Non-intrusive load disaggregation device migration performance evaluation method and system

Abstract

The application discloses a non-intrusive load identification device migration performance evaluation method and system, the method comprises the following steps: obtaining a source domain data set of power user electricity consumption data, and training a load identification model of the non-intrusive load identification device to obtain model parameters; obtaining a target domain data set of power user electricity consumption data; applying the source domain model parameters and the load identification model to the target domain data set to construct a weighted virtual load label of a target domain sample; calculating an empirical conditional distribution of a real load label of the target domain with respect to the weighted virtual load label; and calculating a migration performance evaluation index of the non-intrusive load identification device. The application aims to comprehensively evaluate the performance of the non-intrusive load identification device in a small data environment, and provide more reliable guidance and selection basis for the practical application of the non-intrusive load identification device.

Description

Technical Field

[0001] This invention belongs to the field of electrical load identification technology, specifically relating to a non-intrusive load identification device migration performance evaluation method and system. Background Technology

[0002] Against the backdrop of smart grid construction and the promotion of grid transparency, acquiring granular operational status information of various electrical devices and achieving bidirectional information flow interaction between the grid side and the user side has become a crucial link. In this context, load monitoring has become an important means of acquiring operational status information of electrical devices, which is of great significance for improving enterprise service quality, increasing user satisfaction, and achieving energy conservation and emission reduction.

[0003] Load monitoring technology has evolved to primarily fall into two categories: intrusive and non-intrusive. Traditional intrusive load monitoring (ILM) methods focus on hardware solutions, acquiring information by installing sensors on the monitored equipment. While this method provides accurate data, hardware, maintenance, and time / labor costs increase with the variety and number of loads, making it unsuitable for widespread adoption. Conversely, non-intrusive load monitoring (NILM) emphasizes software algorithms, primarily based on power data such as current, voltage, and power at the user's power input bus. NILM offers advantages such as high user acceptance and ease of maintenance, and while maintaining a certain level of accuracy, it boasts a significant cost advantage compared to intrusive methods.

[0004] Meanwhile, the rise of artificial intelligence has driven the interdisciplinary application of machine learning, providing a more convenient and efficient solution for non-intrusive load monitoring (NILM) technology. Deep learning is one of the most widely used machine learning methods. The similarity of deep learning structures determines its strong transferability; through transfer learning on different databases and modifications to the network structure, it can effectively solve problems such as limited samples, high time consumption, and high hardware costs in practical engineering. Therefore, introducing deep learning technology into the field of load monitoring is of great significance for optimizing existing load identification and classification algorithms and expanding the application scope of NILM technology.

[0005] With the application of deep learning extending to real-world load identification problems where training data is insufficient, transfer learning, as a means to improve performance in such small-data environments, has recently received widespread attention. In this context, the transferability of non-intrusive load identification modules is particularly important, as good transferability estimation helps achieve more accurate load identification, thereby enabling the selection of highly transferable devices. However, traditional transferability evaluation methods suffer from a series of problems, including computational difficulty, symmetry issues, overly stringent application conditions, difficulty in interpreting metrics, and inability to evaluate meta-transfer learning. To address these issues, the field of load identification urgently needs a comprehensive evaluation technique, with particular focus on the transferability of non-intrusive load identification modules. In non-intrusive load identification, load refers to the system's load condition under different operating conditions, and load identification achieves an accurate grasp of the system state by learning and identifying these load conditions. To better evaluate the performance of load identification modules in small-data environments, introducing load transfer learning from the load identification problem becomes an inevitable choice. Load transfer learning in load identification refers to transferring knowledge learned in one domain (source domain) to another domain (target domain) to improve the performance of the target domain.

[0006] In summary, the introduction of a comprehensive evaluation technology for the migration capability of non-intrusive load identification equipment is clearly necessary. However, how to overcome the problems of traditional evaluation methods when evaluating the migration capability of non-intrusive load identification equipment remains a technical problem that urgently needs to be solved. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method and system for evaluating the migration performance of non-intrusive load identification equipment, which aims to comprehensively evaluate the performance of non-intrusive load identification equipment in a small data environment and provide more reliable guidance and selection basis for the practical application of non-intrusive load identification equipment.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A method for evaluating the migration performance of a non-intrusive load identification device includes the following steps:

[0010] S1, Obtain the source domain dataset of power user electricity consumption data in the source domain. ,in and These are the load metric set and load label set of the source domain samples, respectively. Using the source domain dataset... Train the load identification model of the non-intrusive load identification device to obtain model parameters ;

[0011] S2, Obtain the target domain dataset of electricity consumption data for target domain users. ,in and These are the load index set and load label set for the target domain samples, respectively; the source domain model parameters... The load identification model is applied to the target domain dataset. Load index for each target domain sample Obtain the target domain samples in the source domain load label set. Virtual label weights and combined with the target domain load tag set Constructing weighted virtual load labels for target domain samples ;

[0012] S3, based on the weighted virtual load labels of each target domain sample. and its target domain load label set In the real load labels, calculate the empirical conditional distribution of the real load labels in the target domain with respect to the weighted virtual load labels;

[0013] S4. The migration performance evaluation index of the non-intrusive load identification device is calculated based on the empirical condition distribution of the actual load labels in the target domain with respect to the weighted virtual load labels.

[0014] Optionally, the target domain samples obtained in step S2 are in the source domain load label set. Virtual label weights The function expression is:

[0015] ;

[0016] in, ~ The respective load label sets in the source domain The above is section 1~ The probability of each load label For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label.

[0017] Optionally, in step S2, the target domain load label set is combined. Constructing weighted virtual load labels for target domain samples The function expression is:

[0018] ;

[0019] in, For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label. For the j-th source domain sample in the source domain load label set The actual load label in the middle.

[0020] Optionally, the empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels in step S3 includes:

[0021] S3.1, Weighted virtual load labels for all target domain samples and its target domain load label set The corresponding actual load label The constructed label pairs The joint probability density is estimated based on the following statistical empirical formula:

[0022] ;

[0023] in, For tag pairs The empirical joint probability density function is used to describe the overall statistical relationship between the two. For the target domain dataset The number of samples in the target domain; For bandwidth The kernel function, which is used for smoothing statistics of continuous variables; For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set True load labeling and Let be the independent variable used for integration or calculation in kernel density estimation, and let represent any value points of the weighted virtual load label and the real load label, respectively.

[0024] S3.2, the empirical joint probability density is estimated according to the following formula. Integrating along the dimension yields the empirical marginal distribution of the true load labels in the target domain:

[0025] ;

[0026] in, The empirical marginal distribution of the true load labels for the target domain;

[0027] S3.3, calculate the empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels according to the following formula:

[0028] ;

[0029] in, The empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels. For tag pairs The empirical joint probability density function, The empirical marginal distribution of the true load labels for the target domain.

[0030] Optionally, in step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label deviation migration performance index, and the calculation function expression of the load label deviation is:

[0031] ;

[0032] in, This is a performance indicator for load label deviation migration. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle, for and The difference function between them.

[0033] Optionally, in step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label mean square deviation migration performance index, and the calculation function expression of the mean square deviation migration performance index is:

[0034] ;

[0035] in, The mean square deviation of the load label is a migration performance index. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle.

[0036] Optionally, in step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label information gain transfer performance index, and the calculation function expression of the information gain transfer performance index is as follows:

[0037] ;

[0038] in, The load tag information gain transfer performance index For the target domain dataset The number of samples in the target domain. For the first Empirical conditional distribution of samples in the target domain For the first The empirical marginal distribution of a sample in the target domain For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle.

[0039] The present invention also provides a non-intrusive load identification device migration performance evaluation system, including a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the non-intrusive load identification device migration performance evaluation method.

[0040] The present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the non-intrusive load identification device migration performance evaluation method by a processor.

[0041] The present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the non-intrusive load identification device migration performance evaluation method via a processor.

[0042] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: When evaluating the transferability of a load identification module, it is necessary to consider the problems existing in traditional evaluation methods and introduce load characteristics from the load identification field to more comprehensively evaluate module performance. To overcome the problems of traditional evaluation methods, the non-intrusive load identification equipment transfer performance evaluation method of the present invention introduces load characteristics from the load identification field. By considering factors such as load characteristics, ease of calculation, asymmetry, wide applicability, and interpretability, it more comprehensively evaluates module performance. It possesses the characteristics of ease of calculation, applicability to practical applications, asymmetry, applicability to different operating conditions, and interpretability. Furthermore, it can be applied to the evaluation of meta-transfer learning, and can more comprehensively evaluate the performance of non-intrusive load identification equipment in small data environments, providing more reliable guidance and selection criteria for the practical application of non-intrusive load identification equipment. The method of the present invention has the following advantages: 1) Easy to calculate and applicable to practical applications; 2) Asymmetric to suit actual load conditions; 3) Wide applicability to different operating conditions; 4) Evaluation indicators should be interpretable to facilitate result analysis; 5) Applicable to the evaluation of meta-transfer learning. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0045] like Figure 1 As shown, the non-intrusive load identification device migration performance evaluation method in this embodiment includes the following steps:

[0046] S1, Obtain the source domain dataset of power user electricity consumption data in the source domain. ,in and These are the load metric set and load label set of the source domain samples, respectively. Using the source domain dataset... Train the load identification model of the non-intrusive load identification device to obtain model parameters ;

[0047] S2, Obtain the target domain dataset of electricity consumption data for target domain users. ,in and These are the load index set and load label set for the target domain samples, respectively; the source domain model parameters... The load identification model is applied to the target domain dataset. Load index for each target domain sample Obtain the target domain samples in the source domain load label set. Virtual label weights and combined with the target domain load tag set Constructing weighted virtual load labels for target domain samples ;

[0048] S3, based on the weighted virtual load labels of each target domain sample. and its target domain load label set In the real load labels, calculate the empirical conditional distribution of the real load labels in the target domain with respect to the weighted virtual load labels;

[0049] S4. The migration performance evaluation index of the non-intrusive load identification device is calculated based on the empirical condition distribution of the actual load labels in the target domain with respect to the weighted virtual load labels.

[0050] In step S1 of this embodiment, the source domain dataset of source domain power user electricity consumption data is obtained. At that time, the load index set and load label set of the source domain samples and They can be represented as:

[0051] ;

[0052] ;

[0053] in, ~ They are respectively number 1 to The load index of each source domain sample, specifically the active power of electricity users. ~ They are respectively number 1 to Load labels of each source domain sample The number of source domain samples is used as an optional implementation method in this embodiment. =12000. Using the source domain dataset. Train the load identification model of the non-intrusive load identification device to obtain model parameters It should be noted that the load identification model is the machine learning model built into the non-intrusive load identification device being evaluated. The method in this embodiment is applicable to all machine learning models that satisfy the above input-output mapping relationship, utilizing the source domain dataset. Train the machine learning model of the non-invasive load identification device to obtain model parameters Since this is a known existing method, its implementation details will not be elaborated here.

[0054] In step S2 of this embodiment, the target domain dataset for obtaining electricity consumption data of target domain power users is obtained. At that time, the load index set and load label set of the target domain sample and They can be represented as:

[0055] ;

[0056] ;

[0057] in, ~ They are respectively number 1 to The load index for each target domain sample, specifically the active power of electricity users. ~ They are respectively number 1 to Load labels for each target domain sample. As an optional implementation method, the number of samples in the target domain is specified in this embodiment. =2000. and Let represent the i-th target domain sample, where , . Model parameters Applied to the target dataset Each input The source domain load tag set can then be obtained. The predicted weight distribution, specifically the target domain samples obtained in step S2 within the source domain load label set. Virtual label weights The function expression is:

[0058] ;

[0059] in, ~ The respective load label sets in the source domain The above is section 1~ The probability of each load label For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label. In this embodiment... =12000, then:

[0060] .

[0061] The target domain samples are placed in the source domain load label set. Virtual label weights Source domain load tag set By combining these, weighted virtual load labels for the target domain samples can be constructed. In step S2 of this embodiment, the target domain load label set is combined. Constructing weighted virtual load labels for target domain samples The function expression is:

[0062] ;

[0063] in, For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label. For the j-th source domain sample in the source domain load label set The actual load label in the middle.

[0064] The empirical conditional distribution of the target label is calculated to characterize the mapping relationship between the source domain label and the target domain label. In this embodiment, step S3, constructing the empirical conditional distribution of the target domain's true load label with respect to the weighted virtual load label, includes:

[0065] S3.1, Weighted virtual load labels for all target domain samples and its target domain load label set The corresponding actual load label The constructed label pairs The joint probability density is estimated based on the following statistical empirical formula:

[0066] ;

[0067] in, For tag pairs The empirical joint probability density function is used to describe the overall statistical relationship between the two. For the target domain dataset The number of samples in the target domain; For bandwidth The kernel function, which is used for smoothing statistics of continuous variables; For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set True load labeling and Let be the independent variables used for integration or calculation in kernel density estimation, and let represent arbitrary values ​​of the weighted virtual load label and the real load label, respectively. These are related to the subscripted . , The relationship is as follows: the subscript indicates the sample observation value, and the unsubscript indicates the corresponding calculated variable or evaluation point in the probability density function.

[0068] In this embodiment =2000, then:

[0069] ;

[0070] S3.2, the empirical joint probability density is estimated according to the following formula. Integrating along the dimension yields the empirical marginal distribution of the true load labels in the target domain:

[0071] ;

[0072] in, The empirical marginal distribution of the true load labels for the target domain;

[0073] S3.3, calculate the empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels according to the following formula:

[0074] ;

[0075] in, The empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels. For tag pairs The empirical joint probability density function, This is the empirical marginal distribution of the true load labels for the target domain. This conditional distribution reflects the source domain through the model parameters. The statistical relationship between the mapped virtual label values ​​and the target real labels provides a basis for subsequent migration performance evaluation.

[0076] The Transferability Evaluation Index (TEI) is used to measure the model parameters of the source domain model. Transferability performance on target domain data.

[0077] As an optional implementation, in step S4 of this embodiment, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical condition distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label deviation migration performance index, and the calculation function expression of the load label deviation is:

[0078] ;

[0079] in, This is a performance indicator for load label deviation migration. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle, for and The difference function between them can be selected in different forms according to requirements, such as Euclidean distance function, Manhattan distance function, squared error function, mean absolute error function, and relative error function. The performance index of load label deviation migration is used. This allows for the comparison of virtual label values ​​with target real labels to obtain the deviation between the two, thus comprehensively reflecting the performance of this deviation across the entire target domain. In this embodiment... =2000, then:

[0080] .

[0081] As an optional implementation, in step S4 of this embodiment, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical condition distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the Mean Square Deviation Transferability Index (MSDI), and the calculation function expression of the MSDI is as follows:

[0082] ;

[0083] in, The mean square deviation of the load label is a migration performance index. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The model uses the true load labels. The mean squared deviation (MSD) transfer performance index is negative, and the smaller its absolute value (i.e., the closer the MSD is to 0), the better the model's transfer performance in the target domain. In this embodiment... =2000, then:

[0084] .

[0085] As an optional implementation, in step S4 of this embodiment, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical condition distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label information gain transferability index. The calculation function expression of the information gain transferability index is as follows:

[0086] ;

[0087] in, The load tag information gain transfer performance index For the target domain dataset The number of samples in the target domain. For the first Empirical conditional distribution of samples in the target domain For the first The empirical marginal distribution of a sample in the target domain For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The virtual label is the actual payload label. The information gain transfer performance index reflects the information contribution of the virtual label to the target label; the higher the value of the information gain transfer performance index, the stronger the transferability of the source domain model to the target domain. In this embodiment... =2000, then:

[0088] .

[0089] To verify the non-intrusive load identification device migration performance evaluation method of this embodiment, the experimental environment used in this embodiment is as follows: 64-bit Windows 11 operating system, Intel(R) Core(TM) i9-12900H CPU, NVIDIA GeForce RTX 3060 graphics card, 40GB of memory, and Matlab software version R2021b. Six configurations were used in the experiment in this embodiment, as follows:

[0090] Configuration A: (learning rate l) r =1e-3, regularization coefficient L2=1e-4).

[0091] Configuration B: (Learning rate l) r =1e-2, regularization coefficient L2=1e-4).

[0092] Configuration C: (learning rate l) r=5e-2, regularization coefficient L2=5e-5).

[0093] Configuration D: (Learning rate l) r =5e-3, regularization coefficient L1=1e-5).

[0094] Configuration E: (Learning rate l) r =2e-2, random dropout ratio = 0.3).

[0095] Configuration F: (learning rate l) r =3e-3, weight decay coefficient wd=1e-4).

[0096] Wherein, the learning rate l r Regularization coefficients L1 / L2, dropout ratio, and weight decay coefficient wd are all parameters used during training of the load identification model for non-intrusive load identification equipment. The learning rate is l. r The network parameter update step size of the load identification model is controlled by the regularization coefficients L1 / L2 to suppress overfitting. The random dropout ratio is used to control the random dropout rate during training. The weight decay coefficient wd is used to limit parameter amplitudes during optimization and improve model generalization ability. The experiments in this embodiment used target domain MAE (mean absolute error, the smaller the better) and target domain R... 2 (Reflects the degree of fit of the model to the actual load changes in the target domain, with a value range of [0,1]. The larger the value, the better the fit.) (The load label mean square deviation migration performance index measures the mean square deviation between the target domain weighted virtual load labels and the actual load labels. A negative value, closer to 0, indicates better migration performance; ↑→0 indicates performance improvement.) (The load tag information gain transfer performance index, the larger the better) is used as the evaluation index, and the results are shown in Table 1.

[0097] Table 1: Experimental results under different configurations

[0098]

[0099] See Table 1 for the experimental results under configurations A to F: (1) (Mean Square Deviation of Load Labels Migration Performance Index) and (Load tag information gain transfer performance index) Two indicators, along with target domain MAE and target domain R 2 The ranking predictions were highly consistent (Spearman rank correlation coefficients reached 0.95–1.00, with a significance level of p<0.01); (2) with (Mean Square Deviation of Load Labels Migration Performance Index) and (Load label information gain transfer performance index) The regret value of the Top-1 model selection is close to 0, which is significantly better than random selection; (3) In addition, in order to verify the robustness of the proposed transfer performance index under different experimental conditions, this embodiment changes the number of target domain samples in an additional stability verification experiment. (e.g., taking 1000, 1500, 2000, etc.) and core bandwidth (For example, take 0.05, 0.1, 0.2, etc.) to perform repeated tests. The results show that the conclusion remains stable under different target domain sample numbers and kernel bandwidth settings, verifying the robustness and universality of the method in this embodiment; (4) In addition, in the ablation experiments with shuffled labels or weights other than those in Table 1, the correlation degraded to close to 0, verifying that (Mean Square Deviation of Load Labels Migration Performance Index) and The effectiveness and necessity of the (load tag information gain migration performance index) are demonstrated by the above results, proving the validity of the migration performance evaluation method for the non-intrusive load identification equipment in this embodiment. (Mean Square Deviation of Load Labels Migration Performance Index) and (Load tag information gain transfer performance index) can accurately and effectively evaluate the migration performance of equipment and can be used as a basis for model selection and early prediction in actual deployment.

[0100] This embodiment also provides a non-intrusive load identification device migration performance evaluation system, including a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to execute the non-intrusive load identification device migration performance evaluation method. This embodiment also provides a computer-readable storage medium storing a computer program or instructions programmed or configured to execute the non-intrusive load identification device migration performance evaluation method via a processor. This embodiment also provides a computer program product, including a computer program or instructions programmed or configured to execute the non-intrusive load identification device migration performance evaluation method via a processor.

[0101] Those skilled in the art will understand that the technical solutions provided by this invention may take the form of a method, system, or computer program product. Therefore, this invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce an implementation of the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for evaluating the migration performance of a non-intrusive load identification device, characterized in that, Includes the following steps: S1, Obtain the source domain dataset of power user electricity consumption data in the source domain. ,in and These are the load index set and load label set of the source domain samples, respectively. The load index in the load index set represents the active power of electricity users. The source domain dataset is used... Train the load identification model of the non-intrusive load identification device to obtain model parameters ; S2, Obtain the target domain dataset of electricity consumption data for target domain users. ,in and These are the load index set and load label set for the target domain samples, respectively; the source domain model parameters... The load identification model is applied to the target domain dataset. Load index for each target domain sample Obtain the target domain samples in the source domain load label set. Virtual label weights and combined with the target domain load tag set Constructing weighted virtual load labels for target domain samples ; S3, based on the weighted virtual load labels of each target domain sample. and its target domain load label set In the real load labels, calculate the empirical conditional distribution of the real load labels in the target domain with respect to the weighted virtual load labels; S4. The migration performance evaluation index of the non-intrusive load identification device is calculated based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels. The target domain samples obtained in step S2 are in the source domain load label set Virtual label weights The function expression is: ； in, ~ The respective load label sets in the source domain The above is section 1~ The probability of each load label For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label; in step S2, the target domain load label set is combined. Constructing weighted virtual load labels for target domain samples The function expression is: ； in, For source domain dataset The number of source domain samples in the data. For the source domain load tag set The above represents the probability of the j-th load label. For the j-th source domain sample in the source domain load label set The actual load label in the middle.

2. The non-intrusive load identification equipment migration performance evaluation method according to claim 1, characterized in that, Step S3, which constructs the empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels, includes: S3.1, Weighted virtual load labels for all target domain samples and its target domain load label set The corresponding actual load label The constructed label pairs The joint probability density is estimated based on the following statistical empirical formula: ； in, For tag pairs The empirical joint probability density function is used to describe the overall statistical relationship between the two. For the target domain dataset The number of samples in the target domain; For bandwidth The kernel function, which is used for smoothing statistics of continuous variables; For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set True load labeling and Let be the independent variable used for integration or calculation in kernel density estimation, and let represent any value points of the weighted virtual load label and the real load label, respectively. S3.2, the empirical joint probability density is estimated according to the following formula. Integrating along the dimension yields the empirical marginal distribution of the true load labels in the target domain: ； in, The empirical marginal distribution of the true load labels for the target domain; S3.3, calculate the empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels according to the following formula: ； in, The empirical conditional distribution of the target domain's true load labels with respect to the weighted virtual load labels. For tag pairs The empirical joint probability density function, The empirical marginal distribution of the true load labels for the target domain.

3. The non-intrusive load identification equipment migration performance evaluation method according to claim 1, characterized in that, In step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label deviation migration performance index, and the calculation function expression of the load label deviation is: ； in, This is a performance indicator for load label deviation migration. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle, for and The difference function between them.

4. The method for evaluating the migration performance of non-intrusive load identification equipment according to claim 1, characterized in that, In step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label mean square deviation migration performance index. The calculation function expression of the mean square deviation migration performance index is as follows: ； in, The mean square deviation of the load label is a migration performance index. For the target domain dataset The number of samples in the target domain. For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle.

5. The non-intrusive load identification equipment migration performance evaluation method according to claim 3, characterized in that, In step S4, when calculating the migration performance evaluation index of the non-intrusive load identification device based on the empirical conditional distribution of the actual load labels in the target domain with respect to the weighted virtual load labels, the migration performance evaluation index includes the load label information gain transfer performance index, and the calculation function expression of the information gain transfer performance index is as follows: ； in, The load tag information gain transfer performance index For the target domain dataset The number of samples in the target domain. For the first Empirical conditional distribution of samples in the target domain For the first The empirical marginal distribution of a sample in the target domain For the first Weighted virtual load labels for each target domain sample. For the first Each target domain sample in the target domain load label set The actual load label in the middle.

6. A non-intrusive load identification device migration performance evaluation system, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to execute the non-intrusive load identification device migration performance evaluation method according to any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the non-intrusive load identification device migration performance evaluation method according to any one of claims 1 to 5.

8. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the non-intrusive load identification device migration performance evaluation method according to any one of claims 1 to 5.

Images