A non-intrusive load identification method and system based on wavelet transform

Abstract

The application relates to the technical field of load identification, in particular to a non-intrusive load identification method and system based on wavelet transform, which comprises the following steps: wavelet decomposition is performed on the collected current time sequence signal of a household total incoming line to obtain multiple layers of wavelet coefficients; harmonic characteristics in each layer of wavelet coefficients are analyzed and removed, and a current feature matrix of the current time sequence signal is constructed; the current feature matrix is input into a pre-trained label discrimination model, and the type label of an electrical appliance put into the current household total incoming line is output. The application aims to select and correct the wavelet coefficients of wavelet decomposition to remove the influence of harmonic currents, and improve the precision of the multi-dimensional feature matrix of the current time sequence signal constructed by the wavelet coefficients for non-intrusive load identification.

Description

Technical Field

[0001] This application relates to the field of load identification technology, specifically to a non-intrusive load identification method and system based on wavelet transform. Background Technology

[0002] Non-intrusive load identification (NIDI) can monitor the power data of the main circuit in real time by installing sensor devices in the home. By analyzing the power data of individual points in the main circuit, it can separate the power of appliances and identify their operating status, enabling users to accurately manage their electricity consumption and providing real-time data for grid optimization and dispatch. The prerequisite for implementing NIDI is to ensure that the characteristics of the acquired current data are as accurate as possible.

[0003] Current non-invasive load identification methods primarily identify current time-series signals. However, in real-world households, there are often variable frequency appliances, such as variable frequency air conditioners and refrigerators. These appliances typically generate harmonic currents during frequency conversion, causing instantaneous current changes in the current time-series signal. This can be mistakenly interpreted as the connection or disconnection of a new appliance, leading to errors in extracting the feature vectors at the time of connection or disconnection, thus affecting the accuracy of non-invasive load identification. Furthermore, wavelet coefficients obtained using wavelet decomposition often contain harmonic current features, resulting in inaccurate feature matrices for the current time-series signal extracted using wavelet coefficients. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a non-intrusive load identification method and system based on wavelet transform, the specific technical solution of which is as follows: In a first aspect, one embodiment of this application provides a non-intrusive load identification method based on wavelet transform, the method comprising the following steps: Wavelet decomposition was performed on the collected current timing signal of the main incoming line of the household to obtain multi-level wavelet coefficients. Each level of wavelet coefficients includes low-frequency approximation coefficients and high-frequency detail coefficients. By combining the harmonic distortion rate of the low-frequency approximation coefficients in the frequency domain spectrum, the harmonic energy ratio of the high-frequency detail coefficients in the frequency domain spectrum, and the average energy information entropy of the low-frequency approximation coefficients and the high-frequency detail coefficients, the first harmonic feature of each layer of wavelet coefficients is constructed. Abnormal wavelet coefficients in each layer of wavelet coefficients are detected, and their amplitude abnormality characteristics and first harmonic characteristics are combined to construct the second harmonic characteristics of each layer of wavelet coefficients. The correction threshold of each layer of wavelet coefficients is determined by using the second harmonic feature, and soft threshold correction is performed on the wavelet coefficients to remove the harmonic feature; Load identification features are extracted from the low-frequency coefficients of the last corrected layer and the high-frequency detail coefficients of all layers respectively, and feature vectors of the corresponding layers are constructed to form a current feature matrix. Input the current feature matrix into the pre-trained label discrimination model, and output the label of the type of appliance connected to the current household main line.

[0005] Preferably, the first harmonic feature is positively correlated with the harmonic distortion rate, the harmonic energy ratio, and the mean energy information entropy.

[0006] Preferably, the proportion of harmonic energy is determined by calculating the ratio of harmonic energy to the total energy of the frequency domain spectrum of the high-frequency detail coefficients, wherein the harmonic energy is the sum of squares of the amplitudes of the high-frequency detail coefficients in the frequency domain spectrum whose frequencies are integer multiples of the fundamental frequency.

[0007] Preferably, the method for constructing the second harmonic feature is as follows: Calculate the coefficient of variation of the amplitude anomaly characteristics of all anomalous wavelet coefficients in the low-frequency approximation coefficients and high-frequency detail coefficients of each layer of wavelet coefficients; The second harmonic feature is positively correlated with the first harmonic feature and the mean of the two coefficients of variation.

[0008] Preferably, the method for determining the abnormal wavelet coefficients is as follows: the wavelet coefficients in each layer of low-frequency approximation coefficients and high-frequency detail coefficients whose amplitudes are simultaneously higher than the wavelet coefficients on their left and right sides, and whose amplitudes are greater than a preset multiple of the average amplitude of the wavelet coefficients in that layer, are recorded as abnormal wavelet coefficients.

[0009] Preferably, the method for determining the amplitude anomaly feature is as follows: the proportion of the amplitude of the anomalous wavelet coefficient to the mean amplitude of all wavelet coefficients other than the anomalous wavelet coefficient in the low-frequency approximation coefficient or high-frequency detail coefficient where the anomalous wavelet coefficient is located is taken as the amplitude anomaly feature of the anomalous wavelet coefficient.

[0010] Preferably, the rule for soft threshold correction is: when hour, ;otherwise ; in, This represents the s-th corrected wavelet coefficient in the j-th layer of wavelet coefficients. Represents a symbolic function. This represents the s-th wavelet coefficient in the j-th layer of wavelet coefficients. This represents the correction threshold for the wavelet coefficients of the j-th layer.

[0011] Preferably, the correction threshold is obtained by substituting the second harmonic feature into the general wavelet threshold formula.

[0012] Preferably, the load identification features extracted from the feature vector are: peak factor, mean, standard deviation, kurtosis, and peak variation coefficient.

[0013] Secondly, another embodiment of this application provides a non-intrusive load identification system based on wavelet transform, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the non-intrusive load identification method based on wavelet transform.

[0014] This application has at least the following beneficial effects: 1. This application addresses the issue that the accuracy of non-intrusive load identification is easily affected by harmonic currents from frequency converters. It obtains multi-layer wavelet coefficients through wavelet decomposition, determines the correction threshold by calculating the harmonic characteristics of the wavelet coefficients, and removes harmonic characteristics by correcting the wavelet coefficients, thereby avoiding the reduction of load identification accuracy by harmonic currents from frequency converters.

[0015] 2. This application addresses the issue that directly identifying current time-series signals is susceptible to the influence of harmonic currents from frequency converters. It constructs a multi-dimensional feature matrix using wavelet coefficients as the feature matrix for the current time-series signal. By identifying the load using the multi-dimensional feature matrix, the accuracy of identifying the type of load appliance is effectively improved. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a non-intrusive load identification method based on wavelet transform, provided as an embodiment of this application. Detailed Implementation

[0018] Example 1 This application provides an embodiment of a non-intrusive load identification method based on wavelet transform, see details below. Figure 1 The method includes the following steps: Step 1: Perform wavelet decomposition on the collected current timing signal of the main incoming line of the household to obtain multi-layer wavelet coefficients.

[0019] By installing smart meters in each household to collect current timing signals, this embodiment sets the sampling period to 15 minutes and the sampling frequency to 20kHz.

[0020] In general, most households have inverter appliances installed, such as inverter air conditioners and inverter refrigerators. Inverter appliances generate harmonic currents, which cause distortion of the current timing signal. This leads to errors in extracting the feature vector when the current timing signal is accessed or cut off, affecting the accuracy of non-invasive load identification.

[0021] To more intuitively highlight the impact of harmonic current generated by frequency converters on current timing signals, this application uses the current timing signal as the input to the wavelet decomposition algorithm. The wavelet basis type and the number of decomposition layers can be set by the implementer without feature restrictions. In this embodiment, the wavelet basis is selected as the db wavelet (Daubechies wavelet), and the number of decomposition layers is set to 3, resulting in 3 layers of wavelet coefficients. Each layer of wavelet coefficients includes low-frequency approximation coefficients and high-frequency detail coefficients. The wavelet decomposition algorithm is a well-known technology, and its specific implementation will not be described in detail.

[0022] Step 2: Analyze and remove the harmonic features in the wavelet coefficients of each layer, and construct the current feature matrix of the current time series signal.

[0023] S1: Combining the harmonic distortion rate of the low-frequency approximation coefficients in the frequency domain spectrum, the harmonic energy ratio of the high-frequency detail coefficients in the frequency domain spectrum, and the average energy information entropy of the low-frequency approximation coefficients and the high-frequency detail coefficients, construct the first harmonic feature of each layer of wavelet coefficients.

[0024] When unaffected by harmonic currents from frequency converters, the wavelet coefficient energy (fundamental frequency energy) of the current timing signal is generally concentrated in the low-frequency wavelet coefficients. However, frequency converters typically generate harmonic currents during frequency modulation, resulting in odd harmonics such as the 5th and 7th harmonics in the current timing signal, which causes the wavelet coefficient energy distribution at different levels to become scattered.

[0025] To obtain the energy distribution characteristics of wavelet coefficients of the current time-series signal under the influence of harmonic currents of frequency converters, this application performs single-branch reconstruction on each layer of wavelet coefficients. After restoring them to time-series signal components, Fourier transform is then used to obtain the frequency domain spectra of low-frequency approximation coefficients and high-frequency detail coefficients. Based on a fundamental frequency of 50Hz, the harmonic distortion rate of the low-frequency approximation coefficients is calculated through the frequency domain spectrum of the low-frequency approximation coefficients, reflecting the degree of influence of the harmonic currents of frequency converters on the low-frequency approximation coefficients.

[0026] Since the frequency domain spectrum of the high-frequency detail coefficients does not include the fundamental frequency, the ratio of harmonic energy to the total energy of the frequency domain spectrum of the high-frequency detail coefficients is calculated and denoted as the harmonic energy ratio. This ratio reflects the degree of influence of the harmonic current of the frequency converter on the high-frequency detail coefficients. The harmonic energy is the sum of the squares of the amplitudes of the high-frequency detail coefficients in the frequency domain spectrum whose frequencies are integer multiples of the fundamental frequency.

[0027] Furthermore, we consider analyzing the entropy value of the wavelet coefficient energy at each layer to reflect the degree of dispersion of the wavelet coefficients. We calculate the mean value of the energy information entropy of the low-frequency approximation coefficients and the high-frequency detail coefficients, respectively, and denote it as the first mean value, which reflects the degree of dispersion of the wavelet coefficient energy distribution under the influence of harmonic currents in variable frequency appliances.

[0028] Among them, Fourier transform, harmonic distortion rate, harmonic energy, total energy of frequency domain spectrum, and energy information entropy are all well-known technologies, and their specific implementations will not be elaborated here.

[0029] Based on the above analysis, this application combines the harmonic distortion rate of the low-frequency approximation coefficients in the frequency domain spectrum, the harmonic energy ratio of the high-frequency detail coefficients in the frequency domain spectrum, and the average energy information entropy of the low-frequency approximation coefficients and the high-frequency detail coefficients to construct the first harmonic feature of each layer of wavelet coefficients. This feature is used to characterize the significance of the influence of harmonic current of frequency-converted appliances on each layer of wavelet coefficients after wavelet decomposition of household current time-series signals.

[0030] The first harmonic feature is positively correlated with the harmonic distortion rate, the harmonic energy ratio, and the mean value of the energy information entropy.

[0031] It is understandable that a positive correlation means that the dependent variable increases as the independent variable increases, and the dependent variable decreases as the independent variable decreases. This is determined by the actual application and is not subject to any special restrictions in this application.

[0032] Specifically, this embodiment takes the j-th layer wavelet coefficients as an example to calculate the first harmonic characteristic of the j-th layer wavelet coefficients: This represents the first harmonic characteristic of the wavelet coefficients at the j-th level. This represents the harmonic distortion rate of the low-frequency approximation coefficients in the j-th layer wavelet coefficients. This represents the proportion of harmonic energy in the high-frequency detail coefficients of the j-th wavelet coefficients. This represents the first mean of the wavelet coefficients at the j-th level. This represents the hyperbolic tangent function, ensuring the order of magnitude is consistent with the left side of the plus sign. In wavelet coefficients, a higher harmonic distortion rate, a larger proportion of harmonic energy, and a larger first mean indicate the first harmonic characteristic. The larger the value, the more significant the influence of harmonic current of frequency converter appliances on the wavelet coefficients after wavelet decomposition of household current time sequence signal.

[0033] It should be noted that, in order to analyze the significance of the influence of harmonic currents on the wavelet coefficients after wavelet decomposition of the household current time-series signal, the following method was used: and The influence of frequency converter harmonics on the energy distribution of low-frequency approximation coefficients and high-frequency detail coefficients of wavelet coefficients is obtained respectively. Their product reflects the influence of frequency converter harmonics on the overall energy distribution of wavelet coefficients. The average energy information entropy of the high- and low-frequency wavelet coefficients reflects the degree of dispersion of the overall wavelet coefficient energy distribution under the influence of harmonic currents in variable frequency appliances; therefore, the first harmonic characteristic of the wavelet coefficients is obtained through comprehensive calculation. It can reflect the significance of the influence of harmonic current of frequency conversion appliances on the wavelet coefficients after wavelet decomposition of household current time sequence signal.

[0034] S2: Detect anomalous wavelet coefficients in each layer of wavelet coefficients, and construct the second harmonic feature of each layer of wavelet coefficients by combining their amplitude anomalous features and the first harmonic feature.

[0035] Since wavelet decomposition decomposes the current time-series signal into components of different scales, the wavelet coefficients of different layers are affected to different degrees by the harmonic current of the frequency converter. The greater the degree of influence, the greater the abnormal fluctuation of the wavelet coefficient amplitude.

[0036] When the frequency converter performs frequency modulation at a certain moment, it generates harmonic current. The decomposed wavelet coefficients exhibit abnormal abrupt changes, exceeding the amplitude of the wavelet coefficients on the left and right. To analyze the amplitude anomaly characteristics of wavelet coefficients at different levels, the wavelet coefficients in the low-frequency approximation coefficients and high-frequency detail coefficients of each level whose amplitudes are simultaneously higher than the wavelet coefficients on the left and right, and whose amplitudes are greater than the average amplitude of the wavelet coefficients in that level (in this embodiment, the preset multiple is set to 2), are recorded as abnormal wavelet coefficients and used for subsequent analysis of amplitude anomaly characteristics. Note that since the first and last data points lack adjacent points on the left or right, the data points at the first and last ends are not included in the detection of abnormal wavelet coefficients.

[0037] Furthermore, this application utilizes the amplitude anomaly characteristics of the abnormal wavelet coefficients to construct the second harmonic characteristics of each layer of wavelet coefficients, which are used to characterize the significance of amplitude anomaly fluctuations of the wavelet coefficients after wavelet decomposition of household current time-series signals affected by harmonic currents of frequency converters.

[0038] Preferably, the method for constructing the second harmonic feature is as follows: calculate the variation coefficient of the amplitude anomaly feature of all abnormal wavelet coefficients in the low-frequency approximation coefficient and high-frequency detail coefficient in each layer of wavelet coefficients; the second harmonic feature is positively correlated with the first harmonic feature and the mean of the two variation coefficients.

[0039] Specifically, in this embodiment, the method for calculating the second harmonic characteristic of the j-th layer wavelet coefficients is as follows: ; This represents the second harmonic characteristic of the wavelet coefficients at the j-th layer. The coefficient of variation represents the amplitude anomaly characteristics of all anomalous wavelet coefficients in the low-frequency approximation coefficients of the j-th layer wavelet coefficients, reflecting the degree of imbalance in the amplitude anomalies of the anomalous wavelet coefficients. The coefficient of variation represents the amplitude anomaly characteristics of all anomalous wavelet coefficients in the high-frequency detail coefficients of the j-th layer wavelet coefficients, and its calculation method is as follows: The calculation method is the same; the more unbalanced the anomalous amplitude characteristics of the wavelet coefficients in each layer, the better. The larger the value, the more significant the abnormal fluctuations in the amplitude of the j-th layer wavelet coefficients after wavelet decomposition of the household current time-series signal due to the influence of harmonic currents from frequency converters. The calculation process for the coefficient of variation is a well-known technique and will not be elaborated further. It should also be noted that when the number of detected abnormal wavelet coefficients is less than 2, the coefficient of variation is directly assigned a value of 0.

[0040] After wavelet decomposition of household current time-series signals, the mean value of the coefficient of variation of wavelet coefficients under the influence of harmonic currents from frequency converters is determined based on the amplitude anomaly characteristics of high- and low-frequency wavelet coefficients. This reflects the imbalance of the anomalous amplitude characteristics of the wavelet coefficients. To characterize the significance of the influence of harmonic currents from frequency converters on the wavelet coefficients after wavelet decomposition of household current time-series signals, and based on the above indicators, the second harmonic characteristic of the j-th layer wavelet coefficients is calculated. It can reflect the significance of abnormal fluctuations in the amplitude of wavelet coefficients after wavelet decomposition of household current time sequence signal due to the influence of harmonic current of frequency conversion appliances.

[0041] In this application, the amplitude anomalous characteristics of the anomalous wavelet coefficients are determined by the proportion of the amplitude of the anomalous wavelet coefficient to the average amplitude of all wavelet coefficients other than the anomalous wavelet coefficient in the low-frequency approximation coefficients or high-frequency detail coefficients where the anomalous wavelet coefficient is located.

[0042] Specifically, in this embodiment, the method for calculating the amplitude anomaly characteristics of the abnormal wavelet coefficients is as follows: , This represents the amplitude anomaly characteristic of the k-th anomalous wavelet coefficient among the low-frequency approximation coefficients in the j-th layer of wavelet coefficients. This indicates the number of anomalous wavelet coefficients in the low-frequency approximation coefficients. If K is less than 1, it means that there are no anomalous wavelet coefficients. The value is 0. This indicates that the amplitude of the wavelet coefficients is obtained by taking the modulus of the wavelet coefficients. This represents the amplitude of the k-th anomalous wavelet coefficient among the low-frequency approximation coefficients in the j-th layer of wavelet coefficients. This represents the mean amplitude of the low-frequency approximation coefficients in the j-th layer wavelet coefficients, excluding anomalous wavelet coefficients. The larger the value, the better. The greater the difference in amplitude compared to other wavelet coefficients, the more prominent it is among all anomalous wavelet coefficients.

[0043] According to the principle of ratio weighting calculation The value of the amplitude anomaly is determined based on the degree of difference between the amplitude of each anomalous wavelet coefficient and the mean amplitude. The greater the degree of difference, the larger the amplitude anomaly characteristic value.

[0044] It should be noted that, in this application, to avoid the denominator in the formula having a value of 0, a zero-suppression parameter is set for all formula denominators that may have a value of 0. The specific value is set by the implementer according to the dimension of the denominator. In this embodiment, it is set as follows: .

[0045] S3: Use the second harmonic feature to determine the correction threshold of each layer of wavelet coefficients, and perform soft threshold correction on the wavelet coefficients to remove harmonic features.

[0046] The last layer of low-frequency approximation coefficients in wavelet coefficients usually contain most of the features of the current time series signal and are more stable. High-frequency detail features usually contain important information for load identification, such as transient features like electrical switching. Therefore, when constructing the current feature matrix, the last layer of low-frequency coefficients and all layers of high-frequency detail coefficients are selected. However, these wavelet coefficients usually contain current harmonic features, which leads to inaccurate current feature matrix construction.

[0047] In this embodiment, the low-frequency coefficients of the third layer and the high-frequency detail coefficients of all layers are selected to construct the current characteristic matrix.

[0048] Furthermore, in order to remove the harmonic features of the wavelet coefficients at each level of the current time-series signal decomposition and construct a more accurate multidimensional feature matrix of the current time-series signal, the wavelet coefficients need to be corrected. The greater the influence of the harmonic current of the frequency converter on each level of wavelet coefficients, the greater the abnormal fluctuation of the wavelet coefficient amplitude, and the larger the correction threshold should be.

[0049] Accordingly, this application uses the second harmonic feature to determine the correction threshold of each layer of wavelet coefficients, and performs soft threshold correction on the wavelet coefficients to remove the harmonic feature.

[0050] The correction threshold is obtained by substituting the second harmonic feature into the general wavelet threshold formula.

[0051] In this embodiment, the correction threshold of the j-th layer wavelet coefficients The calculation formula is: , Indicates a base e and a base 0. The logarithmic function of the argument, The length of the wavelet coefficients in the j-th layer is represented by the length of the wavelet coefficients in the first two layers, which is equal to the length of the high-frequency detail coefficients. The length of the wavelet coefficients in the third layer is equal to the sum of the lengths of the low-frequency approximation coefficients and the high-frequency detail coefficients.

[0052] Based on the calculation principle of the general wavelet threshold formula The wavelet coefficient length parameter is a necessary parameter in the general wavelet threshold formula. The larger the value, the more harmonic features are contained in the wavelet coefficients, and the greater the influence of harmonic currents from frequency converters on the j-th layer wavelet coefficients. The calculated correction threshold... The larger.

[0053] In other implementations, the correction threshold for each layer of wavelet coefficients can also be obtained using other well-known wavelet thresholding formulas.

[0054] Specifically, the rule for performing soft thresholding correction on the wavelet coefficients of the j-th layer to remove harmonic features is as follows: when hour, ;otherwise ; in, This represents the s-th corrected wavelet coefficient in the j-th layer of wavelet coefficients. This represents a sign function, which takes the value 1, -1, or 0. This represents the s-th wavelet coefficient in the j-th layer of wavelet coefficients. This represents the correction threshold for the wavelet coefficients of the j-th layer.

[0055] It should be noted that the energy of wavelet coefficients containing only harmonic current characteristics is significantly lower than that of wavelet coefficients containing useful characteristics. Therefore, their amplitude is usually significantly smaller than the threshold. Since wavelet coefficients that do not contain useful current characteristics can have their amplitude set to 0, the more harmonic characteristics they contain, the larger the correction threshold. The smaller the value, the more useful features can be retained while effectively removing a large number of harmonic current characteristics.

[0056] Based on the principle of wavelet soft thresholding denoising formula, and combined with correction threshold, harmonic features in wavelet coefficients of each layer are removed.

[0057] S4: Extract the load identification features from the low-frequency coefficients of the last corrected layer and the high-frequency detail coefficients of all layers respectively, and construct the feature vectors of the corresponding layers to form the current feature matrix.

[0058] For the corrected wavelet coefficients of each layer (i.e., the low-frequency approximation coefficients of the last layer and the high-frequency detail coefficients of all layers), the load identification features of each layer of wavelet coefficients are extracted. The load identification features include: average value, peak value, standard deviation, rectified average value, effective value, kurtosis, waveform factor, peak factor, active power, reactive power, 1st harmonic amplitude, 3rd harmonic amplitude, and 5th harmonic amplitude. For details, please refer to Table 2 in the paper He Sheng - "A Non-Intrusive Load Identification Method for Unknown New Equipment Based on Adaptive Adversarial Discriminant Domain".

[0059] The feature extraction dimension of the load identification features can be selected by the implementer without special restrictions. The load identification features extracted from each layer of wavelet coefficients are used to construct the feature vector of each layer of wavelet coefficients to form the current feature matrix.

[0060] This embodiment selects five load identification features, including peak factor, average value, standard deviation, kurtosis, and peak variation coefficient, to construct a multidimensional current feature matrix W as the feature matrix of the current time series signal, where W=[ ],in This represents the eigenvector of the low-frequency approximation coefficients in the third-level wavelet coefficients. , , These represent the feature vectors of the high-frequency approximation coefficients in the wavelet coefficients of the first, second, and third layers, respectively, and are all 5-dimensional feature vectors containing peak factor, mean, standard deviation, kurtosis, and peak variation coefficient.

[0061] Step 3: Input the current feature matrix into the pre-trained label discrimination model and output the labels of the types of appliances connected to the current household main line.

[0062] To accurately identify the types of electrical appliances connected in a household circuit, this embodiment selects four common appliances: inverter air conditioner, television, washing machine, and incandescent light bulb. These can form 10 training labels, including but not limited to {inverter air conditioner} and {inverter air conditioner, television}. For each label, 500 current time-series signals are acquired, and a current feature matrix is ​​extracted. A fully connected deep neural network (DNN) is used for training to obtain a pre-trained label discrimination model, which is used to determine the type of appliance label connected in the main household power line. The training loss function uses the cross-entropy loss function, the optimizer uses Adam, and the training epochs are set to 100.

[0063] Furthermore, the current feature matrix of the current timing signal of the current main line of the current household is input into the pre-trained label discrimination model, and the label of the type of appliance put into the current main line of the household is output.

[0064] Example 2 Another embodiment of this application provides a non-intrusive load identification system based on wavelet transform, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the non-intrusive load identification method based on wavelet transform.

[0065] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not invented in this application.

[0066] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A non-intrusive load identification method based on wavelet transform, characterized in that, The method includes the following steps: Wavelet decomposition was performed on the collected current timing signal of the main incoming line of the household to obtain multi-level wavelet coefficients. Each level of wavelet coefficients includes low-frequency approximation coefficients and high-frequency detail coefficients. By combining the harmonic distortion rate of the low-frequency approximation coefficients in the frequency domain spectrum, the harmonic energy ratio of the high-frequency detail coefficients in the frequency domain spectrum, and the average energy information entropy of the low-frequency approximation coefficients and the high-frequency detail coefficients, the first harmonic feature of each layer of wavelet coefficients is constructed. Abnormal wavelet coefficients in each layer of wavelet coefficients are detected, and their amplitude abnormality characteristics and first harmonic characteristics are combined to construct the second harmonic characteristics of each layer of wavelet coefficients. The correction threshold of each layer of wavelet coefficients is determined by using the second harmonic feature, and soft threshold correction is performed on the wavelet coefficients to remove the harmonic feature; Load identification features are extracted from the low-frequency coefficients of the last corrected layer and the high-frequency detail coefficients of all layers respectively, and feature vectors of the corresponding layers are constructed to form a current feature matrix. Input the current feature matrix into the pre-trained label discrimination model, and output the label of the type of appliance connected to the current household main line.

2. The non-intrusive load identification method based on wavelet transform as described in claim 1, characterized in that, The first harmonic feature is positively correlated with the harmonic distortion rate, the harmonic energy ratio, and the mean value of the energy information entropy.

3. The non-intrusive load identification method based on wavelet transform as described in claim 2, characterized in that, The harmonic energy ratio is determined by calculating the ratio of harmonic energy to the total energy of the frequency domain spectrum of the high-frequency detail coefficients, where the harmonic energy is the sum of the squares of the amplitudes of the high-frequency detail coefficients in the frequency domain spectrum whose frequencies are integer multiples of the fundamental frequency.

4. The non-intrusive load identification method based on wavelet transform as described in claim 1, characterized in that, The method for constructing the second harmonic feature is as follows: Calculate the coefficient of variation of the amplitude anomaly characteristics of all anomalous wavelet coefficients in the low-frequency approximation coefficients and high-frequency detail coefficients of each layer of wavelet coefficients; The second harmonic feature is positively correlated with the first harmonic feature and the mean of the two coefficients of variation.

5. The non-intrusive load identification method based on wavelet transform as described in claim 4, characterized in that, The method for determining the abnormal wavelet coefficients is as follows: the wavelet coefficients in each layer of low-frequency approximation coefficients and high-frequency detail coefficients whose amplitudes are simultaneously higher than the wavelet coefficients on their left and right sides, and whose amplitudes are greater than a preset multiple of the average amplitude of the wavelet coefficients in that layer, are recorded as abnormal wavelet coefficients.

6. The non-intrusive load identification method based on wavelet transform as described in claim 5, characterized in that, The method for determining the amplitude anomaly feature is as follows: the proportion of the amplitude of the anomalous wavelet coefficient to the mean amplitude of all wavelet coefficients other than the anomalous wavelet coefficient in the low-frequency approximation coefficient or high-frequency detail coefficient where the anomalous wavelet coefficient is located is taken as the amplitude anomaly feature of the anomalous wavelet coefficient.

7. The non-intrusive load identification method based on wavelet transform as described in claim 1, characterized in that, The rule for soft threshold correction is as follows: when hour, ;otherwise ; in, This represents the s-th corrected wavelet coefficient in the j-th layer of wavelet coefficients. Represents a symbolic function. This represents the s-th wavelet coefficient in the j-th layer of wavelet coefficients. This represents the correction threshold for the wavelet coefficients of the j-th layer.

8. The non-intrusive load identification method based on wavelet transform as described in claim 7, characterized in that, The correction threshold is obtained by substituting the second harmonic feature into the general wavelet threshold formula.

9. The non-intrusive load identification method based on wavelet transform as described in claim 1, characterized in that, The load identification features extracted from the feature vector are: peak factor, mean, standard deviation, kurtosis, and peak variation coefficient.

10. A non-intrusive load identification system based on wavelet transform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements a non-intrusive load identification method based on wavelet transform as described in any one of claims 1-9.

Images