Anti-interference load identification method and system

Abstract

The invention discloses an anti-interference load identification method and system. The method comprises the following steps: acquiring target resident load data, wherein the target resident load data comprises first data and second data; performing first preprocessing on the target resident load data to obtain third data of a plurality of load samples; and presetting a first load identification model, and performing load identification in combination with the third data. Through the preset load identification model, the influence of label noise on the model performance can be effectively identified and reduced, and the accuracy and robustness of load identification are improved. An improved loss function is adopted, and the evaluation of the label noise level is combined, so that the model training process is further optimized, and the model can still keep good recognition performance when facing noise data.

Description

Technical Field

[0001] The present invention relates to the technical field of load identification, and in particular to an anti-interference load identification method and system. Background Art

[0002] In order to reduce the comprehensive energy consumption of social production and life, the importance of energy consumption metering and energy conservation and emission reduction has become increasingly prominent. Load identification and monitoring, as the basis for realizing home energy management, demand response and two-way interaction between power grid users, is a key mitigation method for achieving energy conservation and safety of residential electricity consumption, and is of great significance to the construction of smart grids. Therefore, the research on load identification technology has received increasing attention.

[0003] The load identification task is to use electrical data to classify load types, which is a typical pattern classification task. Given the diversity of load characteristics, most studies use deep learning methods to build load identification models using mobile phone data sets. However, in machine learning, especially in the supervised learning stage, it is difficult to avoid label noise. Studies have shown that in real-world datasets, the proportion of incorrect labels can be as high as 8% to 38.5%. This is because the dataset is easily affected by subjective judgment and negligence during the labeling process, resulting in sample labeling errors. Due to the powerful ability of deep learning to learn complex functions, it is easy to cause the model to overfit noisy labels, which seriously affects the accuracy of the model. Therefore, reducing label noise is crucial to improving the accuracy and robustness of the model, and is a key problem that needs to be solved in the load identification task. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot 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 an anti-interference load identification method and system, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides an anti-interference load identification method, comprising:

[0009] Acquire target resident load data, where the target resident load data includes first data and second data;

[0010] Perform a first preprocessing on the target residential load data to obtain third data of a number of load samples;

[0011] Preset a first load recognition model and perform load recognition in combination with the third data.

[0012] As a preferred solution of the anti-interference load recognition method described in the present invention, wherein: the first load recognition model includes:

[0013] The first load recognition model is any model with the third data as the input and the load recognition result or relevant parameters that can directly or indirectly obtain the load recognition result as the output.

[0014] As a preferred solution of the anti-interference load recognition method described in the present invention, wherein: the first load recognition model further includes:

[0015] Evaluate the label noise level of the second data according to the first data and the first algorithm;

[0016] Configure a first improved loss function based on the label noise level.

[0017] As a preferred solution of the anti-interference load recognition method described in the present invention, wherein: the configuring of the first improved loss function based on the label noise level includes:

[0018] Preset a first noise threshold;

[0019] Perform a first judgment on the label noise level according to the first noise threshold;

[0020] Configure a first improved loss function according to the first judgment result.

[0021] As a preferred solution of the anti-interference load recognition method described in the present invention, wherein: the third data includes a current cycle sequence and electrical knowledge feature quantities;

[0022] After the current cycle sequence is transformed and calculated through a convolutional layer, a pooling layer, and a batch normalization layer, it is input into an attention layer together with the normalized electrical knowledge feature quantities to obtain a weighted feature vector;

[0023] Output the weighted feature vector through a fully connected layer as the load recognition result or relevant parameters that can directly or indirectly obtain the load recognition result.

[0024] As a preferred solution of the anti-interference load recognition method described in the present invention, wherein: the first algorithm includes:

[0025] Record the first data as labeled samples and the second data as unlabeled samples;

[0026] Obtain a label propagation matrix based on the first data and the second data;

[0027] Perform iterative convergence judgment of label propagation based on the label propagation matrix;

[0028] Evaluate the label noise level of the second data according to the iterative convergence judgment result.

[0029] As a preferred solution of the anti-interference load identification method described in the present invention, wherein: the first load identification model further includes at least a convolutional layer, a pooling layer, a batch normalization layer, a fully connected layer, and an attention layer.

[0030] In a second aspect, the present invention provides an anti-interference load identification system, including:

[0031] A data acquisition module, configured to acquire target residential load data, where the target residential load data includes first data and second data;

[0032] A data processing module, configured to perform first preprocessing on the target residential load data to obtain third data of a plurality of load samples;

[0033] An identification module, configured to preset a first load identification model and perform load identification in combination with the third data.

[0034] 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.

[0035] 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.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes an anti-interference load recognition method and system, which acquires target residential load data, where the target residential load data includes first data and second data; performs first preprocessing on the target residential load data to obtain third data of a number of load samples; presets a first load recognition model and performs load recognition in combination with the third data. Through the preset load recognition model, the influence of label noise on the model performance can be effectively identified and reduced, and the accuracy and robustness of load recognition can be improved. An improved loss function is adopted, combined with the evaluation of the label noise level, to further optimize the model training process and ensure that the model can still maintain good recognition performance when facing noisy data. Technologies such as convolutional layers, pooling layers, batch normalization layers, fully connected layers, and attention layers in deep learning are used to extract effective features from the load data and enhance the model's recognition ability for complex load patterns. Through the label propagation algorithm, labeled samples and unlabeled samples are combined to effectively utilize the information in the unlabeled samples and improve the generalization ability of the model in practical applications. The anti-interference load recognition system and method of the present invention are not only applicable to residential electricity consumption scenarios, but also can be widely applied to other electricity consumption scenarios such as industry and commerce, and have broad application prospects and practical values. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0038] Figure 1 It is a flowchart of a method for an anti-interference load recognition method and system provided by an embodiment of the present invention;

[0039] Figure 2 It is a structural diagram of a first load recognition model for an anti-interference load recognition method and system provided by an embodiment of the present invention;

[0040] Figure 3 It is an internal structural diagram of a computer device for an anti-interference load recognition method and system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Example 1

[0043] Reference Figures 1 - 3 , which is the first embodiment of the present invention. This embodiment provides an anti-interference load identification method and system, including:

[0044] In the existing related technologies, there are some problems. For example, in the residential electricity consumption scenario, due to the diversity and complexity of electrical equipment, the load data often contains a large amount of noise, which may come from the characteristics of the equipment itself, measurement errors, data transmission errors, etc. These noise data will have a negative impact on the accuracy of the load identification model, resulting in the model being unable to accurately identify the load type, and further affecting the effect of home energy management and two-way interaction between power grid users.

[0045] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail on how to implement this anti-interference load identification method;

[0046] Figure 1 The method flow chart of an anti-interference load identification method and system is shown, including:

[0047] S101, obtain target residential load data, where the target residential load data includes first data and second data;

[0048] In an optional embodiment, the target residential load data can be the power load data collected in real time from a smart meter or other data acquisition devices, and these data cover the electricity consumption of residential users at different time periods. Among them, the first data is the labeled load data with clear load type labels, which can be used to train the load identification model; the second data is the unlabeled load data without clear load type labels, but contains rich electricity consumption information, which can be used to expand the learning ability of the model and improve the generalization performance of the model.

[0049] In an optional embodiment, the first data may include the electricity consumption records of various household appliances, such as televisions, air conditioners, refrigerators, etc., and the load types have been determined during the labeling process. The second data may include some load data that is unknown or difficult to accurately label, such as the data generated by some new appliances or unconventional electricity consumption behaviors. By comprehensively processing and analyzing these data, the accuracy and reliability of load identification can be further improved.

[0050] In an optional embodiment, the target residential load data can be obtained in real time through data acquisition devices such as smart meters and sensors. Among them, the first data is the labeled load data with high accuracy and reliability, while the second data is the unlabeled load data, which may have label noise.

[0051] In an embodiment of the present application, a residential load dataset is made from the obtained target residential load data as training data. The dataset includes voltage and current waveform data of each load and its corresponding label information. The dataset is divided into two parts: a professional dataset (i.e., the set made from the first data) and a general dataset (the set made from the second data). Among them, the sample size of the professional dataset is 200, and the sample size of the general dataset with incorrect labels is 1,800. The sampling frequency of the load sample data is 7,500 Hz, including 11 types of residential loads such as fans, vacuum cleaners, refrigerators, and water heaters. The load label values correspond to 0-10 respectively.

[0052] It should be noted that obtaining the target residential load data, which includes the first data and the second data, can provide a rich data basis for subsequent preprocessing, label noise evaluation, and load identification. By using the first data, a preliminary load identification model can be trained. For the second data, although there is label noise, through the label propagation algorithm and the improved loss function proposed in the present invention, the label noise level can be effectively evaluated, and the model training process can be optimized accordingly, so as to realize the effective utilization of noise data and improve the accuracy and robustness of the model. In addition, this way of combining labeled and unlabeled data not only enhances the learning ability of the model but also improves the generalization ability of the model in practical applications.

[0053] S102, perform a first preprocessing on the target residential load data to obtain the third data of a number of load samples;

[0054] In an optional embodiment, the first preprocessing may include steps such as data cleaning, data format conversion, and data normalization. Data cleaning aims to remove outliers and missing values to ensure the integrity and accuracy of the data. Data format conversion is to convert the original data into a format suitable for model training, such as converting time series data into an input format acceptable to the model. Data normalization is to scale the data to the same range to improve the training efficiency and performance of the model.

[0055] It should be noted that through the first preprocessing, effective load samples can be extracted from the original target residential load data, providing reliable data support for subsequent load identification.

[0056] In an optional embodiment, the third data can be specifically restricted according to the steps of the first preprocessing. For example, the third data can be the load data after cleaning and normalization, and its format has been converted into an input format acceptable to the model, such as a two-dimensional array or a tensor. These data contain key features of the load samples, such as current, voltage, power, etc., and can be used to train a load identification model to achieve accurate classification of load types.

[0057] In the embodiments of the present application, the third data includes a current cycle sequence and electrical knowledge feature quantities;

[0058] After the current cycle sequence is transformed and calculated through a convolutional layer, a pooling layer, and a batch normalization layer, it is input into an attention layer together with the normalized electrical knowledge feature quantities to obtain a weighted feature vector;

[0059] The weighted feature vector is output through a fully connected layer as a load identification result or relevant parameters that can directly or indirectly obtain the load identification result.

[0060] In the embodiments of the present application, the third data of each load sample is extracted according to the voltage and current data of the data set, and the third data includes a current cycle sequence and electrical knowledge feature quantities. 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 value of the current, the effective value of the current, the current form factor, the current harmonic distortion rate, the relative current harmonic distortion rate, the active power, the reactive power, the effective values and phases of odd-order (1, 3, 5) harmonic currents within one power frequency cycle when the load is working.

[0061] In an alternative embodiment, the harmonic amplitude and phase information of the current and voltage are obtained through fast Fourier transform calculation, and the specific calculations of other feature quantities are as follows:

[0062] In an alternative embodiment, the peak-to-peak value I of the current pp The calculation method is as follows:

[0063] I pp = max{i} - min{i}

[0064] 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;

[0065] In an alternative embodiment, the calculation method of the effective value of the current is as follows:

[0066]

[0067] 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;

[0068] In an alternative embodiment, the calculation method of the active power P is as follows:

[0069]

[0070] In the formula, v nRepresents the nth voltage sampling point within a power frequency cycle;

[0071] In an alternative embodiment, the reactive power Q is calculated as follows:

[0072]

[0073]

[0074] In an alternative embodiment, the current form factor I wave is calculated as follows:

[0075]

[0076] In an alternative embodiment, the current total harmonic distortion rate I THD is calculated as follows:

[0077]

[0078] In the formula, I a represents the amplitude of the ath harmonic of the current cycle;

[0079] In an alternative embodiment, the relative current total harmonic distortion rate I′ THD is calculated as follows:

[0080]

[0081]

[0082] In the formula, U a represents the amplitude of the ath harmonic of the voltage cycle.

[0083] It should be noted that performing the first preprocessing on the target residential load data to obtain the third data of several load samples can provide high-quality data input for the subsequent training and testing of the load recognition model. By preprocessing the target residential load data, noise and redundant information in the data can be removed, features useful for load recognition can be extracted, thereby improving the recognition accuracy and efficiency of the model. In addition, the preprocessed data format is more unified and standardized, which is beneficial to the training and practical application of the model. Therefore, the first preprocessing step in the embodiments of the present application is an indispensable part of the anti-interference load recognition method and system, and is of great significance for achieving accurate load recognition and improving the robustness of the model.

[0084] S103, preset the first load recognition model, and perform load recognition in combination with the third data.

[0085] In the embodiments of the present application, the first load recognition model includes:

[0086] The first load identification model is any model that takes the third data as input and outputs the load identification result or relevant parameters that can directly or indirectly obtain the load identification result.

[0087] In an alternative embodiment, the first load identification model can be built using a deep learning framework such as TensorFlow or PyTorch. The model structure can include parts such as an input layer, a convolutional layer, a pooling layer, a batch normalization layer, an attention layer, and a fully connected layer. Among them, the input layer receives the preprocessed third data, that is, the current cycle sequence and electrical knowledge feature quantities of the load samples. The convolutional layer is used to extract local features in the data, the pooling layer is used to reduce the dimension of the data and retain important features, and the batch normalization layer is used to accelerate model training and improve the stability of the model. The attention layer is used to perform weighted processing on the features to highlight the influence of important features on the load identification result. Finally, the fully connected layer maps the weighted feature vector to the load type label and outputs the load identification result or relevant parameters.

[0088] In an alternative embodiment, the first load identification model can also be optimized using techniques such as transfer learning and ensemble learning to improve the model's identification performance and generalization ability. Transfer learning can utilize knowledge and experience in related fields to accelerate the training process of the model in new scenarios and improve the identification accuracy of the model. Ensemble learning can reduce the prediction error of a single model and improve the overall identification performance by combining the prediction results of multiple models. The introduction of these techniques enables the first load identification model to better adapt to complex and changing load identification tasks and provide more accurate and reliable load identification results for home energy management and two-way interaction between power grid users.

[0089] In an alternative embodiment, the first load identification model can also be trained using a custom loss function. The custom loss function can include a label noise evaluation module and a model optimization module. The label noise evaluation module is used to evaluate the label noise level of unlabeled data (i.e., the second data), and appropriate evaluation metrics and methods can be designed according to the characteristics of the data. The model optimization module adjusts the training strategy of the model according to the evaluation results, such as changing the learning rate, increasing the regularization term, etc., to reduce the impact of noisy data on model training. By introducing a custom loss function, the robustness and accuracy of the model in a noisy data environment can be further improved.

[0090] In an alternative embodiment, the relevant parameters from which the load identification result can be directly or indirectly obtained may include the probability distribution of the load type, the power factor of the load, the active period of the load, etc. Through the comprehensive analysis and processing of these parameters, a more detailed and accurate load identification result can be obtained, providing more comprehensive information support for home energy management and two-way interaction between power grid users. For example, the probability distribution of the load type can help users understand the electricity consumption of various appliances at home, so as to formulate a more reasonable electricity consumption plan; the power factor of the load can reflect the energy efficiency of the appliances, providing a reference for users to select high-efficiency and energy-saving appliances; the active period of the load can reveal the electricity consumption habits of users, providing data support for power grid dispatching and load forecasting.

[0091] In the embodiment of the present application, the first load identification model further includes:

[0092] Evaluating the label noise level of the second data according to the first data and the first algorithm;

[0093] Configuring a first improved loss function based on the label noise level.

[0094] In an alternative embodiment, the first algorithm may be a graph semi-supervised algorithm, a K-means clustering algorithm, or other algorithms suitable for label noise evaluation. Through the first algorithm, the label noise level of the unlabeled second data can be evaluated using the labeled first data, so as to determine which data may have label errors or uncertainties. Based on the evaluation result, the first improved loss function can be configured, which can be adjusted according to different label noise levels to reduce the impact of noise data on model training and improve the robustness and accuracy of the model. For example, for data with a high label noise level, its weight in model training can be reduced by increasing the regularization term or adjusting the learning rate, so as to avoid the model overfitting to the noise data.

[0095] In the embodiment of the present application, the first algorithm is a graph semi-supervised algorithm, which can make full use of a large amount of unlabeled data for learning while using a small amount of labeled data, improving the generalization ability of the model. In the graph semi-supervised algorithm, data points are represented as nodes in a graph, and the connections between nodes represent the similarity or relationship between data points. The nodes of the labeled data have clear labels, while the nodes of the unlabeled data infer their labels through their relationships with the labeled nodes. In this way, the graph semi-supervised algorithm can effectively utilize the information in the unlabeled data and improve the accuracy of load identification.

[0096] 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, a fully connected layer, and an attention layer.

[0097] In the embodiment of the present application, the first algorithm includes:

[0098] Record the first data as the labeled samples and the second data as the unlabeled samples;

[0099] Obtain a label propagation matrix according to the first data and the second data;

[0100] Perform iterative convergence judgment of label propagation according to the label propagation matrix;

[0101] Evaluate the label noise level of the second data according to the iterative convergence judgment result.

[0102] In the embodiment of the present application, configuring the first improved loss function based on the label noise level includes:

[0103] Preset a first noise threshold;

[0104] Perform a first judgment on the label noise level according to the first noise threshold;

[0105] Configure the first improved loss function according to the first judgment result.

[0106] In the embodiment of the present application, the label noise level is evaluated based on the graph semi-supervised algorithm. The load samples of the professional dataset are regarded as labeled samples, and the load samples of the general dataset are regarded as unlabeled samples. Let the feature vector of each sample be x, and the initial set of labeled samples be X 1 ={(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x M ,y M )}, and the set of unlabeled samples is denoted as X u ={x M+1 ,x M+2 ,...,x M+N}, where M is the number of samples in the professional dataset, N is the number of samples in the general dataset, and the number of load categories is K. Then the initial label matrix Y can be defined as:

[0107]

[0108] Step 1: Construct a graph G=(V,E) based on X 1 and X u , where the node set V={x 1 ,…,x N ,x M+1 ,…,x M+N}, and the edge set E is an affinity matrix calculated based on the Gaussian function. The calculation method is as follows:

[0109]

[0110] where \(i, j\in\{1, 2, \ldots, M + N\}\); \(0\lt\sigma\lt1\)

[0111] Step 2: Construct the label propagation matrix \(S\) based on \(W\)

[0112]

[0113] In the formula, the matrix \(D = diag(d 1 , d 2 , \ldots, d M+N )\)

[0114] Step 3: Iteratively calculate the following formula until convergence, and perform the label propagation process, that is, the process in which labeled samples transfer label information to unlabeled samples according to the similarity of samples

[0115] \(F^{(t + 1)}=\alpha SF^{(t)}+(1 - \alpha)Y\)

[0116] where \(F^{(0)} = Y\); \(\alpha\in(0, 1)\) is used to balance the importance of the label propagation term \(SF^{(t)}\) and the initialization term \(Y\). When \(F\) converges, \(F * =(D-\alpha W) -1 Y can be used to obtain the predicted label of the unlabeled sample \(x * i

[0117] Step 4: Evaluate the label noise level of the general dataset, as shown in the following formula

[0118]

[0119] where \(N'\) represents the number of inconsistencies between the predicted label \(h'\) of the graph semi-supervised algorithm of the general dataset and the original label

[0120] In an alternative embodiment, a dual-branch attention neural network with the current cycle sequence and electrical knowledge feature quantities as inputs is established as the load recognition model. The network structure diagram of the recognition model in this embodiment is as shown in Figure 2 shown. Among them, the parameters in the one-dimensional convolutional layer brackets represent the number of input channels, the number of output channels, the convolutional kernel size, and its stride respectively; the parameters in the max pooling layer brackets represent the window size and stride respectively; the parameters in the fully connected layer brackets represent the input and output dimensions respectively; the Flatten layer represents converting the multi-dimensional input quantity into a one-dimensional vector; the attention layer adopts a multi-head attention mechanism, and the number of attention heads is set to 7; the value in the Dropout layer brackets represents the probability that each neuron connection is discarded.

[0121] ​​​In an alternative embodiment, the first preprocessed load data is input into the recognition model as a training set and trained a preset number of times based on a robust loss function. First, the robust loss function is set according to the evaluated label noise level β, and the calculation method is as follows

[0122]

[0123] where x i is the i-th load sample data, including the current cycle sequence and electrical knowledge feature quantities; represents the load class label of the model predicting sample i; represents the output of the recognition model, indicating that sample x i belongs to the label prediction probability, is the one-hot encoded vector of the label When the label value is There is and The w value is determined according to the label noise level β:

[0124]

[0125] s is a correction coefficient and s = 3 is set.

[0126] Furthermore, the model is trained according to the above loss function, and the training process is carried out in multiple iterations. Each iteration randomly obtains multiple batches of data from the training set, and the number of samples in each batch is 64. The number of training iterations is set to 500 times.

[0127] Furthermore, the load recognition model trained in the above steps is put into practical application to identify unknown load samples.

[0128] In summary, the present invention proposes an anti-interference load recognition method, which obtains target residential load data, where the target residential load data includes first data and second data; performs first preprocessing on the target residential load data to obtain third data of a number of load samples; presets a first load recognition model and performs load recognition in combination with the third data. Through the preset load recognition model, the influence of label noise on the model performance can be effectively recognized and reduced, and the accuracy and robustness of load recognition can be improved. An improved loss function is adopted, combined with the evaluation of the label noise level, to further optimize the model training process and ensure that the model can still maintain good recognition performance when facing noisy data. Technologies such as convolutional layers, pooling layers, batch normalization layers, fully connected layers, and attention layers in deep learning are used to extract effective features in the load data and enhance the model's recognition ability for complex load patterns. Through the label propagation algorithm, the labeled samples and unlabeled samples are combined to effectively utilize the information in the unlabeled samples and improve the generalization ability of the model in practical applications. The anti-interference load recognition system and method of the present invention are not only applicable to residential electricity consumption scenarios, but also can be widely applied to other electricity consumption scenarios such as industry and commerce, and have broad application prospects and practical values.

[0129] Embodiment 2

[0130] This embodiment also provides an anti-interference load recognition system, including:

[0131] A data acquisition module, configured to acquire target residential load data, where the target residential load data includes first data and second data;

[0132] A data processing module, configured to perform first preprocessing on the target residential load data to obtain third data of a number of load samples;

[0133] A recognition module, configured to preset a first load recognition model and perform load recognition in combination with the third data.

[0134] The above-mentioned 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 that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0135] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 3As shown in the figure. 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 implements an anti-interference load identification method. 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 covered on 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.

[0136] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0137] Obtain target residential load data, where the target residential load data includes first data and second data;

[0138] Perform a first preprocessing on the target residential load data to obtain third data of several load samples;

[0139] Preset a first load identification model and perform load identification in combination with the third data.

[0140] Embodiment 3

[0141] Refer to Figure 2 , which is an embodiment of the present invention, provides an anti-interference load identification method and system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0142] 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.

[0143] Comparison method 1: Set a convolutional neural network with only the current cycle sequence as the input as the load identification model, and use the traditional cross-entropy function as the loss function during the model training process, and the rest is the same as the method of the present invention;

[0144] Comparison method 2: Use the dual-branch attention neural network of the present invention as the load identification model, and use the traditional cross-entropy function as the loss function during the model training process, and the rest is the same as the method of the present invention;

[0145] Comparison method 3: Use the dual-branch attention neural network of the present invention as the load recognition model. The fixed generalized cross-entropy function is used in the model training process (set w = 0.5, that is, sβ is always equal to 0.5), and the rest is the same as the method of the present invention.

[0146] Use the public dataset 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 2,510 samples. The data sampling frequency is 7,500 Hz.

[0147] All the above samples are first randomly divided into a training set and a test set according to a ratio of 4:1. Subsequently, the training set is randomly divided into a professional dataset and a general dataset according to a ratio of 1:9, and 20%, 40%, and 60% of different contents of label noise are added to the general dataset. The specific method is to randomly select samples and randomly replace their original labels with other categories.

[0148] The above comparison method and the invention method both 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.

[0149] The experimental accuracies of each method are shown in the following table:

[0150] Label noise content: 20％ 40％ 60％ Comparison method 1 84.9％ 73.5％ 70.2％ Comparison method 2 91.8％ 84.5％ 75.9％ Comparison method 3 92.8％ 89.3％ 82.3％ The method of the present invention 92.4％ 89.6％ 85.7％

[0151] As shown in the above table, the solution of the present invention uses a dual-branch attention network and an adaptive training strategy for load recognition and label noise, and constructs a high-precision recognition model in a dataset containing a large number of incorrect labels. Comparing the method of the present invention with the above three methods, it can be seen that the load recognition model based on the dual-branch attention network in the present invention can improve the robustness to label noise; in addition, the proposed adaptive loss function based on the evaluation of label noise level can effectively reduce the negative impact of label noise on model training compared with the traditional cross-entropy function and the fixed generalized cross-entropy function, and generally improve the accuracy of load recognition.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limitations. 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.

[0153] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a completely hardware embodiment, a completely 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.) that contain 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.

[0154] 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 means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0155] 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 instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0156] 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 performed on the computer or other programmable device to generate a computer-implemented process, so that 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.

[0157] 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 as including the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0158] 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 include these modifications and variations.

Claims

1. A method for identifying an anti-interference load, characterized in that: include: Acquire target resident load data, where the target resident load data includes first data and second data; Performing a first preprocessing on the target resident load data to obtain third data of a plurality of load samples; A first load identification model is preset, and load identification is performed in combination with the third data.

2. The anti-interference load identification method 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 third data and whose output is the load identification result or the relevant parameters of the load identification result that can be obtained directly or indirectly.

3. The anti-interference load identification method according to claim 2, characterized in that: The first load identification model also includes: According to the first data and the first algorithm, evaluating a label noise level of the second data; A first improved loss function based on the label noise level is configured.

4. The anti-interference load identification method according to claim 3, characterized in that: The configuring of a first improved loss function based on the label noise level comprises: Presetting a first noise threshold; Performing a first judgment on the tag noise level according to the first noise threshold; A first improved loss function is configured according to the first judgment result.

5. The anti-interference load identification method according to claim 4, characterized in that: The third data includes a current cycle sequence and electrical knowledge feature quantity; After the current cycle sequence is transformed and calculated by the convolution layer, the pooling layer, and the batch normalization layer, it is input into the attention layer together with the normalized electrical knowledge feature quantity to obtain a weighted feature vector; The weighted feature vector is outputted as a load identification result through a fully connected layer, or related parameters of the load identification result can be directly or indirectly obtained.

6. The anti-interference load identification method according to claim 5, characterized in that: The first algorithm comprises: The first data is recorded as a labeled sample, and the second data is recorded as an unlabeled sample; Acquire a label propagation matrix according to the first data and the second data; Perform iterative convergence judgment of label propagation according to the label propagation matrix; The label noise level of the second data is evaluated according to the iterative convergence judgment result.

7. The anti-interference load identification method according to claim 6, characterized in that: The first load identification model also includes at least a convolutional layer, a pooling layer, a batch normalization layer, a fully connected layer and an attention layer.

8. An anti-interference load identification system, characterized in that: include: A data acquisition module, used to acquire target resident load data, wherein the target resident load data includes first data and second data; A data processing module, used for performing a first preprocessing on the target resident load data to obtain third data of a plurality of load samples; The identification module is used to preset a first load identification model and perform load identification in combination with the third 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