A dynamic electric energy metering compensation method and system with load type self-recognition

Abstract

The present application relates to the technical field of electric energy metering, and more particularly, the present application provides a dynamic electric energy metering compensation method and system with load type self-identification. Multiple signal modal components are obtained through signal decomposition, and then time-frequency features are extracted and input into a load state analysis model. The output is the activation probability distribution of each internal working module of the electrical appliance, rather than only identifying the overall electrical appliance. This module-level identification granularity enables the system to distinguish the electrical differences of the same electrical appliance in different working stages, providing a fine basis for subsequent differentiated electric energy metering compensation. Furthermore, the present application constructs an electrical appliance topology knowledge base, which at least includes the mapping relationship between modules and electrical appliance types, the dependency relationship and mutual exclusion relationship between modules, and the corresponding rules between module combinations and electrical appliance working stages. When multiple modules are activated simultaneously, the system excludes unreasonable combinations according to these rules, thereby improving the reliability and interpretability of the identification results.

Description

Technical Field

[0001] This invention relates to the field of electricity metering technology, and more specifically, to a dynamic electricity metering compensation method and system with self-identification of load type. Background Technology

[0002] In the fields of smart grids and refined electricity management, dynamic power metering compensation technology is crucial for improving the overall efficiency of the power grid, ensuring the fairness of metering for users, and optimizing energy distribution. The core of this technology lies in accurately identifying the constantly changing types of electrical loads at the user end and implementing differentiated power quality compensation and metering correction accordingly, thereby addressing issues such as harmonic distortion and voltage fluctuations caused by nonlinear loads in the power grid. However, existing research mainly relies on analyzing current and voltage waveform characteristics to achieve load identification, and these methods are often limited to a pre-set fixed load characteristic library. This static characteristic library is prone to failure when faced with the introduction of new electrical appliances or when traditional appliances are in atypical operating conditions. For example, the electrical signatures of the motors and sensor modules of new intelligent robotic vacuum cleaners differ significantly from traditional models; uneven operation of the heating elements in old rice cookers during half-load cooking also causes the characteristics to deviate significantly from the preset values.

[0003] A deeper analysis reveals that the root cause of identification failures lies not only in the static nature of the feature library, but also in the complex and dynamic electrical characteristics resulting from the coordination and switching of multiple internal working modules within an appliance in a real operating environment. Take a variable frequency air conditioner as an example: during startup, the high inrush current of the compressor superimposes with the low-frequency harmonics of the fan; during frequency ramp-up, the multi-frequency harmonics generated by the inverter further interfere; during steady-speed operation, if the user adjusts the set temperature, the pulse signal of the control circuit will change the overall waveform pattern in real time. This dynamic superposition between modules not only changes with the operating stage but is also affected by external factors (such as ambient temperature and voltage fluctuations), resulting in a highly non-stationary composite pattern in the total current waveform. Similarly, in a multi-functional washing machine, during the spin-drying stage, the alternating high-speed rotation of the motor and the intermittent starting of the drain pump creates periodically abrupt current peaks and troughs. If the drum heating module is activated simultaneously, a low-frequency thermal effect is introduced, further blurring the boundaries of individual features. Existing methods struggle to capture these inherent dynamic combination patterns because they assume relatively stable features and cannot handle the nonlinear coupling between modules.

[0004] The aforementioned dynamic combination and uncertainty constitute the core technical challenge in this field, directly leading to another closely related problem: achieving high-precision load decomposition and identification becomes exceptionally difficult in complex mixed power consumption scenarios where multiple appliances operate simultaneously. Taking a household evening peak scenario as an example, the high-power heating of a microwave oven in the kitchen, the harmonics of the power supply of a TV in the living room, the pulse current of a charger in the bedroom, and the inverter waveform of an air conditioner all intertwine, causing the characteristics of each appliance in the total current signal to mask and distort each other. For instance, the rectangular wave of a microwave oven overlaps with the sinusoidal fundamental wave of an air conditioner compressor, potentially leading to misidentification as a single high-power device; the spike noise of a TV merges with the harmonics of a charger, amplifying the identification error. This intertwining effect is even more severe in industrial scenarios, such as the parallel operation of multiple inverter fans and induction heating equipment in a small factory, where their composite harmonic spectra highly overlap, making it almost impossible to separate individual loads from the total waveform. Incomplete feature separation not only weakens the accuracy of load type identification but also directly affects the targeting of subsequent power quality compensation. For example, without distinguishing the source of reactive power, it is difficult to implement accurate reactive power compensation, resulting in metering deviations and increased grid losses.

[0005] Therefore, how to adaptively analyze the dynamic combination characteristics within electrical appliances and accurately identify the operating status and contribution share of each specific load in real-world scenarios with overlapping loads has become a key bottleneck restricting the transformation of dynamic energy metering and compensation technology from the laboratory to large-scale practical application. Existing technologies struggle to balance accuracy and robustness, ultimately hindering the realization of fair and efficient electricity management in smart grids. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic energy metering and compensation method and system with self-identification of load type, in order to solve the technical problems in the prior art where the dynamic combination of internal working modules of electrical appliances cannot be identified in fine granular terms, the lack of inter-module dependency and mutual exclusion rules leads to unreasonable inference results, and the fixed feature template library cannot automatically discover unknown load modes and update the model, thus making it impossible to implement targeted dynamic energy metering and compensation.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a dynamic energy metering and compensation method with self-identification of load type, comprising the following steps:

[0008] S1. Obtain real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence;

[0009] S2. Perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components; the signal mode components correspond to the oscillation characteristics of different loads in the electrical system;

[0010] S3. For each of the signal mode components, extract the time-domain statistical features and frequency-domain spectral features of the signal mode components to form a dynamic feature vector;

[0011] S4. Input the dynamic feature vector into the pre-trained load state analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment.

[0012] S5. Based on the activation probability distribution, perform similarity matching in the preset module feature template library, and determine the modules in the running state based on the matching results;

[0013] S6. Based on the modules in operation, and in conjunction with a preset electrical appliance topology knowledge base, deduce the type of electrical appliance currently in operation and its working stage; the electrical appliance topology knowledge base includes at least: the mapping relationship between modules and electrical appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and electrical appliance working stages;

[0014] S7. Based on the type of electrical appliance and the working stage, extract the corresponding standard current waveform template from the module feature template library, and superimpose them to generate the theoretical current waveform of each electrical appliance.

[0015] S8. Subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal;

[0016] S9. Analyze the energy and characteristics of the residual current signal, and determine whether the residual current signal meets the background noise condition. If it does not meet the condition, store the characteristics of the residual current signal in the feature pool to be identified, and use it to update the load state analysis model in the future.

[0017] A further technical solution of this application: Step S2 includes the following sub-steps:

[0018] S21. Perform multi-scale decomposition on the original electrical signal sequence to generate signal components at multiple frequency levels;

[0019] S22. For each frequency level of signal component, identify the transient and periodic characteristics of the signal component, and separate the signal mode components that characterize different load oscillation modes.

[0020] A further technical solution of this application: Step S3 includes the following sub-steps:

[0021] S31. Perform time-domain analysis on the signal modal components to extract the amplitude variation and time distribution characteristics of the signal modal components;

[0022] S32. Perform frequency domain transformation on the signal mode components, extract the spectral distribution and dominant frequency components of the signal mode components, and combine them to form the dynamic feature vector.

[0023] A further technical solution of this application: Step S4 includes: inputting the dynamic feature vector into the load state analysis model according to a preset format; generating the activation probability distribution covering the prediction of the operating status of multiple modules through the internal hierarchical calculation of the load state analysis model.

[0024] A further technical solution of this application: Step S5 includes the following sub-steps:

[0025] S51. Based on the activation probability distribution, modules with probability values ​​higher than a preset threshold are selected as candidate modules.

[0026] S52. Compare the dynamic feature vector corresponding to the candidate module with the standard feature template in the module feature template library one by one, and calculate the similarity index.

[0027] S53. Determine whether the candidate module is in operation based on whether the similarity index exceeds the matching threshold of the corresponding module.

[0028] A further technical solution of this application: In step S52, the similarity index is calculated using cosine similarity, and the formula is as follows:

[0029]

[0030] Among them, subscript Indicates the index of the signal mode component. Indicates the module number; Indicates the first The dynamic eigenvector of the first signal mode component and the second... Cosine similarity between the standard feature templates of each module; Indicates the first The dynamic eigenvectors of each signal mode component This indicates the first module in the feature template library. The standard feature template for each module.

[0031] A further technical solution of this application: Step S6 includes the following sub-steps:

[0032] S61. Based on the module in operation, query the connection relationship between the module and the appliance stored in the appliance topology knowledge base;

[0033] S62. Based on the connection relationship, deduce the combination state of the electrical appliance type and the working stage.

[0034] A further technical solution of this application: Step S7 includes the following sub-steps:

[0035] S71. Based on the electrical appliance type and working stage, locate the corresponding standard current waveform template in the module feature template library;

[0036] S72. The standard current waveform template is superimposed to generate the theoretical current waveform that reflects the current characteristics of multiple electrical appliances operating in parallel.

[0037] A further technical solution of this application: Steps S8 and S9 specifically include:

[0038] S81. Calculate the point-by-point difference between the total current waveform of the original electrical signal sequence and the theoretical current waveform to generate the residual current signal; the calculation formula is as follows:

[0039]

[0040] in, Represents the residual current signal. This represents the original total current waveform. Indicates the first The theoretical current waveform of the electrical appliance;

[0041] S82. Calculate the energy value and characteristic distribution of the residual current signal;

[0042] S83. If the energy value is lower than the preset noise energy threshold and the feature distribution does not match the template in the module feature template library, it is determined to be background noise and ignored.

[0043] S84. If the energy value is higher than or equal to the noise energy threshold, or if the feature distribution does not match the template in the module feature template library, then the features of the residual current signal are stored in the feature pool to be identified for subsequent updates to the load state analysis model.

[0044] A dynamic energy metering and compensation system with self-identifying load type includes:

[0045] The data acquisition module is used to acquire real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence;

[0046] The signal decomposition module is used to perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components;

[0047] The feature extraction module is used to extract the time-domain statistical features and frequency-domain spectral features of each signal mode component to form a dynamic feature vector.

[0048] The load status analysis module is used to input the dynamic feature vector into the pre-trained load status analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment.

[0049] The module matching and determination module is used to perform similarity matching in a preset module feature template library according to the activation probability distribution, and determine the module in the running state based on the matching result;

[0050] An appliance reasoning module is used to deduce the type and working stage of the currently running appliance based on the module in operation and in conjunction with a preset appliance topology knowledge base. The appliance topology knowledge base includes at least: the mapping relationship between modules and appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance working stages.

[0051] The theoretical waveform generation module is used to extract the corresponding standard current waveform template from the module feature template library according to the type of electrical appliance and the working stage, and superimpose them to generate the theoretical current waveform of each electrical appliance.

[0052] The residual signal calculation module is used to subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal.

[0053] The residual signal analysis module is used to analyze the energy and characteristics of the residual current signal, determine whether the residual current signal meets the background noise conditions, and if not, store the characteristics of the residual current signal in the feature pool to be identified for subsequent updates to the load state analysis model.

[0054] Compared with the prior art, the technical solution provided by this invention has the following advantages:

[0055] 1. This invention obtains multiple signal modal components through signal decomposition, then extracts time-frequency features and inputs them into a load state analytical model. The output is the activation probability distribution of the internal working modules of each electrical appliance, rather than just identifying the appliance as a whole. This module-level identification granularity allows the system to distinguish the electrical differences of the same appliance at different operating stages, providing a precise basis for differentiated energy metering compensation in subsequent stages. Furthermore, this invention constructs an appliance topology knowledge base, which includes at least the mapping relationship between modules and appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance operating stages. When multiple modules are activated simultaneously, the system uses these rules to exclude unreasonable combinations, thereby improving the reliability and interpretability of the identification results and avoiding logical errors that may arise from relying solely on probability matching.

[0056] 2. This invention forms a closed loop through theoretical waveform synthesis and residual signal analysis. Based on the identified appliance type and operating stage, corresponding standard current waveform templates are extracted from the module feature template library and superimposed to generate a theoretical current waveform. The theoretical waveform is then subtracted from the original total current waveform to obtain the residual current signal. If the energy of the residual signal exceeds a preset noise energy threshold, or if its features are not in the existing template library, an unknown load mode is determined, and this feature is stored in the feature pool to be identified. This mechanism enables the system to automatically discover new appliances or atypical operating states, continuously optimizing the model without manual intervention. Furthermore, this invention organically integrates signal decomposition, module-level probabilistic recognition, knowledge base constraints, theoretical waveform synthesis, and residual signal feedback, effectively decoupling the superimposed and masked electrical features of multiple appliances operating in parallel. In complex mixed power consumption scenarios, this invention can more accurately identify the operating status and contribution share of various appliances, providing a reliable foundation for subsequent targeted power quality compensation, while also possessing good versatility and scalability. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention will be further described below with reference to the embodiments.

[0059] Example 1

[0060] Please see Figure 1 In one embodiment of this application, a dynamic energy metering compensation method with self-identification of load type includes the following steps:

[0061] S1. Obtain real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence;

[0062] Current and voltage sensors are installed at the main incoming line of the power grid to continuously acquire real-time current and voltage waveforms at a preset sampling frequency. The acquired analog signals are converted into digital signals by an analog-to-digital converter to form the original electrical signal sequence. The sampling frequency should satisfy the Nyquist sampling theorem, i.e., it should be at least twice the highest frequency component of the analyzed signal to ensure no aliasing distortion. The acquisition duration can be dynamically set according to the transient characteristics of the load; for example, a shorter time window is used for fast-switching equipment, and a longer time window is used for steady-state loads.

[0063] S2. Perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components; the signal mode components correspond to the oscillation characteristics of different loads in the electrical system.

[0064] Step S2 includes the following sub-steps:

[0065] S21. Perform multi-scale decomposition on the original electrical signal sequence to generate signal components at multiple frequency levels;

[0066] S22. For each frequency level of signal component, identify the transient and periodic characteristics of the signal component, and separate the signal mode components that characterize different load oscillation modes.

[0067] It is important to note that this step aims to decompose the complex original electrical signal into multiple signal mode components with different oscillation modes, each component corresponding to the contribution of a type of load in the electrical system.

[0068] In one specific implementation, variational mode decomposition (VMD) is used to decompose the original electrical signal sequence. VMD adaptively decomposes the signal into multiple intrinsic mode functions (IMFs), each with a center frequency and a finite bandwidth. The decomposition parameters can be set through optimization algorithms or empirically. In other implementations, empirical mode decomposition (EMD) or wavelet packet decomposition can also be used.

[0069] After decomposition, K signal mode components are obtained, denoted as... These components differ significantly in frequency distribution and transient characteristics. For example, high-frequency components correspond to the fast pulses of electronic switching devices, while low-frequency components correspond to the fundamental oscillations of motor loads.

[0070] This step obtains multiple signal mode components through signal decomposition, rather than directly analyzing the raw total current. This process separates the oscillation modes of different loads, laying the foundation for subsequent module-level identification. Compared with existing technologies that directly extract the overall features of electrical appliances, this step can preserve the independent electrical signatures of different modules, avoiding feature ambiguity caused by the superposition of multiple modules.

[0071] S3. For each of the signal mode components, extract the time-domain statistical features and frequency-domain spectral features of the signal mode components to form a dynamic feature vector.

[0072] Step S3 includes the following sub-steps:

[0073] S31. Perform time-domain analysis on the signal modal components to extract the amplitude variation and time distribution characteristics of the signal modal components;

[0074] S32. Perform frequency domain transformation on the signal mode components, extract the spectral distribution and dominant frequency components of the signal mode components, and combine them to form the dynamic feature vector.

[0075] For each signal mode component Extract the following time-domain statistical features: peak value, peak-to-peak value, root mean square value, variance, skewness, and kurtosis.

[0076] At the same time, it is necessary to analyze each signal mode component. A fast Fourier transform is performed to obtain its spectral distribution, and frequency domain spectral features are extracted. The frequency domain spectral features include: fundamental amplitude, amplitude of each harmonic, total harmonic distortion rate, and weighted average frequency of spectral energy distribution.

[0077] All extracted time-domain and frequency-domain features are combined into a high-dimensional vector, called the dynamic feature vector. This vector reflects the variation pattern of the k-th signal mode component in the time and frequency dimensions.

[0078] This step extracts time-frequency features for each signal mode component individually, forming a dynamic feature vector. These features characterize the subtle variation patterns of the module in the time and frequency dimensions, providing precise input for subsequent identification of different working modules within the same electrical appliance. Existing technologies typically extract statistical features of the total current, failing to distinguish the contributions of different modules.

[0079] S4. Input the dynamic feature vector into the pre-trained load state analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment.

[0080] Step S4 further includes: inputting the dynamic feature vector into the load state analysis model according to a preset format; and generating the activation probability distribution covering the prediction of the operating states of multiple modules through the internal hierarchical calculation of the load state analysis model.

[0081] A load state analytical model is pre-trained; this model can employ a multi-layer neural network or a hidden Markov model. The model's input is a dynamic feature vector. The output is a probability vector. The formula is as follows:

[0082]

[0083] in, The total number of working modules in the module feature template library, each This represents the probability that the m-th module is activated at the current moment.

[0084] By collecting a large amount of historical data on the operating status of known electrical modules, the extracted dynamic feature vectors are used as inputs, and the corresponding module activation labels are used as outputs. The backpropagation algorithm or the Baum-Welch algorithm is then used to optimize the model parameters.

[0085] In this embodiment, the dynamic feature vectors of each signal mode component obtained in step S3 are sequentially input into the trained load state analytical model to obtain the activation probability distribution of each module at each time step.

[0086] This step outputs not the overall operating probability of the appliance, but rather the activation probability distribution of each working module. This module-level probability output allows the system to determine which modules within the same appliance are currently operating, thereby inferring the specific operating stage of the appliance—something traditional appliance-level recognition cannot do.

[0087] S5. Based on the activation probability distribution, perform similarity matching in the preset module feature template library, and determine the modules in the running state based on the matching results.

[0088] Step S5 includes the following sub-steps:

[0089] S51. Based on the activation probability distribution, modules with probability values ​​higher than a preset threshold are selected as candidate modules.

[0090] S52. Compare the dynamic feature vector corresponding to the candidate module with the standard feature template in the module feature template library one by one, and calculate the similarity index.

[0091] S53. Determine whether the candidate module is in operation based on whether the similarity index exceeds the matching threshold of the corresponding module. Different matching thresholds can be set for different modules: for example, the matching threshold of the steady-state operation module can be set to 0.7, and the matching threshold of the transient pulse module can be set to 0.85 to avoid misjudgment.

[0092] Furthermore, in step S52, the similarity index is calculated. In a preferred embodiment, cosine similarity is used for calculation, and the formula is as follows:

[0093]

[0094] Among them, subscript Indicates the index of the signal mode component. Indicates the module number; Indicates the first The dynamic eigenvector of the first signal mode component and the second... Cosine similarity between the standard feature templates of each module; Indicates the first The dynamic eigenvectors of each signal mode component This indicates the first module in the feature template library. The standard feature templates for each module are provided, where the cosine similarity value ranges from [-1, 1], and the closer the value is to 1, the more similar the similarity.

[0095] This step sets different matching thresholds for different modules, improving recognition accuracy. Furthermore, the matching target is the module, not the appliance, enabling the system to identify the combined state of multiple modules within an appliance.

[0096] S6. Based on the modules in operation, and in conjunction with a preset electrical appliance topology knowledge base, deduce the type of electrical appliance currently in operation and its working stage; the electrical appliance topology knowledge base includes at least: the mapping relationship between modules and electrical appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and electrical appliance working stages.

[0097] It should be noted that the Electrical Topology Knowledge Base is a structured database that contains at least the following:

[0098] Mapping relationship between modules and appliance types: For example, the compressor starting winding module belongs to air conditioning appliances; the heating element full power module belongs to electric water heaters.

[0099] Dependencies and mutual exclusions between modules: For example, once the compressor start winding module is activated, the compressor running winding module must be activated as well; heating and cooling modules within the same appliance are mutually exclusive.

[0100] The correspondence between module combinations and electrical operating stages: For example, the compressor start winding module being activated while the compressor running winding module is not activated corresponds to the start-up stage of the air conditioner; the compressor running winding module being activated while the inverter drive module is in the frequency-increasing state corresponds to the frequency-increasing stage of the air conditioner.

[0101] Step S6 includes the following sub-steps:

[0102] S61. Based on the module in operation, query the connection relationship between the module and the appliance stored in the appliance topology knowledge base;

[0103] S62. Based on the connection relationship, deduce the combination state of the electrical appliance type and the working stage.

[0104] The specific reasoning process is as follows: Based on the set of modules in operation determined in step S5, the connection relationships between modules and appliances stored in the appliance topology knowledge base are queried to obtain the appliance type to which each module belongs. Combining the correspondence rules between module combinations and working stages, the specific working stage of each appliance is derived. For example, if the compressor starting winding module is detected to be activated while the compressor running winding module is not activated, it is determined that the air conditioner is in the starting stage; if both are activated and the frequency value of the inverter drive module continues to rise, it is determined that it is in the frequency ramp-up stage.

[0105] The core innovation of this step lies in the introduction of an electrical topology knowledge base, which explicitly stores the dependency and mutual exclusion rules between modules, as well as the correspondence between module combinations and working stages. When multiple modules are activated simultaneously, the system uses these rules to eliminate unreasonable combinations, thereby improving the reliability and interpretability of the reasoning results. Existing technologies lack such structured knowledge constraints, which can easily lead to logical contradictions.

[0106] S7. Based on the type of electrical appliance and the working stage, extract the corresponding standard current waveform template from the module feature template library, and superimpose them to generate the theoretical current waveform of each electrical appliance.

[0107] Step S7 includes the following sub-steps:

[0108] S71. Based on the type of electrical appliance and its operating stage, locate the corresponding standard current waveform template in the module feature template library. Each combination of operating stages of an electrical appliance corresponds to a standard current waveform template, which is a representative current waveform obtained by averaging or clustering a large amount of typical operating data.

[0109] S72. The standard current waveform template is superimposed to generate the theoretical current waveform that reflects the current characteristics of multiple electrical appliances operating in parallel. The specific formula is as follows:

[0110]

[0111] in, This represents the total number of currently operating electrical appliances. Let be the theoretical current waveform of the e-th electrical appliance. When superimposing the waveforms, the phase relationship between the electrical appliances needs to be considered. Usually, it is assumed that each electrical appliance operates independently, and algebraic superposition is performed directly.

[0112] S8. Subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal;

[0113] S9. Analyze the energy and characteristics of the residual current signal, and determine whether the residual current signal meets the background noise condition. If it does not meet the condition, store the characteristics of the residual current signal in the feature pool to be identified, and use it to update the load state analysis model in the future.

[0114] Steps S8 and S9 specifically include:

[0115] S81. Calculate the point-by-point difference between the total current waveform of the original electrical signal sequence and the theoretical current waveform to generate the residual current signal; the calculation formula is as follows:

[0116]

[0117] in, Represents the residual current signal. This represents the original total current waveform. Indicates the first The theoretical current waveform of the electrical appliance;

[0118] S82. Calculate the energy value and characteristic distribution of the residual current signal;

[0119] S83. If the energy value is lower than the preset noise energy threshold and the feature distribution does not match the template in the module feature template library, it is determined to be background noise and ignored.

[0120] S84. If the energy value is higher than or equal to the noise energy threshold, or if the feature distribution does not match the template in the module feature template library, then the features of the residual current signal are stored in the feature pool to be identified, for subsequent updates to the load state analysis model. When the number of features accumulated in the feature pool to be identified reaches a preset batch size, incremental training of the load state analysis model is triggered: the new features are merged with the original training data, the model is fine-tuned again, and the updated model parameters are synchronized to the online system. In this way, the system can gradually learn the features of new electrical appliances or atypical working states, continuously improving its recognition capabilities.

[0121] The above steps form a closed-loop self-optimization mechanism. Specifically: Step S7 synthesizes theoretical waveforms based on the identified appliance type and operating stage; Step S8 obtains the residual signal through difference; and Step S9 analyzes the energy and characteristics of the residual signal. If the residual signal energy exceeds the noise threshold or its characteristics are not in the existing template library, an unknown load mode is determined to exist and stored in the feature pool, triggering incremental model training. This mechanism enables the system to automatically discover new appliances or atypical operating states, continuously optimizing the model without manual annotation. Compared with existing feedback methods that require manual intervention, the closed-loop self-optimization of this invention has higher practicality and adaptability.

[0122] Example 2

[0123] This embodiment provides a dynamic energy metering and compensation system with self-identification of load type, such as... Figure 2 As shown, a dynamic energy metering and compensation system with self-identification of load type includes:

[0124] The data acquisition module is used to acquire real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence;

[0125] The signal decomposition module is used to perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components;

[0126] The feature extraction module is used to extract the time-domain statistical features and frequency-domain spectral features of each signal mode component to form a dynamic feature vector.

[0127] The load status analysis module is used to input the dynamic feature vector into the pre-trained load status analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment.

[0128] The module matching and determination module is used to perform similarity matching in a preset module feature template library according to the activation probability distribution, and determine the module in the running state based on the matching result;

[0129] An appliance reasoning module is used to deduce the type and working stage of the currently running appliance based on the module in operation and in conjunction with a preset appliance topology knowledge base. The appliance topology knowledge base includes at least: the mapping relationship between modules and appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance working stages.

[0130] The theoretical waveform generation module is used to extract the corresponding standard current waveform template from the module feature template library according to the type of electrical appliance and the working stage, and superimpose them to generate the theoretical current waveform of each electrical appliance.

[0131] The residual signal calculation module is used to subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal.

[0132] The residual signal analysis module is used to analyze the energy and characteristics of the residual current signal, determine whether the residual current signal meets the background noise conditions, and if not, store the characteristics of the residual current signal in the feature pool to be identified for subsequent updates to the load state analysis model.

[0133] The specific workflows of the above modules correspond to steps S1 to S9 in Embodiment 1, and will not be repeated here. Those skilled in the art will understand that these modules can be implemented by software, hardware, or a combination of both, for example, by programming in a digital signal processor, field-programmable gate array, or embedded microcontroller.

[0134] In the above embodiments of the present invention, some specific algorithm examples are given. These specific implementations are only used to enable those skilled in the art to implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Those skilled in the art can fully utilize other equivalent signal decomposition methods, similarity measurement methods, or classification models based on their understanding of the concept of the present invention. As long as the core technical solutions of module-level activation probability inference, knowledge base constraints, theoretical waveform synthesis, and residual signal feedback can be achieved, they should all fall within the scope of protection of the present invention.

[0135] For example: the signal decomposition in step S2 can employ empirical mode decomposition, ensemble empirical mode decomposition, or wavelet packet decomposition. The load state analytical model in step S4 can employ a hidden Markov model, conditional random field, or deep belief network. The similarity calculation in step S5 can employ Euclidean distance, Manhattan distance, or Mahalanobis distance. The model update in step S9 can employ online gradient descent, mini-batch learning, or active learning strategies.

[0136] In summary, this invention obtains multiple signal modal components through signal decomposition, then extracts time-frequency features and inputs them into a load state analytical model. The output is the activation probability distribution of the internal working modules of each electrical appliance, rather than simply identifying the appliance as a whole. This module-level identification granularity allows the system to distinguish the electrical differences of the same appliance at different operating stages, providing a precise basis for differentiated energy metering compensation in subsequent stages. Furthermore, this invention constructs an appliance topology knowledge base, which includes at least the mapping relationship between modules and appliance types, the dependencies and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance operating stages. When multiple modules are activated simultaneously, the system uses these rules to eliminate unreasonable combinations, thereby improving the reliability and interpretability of the identification results and avoiding logical errors that may arise from relying solely on probability matching.

[0137] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A dynamic energy metering and compensation method with self-identification of load type, characterized in that, Includes the following steps: S1. Obtain real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence; S2. Perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components; the signal mode components correspond to the oscillation characteristics of different loads in the electrical system; S3. For each of the signal mode components, extract the time-domain statistical features and frequency-domain spectral features of the signal mode components to form a dynamic feature vector; S4. Input the dynamic feature vector into the pre-trained load state analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment. S5. Based on the activation probability distribution, perform similarity matching in the preset module feature template library, and determine the modules in the running state based on the matching results; S6. Based on the module in operation, and combined with the preset electrical topology knowledge base, deduce the type of electrical appliance currently in operation and its working stage. The electrical appliance topology knowledge base includes at least: the mapping relationship between modules and appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance working stages; S7. Based on the type of electrical appliance and the working stage, extract the corresponding standard current waveform template from the module feature template library, and superimpose them to generate the theoretical current waveform of each electrical appliance. S8. Subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal; S9. Analyze the energy and characteristics of the residual current signal, and determine whether the residual current signal meets the background noise condition. If it does not meet the condition, store the characteristics of the residual current signal in the feature pool to be identified, and use it to update the load state analysis model in the future.

2. The dynamic energy metering and compensation method with self-identification of load type according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Perform multi-scale decomposition on the original electrical signal sequence to generate signal components at multiple frequency levels; S22. For each frequency level of signal component, identify the transient and periodic characteristics of the signal component, and separate the signal mode components that characterize different load oscillation modes.

3. The dynamic energy metering and compensation method with self-identification of load type according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Perform time-domain analysis on the signal modal components to extract the amplitude variation and time distribution characteristics of the signal modal components; S32. Perform frequency domain transformation on the signal mode components, extract the spectral distribution and dominant frequency components of the signal mode components, and combine them to form the dynamic feature vector.

4. The dynamic energy metering and compensation method with self-identification of load type according to claim 1, characterized in that, Step S4 includes: inputting the dynamic feature vector into the load state analysis model according to a preset format; and generating the activation probability distribution covering the prediction of the operating status of multiple modules through the internal hierarchical calculation of the load state analysis model.

5. The dynamic energy metering and compensation method with self-identification of load type according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51. Based on the activation probability distribution, modules with probability values ​​higher than a preset threshold are selected as candidate modules. S52. Compare the dynamic feature vector corresponding to the candidate module with the standard feature template in the module feature template library one by one, and calculate the similarity index. S53. Determine whether the candidate module is in operation based on whether the similarity index exceeds the matching threshold of the corresponding module.

6. The dynamic energy metering and compensation method for self-identification of load type according to claim 5, characterized in that, In step S52, the similarity index is calculated using cosine similarity, and the formula is as follows: ； Among them, subscript Indicates the index of the signal mode component. Indicates the module number; Indicates the first The dynamic eigenvector of the first signal mode component and the second... Cosine similarity between the standard feature templates of each module; Indicates the first The dynamic eigenvectors of each signal mode component This indicates the first module in the feature template library. The standard feature template for each module.

7. The dynamic energy metering and compensation method for self-identification of load type according to claim 1, characterized in that, Step S6 includes the following sub-steps: S61. Based on the module in operation, query the connection relationship between the module and the appliance stored in the appliance topology knowledge base; S62. Based on the connection relationship, deduce the combination state of the electrical appliance type and the working stage.

8. The dynamic energy metering and compensation method for self-identification of load type according to claim 1, characterized in that, Step S7 includes the following sub-steps: S71. Based on the electrical appliance type and working stage, locate the corresponding standard current waveform template in the module feature template library; S72. The standard current waveform template is superimposed to generate the theoretical current waveform that reflects the current characteristics of multiple electrical appliances operating in parallel.

9. A dynamic energy metering and compensation method for self-identification of load type according to claim 1, characterized in that, Steps S8 and S9 specifically include: S81. Calculate the point-by-point difference between the total current waveform of the original electrical signal sequence and the theoretical current waveform to generate the residual current signal; the calculation formula is as follows: ； in, Represents the residual current signal. This represents the original total current waveform. Indicates the first The theoretical current waveform of the electrical appliance; S82. Calculate the energy value and characteristic distribution of the residual current signal; S83. If the energy value is lower than the preset noise energy threshold and the feature distribution does not match the template in the module feature template library, it is determined to be background noise and ignored. S84. If the energy value is higher than or equal to the noise energy threshold, or if the feature distribution does not match the template in the module feature template library, then the features of the residual current signal are stored in the feature pool to be identified for subsequent updates to the load state analysis model.

10. A dynamic energy metering and compensation system with self-identification of load type, characterized in that, include: The data acquisition module is used to acquire real-time current and voltage waveform data of the main incoming line and generate the original electrical signal sequence; The signal decomposition module is used to perform a preset signal decomposition operation on the original electrical signal sequence to separate multiple signal mode components; The feature extraction module is used to extract the time-domain statistical features and frequency-domain spectral features of each of the signal modal components to form a dynamic feature vector; The load status analysis module is used to input the dynamic feature vector into the pre-trained load status analysis model and output the activation probability distribution of each potential electrical appliance's internal working module at the current moment. The module matching and determination module is used to perform similarity matching in a preset module feature template library according to the activation probability distribution, and determine the module in the running state based on the matching result; The appliance reasoning module is used to deduce the type of appliance currently in operation and its working stage based on the module in operation and a preset appliance topology knowledge base. The electrical appliance topology knowledge base includes at least: the mapping relationship between modules and appliance types, the dependency and mutual exclusion relationships between modules, and the correspondence rules between module combinations and appliance working stages; The theoretical waveform generation module is used to extract the corresponding standard current waveform template from the module feature template library according to the type of electrical appliance and the working stage, and superimpose them to generate the theoretical current waveform of each electrical appliance. The residual signal calculation module is used to subtract the theoretical current waveform from the total current waveform of the original electrical signal sequence to generate a residual current signal. The residual signal analysis module is used to analyze the energy and characteristics of the residual current signal, determine whether the residual current signal meets the background noise conditions, and if not, store the characteristics of the residual current signal in the feature pool to be identified for subsequent updates to the load state analysis model.

Images