An event detection-based power fingerprinting method, device and equipment

Abstract

The present application relates to the technical field of load identification, and discloses a power fingerprint identification method, device and equipment based on event detection. The present application adopts a sliding window with multiple margin windows at both ends to sample electrical quantity data, and then calculates the mean values of the electrical quantity sampled by the main margin window at the left end of the sliding window, the secondary margin window at the left end, the main margin window at the right end and the secondary margin window at the right end. If the absolute values of the differences between the mean values of the electrical quantity corresponding to the main margin window and the secondary margin window at both ends of the sliding window are greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switching event is greater than the preset time threshold, it is determined that a load switching event occurs and the corresponding load waveform data is extracted. According to the load waveform data corresponding to each load switching event extracted, power fingerprint features are extracted and input into the trained power fingerprint identification model, and the type identification result of the load is output. The present application can effectively improve the identification accuracy of power fingerprint identification.

Description

Technical Field

[0001] This invention relates to the field of load identification technology, and in particular to a power fingerprint identification method, apparatus and equipment based on event detection. Background Technology

[0002] By monitoring electrical data from electrical equipment, artificial intelligence and big data technologies are used to mine feature points that characterize the equipment's properties. The aggregation of these feature points across various dimensions forms the equipment's "electrical fingerprint." Building upon existing research on load characteristics and identification methods, electrical fingerprint technology further expands the connotation of load characteristics. It enables the identification of equipment type, characteristics, parameters, user behavior habits, energy efficiency, health level, and even the equipment's identity.

[0003] Power fingerprint recognition can be divided into two categories: non-event-based recognition and event-based recognition. Non-event-based recognition directly assigns the aggregated features of the bus to each electrical device; event-based recognition focuses on the sudden changes in bus electrical features caused by load device state switching (also known as load events), and identifies electrical devices by detecting and extracting the amount of change. Compared with non-event-based recognition, event-based recognition has better real-time performance and overall recognition accuracy.

[0004] However, when using event detection for power fingerprinting, long transients and surge currents generated when certain loads are connected, as well as abnormal fluctuations in measurements taken by monitoring equipment, can all lead to false detections, resulting in a decrease in overall recognition accuracy. Furthermore, the limited number of load samples can also cause a decrease in the robustness of the power fingerprinting model, leading to a decline in recognition accuracy under combined load disturbances. Summary of the Invention

[0005] This invention provides an event-detection-based power fingerprint identification method, apparatus, and device, which solves the technical problem that the accuracy of existing event-detection-based power fingerprint identification methods still needs to be improved.

[0006] The first aspect of this invention provides an event-detection-based power fingerprint recognition method, comprising:

[0007] Electrical quantity data is sampled using a sliding window with multiple margin windows at both ends; the length of the sliding window is greater than the sum of the lengths of the multiple margin windows at both ends of the sliding window, and the multiple margin windows at each end consist of a main margin window and a secondary margin window;

[0008] After each sampling, the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window is calculated respectively. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, it is determined that a load switch event has occurred and the corresponding load waveform data is extracted.

[0009] Based on the load waveform data corresponding to each load switch event, extract the corresponding power fingerprint features, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

[0010] According to one achievable method of the first aspect of the invention, the electrical quantity data is an active power sequence consisting of active power in each cycle, the active power being calculated from voltage and current measured in the corresponding cycle.

[0011] According to one achievable method of the first aspect of the present invention, the corresponding load waveform data includes voltage and current data before and after the occurrence of the corresponding load switching event, and the step of extracting corresponding power fingerprint features based on the extracted load waveform data corresponding to each load switching event includes:

[0012] Using the zero-crossing point of the voltage rise phase as the boundary, the voltage and current data before and after each load switch event are segmented so that each data segment is a cycle of data.

[0013] If the load switch event is a load connection, the current data before the corresponding event is subtracted from the current data segment after the event to obtain the current data after the disconnection; if the load switch event is a load disconnection, the current data after the corresponding event is subtracted from the current data segment before the event to obtain the current data after the disconnection.

[0014] The separated current data is combined with the voltage data segment after the corresponding event to obtain the separated load data.

[0015] Extract the power fingerprint features of each of the separated load data.

[0016] According to one feasible method of the first aspect of the present invention, the extraction of power fingerprint features of each of the separated load data includes:

[0017] The power characteristics, harmonic characteristics, and waveform characteristics of the separated load data are extracted. The power characteristics include active power, reactive power, apparent power, and power factor. The harmonic characteristics include the effective values ​​and phases of the fundamental current and the 2nd, 3rd, 5th, and 7th harmonic currents and impedances, as well as the aggregated effective values ​​of the 4th-6th-8th-10th, 9th-12th-15th, and 11th-13th-17th-19th-23rd-25th harmonic currents. The waveform characteristics include total harmonic distortion, waveform factor, and peak factor.

[0018] According to one feasible method of the first aspect of the present invention, the extraction of power characteristics, harmonic characteristics, and waveform characteristics of the separated load data includes:

[0019] The effective value of each current harmonic aggregation is calculated using the following formula:

[0020]

[0021] In the formula, I Σ I represents the effective value of current harmonic aggregation. h This represents the effective value of the h-th harmonic current, where H is the set of harmonic orders aggregated.

[0022] According to one achievable embodiment of the first aspect of the invention, the method further includes:

[0023] Collect steady-state load waveform data of various types of individual loads and label them. Extract the power fingerprint features of the corresponding individual loads based on the labeled steady-state load waveform data and construct an initial training set.

[0024] The initial training set is augmented to obtain the augmented training set;

[0025] The power fingerprint recognition model is trained using the data augmented training set to obtain the trained power fingerprint recognition model.

[0026] According to one feasible method of the first aspect of the present invention, the extraction of the power fingerprint features of the corresponding individual load based on the tagged steady-state load waveform data includes:

[0027] The steady-state load waveform data after being tagged is preprocessed; the preprocessing includes filling in missing data values, removing abnormal data, removing data that does not match the corresponding tag and / or removing data that does not match the corresponding switch state;

[0028] The power fingerprint features of the corresponding individual loads are extracted from the preprocessed steady-state load waveform data.

[0029] According to one achievable method of the first aspect of the present invention, the data augmentation of the initial training set includes:

[0030] Select a set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase so that each data segment is a cycle of data. Then, select two unequal current waveforms from each data segment and subtract them to obtain the corresponding disturbance waveform data.

[0031] Repeat the previous step to obtain multiple disturbance waveform data, and construct a disturbance waveform dataset based on the multiple disturbance waveform data;

[0032] Select any set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase. Determine the target current data segment from each data segment. Randomly select a number of disturbance waveform data from the disturbance waveform dataset and superimpose them onto the target current data segment. Do not change the voltage data and label corresponding to the target current data segment to obtain the reconstructed load data.

[0033] Repeat the previous step to obtain multiple reconstruction load data, and construct a reconstruction load dataset based on the multiple reconstruction load data;

[0034] The power fingerprint features of each individual load in the reconstructed load dataset are extracted and added to the initial training set.

[0035] According to one aspect of the present invention, training the power fingerprint recognition model using the data-augmented training set includes:

[0036] A random forest model is constructed as the power fingerprinting model.

[0037] A second aspect of the present invention provides an event-detection-based power fingerprint recognition device, comprising:

[0038] The sampling module is used to sample electrical quantity data using a sliding window with multi-level margin windows at both ends; the length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends of the sliding window, and the multi-level margin window at each end consists of a main margin window and a secondary margin window;

[0039] The event detection module is used to calculate the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window after each sampling. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, it is determined that a load switch event has occurred and the corresponding load waveform data is extracted.

[0040] The power fingerprint recognition module is used to extract corresponding power fingerprint features based on the load waveform data corresponding to each load switching event, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

[0041] According to one achievable method of the second aspect of the invention, the electrical quantity data is an active power sequence consisting of active power in each cycle, the active power being calculated from voltage and current measured in the corresponding cycle.

[0042] According to one embodiment of the second aspect of the present invention, the corresponding load waveform data includes voltage and current data before and after the occurrence of the corresponding load switching event, and the power fingerprint recognition module includes:

[0043] The waveform alignment unit is used to segment the voltage and current data before and after each load switching event, with the zero-crossing point of the voltage rise phase as the boundary, so that each data segment obtained is a cycle of data.

[0044] The first incremental extraction unit is used to subtract the current data before the corresponding event from the current data segment after the corresponding event to obtain the separated current data if the load switch event is a load connection; and to subtract the current data after the corresponding event from the current data segment before the corresponding event to obtain the separated current data if the load switch event is a load disconnection.

[0045] A data merging unit is used to merge the separated current data with the voltage data segment after the corresponding event to obtain the separated load data.

[0046] The first feature extraction unit is used to extract the power fingerprint features of each of the separated load data.

[0047] According to one achievable embodiment of the second aspect of the present invention, the first feature extraction unit comprises:

[0048] The feature extraction subunit is used to extract the power characteristics, harmonic characteristics, and waveform characteristics of the separated load data. The power characteristics include active power, reactive power, apparent power, and power factor. The harmonic characteristics include the effective values ​​and phases of the fundamental wave and the 2nd, 3rd, 5th, and 7th harmonic currents and impedances, as well as the aggregated effective values ​​of the 4th-6th-8th-10th, 9th-12th-15th, and 11th-13th-17th-19th-23rd-25th harmonic currents. The waveform characteristics include total harmonic distortion, waveform factor, and peak factor.

[0049] According to one achievable method of the second aspect of the present invention, the feature extraction subunit is specifically used for:

[0050] The effective value of each current harmonic aggregation is calculated using the following formula:

[0051]

[0052] In the formula, I Σ I represents the effective value of current harmonic aggregation. h This represents the effective value of the h-th harmonic current, where H is the set of harmonic orders aggregated.

[0053] According to one embodiment of the second aspect of the invention, the apparatus further includes:

[0054] The initial training set construction module is used to collect steady-state load waveform data of various types of individual loads and label them. Based on the labeled steady-state load waveform data, the power fingerprint features of the corresponding individual loads are extracted to construct the initial training set.

[0055] The training set augmentation module is used to augment the initial training set to obtain an augmented training set.

[0056] The model training module is used to train the power fingerprint recognition model using the data augmentation training set to obtain the trained power fingerprint recognition model.

[0057] According to one achievable method of the second aspect of the present invention, the initial training set construction module comprises:

[0058] The data preprocessing unit is used to preprocess the labeled steady-state load waveform data; the preprocessing includes filling missing data values, removing abnormal data, removing data that does not match the corresponding label and / or removing data that does not match the corresponding switch state;

[0059] The second feature extraction unit is used to extract the power fingerprint features of the corresponding individual loads from the preprocessed steady-state load waveform data.

[0060] According to one achievable method of the second aspect of the present invention, the training set augmentation module includes:

[0061] The second incremental extraction unit is used to select a set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase so that each data segment is a cycle of data. It also selects two unequal current waveforms from each data segment and subtracts them to obtain the corresponding disturbance waveform data.

[0062] A disturbance waveform dataset construction unit is used to repeat the second incremental extraction unit to obtain multiple disturbance waveform data, and to construct a disturbance waveform dataset based on the multiple disturbance waveform data;

[0063] The third incremental extraction unit is used to select any set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase. It determines the target current data segment from each data segment and randomly selects a number of disturbance waveform data from the disturbance waveform dataset to superimpose on the target current data segment. The voltage data and label corresponding to the target current data segment are not changed to obtain the reconstructed load data.

[0064] A reconstruction load dataset construction unit is used to repeat the third incremental extraction unit to obtain multiple reconstruction load data, and to construct a reconstruction load dataset based on the multiple reconstruction load data.

[0065] The third feature extraction unit is used to extract the power fingerprint features of each individual load in the reconstructed load dataset and add them to the initial training set.

[0066] According to one achievable method of the second aspect of the present invention, the model training module includes:

[0067] The model building unit is used to build a random forest model as the power fingerprint recognition model.

[0068] A third aspect of the present invention provides an event-detection-based power fingerprint recognition device, comprising:

[0069] A memory for storing instructions; wherein the instructions are used to implement the event detection-based power fingerprinting method as described in any of the above embodiments.

[0070] A processor for executing instructions in the memory.

[0071] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the event-detection-based power fingerprinting method as described in any of the above embodiments.

[0072] As can be seen from the above technical solutions, the present invention has the following advantages:

[0073] This invention employs a sliding window with multi-level margin windows at both ends to sample electrical quantity data. The length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends. Each multi-level margin window consists of a primary margin window and a secondary margin window. After each sampling, the average value of the electrical quantities sampled by the primary margin window, the secondary margin window at the left end, the primary margin window at the right end, and the secondary margin window at the right end of the sliding window are calculated respectively. If the absolute value of the difference between the average electrical quantities corresponding to the primary margin window and the secondary margin window at both ends of the sliding window is greater than a preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than a preset time threshold, a load switch event is determined to have occurred, and the corresponding load waveform data is extracted. Based on the extracted load waveform data corresponding to each load switch event, the corresponding power fingerprint features are extracted. The power fingerprint features are input into a trained power fingerprint recognition model, and the load type recognition result is output. This invention, based on a sliding window with multi-level margin windows at both ends for data sampling and event detection, can effectively solve the false detection caused by long transient loads, abnormal value fluctuations, and surge currents, and effectively improve the recognition accuracy of power fingerprint recognition. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 A flowchart illustrating an event-detection-based power fingerprinting method, provided as an optional embodiment of the present invention;

[0076] Figure 2 A flowchart of an event detection-based power fingerprinting method provided as another optional embodiment of the present invention;

[0077] Figure 3 The diagram below shows the structural connection of an event-detection-based power fingerprint recognition device, which is provided as an optional embodiment of the present invention.

[0078] Figure 4 The diagram below shows the structural connection of an event-detection-based power fingerprint recognition device, which is provided as another optional embodiment of the present invention.

[0079] Figure label:

[0080] 1-Sampling module; 2-Event detection module; 3-Electricity fingerprint recognition module; 4-Initial training set construction module; 5-Training set enhancement module; 6-Model training module. Detailed Implementation

[0081] This invention provides an event-detection-based power fingerprint identification method, apparatus, and device to address the technical problem that the accuracy of existing event-detection-based power fingerprint identification methods still needs to be improved.

[0082] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0083] This invention provides an event detection-based power fingerprinting method.

[0084] Please see Figure 1 , Figure 1 A flowchart of an event detection-based power fingerprinting method provided by an embodiment of the present invention is shown.

[0085] This invention provides an event-detection-based power fingerprint recognition method, comprising:

[0086] Step S1: Sampling of electrical quantity data is performed using a sliding window with multi-level margin windows at both ends; the length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends of the sliding window, and the multi-level margin window at each end is composed of a main margin window and a secondary margin window.

[0087] In this embodiment, a sliding window with multi-level margin windows at both ends is used for data sampling. Specifically, let the length of the sliding window be N and the length of the multi-level margin window be N. m The length of the main edge distance window is N m1 The length of the secondary side distance window is N m2 Then each parameter satisfies 2N m ≤N,N m1 +N m2 =N m .

[0088] Step S2: After each sampling, calculate the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window, respectively. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, determine that a load switch event has occurred and extract the corresponding load waveform data.

[0089] In one feasible approach, the electrical quantity data is an active power sequence composed of active power for each cycle, the active power being calculated from the voltage and current measured in the corresponding cycle. Based on this, the specific process for determining a load switch event in step S2 is as follows: if the absolute value of the difference between the average active power values ​​corresponding to the primary and secondary margin windows at both ends of the sliding window is greater than a preset active power threshold, and the time interval between the current sampling time and the previous load switch event is greater than a preset time threshold, a load switch event is determined to have occurred.

[0090] As a specific implementation method, the relevant calculation formula for determining whether a load switch event has occurred is as follows:

[0091]

[0092] In the formula, p j Let be the active power in the j-th cycle. For j k Voltage sample value at time 10:00 For j k The current sample value at time M, where M is the number of voltage / current sampling points per cycle, l and r are the indices of the left and right boundary points of the sliding window at the current time, respectively, and N is the current sample value at time N. m1 N represents the length of the left / right main margin of the window. m2 P is the length of the left / right secondary margin window. 1L P 1R P 2L P 2R ΔP1 and ΔP2 are the average active power values ​​corresponding to the left main edge window, right main edge window, left secondary edge window, and right secondary edge window, respectively. ΔP1 and ΔP2 are the differences between the average active power values ​​corresponding to the main edge window and the secondary edge window, respectively. h is the preset active power threshold for the occurrence of a load switch event, and t is the value of t. lim The preset time threshold is Δt, which is the time interval between the current sampling time and the previous load switch event. event_type=on indicates that a load access event has occurred, event_type=off indicates that a load disconnection event has occurred, and event_type=none indicates that no load switch event has been sent.

[0093] A load switch event indicates either a load connection event or a load disconnection event. In this embodiment, a criterion formula is set to further determine load connection and load disconnection events.

[0094] In one implementation, the voltage / current sample values ​​are obtained by the smart socket. A suitable sampling frequency can be set, for example, a sampling frequency greater than or equal to 1000Hz.

[0095] In another possible implementation, the electrical quantity data can also be a sequence of effective current values ​​calculated from the current sampling values ​​of each cycle. Based on this, the specific process of determining the occurrence of a load switch event in step S2 is as follows: calculate the average effective current values ​​in the main margin window, the secondary margin window, the main margin window, and the secondary margin window at the left and right ends of the sliding window, respectively. If the absolute value of the difference between the average effective current values ​​corresponding to the main margin window and the secondary margin window at both ends of the sliding window is greater than a preset effective current value threshold, and the time interval between the current sampling time and the previous load switch event is greater than a preset time threshold, then a load switch event is determined to have occurred.

[0096] The above embodiments of the present invention employ an event detection scheme based on a multi-level margin sliding window with time intervals, which achieves effective detection of load switch events. The time interval setting can solve the false detection of events caused by long transients when certain loads are connected, and the multi-level margin setting can solve the problem of abnormal value fluctuations in load data during measurement and false detection of events caused by load surge currents, thereby effectively improving the overall accuracy of power fingerprint recognition.

[0097] Step S3: Extract the corresponding power fingerprint features based on the load waveform data corresponding to each load switch event, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

[0098] In one feasible approach, the corresponding load waveform data includes voltage and current data before and after the occurrence of the corresponding load switching event. The step of extracting corresponding power fingerprint features based on the extracted load waveform data corresponding to each load switching event includes:

[0099] Using the zero-crossing point of the voltage rise phase as the boundary, the voltage and current data before and after each load switch event are segmented so that each data segment is a cycle of data.

[0100] If the load switch event is a load connection, the current data before the corresponding event is subtracted from the current data segment after the event to obtain the current data after the disconnection; if the load switch event is a load disconnection, the current data after the corresponding event is subtracted from the current data segment before the event to obtain the current data after the disconnection.

[0101] The separated current data is combined with the voltage data segment after the corresponding event to obtain the separated load data.

[0102] Extract the power fingerprint features of each of the separated load data.

[0103] In this embodiment, by segmenting the voltage and current data before and after each load switching event using the zero-crossing point of the voltage rise phase as the boundary, waveform alignment can be achieved. This facilitates subsequent direct separation of the load through matrix operations to obtain the separated load data. The segmented voltage and current data U of the j-th cycle obtained from this step... j I j It can be represented as:

[0104]

[0105] In the formula, They represent the j1, j2, ..., j-th elements respectively. k ,…,j M The voltage sample value at time M, where M is the number of voltage / current sampling points per cycle (i.e., period). They represent the j1, j2, ..., j-th elements respectively. k ,…,j M The current sample value at time t.

[0106] The expression for separating the load through matrix operations can be represented as:

[0107]

[0108] j_before_event < j_event_flag

[0109] j_after_event > j_event_flag

[0110] In the formula, I dec This represents the separated current data, I. j_after_event This represents the current data segment after the event occurs, I j_before_event This represents the current data segment before the event occurs. j_event_flag represents the cycle number when the event occurs, j_before_event represents the cycle number before the event occurs, and j_after_event represents the cycle number after the event occurs.

[0111] Furthermore, the expression for the separated load data is as follows:

[0112] X dec ={U j_after_event ,I dec}

[0113] In the formula, X dec Represents the load data after separation, U j_after_event This represents the voltage data segment following the corresponding event.

[0114] In one feasible manner, the extraction of power fingerprint features from each of the separated post-load data includes:

[0115] The power characteristics, harmonic characteristics, and waveform characteristics of the separated load data are extracted. The power characteristics include active power, reactive power, apparent power, and power factor. The harmonic characteristics include the effective values ​​and phases of the fundamental current and the 2nd, 3rd, 5th, and 7th harmonic currents and impedances, as well as the aggregated effective values ​​of the 4th-6th-8th-10th, 9th-12th-15th, and 11th-13th-17th-19th-23rd-25th harmonic currents. The waveform characteristics include total harmonic distortion, waveform factor, and peak factor.

[0116] The effective values ​​and phases of each harmonic can be calculated using Fourier transform.

[0117] In one feasible approach, the effective value of the aggregated current harmonics is calculated according to the following formula:

[0118]

[0119] In the formula, I Σ I represents the effective value of current harmonic aggregation. h This represents the effective value of the h-th harmonic current, where H is the set of harmonic orders aggregated.

[0120] In one achievable manner, the total harmonic distortion (THD) i The calculation formula is as follows:

[0121]

[0122] In the formula, I1 represents the effective value of the fundamental current.

[0123] In one feasible manner, the waveform factor k a The calculation formula is as follows:

[0124]

[0125] In the formula, I arv I is the absolute mean of the current. rms i is the effective value of the current. k Let M be the k-th current value, and M be the number of current values ​​in one cycle.

[0126] In one feasible approach, the peak factor k f The calculation formula is as follows:

[0127]

[0128] In the formula, I peakThis represents the peak current.

[0129] The above embodiments of the present invention set power fingerprint features including power features, harmonic features and waveform features, which further expands the connotation of load features on the basis of the prior art, and can more accurately express the power consumption status of the equipment and achieve more refined identification.

[0130] Please see Figure 2 , Figure 2 A flowchart of an event detection-based power fingerprinting method provided by another optional embodiment of the present invention is shown.

[0131] In one feasible way, Figure 1 Based on the method shown, the method described in this embodiment further includes:

[0132] Step S4: Collect steady-state load waveform data of each type of individual load and label them. Extract the power fingerprint features of the corresponding individual load based on the labeled steady-state load waveform data and construct the initial training set.

[0133] Step S5: Perform data augmentation on the initial training set to obtain the data-augmented training set;

[0134] Step S6: Use the data-enhanced training set to train the power fingerprint recognition model to obtain the trained power fingerprint recognition model.

[0135] To address the problems of limited load samples and high data collection costs in existing technologies, this invention augments the initial training set and uses the augmented training set to train the power fingerprint recognition model. This effectively improves the robustness of the model and increases the recognition accuracy of loads under combined load disturbances, even with limited load samples and high data collection costs.

[0136] In one feasible approach, the extraction of the power fingerprint features of the corresponding individual loads based on the tagged steady-state load waveform data includes:

[0137] The steady-state load waveform data after being tagged is preprocessed; the preprocessing includes filling in missing data values, removing abnormal data, removing data that does not match the corresponding tag and / or removing data that does not match the corresponding switch state;

[0138] The power fingerprint features of the corresponding individual loads are extracted from the preprocessed steady-state load waveform data.

[0139] In this embodiment, the accuracy of data augmentation operation is improved by preprocessing the labeled steady-state load waveform data.

[0140] In one feasible manner, the data augmentation of the initial training set includes:

[0141] Select a set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase so that each data segment is a cycle of data. Then, select two unequal current waveforms from each data segment and subtract them to obtain the corresponding disturbance waveform data.

[0142] Repeat the previous step to obtain multiple disturbance waveform data, and construct a disturbance waveform dataset based on the multiple disturbance waveform data;

[0143] Select any set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase. Determine the target current data segment from each data segment. Randomly select a number of disturbance waveform data from the disturbance waveform dataset and superimpose them onto the target current data segment. Do not change the voltage data and label corresponding to the target current data segment to obtain the reconstructed load data.

[0144] Repeat the previous step to obtain multiple reconstruction load data, and construct a reconstruction load dataset based on the multiple reconstruction load data;

[0145] The power fingerprint features of each individual load in the reconstructed load dataset are extracted and added to the initial training set.

[0146] In this embodiment of the invention, the process of generating reconstructed load data is basically the same as the process of separating target load data from combined load. It can simulate target load data under combined load disturbance, allowing the model to explore more essential fingerprint features of the load under the influence of combined load disturbance, and improve the identification accuracy of the load under combined load disturbance.

[0147] In one feasible manner, training the power fingerprint recognition model using the data-augmented training set includes:

[0148] A random forest model is constructed as the power fingerprinting model.

[0149] Random forest is a supervised ensemble learning classification technique. Its classification model consists of a set of decision tree classifiers. The model classifies data by collectively voting on the classification results of individual decision trees. It combines Leo Breiman's Bagging ensemble learning theory with Ho's random subspace method. By injecting randomness into the training sample space and attribute space, it fully ensures the independence and diversity between each decision tree, effectively overcoming the overfitting problem of decision trees, and also showing good robustness to noise and outliers. In this embodiment, a random forest model is used as the power fingerprint recognition model, which can ensure a certain degree of robustness in the power fingerprint recognition process.

[0150] It should be noted that in other embodiments, other applicable machine learning models may also be used as power fingerprint recognition models.

[0151] The present invention also provides an event-detection-based power fingerprint recognition device, which can be used to execute the event-detection-based power fingerprint recognition method described in any of the above embodiments of the present invention.

[0152] Please see Figure 3 , Figure 3 The diagram shows a structural connection block diagram of an event detection-based power fingerprint recognition device provided in an embodiment of the present invention.

[0153] An embodiment of the present invention provides an event-detection-based power fingerprint recognition device, comprising:

[0154] Sampling module 1 is used to sample electrical quantity data using a sliding window with multi-level margin windows at both ends; the length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends of the sliding window, and the multi-level margin window at each end is composed of a main margin window and a secondary margin window;

[0155] Event detection module 2 is used to calculate the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window after each sampling. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, it is determined that a load switch event has occurred and the corresponding load waveform data is extracted.

[0156] The power fingerprint recognition module 3 is used to extract corresponding power fingerprint features based on the load waveform data corresponding to each load switching event, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

[0157] In one feasible approach, the electrical quantity data is an active power sequence consisting of active power for each cycle, which is calculated from the voltage and current measured in the corresponding cycle.

[0158] In one feasible implementation, the corresponding load waveform data includes voltage and current data before and after the corresponding load switching event occurs, and the power fingerprint recognition module 3 includes:

[0159] The waveform alignment unit is used to segment the voltage and current data before and after each load switching event, with the zero-crossing point of the voltage rise phase as the boundary, so that each data segment obtained is a cycle of data.

[0160] The first incremental extraction unit is used to subtract the current data before the corresponding event from the current data segment after the corresponding event to obtain the separated current data if the load switch event is a load connection; and to subtract the current data after the corresponding event from the current data segment before the corresponding event to obtain the separated current data if the load switch event is a load disconnection.

[0161] A data merging unit is used to merge the separated current data with the voltage data segment after the corresponding event to obtain the separated load data.

[0162] The first feature extraction unit is used to extract the power fingerprint features of each of the separated load data.

[0163] In one feasible implementation, the first feature extraction unit includes:

[0164] The feature extraction subunit is used to extract the power characteristics, harmonic characteristics, and waveform characteristics of the separated load data. The power characteristics include active power, reactive power, apparent power, and power factor. The harmonic characteristics include the effective values ​​and phases of the fundamental wave and the 2nd, 3rd, 5th, and 7th harmonic currents and impedances, as well as the aggregated effective values ​​of the 4th-6th-8th-10th, 9th-12th-15th, and 11th-13th-17th-19th-23rd-25th harmonic currents. The waveform characteristics include total harmonic distortion, waveform factor, and peak factor.

[0165] In one feasible implementation, the feature extraction subunit is specifically used for:

[0166] The effective value of each current harmonic aggregation is calculated using the following formula:

[0167]

[0168] In the formula, I Σ I represents the effective value of current harmonic aggregation. hThis represents the effective value of the h-th harmonic current, where H is the set of harmonic orders aggregated.

[0169] Please see Figure 4 , Figure 4 The diagram shows a structural connection block diagram of an event detection-based power fingerprint recognition device provided in an embodiment of the present invention.

[0170] In one feasible way, such as Figure 4 As shown, in Figure 3 Based on the structure of the device shown, the device further includes:

[0171] The initial training set construction module 4 is used to collect steady-state load waveform data of various types of individual loads and label them. Based on the labeled steady-state load waveform data, the power fingerprint features of the corresponding individual loads are extracted to construct the initial training set.

[0172] Training set augmentation module 5 is used to augment the initial training set to obtain an augmented training set.

[0173] Model training module 6 is used to train the power fingerprint recognition model using the data augmentation training set to obtain the trained power fingerprint recognition model.

[0174] In one feasible manner, the initial training set construction module 4 includes:

[0175] The data preprocessing unit is used to preprocess the labeled steady-state load waveform data; the preprocessing includes filling missing data values, removing abnormal data, removing data that does not match the corresponding label and / or removing data that does not match the corresponding switch state;

[0176] The second feature extraction unit is used to extract the power fingerprint features of the corresponding individual loads from the preprocessed steady-state load waveform data.

[0177] In one feasible manner, the training set augmentation module 5 includes:

[0178] The second incremental extraction unit is used to select a set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase so that each data segment is a cycle of data. It also selects two unequal current waveforms from each data segment and subtracts them to obtain the corresponding disturbance waveform data.

[0179] A disturbance waveform dataset construction unit is used to repeat the second incremental extraction unit to obtain multiple disturbance waveform data, and to construct a disturbance waveform dataset based on the multiple disturbance waveform data;

[0180] The third incremental extraction unit is used to select any set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase. It determines the target current data segment from each data segment and randomly selects a number of disturbance waveform data from the disturbance waveform dataset to superimpose on the target current data segment. The voltage data and label corresponding to the target current data segment are not changed to obtain the reconstructed load data.

[0181] A reconstruction load dataset construction unit is used to repeat the third incremental extraction unit to obtain multiple reconstruction load data, and to construct a reconstruction load dataset based on the multiple reconstruction load data.

[0182] The third feature extraction unit is used to extract the power fingerprint features of each individual load in the reconstructed load dataset and add them to the initial training set.

[0183] In one feasible implementation, the model training module 6 includes:

[0184] The model building unit is used to build a random forest model as the power fingerprint recognition model.

[0185] The present invention also provides an event-detection-based power fingerprint recognition device, comprising:

[0186] A memory for storing instructions; wherein the instructions are used to implement the event detection-based power fingerprinting method as described in any of the above embodiments;

[0187] A processor for executing instructions in the memory.

[0188] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the event detection-based power fingerprinting method as described in any of the above embodiments.

[0189] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, equipment, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and the specific beneficial effects of the devices, equipment, and modules described above can be referred to the corresponding beneficial effects in the foregoing method embodiments, and will not be repeated here.

[0190] In the several embodiments provided in this application, it should be understood that the disclosed apparatus, devices, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0191] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0192] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0193] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0194] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power fingerprint recognition method based on event detection, characterized in that, include: Electrical quantity data are sampled using a sliding window with multiple margin windows at both ends; The length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends of the sliding window, and the multi-level margin window at each end consists of a main margin window and a secondary margin window; After each sampling, the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window is calculated respectively. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, it is determined that a load switch event has occurred and the corresponding load waveform data is extracted. Based on the load waveform data corresponding to each load switch event, extract the corresponding power fingerprint features, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

2. The power fingerprint recognition method based on event detection according to claim 1, characterized in that, The electrical quantity data is an active power sequence composed of active power in each cycle, and the active power is calculated from the voltage and current measured in the corresponding cycle.

3. The power fingerprint recognition method based on event detection according to claim 1, characterized in that, The corresponding load waveform data includes voltage and current data before and after the occurrence of the corresponding load switching event. The step of extracting corresponding power fingerprint features based on the extracted load waveform data for each load switching event includes: Using the zero-crossing point of the voltage rise phase as the boundary, the voltage and current data before and after each load switch event are segmented so that each data segment is a cycle of data. If the load switch event is a load connection, the current data before the corresponding event is subtracted from the current data segment after the event to obtain the current data after the disconnection; if the load switch event is a load disconnection, the current data after the corresponding event is subtracted from the current data segment before the event to obtain the current data after the disconnection. The separated current data is combined with the voltage data segment after the corresponding event to obtain the separated load data. Extract the power fingerprint features of each of the separated load data.

4. The power fingerprint recognition method based on event detection according to claim 3, characterized in that, The extraction of power fingerprint features from each of the separated load data includes: The power characteristics, harmonic characteristics, and waveform characteristics of the separated load data are extracted. The power characteristics include active power, reactive power, apparent power, and power factor. The harmonic characteristics include the effective values ​​and phases of the fundamental current and the 2nd, 3rd, 5th, and 7th harmonic currents and impedances, as well as the aggregated effective values ​​of the 4th-6th-8th-10th, 9th-12th-15th, and 11th-13th-17th-19th-23rd-25th harmonic currents. The waveform characteristics include total harmonic distortion, waveform factor, and peak factor.

5. The power fingerprint recognition method based on event detection according to claim 4, characterized in that, The extraction of power characteristics, harmonic characteristics, and waveform characteristics from the separated load data includes: The effective value of each current harmonic aggregation is calculated using the following formula: In the formula, I Σ I represents the effective value of current harmonic aggregation. h This represents the effective value of the h-th harmonic current, where H is the set of harmonic orders aggregated.

6. The power fingerprint recognition method based on event detection according to claim 1, characterized in that, The method further includes: Collect steady-state load waveform data of various types of individual loads and label them. Extract the power fingerprint features of the corresponding individual loads based on the labeled steady-state load waveform data and construct an initial training set. The initial training set is augmented to obtain the augmented training set; The power fingerprint recognition model is trained using the data augmentation training set to obtain the trained power fingerprint recognition model.

7. The power fingerprint recognition method based on event detection according to claim 6, characterized in that, The step of extracting the power fingerprint features of corresponding individual loads based on the tagged steady-state load waveform data includes: The steady-state load waveform data after being tagged is preprocessed; the preprocessing includes filling in missing data values, removing abnormal data, removing data that does not match the corresponding tag and / or removing data that does not match the corresponding switch state; The power fingerprint features of the corresponding individual loads are extracted from the preprocessed steady-state load waveform data.

8. The power fingerprint recognition method based on event detection according to claim 7, characterized in that, The data augmentation of the initial training set includes: Select a set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase so that each data segment is a cycle of data. Then, select two unequal current waveforms from each data segment and subtract them to obtain the corresponding disturbance waveform data. Repeat the previous step to obtain multiple disturbance waveform data, and construct a disturbance waveform dataset based on the multiple disturbance waveform data; Select any set of data from the preprocessed steady-state load waveform data and segment it according to the zero-crossing point of the voltage rise phase. Determine the target current data segment from each data segment. Randomly select a number of disturbance waveform data from the disturbance waveform dataset and superimpose them onto the target current data segment. Do not change the voltage data and label corresponding to the target current data segment to obtain the reconstructed load data. Repeat the previous step to obtain multiple reconstruction load data, and construct a reconstruction load dataset based on the multiple reconstruction load data; The power fingerprint features of each individual load in the reconstructed load dataset are extracted and added to the initial training set.

9. The power fingerprint recognition method based on event detection according to claim 6, characterized in that, The process of training the power fingerprint recognition model using the data-enhanced training set includes: A random forest model is constructed as the power fingerprint recognition model.

10. An event-detection-based power fingerprint recognition device, characterized in that, include: The sampling module is used to sample electrical quantity data using a sliding window with multi-level margin windows at both ends; The length of the sliding window is greater than the sum of the lengths of the multi-level margin windows at both ends of the sliding window, and the multi-level margin window at each end consists of a main margin window and a secondary margin window; The event detection module is used to calculate the average value of the electrical quantities sampled by the main edge window, the secondary edge window, the main edge window, and the secondary edge window at the left and right ends of the sliding window after each sampling. If the absolute value of the difference between the average electrical quantities corresponding to the main edge window and the secondary edge window at both ends of the sliding window is greater than the preset electrical quantity threshold, and the time interval between the current sampling time and the previous load switch event is greater than the preset time threshold, it is determined that a load switch event has occurred and the corresponding load waveform data is extracted. The power fingerprint recognition module is used to extract corresponding power fingerprint features based on the load waveform data corresponding to each load switching event, input the power fingerprint features into the trained power fingerprint recognition model, and output the load type recognition result.

11. An event-detection-based power fingerprint recognition device, characterized in that, include: A memory for storing instructions; wherein the instructions are used to implement the event detection-based power fingerprinting method as described in any one of claims 1-9; A processor for executing instructions in the memory.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the event-detection-based power fingerprinting method as described in any one of claims 1-9.

Images