Method and device for identifying embedded load of electric energy meter and electric energy meter

Abstract

The invention provides an electric energy meter embedded load identification method and device and an electric energy meter, and the method comprises the steps: carrying out the high-frequency sampling of a bus of a power grid, and obtaining a current parameter and a voltage parameter of the bus; segmenting the current parameters and the voltage parameters according to continuous time by adopting a preset edge algorithm, and obtaining windowed power utilization parameters after segmentation; a preset hybrid neural network model is adopted, electric appliance characteristics of the electricity utilization parameters are extracted, electricity utilization analysis results corresponding to different time periods of all the electricity utilization devices are output according to the electric appliance characteristics of the electricity utilization parameters, and the electric appliance characteristics comprise electricity utilization information corresponding to different electricity utilization devices and corresponding electricity utilization time sequence information; the electricity utilization analysis result comprises an electricity utilization equipment identification result and electricity utilization analysis data of each piece of electricity utilization equipment. Through combination of edge algorithm preprocessing and the hybrid neural network, high-precision load identification is realized, power utilization analysis results corresponding to different time periods of each power utilization device are output, and control of each power utilization device is facilitated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy meter, and particularly to an electric energy meter embedded load identification method and device and electric energy meter. BACKGROUND

[0002] With the development of technology, the electric energy meter used in family or public area is no longer a traditional electric quantity metering device, but gradually intelligent, which can record or analyze the energy consumption status of different devices.

[0003] The existing electric energy meter can only record the energy consumption state of the electric device simply, or upload to the service end of the power company for detailed analysis, for example, the power consumption of an electric appliance in a period of time can be analyzed. However, by using the prior art, it is dependent on the work of the service end, the efficiency is low, and the analysis result is simple, which has low reference value for power consumption and energy saving. SUMMARY

[0004] The purpose of the present application is to provide an electric energy meter embedded load identification method and device and electric energy meter, which obtains fine electric signals by high-frequency sampling of bus, combines edge algorithm preprocessing and hybrid neural network, obtains the power consumption analysis result of each electric device in different time periods, and realizes high-precision load identification.

[0005] To achieve the above purpose, the technical scheme adopted by the embodiments of the present application is as follows: In a first aspect, the embodiments of the present application provide an electric energy meter embedded load identification method applied to an electric energy meter, and the method comprises: High-frequency sampling is performed on the bus of the power grid to obtain current parameters and voltage parameters of the bus; wherein the current parameters include instantaneous current value, current effective value, current phase, current harmonic component, which are used to describe the energy consumption intensity and waveform characteristics of the electric device; the voltage parameters include instantaneous voltage value, voltage effective value, voltage phase, voltage fluctuation, which are used to calculate power and analyze the influence of power grid quality on load; The current parameters and the voltage parameters are divided according to continuous time by using a preset edge algorithm to obtain windowed electric parameters after division; A preset hybrid neural network model is used to extract appliance features of the electric parameters, and according to the appliance features of the electric parameters, power consumption analysis results corresponding to each electric device in different time periods are output, the appliance features include electric information and corresponding electric time sequence information of different electric devices, and the power consumption analysis results include electric device identification results and power consumption analysis data of each electric device.

[0006] In an optional implementation, the preset edge algorithm is used to segment the current parameter and the voltage parameter according to continuous time, and segmented and windowed power consumption parameters are obtained, including: The preset edge algorithm is used to segment the current parameter and the voltage parameter according to continuous time, and windowed currents and windowed voltages corresponding to multiple time windows are obtained. According to the windowed currents and the windowed voltages, a voltage-current trajectory graph is generated.

[0007] In an optional implementation, after the preset edge algorithm is used to segment the current parameter and the voltage parameter according to continuous time, and windowed currents and windowed voltages corresponding to multiple time windows are obtained, the method further includes: The power consumption parameters in each time window are calculated, including active power, reactive power, and power factor. According to the power consumption parameters in each time window, a time sequence vector corresponding to the power consumption parameters is generated.

[0008] In an optional implementation, the preset hybrid neural network model includes a convolutional neural network (CNN) model and a long short-term memory (LSTM) model. The preset hybrid neural network model is used to extract appliance features of the power consumption parameters, including: The voltage-current trajectory graph is input into the CNN model, and a voltage-current spatial feature vector is output through image dimension recognition. The time sequence vector corresponding to the power consumption parameters is input into the LSTM model, and a power information and time sequence dependency feature is output. According to the voltage-current spatial feature vector, the power information and the time sequence dependency feature, probabilities that different power consumption devices are turned on or turned off in different time windows are identified.

[0009] In an optional implementation, according to the voltage-current spatial feature vector, the power information and the time sequence dependency feature, probabilities that different power consumption devices are turned on or turned off in different time windows are identified, including: The voltage-current spatial feature vector, the power information and the time sequence dependency feature are fused to obtain a fused feature vector. According to the fused feature vector, different power consumption devices are identified, and probabilities that different power consumption devices are turned on or turned off in different time windows are output.

[0010] In an optional implementation, the preset hybrid neural network model is trained by using the CNN model, the LSTM model, and a preset data set, where the preset data set includes current data and voltage data corresponding to different use states or off states of different electrical equipment.

[0011] In an optional implementation, the preset hybrid neural network model is a model after light processing, where the light processing includes pruning training on the trained hybrid neural network model and re-quantization perception training on the model after the pruning training.

[0012] In an optional implementation, after the preset hybrid neural network model is used to extract the appliance features of the power consumption parameters and output power consumption analysis results corresponding to different time periods of each electrical equipment according to the appliance features of the power consumption parameters, the method further includes: generating a power consumption control strategy according to the power consumption analysis results; sending the power consumption control strategy to an intelligent appliance control platform.

[0013] In a second aspect, an embodiment of the present application provides an in-embedded load identification device of an electric energy meter, applied to an electric energy meter, and the device includes: a sampling module, configured to perform high-frequency sampling on a bus of a power grid to obtain current parameters and voltage parameters of a total incoming line; the current parameters include an instantaneous current value, a current effective value, a current phase, and a current harmonic component, and are used to describe energy consumption intensity and waveform characteristics of electrical equipment; the voltage parameters include an instantaneous voltage value, a voltage effective value, a voltage phase, and voltage fluctuation, and are used to calculate power and analyze the influence of power grid quality on load; a processing module, configured to perform windowing on the current parameters and the voltage parameters according to continuous time by using a preset edge algorithm to obtain windowed power consumption parameters after cutting; an extraction module, configured to extract appliance features of the power consumption parameters by using a preset hybrid neural network model, and output power consumption analysis results corresponding to different time periods of each electrical equipment according to the appliance features of the power consumption parameters, where the appliance features include power consumption information and power consumption time sequence information corresponding to different electrical equipment, and the power consumption analysis results include electrical equipment identification results and power consumption analysis data of each electrical equipment.

[0014] In a third aspect, an embodiment of the present application provides an electric energy meter, including a processor, a storage medium, and a bus, the storage medium stores program instructions executable by the processor, when the computer device runs, the processor and the storage medium communicate through the bus, and the processor executes the program instructions to perform the steps of the in-embedded load identification method of the electric energy meter in the first aspect.

[0015] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the power meter embedded load identification method of any of the first aspect.

[0016] Compared with the prior art, the power meter embedded load identification method, device and power meter provided by the embodiments of the present application have the following advantages. The method comprises: high-frequency sampling of a bus of a power grid to obtain current parameters and voltage parameters of the bus; wherein the current parameters comprise: instantaneous current value, current effective value, current phase, and current harmonic component, which are used to describe energy consumption intensity and waveform characteristics of a power consumption device; the voltage parameters comprise: instantaneous voltage value, voltage effective value, voltage phase, and voltage fluctuation, which are used to calculate power and analyze the influence of power grid quality on the load; a preset edge algorithm is used to divide the current parameters and the voltage parameters according to continuous time to obtain windowed power consumption parameters after division; a preset hybrid neural network model is used to extract appliance features of the power consumption parameters, and to output power consumption analysis results corresponding to each power consumption device in different time periods according to the appliance features of the power consumption parameters. The appliance features comprise: power consumption information corresponding to different power consumption devices and corresponding power consumption time sequence information, and the power consumption analysis results comprise: power consumption device identification results and power consumption analysis data of each power consumption device. The method provided by the present application can obtain fine electrical signals by high-frequency sampling of the bus, and can realize high-precision load identification by combining edge algorithm preprocessing and a hybrid neural network, thereby reducing data transmission power consumption and privacy leakage risk. The total power consumption is decomposed into each power consumption device, and power consumption analysis results corresponding to each power consumption device in different time periods are output, which is helpful for controlling each power consumption device.

[0017] In order to make the above objectives, characteristics and advantages of the present application more apparent, clear and easy to understand, the following will describe the preferred embodiments in detail, and the accompanying drawings will be referred to, and the detailed description will be as follows. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0019] Figure 1 One of the flowcharts of the power meter embedded load identification method provided by the embodiments of the present application; Figure 2 The second flowchart of the power meter embedded load identification method provided by the embodiments of the present application; Figure 3A flowchart of a third electric energy meter embedded load identification method provided by an embodiment of the present application; Figure 4 A flowchart of a fourth electric energy meter embedded load identification method provided by an embodiment of the present application; Figure 5 A flowchart of a fifth electric energy meter embedded load identification method provided by an embodiment of the present application; Figure 6 A flowchart of a sixth electric energy meter embedded load identification method provided by an embodiment of the present application; Figure 7 A functional module diagram of an electric energy meter embedded load identification device provided by an embodiment of the present application; Figure 8 A schematic diagram of an electric energy meter provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor, fall within the scope of protection of the present application.

[0022] In the description of the present application, it should be noted that if the terms "upper", "lower", etc. indicate the orientation or positional relationship shown in the drawings, or the orientation or positional relationship of the product in use, only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation on the present application.

[0023] Moreover, the terms "first", "second", and the like, in the description and in the claims of the present application, as well as above-mentioned drawings, are used to distinguish similar objects and are not necessarily used to describe a specific sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the present application described herein are capable of accomplishing the same object for example, an embodiment of the present application described herein and hereinafter is capable of accomplishing the same object regardless of the specific order or sequence previously illustrated or otherwise described herein. Furthermore, the terms "comprise" and "have", and variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product, or apparatus that comprises a list of steps or units not necessarily limited to those explicitly listed, but can include other steps or units not expressly listed or inherent to such process, method, product, or apparatus.

[0024] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0025] The electric energy meter embedded load identification method provided by the embodiments of the present application is applied to an electric energy meter. Please refer to Figure 1 , Figure 1 This is one of the flowcharts of the electric energy meter embedded load identification method provided by the embodiments of the present application. The method comprises the following steps. S101, high-frequency sampling is performed on the bus of the power grid to obtain current parameters and voltage parameters of the bus.

[0026] In this embodiment, the electric energy meter is used as a terminal measurement device of the power system to monitor the total power consumption in a household or industrial scene in real time. The purpose of high-frequency sampling is to obtain high-resolution electrical signal data to capture subtle power consumption characteristics of different electrical equipment, such as current mutation at switching moment, power fluctuation during steady-state operation, etc., to provide a raw data basis for subsequent load identification. Specifically, the electric energy meter can perform high-frequency synchronous sampling on the voltage and current of the signals transmitted by the bus of the power grid.

[0027] Among them, the electric energy meter integrates a high-precision analog-to-digital converter (ADC) chip such as ATT7022E at the hardware level, which internally uses a 24-bit ADC to achieve high-precision analog signal conversion. These ADCs can convert the analog voltage and current signals of the bus into digital signals, which have high resolution and can accurately capture the tiny changes in voltage and current. In order to improve the measurement accuracy and anti-interference ability, the signal input part of the electric energy meter often uses differential input mode. By comparing the voltage difference between the two input terminals to obtain the signal, common-mode noise can be effectively suppressed to ensure the accuracy of the collected current and voltage signals.

[0028] High-frequency sampling generally refers to a sampling frequency higher than the conventional electric energy metering (e.g. more than 10 times of the 50Hz power grid frequency, commonly 1kHz to 10kHz), which ensures the capture of high-frequency components such as appliance start-stop and harmonics, and requires an accurate clock signal to control the sampling frequency and timing, ensuring the accuracy and consistency of the sampling points, meeting the Nyquist sampling theorem, and avoiding aliasing.

[0029] Specifically, the microprocessor of the electric energy meter triggers sampling at a preset frequency through timer interrupts and the like. For example, 1500 timer interrupts are triggered per second, and the current voltage and current data are recorded each time the interrupt occurs, thereby achieving high-frequency accurate sampling of the transient changes of voltage and current thousands of times per second. The sampled data may contain noise and interference, and the electric energy meter will process the sampling data through a digital filter to remove high-frequency noise and other interference signals, ensuring the accuracy of the data used for final calculation and analysis. Finally, the collected high-frequency data will be temporarily stored in the internal storage area of the microprocessor of the electric energy meter, and when certain conditions are met, such as the number of recorded data in the internal storage area exceeding the preset number, the data will be transferred to the external storage, for subsequent analysis and processing, and the data can also be transmitted to the main station of the power system or other terminal devices through the communication module.

[0030] Among them, the current parameters include: instantaneous current value, current effective value, current phase, current harmonic component, which are used to describe the energy consumption intensity and waveform characteristics of the electrical equipment; the voltage parameters include: instantaneous voltage value, voltage effective value, voltage phase, voltage fluctuation, which are used to calculate power and analyze the influence of power grid quality on load.

[0031] S102, using a preset edge algorithm to divide the current parameters and voltage parameters according to continuous time to obtain windowed electrical parameters after division.

[0032] Specifically, the preset edge algorithm refers to a lightweight algorithm running locally on the electric energy meter, which is used to initially process the current parameters and voltage parameters to obtain electrical parameters. The purpose of initial processing is to clean, compress and preliminarily extract features from the original sampling data.

[0033] Using the preset edge algorithm, the continuously collected high-frequency current and voltage signals are divided into several time segments according to continuous time, and the dynamically changing electrical signals are converted into local data units that can be independently analyzed, so as to capture the electrical characteristics in different time intervals.

[0034] S103, using a preset hybrid neural network model to extract appliance features of the electrical parameters, and outputting electrical analysis results corresponding to different time periods of each electrical equipment according to the appliance features of the electrical parameters.

[0035] Among them, the electrical appliance characteristics include: the power consumption information and the corresponding power consumption time sequence information for different electrical appliances, and the power consumption analysis results include: the identification results of electrical appliances, and the power consumption analysis data of each electrical appliance.

[0036] Specifically, the pre-defined hybrid neural network combines the advantages of different networks to more comprehensively capture the characteristics of electrical appliances. These characteristics include power consumption information such as power output, waveform shape, and power consumption timing information such as start-up time, runtime, and state change patterns. The final output is the power consumption analysis results for each electrical device at different time periods. For devices including smart home devices, the power consumption analysis results may include: device identification results, i.e., identifying which devices are using the appliances at different time periods; and power consumption analysis data, including the on / off status of each device, and possibly real-time power, power consumption, and runtime of each device.

[0037] In summary, this application provides an embedded load identification method for electricity meters, applied to electricity meters. The method includes: high-frequency sampling of the power grid bus to obtain current and voltage parameters; wherein the current parameters include: instantaneous current value, effective current value, current phase, and current harmonic components, used to describe the energy consumption intensity and waveform characteristics of electrical equipment; the voltage parameters include: instantaneous voltage value, effective voltage value, voltage phase, and voltage fluctuation, used to calculate power and analyze the impact of power grid quality on the load; using a preset edge algorithm to segment the current and voltage parameters according to continuous time, obtaining windowed electricity consumption parameters after segmentation; using a preset hybrid neural network model to extract the electrical appliance features of the electricity consumption parameters, and outputting electricity consumption analysis results corresponding to different time periods for each electrical device based on the electrical appliance features of the electricity consumption parameters. The electrical appliance features include: electricity consumption information and corresponding electricity consumption time sequence information corresponding to different electrical devices; the electricity consumption analysis results include: electrical device identification results and electricity consumption analysis data for each electrical device. The method of this application obtains fine electrical signals by high-frequency sampling of the bus, and combines edge algorithm preprocessing and hybrid neural network to achieve high-precision load identification, reduce data transmission power consumption and privacy leakage risks; it decomposes the total power consumption to each power-consuming device and outputs the power consumption analysis results of each power-consuming device at different time periods, which helps to control each power-consuming device.

[0038] Figure 2 This is a second schematic flowchart illustrating an embedded load identification method for an electricity meter, as provided in this application embodiment. Figure 2 As shown, a preset edge algorithm is used to segment the current and voltage parameters according to continuous time, obtaining the segmented windowed power consumption parameters, including: S201. The current and voltage parameters are divided into continuous time segments using a preset edge algorithm to obtain the window current and window voltage corresponding to multiple time windows.

[0039] In this embodiment, the definition of the time window is to set a fixed time length (such as 10 ms, 50 ms) by a preset edge algorithm, to divide the continuous current / voltage sampling sequence into multiple non-overlapping or partially overlapping subsequences, that is, to split the current parameters and voltage parameters according to continuous time to obtain window currents and window voltages corresponding to multiple time windows. For example, if the sampling frequency is 1 kHz (1000 points per second), and a 50 ms window is set, then each window contains 50 sampling points. Among them, the window length needs to be set according to the power consumption characteristic scale: a short window, such as 10 ms, is suitable for capturing transient changes such as device start-stop, such as the current peak of motor start; a long window, such as 100 ms, is suitable for analyzing steady-state operation characteristics, such as the stable current of an incandescent lamp, and in addition, an overlap rate such as 50% can be set: to avoid feature loss at the window boundary, to ensure the integrity of the continuous signal.

[0040] S202, generating a voltage-current trajectory diagram according to the window current and the window voltage.

[0041] Specifically, the current data and voltage data in each time window are converted into two-dimensional image features, and the corresponding relationship between the voltage and the current in the time period is intuitively presented.

[0042] For example, for each power frequency cycle in each window, a closed V-I trajectory curve is drawn with the instantaneous voltage v(t) in the cycle as the X axis and the instantaneous current i(t) as the Y axis. The trajectory diagram is rasterized and normalized to generate a fixed-size two-dimensional grayscale image (for example, 32x32 pixels). Different types of electrical equipment will produce V-I trajectory diagrams with different shapes, resulting in an extremely recognizable appliance fingerprint. Among them, the power consumption parameters include: the voltage-current trajectory diagram.

[0043] In the method provided by the embodiment of the application, the preset edge algorithm is used to split the current parameters and voltage parameters according to continuous time, to obtain window currents and window voltages corresponding to multiple time windows; and a voltage-current trajectory diagram is generated according to the window current and the window voltage. The original high-frequency current parameters and voltage parameters are converted into a structured and visualized voltage-current trajectory diagram, which not only retains the detailed features of the local time interval, but also adapts to the feature extraction needs of the neural network in the form of an image, laying a foundation for accurate identification of the hybrid model, and relying on edge computing to realize efficient local processing.

[0044] Figure 3 As shown in FIG. 3, after the preset edge algorithm is used to split the current parameters and voltage parameters according to continuous time to obtain window currents and window voltages corresponding to multiple time windows, the method further includes: Figure 3 ​S301, calculate the power consumption parameter in each time window.

[0045] The power consumption parameter includes active power, reactive power, and power factor.

[0046] In this embodiment, key numerical parameters reflecting the energy consumption characteristics and electrical operating state of the power consumption equipment are extracted from the current and voltage data in the time window, to provide quantitative indicators for subsequent time series analysis.

[0047] Specifically, in the same window, macro parameters such as total active power P, reactive power Q, and power factor PF are calculated to form a time series vector S = [P(t), Q(t), PF(t)]. The active power refers to the power actually consumed in the circuit for doing work, and is calculated by integrating the instantaneous power (instantaneous voltage x instantaneous current) in the time window and then taking the average. The reactive power is calculated based on the phase difference between the voltage and the current through the orthogonal component of the instantaneous power. The power factor is the ratio of the active power to the apparent power (voltage effective value x current effective value), and is used to reflect the power consumption efficiency.

[0048] S302, generate a time series vector corresponding to the power consumption parameter according to the power consumption parameter in each time window.

[0049] Specifically, the power consumption parameters (active power, reactive power, and power factor) of the discrete time windows are concatenated in time sequence to form a continuous time series feature sequence, i.e., a time series vector, to capture the dynamic change law of the power consumption parameter over time.

[0050] The structure of the time series vector can be that N continuous windows (such as window 1, window 2, …, window N) are divided in time sequence, and the power consumption parameters of each window are [Pi, Qi, PFi] (i is the window number), and the time series vector is: [[P1, Q1, PF1], [P2, Q2, PF2], …, [Pn, Qn, PFn]], to obtain a two-dimensional vector of n x 3 (n is the number of windows, and 3 is the parameter dimension). The processed power consumption parameters include the time series vector.

[0051] In the method provided by the embodiment of the application, the power consumption parameter in each time window is calculated, and the power consumption parameter includes active power, reactive power, and power factor. The time series vector corresponding to the power consumption parameter is generated according to the power consumption parameter in each time window. The electrical signal in the time window is converted into a data form with quantitative characteristics and time correlation: parameters such as active / reactive power, which quantifies the energy consumption and electrical characteristics of the power consumption equipment, and the time series vector retains the dynamic change law of these parameters over time.

[0052] Figure 4FIG. 4 is a flowchart of a fourth embodiment of an embedded load identification method of an electric energy meter according to the present application. The preset hybrid neural network model includes a convolutional neural network (CNN) model and a long short-term memory (LSTM) model. As shown in FIG. 4, the preset hybrid neural network model is used to extract appliance features of the power consumption parameters, including: Figure 4 S401, inputting the voltage-current trajectory graph into the CNN model, identifying the image dimension, and outputting a voltage-current spatial feature vector.

[0053] In this embodiment, the convolutional neural network (CNN) is used to capture the spatial fingerprints of different electrical equipment from the voltage-current (V-I) trajectory graph by virtue of the strong extraction capability of the image features, and the visual features of the image are converted into quantifiable vector representation.

[0054] Specifically, the 32x32 pixel voltage-current trajectory graph is inputted. The CNN model includes a convolution layer C1 with 16 3x3 convolution kernels and a ReLU activation function, which is used to extract low-level features such as edges and corner points of the trajectory graph. A maximum pooling layer P1 with a 2x2 pooling window is used to reduce the dimension of the feature map and retain the most significant features. A convolution layer C2 with 32 3x3 convolution kernels and a ReLU activation function is used to combine the low-level features to form more complex shape features. A maximum pooling layer P2 with a 2x2 pooling window is used. A flattening layer is used to convert the two-dimensional feature map output by C2 into a one-dimensional feature vector, and thus the voltage-current spatial feature vector corresponding to the voltage-current trajectory graph is obtained.

[0055] S402, inputting the time sequence vector corresponding to the power consumption parameters into the LSTM model, and outputting the power information and the time sequence dependency feature.

[0056] The long short-term memory (LSTM) network is used to model the time sequence data, capture the dynamic change rule of the power consumption parameters (active power, reactive power, power factor) over time, and extract the time sequence fingerprints of the electrical equipment operation.

[0057] Specifically, the input is a time sequence vector S_seq = [S(t-N+1),..., S(t)] with a length of N (for example, N = 10, representing 10 consecutive time steps). The LSTM model includes an LSTM layer L1 with 64 hidden units, which can learn the time dependence of the time sequence vector. The output is the output vector of the LSTM layer at the last time step t. Thus, the power information and the time sequence dependency feature are obtained.

[0058] S403, identifying the probability of turning on or off of different electrical equipment in different time windows according to the voltage-current spatial feature vector, the power information, and the time sequence dependency feature. ​

[0059] Specifically, the voltage-current space feature vector extracted by the CNN model and the power information and time sequence dependent relationship feature extracted by the LSTM model are comprehensively judged to determine the running state of each electrical equipment in each time window, and a probability value is output to reflect the reliability of the judgment.

[0060] It should be noted that after identifying the probabilities of different electrical equipment being turned on or off in different time windows, the power consumption information of the electrical equipment in the on state can be further obtained, and the power consumption analysis results of each electrical equipment in different time periods can be output.

[0061] In the method provided by the embodiment of the application, the voltage-current trajectory graph is input into the CNN model, the voltage-current space feature vector is output through image dimension recognition, the time sequence vector corresponding to the power consumption parameter is input into the LSTM model, and the power information and time sequence dependent relationship feature are output. According to the voltage-current space feature vector, the power information and the time sequence dependent relationship feature, the probabilities of different electrical equipment being turned on or off in different time windows are identified. The electrical appliance features are comprehensively characterized, the spatial features distinguish the inherent electrical properties of different equipment, the time sequence features capture the running mode of the equipment over time, and the combination of the two greatly improves the accuracy and robustness of load identification. The final output probability value can not only determine the state of each electrical equipment, but also reflect the reliability of identification through the probability size, and provide fine basic data for subsequent power consumption analysis.

[0062] Figure 5 A flowchart of a power meter embedded load identification method provided by the embodiment of the application is shown in Figure 5. Figure 5 As shown in Figure 5, according to the voltage-current space feature vector, the power information and the time sequence dependent relationship feature, the probabilities of different electrical equipment being turned on or off in different time windows are identified, which includes: S501, the voltage-current space feature vector, the power information and the time sequence dependent relationship feature are fused to obtain a fused feature vector.

[0063] In this embodiment, the voltage-current space feature vector, the power information and the time sequence dependent relationship feature are integrated to form a comprehensive feature containing multi-dimensional information, avoiding the limitation of single feature dimension and improving the accuracy of subsequent identification.

[0064] The following methods can be used for fusion, specifically: (1) Directly concatenate the two types of feature vectors by dimension to form a longer comprehensive vector. For example: the voltage-current space feature vector is 256 (encoding trajectory shape, harmonic characteristics, etc.); the time-dependent relationship feature vector is 128 dimensions (encoding power change trend, start-stop timing, etc.); after fusion, it is a 256+128=384-dimensional feature vector, retaining the complete information of the two types of features. The advantage is simple and direct, without losing the details of the original features, suitable for scenarios where the two types of features are equally important.

[0065] (2) Assign different weights to the two types of feature vectors and sum them up, with the weights automatically learned through model training or manually set. For example: fusion vector = a x space feature vector + (1-a) x relationship feature vector (a is the weight, 0

[0066] (3) Introduce an attention layer to let the model automatically learn the importance of different features in different time windows. For example: calculate the attention weights of space features and relationship features respectively (such as the weight of space features in a certain window is 0.7, and the weight of relationship features is 0.3); weight the sum according to the weight to get the fusion vector, and realize dynamic focusing on key features. The advantage is that it can adapt to different scenarios (such as when complex loads are superimposed, it automatically amplifies the weight of features with high discrimination).

[0067] S502, according to the fused feature vector, identify different electrical equipment, and output the probability of different electrical equipment being turned on or off in different time windows.

[0068] Based on the fused comprehensive features, the classifier determines which appliances are in operation in each time window, and quantifies the reliability of the judgment in the form of probability, providing a detailed state basis for subsequent power analysis.

[0069] Specifically, input the fused feature vector, and the classification model uses a structure of fully connected layer and softmax activation function. The fully connected layer: maps high-dimensional fused features to a dimension corresponding to the type of appliance to be identified, completing the conversion of features to class probabilities. The softmax function: normalizes the output of the fully connected layer to a probability value (each dimension probability ∈ [0, 1], and all dimension probabilities sum to 1), ensuring that the output conforms to the probability distribution characteristics. For example, output an M-dimensional probability vector Y = [y_1, y_2,..., y_M], where y_i represents the probability of the i-th appliance being in the on state.

[0070] The method provided in the embodiments of the present application fuses the dependence relationship features of the voltage-current space feature vector, power information and time sequence, and obtains a fused feature vector; different electrical equipment is identified according to the fused feature vector, and the probability of turning on or off of different electrical equipment corresponding to different time windows is output. The precise mapping from multi-dimensional features to electrical appliance states is realized: the fusion process integrates the advantages of space and time sequence features, and makes up for the information loss of a single feature; the probability output provides the running state and reliability of each electrical appliance in a quantitative form, which supports single device recognition and can also handle complex scenes of multiple devices running, and provides basic data for subsequent power consumption analysis.

[0071] The possible implementation manner of the power meter embedded load identification method provided in the embodiments of the present application is that a preset hybrid neural network model is obtained by using a CNN model, an LSTM model and a preset data set, wherein the preset data set includes current data and voltage data corresponding to different electrical equipment use states or off states.

[0072] In the embodiments, the architecture of the preset hybrid neural network model is built, wherein the CNN model part includes an initialized convolution layer, a pooling layer and a full connection layer (such as 3 layers of convolution + 2 layers of full connection), which are used to extract the space features of the V-I trajectory graph; the LSTM model part includes an initialized LSTM unit and a subsequent full connection layer, which are used to extract the dynamic features of the time sequence vector; and the fusion layer includes an initialized feature fusion manner (such as splicing and weighted fusion) and a final classification layer (full connection layer + softmax), which are used to output the electrical appliance switching probability.

[0073] The preset data set is preprocessed to obtain training samples including the voltage-current trajectory graph and the time sequence vector corresponding to the electrical parameter, and the training samples are respectively input into the CNN model and the LSTM model, wherein the V-I trajectory graph is input into the CNN model, the voltage-current space feature vector is output through convolution and pooling operation, the time sequence vector is input into the LSTM model, and the power time sequence dependence feature vector is output through the gating mechanism; and finally, the fusion layer integrates the two types of feature vectors, and the classification layer outputs the predicted switching probability of each electrical appliance (such as the predicted air conditioner turning-on probability of a certain sample being 0.8).

[0074] The cross-entropy loss is used to calculate the difference between the model prediction probability and the true label (such as the air conditioner label being 1). The smaller the loss value is, the closer the prediction is to the true state. Through the gradient descent algorithm, the loss value is reversely transmitted from the output layer to each layer of the CNN and the LSTM, and the network parameters (weights and biases) are adjusted to reduce the loss. For example, if the model prediction probability of the microwave oven turning on is low, the convolution kernel weight related to the microwave oven V-I trajectory feature in the CNN is increased through the back propagation.

[0075] Finally, the preset data set is divided into a training set (70%, used for parameter learning), a validation set (20%, used for monitoring overfitting), and a test set (10%, used for final performance evaluation). The forward propagation, loss calculation, and back propagation process are repeated, and the performance is evaluated on the validation set after each iteration. If the performance on the validation set decreases, it indicates that the model is overfitting the training data, and optimization methods such as early stopping and regularization need to be used. When the training set loss stabilizes and the validation set performance no longer improves, the training is stopped, and the model parameters at this time are saved, which is the trained preset hybrid neural network model.

[0076] Optionally, the preset hybrid neural network model is a light processing model, wherein the light processing includes pruning training on the trained hybrid neural network model, and re-quantization perception training on the pruned model.

[0077] Specifically, the essence of pruning training is to remove redundant or unimportant parameters / structures in the model, and to retain the core part that is critical to identification performance. While reducing the number of parameters and computational load, it avoids significant accuracy degradation.

[0078] After the model training is completed, the importance of each layer parameter is analyzed (usually measured by the absolute value of the weight), and redundant parameters with little impact on the output are selected: CNN part: identify low-contribution convolution kernels, unimportant neurons in the fully connected layer; LSTM part: identify parameters with weights close to 0 in the gating mechanism (input gate, forget gate, output gate), or hidden layer units with low contribution to time series dependence modeling. Set a threshold (e.g., retain the top 80% of parameters by absolute weight value), and directly remove redundant parameters below the threshold: for CNN: delete redundant convolution kernels, or delete corresponding neurons in the fully connected layer; for LSTM: delete redundant weights in the gating mechanism, or reduce the number of hidden layer units; after pruning, the number of model parameters can be reduced, and the network structure is more streamlined.

[0079] Pruning may disrupt the original parameter balance, resulting in temporary accuracy degradation. Through quantization perception training on the preset data set, the accuracy redundancy problem is solved, and the storage demand and running energy consumption are further reduced; the final light model can maintain load identification accuracy under the limited hardware resources, low computing power, small memory, and low power consumption of the electric energy meter, achieving a balance between high accuracy and embedded deployment.

[0080] Figure 6 A flowchart of an electric energy meter embedded load identification method provided by an embodiment of the present application is shown in Figure 6, which uses a preset hybrid neural network model to extract appliance features of electric parameters, and outputs electric analysis results of each electric device corresponding to different time periods according to the appliance features of electric parameters. Figure 6 ​S601, generating a power consumption control strategy according to the power consumption analysis result.

[0081] S602, sending the power consumption control strategy to the electric appliance intelligent control platform.

[0082] In the embodiment, based on the power consumption analysis result output by the hybrid neural network, including the power consumption equipment identification result and the power consumption analysis data of each power consumption equipment, an executable power consumption control strategy is formulated, passive monitoring is converted into active control, and intelligent power consumption management is realized.

[0083] For example, in the power consumption analysis result, the power consumption equipment A is in the on state at 12 o'clock-2 o'clock and 18 o'clock-21 o'clock, and the power consumption control strategy for the power consumption equipment A is generated, that is, the power consumption equipment A is automatically controlled by the electric appliance intelligent control platform to be turned on at 12 o'clock-2 o'clock and 18 o'clock-21 o'clock in the future, and the power consumption equipment A is not controlled at other times. In the same way, the same method is adopted for other power consumption equipment in the power consumption analysis result, and the power consumption control strategy is generated.

[0084] Based on the power consumption analysis result, the on-off state of each power consumption equipment in different time periods is determined, and an automatic control strategy is generated, which can accurately match the future start-stop period of the equipment, reduce standby and invalid operation energy consumption, and reduce the user's power consumption cost; it can also actively adapt to the user's historical power consumption habits, realize the automatic start-stop of the power consumption equipment, and eliminate the manual operation burden, improve the use convenience; at the same time, it can avoid the disordered start-stop or overload operation of the power consumption equipment, prolong the service life of the equipment, and also assist the power grid to balance the peak-valley load and optimize the overall energy utilization efficiency.

[0085] The following continues to explain the electric energy meter embedded load identification device and the electric energy meter provided by any of the above embodiments of the present application, and the specific implementation process and the technical effects produced are the same as those of the corresponding method embodiments. For brevity, the parts not mentioned in this embodiment can be referred to the corresponding contents in the method embodiments.

[0086] Figure 7 A functional module schematic diagram of an electric energy meter embedded load identification device provided by an embodiment of the present application is shown in FIG. 1. Figure 7 As shown in FIG. 1, the electric energy meter embedded load identification device 100 includes: A sampling module 110 is configured to sample a bus of a power grid at a high frequency to obtain current parameters and voltage parameters of the bus; wherein the current parameters include: instantaneous current value, current effective value, current phase, and current harmonic component, which are used to describe the energy consumption intensity and waveform characteristics of the power consumption equipment; and the voltage parameters include: instantaneous voltage value, voltage effective value, voltage phase, and voltage fluctuation, which are used to calculate power and analyze the influence of power grid quality on the load. The processing module 120 is configured to divide the current parameter according to continuous time by using a preset edge algorithm to obtain a windowed power consumption parameter after division. The extraction module 130 is configured to extract an electrical appliance feature of the power consumption parameter by using a preset hybrid neural network model, and output power consumption analysis results of each power consumption device corresponding to different time periods according to the electrical appliance feature of the power consumption parameter. The electrical appliance feature includes power consumption information and corresponding power consumption timing information of different power consumption devices, and the power consumption analysis result includes a power consumption device identification result and power consumption analysis data of each power consumption device.

[0087] Optionally, the processing module 120 is further configured to divide the current parameter and the voltage parameter according to continuous time by using a preset edge algorithm to obtain windowed current and windowed voltage corresponding to a plurality of time windows; and generate a voltage-current trajectory graph according to the windowed current and the windowed voltage.

[0088] Optionally, the processing module 120 is further configured to calculate a power consumption parameter in each time window, the power consumption parameter including active power, reactive power, and power factor; and generate a timing vector corresponding to the power consumption parameter according to the power consumption parameter in each time window.

[0089] Optionally, the preset hybrid neural network model includes a convolutional neural network (CNN) model and a long short-term memory (LSTM) model; the extraction module 130 is further configured to input the voltage-current trajectory graph into the CNN model, identify the voltage-current trajectory graph by image dimension, and output a voltage-current spatial feature vector; input the timing vector corresponding to the power consumption parameter into the LSTM model, and output power information and a timing dependency feature; and identify a probability of turning on or off of different power consumption devices corresponding to different time windows according to the voltage-current spatial feature vector, the power information, and the timing dependency feature.

[0090] Optionally, the extraction module 130 is further configured to fuse the voltage-current spatial feature vector, the power information, and the timing dependency feature to obtain a fused feature vector; and identify different power consumption devices and output the probability of turning on or off of different power consumption devices corresponding to different time windows according to the fused feature vector.

[0091] Optionally, the preset hybrid neural network model is obtained by training a CNN model, an LSTM model, and a preset data set, wherein the preset data set includes current data and voltage data corresponding to a use state or an off state of different power consumption devices.

[0092] Optionally, the preset hybrid neural network model is a model after light-weight processing, wherein the light-weight processing includes pruning training on the trained hybrid neural network model and re-quantization perception training on the model after pruning training.

[0093] Optionally, the apparatus further comprises: a generating module configured to generate a power consumption control strategy according to the power consumption analysis result; a sending module configured to send the power consumption control strategy to an electric appliance intelligent control platform.

[0094] The apparatus is configured to execute the method provided by the foregoing embodiments, and has similar implementation principles and technical effects, which will not be described here again.

[0095] The above modules can be one or more integrated circuits configured to implement the above method, for example, one or more application specific integrated circuits (ASICs), or one or more microprocessors, or one or more field programmable gate arrays (FPGAs), etc. For another example, when a certain module above is implemented in the form of a processing element scheduling code, the processing element can be a general-purpose processor, for example, a central processing unit (CPU) or other processor that can invoke program code. For another example, the modules can be integrated together to be implemented in the form of a system on a chip (SOC).

[0096] Figure 8 A schematic diagram of an electric energy meter provided by an embodiment of the present application is shown in FIG. 1. The electric energy meter can be used for electric energy meter embedded load identification. As shown in FIG. 1, the computer device includes a processor 210, a storage medium 220, and a bus 230. Figure 8

[0097] The storage medium 220 stores machine readable instructions executable by the processor 210. When the computer device is running, the processor 210 and the storage medium 220 communicate with each other through the bus 230. The processor 210 executes the machine readable instructions to execute the steps of the above method embodiments. The specific implementation manners and technical effects are similar, which will not be described here again.

[0098] Optionally, the present application further provides a storage medium 220, and the storage medium 220 stores a computer program. When the computer program is run by the processor, the steps of the above method embodiments are executed. The specific implementation manners and technical effects are similar, which will not be described here again.

[0099] ​In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. The units as divided can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0100] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0101] In addition, each functional unit in the various embodiments of the present application can be integrated into a processing unit, or each unit can be a physically independent unit, or two or more units can be integrated into a unit. The integrated unit can be implemented in the form of hardware, or in the form of hardware plus software functional units.

[0102] The integrated unit implemented in the form of software functional units can be stored in a computer readable storage medium. The software functional units stored in the storage medium include a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to execute some steps of the method described in the various embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a magnetic disk or an optical disk, and various program code storage media.

[0103] The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for embedded load identification in an electricity meter, characterized in that, Applied to electricity meters, the method includes: The power grid bus is sampled at high frequency to obtain the current and voltage parameters of the bus. The current parameters include instantaneous current value, effective current value, current phase, and current harmonic components, which are used to describe the energy consumption intensity and waveform characteristics of electrical equipment. The voltage parameters include instantaneous voltage value, effective voltage value, voltage phase, and voltage fluctuation, which are used to calculate power and analyze the impact of power grid quality on the load. The current parameters and voltage parameters are segmented according to continuous time using a preset edge algorithm to obtain the segmented windowed power consumption parameters; A preset hybrid neural network model is used to extract the electrical appliance features of the power consumption parameters, and output the power consumption analysis results of each electrical device for different time periods based on the electrical appliance features of the power consumption parameters. The electrical appliance features include: power consumption information and corresponding power consumption time sequence information for different electrical devices. The power consumption analysis results include: electrical device identification results and power consumption analysis data for each electrical device.

2. The method according to claim 1, characterized in that, The step of using a preset edge algorithm to segment the current and voltage parameters according to continuous time to obtain windowed power consumption parameters after segmentation includes: The preset edge algorithm is used to divide the current parameter and the voltage parameter into continuous time segments to obtain window current and window voltage corresponding to multiple time windows; A voltage-current trajectory diagram is generated based on the window current and the window voltage.

3. The method according to claim 2, characterized in that, After using the preset edge algorithm to segment the current and voltage parameters according to continuous time to obtain window current and window voltage corresponding to multiple time windows, the method further includes: Calculate the power consumption parameters within each time window, including: active power, reactive power, and power factor; Generate a time-series vector corresponding to the electricity consumption parameters based on the electricity consumption parameters within each time window.

4. The method according to claim 3, characterized in that, The preset hybrid neural network model includes: a convolutional neural network (CNN) model and a long short-term memory (LSTM) network model; The method of extracting electrical appliance features from the electricity consumption parameters using a preset hybrid neural network model includes: The voltage-current trajectory diagram is input into the CNN model, and the voltage-current spatial feature vector is output through image dimension recognition. The time-series vector corresponding to the power consumption parameters is input into the LSTM model, and the power information and time-series dependency features are output. Based on the voltage-current spatial feature vector, the power information, and the time-series dependency characteristics, the probability of different electrical devices being turned on or off in different time windows is identified.

5. The method according to claim 4, characterized in that, The step of identifying the probability of different electrical devices being turned on or off in different time windows based on the voltage-current spatial feature vector, the power information, and the time-series dependency features includes: The voltage-current spatial feature vector, the power information, and the time-series dependency features are fused to obtain the fused feature vector; Based on the fused feature vector, different electrical devices are identified, and the probability of different electrical devices being turned on or off in different time windows is output.

6. The method according to claim 5, characterized in that, The preset hybrid neural network model is trained using the CNN model, the LSTM model, and a preset dataset, wherein the preset dataset includes current data and voltage data corresponding to different power equipment usage states or off states.

7. The method according to claim 6, characterized in that, The preset hybrid neural network model is a lightweight model, wherein the lightweight processing includes: pruning the trained hybrid neural network model and then performing quantization-aware training on the pruned model.

8. The method according to claim 1, characterized in that, After extracting the electrical appliance characteristics of the power consumption parameters using a preset hybrid neural network model, and outputting the power consumption analysis results for each electrical device at different time periods based on the electrical appliance characteristics of the power consumption parameters, the method further includes: Based on the power consumption analysis results, a power consumption control strategy is generated; Send the power control strategy to the appliance intelligent control platform.

9. A load identification device embedded in an electricity meter, characterized in that, Applied to electricity meters, the device includes: The sampling module is used to perform high-frequency sampling of the power grid bus to obtain the current and voltage parameters of the main incoming line. The current parameters include: instantaneous current value, effective current value, current phase, and current harmonic components, used to describe the energy consumption intensity and waveform characteristics of electrical equipment. The voltage parameters include: instantaneous voltage value, effective voltage value, voltage phase, and voltage fluctuation, used to calculate power and analyze the impact of power grid quality on the load. The processing module is used to segment the current parameters and the voltage parameters according to continuous time using a preset edge algorithm to obtain the segmented windowed power consumption parameters; The extraction module is used to extract the electrical appliance features of the power consumption parameters using a preset hybrid neural network model, and output the power consumption analysis results corresponding to different time periods for each electrical appliance based on the electrical appliance features of the power consumption parameters. The electrical appliance features include: power consumption information and corresponding power consumption time sequence information for different electrical appliances. The power consumption analysis results include: electrical appliance identification results and power consumption analysis data for each electrical appliance.

10. An electricity meter, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores program instructions executable by the processor. When the computer device is running, the processor communicates with the storage medium via the bus. The processor executes the program instructions to perform the steps of the embedded load identification method for an energy meter as described in any one of claims 1 to 8.

Images