Front-end incremental detection-based end-cloud collaborative real-time load identification method and system

Abstract

The application relates to an end-cloud cooperative real-time load identification method and system based on front-end incremental detection. The method comprises the following steps: collecting current signals and voltage signals of a user at the front end, and extracting load characteristic information and current instantaneous value difference values based on the current signals and the voltage signals; the front end matches the current instantaneous value difference values with a front-end lightweight load characteristic library for detection or carries out front-end lightweight incremental neural network identification, if successful identification is achieved, the identified load is output, and if the identification is not successful, the load characteristic information and the current instantaneous value difference values are sent to the cloud end; the cloud end carries out matching identification processing according to the load characteristic information and the current instantaneous value difference values, so as to complete end-cloud cooperative real-time synchronous load identification. The application can guarantee the timeliness and accuracy of load identification while reducing the implementation complexity.

Description

Technical Field

[0001] This invention relates to the field of intelligent fire protection electrical safety monitoring technology, and in particular to a method and system for real-time synchronous load identification based on front-end lightweight incremental detection and end-cloud collaboration. Background Technology

[0002] Statistics show that my country's power grid coverage has reached 99.95%, and electricity has become the preferred energy source for people's production and daily life. However, due to aging lines, improper use of electrical appliances, and many other reasons, electrical fires have gradually become a major factor affecting fire safety. From 2012 to 2021, a total of 1.324 million residential fires occurred nationwide, of which electrical fires accounted for as much as 42.7%. Faced with the severe challenges to electrical safety, there is an urgent need to carry out intelligent fire protection and electrical safety monitoring.

[0003] Electricity monitoring technologies are broadly categorized into invasive and non-invasive methods. Invasive solutions require the installation of sensors on various electrical devices used by the user to collect data, resulting in poor operability, high implementation costs, and low user acceptance. Non-invasive solutions, on the other hand, only require the installation of a monitoring device at the user's main power line, using data analysis and other technical means to monitor electricity consumption. This method eliminates the need for cumbersome intrusion into the user's premises, offering strong operability, high user acceptance, and suitability for large-scale deployment.

[0004] Currently, non-invasive load monitoring has developed into a technical approach based on condition detection, feature extraction, and load identification. There are two main implementation methods: front-end sensors collect and process load identification data, and front-end sensors collect data and send it to the cloud for load identification.

[0005] The load identification method that uses front-end sensors to collect and process data needs to consider the limitations of the front-end sensor's processing capabilities. Low-frequency sampling is usually used, and the effective values ​​of electrical parameters are taken as the main distinguishing features. The model is simple, and feature value calculation and decision can be completed at the front end, but the recognition accuracy is low and it is easy to produce missed judgments and false judgments.

[0006] The front-end sensor-based load identification method uses a high-frequency, multi-bit quantization AD high-speed sampling module to collect and send high-resolution data to a cloud server. Load identification is then performed in the cloud based on a complex AI-powered algorithm. While this method boasts high accuracy, it requires significant cloud resources. Furthermore, it places extremely high demands on network communication; any transmission failure will halt the identification process and lead to missed detections. Additionally, uploading all user data to the cloud increases the risk of privacy breaches and may draw criticism from users during product promotion. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a real-time load identification method and system based on front-end incremental detection for end-to-cloud collaboration, which can reduce the implementation complexity while ensuring the timeliness and accuracy of load identification.

[0008] The technical solution adopted by this invention to solve its technical problem is: to provide a method for real-time synchronous load identification of end-to-cloud collaboration based on lightweight incremental detection at the front end, including the following steps:

[0009] The front end collects the user's current and voltage signals in real time, and extracts load characteristic information and instantaneous current value difference based on the current and voltage signals;

[0010] The front end performs a matching detection between the instantaneous current value difference and the front end lightweight load feature library. If the matching is successful, the matching load is output. If the matching is unsuccessful, the load increment lightweight neural network recognition is performed. If the recognition is successful, the recognized load is output. If the recognition is unsuccessful, the load feature information and the instantaneous current value difference are sent to the cloud.

[0011] The cloud performs matching and identification processing based on the load characteristic information and the difference value of the instantaneous current value, and sends the identification result features and identification framework parameters to the corresponding front end to complete the real-time synchronous load identification of end-cloud collaboration.

[0012] The front end collects the user's current and voltage signals in real time, and extracts load characteristic information and instantaneous current value difference based on the current and voltage signals, specifically including:

[0013] The system collects timing current and voltage data at a preset sampling frequency and saves the data at a set time interval.

[0014] The system collects time-series current and voltage data in real time at a preset sampling frequency, and saves the data at a set time interval.

[0015] The time-series current and voltage data are sliced ​​according to the time slot length, and the effective value of the current in each slice time slot is calculated;

[0016] Calculate the absolute value of the difference between the current effective value of the current slot and the previous effective value of the current slot, and compare the absolute value of the difference with a set threshold.

[0017] If the absolute value of the difference is greater than or equal to a set threshold, then current phase synchronization is initiated to obtain an instantaneous current sequence;

[0018] Traverse the instantaneous current sequence to find the point where the current changes abruptly, and obtain two sets of time-adjacent current time slot sequences and corresponding voltage time slot sequences. Calculate the difference between the two sets of time-adjacent current time slot sequences to obtain the instantaneous current value difference value, and use the two sets of time-adjacent current time slot sequences and the corresponding voltage time slot sequences as load characteristic information.

[0019] When calculating the effective value of the current time slot based on the time slot length and the time-series current data, through... Perform calculations, where I n The effective value of the current time slot current is Δt. n i is the time slot length. k For time-series current data, k∈[(n-1)*f s *Δt n +1,(n)*f s *Δt n ], f s The sampling frequency.

[0020] The instantaneous current sequence obtained by synchronizing the starting current phase is specifically as follows:

[0021] Using the excitation start-up current phase synchronization time slot as a reference, read the timing current data and timing voltage data of the nearest N time slots before and after the time slot, calculate the first voltage phase zero point of the timing voltage data in the reading time slot, and use the first voltage phase zero point as a reference to extract an integer number of cycles from the corresponding timing current data to obtain the instantaneous current sequence.

[0022] The cloud platform performs matching and identification processing based on the load characteristic information and the difference in instantaneous current value to complete real-time synchronized and coordinated load identification between the end and cloud, specifically including:

[0023] Using the instantaneous current difference value, perform local cloud load feature library matching detection on the cloud local load feature library corresponding to the front end that uploaded the instantaneous current difference value;

[0024] If a match is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification.

[0025] If a match is not found, the instantaneous current value difference will be used for matching and detection using the cloud-based global load feature library.

[0026] If a match is successful, the identification result and corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification; if a match is unsuccessful, the cloud uses the instantaneous current value difference to carry out incremental data neural network load identification.

[0027] If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification; if the identification is unsuccessful, the cloud combines the known single-type load characteristics of each load in the cloud global load feature library, and uses two sets of time-adjacent current time slot sequences and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences to carry out neural network load decomposition identification of load type and corresponding load quantity.

[0028] If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification; if the identification is still unsuccessful, the cloud stores the instantaneous current difference value of the front end, the two sets of time-adjacent current time slot sequences, and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences into the unknown load database; the unknown load database records the unidentified load data of different times and different users according to the user number.

[0029] The aforementioned end-to-cloud collaborative real-time load identification method based on front-end incremental detection also includes:

[0030] When the accumulated data in the unknown load database exceeds the data volume threshold or reaches the set tag identification period, the cloud will count the unknown load type tags according to the stored unknown load type F. For the unknown load that appears in the top P positions among all unknown load types, the cloud administrator will be reminded to perform load tag identification. After the tag is determined, the identification result and the corresponding load characteristics will be sent to the front end of all users with the load characteristics to complete the end-cloud collaborative global load identification.

[0031] The aforementioned end-to-cloud collaborative real-time load identification method based on front-end incremental detection also includes:

[0032] If the absolute value of the difference is less than a set threshold, front-end anchor point calibration flag information is generated, and current phase synchronization is initiated to obtain the instantaneous current sequence.

[0033] Based on the preset sampling frequency and time slot, the instantaneous current sequence is divided into two sets of time-adjacent current time slot sequences in chronological order. The difference between the two sets of time-adjacent current time slot sequences is calculated to obtain the instantaneous current value difference. The front-end anchor point calibration flag information, the two sets of time-adjacent current time slot sequences, and the voltage time slot sequences corresponding to the two sets of time-adjacent current time slot sequences are used as load characteristic information.

[0034] The cloud platform performs matching and identification processing based on the load characteristic information and the difference in instantaneous current value to complete real-time synchronous load identification in a collaborative manner between the end and the cloud, specifically including:

[0035] Based on the instantaneous current value difference and the front-end anchor point calibration flag information, no-load incremental feature matching detection is performed in the cloud load feature database;

[0036] If a match is successful, empty calibration information is sent to the corresponding data upload front-end to complete the real-time synchronization of anchor point calibration between the end and cloud.

[0037] If a match is not found, the instantaneous current difference value is used to perform a local cloud load feature library matching detection on the cloud local load feature library corresponding to the front-end that uploaded the instantaneous current difference value.

[0038] If a match is successful, the identification result and the corresponding load features are sent to the front end to complete the real-time synchronous anchor point calibration of the end-cloud collaboration.

[0039] If a match is not found, the instantaneous current value difference will be used for matching and detection using the cloud-based global load feature library.

[0040] If a match is successful, the identification result and the corresponding load features are sent to the front end to complete the real-time synchronous anchor point calibration of the end-cloud collaboration.

[0041] If the match is unsuccessful, the cloud uses the instantaneous current value difference to perform neural network load decomposition and identification, and after successful identification, sends calibration information containing the identification results and corresponding load characteristics to the front end to update the front end lightweight load feature library;

[0042] If identification fails, the cloud uses two sets of time-adjacent current time slot sequences and the corresponding voltage time slot sequences to perform neural network load decomposition identification. After successful identification, calibration information containing the identification results and corresponding load feature identification framework parameters is sent to the front end to update the front end lightweight load feature library.

[0043] The technical solution adopted by the present invention to solve its technical problem is: to provide a front-end lightweight incremental detection-based end-cloud collaborative real-time synchronous load identification system, including a front-end lightweight incremental matching load identification device and a cloud-based neural network load decomposition identification device;

[0044] The front-end lightweight incremental matching load identification device includes: a data acquisition module for real-time acquisition of the user's current and voltage signals; a processing module for extracting load feature information and instantaneous current difference values ​​based on the current and voltage signals, and performing matching detection between the instantaneous current difference values ​​and the front-end lightweight load feature library. If a match is successful, the matched load is output; if a match is unsuccessful, a lightweight incremental neural network identification of the load is performed. If the identification is successful, the identified load is output; if the identification is unsuccessful, the load feature information and instantaneous current difference values ​​are sent to the cloud; and a communication module for communicating with the cloud.

[0045] The cloud-based neural network load decomposition and identification device includes: a receiving module for receiving the load feature information and instantaneous current difference value sent by the front-end lightweight incremental matching load identification device; and an identification module for performing matching and identification processing based on the load feature information and instantaneous current difference value, and sending the identification result features and identification framework parameters to the corresponding front end to complete real-time synchronous load identification in a cloud-end collaborative manner.

[0046] The processing module includes: a current and voltage storage unit for storing time-series current and voltage data according to a set storage period; a phase synchronization unit for synchronously truncating the current signal and extracting the current signal from adjacent time slots by using the zero point of a reference voltage signal; an effective value difference calculation unit for calculating the absolute value of the difference between the effective values ​​of the current signals in adjacent time slots; a difference comparison unit for comparing the absolute value of the difference between the effective values ​​with a set threshold; and a feature matching unit for performing feature matching based on the instantaneous current difference value output by the instantaneous value difference calculation module with a front-end lightweight load feature library to achieve load identification. The system includes: an incremental lightweight neural network recognition unit for performing incremental lightweight neural network recognition of load based on the incremental instantaneous current value difference; a front-end lightweight load library for storing known load characteristics; an instantaneous value difference calculation unit for calculating the instantaneous current value difference based on the difference between adjacent current time slot sequences output by the phase synchronization unit; and a transmission buffer unit for buffering the instantaneous current value difference to be sent to the cloud, two sets of time-adjacent current time slot sequences, the voltage time slot sequences corresponding to the two sets of time-adjacent current time slot sequences, and the corresponding front-end anchor point calibration flag information.

[0047] The identification module includes: a cloud-based load feature library for storing known load features in the cloud, including local load features stored according to user front-end classification and global load features stored according to load classification; an unknown load database for storing the instantaneous current difference value, two sets of time-adjacent current time slot sequences, and voltage time slot sequences corresponding to the unknown category loads uploaded by the front-ends that were not successfully identified; a tag identification unit for receiving unknown load tag information identified by the cloud administrator; a neural network load identification unit for implementing decomposition training and identification of unknown loads uploaded by each front-end based on neural networks; and an incremental feature matching unit for implementing feature matching operations based on the instantaneous value difference value and the cloud-based load feature library to achieve corresponding load identification.

[0048] Beneficial effects

[0049] By adopting the above-mentioned technical solution, this invention has the following advantages and positive effects compared with the prior art: This invention achieves real-time synchronous identification of known loads that have already been trained by building a localized lightweight incremental judgment model at the front end, and collaboratively conducts training and identification of unknown loads in the cloud, constructing a cloud-based artificial intelligence model based on neural networks. Simultaneously, it supports real-time synchronous cloud anchor point functionality for end-to-cloud collaboration, meeting front-end calibration requirements, and provides the implementation method for the corresponding device. Through end-to-cloud collaboration, this invention can reduce implementation complexity while ensuring the timeliness and accuracy of load identification, strongly supporting the construction of intelligent fire-fighting electrical safety monitoring systems for intelligent perception and refined management of user electricity consumption characteristics. Attached Figure Description

[0050] Figure 1 This is a block diagram of the method according to the first embodiment of the present invention;

[0051] Figure 2 This is a flowchart of the front-end lightweight incremental detection process in the first embodiment of the present invention;

[0052] Figure 3 This is a flowchart of cloud-based neural network load identification in the first embodiment of the present invention;

[0053] Figure 4 This is a flowchart of the end-to-cloud collaborative anchor point calibration process in the first embodiment of the present invention;

[0054] Figure 5 This is a structural block diagram of the front-end lightweight incremental detection load identification device in the second embodiment of the present invention;

[0055] Figure 6 This is a structural block diagram of the cloud neural network load decomposition and identification device in the second embodiment of the present invention. Detailed Implementation

[0056] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0057] The first embodiment of the present invention relates to a method for real-time synchronous load identification in end-to-cloud collaboration based on lightweight incremental detection at the front end, such as... Figure 1As shown, the method includes the following steps: The front end collects the user's current and voltage signals in real time, and extracts load feature information and instantaneous current value difference based on the current and voltage signals; the front end performs matching detection between the instantaneous current value difference and the front end's lightweight load feature library. If a match is successful, the matched load is output; if not, the front end performs incremental lightweight neural network recognition of the load. If the recognition is successful, the recognized load is output; if not, the load feature information and instantaneous current value difference are sent to the cloud; the cloud performs matching recognition processing and neural network load decomposition recognition based on the load feature information and instantaneous current value difference, completing end-cloud collaborative real-time synchronous load recognition and anchor point calibration. Specifically, this method includes two parts: the first part is the front end's lightweight incremental detection load recognition part, and the second part is the end-cloud collaborative real-time synchronous load recognition.

[0058] like Figure 2 As shown, the front-end lightweight incremental detection load identification part includes:

[0059] (1) Using sampling frequency f s The user's current signal I and voltage signal U are collected and stored at a time period T, where the sampling frequency f s ≥1kHz, T≥1s, current signal I=[i1,i2,...,i m The voltage signal U = [u1, u2, ..., u] m ], m=T×f s In this embodiment, f s =1000KHz, T=1s, m=1000;

[0060] (2) The time-series current data and voltage data are processed according to the time slot length Δt. n Slice the data and calculate the effective current value within each slice time slot according to the time slot length Δt. n Wherein, the time slot length Δt n ≥20ms k∈[(n-1)*f s *Δt n +1,(n)*f s *Δt n ], In this embodiment, Δt n =20ms, n∈[1,50],

[0061] (3) Calculate the absolute value ΔI of the difference between the effective current value in the current time slot and the effective current value in the previous time slot. n and with the set threshold ΔI th Comparison, if the absolute value of the difference ΔI n Greater than or equal to the set threshold ΔIth Then the starting current phase is synchronized, where ΔI th ≥0.05A, ΔI n =|I n -I n-1 In this embodiment, ΔI th =0.05A;

[0062] (4) Using the excitation start-up current phase synchronization time slot as a reference, read the timing current data and timing voltage data of the N nearest time slots before and after this time slot, and calculate the first voltage phase zero point u of the timing voltage data within the reading time slot. k Based on this, an integer number of cycles are truncated from the corresponding time-series current data to obtain the instantaneous current sequence I. syn Voltage phase zero point u k =0, and u k-j <0,u k+j >0, where j is typically taken as 0.005f s And round up, N∈[2,5]. In this embodiment, j=5, N=5;

[0063] (5) Traverse the instantaneous current sequence I syn Find the point where the current changes abruptly, thus obtaining two sets of time-adjacent current sequences, the current time slot sequence I. n (k) and the previous time slot sequence I n-1 (k) and perform difference calculation to obtain the instantaneous current difference value ΔI. n (k), where k∈[1,20]; let I n (k), I n-1 (k) and I n (k), I n-1 (k) corresponds to the voltage time slot sequence U n (k), U n-1 (k) serves as load characteristic information;

[0064] (6) The difference between the instantaneous current values ​​ΔI n (k) Perform matching detection with the front-end lightweight load feature library. If a match is successful, output the corresponding incremental matching load. At the same time, based on the positive and negative values ​​of the incremental sign, give the opening and closing behavior of the corresponding load respectively.

[0065] (7) If identification fails, the front end performs incremental lightweight neural network identification. If identification is successful, the identified load is output, and the opening and closing behavior of the corresponding load is given based on the positive and negative values ​​of the increment sign. If detection fails, the corresponding I... n (k), I n-1 (k), ΔI n (k), U n(k), U n-1 (k) Send to the cloud.

[0066] like Figure 3 As shown, the edge-cloud collaborative load identification part includes:

[0067] (1) I based on front-end upload n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k), using incremental data ΔI in the cloud n (k) Perform cloud-based local load feature database matching and detection. If a match is successful, send the identification result and the corresponding load features to the front end to complete the real-time synchronous load identification of the end-cloud collaboration.

[0068] If a match is not found, incremental data ΔI will be used in the cloud. n (k) Perform cloud-based global load feature database matching and detection. If a match is successful, send the identification result and corresponding load features to the corresponding data upload front-end to complete real-time synchronized load identification between the end and cloud. If identification fails, the cloud uses ΔI... n (k) Conduct incremental data neural network load identification;

[0069] (3) If identification is successful, the identification result and corresponding load characteristics are sent to the corresponding data upload front-end to complete real-time synchronous load identification in a cloud-edge collaborative manner; if identification is unsuccessful, the cloud combines the known single-type load characteristics in the cloud's global load characteristic library and uses I n (k), I n-1 (k), U n (k), U n-1 (k) Conduct neural network load decomposition and identification of load types and corresponding load quantities;

[0070] (4) If the identification is successful, the identification result and corresponding load characteristics are sent to the corresponding data upload front-end to complete the real-time synchronous load identification of the end-to-cloud collaboration; if the identification is unsuccessful, the cloud will send the corresponding front-end's I n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k) Store in the unknown load database; set the label number for the unknown load type, and send an identification failure message to the corresponding data upload front end.

[0071] (5) The unknown load database records unidentified load data for different users at different times, according to user ID;

[0072] (6) When the amount of data in the unknown load database exceeds the data volume threshold or the set label identification period is reached, the cloud server counts the unknown load type labels according to the stored unknown load type F. For the unknown load that appears in the top P positions among all unknown load types, the cloud administrator is reminded to perform load label identification operation. After the label is determined, the identification result and the corresponding load characteristics are sent to the front end of all users with the load characteristics to complete the global load identification of the end-cloud collaboration; in this embodiment, P=1.

[0073] This implementation addresses the issue of accumulated load identification errors caused by the effective current consistently falling below the decision threshold during incremental matching detection. In this case, cloud-based anchor points can be used for end-to-cloud collaborative error elimination, as detailed below:

[0074] like Figure 4 As shown, if in step (3) of the front-end lightweight incremental detection load identification part, the absolute value of the difference ΔI n Less than the set threshold ΔI th Furthermore, if the end-cloud collaborative anchor point calibration is set, then the front-end anchor point calibration flag information cf is generated, and current phase synchronization is started; current phase synchronization is performed in step (4), and in step (5), the sampling frequency f is used as the basis for the current phase synchronization. s and time slot Δt n In the instantaneous current sequence I syn The current sequence is divided into two groups of time-adjacent current sequences according to the chronological order, namely the current time slot sequence I. n (k) and the previous time slot sequence I n-1 (k), and perform difference calculation to obtain ΔI. n (k); In step (6), I will be... n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k) and the corresponding front-end anchor point calibration flag information cf are sent to the cloud; during the real-time synchronization of load identification and anchor point calibration in the end-cloud collaborative process, the cloud is based on ΔI n(k) and cf, perform no-load incremental feature matching detection in the cloud load feature library. If a match is successful, send empty calibration information to the corresponding data upload front-end to complete the end-to-cloud collaborative real-time synchronous anchor point calibration. If a match is unsuccessful, use the instantaneous current value difference to perform local cloud load feature library matching detection in the cloud local load feature library corresponding to the front-end that uploaded the instantaneous current value difference. If a match is successful, send the identification result and the corresponding load feature to the front-end to complete the end-to-cloud collaborative real-time synchronous anchor point calibration. If a match is unsuccessful, use the instantaneous current value difference to perform cloud global load feature library matching detection. If a match is successful, send the identification result and the corresponding load feature to the front-end to complete the end-to-cloud collaborative real-time synchronous anchor point calibration. If a match is unsuccessful, the cloud uses ΔI n (k) Conduct neural network load decomposition and identification. If successful, send calibration information containing the identification results and corresponding load feature identification framework parameters to the corresponding data upload front-end, update the front-end lightweight load feature library, and complete the real-time synchronous anchor point calibration of end-cloud collaboration. If unsuccessful, use I in the cloud. n (k), I n-1 (k), U n (k), U n-1 (k) Conduct neural network load decomposition identification. If identification is successful, send calibration information containing the identification results and corresponding load feature identification framework parameters to the corresponding data upload front-end, update the front-end lightweight load feature library, and complete the end-cloud collaborative real-time synchronous anchor point calibration. If identification is unsuccessful, send calibration failure information to the corresponding data upload front-end. The front-end reads the corresponding uploaded I based on the received calibration information. n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k) Calculate the absolute value ΔI of the difference between the effective current value of the current in the current slot and the previous time slot. n And update the preset threshold, setting the preset threshold ΔI th =ΔI n Complete the real-time synchronization anchor point calibration for end-to-cloud collaboration.

[0075] It is easy to see that this invention achieves real-time identification of known loads that have already been trained by building a localized, lightweight incremental judgment model at the front end. Unknown loads are then collaboratively trained and identified in the cloud, constructing a cloud-based artificial intelligence model based on neural networks. Simultaneously, it supports cloud anchor point functionality for end-to-cloud collaboration, meeting front-end calibration requirements, and provides the implementation method for the corresponding device. Through end-to-cloud collaboration, this invention can reduce implementation complexity while ensuring the timeliness and accuracy of load identification, strongly supporting the construction of intelligent fire-fighting electrical safety monitoring systems for scenarios such as intelligent perception and refined management of user electricity consumption characteristics.

[0076] The second embodiment of the present invention relates to an end-to-cloud collaborative real-time synchronous load identification system based on front-end lightweight incremental detection, including a front-end lightweight incremental matching load identification device and a cloud-based neural network load decomposition identification device.

[0077] like Figure 5 As shown, the front-end lightweight incremental load identification device includes a data acquisition module, a processing module, a communication module, and a power supply module. The data acquisition module includes a current signal acquisition unit, a voltage signal acquisition unit, and a sampling control unit. The current signal acquisition unit acquires the user's current signal, the voltage signal acquisition unit acquires the user's voltage signal, and the sampling control unit is used to set and store the sampling frequency f. s This enables the user's current and voltage signals to be sampled at a set frequency f. s Real-time data acquisition is performed; the processing module includes a current and voltage storage unit, a phase synchronization unit, an effective value differential calculation unit, a differential comparison unit, an anchor point calibration control unit, a decision threshold storage unit, a feature matching unit, a front-end lightweight load library, an instantaneous value differential calculation unit, an incremental lightweight neural network recognition unit, a transmit buffer unit, and a receive buffer unit. The current and voltage storage unit stores time-series current data I = [i1, i2, ..., i...] according to a set storage period T. m ] and voltage data U=[u1,u2,...,u m The phase synchronization unit synchronizes the current signal cutoff and extracts the current signal from adjacent time slots by using the zero point of the reference voltage signal; the effective value difference calculation unit calculates the absolute value ΔI of the effective value difference between adjacent time slot current signals. n The differential comparison unit is used to calculate the absolute value of the difference between the effective values, ΔI. n With the threshold ΔI in the decision threshold storage unit th Comparison; Anchor point calibration control unit, used to control whether to perform edge-cloud collaborative anchor point calibration; Decision threshold storage unit, used to store the effective value comparison threshold ΔI. th The feature matching unit is used to calculate the instantaneous current difference ΔI between adjacent time slots based on the instantaneous value difference calculation unit. n (k) performs feature matching with the front-end lightweight load feature library to achieve load identification; the incremental lightweight neural network identification unit is used to perform incremental lightweight neural network identification of the load based on the incremental current instantaneous value difference; the front-end lightweight load library is used to store known load characteristics; the instantaneous value difference calculation unit is used to perform differential calculation to obtain ΔI based on the instantaneous current values ​​of adjacent time slots output by the phase synchronization unit. n (k); Send buffer unit, used to buffer I to be sent to the cloud. n (k), I n-1 (k), ΔI n(k), U n (k), U n-1 (k) and the corresponding front-end anchor point calibration flag information cf; receive buffer unit, used to cache the cloud-sent calibration information containing the identification result and the corresponding load characteristic information or the identification result and the corresponding load characteristic; communication module can be 5G / 4G / WiFi to complete communication with the cloud; power supply module provides power to each module of the front-end lightweight incremental detection load identification device.

[0078] like Figure 6 As shown, the cloud-based neural network load decomposition and identification device includes a receiving module for receiving the load feature information and instantaneous current value difference value sent by the front-end lightweight incremental matching load identification device; and an identification module for performing matching and identification processing based on the load feature information and instantaneous current value difference value to complete end-cloud collaborative load identification and anchor point calibration. The identification module includes a cloud-based load feature library, an unknown load database, a tag identification unit, a neural network load decomposition and identification unit, an incremental feature matching unit, a received data buffer unit, a transmitted data buffer unit, and a 5G / 4G / WiFi communication unit. The cloud-based load feature library stores known load features in the cloud, including local load features stored according to user front-end classifications and global load features stored according to load classifications; the unknown load database stores the I corresponding to unknown-class loads uploaded by user front-ends that were not successfully identified. n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k); Tag identification unit, used to receive unknown load tag information identified by the cloud administrator; Neural network load identification unit, used to implement unknown load decomposition training and identification and anchor point calibration based on neural networks uploaded by each user front-end; Incremental feature matching unit, used to implement the instantaneous value difference ΔI n (k) Feature matching operation with the cloud load feature database to achieve corresponding load identification; Tag identification unit, used to receive unknown load tag information identified by the cloud administrator; Data receiving cache unit, used to cache the I corresponding to unknown category loads uploaded by the user front end. n (k), I n-1 (k), ΔI n (k), U n (k), U n-1 (k); The data caching module is used to cache the identification results and corresponding load characteristic information or the calibration information containing the identification results and corresponding load characteristics to be sent to the user front end; The 5G / 4G / WiFi communication unit completes the communication between the cloud and the front end.

Claims

1. A method for real-time synchronous load identification in edge-cloud collaboration based on lightweight incremental detection at the front end, characterized in that, Includes the following steps: The front end collects the user's current and voltage signals in real time, and extracts load characteristic information and instantaneous current value difference based on the current and voltage signals; The front end performs a matching detection between the instantaneous current value difference and the front end lightweight load feature library. If the matching is successful, the matching load is output. If the matching is unsuccessful, the load increment lightweight neural network recognition is performed. If the recognition is successful, the recognized load is output. If the recognition is unsuccessful, the load feature information and the instantaneous current value difference are sent to the cloud. The cloud performs matching and identification processing based on the load characteristic information and the difference in instantaneous current value, and sends the identification result features and identification framework parameters to the corresponding front end to complete real-time synchronous load identification in a collaborative manner between the cloud and the end, specifically including: Using the instantaneous current difference value, perform local cloud load feature library matching detection on the cloud local load feature library corresponding to the front end that uploaded the instantaneous current difference value; If a match is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If a match is not found, the instantaneous current value difference will be used for matching and detection against the global load feature library in the cloud. If a match is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If a match is not found, the cloud uses the difference in instantaneous current values ​​to perform incremental data neural network load identification. If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If identification is unsuccessful, the cloud will combine the known single-type load characteristics of the global load feature library in the cloud, and use two sets of time-adjacent current time slot sequences and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences to carry out neural network load decomposition identification of load type and corresponding load quantity. If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If identification is still unsuccessful, the cloud will store the instantaneous current difference value of the front end, the two sets of time-adjacent current time slot sequences, and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences into the unknown load database; the unknown load database records the load data that was not successfully identified by different users at different times according to the user number.

2. The method for real-time synchronous load identification based on lightweight incremental detection at the front end and cloud collaboration according to claim 1, characterized in that, The front end collects the user's current and voltage signals, and extracts load characteristic information and instantaneous current value difference based on the current and voltage signals, specifically including: The system collects time-series current and voltage data in real time at a preset sampling frequency, and saves the data at a set time interval. The time-series current and voltage data are sliced ​​according to the time slot length, and the effective value of the current in each slice time slot is calculated; Calculate the absolute value of the difference between the current effective value of the current slot and the previous effective value of the current slot, and compare the absolute value of the difference with a set threshold. If the absolute value of the difference is greater than or equal to a set threshold, then current phase synchronization is initiated to obtain an instantaneous current sequence; Traverse the instantaneous current sequence to find the point where the current changes abruptly, and obtain two sets of time-adjacent current time slot sequences and corresponding voltage time slot sequences. Calculate the difference between the two sets of time-adjacent current time slot sequences to obtain the instantaneous current value difference value, and use the two sets of time-adjacent current time slot sequences and the corresponding voltage time slot sequences as load characteristic information.

3. The method for real-time synchronous load identification based on lightweight incremental detection at the front end and cloud collaboration according to claim 2, characterized in that, When calculating the effective value of the current time slot based on the time slot length and the time-series current data, through... Perform calculations, where, This is the current effective value of the time slot current. The time slot length, This is time-series current data. , , The sampling frequency.

4. The method for real-time synchronous load identification based on lightweight incremental detection at the front end and cloud collaboration according to claim 2, characterized in that, The instantaneous current sequence obtained by synchronizing the starting current phase is specifically as follows: Using the excitation start-up current phase synchronization time slot as a reference, read the timing current data and timing voltage data of the nearest N time slots before and after the time slot, calculate the first voltage phase zero point of the timing voltage data in the reading time slot, and use the first voltage phase zero point as a reference to extract an integer number of cycles from the corresponding timing current data to obtain the instantaneous current sequence.

5. The method for real-time synchronous load identification of end-to-cloud collaboration based on lightweight incremental detection according to claim 1, characterized in that, Also includes: When the accumulated data in the unknown load database exceeds the data volume threshold or reaches the set tag identification period, the cloud counts unknown load type tags according to the stored unknown load type F. For unknown loads that appear in the top P positions among all unknown load types, the cloud administrator is reminded to perform load tag identification. After the tag is determined, the identification result and the corresponding load characteristics are sent to the front end of all users with the load characteristics, thus completing the end-cloud collaborative global load identification.

6. The method for real-time synchronous load identification based on lightweight incremental detection at the front end and cloud collaboration according to claim 2, characterized in that, Also includes: If the absolute value of the difference is less than a set threshold, front-end anchor point calibration flag information is generated, and current phase synchronization is initiated to obtain the instantaneous current sequence. Based on the preset sampling frequency and time slot, the instantaneous current sequence is divided into two sets of time-adjacent current time slot sequences in chronological order. The difference between the two sets of time-adjacent current time slot sequences is calculated to obtain the instantaneous current value difference. The front-end anchor point calibration flag information, the two sets of time-adjacent current time slot sequences, and the voltage time slot sequences corresponding to the two sets of time-adjacent current time slot sequences are used as load characteristic information.

7. The method for real-time synchronous load identification based on lightweight incremental detection at the front end and cloud collaboration according to claim 6, characterized in that, The cloud platform performs matching and identification processing based on the load characteristic information and the difference in instantaneous current value to complete the end-cloud collaborative load identification, specifically including: Based on the instantaneous current value difference and the front-end anchor point calibration flag information, no-load incremental feature matching detection is performed in the cloud load feature database; If a match is successful, empty calibration information is sent to the corresponding data upload front-end to complete the real-time synchronization of anchor point calibration between the end and cloud. If a match is not found, the instantaneous current difference value is used to perform a local cloud load feature library matching detection on the cloud local load feature library corresponding to the front-end that uploaded the instantaneous current difference value. If a match is successful, the identification result and the corresponding load features are sent to the front end to complete the real-time synchronous anchor point calibration of the end-cloud collaboration. If a match is not found, the instantaneous current value difference will be used for matching and detection against the global load feature library in the cloud. If a match is successful, the identification result and the corresponding load features are sent to the front end to complete the real-time synchronous anchor point calibration of the end-cloud collaboration. If the match is unsuccessful, the cloud uses the instantaneous current value difference to perform neural network load decomposition and identification, and after successful identification, sends calibration information containing the identification results and corresponding load characteristics to the front end to update the front end lightweight load feature library; If identification fails, the cloud uses two sets of time-adjacent current time slot sequences and the corresponding voltage time slot sequences of the two sets of time-adjacent current time slot sequences to perform neural network load decomposition identification. After successful identification, calibration information containing the identification results and the corresponding load feature identification framework parameters is sent to the front end to update the front end lightweight load feature library.

8. A real-time synchronous load identification system for end-to-cloud collaboration based on lightweight incremental detection at the front end, characterized in that, This includes a front-end lightweight incremental matching load identification device and a cloud-based neural network load decomposition identification device; The front-end lightweight incremental matching load identification device includes: a data acquisition module for real-time acquisition of the user's current and voltage signals; a processing module for extracting load feature information and instantaneous current difference values ​​based on the current and voltage signals, and performing matching detection between the instantaneous current difference values ​​and the front-end lightweight load feature library. If a match is successful, the matched load is output; if a match is unsuccessful, a lightweight incremental neural network identification of the load is performed. If the identification is successful, the identified load is output; if the identification is unsuccessful, the load feature information and instantaneous current difference values ​​are sent to the cloud; and a communication module for communicating with the cloud. The cloud-based neural network load decomposition and identification device includes: a receiving module for receiving the load feature information and instantaneous current value difference value sent by the front-end lightweight incremental matching load identification device; and an identification module for performing matching and identification processing based on the load feature information and instantaneous current value difference value, and sending the identification result features and identification framework parameters to the corresponding front-end to complete real-time synchronous load identification in a cloud-edge collaborative manner, specifically including: Using the instantaneous current difference value, perform local cloud load feature library matching detection on the cloud local load feature library corresponding to the front end that uploaded the instantaneous current difference value; If a match is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If a match is not found, the instantaneous current value difference will be used for matching and detection against the global load feature library in the cloud. If a match is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If a match is not found, the cloud uses the difference in instantaneous current values ​​to perform incremental data neural network load identification. If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If identification is unsuccessful, the cloud will combine the known single-type load characteristics of the global load feature library in the cloud, and use two sets of time-adjacent current time slot sequences and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences to carry out neural network load decomposition identification of load type and corresponding load quantity. If the identification is successful, the identification result and the corresponding load characteristics are sent to the front end to complete the end-cloud collaborative load identification. If identification is still unsuccessful, the cloud will store the instantaneous current difference value of the front end, the two sets of time-adjacent current time slot sequences, and the voltage sequences corresponding to the two sets of time-adjacent current time slot sequences into the unknown load database; the unknown load database records the load data that was not successfully identified by different users at different times according to the user number.

9. The end-to-cloud collaborative real-time synchronous load identification system based on lightweight incremental detection at the front end according to claim 8, characterized in that, The processing module includes: a current and voltage storage unit for storing time-series current and voltage data according to a set storage period; a phase synchronization unit for synchronously truncating the current signal and extracting the current signal from adjacent time slots by using the zero point of a reference voltage signal; an effective value difference calculation unit for calculating the absolute value of the difference between the effective values ​​of the current signals in adjacent time slots; a difference comparison unit for comparing the absolute value of the difference between the effective values ​​with a set threshold; and a feature matching unit for performing feature matching with a front-end lightweight load feature library based on the instantaneous current difference value output by the instantaneous value difference calculation unit to achieve load identification. The system includes: an incremental lightweight neural network recognition unit for performing incremental lightweight neural network recognition of load based on the incremental instantaneous current value difference; a front-end lightweight load feature library for storing known load features; an instantaneous value difference calculation unit for calculating the instantaneous current value difference based on the difference between adjacent current time slot sequences output by the phase synchronization unit; and a transmission buffer unit for buffering the instantaneous current value difference to be sent to the cloud, two sets of time-adjacent current time slot sequences, the voltage time slot sequences corresponding to the two sets of time-adjacent current time slot sequences, and the corresponding front-end anchor point calibration flag information. The identification module includes: a cloud-based load feature library for storing known load features in the cloud, including local load features stored according to user front-end classification and global load features stored according to load classification; an unknown load database for storing the instantaneous current difference value, two sets of time-adjacent current time slot sequences, and voltage time slot sequences corresponding to the unknown category loads uploaded by the front-ends that were not successfully identified; a tag identification unit for receiving unknown load tag information identified by the cloud administrator; a neural network load identification unit for implementing decomposition training and identification of unknown loads uploaded by each front-end based on neural networks; and an incremental feature matching unit for implementing feature matching operations based on the instantaneous value difference value and the cloud-based load feature library to achieve corresponding load identification.

Images