Resident load identification method and system for label noise

Abstract

The invention discloses a resident load identification method and system aiming at label noise. The method comprises the following steps: acquiring first real-time target data, wherein the first real-time target data comprises voltage and current waveform data of a plurality of loads and corresponding label information; performing first preprocessing on the first real-time target data to obtain second real-time target data; and presetting a first load identification model, and performing load identification in combination with the second real-time target data. According to the technical scheme, the accuracy and robustness of resident load identification can be remarkably improved under the condition that label noise exists.

Description

Technical Field

[0001] The present invention relates to the technical field of load identification, and particularly to a method and system for residential load identification against label noise. Background Art

[0002] To reduce overall energy consumption, the importance of energy conservation and efficient energy management technologies has become increasingly prominent. Load identification is the basis for realizing home energy management, demand response, and two-way interaction between power grid users, and is of great significance for building a smart grid. Therefore, the research on load identification technology has received increasing attention.

[0003] The load identification task refers to classifying load types using electrical data, which is a typical pattern classification task. Due to the diverse load characteristics, the vast majority of research follows the deep learning route and constructs a load identification model using a dataset. However, in machine learning, especially supervised learning, label noise is inevitable. Research statistics show that in real-scene datasets, the proportion of incorrect labels can range from 8% to 38.5%. This is because the dataset is prone to subjective judgment and negligence during the annotation process, resulting in incorrect sample annotation. And deep learning has a powerful ability to learn complex functions, which easily leads to overfitting of the deep learning model to noise labels, thus seriously affecting the model accuracy. Therefore, reducing the impact of label noise is crucial for improving the accuracy and robustness of the model, which is a key problem to be solved in the load identification task. Summary of the Invention

[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method and system for residential load identification against label noise, which can solve the problems mentioned in the background art.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for residential load identification against label noise, including:

[0009] Obtain first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information;

[0010] Perform a first preprocessing on the first real-time target data to obtain second real-time target data;

[0011] Preset a first load identification model and perform load identification in combination with the second real-time target data.

[0012] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the first load identification model includes:

[0013] The first load identification model is any model with the second historical target data or the second real-time target data as the input and the load identification result or relevant parameters that can directly or indirectly obtain the load identification result as the output.

[0014] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the first load identification model further includes configuring a first improved loss function;

[0015] The first improved loss function is a composite function;

[0016] The first improved loss function is used to reduce the overfitting of wrong labels during the training process of the first load identification model.

[0017] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the performing a first preprocessing on the first real-time target data to obtain second real-time target data includes:

[0018] The first preprocessing at least includes performing a first extraction operation;

[0019] The first extraction operation is used to extract the current cycle sequence and electrical knowledge feature quantities in the first real-time target data;

[0020] The second real-time target data at least includes the current cycle sequence and electrical knowledge feature quantities.

[0021] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the electrical knowledge feature quantities at least include the peak-to-peak value of current, the effective value of current, the current form factor, the current harmonic distortion rate, the relative current harmonic distortion rate, the active power, the reactive power, the effective value and phase of odd harmonic current within several power frequency cycles.

[0022] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the current cycle sequence at least includes current data corresponding to several complete voltage cycles, and the current cycle sequence starts from the current sampling point corresponding to the positive zero point of the voltage, and its data length is equal to the product of the sampling frequency and the power grid frequency.

[0023] As a preferred solution of the residential load identification method for label noise according to the present invention, wherein: the first load identification model further includes at least a convolutional layer, a pooling layer, a batch normalization layer, and a fully connected layer;

[0024] After the current cycle sequence is transformed and calculated by a one-dimensional convolutional layer, a pooling layer, and a batch normalization layer, it is input together with the normalized electrical knowledge feature quantity into the fully connected layer to obtain the final output load type.

[0025] In a second aspect, the present invention provides a residential load identification system for label noise, including:

[0026] A data acquisition module, configured to acquire first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information;

[0027] A preprocessing module, configured to perform first preprocessing on the first real-time target data to obtain second real-time target data;

[0028] An identification module, configured to preset a first load identification model and perform load identification in combination with the second real-time target data.

[0029] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0030] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a residential load identification method and system for label noise, acquires first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information; performs first preprocessing on the first real-time target data to obtain second real-time target data; presets a first load identification model and performs load identification in combination with the second real-time target data. Through the above technical solutions, the present invention can significantly improve the accuracy and robustness of residential load identification in the presence of label noise. Specifically, by configuring the first improved loss function, the overfitting phenomenon of the model to incorrect labels during training is effectively reduced, thereby improving the generalization ability of the model. In addition, through refined first preprocessing of the first real-time target data, key current cycle sequences and electrical knowledge feature quantities are extracted, making the data input into the first load identification model more pure and representative, and further improving the identification accuracy.

[0032] In practical applications, the methods and systems of the present invention can be widely applied to multiple fields such as smart grids, home energy management, and demand response, providing strong support for energy conservation, emission reduction, and efficient energy management. For example, in the field of smart grids, by accurately identifying residential loads, grid dispatching can be optimized to reduce energy waste; in home energy management, the system can intelligently adjust electricity consumption strategies based on the load identification results to improve energy usage efficiency; in demand response, the present invention can effectively identify high-energy-consuming loads, providing data support for implementing dynamic electricity prices and demand-side management. In addition, the methods and systems of the present invention also have the advantages of strong scalability and good adaptability, being able to flexibly respond to the demand changes in different application scenarios and further promoting the development of intelligent energy management technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0034] Figure 1 is a method flow chart of a method and system for identifying residential loads against label noise provided by an embodiment of the present invention;

[0035] Figure 2 is a schematic structural diagram of a first load identification model for a method and system for identifying residential loads against label noise provided by an embodiment of the present invention;

[0036] Figure 3 is an internal structure diagram of a computer device for a method and system for identifying residential loads against label noise provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] Embodiment 1

[0039] Refer to Figures 1 - 3 , which is the first embodiment of the present invention. This embodiment provides a method and system for identifying residential loads against label noise, including:

[0040] In the existing related technologies, there are some problems, such as the impact of label noise on the accuracy of the load identification model. Label noise refers to the sample annotation errors caused by subjective judgment, negligence and other factors during the data annotation process. In the load identification task, due to the diversity of load characteristics, the annotation process of the data set is complex and cumbersome, so the label noise problem is particularly prominent. Incorrect labels will cause the deep learning model to overfit the noise labels during the training process, thus seriously affecting the accuracy and robustness of the model.

[0041] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail how to implement the residential load identification method for label noise;

[0042] Figure 1 The method flow chart of a residential load identification method and system for label noise is shown, including:

[0043] S101, obtain first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information;

[0044] In an optional embodiment, the first real-time target data is the relevant data of the target residential load, and these data can be obtained by real-time collection through sensors installed on residential electrical equipment. The voltage and current waveform data reflects the voltage and current changes of the load during operation, while the corresponding label information identifies the type or category of the load. These data provide a basis for subsequent processing and analysis.

[0045] In an optional embodiment, the first real-time target data may include voltage and current waveform data of various types of loads such as household appliances, lighting equipment, and power tools. These data not only contain the basic operation information of the load, but also imply the characteristics and behavior patterns of the load, which are crucial for subsequent load identification.

[0046] In the embodiment of this application, the first real-time target data is obtained to make a residential load data set. The residential load data set formed by the first real-time target data includes the voltage and current waveform data of each load and its corresponding label information. The sampling frequency of the data set is 7500Hz, including 11 types of residential loads such as fans, vacuum cleaners, refrigerators, and water heaters, and the load label values correspond to 0 to 10 respectively.

[0047] It should be noted that obtaining the first real-time target data, where the first real-time target data includes the voltage and current waveform data of several loads and their corresponding tag information, can provide an accurate and rich information basis for subsequent load identification. Through the real-time collected voltage and current waveform data, the subtle changes in the load during operation can be captured, and these changes can often reflect the type and characteristics of the load. At the same time, the corresponding tag information provides clear guidance for load identification, enabling the model to learn the correct classification rules. In addition, making a dataset containing multiple load types helps improve the generalization ability of the model, enabling it to handle load identification tasks in different scenarios.

[0048] S102, perform a first preprocessing on the first real-time target data to obtain second real-time target data;

[0049] In an optional embodiment, the first preprocessing generally can include steps such as data cleaning, data normalization, and feature extraction. Data cleaning aims to remove outliers and noise to ensure the accuracy and reliability of the data; data normalization is to scale the data so that it is at the same order of magnitude, which helps improve the training efficiency and identification accuracy of the model; feature extraction is to extract key information useful for load identification from the original data, such as current cycle sequences and electrical knowledge feature quantities.

[0050] It should be noted that through these preprocessing steps, the original data can be transformed into a purer and more representative second real-time target data, providing strong support for subsequent load identification.

[0051] In the embodiment of the present application, performing a first preprocessing on the first real-time target data to obtain second real-time target data includes:

[0052] The first preprocessing at least includes performing a first extraction operation;

[0053] The first extraction operation is used to extract the current cycle sequence and electrical knowledge feature quantity in the first real-time target data;

[0054] The second real-time target data at least includes the current cycle sequence and electrical knowledge feature quantity.

[0055] In an optional embodiment, the first extraction operation can be carried out based on digital signal processing technology and electrical knowledge. Specifically, the extraction of the current cycle sequence can be achieved by filtering and resampling the original current signal to obtain the current data corresponding to several complete voltage cycles. These current data start from the current sampling point corresponding to the positive zero point of the voltage, and the data length is equal to the product of the sampling frequency and the power grid frequency, thus ensuring the accuracy and integrity of the current cycle sequence.

[0056] In an alternative embodiment, the extraction of electrical knowledge feature quantities can be based on electrical principles and related algorithms to calculate key parameters such as the peak-to-peak current, root mean square (RMS) current, form factor of current, total harmonic distortion (THD) of current, etc. from the original data. These parameters can reflect the characteristics and behavior patterns of the load and are of great value for load identification.

[0057] In the embodiment of the present application, the electrical knowledge feature quantities at least include the peak-to-peak current, root mean square (RMS) current, form factor of current, total harmonic distortion (THD) of current, relative THD of current, active power, reactive power, RMS value and phase of odd harmonic currents within several power frequency cycles.

[0058] In the embodiment of the present application, the current cycle sequence at least includes current data corresponding to several complete voltage cycles. The current cycle sequence starts from the current sampling point corresponding to the positive zero point of the voltage, and its data length is equal to the product of the sampling frequency and the power grid frequency.

[0059] In the embodiment of the present application, the current cycle sequence and electrical knowledge feature quantities of each load sample are extracted according to the voltage and current data of the data set. The current cycle sequence of the load sample refers to the current data corresponding to a complete voltage cycle when the load is working. The current cycle sequence starts from the current sampling point corresponding to the positive zero point of the voltage, and its data length is 150 sampling points. The electrical knowledge feature quantities of the load sample refer to the peak-to-peak current, root mean square (RMS) current, form factor of current, total harmonic distortion (THD) of current, relative THD of current, active power, reactive power, RMS value and phase of odd (1, 3, 5) harmonic currents within one power frequency cycle when the load is working. The harmonic amplitude and phase information of the current and voltage are obtained by fast Fourier transform. The specific calculation methods of other feature quantities are as follows:

[0060] Peak-to-peak current I pp The calculation method is as follows:

[0061] I pp = max{i} - min{i}

[0062] max{i} is the maximum value of the current within one power frequency cycle, and min{i} is the minimum value of the current within one power frequency cycle;

[0063] The calculation method of the root mean square (RMS) current is as follows:

[0064]

[0065] In the formula, i n represents the nth current sampling point within one power frequency cycle, and N is the number of sampling points within one power frequency cycle;

[0066] The calculation method of the active power P is as follows:

[0067]

[0068] wherein \(v\) n represents the \(n\)th voltage sampling point within a power frequency period;

[0069] The calculation method of the reactive power \(Q\) is as follows:

[0070]

[0071] The form factor \(I\) of the current waveform wave is calculated as follows:

[0072]

[0073] The current harmonic distortion rate \(I\) THD is calculated as follows:

[0074]

[0075] wherein \(I\) a represents the amplitude of the \(a\)th harmonic of the current cycle;

[0076] The relative current harmonic distortion rate \(I'\) THD is calculated as follows:

[0077]

[0078] wherein \(U\) a represents the amplitude of the \(a\)th harmonic of the voltage cycle.

[0079] It should be noted that performing the first preprocessing on the first real-time target data to obtain the second real-time target data can significantly improve the accuracy and efficiency of load identification. Through refined preprocessing steps, such as extracting the current cycle sequence and electrical knowledge feature quantities, the noise and redundant information in the original data can be effectively removed, and the features most critical for load identification are retained. These features not only reflect the basic operating conditions of the load but also contain the characteristics and behavior patterns of the load, providing a more accurate and comprehensive information basis for subsequent load identification. In addition, the preprocessing steps can also make the data format more unified and standardized, which helps to improve the training speed and identification accuracy of the model. Therefore, performing the first preprocessing on the first real-time target data is an essential and important step in the present invention and is of crucial significance for realizing the residential load identification for label noise.

[0080] S103, preset the first load identification model, and perform load identification in combination with the second real-time target data.

[0081] In the embodiment of the present application, the first load identification model includes:

[0082] The first load recognition model is any model that takes the second historical target data or the second real-time target data as input and outputs the load recognition result or the relevant parameters that can directly or indirectly obtain the load recognition result.

[0083] In an optional embodiment, the first load recognition model can be constructed using deep learning algorithms, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), or their improved versions, such as long short-term memory networks (LSTMs), gated recurrent units (GRUs), etc. These models can automatically learn and extract the feature representations of the load from complex voltage and current waveform data, thereby achieving accurate load recognition. By combining the second real-time target data, that is, the preprocessed current cycle sequence and electrical knowledge feature quantities, the first load recognition model can make full use of these key information to efficiently and accurately identify the residential load.

[0084] In another optional embodiment, the design of the first load recognition model also considers the robustness and generalization ability of the model, enabling it to handle load recognition tasks in different scenarios and further enhancing the practicality and application value of the present invention.

[0085] In an optional embodiment, the first load recognition model can also be trained in combination with a transfer learning strategy to improve the model's recognition ability for unknown loads. Transfer learning is a method of transferring the knowledge learned from one task to another related task.

[0086] In an optional embodiment, an existing load recognition model and dataset can be used as a pre-trained model, and then fine-tuned in combination with the new second real-time target data, thereby accelerating the model training process and improving the recognition performance.

[0087] It should be noted that in this way, the first load recognition model can better adapt to new load types and scenarios and achieve a wider range of applications.

[0088] In the embodiment of the present application, a dual-branch neural network structure is used as the first load recognition model. The dual-branch neural network includes two parallel branches that respectively process the current cycle sequence and electrical knowledge feature quantities. Each branch consists of multiple convolutional layers and pooling layers for extracting the features of the input data. The outputs of the two branches are fused in the subsequent fully connected layer, and the load recognition result is output through the softmax function. This dual-branch neural network structure can make full use of the complementary information in the current cycle sequence and electrical knowledge feature quantities to improve the accuracy and robustness of load recognition.

[0089] In an optional embodiment, the relevant parameters that can directly or indirectly obtain the load identification result may include key information such as the type, power, and working status of the load. These information are of great significance for the monitoring, management, and optimization of the power system. For example, by accurately identifying the type and power of residential loads, refined management of power loads can be achieved, improving the utilization efficiency of power resources. At the same time, the working status information of the load can also provide strong support for the fault diagnosis and prediction of the power system.

[0090] In the embodiment of the present application, the first load identification model can output these key information by combining the second real-time target data for load identification, providing strong guarantee for the intelligent management of the power system.

[0091] In the embodiment of the present application, the first load identification model further includes configuring a first improved loss function;

[0092] The first improved loss function is a composite function;

[0093] The first improved loss function is used to reduce the overfitting error labels during the training process of the first load identification model.

[0094] In the embodiment of the present application, the first load identification model further includes at least a convolutional layer, a pooling layer, a batch normalization layer, and a fully connected layer;

[0095] After the current cycle sequence is transformed and calculated by the one-dimensional convolutional layer, the pooling layer, and the batch normalization layer, it is input into the fully connected layer together with the normalized electrical knowledge feature quantity to obtain the final output load type.

[0096] In the embodiment of the present application, a dual-branch neural network with the current cycle sequence and the electrical knowledge feature quantity as inputs is established as the load identification model. The network structure diagram of the identification model in this embodiment is as Figure 2 shown. Among them, the parameters in the brackets of the one-dimensional convolutional layer respectively represent the number of input channels, the number of output channels, the size of the convolutional kernel, and its stride; the parameters in the brackets of the max pooling layer respectively represent the size of the window and the stride; the parameters in the brackets of the fully connected layer respectively represent the input and output dimensions; the Flatten layer represents converting the multi-dimensional input quantity into a one-dimensional vector; the value in the brackets of the Dropout layer represents the probability that each neuron connection is discarded.

[0097] The preprocessed load data is used as the training set and input into the identification model, and is trained a preset number of times based on the robust loss function. The model training process is carried out in multiple iterations. Each time, multiple batches of data are randomly obtained from the training set. The number of samples in each batch is 64, and the number of training iterations is set to 500 times. The robust loss function is composed of a generalized cross-entropy function and an anti-cross-entropy function, and the calculation method is as follows:

[0098] L = LGCE +λ(t)×L RCE

[0099]

[0100] where λ(t) is a weight function with respect to the number of training times t, and μ = 0.005 in this embodiment; L GCE is the generalized cross-entropy function, L RCE is the inverse cross-entropy function, and its calculation method is as follows:

[0101]

[0102] In the formula, x i is the i-th load sample data, including the current cycle sequence and electrical knowledge feature quantities; represents the load category label of sample i in the dataset; K is the number of load categories in the dataset, that is represents the output of the recognition model, indicating that the sample x i belongs to the label the predicted probability of, is the one-hot encoded vector of the label , when the label value is , there is and c ik tends to w is a hyperparameter, and w = 0.5 is set in this embodiment.

[0103] Put the load recognition model trained in the above steps into practical application to identify unknown load samples.

[0104] In an alternative embodiment, the specific determination of identifying unknown load samples can be made by calculating the similarity between the current cycle sequence of the unknown load sample and the current cycle sequence template in the trained load recognition model, and the matching degree between the electrical knowledge feature quantities of the unknown load sample and the electrical knowledge feature quantity template in the trained load recognition model, and making a comprehensive judgment.

[0105] Specifically, thresholds for the similarity and the matching degree can be set. When both the similarity of the current cycle sequence and the matching degree of the electrical knowledge feature quantities of the unknown load sample exceed the set thresholds, it is considered that the unknown load sample matches successfully with a certain load type in the trained load recognition model, thereby determining the load type of the unknown load sample.

[0106] Exemplarily, set the threshold as the first threshold,

[0107] When both the similarity of the current cycle sequence and the matching degree of the electrical knowledge feature quantities of the unknown load sample exceed the first threshold, then classify the unknown load sample into the corresponding load type with successful matching.

[0108] If the similarity of the current cycle sequence and the matching degree of the electrical knowledge feature quantity of the unknown load sample do not exceed the first threshold, or only one of them exceeds the first threshold, then the unknown load sample is regarded as an unidentifiable load type, or other identification strategies are further adopted for identification.

[0109] In this way, the efficient and accurate identification of unknown load samples can be realized, providing strong support for the intelligent management and optimization of the power system.

[0110] In summary, the present invention proposes a method for identifying residential loads against label noise, obtaining first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information; performing first preprocessing on the first real-time target data to obtain second real-time target data; presetting a first load identification model, and combining the second real-time target data to perform load identification. Through the above technical solutions, the present invention can significantly improve the accuracy and robustness of residential load identification in the presence of label noise. Specifically, by configuring the first improved loss function, the overfitting phenomenon of the model to incorrect labels during training is effectively reduced, thereby improving the generalization ability of the model. In addition, through refined first preprocessing of the first real-time target data, key current cycle sequences and electrical knowledge feature quantities are extracted, making the data input into the first load identification model more pure and representative, and further improving the identification accuracy.

[0111] In practical applications, the method and system of the present invention can be widely applied to multiple fields such as smart grids, home energy management, and demand response, providing strong support for energy conservation and emission reduction and efficient energy management. For example, in the field of smart grids, by accurately identifying residential loads, grid dispatching can be optimized and energy waste can be reduced; in home energy management, the system can intelligently adjust the electricity consumption strategy according to the load identification result to improve energy use efficiency; in demand response, the present invention can effectively identify high-energy-consuming loads and provide data support for implementing dynamic electricity prices and demand-side management. In addition, the method and system of the present invention also have the advantages of strong scalability and good adaptability, and can flexibly respond to the demand changes in different application scenarios, further promoting the development of intelligent energy management technology.

[0112] Embodiment 2

[0113] In this embodiment, a system for identifying residential loads against label noise is further provided, including:

[0114] A data acquisition module, configured to acquire first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information;

[0115] A preprocessing module for performing a first preprocessing on the first real-time target data to obtain second real-time target data;

[0116] An identification module for presetting a first load identification model and performing load identification in combination with the second real-time target data.

[0117] Each of the above unit modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0118] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 3 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it realizes a method for identifying residential load against label noise. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0119] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by the processor, the following steps are realized:

[0120] Obtain first real-time target data, where the first real-time target data includes voltage and current waveform data of several loads and their corresponding label information;

[0121] Perform a first preprocessing on the first real-time target data to obtain second real-time target data;

[0122] Preset a first load identification model and perform load identification in combination with the second real-time target data.

[0123] Embodiment 3

[0124] Refer to Figure 2, which is an embodiment of the present invention, provides a method and system for identifying residential loads against label noise. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0125] In this embodiment, a traditional technical solution is used for comparative testing with the method of the present invention to verify the effectiveness of this method. The structure of the dual-branch neural network adopted in this application is as Figure 2 shown. Among them, input layer: The input is "load voltage and current data", including two parts: current cycle sequence and electrical knowledge feature quantity.

[0126] Current cycle sequence processing path: One-dimensional convolutional layer (1, 32, 16, 8): Perform one-dimensional convolutional operation on the current cycle sequence, and the output dimension is (1, 32, 16, 8). Batch normalization layer: Perform batch normalization processing on the output of the previous layer to accelerate the training process and improve the model performance. Max pooling layer (2, 2): Perform max pooling operation to reduce the size of the feature map, and the output dimension is (1, 32, 8, 4). One-dimensional convolutional layer (32, 64, 5, 1): Perform one-dimensional convolutional operation again, and the output dimension is (32, 64, 5, 1). Batch normalization layer: Perform batch normalization processing on the output of the previous layer. Max pooling layer (2, 2): Perform max pooling operation to reduce the size of the feature map, and the output dimension is (32, 64, 2, 0.5). One-dimensional convolutional layer (64, 64, 5, 1): Perform one-dimensional convolutional operation again, and the output dimension is (64, 64, 5, 1). Batch normalization layer: Perform batch normalization processing on the output of the previous layer. Max pooling layer (2, 2): Perform max pooling operation to reduce the size of the feature map, and the output dimension is (64, 64, 2, 0.5). Flatten layer: Flatten the multi-dimensional feature map into a one-dimensional vector, and the output dimension is (77, 64).

[0127] Electrical knowledge feature quantity processing path: Normalization: Perform normalization processing on the electrical knowledge feature quantity. Fusion layer: Fuse the output of the current cycle sequence processing path and the output of the electrical knowledge feature quantity processing path. Fully connected layer: Fully connected layer (77, 64): Perform a fully connected operation on the fused features, and the output dimension is (77, 64). Dropout layer (0.25): Perform Dropout operation to prevent overfitting, and the retention probability is 0.75. Fully connected layer (64, K): Perform a fully connected operation again, and the output dimension is (64, K), where K is the number of categories.

[0128] Output layer: Output the recognition result, that is, the final prediction result of the model.

[0129] Comparison method 1: Set a convolutional neural network with only the current cycle sequence as the input as the load recognition model. The traditional cross-entropy function is used as the loss function during the model training process, and the rest is the same as the method of the present invention;

[0130] Comparison method 2: The double-branch neural network of the present invention is used as the load recognition model. The traditional cross-entropy function is used as the loss function during the model training process, and the rest is the same as the method of the present invention.

[0131] An open dataset is used as the experimental data. The types of experimental data include 11 types, namely, fan, vacuum cleaner, refrigerator, water heater, fluorescent lamp, hair dryer, light bulb, microwave oven, air conditioner, washing machine, and laptop computer, with a total of 2510 samples. The data sampling frequency is 7500 Hz for all of the above. All the above samples are randomly divided into a training set and a test set according to a ratio of 4:1, and 40% of the samples are randomly selected from the training set, and the original labels of the samples are randomly replaced with other categories, that is, label noise is added to the training set.

[0132] Both the above comparison method and the inventive method use the training set to train the recognition model. The trained recognition models are all applied to the test set. By comparing the recognition accuracies of these three methods on the test set, the effectiveness of the method of the present invention is verified.

[0133] The experimental results are shown in the following table:

[0134] Accuracy rate (%) Comparison method 1 73.9％ Comparison method 2 85.1％ Method of the present invention 89.6％

[0135] As shown in the above table, the solution of the present invention uses a residential load recognition model and training method for label noise, and constructs a high-precision recognition model in a dataset containing a large number of incorrect labels. From the comparison between Method 1 and Method 2, it can be seen that the double-branch neural network recognition model of the present invention improves the anti-interference ability to label noise by combining current waveform features and electrical knowledge. From the comparison between the method of the present invention and Method 2, it can be seen that the loss function proposed in this patent improves the training effect of the model in label noise and improves the recognition accuracy of the recognition model.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

[0137] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0139] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction means that implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0141] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0142] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to cover these changes and modifications.

Claims

1. A method for identifying resident loads against label noise, characterized in that: include: Acquire first real-time target data, wherein the first real-time target data includes voltage and current waveform data of a plurality of loads and corresponding label information; Performing a first preprocessing on the first real-time target data to obtain second real-time target data; A first load identification model is preset, and load identification is performed in combination with the second real-time target data.

2. The method for identifying resident loads against label noise according to claim 1, characterized in that: The first load identification model includes: The first load identification model is any model whose input is the second historical target data or the second real-time target data, and whose output is the load identification result or the relevant parameters that can directly or indirectly obtain the load identification result.

3. The method for identifying resident loads against label noise according to claim 2, characterized in that: The first load identification model further includes configuring a first improved loss function; The first improved loss function is a composite function; The first improved loss function is used to reduce overfitting erroneous labels during the training process of the first load recognition model.

4. The method for identifying resident loads against label noise according to claim 3, characterized in that: The performing a first preprocessing on the first real-time target data to obtain the second real-time target data comprises: The first pretreatment at least includes performing a first extraction operation; The first extraction operation is used to extract the current cycle sequence and electrical knowledge feature quantity in the first real-time target data; The second real-time target data at least includes the current cycle sequence and electrical knowledge feature quantities.

5. The method for identifying resident loads against label noise according to claim 4, characterized in that: The electrical knowledge characteristic quantities include at least the current peak-to-peak value, current effective value, current waveform factor, current harmonic distortion rate, current relative harmonic distortion rate, active power, reactive power, odd harmonic current effective value and phase within several power frequency cycles.

6. The method for identifying resident loads against label noise according to claim 5, characterized in that: The current cycle sequence includes at least current data corresponding to several complete voltage cycles. The current cycle sequence starts with a current sampling point corresponding to the positive zero point of the voltage, and its data length is equal to the product of the sampling frequency and the grid frequency.

7. The method for identifying resident loads against label noise according to claim 6, characterized in that: The first load identification model further includes at least a convolutional layer, a pooling layer, a batch normalization layer, and a fully connected layer; After the current cycle sequence is transformed and calculated by the one-dimensional convolution layer, the pooling layer, and the batch normalization layer, it is input into the fully connected layer together with the normalized electrical knowledge feature quantity to obtain the final output load type.

8. A resident load identification system for label noise, characterized in that: include: A data acquisition module, used to acquire first real-time target data, wherein the first real-time target data includes voltage and current waveform data of a plurality of loads and corresponding label information; A preprocessing module, used for performing a first preprocessing on the first real-time target data to obtain second real-time target data; The identification module is used to preset a first load identification model and perform load identification in combination with the second real-time target data.

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

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

Images