Electricity stealing detection method and system based on IVMD-WVD and CNN

Abstract

The invention discloses an electricity stealing detection method and system based on IVMD-WVD and CNN. Relates to the technical field of electricity stealing detection. According to the method, feature extraction and generalization processing are carried out on the basis of IVMD-WVD and CNN, so that gradient disappearance and other phenomena during CNN training are effectively avoided or inhibited, and the electricity stealing detection rate is increased; the obtained IVMD-WVD gray scale map can well keep the characteristics of data edge characteristics, instantaneous frequency and the like, and the training speed of a network model is increased; according to the scheme, the improved deep convolutional neural network is trained, and the electricity stealing type is identified and detected; compared with manual experience judgment, the method can effectively improve the recognition speed and recognition rate, and avoids the problems of high subjectivity and insufficient feature extraction capability of manual search.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity theft detection, and in particular to an electricity theft detection method and system based on IVMD-WVD and CNN. Background Art

[0002] With the development of science and technology, the methods of electricity theft are endless, and the detection methods of electricity theft are becoming more and more advanced. At the same time, the popularization of smart meters and electricity information collection systems has accumulated a large amount of metering data for data-driven electricity theft detection technology research. Electricity theft not only seriously affects the operating efficiency of power grid companies, but also has a relatively negative social impact.

[0003] At present, there are two main methods for electricity theft: manual inspection and data-driven. Manual inspection mainly involves manual household screening and data analysis based on the high-loss areas warned by the system, which has the disadvantages of low efficiency, strong subjectivity, and difficulty in detecting electricity theft households in low-loss areas. Data-driven methods based on artificial neural networks, support vector machines and other methods are currently partially applied to power systems, but due to problems such as sample quality, gradient diffusion and overfitting of traditional neural networks, they have disadvantages such as low accuracy and difficulty in feature extraction. The average detection rate of this type of method can reach about 85%. In addition, sample data preprocessing, activation function selection, deep network gradient diffusion problems, overfitting, and model generalization ability in neural networks will affect the final detection results, thereby affecting the judgment of whether the user is an electricity theft household and the type of electricity theft.

[0004] In summary, the problems of low efficiency, strong subjectivity and insufficient feature extraction capability of manual household screening in electricity theft detection need to be solved urgently in order to achieve accurate detection of electricity theft users. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the traditional electricity theft detection method has the problems of low efficiency, strong subjectivity and insufficient feature extraction ability of manual household screening. The purpose of the present invention is to provide an electricity theft detection method and system based on IVMD-WVD and CNN, based on the improvement of IVMD-WVD and CNN, feature extraction and generalization processing are performed, the phenomena such as gradient disappearance during CNN training are effectively avoided or suppressed, and the electricity theft detection rate is improved; the obtained IVMD-WVD grayscale map can well retain the data edge characteristics, instantaneous frequency and other characteristics, and increase the network model training speed.

[0006] The present invention is achieved through the following technical solutions:

[0007] This solution provides a power theft detection method based on IVMD-WVD and CNN, including:

[0008] Collecting data: collecting historical electricity consumption data of users and preprocessing the historical electricity consumption data;

[0009] Build a power theft detection model:

[0010] Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample.

[0011] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0012] Electricity theft detection: Collect the user's real-time electricity usage data, pre-process the real-time electricity usage data, input the pre-processed real-time electricity usage data into the constructed electricity theft detection model, and obtain the user's electricity theft results.

[0013] Working principle of this scheme: The electricity theft detection method and system based on IVMD-WVD and CNN provided by this scheme are based on improved IVMD-WVD and CNN, feature extraction and generalization processing are performed, which effectively avoids or suppresses the phenomenon of gradient disappearance during CNN training, and improves the electricity theft detection rate; the obtained IVMD-WVD grayscale map can well retain the data edge characteristics, instantaneous frequency and other characteristics, and increase the training speed of the network model; this scheme trains the improved deep convolutional neural network to identify and detect the type of electricity theft; compared with manual experience judgment, it can effectively improve the recognition speed and recognition rate, and avoid the problems of strong subjectivity in manual search and insufficient feature extraction ability.

[0014] A further optimization scheme is that the preprocessing method includes: cleaning the user's historical electricity consumption data to remove invalid data.

[0015] A further optimization scheme is that the samples include a normal sample and a plurality of different types of electricity theft samples; the electricity theft samples include: a fixed ratio reduction metering power sample S1 (w t ), random threshold reduction power sample S2(w t ), random time period reduction power sample S3 (w t ), random threshold reduction of power samples S4 (w t ), reverse power consumption time sample S5 (w t ) and average power consumption sample S6(w t ).

[0016] A further optimization scheme is to reduce the metering power sample S1(w t )for:

[0017] S1(w t )=μw t (0.1<μ<0.8);

[0018] Among them, μ represents an interval random constant;

[0019] Random threshold reduction power sample S2(w t )for:

[0020]

[0021] Among them, λ represents a random constant between the minimum power value and the maximum power value;

[0022] Random time period reduction power sample S3(w t )for:

[0023]

[0024] Where t1 and t2 represent any time of the day, and t2 ≥ t1 + 3;

[0025] Random threshold reduces the power sample S4(w t )for:

[0026] S4(w t )=max{w t -δ,0};

[0027] Among them, δ represents a random constant, taking the average or minimum value;

[0028] Reversed power consumption time sample S5(w t )for:

[0029] S5(w t )=w 97-t ;

[0030] Among them, w 97-t Indicates the user's electricity consumption data after time t;

[0031] Average power consumption sample S6 (w t )for:

[0032] S6(w t )=mean(w t );

[0033] Among them, mean(w t ) means taking the average value of the user's electricity consumption.

[0034] A further optimization scheme is to introduce the energy difference method to improve the variational modal decomposition model, and input the samples into the improved variational modal decomposition model to decompose the modal components of each sample; the method includes:

[0035] Configure the initial decomposition level K or , construct a constrained electricity theft sample variational problem model;

[0036] The Lagrange multiplier method and penalty factor α are introduced to optimize the constrained electricity theft sample variational problem model;

[0037] The energy difference method is introduced to improve the optimization process of the constraint electricity theft sample variational problem model and obtain the optimal decomposition layer number K opt ;

[0038] Based on the optimal decomposition level K opt Decompose the modal components of each sample.

[0039] A further optimization scheme is that the constraint electricity theft sample variational problem model includes:

[0040]

[0041] Among them, S(w t ) represents the user's electricity consumption data; ω k represents the center frequency of each mode function; represents the partial derivative with respect to time t; δ(t) represents the impulse function; Indicates the square of the Euclidean distance; S IMFk It represents 10 partial discharge signal modal functions obtained by decomposition, and i is an imaginary unit.

[0042] A further optimization scheme is that the method for solving the IVMD-WVD spectrum of each sample includes:

[0043] The IVMD-WVD spectrum was calculated according to the following formula:

[0044]

[0045] Where IVMD-WVD(t,ω) represents the IVMD-WVD spectrum, K represents S IMFk The total number of S; k represents the kth S IMFk ; S IMFk represents the modal function of electricity consumption samples; α represents the time window; t represents the sample time span; ω represents the frequency of the sample at time t; i represents the imaginary unit.

[0046] A further optimization scheme is that the grayscale processing of each IVMD-WVD spectrum includes the following method:

[0047] Each IVMD-WVD spectrum was normalized according to the following formula:

[0048]

[0049] The normalized IVMD-WVD spectrum is grayed according to the following formula:

[0050]

[0051] Where (a, b) represents the value of IVMD-WVD spectrum at (a, b); IVMD-WVD nor 、IVMD-WVD nor-gray They represent IVMD-WVD normalized spectrum and IVMD-WVD normalized grayscale spectrum respectively; They represent the red, green and blue components of the IVMD-WVD normalized spectrum at (m, n), respectively.

[0052] A further optimization scheme is to train the improved CNN network based on the grayscale processed IVMD-WVD atlas to obtain the electricity theft detection model, including the following method:

[0053] Build the AlexNet network and randomly divide the IVMD-WVD atlas of each sample into training set, validation set, and test set according to the proportion;

[0054] Training the AlexNet network based on the training set and the validation set;

[0055] Test the trained AlexNet network based on the test set.

[0056] This solution also provides an electricity theft detection system based on IVMD-WVD and CNN, which is used to implement the above-mentioned electricity theft detection method based on IVMD-WVD and CNN. The system includes:

[0057] A collection module, used to collect historical electricity consumption data of users and pre-process the historical electricity consumption data;

[0058] Model building module, used to build electricity theft detection model:

[0059] Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample.

[0060] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0061] The detection module is used for electricity theft detection: collecting the user's real-time electricity consumption data, preprocessing the real-time electricity consumption data, inputting the preprocessed real-time electricity consumption data into the constructed electricity theft detection model, and obtaining the user's electricity theft results.

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] The invention provides an electricity theft detection method and system based on IVMD-WVD and CNN; based on improved IVMD-WVD and CNN, feature extraction and generalization processing are performed to effectively avoid or suppress the phenomenon of gradient disappearance during CNN training, thereby improving the electricity theft detection rate; the obtained IVMD-WVD grayscale map can well retain data edge characteristics, instantaneous frequency and other features, thereby increasing the network model training speed; the scheme trains the improved deep convolutional neural network to identify and detect the type of electricity theft; compared with manual experience judgment, the scheme can effectively improve the recognition speed and recognition rate, thereby avoiding the problems of strong subjectivity in manual search and insufficient feature extraction capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0065] Figure 1 Schematic diagram of electricity theft detection method based on IVMD-WVD and CNN;

[0066] Figure 2 This is the structure diagram of the electricity theft detection system based on IVMD-WVD and CNN;

[0067] Figure 3 It is a flow chart of the electricity theft detection method based on IVMD-WVD and CNN;

[0068] Figure 4 Examples of different types of electricity theft for a certain user sample;

[0069] Figure 5 It is a schematic diagram of normal samples in the normalized IVMD-WVD spectrum;

[0070] Figure 6 It is a schematic diagram of electricity theft type 1 in the normalized IVMD-WVD spectrum;

[0071] Figure 7It is a schematic diagram of power theft type 2 in the normalized IVMD-WVD spectrum;

[0072] Figure 8 It is a schematic diagram of power theft type 3 in the normalized IVMD-WVD spectrum;

[0073] Fig. 9 It is a schematic diagram of power theft type 4 in the normalized IVMD-WVD spectrum;

[0074] Fig.10 It is a schematic diagram of power theft type 5 in the normalized IVMD-WVD spectrum;

[0075] Fig.11 It is a schematic diagram of power theft type 6 in the normalized IVMD-WVD spectrum;

[0076] Fig.12 It is a schematic diagram of a normal sample in the normalized IVMD-WVD grayscale spectrum;

[0077] Fig.13 It is a schematic diagram of power theft type 1 in the normalized IVMD-WVD grayscale spectrum;

[0078] Fig.14 It is a schematic diagram of power theft type 2 in the normalized IVMD-WVD grayscale spectrum;

[0079] Fig.15 It is a schematic diagram of power theft type 3 in the normalized IVMD-WVD grayscale spectrum;

[0080] Fig.16 It is a schematic diagram of power theft type 4 in the normalized IVMD-WVD grayscale spectrum;

[0081] Fig.17 It is a schematic diagram of power theft type 5 in the normalized IVMD-WVD grayscale spectrum;

[0082] Fig.18 It is a schematic diagram of power theft type 6 in the normalized IVMD-WVD grayscale spectrum;

[0083] Fig.19 It is a schematic diagram of the network structure;

[0084] Fig. 20 This is a schematic diagram of the detection results of Example 3. DETAILED DESCRIPTION

[0085] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0086] Example 1

[0087] This embodiment provides a method for detecting electricity theft based on IVMD-WVD and CNN. Figure 1 As shown, including:

[0088] Step 1, collecting data: collecting the user's historical electricity usage data, and preprocessing the historical electricity usage data; the preprocessing method includes: cleaning the user's historical electricity usage data to remove invalid data.

[0089] Step 2: Build a power theft detection model:

[0090] Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample.

[0091] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0092] The samples include a normal sample and a plurality of different types of electricity theft samples; the electricity theft samples include: a fixed ratio reduced metering power sample S1 (w t ), random threshold reduction power sample S2(w t ), random time period reduction power sample S3 (w t ), random threshold reduction of power samples S4 (w t ), reverse power consumption time sample S5 (w t ) and average power consumption sample S6(w t ).

[0093] Fixed ratio reduction of metering power sample S1(w t )for:

[0094] S1(w t )=μw t (0.1<μ<0.8); (1)

[0095] Among them, μ represents the interval random constant;

[0096] Random threshold reduction power sample S2(w t )for:

[0097]

[0098] Among them, λ represents a random constant between the minimum power value and the maximum power value;

[0099] Random time period reduction power sample S3(w t )for:

[0100]

[0101] Where t1 and t2 represent any time of the day, and t2 ≥ t1 + 3;

[0102] Random threshold reduces the power sample S4(w t )for:

[0103] S4(w t )=max{w t -δ,0}; (4)

[0104] Among them, δ represents a random constant, taking the average or minimum value;

[0105] Reversed power consumption time sample S5(w t )for:

[0106] S5(w t )=w 97-t ; (5)

[0107] Among them, w 97-t Indicates the user's electricity consumption data after time t;

[0108] Average power consumption sample S6 (w t )for:

[0109] S6(w t )=mean(w t ); (6)

[0110] Among them, mean(w t ) means taking the average value of the user's electricity consumption.

[0111] Configure the initial decomposition level K or , construct a constrained electricity theft sample variational problem model;

[0112]

[0113] Among them, S(w t ) represents the user's electricity consumption data; ω k represents the center frequency of each mode function; represents the partial derivative with respect to time t; δ(t) represents the impulse function; Indicates the square of the Euclidean distance; S IMFk It represents 10 partial discharge signal modal functions obtained by decomposition, and i is an imaginary unit.

[0114] The energy difference method is introduced to improve the variational modal decomposition model, and samples are input into the improved variational modal decomposition model to decompose the modal components of each sample; the method includes:

[0115] Configure the initial decomposition level K or , construct a constrained electricity theft sample variational problem model;

[0116] The Lagrange multiplier method and penalty factor α are introduced to optimize the constrained electricity theft sample variational problem model;

[0117] The energy difference method is introduced to improve the optimization process of the constraint electricity theft sample variational problem model and obtain the optimal decomposition layer number K opt ;

[0118] Based on the optimal decomposition level K opt Decompose the modal components of each sample.

[0119] The specific number of initial decomposition layers K or = 10, the energy difference method is introduced to improve the variational mode decomposition (IVMD), and the optimal decomposition layer number K is obtained to convert the normal sample S (w t ) and 6 types of electricity theft samples S1(w t )、S2(w t )、S3(w t )、S4(w t )、S5(w t )、S6(w t ) are used to calculate IVMD respectively, and the modal components S of 7 types of samples are obtained. IMF (w t ).

[0120] The Wigner-Ville distribution (WVD) of the sample is obtained according to the following formula, and the IVMD-WVD spectra of the seven types of samples are obtained;

[0121]

[0122] Where IVMD-WVD(t,ω) represents the IVMD-WVD spectrum, K represents S IMFk The total number of S; k represents the kth S IMFk ; S IMFk represents the modal function of electricity consumption samples; α represents the time window; t represents the sample time span; ω represents the frequency of the sample at time t; i represents the imaginary unit.

[0123] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0124] Each IVMD-WVD spectrum was normalized according to the following formula:

[0125]

[0126] The normalized IVMD-WVD spectrum is grayed according to the following formula:

[0127]

[0128] Where (a, b) represents the value of IVMD-WVD spectrum at (a, b); IVMD-WVD nor 、IVMD-WVD nor-gray They represent IVMD-WVD normalized spectrum and IVMD-WVD normalized grayscale spectrum respectively; They represent the red, green and blue components of the IVMD-WVD normalized spectrum at (m, n), respectively.

[0129] The improved CNN network is trained based on the grayscale processed IVMD-WVD graph to obtain the electricity theft detection model, including the following method:

[0130] Build the AlexNet network and randomly divide the IVMD-WVD atlas of each sample into training set, validation set, and test set according to the proportion;

[0131] Training the AlexNet network based on the training set and the validation set;

[0132] Test the trained AlexNet network based on the test set.

[0133] Step three, electricity theft detection: collect the user's real-time electricity usage data, pre-process the real-time electricity usage data, input the pre-processed real-time electricity usage data into the constructed electricity theft detection model, and obtain the user's electricity theft result.

[0134] At present, electricity theft detection is mainly based on manual analysis and data-driven machine learning methods; manual analysis is to manually obtain abnormal areas in the metering system, analyze users in the area house by house, determine abnormal users and go to the site for detection. However, the method based on manual analysis requires the use of the rich experience of the staff, and there are problems such as strong subjectivity and low efficiency. The data-driven machine learning method mainly establishes feature libraries such as area loss power, user power, and area line loss rate, and uses machine learning algorithms such as support vector machines and K-means clustering to detect electricity theft users. However, the establishment of such features also needs to rely on manual experience selection, the algorithm has weak ability to extract high-dimensional features, and the probability of misjudging electricity theft users is high. Compared with the above method, the embodiment method provided by the present invention is based on the AlexNet network of the improved CNN, which has extremely excellent feature extraction processing capabilities and generalization processing capabilities, and effectively avoids or suppresses phenomena such as gradient disappearance during CNN training. The AlexNet network model has a stable convergence speed, fast model training speed, and improves the detection rate of electricity theft. This scheme performs IVMD-WVD on the user's power curve. This method has the advantages of ultra-high time-frequency resolution, good mathematical properties, fast calculation speed, excellent stability, and strong anti-interference ability. The obtained IVMD-WVD grayscale map can well retain the data edge characteristics, instantaneous frequency and other characteristics, and increase the training speed of the network model.

[0135] Example 2

[0136] This embodiment provides an electricity theft detection system based on IVMD-WVD and CNN, such as Figure 2 As shown, for implementing the above-mentioned electricity theft detection method based on IVMD-WVD and CNN, the system includes:

[0137] A collection module, used to collect historical electricity consumption data of users and pre-process the historical electricity consumption data;

[0138] Model building module, used to build electricity theft detection model:

[0139] Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample.

[0140] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0141] The detection module is used for electricity theft detection: collecting the user's real-time electricity usage data, preprocessing the real-time electricity usage data, inputting the preprocessed real-time electricity usage data into the constructed electricity theft detection model, and obtaining the user's electricity theft result.

[0142] Example 3

[0143] This embodiment provides a computer-readable medium on which a computer program is stored. The computer program is executed by a processor to implement the electricity theft detection method based on IVMD-WVD and CNN as described in Embodiment 1; specifically, the following steps are performed:

[0144] Step 1: Collecting data: collecting historical electricity consumption data of users and preprocessing the historical electricity consumption data;

[0145] Step 2: Build a power theft detection model:

[0146] Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample.

[0147] The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model.

[0148] Step three, electricity theft detection: collect the user's real-time electricity usage data, pre-process the real-time electricity usage data, input the pre-processed real-time electricity usage data into the constructed electricity theft detection model, and obtain the user's electricity theft result.

[0149] In order to verify the actual effect of the method of the present invention, this embodiment selects the power consumption data of 1000 normal users in the system for the past 7 days. Each data record records the power consumption of the user in 672 time periods in 7 days in units of 15 minutes. Figure 3 and Figure 4 As shown, it mainly includes the following steps:

[0150] S1, the user's electricity consumption data is used to construct 6 types of electricity theft samples according to the electricity theft model of formulas (1)-(6); see the attached Figure 5-Figure 12 ;

[0151] S2, initially set the number of VMD calculation layers K or =10, construct the constrained electricity theft sample variational problem model, as shown in formula (8);

[0152] S3, introduces the Lagrange multiplier method and penalty factor α to solve the constrained optimization problem of formula (8), and iterates the optimization sequence through formulas (9)-(11) to obtain the sample mode function and center frequency, and iteratively updates the sample mode function and center frequency. The iteration is stopped when the convergence condition of formula (12) is met.

[0153]

[0154] in, represent the original data after Fourier transformation, the mode function and the Lagrange multiplier operator respectively; ω, ω k represents the original data and the center frequency of each mode function; α represents the penalty factor; β represents the update factor; ε represents the convergence threshold.

[0155] S4, introduce equations (13)-(15) to improve VMD calculation and obtain the optimal decomposition level K opt , re-substitute into equations (8)-(12) to obtain the final modal components of the sample.

[0156]

[0157] Where E represents the energy of the data; E OR 、E IMFk Represent the energy of the original data and the kth mode function respectively; θ k Represents the energy ratio of the kth layer; K opt represents the optimal number of decomposition levels.

[0158] S5, press (The Wignerville distribution (WVD) of the sample is obtained by calculating according to equations (16) and (17) in sequence to obtain the normalized and grayscaled sample IVMD-WVD grayscale spectrum, see Appendix Figure 13-Figure 18 .

[0159] S6, such as Fig.19 As shown in the figure, the AlexNet network is built, and the IVMD-WVD grayscale maps of each sample are randomly divided into training set, verification set, and test set in a ratio of 6:2:2. The network is trained to obtain the optimal number of iterations and learning rate. Finally, the optimal training network is used to detect normal samples and electricity theft samples to obtain the final electricity theft user detection rate. The detection results are shown in Fig. 20 shown.

[0160] From steps S1 to S6, the average detection rate of the method of the present invention can reach 94.29%, and the detection effect is higher than that of manual screening and data feature-based screening. The results of the present invention can avoid the problems of strong subjectivity, low efficiency, and weak feature extraction ability of manual screening, reduce the probability of misjudgment of electricity theft users, and realize rapid and accurate screening of electricity theft users.

[0161] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The electricity theft detection method based on IVMD-WVD and CNN is characterized by: include: Collecting data: collecting historical electricity consumption data of users and preprocessing the historical electricity consumption data; Build a power theft detection model: Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample. The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model. Electricity theft detection: Collect the user's real-time electricity usage data, pre-process the real-time electricity usage data, input the pre-processed real-time electricity usage data into the constructed electricity theft detection model, and obtain the user's electricity theft results.

2. The electricity theft detection method based on IVMD-WVD and CNN according to claim 1 is characterized in that: The preprocessing method includes: cleaning the historical electricity consumption data of the user to remove invalid data.

3. The electricity theft detection method based on IVMD-WVD and CNN according to claim 1 is characterized in that: The samples include a normal sample and a plurality of different types of electricity theft samples; the electricity theft samples include: a fixed ratio reduced metering power sample S1 (w t ), random threshold reduction power sample S2(w t ), random time period reduction power sample S3 (w t ), random threshold reduction of power samples S4 (w t ), reverse power consumption time sample S5 (w t ) and average power consumption sample S6(w t ).

4. The electricity theft detection method based on IVMD-WVD and CNN according to claim 3 is characterized in that: Fixed ratio reduction of metering power sample S1(w t )for: S1(w t )=μw t (0.1<μ<0.8); Among them, μ represents an interval random constant; Random threshold reduction power sample S2(w t )for: Among them, λ represents a random constant between the minimum power value and the maximum power value; Random time period reduction power sample S3(w t )for: Where t1 and t2 represent any time of the day, and t2 ≥ t1 + 3; Random threshold reduces the power sample S4(w t )for: S4(in t )=max{in t -δ,0}; Among them, δ represents a random constant, taking the average or minimum value; Reversed power consumption time sample S5(w t )for: S5(w t )=w 97-t ; Among them, w 97-t Indicates the user's electricity consumption data after time t; Average power consumption sample S6 (w t )for: S6(w t )=mean(w t ); Among them, mean(w t ) means taking the average value of the user's electricity consumption.

5. The electricity theft detection method based on IVMD-WVD and CNN according to claim 3 is characterized in that: The energy difference method is introduced to improve the variational modal decomposition model, and samples are input into the improved variational modal decomposition model to decompose the modal components of each sample; the method includes: Configure the initial decomposition level K or , construct a constrained electricity theft sample variational problem model; The Lagrange multiplier method and penalty factor α are introduced to optimize the constrained electricity theft sample variational problem model; The energy difference method is introduced to improve the optimization process of the constraint electricity theft sample variational problem model and obtain the optimal decomposition layer number K opt ; Based on the optimal decomposition level K opt Decompose the modal components of each sample.

6. The electricity theft detection method based on IVMD-WVD and CNN according to claim 5 is characterized in that: The constrained electricity theft sample variational problem model includes: Among them, S(w t ) represents the user's electricity consumption data; ω k represents the center frequency of each mode function; represents the partial derivative with respect to time t; δ(t) represents the impulse function; Indicates the square of the Euclidean distance; S IMFk It represents 10 partial discharge signal modal functions obtained by decomposition, and i is an imaginary unit.

7. The electricity theft detection method based on IVMD-WVD and CNN according to claim 3 is characterized in that: The method for solving the IVMD-WVD spectrum of each sample includes: The IVMD-WVD spectrum was calculated according to the following formula: Where IVMD-WVD(t,ω) represents the IVMD-WVD spectrum, K represents S IMFk The total number of S; k represents the kth S IMFk ; S IMFk represents the modal function of electricity consumption samples; α represents the time window; t represents the sample time span; ω represents the frequency of the sample at time t; i represents the imaginary unit.

8. The electricity theft detection method based on IVMD-WVD and CNN according to claim 1 is characterized in that: The grayscale processing of each IVMD-WVD spectrum includes the following method: Each IVMD-WVD spectrum was normalized according to the following formula: The normalized IVMD-WVD spectrum is grayed according to the following formula: Where (a, b) represents the value of IVMD-WVD spectrum at (a, b); IVMD-WVD nor 、IVMD-WVD nor-gray They represent IVMD-WVD normalized spectrum and IVMD-WVD normalized grayscale spectrum respectively; They represent the red, green and blue components of the IVMD-WVD normalized spectrum at (m, n), respectively.

9. The electricity theft detection method based on IVMD-WVD and CNN according to claim 1, characterized in that: The improved CNN network is trained based on the grayscale processed IVMD-WVD graph to obtain the electricity theft detection model, including the following method: Build the AlexNet network and randomly divide the IVMD-WVD atlas of each sample into training set, validation set, and test set according to the proportion; Training the AlexNet network based on the training set and the validation set; Test the trained AlexNet network based on the test set.

10. The electricity theft detection system based on IVMD-WVD and CNN is characterized by: The system is used to implement the electricity theft detection method based on IVMD-WVD and CNN as described in any one of claims 1 to 9, and comprises: A collection module, used to collect historical electricity consumption data of users and pre-process the historical electricity consumption data; Model building module, used to build electricity theft detection model: Based on the preprocessed historical electricity consumption data, multiple samples are constructed, and the energy difference method is introduced to improve the variational modal decomposition model. The samples are input into the improved variational modal decomposition model to decompose the modal components of each sample. The IVMD-WVD spectrum of each sample is solved according to the modal components of each sample, and each IVMD-WVD spectrum is gray-processed. The improved CNN network is trained based on the IVMD-WVD spectrum after gray-processing to obtain the electricity theft detection model. The detection module is used for electricity theft detection: collecting the user's real-time electricity usage data, preprocessing the real-time electricity usage data, inputting the preprocessed real-time electricity usage data into the constructed electricity theft detection model, and obtaining the user's electricity theft result.

Images