a plurality of local user nodes, carrying out high-frequency sampling on total current data of trunk circuits in the local user nodes, and carrying out low-frequency sampling on power data of each parallel branch circuit in the local user nodes; constructing a historical data set; constructing a deep learning neural network, arranging a feature extraction sub-network in a local user node, and arranging an equipment prediction sub-network in a cloud; training a deep learning neural network through a historical data set, and performing data transmission between a local user node and a cloud end through an encryption algorithm in the training process; and recognizing the state of the equipment by using the recognition model obtained by training. Compared with the prior art, the method solves the problem of data reliability, the problem of data transmission bottleneck under high-frequency sampling and the problem of user privacy safety in the training and predicting process.

Description

Non-invasive equipment state identification method and device based on federal learning
Technical Field
The invention relates to the field of non-intrusive equipment identification, in particular to a method and a device for identifying a state of non-intrusive equipment based on federal learning.
Background
In complex power systems with a large number of electrical devices, such as buildings, identifying and determining the state of the devices is a major concern for researchers. Non-invasive device identification is a method of measuring only the summary data using only one set of sensors to obtain the operating status of various devices in the internal system. The method does not need to relate to a plurality of groups of sensors, has the characteristics of convenient deployment, low price and easy popularization, and is widely applied.
In the non-invasive device identification method, the first type is to analyze 50Hz alternating current from the point of view of the collected signals, and the sampling frequency is from a fraction of Hz to several Hz. Generally, such low frequency data is used to analyze the state switching (e.g., on and off) between steady states, and the data itself is mainly average power, power factor, etc. The second type is to analyze 50Hz AC and its harmonics, with sampling frequencies from 1kHz to tens of kHz. Present household appliances are generally non-linear, such as televisions and LED lamps, and therefore contain important harmonic distortion information in the waveform of the current. To capture such detailed information, a higher sampling rate is necessary. The third type is to analyze high frequency transient signals, the sampling frequency is from tens of kHz to tens of MHz, most electronic devices use a Switch Mode Power Supply (SMPS), and the frequency band is in this interval. From a data processing perspective, most of the existing research work is focused on supervised machine learning techniques, including traditional markov models and deep learning models.
In data acquisition, the data acquisition can be a high-frequency transient signal or a low-frequency steady-state signal. Undoubtedly, the high-frequency transient signal contains more information, which can provide more accurate identification, and meanwhile, the low-frequency steady-state signal can be obtained by sampling, which is undoubtedly the first choice of data acquisition. However, high frequency data sampling will result in large data transmission flow, will replace the effort bottleneck and become a bigger bottleneck, and significantly limits the carrying capacity of the server.
The deep learning model is obviously superior in data processing. Invasive device identification systems require a large amount of training data from a large number of users, and require that the data of different users be aggregated to form a training set for neural network training. In the data summarization process, how to protect the privacy of each user becomes a challenge that must be faced. A potential attacker can steal original training data, identify the use condition of the equipment, further deduce the state information of the user and further achieve the purpose of stealing the privacy of the user.
In summary, large-traffic data transmission and user privacy security are problems to be solved urgently in the non-invasive device state identification method.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a method and a device for identifying the state of non-invasive equipment based on federal learning.
The purpose of the invention can be realized by the following technical scheme:
a non-intrusive equipment state identification method based on federal learning comprises the following steps:
s1, arranging a non-invasive equipment data acquisition system in a plurality of local user nodes, carrying out high-frequency sampling on total current data of trunk circuits in the local user nodes, and carrying out low-frequency sampling on power data of each parallel branch circuit in the local user nodes, wherein the local user nodes are in environments such as families, laboratories, offices and the like;
s2, respectively constructing historical data sets according to the data of each local user node;
s3, constructing a deep learning neural network, wherein the deep learning neural network comprises a feature extraction sub-network and a device prediction sub-network, the feature extraction sub-network is arranged in a local user node, and the device prediction sub-network is arranged in a cloud end;
s4, training the deep learning neural network through a historical data set, and performing data transmission between the local user node and the cloud end through an encryption algorithm in the training process to obtain a trained recognition model;
and S5, recognizing the equipment state by using the trained recognition model, and monitoring the running state of each equipment.
Further, the non-invasive device data collecting system in step S1 non-invasively collects total current data of the trunk in the local user node at a frequency of 180kHz, and collects power data of each parallel branch in the local user node at a frequency of 1Hz, where each device in the local user node is connected to the parallel branch.
Further, step S2 includes the following steps:
s21, acquiring total current data and power data continuously acquired by a non-invasive equipment data acquisition system;
s22, setting the minimum starting power P of each devicemin_startMaximum stopping power Pmax_endMinimum duration tminAnd a maximum interruption time tmaxProcessing the power data, and judging the working state of each device according to the power data by using a streaming algorithm, wherein the working state comprises starting and closing;
s23, introducing a sliding window, converging the working state of each device by applying a streaming algorithm to the data between the current time t and the previous time t-1, retaining valid and effective data according to a preset judgment rule, adding a label to the device according to the working state, and constructing a historical data set in each local user node.
Further, the preset judgment rule is as follows: in the same window, if the working states of all the devices are not switched from starting to closing or from closing to starting, the data of the window is considered to be legal; within the same window, if all devices are in the off state, the window is considered invalid.
Furthermore, the deep learning neural network is used for realizing multi-label classification, the feature extraction sub-network is a residual error network with a basic module as a one-dimensional convolution unit, the equipment prediction sub-network is a residual error network of a multilayer perceptron, a Loss function used in multi-label classification is Asymmetric Loss, the feature extraction sub-network and the equipment prediction sub-network are simplified and optimized by methods such as model compression, pruning and the like, relevant parameters are adjusted, accuracy is guaranteed, and meanwhile, the calculation amount is reduced, so that the feature extraction sub-network and the equipment prediction sub-network can be deployed on edge equipment.
Further, the training process of the deep learning neural network is as follows:
s41, for each local user node, acquiring and preprocessing a historical data set, and sending the historical data set into a feature extraction sub-network to obtain a calculation result;
s42, uploading the calculation results in each local user node to the cloud after being encrypted through an encryption algorithm;
s43, the cloud end collects the calculation results of all local user nodes, trains the equipment prediction sub-network at the cloud end, and transmits the gradient back to each local user node;
and S44, the local user node extracts the sub-network according to the received gradient updating characteristics.
Further, the encryption algorithm is a differential encryption algorithm.
Further, the cloud end transmits the gradient back to each local user node, the structures of the feature extraction sub-networks of each local user node are mutually independent, and the updating of the feature extraction sub-networks of each local user node is mutually independent.
A non-intrusive equipment state identification device based on federal learning comprises the following components:
the non-invasive equipment data acquisition system is respectively arranged in each local user node and is used for carrying out high-frequency sampling on total current data of a trunk circuit in the local user node and carrying out low-frequency sampling on power data of each parallel shunt circuit in the local user node;
the local data set building module is respectively arranged in each local user node and used for respectively building a historical data set according to the data of each local user node;
the local computing units are respectively arranged in each local user node, are in communication connection with the local data set building module, and store a feature extraction sub-network for computing and updating the feature extraction sub-network in the training process;
the cloud computing unit is arranged at the cloud, stores the equipment prediction sub-network and is used for computing and updating the equipment prediction sub-network in the training process, and the local computing unit and the cloud computing unit perform data transmission through an encryption algorithm in the training process;
and the detection module is used for identifying the equipment state by using the trained identification model.
Furthermore, a high-frequency digital-to-analog conversion chip is used for non-invasively acquiring total current data of a trunk circuit in the local user node at the frequency of 180kHz, an intelligent socket with a power measurement function is used for acquiring power data of each parallel branch circuit in the local user node at the frequency of 1Hz, and all devices in the local user node are connected into the parallel branch circuits.
Further, the total current data is transmitted to the local data set building module in a wired transmission mode, the intelligent socket has a wireless communication function, and the power data is transmitted to the local data set building module in a wireless transmission mode.
Compared with the prior art, the invention has the following beneficial effects:
(1) the total current is sampled at high frequency, data processing is carried out through a deep learning neural network, more information can be kept through high-frequency sampling, power data on parallel branches of the equipment are collected simultaneously, validity and validity check are carried out on the basis of the power data, tags are added, and the problem of data reliability is solved.
(2) The deep learning neural network is divided into a local feature extraction sub-network and a cloud device prediction sub-network, locally acquired high-frequency data are directly calculated locally, calculation results and gradients are transmitted between the cloud and the local, data transmission quantity is small, and the problem of data transmission bottleneck under high-frequency sampling is solved.
(3) The features extracted by the local feature extraction sub-network are sent to the cloud end through an encryption algorithm, the cloud end returns the gradient to be updated, on one hand, the encryption algorithm ensures the security of data transmission, on the other hand, the calculation result and the gradient cannot be traced to obtain the original state data of the user, and the problem of privacy security of the user in the training and prediction processes is solved.
Drawings
FIG. 1 is a flow chart of a method of device status identification;
FIG. 2 is an architecture diagram of an equipment status identification system;
reference numerals: 1. the system comprises a non-invasive equipment data acquisition system, 2, a local data set construction module, 3, a local calculation unit, 4 and a cloud calculation unit.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
In the drawings, structurally identical elements are represented by like reference numerals, and structurally or functionally similar elements are represented by like reference numerals throughout the several views. The size and thickness of each component shown in the drawings are arbitrarily illustrated, and the present invention is not limited to the size and thickness of each component. Parts are exaggerated in the drawing where appropriate for clarity of illustration.
Example 1:
a method for identifying a state of a non-intrusive device based on federal learning, as shown in fig. 1, includes the following steps:
s1, arranging a non-invasive equipment data acquisition system in a plurality of local user nodes, carrying out high-frequency sampling on total current data of trunk circuits in the local user nodes, and carrying out low-frequency sampling on power data of each parallel branch circuit in the local user nodes, wherein the local user nodes are in the environments of families, laboratories, offices and the like;
the non-invasive equipment data acquisition system non-invasively acquires total current data of a trunk circuit in a local user node at the frequency of 180kHz, acquires power data of each parallel branch circuit in the local user node at the frequency of 1Hz, and all equipment in the local user node is connected into the parallel branch circuit.
Specifically, the embodiment is as follows:
step S11: a high-frequency digital-to-analog conversion chip is used for designing a sampling circuit and non-invasively acquiring total current data I at the frequency of 180kHzGeneral assemblyThe total current being the sum of the currents of the individual parallel branches, theoretically
Figure BDA0003454904230000051
n represents the total number of devices, wherein IiEach device is connected into the parallel branches for the current of each parallel branch, and a high-frequency filter device is not arranged between each branch;
step S12: acquiring power data P of each parallel branch at low frequency of 1Hz by using a plurality of intelligent sockets with wifi communication and power measurement functionsi；
Step S13: the method comprises the steps that a micro host is deployed as a local computing center, high-frequency total current data are obtained in a usb wired transmission mode, low-frequency power data are obtained in a local area network wifi mutual transmission mode, an application software buffer layer writes the data into a memory quickly, and the data are continuously written into a hard disk of the micro host, so that reliable transmission of the data is guaranteed.
S2, respectively constructing historical data sets according to the data of each local user node;
s21, acquiring total current data and power data continuously acquired by the non-invasive equipment data acquisition system, for example, continuously acquiring power data and total current data of equipment with high frequency characteristics under actual use conditions within one week, half month and the like after the non-invasive equipment data acquisition system is deployed;
s22, setting the minimum starting power P of each equipment according to expert experiencemin_startMaximum stopping power Pmax_endMinimum duration tminAnd a maximum interruption time tmaxProcessing the power data, and judging the working state of each device according to the power data by using a streaming algorithm, wherein the working state comprises starting and closing; when the real-time power of the equipment is more than Pmin_startAnd has a duration longer than tminThe device may be considered powered up when the real-time power of the device is less than Pmax_endAnd has a duration longer than tmaxThe device may be considered to have been turned off;
s23, introducing a sliding window, converging the working state of each device by applying a streaming algorithm to the data between the current time t and the previous time t-1S, keeping legal and effective data according to a preset judgment rule, adding a label to the device according to the working state, and constructing a historical data set in each local user node. Such as: in the same window, if the working states of all the devices are not switched from starting to closing or from closing to starting, the data of the window is considered to be legal; within the same window, if all devices are in the off state, the window is considered invalid. And recording and storing the labels for legal and effective data, constructing a historical data set, wherein data samples are total current data and corresponding equipment states, and the rest data are discarded. The power data is used for judging the legal validity of the data and determining the label, so that label addition and data screening can be automatically carried out according to the power, and the labor work is greatly reduced.
S3, constructing a deep learning neural network, wherein the deep learning neural network comprises a feature extraction sub-network and a device prediction sub-network, the feature extraction sub-network is arranged in a local user node, and the device prediction sub-network is arranged in a cloud end;
and constructing a neural network-based hybrid model, namely a deep learning neural network, wherein the deep learning neural network is used for realizing multi-label classification of input data, transforming a very-matched model and dividing the model into a local calculation part and a cloud calculation part. In this embodiment, the feature extraction sub-network is a residual error network whose basic module is a one-dimensional convolution unit, and is disposed in a local micro host (edge device), the device prediction sub-network is a residual error network of a multi-layer perceptron, and is disposed in a cloud server, and the Loss function used in multi-label classification is Asymmetric Loss.
The high-frequency total current data collected at 180kHz is high-dimensional data, the original purpose of reserving more information by using the high-frequency data is overcome if dimension reduction processing is carried out, and more input neurons need to be designed if a conventional neural network is used, so that the network structure is complex. A convolutional network is generally used for image processing, and takes a 240 × 240 picture as an example, and a pixel matrix of 240 × 240 is actually input thereto, and data of a higher dimension can be processed. Therefore, the feature extraction sub-network is a residual error network with a basic module as a one-dimensional convolution unit, feature extraction is performed on high-frequency data through convolution operation, complexity of the feature extraction sub-network is reduced, and the calculation amount of local edge equipment can be met.
In order to deal with the output of the multi-label classification, the device predicts that the sub-network is a residual network of the multi-layer perceptron, and the Loss function used by the multi-label classification is Asymmetric Loss.
Finally, the feature extraction sub-network and the equipment prediction sub-network are simplified and optimized through methods such as model compression, pruning and the like, relevant parameters are adjusted, accuracy is guaranteed, and meanwhile, the calculation amount is reduced, so that the feature extraction sub-network and the equipment prediction sub-network can be deployed on local edge equipment.
In this embodiment, specific structures of the feature extraction sub-network and the device prediction sub-network are not limited, and relevant practitioners may design as needed, and may apply various high-performance neural network hybrid modeling, and ensure a high accuracy rate with a low calculation amount.
S4, training the deep learning neural network through a historical data set, and performing data transmission between the local user node and the cloud end through an encryption algorithm in the training process to obtain a trained recognition model;
the training process is as follows:
s41, for each local user node, acquiring a historical data set, carrying out preprocessing, including preprocessing operations such as denoising and normalization, and sending the historical data set into a feature extraction sub-network to obtain a calculation result;
s42, uploading the calculation results in each local user node to a cloud after being encrypted through an encryption algorithm, such as a differential encryption algorithm;
s43, the cloud end is provided with an equipment prediction sub-network, the calculation results of all local user nodes are collected, the calculation results are integrated, the equipment prediction sub-network is trained in the cloud end, and the gradient is transmitted back to each local user node;
and S44, the local user node extracts the sub-network according to the received gradient updating characteristics.
Considering that the number of devices and the working modes in different local user nodes are different, the structures of the local user node feature extraction sub-networks are independent from each other, and different structures can be used, so that the gradient is transmitted back to each local user node by the cloud, updating of different feature extraction sub-networks is not influenced, and finally obtained recognition models can also be adapted to each local user node.
The updating of each local user node feature extraction sub-network is independent, the updating according to the gradient is different due to different network structures, and the gradient can be returned in different time periods and each feature extraction sub-network is updated asynchronously in consideration of the high and low peaks, the network state and the like of the equipment of different local user nodes.
And S5, recognizing the equipment state by using the trained recognition model, and monitoring the running state of each equipment.
After the training is completed, the device prediction sub-network can be arranged locally, that is, the local edge device is loaded with the recognition model. And acquiring real-time total current data locally, and sending the real-time total current data into the identification model, so that the state of each device can be monitored in real time, and analysis can be performed, such as the running state of each device in a target time period. The device state identification method can also be used for arranging the device prediction sub-network at the cloud end, carrying the feature extraction sub-network on local edge devices, sending real-time total current data into the feature extraction sub-network, and sending the features extracted by the feature extraction sub-network into the cloud end device prediction sub-network to obtain a device state identification result.
If a device is newly introduced into the local user node, steps S1-S4 may be repeated to update the feature extraction sub-network and the device prediction sub-network to obtain a new recognition model.
Example 2:
a non-intrusive device status recognition apparatus based on federal learning, as shown in fig. 2, includes:
the non-invasive equipment data acquisition system 1 is respectively arranged in each local user node and is used for carrying out high-frequency sampling on total current data of a trunk circuit in the local user node and carrying out low-frequency sampling on power data of each parallel shunt circuit in the local user node;
the local data set building module 2 is respectively arranged in each local user node and used for respectively building a historical data set according to the data of each local user node collected by the non-invasive equipment data collection system 1;
the local computing units 3 are respectively arranged in each local user node, are in communication connection with the local data set building module 2, and store a feature extraction sub-network for computing and updating the feature extraction sub-network in the training process;
the cloud computing unit 4 is arranged at the cloud, stores a device prediction sub-network and is used for computing and updating the device prediction sub-network in the training process, and data transmission is performed between the local computing unit 3 and the cloud computing unit 4 through an encryption algorithm in the training process;
and the detection module is used for identifying the equipment state by using the trained identification model.
As shown in fig. 2, in the process of building the non-invasive device data acquisition system 1, a high-frequency digital-to-analog conversion chip is used to design a sampling circuit, total current data of a trunk circuit in a local user node is acquired in a non-invasive manner at a frequency of 180kHz, power data of each parallel shunt circuit in the local user node is acquired at a frequency of 1Hz by using an intelligent socket with a power measurement function, and each device in the local user node is connected to the parallel shunt circuit.
The local data set building module 2 and the local computing unit 3 may be integrated on a micro-mainframe, i.e. an edge computing device, and transmit the total current data to the local data set building module 2 in a USB wired transmission manner, and the smart socket has a wireless communication function and transmits the power data to the local data set building module 3 in a WIFI wireless transmission manner.
The technology of high-frequency sampling by the high-frequency digital-to-analog conversion chip and the sampling circuit is mature, the micro host and the intelligent socket are also designed in the prior art, and the high-frequency sampling can be realized by using a common computer and the intelligent socket, so the cost is low and the reliability is high.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims (10)

1. A non-intrusive equipment state identification method based on federal learning is characterized by comprising the following steps:
s1, arranging a non-invasive equipment data acquisition system in a plurality of local user nodes, carrying out high-frequency sampling on total current data of trunk circuits in the local user nodes, and carrying out low-frequency sampling on power data of each parallel branch circuit in the local user nodes;
s2, respectively constructing historical data sets according to the data of each local user node;
s3, constructing a deep learning neural network, wherein the deep learning neural network comprises a feature extraction sub-network and a device prediction sub-network, the feature extraction sub-network is arranged in a local user node, and the device prediction sub-network is arranged in a cloud end;
s4, training the deep learning neural network through a historical data set, and performing data transmission between the local user node and the cloud end through an encryption algorithm in the training process to obtain a trained recognition model;
and S5, recognizing the equipment state by using the trained recognition model.
2. The method according to claim 1, wherein the non-intrusive device data collection system in step S1 non-intrusively collects total current data of a trunk in the local user node at a frequency of 180kHz and collects power data of each parallel branch in the local user node at a frequency of 1Hz, and each device in the local user node is connected to the parallel branch.
3. The method for non-intrusive equipment state identification based on federal learning of claim 1, wherein the step S2 includes the following steps:
s21, acquiring total current data and power data continuously acquired by a non-invasive equipment data acquisition system;
s22, setting the minimum starting power P of each devicemin_startMaximum stopping power Pmax_endMinimum duration tminAnd a maximum interruption time tmaxProcessing the power data, and judging the working state of each device according to the power data by using a streaming algorithm;
s23, introducing a sliding window, converging the working state of each device by applying a streaming algorithm to the data between the current time t and the previous time t-1, retaining valid and effective data according to a preset judgment rule, adding a label to the device according to the working state, and constructing a historical data set in each local user node.
4. The method as claimed in claim 1, wherein a deep learning neural network is used to implement multi-label classification, the feature extraction sub-network is a residual error network whose basic module is a one-dimensional convolution unit, the device prediction sub-network is a residual error network of a multi-layer perceptron, and a Loss function used in multi-label classification is Asymmetric Loss.
5. The method for non-intrusive equipment state recognition based on federal learning as claimed in claim 1, wherein the training process of the deep learning neural network is as follows:
s41, for each local user node, acquiring and preprocessing a historical data set, and sending the historical data set into a feature extraction sub-network to obtain a calculation result;
s42, uploading the calculation results in each local user node to the cloud after being encrypted through an encryption algorithm;
s43, the cloud end collects the calculation results of all local user nodes, trains the equipment prediction sub-network at the cloud end, and transmits the gradient back to each local user node;
and S44, the local user node extracts the sub-network according to the received gradient updating characteristics.
6. The method of claim 5, wherein the encryption algorithm is a differential encryption algorithm.
7. The method of claim 5, wherein the cloud returns the gradient to each local user node, the structure of each local user node feature extraction sub-network is independent, and the update of each local user node feature extraction sub-network is independent.
8. A federal learning-based non-intrusive equipment state identification device, which is based on a federal learning-based non-intrusive equipment state identification method as claimed in any one of claims 1 to 7, and comprises:
the non-invasive equipment data acquisition system is respectively arranged in each local user node and is used for carrying out high-frequency sampling on total current data of a trunk circuit in the local user node and carrying out low-frequency sampling on power data of each parallel shunt circuit in the local user node;
the local data set building module is respectively arranged in each local user node and used for respectively building a historical data set according to the data of each local user node;
the local computing units are respectively arranged in each local user node, are in communication connection with the local data set building module, and store a feature extraction sub-network for computing and updating the feature extraction sub-network in the training process;
the cloud computing unit is arranged at the cloud, stores the equipment prediction sub-network and is used for computing and updating the equipment prediction sub-network in the training process, and the local computing unit and the cloud computing unit perform data transmission through an encryption algorithm in the training process;
and the detection module is used for identifying the equipment state by using the trained identification model.
9. The device for identifying the state of non-invasive equipment based on federal learning as claimed in claim 8, wherein a high frequency digital-to-analog conversion chip is used to collect the total current data of the trunk circuit in the local user node in a non-invasive manner at a frequency of 180kHz, and a smart socket with a power measurement function is used to collect the power data of each parallel branch circuit in the local user node at a frequency of 1Hz, and each device in the local user node is connected to the parallel branch circuit.
10. The non-intrusive equipment state identification device based on federal learning of claim 9, wherein the total current data is transmitted to the local data set building module in a wired transmission manner, and the smart socket has a wireless communication function and transmits the power data to the local data set building module in a wireless transmission manner.
CN202210001823.0A 2022-01-04 2022-01-04 Non-invasive equipment state identification method and device based on federal learning Pending CN114462501A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210001823.0A CN114462501A (en) 2022-01-04 2022-01-04 Non-invasive equipment state identification method and device based on federal learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210001823.0A CN114462501A (en) 2022-01-04 2022-01-04 Non-invasive equipment state identification method and device based on federal learning

Publications (1)

Publication Number Publication Date
CN114462501A true CN114462501A (en) 2022-05-10

Family

ID=81407035

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210001823.0A Pending CN114462501A (en) 2022-01-04 2022-01-04 Non-invasive equipment state identification method and device based on federal learning

Country Status (1)

Country Link
CN (1) CN114462501A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116108350A (en) * 2023-01-06 2023-05-12 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116108350A (en) * 2023-01-06 2023-05-12 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning
CN116108350B (en) * 2023-01-06 2023-10-20 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning

Similar Documents

Publication Publication Date Title
CN110298254A (en) A kind of analysis method and system for personnel's abnormal behaviour
CN103107886A (en) Identification recognition system and method
CN105143892A (en) Power entry point energy measuring apparatus and energy measurement information labeling system using same
CN105824837B (en) A kind of log processing method and device
CN204303005U (en) Intelligent image recognition in-out control device
CN110687404A (en) Portable partial discharge detection equipment
CN102208983A (en) Identity recognition management system and method focused on households in social security housing
CN109391923A (en) A kind of building energy consumption management method and system based on 5G framework
CN114462501A (en) Non-invasive equipment state identification method and device based on federal learning
CN108804696A (en) The creation method and its system of power consumer electric appliance fingerprint base
CN105974220B (en) A kind of residential area power load identifying system
CN111198525A (en) Audio monitoring system based on operation of oilfield field equipment and fault detection method
CN105046882A (en) Fall detection method and device
CN107843781A (en) Electric energy quality monitoring analysis system
CN113723184A (en) Scene recognition system, method and device based on intelligent gateway and intelligent gateway
CN109064467A (en) Analysis method, device and the electronic equipment of community security defence
CN203942602U (en) Bank monitoring device
CN112257658B (en) Electroencephalogram signal processing method and device, electronic equipment and storage medium
CN113627547B (en) Training method, arc detection device, electronic equipment and storage medium
CN108460363A (en) Monitor Work attendance method, apparatus and system
CN112651273A (en) AI intelligent camera tracking method
CN114895609A (en) Machine room monitoring method, device, equipment and medium
CN104183045A (en) Capacitive sensing technology-based access control system
CN203982468U (en) Job site personnel attendance management equipment
CN206684798U (en) A kind of Multifunctional attendance machine with PM2.5 monitoring functions

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination