Load identification method and system

Abstract

The invention discloses a load identification method and system. The method comprises the following steps: acquiring a first operation parameter of target equipment, and preprocessing the first operation parameter; performing feature extraction operation on the preprocessed first operation parameter to obtain a total feature vector; pre-training a first load identification model, and taking the total feature vector as the input of the first load identification model; and carrying out load identification according to the output of the first identification model. According to the method, the power feature, the waveform feature, the harmonic feature, the V-I track feature and the wavelet scattering feature are combined for identification, the features of the load data can be excavated from multiple angles and fully utilized, and the precision of load identification is improved; according to the method, the wavelet scattering network is constructed, the multi-scale analysis capability and the non-linear approximation capability are achieved, complex signals can be better processed, and the multi-scale features of the signals are effectively extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of load identification, and in particular to a load identification method and system. Background Art

[0002] In order to address the problems of overuse of fossil energy and deterioration of the global climate environment, load monitoring technology with load identification as the core realizes the acquisition of information at the user internal equipment level by monitoring and analyzing user power consumption data, and provides suggestions for energy conservation and demand response.

[0003] The core of load identification lies in extracting highly representative features from power consumption data. The commonly used features in existing methods are mainly divided into the following categories: (1) Power features, such as active power, reactive power, and power factor, which reflect the most intuitive characteristics of the load; (2) Harmonic features, such as the effective value and phase of each harmonic of the current, which analyze the discrimination of load data from the frequency domain perspective; (3) Image-based features, such as V-I trajectories and Gram angle fields, which can reflect the dynamic change characteristics of load data; (4) Deep network features, such as features extracted by convolutional neural networks or autoencoders, which mine more representative features through deep networks.

[0004] However, the above features all have their limitations. For example, the overlap degree of power features is relatively high, the redundancy degree of harmonic features is relatively large, most image-based features lose power information due to normalization processing, and deep network features lack interpretability. Therefore, it is necessary to further consider feature extraction methods from more perspectives, and select, process, and combine them in combination with the characteristics of each feature to achieve the full utilization of data and improve the accuracy of load identification. Summary of the Invention

[0005] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the present invention provides a load identification method and system, which can solve the problems mentioned in the background art.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a load identification method, including:

[0010] Obtain the first operating parameters of the target device, and preprocess the first operating parameters;

[0011] Perform feature extraction on the preprocessed first operating parameter to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector;

[0012] Pre-train a first load identification model and use the total feature vector as the input of the first load identification model;

[0013] Perform load identification according to the output of the first identification model.

[0014] As a preferred solution of the load identification method described in the present invention, wherein: the first load identification model includes:

[0015] The first load identification model is any model with a feature vector as the input and a load identification result or relevant parameters that can directly or indirectly obtain the load identification result as the output.

[0016] As a preferred solution of the load identification method described in the present invention, wherein: the feature extraction operation includes a first feature extraction operation and a second feature extraction operation;

[0017] The first feature extraction operation is used to extract the knowledge features in the first operating parameter as the first feature vector;

[0018] The second feature extraction operation is used to extract the target scattering features in the first operating parameter as the second feature vector.

[0019] As a preferred solution of the load identification method described in the present invention, wherein: the extraction of the target scattering features in the first operating parameter includes:

[0020] Decompose the current waveform data in the first operating parameter into active current waveform and reactive current waveform;

[0021] Construct a three-layer wavelet scattering network;

[0022] Input the reactive current waveform into the wavelet scattering network to obtain wavelet scattering coefficient vectors of each layer;

[0023] Perform dimensionality reduction on the wavelet scattering coefficient vectors to obtain target scattering features.

[0024] As a preferred solution of the load identification method described in the present invention, wherein: the dimensionality reduction operation on the wavelet scattering coefficient vectors includes: performing dimensionality reduction on the wavelet scattering coefficient vectors through a first dimensionality reduction operation.

[0025] As a preferred solution of the load identification method described in the present invention, wherein: the first operating parameters at least include voltage waveform data and current waveform data during the operation of the target device.

[0026] As a preferred solution of the load identification method described in the present invention, wherein: the knowledge features at least include power features, waveform features, harmonic features, and V-I trajectory features;

[0027] The power features include active power, reactive power, and power factor;

[0028] The waveform features include current peak-to-peak value, form factor, and peak factor;

[0029] The harmonic features include total harmonic distortion rate of current and effective values of 3rd and 5th harmonics;

[0030] The V-I trajectory features include V-I trajectory area, sum of areas of the left and right parts of the V-I trajectory, and variance of the middle section of the V-I trajectory.

[0031] In a second aspect, the present invention provides a load identification system, including:

[0032] A data processing module, configured to obtain the first operating parameters of the target device and preprocess the first operating parameters;

[0033] A feature vector extraction module, configured to perform feature extraction operations on the preprocessed first operating parameters to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector;

[0034] A model establishment module, configured to pre-train a first load identification model and use the total feature vector as the input of the first load identification model;

[0035] An identification module, configured to perform load identification according to the output of the first identification model.

[0036] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a load identification method and system, which acquires the first operating parameters of a target device and preprocesses the first operating parameters; performs feature extraction operations on the preprocessed first operating parameters to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector; pre-trains a first load identification model and uses the total feature vector as the input of the first load identification model; performs load identification according to the output of the first identification model. This application combines power features, waveform features, harmonic features, V-I trajectory features, and wavelet scattering features for identification, can excavate the features of load data from multiple perspectives and make full use of them, and improve the accuracy of load identification; this application constructs a wavelet scattering network, which has multi-scale analysis ability and nonlinear approximation ability, can better process complex signals, and effectively extract multi-scale features of signals; when calculating wavelet scattering features, this application separates reactive current information based on the fryze power theory, avoids the situation where the active current of high-power devices accounts for too large a proportion and the current waveforms are too similar, and enhances the feature discrimination. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0040] Figure 1 FIG. is a flowchart of a method for a load identification method and system provided by an embodiment of the present invention;

[0041] Figure 2 FIG. is a detailed operation flowchart of a load identification method and system provided by an embodiment of the present invention;

[0042] Figure 3 FIG. is a schematic diagram for extracting V-I trajectory features of a load identification method and system provided by an embodiment of the present invention;

[0043] Figure 4 FIG. is a structural diagram of a wavelet scattering network of a load identification method and system provided by an embodiment of the present invention;

[0044] Figure 5 FIG. is an internal structure diagram of a computer device of a load identification method and system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0046] Embodiment 1

[0047] Referring to Figures 1-5 , which is the first embodiment of the present invention. This embodiment provides a load identification method and system, including:

[0048] In the existing related technologies, there are some problems. For example, the feature extraction of load data is not comprehensive enough, resulting in limited accuracy of load identification; or there is redundancy between features, increasing the computational complexity and reducing the identification efficiency.

[0049] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail on how to implement this load identification method;

[0050] Figure 1 shows a method flow chart of a load identification method and system, including:

[0051] S101, obtain the first operating parameters of the target device and preprocess the first operating parameters;

[0052] In an optional embodiment, the target device may include various electrical equipment in the power system, such as motors, lighting equipment, air conditioners, etc. The preprocessing operations may include, but are not limited to, steps such as data cleaning, denoising, and normalization to improve the accuracy and efficiency of subsequent feature extraction.

[0053] In an optional embodiment, the first operating parameters may include voltage waveform data, current waveform data, and other information that can reflect the operating state of the target device. By preprocessing these parameters, a reliable data basis can be provided for subsequent feature extraction and load identification.

[0054] In an optional embodiment, the first operating parameters can be obtained in different ways. For example, they can be obtained by real-time monitoring through sensors installed on the target device, or extracted from the historical data of the power system. After obtaining the first operating parameters, these parameters need to be preprocessed to eliminate noise and outliers and improve the accuracy and reliability of the data.

[0055] It should be noted that the preprocessing steps can be selected according to the actual situation. For example, methods such as filtering, smoothing, and detrending can be used to convert the original data into a form suitable for subsequent processing.

[0056] In the embodiment of the present application, the first operating parameter at least includes voltage waveform data and current waveform data when the target device is operating.

[0057] Exemplarily, voltage and current waveform data when the load device is operating are obtained, and a time interval is also set to segment the data. In this embodiment, an intelligent socket is used to collect data from 10 common load devices including electric blankets, electric heaters, hair dryers, induction cookers, electric fans, energy-saving lamps, laptop computers, microwave ovens, rice cookers, and refrigerators, and the sampling frequency is 6.4 kHz. The high sampling frequency can capture the detailed information of the waveform data and is the basis for realizing the extraction of subsequent knowledge features and wavelet scattering.

[0058] In the embodiment of the present application, the time interval is set to 0.08 s, that is, each segment of data contains voltage and current data of 4 cycles.

[0059] It should be noted that obtaining the first operating parameter of the target device and preprocessing the first operating parameter can ensure the accuracy and reliability of subsequent feature extraction. Specifically, by preprocessing the first operating parameter of the target device, noise and outliers in the data can be eliminated, improving the accuracy and reliability of the data, thereby providing a more accurate and reliable data basis for subsequent feature extraction and load identification. In addition, the preprocessing steps can also be selected according to the actual situation of the data to adapt to different types and scales of load data, further improving the accuracy and efficiency of load identification.

[0060] S102, perform a feature extraction operation on the preprocessed first operating parameter to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector;

[0061] In the embodiment of the present application, the feature extraction operation includes a first feature extraction operation and a second feature extraction operation;

[0062] The first feature extraction operation is used to extract the knowledge features in the first operating parameter as the first feature vector;

[0063] The second feature extraction operation is used to extract the target scattering features in the first operating parameter as the second feature vector.

[0064] In the embodiment of the present application, the knowledge features at least include power features, waveform features, harmonic features, and V-I trajectory features;

[0065] The power features include active power, reactive power, and power factor;

[0066] The waveform features include the current peak-to-peak value, the form factor, and the peak factor;

[0067] The harmonic features include the total harmonic distortion rate of the current and the effective values of the 3rd and 5th harmonics;

[0068] The V-I trajectory features include the area of the V-I trajectory, the sum of the areas of the left and right parts of the V-I trajectory, and the variance of the middle section of the V-I trajectory.

[0069] In an optional embodiment, the power features, waveform features, harmonic features, and V-I trajectory features may further include other parameters that can reflect the load characteristics, such as the voltage fluctuation rate, current unbalance degree, etc. These features can describe the operating state of the load from different angles and provide more comprehensive information for load identification.

[0070] In an optional embodiment, when extracting knowledge features, existing feature extraction algorithms or models can be used, such as power factor calculation, harmonic analysis, V-I trajectory plotting, etc. These algorithms or models can be selected and optimized according to the actual situation of the load data to improve the accuracy and efficiency of feature extraction.

[0071] In the embodiment of the present application, extracting knowledge features from current waveform data includes power features, waveform features, harmonic features, and V-I trajectory features. Furthermore:

[0072] (1) Power features: including active power P, reactive power Q, power factor cos

[0073] (2) Waveform features: including the current peak-to-peak value I p-p , form factor α w and peak factor α p , and the calculation methods are as follows:

[0074] I p-p = max(I) - min(I)

[0075]

[0076] In the formula, I rms and I arv are the effective value (root mean square) and the absolute mean value of the current respectively, and n s is the number of sampling points.

[0077] (3) Harmonic features: including the total harmonic distortion rate of the current I THD , the effective value of the 3rd harmonic of the current I 3 and the effective value of the 5th harmonic of the current I 5 . Among them, the effective values of each harmonic are calculated through Fourier transform, and I THDCalculated by the following formula:

[0078]

[0079] In the formula, I h represents the effective value of the h-th harmonic (the fundamental effective value when h = 1).

[0080] (4) V-I trajectory characteristics: including the V-I trajectory area, the sum of the areas of the left and right parts of the V-I trajectory, and the variance of the middle section of the V-I trajectory. The specific calculation methods are as follows:

[0081] In an optional embodiment, as Figure 3 shown, according to the maximum and minimum values of the voltage, the V-I trajectory is divided into two parts, A and B, denoted as set TA and set TB. Then the V-I trajectory area σ a is calculated as follows:

[0082]

[0083] In the formula, M(u M , i M ) and N(u N , i N ) are two consecutive points on the V-I trajectory of part A. M'(u M' , i M' ) and N'(u N' , i N' ) respectively represent the points on part B where the voltage is closest to points M and N, that is:

[0084]

[0085] In an optional embodiment, find the turning point K A of the V-I trajectory in parts A and B B :

[0086]

[0087] In the formula, J A , K A , L A are three consecutive points on the V-I trajectory of part A, and J B , K B , L B are three consecutive points on the V-I trajectory of part B. Divide the left, middle, and right three parts according to the following criteria, denoted as set T L , T M , T N :

[0088]

[0089] In an alternative embodiment, the sum of the areas of the left and right parts of the V-I trajectory, σ, can be calculated based on three parts lr and the variance var(T of the middle section of the V-I trajectory M ). σ lr The calculation is as follows:

[0090]

[0091] It should be noted that the above feature extraction operations can comprehensively and accurately reflect the operating state of the target device, providing reliable feature information for subsequent load identification. Specifically, the power feature can reflect the energy consumption and power factor of the load; the waveform feature can describe the fluctuations and shapes of the current; the harmonic feature can reveal the degree of harmonic pollution generated by the load; and the V-I trajectory feature can reflect the characteristics and behavior patterns of the load from the perspective of the voltage-current relationship. These features together form the basis of load identification, providing strong support for subsequent model training and identification.

[0092] In the embodiments of the present application, extracting the target scattering feature from the first operating parameter includes:

[0093] Decompose the current waveform data in the first operating parameter into active current waveform and reactive current waveform;

[0094] Construct a three-layer wavelet scattering network;

[0095] Input the reactive current waveform into the wavelet scattering network to obtain wavelet scattering coefficient vectors of each layer;

[0096] Perform a dimensionality reduction operation on the wavelet scattering coefficient vectors to obtain the target scattering feature.

[0097] In an alternative embodiment, the dimensionality reduction operation on the wavelet scattering coefficient vectors can be implemented by algorithms such as principal component analysis (PCA) or linear discriminant analysis (LDA), etc., to reduce the feature dimension, lower the computational complexity, and retain key information at the same time. Through the dimensionality reduction operation, more concise and effective target scattering features can be obtained, providing more efficient and accurate feature information for subsequent load identification.

[0098] In the embodiments of the present application, the dimensionality reduction operation on the wavelet scattering coefficient vectors includes: performing a dimensionality reduction operation on the wavelet scattering coefficient vectors through a first dimensionality reduction operation.

[0099] In the embodiments of the present application, the first dimensionality reduction operation is dimensionality reduction by principal component analysis.

[0100] In the embodiments of the present application, based on the fryze power theory, the current waveform data is decomposed into an active current waveform and a reactive current waveform. The specific content of the fryze power theory is as follows: The line current i(t) can be decomposed into an active current a(t) and a reactive current f(t) according to the line voltage u(t). Among them, the information richness of the active current waveform is relatively low, only reflecting the resistance information of the load; the information richness of the reactive current is relatively high, reflecting the non-resistance information of the load. For some high-power devices, the proportion of the active current is too large, and it is easy to have a situation where the current waveforms are highly similar. Separating the reactive current information is beneficial to avoiding the above problems and enhancing the feature distinguishability. Its calculation method is as follows:

[0101]

[0102] f(t) = i(t) - a(t)

[0103] In the formula, u(t) and i(t) are the sampling values of the voltage and current at time t respectively, a(t) and f(t) are the magnitudes of the active current and reactive current at time t respectively, U rms is the effective value (root mean square) of the voltage data of this section, and P is the average power of this section of data.

[0104] In an optional embodiment, a three-layer wavelet scattering network is constructed. The reactive current waveform is input into the wavelet scattering network, and the wavelet scattering coefficients of each layer are obtained and dimensionally reduced through principal component analysis, denoted as: wavelet scattering features.

[0105] The wavelet scattering network is a deep network structure based on wavelet transform. It combines the multi-scale time-frequency analysis ability of wavelet transform with the hierarchical structure of deep neural networks, providing effective features for complex pattern recognition and prediction tasks. The three-layer wavelet scattering network constructed in this embodiment is as Figure 4 shown. The calculation process of the reactive current waveform in the wavelet scattering network is as follows:

[0106] (1) Perform a convolution operation of the scale function on the reactive current waveform f to obtain the low-frequency information of the signal, and obtain the first-layer scattering wavelet coefficient S 0 f:

[0107] S 0 f = f * φ J

[0108] In the formula, * is the convolution operation, and φ J is the scale function that is essentially a low-pass filter. In this embodiment, a Gaussian low-pass filter with a scale of 2J is used.

[0109] (2) Multiply the reactive current waveform f by the first-layer complex wavelet ψ ξ1Convolution, corresponding to the convolution operation of a convolutional neural network.

[0110] In this embodiment, the Morlet wavelet is selected as the wavelet function. Subsequently, the modulus of the complex signal after convolution is taken, corresponding to the non-linear transformation operation of the convolutional neural network, to obtain the first-layer wavelet modulus coefficient U 1 f:

[0111]

[0112] (3) Convolve the above U 1 f with the scaling function φ J Convolution, corresponding to the averaging operation of the convolutional neural network, to obtain the second-layer wavelet scattering coefficient output S 1 f:

[0113]

[0114] (4) Convolve U 1 f with the second-layer complex wavelet and take the modulus to obtain the second-layer wavelet modulus coefficient U 2 f. Convolve U 2 f with the scaling function φ J to obtain the third-layer scattering wavelet coefficient S 2 f:

[0115]

[0116]

[0117] (5) Concatenate S 0 f, S 1 f, and S 2 f to obtain the wavelet scattering coefficient vector S.

[0118] It should be noted that extracting features from the preprocessed first operating parameters to obtain the total feature vector, where the total feature vector includes the first feature vector and the second feature vector, can more comprehensively reflect the operating state and characteristics of the target device, providing richer and more accurate information for subsequent load identification. Specifically, the knowledge features included in the first feature vector, such as power features, waveform features, harmonic features, and V-I trajectory features, can describe the energy consumption, current fluctuations, harmonic pollution degree, and voltage-current relationship of the load from different perspectives, providing basic feature information for load identification. The target scattering features included in the second feature vector are obtained by performing deep feature extraction on the reactive current waveform through a wavelet scattering network, which can further reveal the non-resistive information and complex patterns of the load, enhancing the feature discrimination. Combining these two feature vectors can form a more complete and effective feature set, providing more reliable and efficient feature support for subsequent model training and identification. In addition, this feature extraction method can also adapt to different types and scales of load data, improving the accuracy and generalization ability of load identification.

[0119] S103, pre-train the first load identification model, and use the total feature vector as the input of the first load identification model;

[0120] In the embodiment of the present application, the first load identification model includes:

[0121] The first load identification model is any model with a feature vector as the input and a load identification result or relevant parameters that can directly or indirectly obtain the load identification result as the output.

[0122] In an alternative embodiment, the first load identification model can use a deep learning model, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM), etc. These deep learning models have powerful feature learning and pattern recognition capabilities, and can extract key information from complex feature vectors to achieve accurate load identification.

[0123] In specific implementation, an appropriate deep learning model can be selected according to the characteristics of the load data and the identification requirements, and the model can be pre-trained with a large amount of training data to improve its identification accuracy and generalization ability. During the pre-training process, techniques such as cross-validation, regularization, and dropout can be used to prevent the model from overfitting and ensure the stability and reliability of the model. Through pre-training, the first load identification model can learn the internal laws and features of the load data, providing strong support for subsequent practical applications.

[0124] In an alternative embodiment, the first load identification model can also be constructed using ensemble learning methods, such as random forest, Gradient Boosting Decision Trees (GBDT), etc. Ensemble learning methods improve the identification accuracy and generalization ability of the overall model by combining the prediction results of multiple weak classifiers. In specific implementation, different base classifiers can be selected, the ensemble strategy can be adjusted, and the ensemble model can be trained and optimized using training data to obtain better load identification results.

[0125] In an alternative embodiment, the first load identification model can also integrate the advantages of traditional machine learning algorithms and deep learning algorithms to construct a hybrid model. Specifically, traditional machine learning algorithms, such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), etc., can be used to initially classify or select features from the total feature vector to remove redundant information and retain key features. Then, the processed features are input into a deep learning model for further feature learning and pattern recognition. This hybrid model can make full use of the advantages of different algorithms to improve the accuracy and efficiency of load identification.

[0126] In the embodiment of the present application, the obtained knowledge features and the obtained wavelet scattering features are concatenated to form a total feature vector, which is input into the trained extremely randomized trees model (i.e., the first load identification model) to obtain the load type identification result.

[0127] Extra Trees is a variant of random forest that improves the generalization ability of the model by constructing a large number of random decision trees. Different from traditional random forest, Extra Trees uses all training samples when constructing each tree and randomly selects features and splitting thresholds during the splitting process of each node, thus introducing more randomness. This method not only accelerates the training process but also effectively prevents overfitting and performs well in various tasks. Its construction method is as follows:

[0128] (1) Generate a single decision tree: Randomly select a subset of features. For each feature in the feature subset, randomly select a splitting threshold and calculate the splitting score. Select the feature with the highest splitting score as the splitting feature for this node. Subsequently, continue to split the split subtree in the above manner until the stopping condition is reached. In this embodiment, the information gain is used as the splitting score, and the calculation method is as follows:

[0129]

[0130] where IG(D,A) is the information gain of feature A for training set D, Entropy(D) is the entropy of training set D, Values(A) is all the values of feature A, D vis the data subset when the value of feature A is v, which are subsets D v and the number of samples in the training set D, p j is the proportion of samples in the j-th class, n C is the total number of classes.

[0131] (2) Construct multiple decision trees according to the above method, and obtain the final identification result by means of majority voting for the prediction results of each tree.

[0132] In an optional embodiment, the relevant parameters that can directly or indirectly obtain the load identification result may include the power factor of the load, current waveform parameters, harmonic content, and specific statistics of the voltage-current trajectory, etc. These parameters can directly reflect the operating state and characteristics of the load, and provide important reference information for load identification. For example, the power factor can reflect the energy consumption efficiency and power quality of the load; the current waveform parameters can describe the fluctuation and morphological characteristics of the current; the harmonic content can reveal the degree of harmonic pollution and spectral characteristics generated by the load; and the specific statistics of the voltage-current trajectory can grasp the voltage-current relationship of the load as a whole, and further reveal the dynamic behavior and response characteristics of the load.

[0133] It should be noted that the design architecture of the final first load identification model can be selected according to specific output requirements. This application does not limit the output of the first load identification model, as long as the first load identification model can output the load identification result or the relevant parameters that can directly or indirectly obtain the load identification result based on the input total feature vector. In this way, the extracted knowledge features and target scattering features can be fully utilized to achieve accurate load identification, providing strong support for subsequent equipment management and operation optimization.

[0134] It should also be noted that pre-training the first load identification model and using the total feature vector as the input of the first load identification model can enable the first load identification model to have certain load identification capabilities and generalization capabilities before formal application. Since the total feature vector contains rich knowledge features and target scattering features, which can comprehensively and accurately reflect the operating state and characteristics of the target device, using the total feature vector as the input can enable the first load identification model to learn more in-depth and detailed load data laws during the training process. In this way, in actual application, the first load identification model can complete the load identification task more quickly and accurately, improving the identification accuracy and efficiency. At the same time, the pre-training process can also help the model better adapt to different types and scales of load data, further enhancing the generalization ability and robustness of the model.

[0135] S104, perform load identification according to the output of the first identification model.

[0136] In the embodiments of this application, such asFigure 2 For detailed implementable operation steps, where:

[0137] Start: The starting point of the process.

[0138] Obtain the voltage and current waveform data during the operation of the load device, and set a time interval to segment the data: First, collect the voltage and current waveform data generated by the load device during operation, and segment these data according to the set time interval for subsequent analysis.

[0139] Decompose the current waveform data into active current waveform and reactive current waveform based on Fryze power theory: Utilize Fryze power theory to decompose the current waveform data into two parts, namely the active current waveform and the reactive current waveform.

[0140] Extract knowledge features from the current waveform data: Extract useful knowledge features from the decomposed current waveform data, including power features, waveform features, harmonic features, and V-I trajectory features, etc.

[0141] Input the reactive current waveform into the wavelet scattering network to obtain wavelet scattering coefficients of each layer: Input the reactive current waveform into the wavelet scattering network, and obtain wavelet scattering coefficients of different layers through this network.

[0142] Principal component analysis for dimensionality reduction: Perform principal component analysis (PCA) on the obtained wavelet scattering coefficients to reduce the data dimension and simplify the subsequent processing process.

[0143] Wavelet scattering features: Wavelet scattering features obtained after principal component analysis.

[0144] Stitch to form a total feature vector: Stitch together all the above-extracted features (including power features, waveform features, harmonic features, V-I trajectory features, and wavelet scattering features) to form a total feature vector.

[0145] Input it into the trained load identification model to obtain the load type identification result: Input the formed total feature vector into a trained load identification model to obtain the identification result of the load type.

[0146] End: The end point of the process.

[0147] In an alternative embodiment, load identification based on the output of the first identification model can directly obtain the load type identification result, or further analyze and process the output parameters of the first identification model to obtain the load type identification result. For example, the output parameters of the first identification model can be compared with a preset threshold or range, and the load type can be determined according to the comparison result. Alternatively, the output parameters of the first identification model can be used as the input of other algorithms or models for further classification or identification.

[0148] It should be noted that in this way, the output information of the first identification model can be fully utilized to improve the accuracy and reliability of load identification. In practical applications, appropriate load identification methods and processes can be selected according to specific identification requirements and scenarios to achieve efficient and accurate load identification.

[0149] Exemplarily, when setting a first threshold and a second threshold, if the output parameter of the first identification model is greater than the first threshold and less than the second threshold, the load type can be determined as a certain specific type, such as a motor load.

[0150] If the output parameter exceeds the range of these two thresholds, other identification methods or models can be further used for subdivision and confirmation.

[0151] This threshold-based determination method is simple and direct, and can quickly give a preliminary load type identification result, which is suitable for situations with high requirements for identification speed and real-time performance.

[0152] In an alternative embodiment, the result of load identification can be output in the form of a load type label or load characteristic parameters. Specifically, the load type label can be a predefined load classification, such as a motor load, a lighting load, a resistive load, etc., and these labels can directly reflect the electrical characteristics and operating modes of the load. The load characteristic parameters can be specific values describing the dynamic behavior and response characteristics of the load, such as power factor, current volatility, harmonic distortion rate, etc., and these parameters can provide more detailed and specific reference information for subsequent equipment management and operation optimization.

[0153] In summary, the present invention proposes a load identification method, which obtains the first operating parameter of the target device and preprocesses the first operating parameter; performs feature extraction operation on the preprocessed first operating parameter to obtain a total feature vector, which includes a first feature vector and a second feature vector; pretrains a first load identification model, and uses the total feature vector as the input of the first load identification model; and performs load identification according to the output of the first identification model. The present application combines power characteristics, waveform characteristics, harmonic characteristics, VI trajectory characteristics, and wavelet scattering characteristics for identification, and can mine and fully utilize the characteristics of load data from multiple angles to improve the accuracy of load identification; the present application constructs a wavelet scattering network with multi-scale analysis capabilities and nonlinear approximation capabilities, which can better process complex signals and effectively extract multi-scale features of signals; the present application separates reactive current information based on the Fryze power theory when calculating wavelet scattering features, to avoid the situation where the active current of high-power equipment accounts for too large a proportion and the current waveforms are too similar, thereby enhancing feature differentiation.

[0154] Example 2

[0155] In a preferred embodiment, the dimension of the solved wavelet scattering coefficient vector is too high, which is not conducive to model identification and consumes a lot of computing resources. Therefore, the principal component analysis method is used to reduce its dimension. Principal component analysis (PCA) transforms the original coordinate system into a new orthogonal coordinate system through a set of orthogonal transformations, maps high-dimensional data to a low-dimensional space, simplifies the data structure, removes redundancy and noise, and retains the most important information, thereby reducing computational complexity and improving the performance of subsequent models. The calculation process is as follows:

[0156] (1) The wavelet scattering coefficient vectors obtained from the training samples in the above steps are organized into a matrix X by columns, and the data in each row is Z-score standardized so that the mean of each dimension is 0 and the standard deviation is 1.

[0157] (2) Calculate the covariance matrix C:

[0158]

[0159] In the formula, nb is the total number of training samples.

[0160] (3) Solve the characteristic equation λE-C|=0 to obtain the characteristic value λ j (j=1,2,…,n d , nd is the dimension of the wavelet scattering coefficient vector) and sort them from large to small to calculate the cumulative contribution rate Select the top n with cumulative contribution rates exceeding 85% p principal components, and substitute Cb=λb for each selected principal component to obtain the eigenvector b corresponding to each principal component. k (k=1,2,…,np ) and form matrix B row by row.

[0161] (4) Multiply the wavelet scattering coefficient vector to be dimension-reduced by matrix B on the left to obtain the dimension-reduced wavelet feature vector.

[0162] Embodiment 3

[0163] This embodiment also provides a load identification system, including:

[0164] A data processing module, configured to obtain the first operating parameter of the target device and preprocess the first operating parameter;

[0165] A feature vector extraction module, configured to perform feature extraction operations on the preprocessed first operating parameter to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector;

[0166] A model establishment module, configured to pre-train a first load identification model and use the total feature vector as the input of the first load identification model;

[0167] An identification module, configured to perform load identification according to the output of the first identification model.

[0168] The above-mentioned each unit module can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned each module.

[0169] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 5 shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication) or other technologies. The computer program, when executed by the processor, implements a load identification method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball or a touchpad provided on the shell of the computer device, or an external keyboard, a touchpad or a mouse, etc.

[0170] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0171] Obtain the first operating parameter of the target device, and preprocess the first operating parameter;

[0172] Perform a feature extraction operation on the preprocessed first operating parameter to obtain a total feature vector, where the total feature vector includes a first feature vector and a second feature vector;

[0173] Pretrain the first load identification model, and use the total feature vector as the input of the first load identification model;

[0174] Perform load identification according to the output of the first identification model.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0176] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0177] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0178] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the function specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0179] These computer program instructions can 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, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the function specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0180] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0181] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A load identification method, characterized in that: include: Acquire a first operating parameter of the target device, and preprocess the first operating parameter; Performing a feature extraction operation on the preprocessed first operating parameter to obtain a total feature vector, wherein the total feature vector includes a first feature vector and a second feature vector; pre-training a first load identification model, and using the total feature vector as an input of the first load identification model; Load identification is performed according to the output of the first identification model.

2. The load identification method according to claim 1, characterized in that: The first load identification model includes: The first load identification model is any model whose input is a feature vector and whose output is a load identification result or a parameter related to the load identification result that can be obtained directly or indirectly.

3. The load identification method according to claim 2, characterized in that: The feature extraction operation includes a first feature extraction operation and a second feature extraction operation; The first feature extraction operation is used to extract knowledge features in the first operating parameter as a first feature vector; The second feature extraction operation is used to extract the target scattering feature in the first operating parameter as a second feature vector.

4. The load identification method according to claim 3, characterized in that: The extracting the target scattering feature in the first operating parameter comprises: Decomposing the current waveform data in the first operating parameter into an active current waveform and a reactive current waveform; Construct a three-layer wavelet scattering network; Inputting the reactive current waveform into the wavelet scattering network to obtain the wavelet scattering coefficient vectors of each layer; A dimension reduction operation is performed on the wavelet scattering coefficient vector to obtain a target scattering feature.

5. The load identification method according to claim 4, characterized in that: The performing a dimensionality reduction operation on the wavelet scattering coefficient vector includes: performing a dimensionality reduction operation on the wavelet scattering coefficient vector through a first dimensionality reduction operation.

6. The load identification method according to claim 5, characterized in that: The first operating parameter at least includes voltage waveform data and current waveform data when the target device is operating.

7. The load identification method according to claim 6, characterized in that: The knowledge features at least include power features, waveform features, harmonic features, and VI trajectory features; The power characteristics include active power, reactive power, and power factor; The waveform characteristics include current peak-to-peak value, waveform factor and crest factor; The harmonic characteristics include the total harmonic distortion rate of current and the effective values ​​of the third and fifth harmonics; The VI track features include the VI track area, the sum of the areas of the left and right parts of the VI track, and the variance of the middle section of the VI track.

8. A load identification system, characterized in that: include: A data processing module, used for acquiring a first operating parameter of a target device and preprocessing the first operating parameter; A feature vector extraction module, used for performing a feature extraction operation on the preprocessed first operating parameter to obtain a total feature vector, wherein the total feature vector includes a first feature vector and a second feature vector; A model building module, used for pre-training a first load identification model, and taking the total feature vector as an input of the first load identification model; An identification module is used to perform load identification according to the output of the first identification model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images