Load identification method and electronic equipment

Abstract

The invention discloses a load identification method and electronic equipment, and belongs to the technical field of electric power. The load identification method is applied to the electric energy meter, and the method comprises the steps: obtaining total power data which are collected by a target electric energy meter; extracting time-frequency characteristics of the total power data to obtain first characteristic information; extracting a time sequence feature of the total power data to obtain second feature information; and based on the first feature information and the second feature information, obtaining a load identification result of the target electric energy meter, the load identification result including load information of an electric device connected with the target electric energy meter. The method can improve the discrimination capability of the total power data features, thereby improving the accuracy of the load recognition result of the target electric energy meter.

Description

Technical Field

[0001] This application belongs to the field of power technology, and in particular relates to a load identification method and electronic device. Background Technology

[0002] Current load identification technologies are mainly divided into two types based on their monitoring methods: invasive and non-invasive. Among them, non-invasive load monitoring (NILM) uses algorithms to identify the status of each appliance by collecting total user electricity consumption data from smart meters. It has advantages such as convenient deployment and low cost, and has become the mainstream solution for residential electricity consumption analysis.

[0003] Currently, non-invasive load identification methods mostly focus on a single feature and cannot take into account global features, resulting in low accuracy of load identification. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a load identification method and electronic device, which can improve the discrimination capability of total power data characteristics, thereby improving the accuracy of load identification results of the target energy meter.

[0005] In a first aspect, this application provides a load identification method, which is applied to an electricity meter, the method comprising: Acquire total power data, which is collected by the target energy meter; Extract the time-frequency features of the total power data to obtain the first feature information; Extract the time-series features of the total power data to obtain the second feature information; Based on the first feature information and the second feature information, the load identification result of the target energy meter is obtained, and the load identification result includes the load information of the electrical equipment connected to the target energy meter.

[0006] According to the load identification method of this application, by acquiring the total power data collected by the target energy meter, the time-frequency features and time-series features of the total power data are extracted to obtain the first feature information and the second feature information, respectively. The time-frequency features can capture the transient characteristics of the starting and stopping of electrical equipment and frequency changes, while the time-series features can reflect the power change trend and periodic pattern of the electrical equipment. By comprehensively analyzing the local details of the total power data reflected by the first feature information and the global dynamics of the total power data on the time axis reflected by the second feature information, the time-frequency features and the time-series features are complementary, which can improve the discrimination ability of the total power data features, thereby improving the accuracy of the load identification results of the target energy meter.

[0007] According to one embodiment of this application, the step of extracting the time-frequency features of the total power data to obtain first feature information includes: The total power data is converted into a time-frequency image; The spatial texture features of the time-frequency image are extracted to obtain the first feature information.

[0008] According to one embodiment of this application, converting the total power data into a time-frequency image includes: The total power data is converted into a time-frequency image by local maximum synchronous compression transformation.

[0009] According to one embodiment of this application, the step of extracting the spatial texture features of the time-frequency image to obtain the first feature information includes: The time-frequency image is input into a convolutional neural network with a three-level depth separable convolutional structure. The spatial texture features of the time-frequency image are extracted by the convolutional neural network to obtain the first feature information output by the convolutional neural network.

[0010] According to one embodiment of this application, the end of each level of the depth-separable convolutional structure in the convolutional neural network includes a max-pooling layer.

[0011] According to one embodiment of this application, the step of extracting the time-series features of the total power data to obtain second feature information includes: The total power data is input into a temporal convolutional network, and the temporal features of the total power data are extracted through the temporal convolutional network to obtain the second feature information output by the temporal convolutional network.

[0012] According to one embodiment of this application, the temporal convolutional network is pruned.

[0013] According to one embodiment of this application, obtaining the load identification result of the target energy meter based on the first feature information and the second feature information includes: The first feature information and the second feature information are assigned corresponding weights through an attention mechanism; Based on the weights corresponding to the first feature information and the second feature information, the first feature information and the second feature information are weighted and summed to obtain the third feature information; The load identification result is obtained based on the third feature information.

[0014] According to one embodiment of this application, the load identification method is executed through a target load identification model. The target load identification model receives the total power data, extracts the time-frequency features of the total power data to obtain first feature information, extracts the time-series features of the total power data to obtain second feature information, and outputs the load identification result based on the first feature information and the second feature information. The target load identification model is trained through the following steps: Acquire sample power data of each electrical device connected to the target energy meter; Based on the power data of each sample, the total power data of the target energy meter is obtained. The target load identification model is trained using total power data from the samples.

[0015] In a second aspect, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the load identification method as described in the first aspect above.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is one of the flowcharts illustrating the load identification method provided in the embodiments of this application; Figure 2 This is a second schematic flowchart of the load identification method provided in the embodiments of this application; Figure 3 This is the third flowchart illustrating the load identification method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the convolutional neural network provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the temporal convolutional network provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of the residual block provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] The load identification method, electronic device, and readable storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0021] The load identification method can be applied to the terminal, and can be executed by the hardware or software in the terminal.

[0022] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0023] The following embodiments describe a terminal including a display and a touch-sensitive surface. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, mouse, and joystick.

[0024] The load identification method provided in this application embodiment can be executed by an electronic device or a functional module or entity in an electronic device that can implement the load identification method. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The load identification method provided in this application embodiment will be described below using an electronic device as the execution subject.

[0025] The load identification method provided in this application is applied to electricity meters.

[0026] In this embodiment, load identification is performed in a non-intrusive manner, based on the total power consumption data of multiple electrical devices collected by the electricity meter, to identify the power load of each electrical device.

[0027] like Figure 1 As shown, the load identification method includes steps 110-140.

[0028] Step 110: Obtain total power data.

[0029] The total power data is collected by the target energy meter. The total power data can be a sequence of active power data collected by the target energy meter recorded in chronological order. The target energy meter is the energy meter to be identified as the load.

[0030] In this step, the target energy meter detects the power of the connected electrical equipment in real time, obtains the total power data, and reads the detected total power data from the target energy meter.

[0031] Step 120: Extract the time-frequency features of the total power data to obtain the first feature information.

[0032] Among them, the first feature information is the information that can characterize the time-frequency characteristics of the total power data. The time-frequency characteristics are the features obtained by analyzing the total power data in two dimensions: time and frequency. They can reveal the pattern of the total power data changing over time and the distribution of the total power data in different frequency components.

[0033] In this step, time-frequency analysis of the total power data can be performed using short-time Fourier transform or wavelet transform to extract the energy distribution of the total power data at different times and frequencies, thereby obtaining the first feature information.

[0034] Step 130: Extract the time-series features of the total power data to obtain the second feature information.

[0035] The second feature information is information that can characterize the time-series features of the total power data. The time-series features are extracted from the total power data and can reflect the characteristics of the total power data changing over time. They can indicate the internal structure, trend, periodicity, and randomness of the total power data.

[0036] In this step, the sliding window can be used to calculate the mean, variance, maximum and minimum values ​​of the total power data, or to extract the periodic features of the total power data, so as to obtain the second feature information that reflects the time dynamics of the total power data.

[0037] Step 140: Based on the first feature information and the second feature information, obtain the load identification result of the target electricity meter.

[0038] The load identification results include load information of electrical equipment connected to the target electricity meter.

[0039] The first and second feature information can be fused and input into a trained classification model. The model learns the power characteristic patterns of different electrical devices, thereby identifying the working status of the electrical devices connected to the target electricity meter and obtaining load identification results containing information such as the start and stop time and power consumption of the electrical devices.

[0040] According to the load identification method provided in the embodiments of this application, by acquiring the total power data collected by the target energy meter, the time-frequency features and time-series features of the total power data are extracted to obtain the first feature information and the second feature information, respectively. The time-frequency features can capture the transient characteristics of the start-up and shutdown and frequency changes of the electrical equipment, while the time-series features can reflect the power change trend and periodic pattern of the electrical equipment. By comprehensively analyzing the local details of the total power data reflected by the first feature information and the global dynamics of the total power data on the time axis reflected by the second feature information, the time-frequency features and the time-series features are complementary, which can improve the discrimination ability of the total power data features, thereby improving the accuracy of the load identification result of the target energy meter.

[0041] In some embodiments, the time-frequency features of the total power data are extracted to obtain first feature information, including: Convert the total power data into a time-frequency image; Spatial texture features are extracted from the time-frequency image to obtain the first feature information.

[0042] Among them, the time-frequency image is a two-dimensional image that visualizes the one-dimensional total power data in both time and frequency dimensions.

[0043] In this embodiment, the total power data is converted into a time-frequency image through time-frequency transformation, which can intuitively display the energy distribution of the total power data in time and frequency. The spatial texture features of the time-frequency image are extracted to reflect the local patterns, contrast and regularity of the energy distribution in the time-frequency image, thereby capturing the typical start-up and shutdown and spectrum change features of different electrical devices and obtaining the first feature information.

[0044] In some embodiments, converting total power data into a time-frequency image includes: The total power data is converted into a time-frequency image by using local maximum synchronous compression transformation.

[0045] The Local Maximum Synchrosqueezing Transform (LMSST) is a time-frequency analysis method that rearranges time-frequency coefficients to local maximum positions, thereby increasing the energy concentration of the time-frequency representation and more accurately capturing the instantaneous frequency and time-varying characteristics of the signal.

[0046] In this embodiment, the total power data is first analyzed using time and frequency to find the local maximum point at each time point as an estimate of the instantaneous frequency. Then, the energy in the time and frequency representation is compressed to the vicinity of these local maximum points, thereby generating a clear time and frequency image.

[0047] In some embodiments, spatial texture features of the time-frequency image are extracted to obtain first feature information, including: The time-frequency image is input into a convolutional neural network with a three-level depth separable convolutional structure. The spatial texture features of the time-frequency image are extracted by the convolutional neural network, and the first feature information output by the convolutional neural network is obtained.

[0048] Among them, a three-level deep separable convolutional network (CNN) is a convolutional neural network that contains at least three levels of deep separable convolutional layers.

[0049] In this embodiment, the time-frequency image is input into a convolutional neural network with a three-level depth separable convolutional structure, and the spatial texture features contained in the time-frequency image are extracted layer by layer, and finally the first feature information is output.

[0050] In some embodiments, the ends of the depthwise separable convolutional structures in a convolutional neural network include max pooling layers.

[0051] In this embodiment, in the convolutional neural network, the end of each depthwise separable convolutional structure includes a max pooling layer. The max pooling layer downsamples the input feature map and extracts the maximum value in each local region, thereby reducing the spatial dimension of the feature map, reducing the amount of computation, and also helping to improve the model's generalization ability and invariance to small positional changes.

[0052] In some embodiments, the time-series features of the total power data are extracted to obtain second feature information, including: The total power data is input into a temporal convolutional network, which extracts the temporal features of the total power data to obtain the second feature information output by the temporal convolutional network.

[0053] Among them, Temporal Convolutional Network (TCN) is a deep learning model used to process sequential data. Based on the idea of ​​convolutional neural networks, it effectively captures long-term dependencies in sequential data by using techniques such as causal convolution and dilated convolution.

[0054] In this embodiment, the total power data is input into a temporal convolutional network. The temporal convolutional network captures the temporal dependencies in the data through a series of causal convolutional layers and dilated convolutional layers, thereby extracting temporal features and integrating them into second feature information, which is used as the output of the temporal convolutional network.

[0055] In some embodiments, the temporal convolutional network is pruned.

[0056] In this embodiment, the temporal convolutional network undergoes pruning, resulting in a lighter structure that improves computational efficiency and reduces resource consumption.

[0057] In some embodiments, the load identification result of the target energy meter is obtained based on the first feature information and the second feature information, including: The first and second feature information are assigned corresponding weights through an attention mechanism; Based on the weights corresponding to the first and second feature information, the first and second feature information are weighted and summed to obtain the third feature information; The load identification result is obtained based on the third feature information.

[0058] In this embodiment, an attention mechanism is used to adaptively assign importance weights to the first and second feature information to reflect the degree of contribution of different features to the recognition task. The first and second feature information are weighted and summed according to the assigned weights to generate the third feature information. Finally, classification or regression is performed based on the third feature information to obtain the load recognition result.

[0059] In some embodiments, the load identification method is executed through a target load identification model. The target load identification model receives total power data, extracts the time-frequency features of the total power data to obtain first feature information, extracts the time-series features of the total power data to obtain second feature information, and outputs the load identification result based on the first and second feature information. The target load identification model is trained through the following steps: Acquire sample power data of each electrical device connected to the target energy meter; Based on the power data of each sample, the total power data of the target energy meter is obtained. The target load identification model is trained using total power data from the samples.

[0060] Among them, the target load identification model is a model that can execute the load identification method, the sample power data is the power data of individual electrical devices obtained in the experimental phase, and the sample total power data is the total power data of each electrical device connected to the target energy meter in the experimental phase.

[0061] In this embodiment, a sample power meter is set up during the experimental phase, and one electrical device is connected to the sample power meter each time. The power data of a single electrical device during startup, shutdown, and operation is obtained to obtain sample power data.

[0062] The power data of each sample are summed to obtain the total power data of the sample.

[0063] By inputting the total power data of the samples into the target load identification model, the model parameters are iteratively optimized by backpropagating the error between the output of the target load identification model and the corresponding real load label, thereby completing the training.

[0064] The following is a specific embodiment of a load identification method.

[0065] like Figure 2 As shown, the load identification method provided in this application uses the total active power time series data (i.e., total power data) collected by smart meters and the power consumption of each electrical device as the original dataset, and adopts a CNN-TCN dual-branch heterogeneous parallel architecture to achieve multimodal feature fusion.

[0066] In terms of the time-frequency feature branch, the original total active power sequence is used to generate a logarithmic-scale time-frequency image through the local maximum synchronization compression transform algorithm, which is then input into the CNN model to extract spatial texture features.

[0067] In terms of time-series dynamic branching, the time-series data, which has been preprocessed and normalized, i.e., the total power data, is input into the TCN module. The long-term dependencies of load changes are effectively captured through dilated causal convolution and residual connections.

[0068] Finally, an attention-weighted feature fusion layer is used to dynamically balance the contributions of time-frequency and time-series features, outputting high-precision load state classification results. The network model is trained and validated using training and test datasets to obtain a trained network model for power load identification.

[0069] First, data preprocessing is performed to obtain the total active power of household electricity consumption (i.e., total power data) and the power consumption of each appliance. Preprocessing steps include data cleaning, data segmentation, dataset partitioning, and time-series dynamic branch data normalization.

[0070] Data cleaning involves removing outliers from the total active power sequence and filling in missing values ​​using linear interpolation.

[0071] The data is segmented into a total active power sequence of L cleaned sampling points. A sliding window is used to segment it into active power subsequences of fixed length N. The window size is twice the step size to ensure a 50% overlap rate. At the same time, the midpoint of the power sequence of each appliance is taken as the label value.

[0072] The dataset is divided into training dataset, validation dataset, and test dataset by dividing all the total active power subsequences and labels after segmentation into a set ratio.

[0073] The time-series dynamic branching data is normalized to calculate the arithmetic mean μ of the total active power subsequence at the window level for each independent window level in the training, validation, and test datasets. w and variance σ w Window-level normalization is performed, and the calculation formula is as follows:

[0074] in, These are the power values ​​in the total active power subsequence after sliding window segmentation. The average value of the total active power subsequence. Let V be the variance of the total active power subsequence.

[0075] For CNN network branches based on LMSST time-frequency maps, time-frequency transformation is performed. That is, the Local Maximum Synchronous Compression Transform (LMSST) algorithm solves the time-frequency ambiguity problem through energy redistribution and generates high-resolution time-frequency images as input to the CNN model. This complements the local feature extraction capability of CNN and has significant advantages in accuracy, efficiency and robustness.

[0076] In this embodiment, LMSST is specifically implemented by using a Gaussian window as the window function for the short-time Fourier transform, extending the one-dimensional time series to a two-dimensional time-frequency plane, and extracting the time-frequency distribution map of the short-time Fourier transform. The window function length and frame shift are set, and the results of the short-time Fourier transform are post-processed. The obtained short-time Fourier transform time-frequency distribution is then subjected to LMSST transform. By detecting local maxima points on the time-frequency plane, the energy is compressed to its corresponding instantaneous frequency, generating a high-resolution time-frequency matrix of size 128×128. The obtained LMSST time-frequency matrix is ​​then grayscale mapped to generate a grayscale image.

[0077] The formula for calculating the short-time Fourier transform is as follows:

[0078] Where x[n] is the nth sampling point of the discrete-time signal, ω is a Gaussian window function, centered at time t, where ω is the angular frequency.

[0079] The formula for calculating local maxima detection is as follows:

[0080] in, The sum of squares of time-frequency energy. The detected local maxima correspond to the instantaneous frequency and amplitude of the signal.

[0081] The formula for calculating energy compression is as follows:

[0082] in, δ is the instantaneous frequency correction value based on the phase derivative estimation, where δ is the Dirac function, t is time, and ω is the angular frequency.

[0083] When the time-frequency matrix output by LMSST is used as input to the CNN, it is processed through a complete grayscale image generation process before being input. The grayscale image generation process includes logarithmic transformation, normalization, and grayscale mapping of the LMSST time-frequency matrix.

[0084] Among them, the logarithmic transformation uses a logarithmic function to process the time-frequency energy values ​​in the LMSST time-frequency matrix, compressing the dynamic range that may originally span multiple orders of magnitude into a more suitable interval.

[0085] The normalization process employs a linear scaling method to map all values ​​to the standard range [0,1], ensuring that images generated from different signals have a consistent brightness reference.

[0086] Gray-scale mapping converts normalized numerical values ​​into specific gray-scale pixel values. A value of 0 corresponds to pure black, a value of 1 corresponds to pure white, and intermediate values ​​are linearly interpolated proportionally to generate a 256-level gray-scale gradient.

[0087] Table 1

[0088] like Figure 4 As shown, for the CNN network branch, the CNN branch takes the time-frequency image of LMSST transformation as input. The architecture adopts a lightweight design of three-level depthwise separable convolution. Each level includes three stages: depthwise convolution, pointwise convolution and max pooling. As shown in Table 1, the key parameter settings of the CNN model are as follows.

[0089] Depthwise convolution performs spatial convolution on the input image to extract local time-frequency features and reduce the number of parameters. Pointwise convolution performs 1×1 cross-channel feature fusion to control the output dimension. Max pooling uses a max pooling layer at the end of each stage to compress the feature map size and enhance spatial invariance.

[0090] Finally, global average pooling is used to aggregate features and output a one-dimensional feature vector.

[0091] like Figure 5As shown, for the TCN network branch, the core architecture of the TCN network is based on the collaborative design of causal convolution and dilated convolution. The network uses a stacked structure of multi-layer convolutional modules to gradually expand the receptive field range by using a dilation mechanism while strictly maintaining the temporal dependency relationship. This achieves efficient modeling of long-term temporal dependencies while maintaining low parameter complexity.

[0092] In this embodiment, the TCN network branch input is a power subsequence data of length N that has undergone normalization preprocessing, such as... Figure 6 As shown, the main body of the residual block adopts an 8-layer dilated causal convolution stacked structure. To ensure training stability, each convolution layer includes weight normalization, a rectified linear unit (ReLU) activation function, and dropout regularization, which can effectively alleviate the gradient vanishing or exploding problem. At the same time, a residual connection mechanism is adopted, and each residual module consists of two levels of convolutional layers and nonlinear transformation units.

[0093] For the computation process of each residual module at time step t, the calculation formulas for dilated causal convolution and regularization are as follows:

[0094]

[0095] Where i is the number of layers (8 layers in this model), j is the number of residual blocks, and k is the kernel size. Let be the expansion rate of the i-th layer, which increases exponentially with the number of layers. These are the weighting coefficients. For the (th)th in the sequence The values ​​at ) time points. RELU refers to nonlinear transformation, and BN is weight normalization.

[0096] The formula for calculating the residual connection of each residual block is as follows:

[0097] in, This is the output of the last layer of the current residual block. This is an optional 1×1 convolution transformation, enabled only when the number of input / output channels is inconsistent; This is the input to the current residual block, i.e., the output of the previous residual block. .

[0098] Finally, to achieve feature fusion with the output of the CNN branches, global average pooling is used for temporal dimension compression, and 1×1 convolution is used for feature dimension alignment.

[0099] Feature fusion and classification are performed. Multimodal feature fusion converts the time-frequency feature vector T output by the CNN branch into a multimodal feature vector. CNN The temporal feature vector T output by the TCN branch TCN After concatenation, the data is input into the attention module to calculate the feature weights.

[0100] Among them, W a Let b be the learnable weight matrix, σ be the sigmoid activation function, and b be the weight matrix. a For bias terms, This is the output of the CNN layer. This is the output of the TCN layer.

[0101] The final formula for generating fused features is:

[0102] The fused features are mapped to the class space through a fully connected layer, and the load state probability distribution P is output through Softmax.

[0103] The loss function is designed using cross-entropy loss, with L2 regularization added to prevent overfitting. The total loss is calculated using the following formula:

[0104] in, Let C be the cross-entropy loss function, and C be the number of classes. One-hot encoding of the real label. To predict probabilities; and These are the weight matrices for the CNN branch and the TCN branch, respectively. is a hyperparameter, used as a coefficient.

[0105] The load identification method provided in this application can be executed by a target load identification model. The following describes a specific embodiment for training a target load identification model.

[0106] like Figure 3 As shown, a time-series dataset, i.e., the dataset corresponding to the sample power data, is obtained. The time-series dataset is preprocessed by data cleaning, data segmentation, etc., and then input into the input layer of the target load identification model. LMSST time-frequency transformation is performed to obtain time-frequency images in grayscale image form. The time-frequency images are input into a CNN network to extract time-frequency features. The time-series dataset is normalized and then input into a TCN network to extract time-series features. The time-frequency features and time-series features are fused and then input into a fully connected layer, and then output through the output layer. The output result of the output layer is calculated by the loss function, and the network parameters of each layer in the target load identification model are updated using the gradient optimizer (adam).

[0107] The load identification method provided in this application is based on a hybrid CNN and TCN dual-branch model of Local Maximum Synchronous Compression Transform (LMSST). This method not only helps to extract local features of power data and capture long-term dependencies of data, thus improving the accuracy of power load identification, but also enables the model to be lightweighted through quantization, pruning and other techniques, allowing the model to run efficiently on edge devices such as electricity meters with limited computing resources.

[0108] In the multimodal heterogeneous feature fusion architecture, the CNN branch adopts a three-level depthwise separable convolutional structure, which combines spatial feature extraction of depthwise convolution, channel fusion of pointwise convolution, and scale compression of max pooling. This significantly reduces computational complexity while ensuring the expressive power of time-frequency features. The max pooling layer at the end of each level is designed to achieve exponential expansion of the receptive field and enhance spatial invariance.

[0109] A lightweight time-frequency feature extraction design is proposed, and the time-frequency branch uses the LMSST algorithm to generate logarithmic scale time-frequency images.

[0110] By employing a CNN-TCN dual-branch parallel structure, the local time-frequency features and long-term dynamic dependencies of load characteristics are captured through time-frequency images and raw time-series data, respectively, thus overcoming the limitations of traditional single-modal modeling.

[0111] Improve TCN temporal processing by using global average pooling to unify the temporal dimensions of the two branches, avoiding information loss from manual feature splicing, and using 1×1 convolutional feature alignment to eliminate dimensional differences between modalities and improve fusion efficiency.

[0112] Suitable for edge deployment, the model achieves lightweight design by combining the parallel computing characteristics of CNN and TCN with techniques such as depthwise separable convolution and pruning, which significantly reduces the computing resource requirements and enables it to efficiently adapt to the limited computing power of edge devices such as edge energy meters, thus achieving low-cost deployment on the edge.

[0113] The load identification method provided in this application can effectively improve the identification accuracy of residential electricity load and enhance the stability of the model in complex electricity consumption scenarios. Furthermore, the lightweight model size and low inference latency allow it to be deployed on edge devices such as smart meters and terminals, providing an efficient and reliable technical solution for load monitoring and energy efficiency optimization in smart grids.

[0114] The load identification method provided in this application can be executed by a load identification device. This application uses the example of a load identification device executing the load identification method to illustrate the load identification device provided in this application.

[0115] This application also provides a load identification device.

[0116] The load identification device includes: The acquisition module is used to acquire total power data, which is collected by the target energy meter. The first processing module is used to extract the time-frequency features of the total power data to obtain the first feature information; The second processing module is used to extract the time-series features of the total power data to obtain the second feature information; The third processing module is used to obtain the load identification result of the target energy meter based on the first feature information and the second feature information. The load identification result includes the load information of the electrical equipment connected to the target energy meter.

[0117] According to the load identification device provided in the embodiments of this application, by acquiring the total power data collected by the target energy meter, the time-frequency features and time-series features of the total power data are extracted to obtain first feature information and second feature information respectively. The time-frequency features can capture the transient characteristics of the start-up and shutdown and frequency changes of the electrical equipment, while the time-series features can reflect the power change trend and periodic pattern of the electrical equipment. By comprehensively analyzing the local details of the total power data reflected by the first feature information and the global dynamics of the total power data on the time axis reflected by the second feature information, the time-frequency features and time-series features are complementary, which can improve the discrimination ability of the total power data features, thereby improving the accuracy of the load identification result of the target energy meter.

[0118] In some embodiments, the first processing module is used to convert total power data into a time-frequency image; Spatial texture features are extracted from the time-frequency image to obtain the first feature information.

[0119] In some embodiments, the first processing module is used to convert the total power data into a time-frequency image through local maximum synchronous compression transformation.

[0120] In some embodiments, the first processing module is used to input the time-frequency image into a convolutional neural network with a three-level depth separable convolutional structure, extract the spatial texture features of the time-frequency image through the convolutional neural network, and obtain the first feature information output by the convolutional neural network.

[0121] In some embodiments, the ends of the depthwise separable convolutional structures in a convolutional neural network include max pooling layers.

[0122] In some embodiments, the second processing module is used to input the total power data into a temporal convolutional network, extract the temporal features of the total power data through the temporal convolutional network, and obtain the second feature information output by the temporal convolutional network.

[0123] In some embodiments, the temporal convolutional network is pruned.

[0124] In some embodiments, the third processing module is used to assign corresponding weights to the first feature information and the second feature information through an attention mechanism; Based on the weights corresponding to the first and second feature information, the first and second feature information are weighted and summed to obtain the third feature information; The load identification result is obtained based on the third feature information.

[0125] In some embodiments, the load identification method is executed through a target load identification model. The target load identification model receives total power data, extracts the time-frequency features of the total power data to obtain first feature information, extracts the time-series features of the total power data to obtain second feature information, and outputs the load identification result based on the first and second feature information. The target load identification model is trained through the following steps: Acquire sample power data of each electrical device connected to the target energy meter; Based on the power data of each sample, the total power data of the target energy meter is obtained. The target load identification model is trained using total power data from the samples.

[0126] The load identification device in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the specific type of device.

[0127] The load identification device in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system used.

[0128] The load identification device provided in this application embodiment can achieve... Figures 1 to 3 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0129] In some embodiments, such as Figure 7 As shown, this application embodiment also provides an electronic device 700, including a processor 701, a memory 702, and a computer program stored in the memory 702 and executable on the processor 701. When the program is executed by the processor 701, it implements the various processes of the above-described load identification method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0130] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0131] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described load identification method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0132] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0133] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described load identification method.

[0134] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0135] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described load identification method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0136] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0137] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0139] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0141] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A load identification method, characterized in that, The load identification method is applied to electricity meters, and the method includes: Acquire total power data, which is collected by the target energy meter; Extract the time-frequency features of the total power data to obtain the first feature information; Extract the time-series features of the total power data to obtain the second feature information; Based on the first feature information and the second feature information, the load identification result of the target energy meter is obtained, and the load identification result includes the load information of the electrical equipment connected to the target energy meter.

2. The load identification method according to claim 1, characterized in that, The step of extracting the time-frequency features of the total power data to obtain the first feature information includes: The total power data is converted into a time-frequency image; The spatial texture features of the time-frequency image are extracted to obtain the first feature information.

3. The load identification method according to claim 2, characterized in that, The step of converting the total power data into a time-frequency image includes: The total power data is converted into a time-frequency image by local maximum synchronous compression transformation.

4. The load identification method according to claim 2, characterized in that, The step of extracting the spatial texture features of the time-frequency image to obtain the first feature information includes: The time-frequency image is input into a convolutional neural network with a three-level depth separable convolutional structure. The spatial texture features of the time-frequency image are extracted by the convolutional neural network to obtain the first feature information output by the convolutional neural network.

5. The load identification method according to claim 4, characterized in that, The end of each level of the depth-separable convolutional structure in the convolutional neural network includes a max-pooling layer.

6. The load identification method according to claim 1, characterized in that, The extraction of time-series features from the total power data to obtain second feature information includes: The total power data is input into a temporal convolutional network, and the temporal features of the total power data are extracted through the temporal convolutional network to obtain the second feature information output by the temporal convolutional network.

7. The load identification method according to claim 6, characterized in that, The temporal convolutional network has been pruned.

8. The load identification method according to any one of claims 1-7, characterized in that, The process of obtaining the load identification result of the target energy meter based on the first feature information and the second feature information includes: The first feature information and the second feature information are assigned corresponding weights through an attention mechanism; Based on the weights corresponding to the first feature information and the second feature information, the first feature information and the second feature information are weighted and summed to obtain the third feature information; The load identification result is obtained based on the third feature information.

9. The load identification method according to any one of claims 1-7, characterized in that, The load identification method is executed through a target load identification model. The target load identification model receives the total power data, extracts the time-frequency features of the total power data to obtain first feature information, extracts the time-series features of the total power data to obtain second feature information, and outputs the load identification result based on the first and second feature information. The target load identification model is trained through the following steps: Acquire sample power data of each electrical device connected to the target energy meter; Based on the power data of each sample, the total power data of the target energy meter is obtained. The target load identification model is trained using total power data from the samples.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the load identification method for the electricity meter as described in any one of claims 1-9.

Images