Non-invasive electricity load identification method and system, circuit breaker and medium

Abstract

The invention provides a non-invasive electricity load identification method and system, a circuit breaker and a medium, and belongs to the technical field of load identification, and the method comprises the steps: firstly, obtaining a total power signal of an electricity environment, carrying out the feature extraction of the total power signal, obtaining a feature signal of the total power signal, and then, carrying out the feature extraction of the feature signal; and respectively inputting the characteristic signal into a plurality of load identification models for identifying different load types, respectively identifying the respective corresponding load type through each load identification model, and outputting the prediction confidence corresponding to each load type. And finally, obtaining a load identification result according to the prediction confidence corresponding to each load type. Therefore, the independent load identification model is adopted to identify each load type, so that the accuracy of identifying each power utilization load under the scene of simultaneous operation of multiple loads is improved.

Description

Technical Field

[0001] The present application relates to the technical field of load identification, and specifically relates to a non-intrusive method, system, circuit breaker and medium for identifying electrical loads. Background Art

[0002] Non-Intrusive Load Monitoring (NILM) is a method of identifying the types of electrical devices and monitoring their states by collecting the total voltage and current signals at the power supply inlet (such as a smart meter) and combining feature extraction and pattern recognition techniques. Compared with the intrusive solution that requires installing sensors for each device, non-intrusive load identification has significant advantages such as low deployment cost and convenient maintenance, and thus has been widely used in the fields of smart home, industrial energy conservation, and power grid demand-side management.

[0003] However, in the scenario of multiple loads running simultaneously, such as air conditioners, lighting, and household appliances running in parallel in a home environment, interference will occur due to the superposition of the total voltage or current signal waveforms. In addition, the waveform superposition caused by multiple loads running simultaneously will also lead to blurred features, resulting in a significant decrease in the recognition rate of features used for load identification, such as steady-state features (effective value, harmonics, etc.).

[0004] Therefore, how to improve the accuracy of identifying each electrical load in the scenario of multiple loads running simultaneously has been a research topic that the field has been committed to. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present application provides a non-intrusive method, system, circuit breaker and medium for identifying electrical loads.

[0006] In a first aspect, a non-intrusive method for identifying electrical loads provided by the present application includes:

[0007] Obtain the total power supply signal of the electrical environment;

[0008] Extract features from the total power supply signal to obtain the feature signal of the total power supply signal;

[0009] Input the feature signal into multiple load identification models for identifying different load types respectively, and identify the corresponding load types through each load identification model, and output the prediction confidence levels corresponding to each load type;

[0010] Obtain the load identification result according to the prediction confidence levels corresponding to each load type.

[0011] Optionally, the feature signal includes time-domain features, frequency-domain features, and time-frequency domain features;

[0012] The extracting features of the total power signal to obtain a feature signal includes:

[0013] Performing statistical and morphological processing on the total power signal to obtain the time domain feature;

[0014] Performing Fourier transform processing on the total power supply signal to obtain the frequency domain feature;

[0015] The total power supply signal is subjected to wavelet transform processing to obtain the time-frequency domain features.

[0016] Optionally, the time domain features include morphological features and statistical features, wherein the morphological features include at least one of the kurtosis, skewness, kurtosis and waveform factor of the total power signal; the statistical features include at least one of the mean, variance, peak value and zero-crossing rate of the total power signal;

[0017] The frequency domain feature includes at least one of a fundamental wave, a harmonic wave, a power spectrum density, and a spectrum entropy of the total power signal;

[0018] The time-frequency domain feature includes at least one of a high-frequency wavelet coefficient and a low-frequency wavelet coefficient.

[0019] Optionally, the method further includes training a preset prediction network to obtain a training step of each of the load identification models, wherein the training step includes:

[0020] Creating training samples for each of the load identification models respectively, and using the load type corresponding to the training sample as a training label; wherein the training sample includes at least one of the time domain feature, the frequency domain feature, and the time-frequency domain feature;

[0021] Inputting each of the training samples into each of the preset prediction networks respectively to obtain the prediction confidence of each of the load types;

[0022] According to the prediction confidence of the load type and the corresponding training label, each preset prediction network is iteratively updated to obtain a plurality of load identification models for identifying different load types.

[0023] Optionally, the preset prediction network includes one or more combinations of a lightweight gradient boosting machine network, an extreme gradient boosting network, a shallow convolutional neural network, and a recurrent neural network.

[0024] Optionally, the multiple load identification models are parallel structures, and the load identification result is obtained according to the prediction confidence corresponding to each load type, including:

[0025] Determine a combination of target load types from each of the load types according to a first preset threshold and each of the prediction confidence levels;

[0026] Query the weight value of each of the prediction confidence levels from a weight assignment table according to the combination of the target load types; wherein, the weight assignment table includes the corresponding relationship between the combination of the target load types and the weight values of each of the prediction confidence levels;

[0027] Adjust each of the prediction confidence levels according to the weight value to obtain an optimized confidence level;

[0028] Compare the optimized confidence level with a second preset threshold, and obtain the load identification result according to the comparison result;

[0029] Wherein, the load identification result is the load type for which the optimized confidence level is greater than the second preset threshold.

[0030] Optionally, multiple of the load identification models are in a master-slave architecture, and the multiple load identification models include a master identification model and several slave identification models;

[0031] The obtaining the load identification result according to each of the prediction confidence levels corresponding to the load types includes:

[0032] Correct the prediction confidence levels of several of the slave identification models according to the prediction confidence level of the master identification model and a third preset threshold to obtain the corrected confidence levels of each of the slave identification models;

[0033] Perform a fusion adjustment on the corrected confidence levels and the prediction confidence level of the master identification model to respectively obtain the optimized confidence levels of the master identification model and the slave identification models;

[0034] Compare the optimized confidence levels with a second preset threshold, and obtain the load identification result according to the comparison result.

[0035] Optionally, the correcting the prediction confidence levels of several of the slave identification models according to the prediction confidence level of the master identification model and a third preset threshold to obtain the corrected confidence levels of each of the slave identification models includes:

[0036] If the prediction confidence level of the master identification model is greater than or equal to the third preset threshold, reduce the prediction confidence levels of the slave identification models to obtain the corrected confidence levels of each of the slave identification models;

[0037] If the prediction confidence level of the master identification model is less than the third preset threshold, maintain or increase the prediction confidence levels of the slave identification models to obtain the corrected confidence levels of each of the slave identification models.

[0038] Second aspect, in one embodiment, the present application provides a non-invasive electrical load identification system, including:

[0039] A signal acquisition module, configured to acquire the total power supply signal of the electrical environment;

[0040] A feature extraction module, configured to extract features from the total power supply signal to obtain the feature signal of the total power supply signal;

[0041] An independent model prediction module, configured to respectively input the feature signal into multiple load identification models for identifying different load types, identify the respective corresponding load types through each load identification model, and output the prediction confidence levels corresponding to each load type;

[0042] A result fusion module, configured to obtain the load identification result according to the prediction confidence levels corresponding to each load type.

[0043] Third aspect, in one embodiment, the present application provides a circuit breaker, including a memory and a processor; the memory stores a computer program, and the processor is configured to run the computer program in the memory to execute the steps in the non-invasive electrical load identification method in any of the above embodiments.

[0044] Fourth aspect, in one embodiment, the present application provides a storage medium, the storage medium stores a computer program, and the computer program is loaded by a processor to execute the steps in the non-invasive electrical load identification method in any of the above embodiments.

[0045] In summary, in the present application, first, the total power supply signal of the electrical environment is acquired, and features are extracted from the acquired total power supply signal to obtain the feature signal of the total power supply signal, so as to obtain the key information in the total power supply signal, which helps to more accurately identify the load type subsequently. Then, the feature signal is respectively input into multiple load identification models for identifying different load types. Each load identification model focuses on identifying a specific load type to respectively identify the respective corresponding load types and output the prediction confidence levels corresponding to each load type. The prediction confidence level reflects the reliability of the model for the identification result. Finally, the load identification result is obtained according to the prediction confidence levels corresponding to each load type. In this way, independent load identification models are used to respectively identify each load type, thereby improving the accuracy of identifying each electrical load in a scenario where multiple loads are running simultaneously. Description of the Drawings

[0046] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0047] Figure 1 Schematic diagram of the application scenario of the non-intrusive electrical load identification method in an embodiment of the present application;

[0048] Figure 2 Flowchart of the non-intrusive electrical load identification method in an embodiment of the present application;

[0049] Figure 3 Flowchart of the method for training the load identification model in an embodiment of the present application;

[0050] Figure 4 Flowchart of the method for generating the load identification result of the parallel architecture in an embodiment of the present application;

[0051] Figure 5 Schematic diagram of the load identification model of the parallel architecture in an embodiment of the present application;

[0052] Figure 6 Flowchart of the method for generating the load identification result of the master-slave architecture in an embodiment of the present application;

[0053] Figure 7 Schematic diagram of the load identification model of the master-slave architecture in an embodiment of the present application;

[0054] Figure 8 Block diagram of the non-intrusive electrical load identification system in an embodiment of the present application;

[0055] Figure 9 Schematic diagram of a circuit breaker in an embodiment of the present application. Detailed implementation manners

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0057] In the description of the present application, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present application, the meaning of "a plurality" is two or more, unless otherwise specifically defined. In the present application, the term "exemplary" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "exemplary" in the present application is not necessarily to be construed as more preferred or advantageous than other embodiments. In order for any person skilled in the art to implement and use the present application, the following description is provided. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present application can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present application with unnecessary details. Therefore, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed in the present application.

[0058] As Figure 1 shown, Figure 1 FIG. is a schematic diagram of an application scenario of a non-intrusive electrical load identification method according to an embodiment of the present application. The application scenario of the non-intrusive electrical load identification method according to an embodiment of the present application includes a circuit breaker 100, in which a non-intrusive electrical load identification system is integrated, and a computer-readable storage medium in which a non-intrusive electrical load identification method corresponding to the non-intrusive electrical load identification method runs to execute the steps of the non-intrusive electrical load identification method.

[0059] It can be understood that Figure 1 the circuit breaker in the application scenario of the non-intrusive electrical load identification method shown, or the devices included in the circuit breaker do not constitute a limitation to the embodiments of the present application. That is, the number of devices, the types of devices included in the application scenario of the non-intrusive electrical load identification method, or the number of devices and the types of devices included in each device do not affect the overall implementation of the technical solution in the embodiments of the present application, and can all be regarded as equivalent replacements or derivatives of the technical solution required to be protected in the embodiments of the present application.

[0060] Those skilled in the art can understand that Figure 1 the application scenario shown in FIG. is only an application scenario corresponding to the technical solution of the present application, and does not constitute a limitation to the application scenario of the technical solution of the present application. The circuit breaker 100 may further include a memory for storing information related to the non-intrusive electrical load identification method.

[0061] In addition, in the application scenario of the non-intrusive electrical load identification method in the embodiments of the present application, a display device may be provided in the circuit breaker 100, or the circuit breaker 100 may not be provided with a display device and is communicatively connected to an external display device 200. The display device 200 is used to output the result of the execution of the non-intrusive electrical load identification method in the circuit breaker. The circuit breaker 100 can access the background database 300. The background database 300 can be the local memory of the circuit breaker 100, and the background database 300 can also be set in the cloud. The background database 300 stores information related to the non-intrusive electrical load identification method.

[0062] It should be noted that Figure 1 The application scenario of the non-intrusive electrical load identification method shown is only an example. The application scenario of the non-intrusive electrical load identification method described in the embodiments of the present application is for more clearly explaining the technical solution of the embodiments of the present application, and does not constitute a limitation on the technical solution provided by the embodiments of the present application.

[0063] Based on the above application scenario of the non-intrusive electrical load identification method, embodiments of the non-intrusive electrical load identification method are proposed.

[0064] In the first aspect, as Figure 2 shown, in one embodiment, the present application provides a non-intrusive electrical load identification method. The non-intrusive electrical load identification method includes steps S101 - S104, which will be introduced in detail below.

[0065] Step S101: Obtain the total power supply signal of the electrical environment.

[0066] As an example, the electrical environment can be various scenarios with electrical loads, including but not limited to home environments, industrial environments, and commercial environments, etc. The total power supply signal is used to provide energy for each electrical device in the electrical environment.

[0067] For example, in a home environment, the total power supply signal can come from the power supply line at the household distribution box and contains the comprehensive electrical information of the electrical energy consumed by all the electrical loads running in the home. For example, electrical loads such as lighting devices, TVs, refrigerators, and air conditioners in the home will all affect the total power supply signal when running. For another example, in an industrial environment, the total power supply signal needs to provide electrical energy for large industrial electrical loads, such as various machine tools, motors, heating devices, etc., and the electrical information it carries reflects the overall power consumption of these industrial loads.

[0068] As an example, the total power supply signal includes a current signal and a voltage signal. To obtain the total power supply signal, specific electrical measurement equipment is required. In actual operation, commonly used equipment includes voltage transformers and current transformers, which collect the current signal and voltage signal in a non-invasive manner to obtain the total power supply signal. A voltage transformer can convert a high voltage into a low voltage signal that can be measured and processed according to a certain ratio. For example, in a high-voltage transmission line, the voltage may reach several thousand volts or even higher, and the voltage transformer can convert it into a suitable low voltage, such as 100V or 220V, etc. A current transformer can convert a large current into a small current in proportion, such as converting a current of several hundred amperes into a small current signal of 5A or 1A, improving the safety of measurement.

[0069] Step S102: Extract the features of the total power supply signal to obtain the feature signal of the total power supply signal.

[0070] Among them, the feature signal includes time-domain features, frequency-domain features, and time-frequency domain features.

[0071] As an example, before step S102, the total power supply signal can be first subjected to filtering, denoising, and normalization processing to eliminate noise and interference and enhance the processability of the total power supply signal.

[0072] Step S103: Input the feature signal into multiple load recognition models for identifying different load types respectively, and each load recognition model identifies its corresponding load type respectively and outputs the prediction confidence corresponding to each load type.

[0073] As an example, the load recognition model is a model that has been pre-trained for identifying electrical loads of specific load types. For example, in a household electricity environment, the load types can include air conditioner types, refrigerator types, and light bulb types. Therefore, for electrical loads of different load types, load recognition models for identifying air conditioner loads, load recognition models for identifying refrigerator loads, and load recognition models for identifying light bulb loads can be set respectively. Each load recognition model can be constructed based on different algorithms or the same algorithm.

[0074] As an example, the prediction confidence is used to characterize the probability that the input feature signal belongs to its corresponding load type, and the value of the prediction confidence usually ranges from 0 to 1. The prediction confidence output by each load recognition model can reflect the degree of belief of the model that the current input feature signal belongs to its corresponding load type.

[0075] Step S104: Obtain the load recognition result according to the prediction confidence corresponding to each load type.

[0076] In the above embodiments, first, the total power supply signal of the power consumption environment is obtained, and feature extraction is performed on the obtained total power supply signal to obtain the feature signal of the total power supply signal, so as to obtain the key information in the total power supply signal, which helps to more accurately identify the load type subsequently. Then, the feature signals are respectively input into a plurality of load recognition models for identifying different load types. Each load recognition model focuses on identifying a specific load type to respectively identify the corresponding load types and output the prediction confidence levels corresponding to the respective load types. The prediction confidence level reflects the reliability of the model for the recognition result. Finally, the load recognition result is obtained according to the prediction confidence levels corresponding to the respective load types. In this way, independent load recognition models are used to respectively identify each load type, thereby improving the accuracy of identifying each power consumption load in the scenario where multiple loads are operating simultaneously.

[0077] In some embodiments, step S102 may include steps S1021 - S2023, which will be introduced in detail below.

[0078] Step S1021: Perform statistical and morphological processing on the total power supply signal to obtain time-domain features.

[0079] Among them, the time-domain features include morphological features and statistical features. Among them, the morphological features include at least one of the kurtosis, skewness, kurtosis coefficient, and waveform factor of the total power supply signal. The statistical features include at least one of the mean value, variance, peak value, and zero-crossing rate of the total power supply signal.

[0080] As an example, the mean value can reflect the average level of the total power supply signal within a certain time range. For some electrical load combinations that operate continuously and stably with relatively small power fluctuations, the mean value of their total power supply signal has a certain representativeness. For example, multiple lighting fixtures with the same power in a relatively simple electrical usage scenario. Since the power of the lighting fixtures is relatively stable, the mean value of the total power supply signal will also remain relatively stable over a period of time. The variance reflects the degree of dispersion of the total power supply signal relative to the mean value. Different types of electrical loads have different effects on the variance of the total power supply signal due to differences in their operating modes. Taking an electrical usage environment containing intermittent operating loads as an example, such as an office environment with normally operating computers and printers, and at the same time there is a water dispenser with an automatic energy-saving mode. When the water dispenser is in the heating state, the current is large, and when it enters the insulation state, the current decreases. This intermittent operating mode will increase the degree of dispersion of the total power supply signal at different time points, resulting in an increase in the variance. The peak value is the maximum value reached by the total power supply signal within a certain time, and it can be used to distinguish electrical loads with a large starting current or instantaneous large current during operation from ordinary stably operating loads. For example, in an industrial workshop environment, there is a large motor. When the motor starts, a large starting current will be generated, and the starting current will cause a relatively high peak value in the total power supply signal. For some small electrical loads with stable power, such as ordinary LED indicators, almost no high current peak will be generated during normal operation. The zero-crossing rate represents the number of times the total power supply signal crosses the zero point, and it can be used to make a preliminary distinction between electrical loads with different operating characteristics (such as AC loads and DC loads). For example, for some periodic electrical loads, within one cycle, the total power supply signal will alternately be positive and negative regularly.

[0081] In this way, through statistical characteristics, the total power supply signal can be described from a macroscopic perspective, capturing the operating characteristics of electrical loads in the time dimension, and providing basic information for more accurate subsequent load identification.

[0082] As an example, kurtosis is a statistic that describes the distribution shape of the total power signal, which reflects the tail thickness of the signal probability density function. Taking an electronic device with a complex filtering circuit as an example, the current waveform of such an electronic device may have spikes at certain moments, resulting in an increase in the kurtosis of the total power signal. Skewness measures the asymmetry of the signal distribution. For some unidirectionally conducting electrical loads, such as a diode rectifier circuit, since the current can only flow in one direction, the total power signal will present an asymmetric distribution feature in the time dimension, so that the skewness of the corresponding total power signal will present a specific value. Kurtosis is a normalized fourth-order moment of signal distribution, which is sensitive to spikes and heavy tails in the signal. For example, in some electrical loads with pulsed current demand, such as pulsed lasers, the total power signal will frequently have spikes, which will significantly increase the kurtosis value of the total power signal. The form factor is the ratio of the effective value of the signal to the average value. Different types of electrical loads have different form factors due to their different electrical characteristics. For example, for a purely resistive load, the current and voltage are basically in phase, and the waveform factor is relatively stable. However, for a load containing inductance or capacitance, the waveform factor will change with the operating state of the load because the inductance and capacitance affect the phase of the current.

[0083] In this way, through morphological features, the characteristics of the total power signal in the time domain can be described more carefully, which can help distinguish power loads with special electrical characteristics and improve the accuracy of load identification.

[0084] Step S1022: Perform Fourier transform processing on the total power signal to obtain frequency domain characteristics.

[0085] The frequency domain features include at least one of the fundamental wave, harmonics, power spectrum density and spectrum entropy of the total power signal.

[0086] As an example, the Fourier transform can convert the total power supply signal in the time domain into a frequency domain signal, thereby revealing the different frequency components contained in the signal. The fundamental wave is the main frequency component in the total power supply signal and is related to the frequency of the power supply. For example, in a power supply system with a mains frequency of 50 Hz, the fundamental wave frequency is 50 Hz. Different electrical loads have different effects on the fundamental wave. Harmonics refer to components with frequencies that are integer multiples of the fundamental wave frequency. For example, a non-linear load such as a fluorescent lamp with an electronic ballast will generate 3rd harmonics (150 Hz), 5th harmonics (250 Hz), etc. Different types of non-linear loads will generate harmonics of different orders and amplitudes. By analyzing the harmonic components, non-linear loads and linear loads can be effectively distinguished. The power spectral density reflects the power distribution of the signal at different frequencies. For some electrical loads with complex frequency characteristics, such as some air conditioning equipment using frequency conversion technology, its power spectral density will have different distributions at different operating frequencies. For example, in the low-frequency operating state of the air conditioner, the energy of its power spectral density accounts for a relatively large proportion in the low-frequency band; while in the high-frequency operating state, the energy of the power spectral density in the high-frequency band will increase. By analyzing the power spectral density, the power consumption of the electrical load at different frequencies can be analyzed, thereby further distinguishing different types of electrical loads. The spectral entropy is an index that measures the complexity of the signal spectrum. Due to the different internal circuit structures and operating modes of different electrical loads, the spectral entropy of their total power supply signals will also vary. For example, a load powered by a simple DC power supply has a relatively simple spectrum and a low spectral entropy; while a complex electronic device, such as a device with multiple sub-circuits and frequency conversion functions, has a more complex spectrum and a high spectral entropy. The spectral entropy can quantitatively describe the complexity of the spectrum of the electrical load, which helps to distinguish electrical loads of different complexities.

[0087] Step S1023: Perform wavelet transform processing on the total power supply signal to obtain time-frequency domain features.

[0088] Among them, the time-frequency domain features include at least one of high-frequency wavelet coefficients and low-frequency wavelet coefficients.

[0089] As an example, wavelet transform can analyze the signal in both the time and frequency dimensions simultaneously, which helps to analyze non-stationary signals. High-frequency wavelet coefficients reflect the characteristics of the signal in the high-frequency part, and the characteristics of the high-frequency part are usually related to the rapidly changing part of the signal. For example, for some electrical loads with rapid switching actions, such as loads controlled by relay switches in electronic devices, when the relay closes or opens, it will generate rapid changes in the total power supply signal, and the rapid changes will be reflected in the high-frequency wavelet coefficients. The low-frequency wavelet coefficients are related to the slowly changing part of the signal. For some loads that start slowly or operate relatively stably, such as large industrial motors, the current gradually increases during the startup process, and the slow change process will be reflected in the low-frequency wavelet coefficients.

[0090] Thus, by extracting time-frequency domain features, the transient and dynamic behaviors of electrical loads can be better captured. In an actual electrical usage environment, the operating state of an electrical load may change at any time, such as load startup, shutdown, and adjustment of load magnitude. Time-frequency domain features can promptly reflect these changes and provide more comprehensive information for load identification.

[0091] In the above-described embodiments, by extracting features of the total power supply signal from three perspectives: time domain, frequency domain, and time-frequency domain, the electrical load information contained in the total power supply signal can be comprehensively and meticulously described, which helps the subsequent load identification model to more accurately identify different load types and improves the accuracy of non-intrusive electrical load identification.

[0092] Referring to Figure 3 , in some embodiments, the non-intrusive electrical load identification method further includes a training step of training a preset prediction network to obtain each load identification model. The training step includes step S201 - step S203, which will be introduced in detail below.

[0093] Step S201: Create training samples for each load identification model respectively, and use the load type corresponding to the training samples as training labels.

[0094] Among them, the training samples include at least one of time domain features, frequency domain features, and time-frequency domain features.

[0095] As an example, each load identification model is used to identify different load types. For each load type, appropriate training samples can be selected according to its own characteristics. Taking the load type of air conditioner as an example, in the time domain signal, since a large current peak will be generated when the air conditioner compressor starts, and there will be specific rules for the current mean and variance during its operation; in the frequency domain signal, since components such as the compressor and fan motor in the air conditioner may be non-linear loads, specific harmonic components will be generated, and the power spectral density will have different distributions under different operating modes (such as cooling, heating, different wind speeds, etc.). Therefore, peak value, mean value, variance, harmonics, and power spectral density can be selected as the training samples for the air conditioner.

[0096] Step S202: Input each training sample into each preset prediction network respectively to obtain the prediction confidence of each load type.

[0097] Step S203: Iteratively update each preset prediction network according to the prediction confidence of the load type and the corresponding training labels to obtain multiple load identification models for identifying different load types.

[0098] In the above embodiments, first, training samples for each load recognition model are created, and the load type corresponding to the training samples is used as the training label, providing targeted data for the subsequent training of each load recognition model. Second, each training sample is separately input into a preset prediction network to obtain the prediction confidence of each load type. This prediction confidence can reflect the credibility of the prediction network regarding whether a training sample belongs to a certain load type. Finally, based on the prediction confidence of the load type and the corresponding training label, each preset prediction network is iteratively updated. Through continuous adjustment and optimization of the network parameters, multiple load recognition models that can be used to identify different load types are finally obtained.

[0099] In some embodiments, the preset prediction network includes one or a combination of a Light Gradient Boosting Machine network, an Extreme Gradient Boosting network, a Shallow Convolutional Neural Network, and a Recurrent Neural Network.

[0100] As an example, the Light Gradient Boosting Machine (LGBM) is a fast and efficient gradient boosting framework that uses optimization techniques such as the histogram algorithm. By iteratively training multiple weak classifiers, it combines them into a strong classifier. The Light Gradient Boosting Machine can automatically learn the relationships between feature signals, accurately classify the feature signals, and output the confidence of each classification. Extreme Gradient Boosting (XGB) can handle missing values and has a certain robustness to abnormal data when the amount of data of the feature signal is large and there are many features. Thus, after classifying the load type, it can output a relatively reliable prediction confidence. The Convolutional Neural Network (CNN) extracts the local features of the feature signal through the convolutional layer, performs dimensionality reduction through the pooling layer, and finally classifies through the fully connected layer. The convolutional operation slides the filter over the data to capture different features. The Recurrent Neural Network (RNN) enables the network to remember previous input information through a recurrent structure, enabling it to process sequential data. Each node receives not only the current input but also the output of the previous node, thus forming a memory ability. In the scenario where the feature signal is recorded in time series, the Recurrent Neural Network can capture the temporal information in the data. Thus, it can analyze and classify the change of the load type over time and output the confidence.

[0101] In this way, according to the differences in the training samples corresponding to each load type, a suitable preset prediction network can be selected as the architecture of the corresponding load recognition model to improve the accuracy of the load recognition model.

[0102] Refer toFigure 4 and Figure 5 , in some embodiments, the multiple load recognition models may be in a parallel architecture, and step S104 may include steps S1041 - S1044, which will be introduced in detail below.

[0103] Step S1041: Determine a combination of target load types from each load type according to a first preset threshold and each prediction confidence level.

[0104] As an example, the load type with a prediction confidence level greater than the first preset threshold may be used as the target load type. Assume that the first preset threshold is 0.6. The prediction confidence level of load type A is 0.8, the prediction confidence level of load type B is 0.3, and the prediction confidence level of load type C is 0.7. Since the prediction confidence levels of load type A and load type C are greater than the first preset threshold of 0.6, the combination of target load types may be {A, C}.

[0105] Step S1042: Query the weight value of each prediction confidence level from the weight allocation table according to the combination of target load types.

[0106] Among them, the weight allocation table includes the corresponding relationship between the combination of target load types and the weight values of each prediction confidence level. As an example, the weight allocation table can be referred to Table 1.

[0107] Table 1

[0108] Combinations of target load types Load type A Load type B Load type C {A, C} 0.8 0.5 0.9 {A, B} 0.7 0.8 0.5 {B, C} 0.5 0.7 0.8

[0109] According to the combination of target load types {A, C}, it is queried that the weight value of load type A is 0.8, the weight value of load type B is 0.5, and the weight value of load type C is 0.9.

[0110] Step S1043: Adjust each prediction confidence level according to the weight value to obtain an optimized confidence level.

[0111] Step S1044: Compare the optimized confidence level with a second preset threshold, and obtain a load recognition result according to the comparison result.

[0112] Among them, the load recognition result is the load type whose optimized confidence level is greater than the second preset threshold.

[0113] As an example, assume that the second preset threshold is 0.6. The optimized confidence level of load type A, 0.64, is greater than the second preset threshold of 0.6; the optimized confidence level of load type B is 0.3, which is less than the second preset threshold of 0.6; the optimized confidence level of load type C is 0.63, which is greater than the second preset threshold of 0.6. Then the load recognition result is load type A and load type C.

[0114] Refer to Figure 6 andFigure 7 , in another embodiment, the multiple load recognition models may also be in a master-slave architecture. The multiple load recognition models include a master recognition model and several slave recognition models. Step S104 may include steps S1045 to S1047, which are introduced in detail below.

[0115] Step S1045: According to the prediction confidence of the master recognition model and the third preset threshold, correct the prediction confidence of several slave recognition models to obtain the corrected confidence of each slave recognition model.

[0116] As an example, the corrected confidence can be expressed as p′ subi = g i (P main , P subi ); where g i is the prediction adjustment function of slave model i. P main is the prediction confidence of the master model, and P subi is the prediction confidence of slave recognition model i; where P main = f main (X); P subi = f subi (X). f main is the prediction function of the master recognition model, and f subi is the prediction function of slave recognition model i.

[0117] Step S1046: Perform fusion adjustment on the corrected confidence and the prediction confidence of the master recognition model to obtain the optimized confidence of the master recognition model and the slave recognition models respectively.

[0118] As an example, the method for obtaining the optimized confidence of the master recognition model and the slave recognition models can specifically refer to steps S1041 to S1043 in the parallel architecture embodiment, which will not be elaborated here.

[0119] Step S1047: Compare the optimized confidence with the second preset threshold, and obtain the load recognition result according to the comparison result.

[0120] Among them, the load recognition result is the load type whose optimized confidence is greater than the second preset threshold. The specific implementation of comparing the optimized confidence with the second preset threshold and obtaining the load recognition result according to the comparison result can refer to step S1044 in the parallel architecture embodiment, which will not be elaborated here.

[0121] In some embodiments, step S1045 may include steps S10451 to S10452, which are introduced in detail below.

[0122] Step S10451: If the prediction confidence of the main recognition model is greater than or equal to the third preset threshold, then reduce the prediction confidence of the slave recognition models to obtain the corrected confidence of each slave recognition model.

[0123] Step S10452: If the prediction confidence of the main recognition model is less than the third preset threshold, then maintain or increase the prediction confidence of the slave recognition models to obtain the corrected confidence of each slave recognition model.

[0124] As an example, if the prediction confidence of the main recognition model is greater than or equal to the third preset threshold, it indicates that the load type recognized by the main recognition model exists at this time. At this time, the load types of other slave recognition models can be masked, so the corrected confidence of the slave recognition models can be reduced. When the prediction confidence of the main recognition model is less than the third preset threshold, it indicates that the load type recognized by the main recognition model does not exist. At this time, the load types of other slave recognition models can be recognized, so the prediction confidence of the slave recognition models can be maintained or increased.

[0125] As an example, the parallel architecture of the load recognition model can be applied to scenarios where multiple load types are relatively independent and there is no obvious primary or secondary relationship. For example, in a smart home environment, it is necessary to simultaneously identify multiple different types of small electrical appliances, such as table lamps, speakers, air purifiers, etc. The master-slave architecture is applicable to scenarios where there are critical tasks or load types that need to be focused on. By adjusting the model priorities, the main recognition model can be made to identify first, and other models can identify in parallel. For example, in a high-risk recognition scenario for electric vehicle charging, the electric vehicle charging recognition is taken as the main task, and the main recognition model is used for priority recognition. At the same time, other normal load types can be recognized in parallel by the slave models, taking into account the recognition of other load types while ensuring that key tasks are processed first.

[0126] In a second aspect, as Figure 8 shown, in one embodiment, the present application provides a non-intrusive electrical load recognition system. The non-intrusive electrical load recognition system includes a signal acquisition module, a feature extraction module, an independent model prediction module, and a result fusion module.

[0127] Among them, the signal acquisition module is used to acquire the total power supply signal of the electrical environment;

[0128] The feature extraction module is used to extract features from the total power supply signal to obtain the feature signal of the total power supply signal;

[0129] The independent model prediction module is used to input the feature signal into multiple load recognition models for identifying different load types respectively, identify the corresponding load types by each load recognition model, and output the prediction confidence corresponding to each load type;

[0130] A result fusion module is configured to obtain a load identification result according to the prediction confidence corresponding to each load type.

[0131] As an example, the signal acquisition module, the feature extraction module, the independent model prediction module, and the result fusion module are respectively configured to execute steps S101 - S104. For details, refer to the method implementation and will not be elaborated here.

[0132] In a third aspect, in an embodiment, the present application provides a circuit breaker, as Figure 9 shown, which shows the structure of the circuit breaker involved in the present application. Specifically:

[0133] The circuit breaker may include a processor 401 with one or more processing cores, a memory 402 of one or more computer-readable storage media, a power supply 403, an input unit 404, and other components. Those skilled in the art can understand that Figure 9 the structure of the circuit breaker shown in

[0134] does not limit the circuit breaker, and it may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0135] The processor 401 is the control center of the circuit breaker, connecting various parts of the entire circuit breaker through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 402, and by calling data stored in the memory 402, it executes various functions of the circuit breaker and processes data, thereby monitoring the circuit breaker as a whole. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modulation and demodulation processor. Among them, the application processor mainly processes the operating system, user interface, computer programs, etc., and the modulation and demodulation processor mainly processes wireless communication. It can be understood that the above modulation and demodulation processor may not be integrated into the processor 401.

[0136] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. Among them, the program storage area may store the operating system, computer programs required for at least one function (such as the sound playback function, image playback function, etc.); the data storage area may store data created according to the use of the server. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0137] The circuit breaker further includes a power source 403 for supplying power to each component. Preferably, the power source 403 can be logically connected to the processor 401 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power source 403 may further include any components such as one or more DC or AC power sources, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0138] The circuit breaker may further include an input unit 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0139] Although not shown, the circuit breaker may further include a display unit, etc., which will not be elaborated here. Specifically, in this embodiment, when the circuit breaker is a model training circuit breaker, the processor 401 in the circuit breaker will load the executable files corresponding to the processes of one or more computer programs into the memory 402 according to the following instructions, and the processor 401 will run the computer programs stored in the memory 402 to execute the above steps.

[0140] Those of ordinary skill in the art can understand that all or part of the steps in any of the above methods can be completed by a computer program, or by controlling related hardware through a computer program. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0141] In a fourth aspect, in an embodiment, the present application provides a storage medium, in which multiple computer programs are stored, and the computer programs can be loaded by a processor to execute the above steps.

[0142] Those of ordinary skill in the art can understand that any reference to a memory, storage, database, or other medium used in the embodiments provided in this application may include non-volatile and / or volatile memories. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink), DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0143] Since the computer program stored in the storage medium can execute the steps in the non-intrusive electrical load identification method in any of the embodiments provided in this application, the beneficial effects achievable by the non-intrusive electrical load identification method in any of the embodiments provided in this application can be realized. For details, refer to the previous embodiments and will not be elaborated here.

[0144] The specific implementation of each of the above operations can be referred to the previous embodiments and will not be elaborated here.

[0145] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not elaborated in a certain embodiment, reference can be made to the detailed descriptions of other embodiments above, and will not be elaborated here.

[0146] The above has introduced in detail a non-intrusive electrical load identification method, system, circuit breaker, and medium provided in this application. Specific examples are used in this article to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those skilled in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

[0147] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

Claims

1. A non-intrusive method for identifying electrical loads, characterized in that, Including: Obtain the total power supply signal of the power consumption environment; Extract features from the total power supply signal to obtain the feature signal of the total power supply signal; Input the feature signal into multiple load recognition models for identifying different load types respectively, identify the corresponding load types through each load recognition model respectively, and output the predicted confidence levels corresponding to each load type; Obtain the load recognition result according to the predicted confidence levels corresponding to each load type.

2. The non-intrusive electrical load identification method according to claim 1, wherein The feature signal includes time domain features, frequency domain features, and time-frequency domain features; The extracting features from the total power supply signal to obtain the feature signal includes: Perform statistical and morphological processing on the total power supply signal to obtain the time domain features; Perform Fourier transform processing on the total power supply signal to obtain the frequency domain features; Perform wavelet transform processing on the total power supply signal to obtain the time-frequency domain features.

3. The non-intrusive power load recognition method according to claim 2, characterized in that: The time domain features include morphological features and statistical features. Among them, the morphological features include at least one of the kurtosis, skewness, kurtosis, and waveform factor of the total power supply signal; the statistical features include at least one of the mean, variance, peak value, and zero-crossing rate of the total power supply signal; The frequency domain features include at least one of the fundamental wave, harmonic, power spectral density, and spectral entropy of the total power supply signal; The time-frequency domain features include at least one of the high-frequency wavelet coefficients and low-frequency wavelet coefficients.

4. The non-invasive method for identifying an electrical load according to claim 3, wherein It further includes a training step of training a preset prediction network to obtain each load recognition model, and the training step includes: Create training samples for each load recognition model respectively, and use the load type corresponding to the training sample as the training label; wherein, the training sample includes at least one of the time domain features, the frequency domain features, and the time-frequency domain features; Input each training sample into each preset prediction network respectively to obtain the predicted confidence level of each load type; Iteratively update each preset prediction network respectively according to the predicted confidence level of the load type and the corresponding training label to obtain multiple load recognition models for identifying different load types.

5. The non-invasive method for identifying an electrical load according to claim 4, wherein The preset prediction network includes one or a combination of a lightweight gradient boosting machine network, an extreme gradient boosting network, a shallow convolutional neural network, and a recurrent neural network.

6. The non-invasive method for identifying an electrical load according to claim 1, characterized in that, The multiple load recognition models are in a parallel architecture, and the obtaining the load recognition result according to the predicted confidence levels corresponding to each load type includes: Determine the combination of target load types from each load type according to a first preset threshold and each predicted confidence level; Query the weight value of each predicted confidence level from the weight distribution table according to the combination of the target load types; wherein, the weight distribution table includes the corresponding relationship between the combination of the target load types and the weight values of each predicted confidence level; Adjust each predicted confidence level according to the weight value to obtain the optimized confidence level; Compare the optimized confidence level with a second preset threshold, and obtain the load identification result according to the comparison result; Wherein, the load identification result is the load type for which the optimized confidence level is greater than the second preset threshold.

7. The non-intrusive electrical load identification method according to claim 1, characterized in that The multiple load identification models are in a master-slave architecture, and the multiple load identification models include a master identification model and several slave identification models; The obtaining the load identification result according to the prediction confidence levels corresponding to the respective load types includes: According to the prediction confidence level of the master identification model and a third preset threshold, correct the prediction confidence levels of the several slave identification models to obtain the corrected confidence levels of the respective slave identification models; Perform fusion adjustment on the corrected confidence levels and the prediction confidence level of the master identification model to respectively obtain the optimized confidence levels of the master identification model and the slave identification models; Compare the optimized confidence level with a second preset threshold, and obtain the load identification result according to the comparison result.

8. The non-intrusive electrical load identification method according to claim 7, characterized in that The correcting the prediction confidence levels of the several slave identification models according to the prediction confidence level of the master identification model and a third preset threshold to obtain the corrected confidence levels of the respective slave identification models includes: If the prediction confidence level of the master identification model is greater than or equal to the third preset threshold, reduce the prediction confidence levels of the slave identification models to obtain the corrected confidence levels of the respective slave identification models; If the prediction confidence level of the master identification model is less than the third preset threshold, maintain or increase the prediction confidence levels of the slave identification models to obtain the corrected confidence levels of the respective slave identification models.

9. A non-invasive electrical load identification system, characterized in that, Comprising: A signal acquisition module, configured to acquire the total power supply signal of the power consumption environment; A feature extraction module, configured to extract features from the total power supply signal to obtain the feature signal of the total power supply signal; An independent model prediction module, configured to respectively input the feature signal into multiple load identification models for identifying different load types, identify the respective corresponding load types through the respective load identification models, and output the prediction confidence levels corresponding to the respective load types; A result fusion module, configured to obtain a load identification result according to the prediction confidence levels corresponding to the respective load types.

10. A circuit breaker, characterized in that, Comprising a memory and a processor; the memory stores a computer program, and the processor is configured to run the computer program in the memory to execute the steps in a non-intrusive method for identifying electrical loads according to any one of claims 1 to 8.

11. A storage medium, characterized in that, The storage medium stores a computer program, and the computer program is loaded by a processor to execute the steps in a non-intrusive method for identifying electrical loads according to any one of claims 1 to 8.

Images