A non-intrusive load monitoring data generation method, system, device and medium

Abstract

The application discloses a non-intrusive load monitoring data generation method, system, device and medium, comprising: extracting voltage and current data from a preset database according to the voltage and current waveform state of an electrical appliance to obtain sampling data; determining the cycle length through the zero-crossing point of the voltage signal, and presetting the cycle point number, performing nearest neighbor interpolation on the sampling data to obtain sample data with a fixed sampling point number; inputting the sample data into a time series generation adversarial network model to generate one-dimensional time series of different waveform states of the electrical appliance, thereby obtaining a data set; randomly selecting one-dimensional time series corresponding to the waveform state of a preset specific electrical appliance from the data set, and cutting and splicing the length of the selected one-dimensional time series according to the zero-crossing point and the preset cycle point number to generate waveform data. Thus, the problem that the data generated by the prior art has a large difference from the real data is solved.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a non-intrusive load monitoring data generation method, system, device and medium. Background Technology

[0002] In the continuous evolution and deepening of non-invasive load identification technology, researchers have consistently strived to build more accurate and efficient identification algorithms. Existing technological experience shows that the core of evaluating the advancement of such technologies lies in their ability to demonstrate universality and robustness on large and diverse datasets. This places new demands on the scale and scenario representativeness of test datasets. In practical applications, constructing and maintaining a large and diverse database undoubtedly poses a significant challenge. To reduce the burden of database construction while adapting to the needs of rapid technological iteration, open-source data sharing platforms and synthetic data technologies have gradually become valuable resources for the research community.

[0003] Currently, non-invasive synthetic data techniques mostly employ neural network-based methods, including but not limited to advanced artificial intelligence techniques such as autoregressive models, variational autoencoders, and generative adversarial networks (GANs). These methods can learn the latent distribution of the original data and generate new samples accordingly, but they still differ somewhat from real data. Furthermore, data synthesis techniques are widely used to generate non-invasive decomposition / recognition training data, aiming to alleviate the insufficiency of model training data and enhance its generalization performance; however, the test sets used to evaluate the model's generalization ability are still limited to finite real-world datasets. Summary of the Invention

[0004] This application provides a non-intrusive load monitoring data generation method, system, device, and medium to address the problem that the data generated by existing technologies differs significantly from the actual data.

[0005] In view of this, the first aspect of this application provides a non-intrusive load monitoring data generation method, the method comprising:

[0006] Based on the waveform states of the voltage and current of the electrical appliance, voltage data and current data are extracted from a preset database to obtain sampled data, wherein the waveform states include steady state and non-steady state;

[0007] Based on the theory of frequency invariance transformation of periodic signals, the period length is determined by the zero-crossing point of the voltage signal, and the number of period points is preset. The sampled data is then subjected to nearest neighbor interpolation to obtain sample data with a fixed number of sampling points.

[0008] The sample data is input into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of electrical appliances, thereby obtaining the dataset of the one-dimensional time series.

[0009] A one-dimensional time series corresponding to the waveform state of a specific electrical appliance is randomly selected from the dataset. Based on the zero-crossing point and the preset number of period points, the length of the selected one-dimensional time series is trimmed and spliced ​​to generate waveform data, wherein the phase at the splicing point is consistent.

[0010] Optionally, the step of randomly selecting a one-dimensional time series corresponding to the waveform state of a preset specific electrical appliance from the dataset, and trimming and splicing the length of the selected one-dimensional time series according to the zero-crossing point and the preset number of period points to generate waveform data, wherein the phase at the splicing point is consistent, further includes:

[0011] The matching degree between steady-state and non-steady-state waveform scenarios is evaluated.

[0012] Optionally, the step of evaluating the consistency between the waveform data and the real data extracted from the preset database, and evaluating the load scenario data quality of the waveform data, thereby completing the evaluation of the matching degree between steady-state and non-steady-state waveform scenarios, specifically includes:

[0013] The consistency between the waveform data and the real data extracted from the preset database is evaluated using the coefficient of determination, root mean square error, regression curve quality standard and angle cosine coefficient as evaluation indicators.

[0014] The load scenario data quality of the waveform data is evaluated using the ratio of total power consumption, peak signal-to-noise ratio, and structural similarity as evaluation indicators.

[0015] Optionally, the step of determining the period length based on the frequency-invariant transformation theory of periodic signals, establishing the zero-crossing point of the voltage signal, and pre-setting the number of period points, and performing nearest-neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points, specifically includes:

[0016] Collect voltage data and analyze the relative positional relationship between the voltage sampling points and the zero point. When the adjacent first voltage sampling point and second voltage sampling point are located on both sides of the zero point, the first voltage sampling point and the second voltage sampling point are determined to be zero-crossing points.

[0017] The number of sampling points between the first voltage sampling point and the second voltage sampling point is used as the period length. The current data is separated by voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between current data and voltage data.

[0018] When the period length of each integer cycle is different, the same number of sampling points is selected for each integer cycle, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points, thereby obtaining sample data.

[0019] A second aspect of this application provides a non-intrusive load monitoring data generation system, the system comprising:

[0020] The sampling unit is used to extract voltage data and current data from a preset database based on the waveform state of the voltage and current of the electrical appliance to obtain sampled data, wherein the waveform state includes steady state and non-steady state;

[0021] The interpolation unit is used to determine the period length by the zero-crossing point of the voltage signal based on the frequency invariant transformation theory of periodic signals and to perform nearest neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points by preset period points.

[0022] The first generation unit is used to input the sample data into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of electrical appliances, thereby obtaining the dataset of the one-dimensional time series.

[0023] The second generation unit is used to randomly select a one-dimensional time series corresponding to the waveform state of a preset specific electrical appliance from the dataset, and to trim and splice the length of the selected one-dimensional time series according to the zero crossing point and the preset number of period points to generate waveform data, wherein the phase at the splicing point is consistent.

[0024] Optionally, it also includes an evaluation unit for evaluating the matching degree between steady-state and non-steady-state waveform scenarios.

[0025] Optionally, the evaluation unit is specifically used for:

[0026] The consistency between the waveform data and the real data extracted from the preset database is evaluated using the coefficient of determination, root mean square error, regression curve quality standard, and angle cosine coefficient as evaluation indicators. The load scenario data quality of the waveform data is evaluated using the ratio of total power consumption of the entire waveform scenario, peak signal-to-noise ratio, and structural similarity as evaluation indicators, thereby completing the evaluation of the matching degree between steady-state and non-steady-state waveform scenarios.

[0027] Optionally, the interpolation unit is specifically used for:

[0028] Collect voltage data and analyze the relative positional relationship between the voltage sampling points and the zero point. When the adjacent first voltage sampling point and second voltage sampling point are located on both sides of the zero point, the first voltage sampling point and the second voltage sampling point are determined to be zero-crossing points.

[0029] The number of sampling points between the first voltage sampling point and the second voltage sampling point is used as the period length. The current data is separated by voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between current data and voltage data.

[0030] When the period length of each integer cycle is different, the same number of sampling points is selected for each integer cycle, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points, thereby obtaining sample data.

[0031] A third aspect of this application provides a non-intrusive load monitoring data generation device, the device comprising a processor and a memory:

[0032] The memory is used to store program code and transmit the program code to the processor;

[0033] The processor is configured to execute the steps of the non-intrusive load monitoring data generation method as described in the first aspect above, according to the instructions in the program code.

[0034] A fourth aspect of this application provides a computer-readable storage medium for storing program code for executing the non-intrusive load monitoring data generation method described in the first aspect above.

[0035] As can be seen from the above technical solutions, this application has the following advantages:

[0036] This application provides a non-intrusive load monitoring data generation method, comprising: extracting voltage and current data from a preset database based on the waveform states of the voltage and current of electrical appliances to obtain sampled data, wherein the waveform states include steady-state and non-steady-state; based on the frequency-invariant transformation theory of periodic signals, determining the period length through the zero-crossing point of the voltage signal and preset the number of period points, performing nearest-neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points; inputting the sample data into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of the electrical appliances, thereby obtaining a one-dimensional time series dataset; randomly selecting a one-dimensional time series corresponding to the waveform state of a preset specific electrical appliance from the dataset, and trimming and splicing the length of the selected one-dimensional time series according to the zero-crossing point and the preset number of period points to generate waveform data. Furthermore, the matching degree between steady-state and non-steady-state waveform scenarios is evaluated.

[0037] Compared with existing technologies, this application generates non-intrusive workload data based on a limited real dataset and further verifies the consistency between the generated data and the actual situation. The generated data shows good similarity and consistency with the real data. This solves the problem of significant discrepancies between the data generated by existing technologies and real data. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a non-intrusive load monitoring data generation method provided in the embodiments of this application;

[0039] Figure 2 This is a model structure diagram of the TimeGAN architecture provided in the embodiments of this application;

[0040] Figure 3 This is a schematic diagram of a non-intrusive load monitoring data generation system provided in the embodiments of this application. Detailed Implementation

[0041] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0042] Please see Figure 1 The present application provides a non-intrusive load monitoring data generation method, comprising:

[0043] Step 101: Based on the waveform state of the voltage and current of the electrical appliance, extract voltage data and current data from the preset database to obtain sampled data. The waveform state includes steady state and non-steady state.

[0044] It should be noted that, firstly, the state of the voltage and current waveforms of the electrical appliance is determined to be steady-state or non-steady-state. Then, based on the state of the waveforms, voltage and current data of a specific length are extracted from the historical waveform data of the electrical appliance in the preset database to obtain the sampled data.

[0045] Step 102: Based on the theory of frequency invariance transformation of periodic signals, the period length is determined by the zero-crossing point of the voltage signal, and the number of period points is preset. The sampled data is then subjected to nearest neighbor interpolation to obtain sample data with a fixed number of sampling points.

[0046] In one embodiment, step 102 specifically includes:

[0047] Collect voltage data and analyze the relative positional relationship between the voltage sampling points and the zero point. When the adjacent first voltage sampling point and second voltage sampling point are located on both sides of the zero point, determine that the first voltage sampling point and the second voltage sampling point are zero-crossing points.

[0048] The number of sampling points between the first voltage sampling point and the second voltage sampling point is used as the period length. The current data is separated by voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between current data and voltage data.

[0049] When the period length of each integer cycle is different, the same number of sampling points is selected for each integer cycle, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points, thereby obtaining sample data.

[0050] It should be noted that the Frequency Invariant Transform of Periodic Signals (FIT-PS) theory is a signal decomposition method. Its key characteristic is that the decomposition process does not depend on the dominant frequency of the signal, effectively constructing a signal representation without periodic oscillations. This method establishes the positions of subsequent sampling points within each period of voltage data based on the starting point of each period, thereby achieving the segmentation of current data across different periods. Specifically:

[0051] (1) Collect voltage data, and then analyze the relative positional relationship between the voltage sampling point and the zero point. When adjacent voltage sampling points are located on both sides of the zero point, identify these two zero points as zero-crossing points;

[0052] (2) Define the period length as the number of sampling points between two adjacent zero crossings. Separate the current data by voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between current and voltage.

[0053] (3) When the length of each integer cycle is different, it is necessary to select the same number of sampling points for each integer cycle. Usually, the mode of the number of integer cycles is taken, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points. This avoids the error caused by outward interpolation and ensures that the voltage and current data under the same waveform have the same number of sampling points in each cycle.

[0054] Step 103: Input the sample data into the time series to generate an adversarial network model, generate a one-dimensional time series of different waveform states of the electrical appliances, and thus obtain a one-dimensional time series dataset.

[0055] It should be noted that the Time Series Generative Adversarial Network (TimeGAN) model is a GAN-based architecture capable of efficiently generating real-world time series data across multiple domains. This architecture consists of four main networks: an embedding part, a recovery part, a sequence generator, and a sequence discriminator. Its model structure is as follows: Figure 2 As shown.

[0056] The frame's input consists of two elements: s represents the static feature vector at the encoder input, and x... t This represents the temporal feature vector. Embedding and retrieval functions provide a mapping between the feature space (static and temporal) and the latent space, enabling adversarial networks to learn the latent temporal dynamics of data through low-dimensional representations. A GRU neural network is chosen for feature mapping and retrieval, with the loss function being:

[0057]

[0058] in, and These are the data for reconstructing the network. The sequence generator uses tuples of static and temporal random feature vectors extracted from known distributions: real and synthetic latent codes. The embedding function is used; the sequence discriminator is implemented using a BiGRU neural network, which receives tuples of real and synthetic latent codes and classifies them as real or synthetic, i.e., real or fake (0 / 1); the generator accepts two types of input during training: and and The gradient is calculated based on the unsupervised loss, and the probability of providing a correct classification is maximized (for the discriminator) or minimized (for the generator). The loss function is:

[0059]

[0060] The binary adversarial feedback from the discriminator alone may not be sufficient to motivate the generator to capture the step conditional distribution in the data; therefore, an additional loss is introduced to further constrain the learning. Since the network is trained in a closed-loop mode, where the generator receives actual data... The embedding sequence generates the next latent vector. ,in Time-featured loop generator Given a random vector in a known distribution vector space, following a stochastic process, the maximum likelihood method is then applied to capture the actual next potential vector. With the next step latent vector of synthesis The difference between them, the expression for the supervision loss is as follows:

[0061]

[0062] By linearly combining the three errors with different weighting coefficients and iteratively finding the comprehensive minimum loss, the expression is as follows:

[0063]

[0064] This combination allows TimeGAN to train encoding (feature vectors), generation (latent representations), and iteration (across time) simultaneously. Furthermore, because adversarial learning occurs in a low-dimensional latent space and the supervised loss also constrains the stepwise dynamics of the generator, the training difficulty of TimeGAN is not increased.

[0065] Step 104: Randomly select a one-dimensional time series from the dataset that corresponds to the waveform state of a specific electrical appliance, and trim and splice the length of the selected one-dimensional time series according to the zero-crossing point and the preset number of period points to generate waveform data, wherein the phase at the splicing point is consistent.

[0066] It should be noted that a one-dimensional time series corresponding to the state of a specific electrical appliance is randomly selected from the dataset in step 103. The length of these one-dimensional time series is then trimmed and spliced ​​based on the voltage zero-crossing point and a pre-set number of period points to ensure phase consistency at the splicing points, thereby generating waveform data. The specific electrical appliance can be a typical appliance, and those skilled in the art can select it according to actual needs; no limitation is made here.

[0067] Furthermore, in one embodiment, step 104 is followed by: evaluating the matching degree between steady-state and non-steady-state waveform scenarios.

[0068] Specifically:

[0069] The consistency between waveform data and real data extracted from a pre-set database is evaluated using the coefficient of determination, root mean square error, regression curve quality standard, and angle cosine coefficient as evaluation indicators.

[0070] The load scenario data quality of the waveform data is evaluated using the ratio of total power consumption, peak signal-to-noise ratio, and structural similarity as evaluation indicators.

[0071] It should be noted that the matching degree between waveform scenarios is mainly determined from two aspects: first, judging the consistency between the waveform synthesized by the TimeGAN model and the waveform of the real sample set; and second, evaluating the quality of the generated load scenario data.

[0072] Among them, waveform consistency is evaluated using four indicators: coefficient of determination, root mean square error, regression curve quality standard, and angle cosine coefficient; load scenario data quality is evaluated using three indicators: the ratio of total power consumption in the entire waveform scenario, peak signal-to-noise ratio, and structural similarity index.

[0073] Waveform Consistency

[0074] (1) The coefficient of determination, also known as the goodness of fit, is commonly used as an indicator to evaluate the fit of linear / nonlinear regression. Its formula is shown below:

[0075]

[0076] In the formula: For the sample true value, For composite values, This represents the sample size, and the meaning of the formula parameters in the following text is the same. The closer a value is to 1, the higher the correlation between the actual value and the composite value caused by the independent variable, and it can reflect the curve fitting effect to a certain extent. This reflects the proportion of the composite value explained by the true value through the regression relationship, i.e. This indicates that the regression relationship can explain 80% of the composite value.

[0077] (2) The most commonly used indicators for measuring the absolute error between data are the root mean square error (RMSE) and the symmetric mean absolute percentage error (SMAPE), and their expressions are as follows:

[0078]

[0079]

[0080] RMSE characterizes the model's ability to control absolute error and is more sensitive to outlier data; a smaller RMSE is better. SMAPE evaluates the percentage error of the prediction results, is more stable to outliers, and can be used for data of different orders of magnitude.

[0081] (3) If y and Considering them as vectors in N-dimensional space, the smaller the angle θ between them, the better the effect is considered. It is defined as the angle cosine coefficient (FR), and the formula is as follows:

[0082]

[0083] Data quality in load scenarios

[0084] (1) Different loads have different components and functions, and therefore exhibit different operating characteristics and load features during operation. Load characteristics are the basis for determining load categories and a key factor in achieving non-intrusive power load monitoring. Considering that the most commonly used load characteristic in non-intrusive power load monitoring research is the power characteristic, the ratio of active power accumulated during operation to electrical energy (E_Ratio) is used as the evaluation index for the generated data:

[0085]

[0086] (2) Peak signal-to-noise ratio (PSNR) is often used to evaluate the similarity between generated data and real data. It is generally used to measure the quality reference value between the maximum signal and background noise. Its formula is as follows:

[0087]

[0088] A higher PSNR value indicates that the generated data is closer to the real data. When the PSNR value is below 20, the generated data quality is considered unacceptable; between 20 and 30, the generated data quality is considered poor; between 30 and 40, the generated data quality is good; and above 40, the generated data quality is excellent.

[0089] (3) Structural similarity (SSIM) measures the similarity between generated data and real data in terms of structure and contrast. The SSIM value ranges from [0,1], with a larger value indicating greater similarity. From the perspective of image composition, the SSIM index defines structural information as an attribute that is independent of brightness and contrast, reflecting the structure of objects in the scene, and models distortion as a combination of three different factors: brightness, contrast, and structure.

[0090]

[0091] Among them, the mean For the estimation of brightness, the standard deviation For the estimation of contrast, covariance As a measure of structural similarity, and It is a constant used to maintain stability and can be set as needed.

[0092] The above describes a non-intrusive load monitoring data generation method provided in the embodiments of this application. The following describes a non-intrusive load monitoring data generation system provided in the embodiments of this application.

[0093] Please see Figure 3 The present application provides a non-intrusive load monitoring data generation system, comprising:

[0094] The sampling unit 201 is used to extract voltage data and current data from a preset database based on the waveform state of the voltage and current of the electrical appliance to obtain sampling data, wherein the waveform state includes steady state and non-steady state.

[0095] The interpolation unit 202 is used to determine the period length by the zero-crossing point of the voltage signal based on the frequency invariant transformation theory of periodic signals and to preset the number of period points, and to perform nearest neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points.

[0096] The first generation unit 203 is used to input sample data into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of electrical appliances, thereby obtaining a one-dimensional time series dataset.

[0097] The second generation unit 204 is used to randomly select a one-dimensional time series corresponding to the waveform state of a specific electrical appliance from the dataset, and to trim and splice the length of the selected one-dimensional time series according to the zero crossing point and the preset number of period points to generate waveform data, wherein the phase at the splicing point is consistent.

[0098] Furthermore, this application embodiment also provides a non-intrusive load monitoring data generation device, the device including a processor and a memory:

[0099] The memory is used to store program code and transmit the program code to the processor;

[0100] The processor is used to execute the steps of the non-intrusive load monitoring data generation method as described in the above method embodiments, according to the instructions in the program code.

[0101] Furthermore, this application embodiment also provides a computer-readable storage medium for storing program code for executing the non-intrusive load monitoring data generation method described in the above method embodiments.

[0102] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0103] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0104] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0105] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0106] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0107] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0108] If the integrated unit is implemented as a software functional unit 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 this application, 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 this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0109] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application.

Claims

1. A non-intrusive method for generating load monitoring data, characterized in that, include: Based on the waveform states of the voltage and current of the electrical appliance, voltage data and current data are extracted from a preset database to obtain sampled data, wherein the waveform states include steady state and non-steady state; Based on the theory of frequency invariance transformation of periodic signals, the period length is determined by the zero-crossing point of the voltage signal, and the number of period points is preset. The sampled data is then subjected to nearest neighbor interpolation to obtain sample data with a fixed number of sampling points. The sample data is input into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of electrical appliances, thereby obtaining the dataset of the one-dimensional time series. A one-dimensional time series corresponding to the waveform state of a specific electrical appliance is randomly selected from the dataset. Based on the zero-crossing point and the preset number of period points, the length of the selected one-dimensional time series is trimmed and spliced ​​to generate waveform data, wherein the phase at the splicing point is consistent. The matching degree between steady-state and non-steady-state waveform scenarios is evaluated, including: using the coefficient of determination, root mean square error, regression curve quality standard, and angle cosine coefficient as evaluation indicators to evaluate the consistency between the waveform data and the real data extracted from the preset database; and using the ratio of total power consumption of the entire waveform scenario, peak signal-to-noise ratio, and structural similarity as evaluation indicators to evaluate the load scenario data quality of the waveform data. Specifically, the method based on the frequency-invariant transformation theory of periodic signals, which establishes the period length through the zero-crossing point of the voltage signal and presets the number of period points, and performs nearest-neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points, includes: Collect voltage data and analyze the relative positional relationship between the voltage sampling points and the zero point. When the adjacent first voltage sampling point and second voltage sampling point are located on both sides of the zero point, the first voltage sampling point and the second voltage sampling point are determined to be zero-crossing points. The number of sampling points between the first voltage sampling point and the second voltage sampling point is used as the period length. The current data is separated by the voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between the current data and the voltage data. When the period length of each integer cycle is different, the same number of sampling points is selected for each integer cycle, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points, thereby obtaining sample data.

2. A non-intrusive load monitoring data generation system, characterized in that, include: The sampling unit is used to extract voltage data and current data from a preset database based on the waveform state of the voltage and current of the electrical appliance to obtain sampled data, wherein the waveform state includes steady state and non-steady state; The interpolation unit is used to determine the period length by the zero-crossing point of the voltage signal based on the frequency invariant transformation theory of periodic signals and to perform nearest neighbor interpolation on the sampled data to obtain sample data with a fixed number of sampling points by preset period points. The first generation unit is used to input the sample data into a time series generation adversarial network model to generate a one-dimensional time series of different waveform states of electrical appliances, thereby obtaining the dataset of the one-dimensional time series. The second generation unit is used to randomly select a one-dimensional time series corresponding to the waveform state of a preset specific electrical appliance from the dataset, and to trim and splice the length of the selected one-dimensional time series according to the zero crossing point and the preset number of period points to generate waveform data, wherein the phase at the splicing point is consistent. The evaluation unit is used to evaluate the matching degree between steady-state and non-steady-state waveform scenarios; The evaluation unit is specifically used for: The consistency between the waveform data and the real data extracted from the preset database is evaluated using the coefficient of determination, root mean square error, regression curve quality standard and angle cosine coefficient as evaluation indicators. The load scenario data quality of the waveform data is evaluated using the ratio of total power consumption of the entire waveform scenario, peak signal-to-noise ratio and structural similarity as evaluation indicators, thereby completing the evaluation of the matching degree between steady-state waveform and non-steady-state waveform scenarios. The interpolation unit is specifically used for: Collect voltage data and analyze the relative positional relationship between the voltage sampling points and the zero point. When the adjacent first voltage sampling point and second voltage sampling point are located on both sides of the zero point, the first voltage sampling point and the second voltage sampling point are determined to be zero-crossing points. The number of sampling points between the first voltage sampling point and the second voltage sampling point is used as the period length. The current data is separated by the voltage data and the number of sampling points, so that each period is a synchronous signal, thereby preserving the phase angle information between the current data and the voltage data. When the period length of each integer cycle is different, the same number of sampling points is selected for each integer cycle, and the voltage and current data are interpolated according to the fixed number of sampling points for each cycle to obtain new sampling points, thereby obtaining sample data.

3. A non-invasive load monitoring data generation device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the non-intrusive load monitoring data generation method of claim 1 according to the instructions in the program code.

4. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the non-intrusive load monitoring data generation method of claim 1.

Images