Concentrator Data Compression and Remote Calibration Optimization Method and System

Abstract

The present invention discloses a method and system for optimizing data compression and remote calibration of a concentrator. The method includes: acquiring the original data of the concentrator and performing preprocessing operations to obtain an operation result; inputting the operation result into a data compression model for data compression; wherein, the data compression model first extracts shallow features and deep features from the operation result, fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing; inputting the operation result into a remote calibration model, and using the compression result as auxiliary information for remote calibration; the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference object system, and analyzing the relationship between different reference objects and calibration parameters; outputting corresponding content. Through the method of the present invention, the data transmission efficiency is improved, the transmission cost is reduced, and the automation and high precision of remote calibration are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of concentrator data processing, and more specifically to a method and system for optimizing data compression and remote calibration of a concentrator. Background Art

[0002] With the rapid development of Internet of Things technology, the concentrator plays a crucial role in the process of data collection and transmission. As a key hub connecting numerous terminal devices and the data center, the concentrator is responsible for collecting a vast amount of data generated by various sensors, monitoring devices, etc., and transmitting it to the data center for further analysis and processing.

[0003] However, with the large-scale deployment of Internet of Things devices, the amount of data that the concentrator needs to process has increased explosively. A large amount of data will occupy a large amount of bandwidth resources during transmission, leading to network congestion, increasing transmission costs and energy consumption. Traditional data transmission methods are difficult to efficiently process large-scale data, with slow data transmission speeds and prone to delays and packet loss. This not only affects the real-time nature of the data but also may result in incomplete data transmission, reducing the availability and timeliness of the data. Traditional remote calibration methods require manual intervention. Usually, professional technicians need to go to the site for equipment debugging and parameter calibration, or perform manual intervention through a complex remote operation interface. This not only increases labor costs and time costs but also may lead to calibration errors caused by human operation mistakes. Existing remote calibration methods are difficult to accurately identify and adjust device parameters, often only able to roughly calibrate the device, and cannot meet the requirements of high-precision applications. For example, in an intelligent meter reading system, when the concentrator remotely calibrates metering devices such as electricity meters and water meters, insufficient accuracy may lead to errors in meter reading data, affecting the accuracy of billing. Traditional compression algorithms are often difficult to achieve efficient compression for complex and high-dimensional data, with limited compression ratios and unable to significantly reduce the amount of data, making it difficult to meet the requirements of fast data transmission. Traditional compression methods are usually based on fixed algorithms and rules, difficult to adaptively adjust according to different characteristics and distributions of data, and often require different compression methods for different types of data, with poor versatility.

[0004] Therefore, it is necessary to design a new method to improve data transmission efficiency, reduce transmission costs and energy consumption, and at the same time achieve the automation and high precision of remote calibration, enhance the performance and intelligent level of the concentrator, and solve the problems faced by existing concentrator data transmission and remote calibration. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a method and system for optimizing data compression and remote calibration of a concentrator.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for optimizing data compression and remote calibration of a concentrator, including:

[0007] Obtain the original data of the concentrator and perform preprocessing operations to obtain an operation result;

[0008] Input the operation result into a data compression model for data compression to obtain a compression result; wherein, the data compression model first extracts shallow features and deep features from the operation result, and fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result;

[0009] Input the operation result into a remote calibration model, and use the compression result as auxiliary information for remote calibration to obtain calibration parameters; wherein, the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference system, and analyzing the relationship between different references and calibration parameters.

[0010] Output the compression result and the calibration parameters.

[0011] A further technical solution thereof is: The shallow features include short-term fluctuation patterns in sensor data and instantaneous change features in device status data; the deep features include long-term trends in sensor data and periodic change patterns in device status data.

[0012] A further technical solution thereof is: The inputting the operation result into a data compression model for data compression to obtain a compression result includes:

[0013] Use a convolutional layer to perform convolution operations and downsampling on the operation result, extract local features of the data and features corresponding to basic patterns, and enhance the feature expression ability and network performance through a batch normalization layer and a ReLU activation function to obtain shallow features;

[0014] Use a deep neural network structure to perform deep feature extraction on the operation result, obtain complex patterns and high-level features in the data, and perform secondary processing through residual links to obtain deep features;

[0015] Adopt a feature fusion network, use upsampling and downsampling techniques to adjust the spatial dimensions of the shallow features and the deep features, and then fuse them to form a comprehensive representation feature;

[0016] Use a compression network to compress the comprehensive representation feature to obtain a compression result.

[0017] Its further technical solution is: the data compression model includes a feature extraction network, a feature fusion network and a compression network, which is trained using the collected data and the BCDELoss loss function is used to optimize the model.

[0018] The further technical solution is: the training process of the remote calibration model includes:

[0019] The teacher network is trained using the labeled dataset, and the teacher network is used to assign pseudo labels to the unlabeled dataset to form a pseudo-labeled dataset, which is then used together with the labeled dataset to train the student network.

[0020] The further technical solution is: the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference system, and analyzing the relationship between different reference objects and calibration parameters, including:

[0021] According to the working environment and data characteristics of the concentrator, multiple reference objects with different characteristics are selected;

[0022] Use image processing technology and positioning technology to extract the shape, size, color and position feature information of the selected reference object to obtain the feature information of the reference object;

[0023] With the help of data mining and machine learning algorithms, the association between the characteristic information of different reference objects and the calibration parameters is mined to obtain labeled data;

[0024] The annotated data is used to train a teacher network to generate a pseudo-label extended data set, and then the pseudo-label extended data set is used to train a student network to construct a remote calibration model based on a multi-reference system.

[0025] A further technical solution is: the operation result is input into a remote calibration model, and the compression result is used as auxiliary information for remote calibration to obtain calibration parameters, including:

[0026] The operation results are input into the remote calibration model to analyze the performance of the concentrator under different working conditions and identify various characteristic information related to the calibration. At the same time, the compression results are input into the remote calibration model as auxiliary information, and the compression results are used to optimize feature matching, filter outliers, and assist teachers and students in network learning to obtain calibration parameters.

[0027] A further technical solution is: the compression result optimizes feature matching, including:

[0028] In a multi-reference object system, each reference object has characteristic information. By comparing the key features in the compression result with the characteristic information of the reference object, the working status of the positioning concentrator and the calibration parameters that need to be adjusted can be determined.

[0029] The present invention also provides a concentrator data compression and remote calibration optimization system, including:

[0030] An acquisition unit, configured to acquire the original concentrator data and perform preprocessing operations to obtain an operation result;

[0031] A compression unit, configured to input the operation result into a data compression model for data compression to obtain a compression result; wherein, the data compression model first extracts shallow features and deep features from the operation result, fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result;

[0032] A remote calibration unit, configured to input the operation result into a remote calibration model and use the compression result as auxiliary information for remote calibration to obtain calibration parameters; wherein, the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference object system, and analyzing the relationship between different reference objects and the calibration parameters;

[0033] An output unit, configured to output the compression result and the calibration parameters.

[0034] A further technical solution thereof is that: the compression unit includes:

[0035] A shallow feature extraction subunit, configured to use a convolutional layer to perform convolution operations and downsampling on the operation result, extract the local features of the data and the features corresponding to the basic patterns, and enhance the feature expression ability and network performance through a batch normalization layer and a ReLU activation function to obtain shallow features;

[0036] A deep feature extraction subunit, configured to use a deep neural network structure to perform deep feature extraction on the operation result, obtain complex patterns and high-level features in the data, and perform secondary processing through a residual link to obtain deep features;

[0037] A fusion subunit, configured to use a feature fusion network, perform spatial dimension adjustment on the shallow features and the deep features by using upsampling and downsampling techniques, and then perform fusion to form a comprehensive representation feature;

[0038] A compression subunit, configured to use a compression network to compress the comprehensive representation feature to obtain a compression result.

[0039] The beneficial effects of the present invention compared with the prior art are as follows: By acquiring the original data of the concentrator and performing preprocessing, the present invention then uses a specially designed data compression model to extract shallow features such as short-term fluctuation patterns and deep features such as long-term trends, and fuses these features to form a comprehensive representation feature. Further, an efficient compression result is obtained through a compression network, thereby significantly improving the data transmission efficiency, reducing the transmission cost and energy consumption. At the same time, the operation result and the compression result are used as auxiliary information and input into a remote calibration model. This model is constructed based on analyzing the relationships between different references and calibration parameters in a multi-reference system, and can achieve high-precision automatic calculation of calibration parameters. Finally, the output compression result and calibration parameters not only reduce the data volume, lower the transmission burden, but also automate and precisely perform the remote calibration process, greatly enhancing the intelligent level of the concentrator in data transmission and performance calibration, and effectively solving the problems of low data transmission efficiency, high cost and inaccurate remote calibration in the prior art.

[0040] The following further describes the present invention in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are 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.

[0042] Figure 1 It is a schematic flowchart of the optimization method for concentrator data compression and remote calibration provided by the embodiment of the present invention;

[0043] Figure 2 It is a schematic block diagram of the optimization system for concentrator data compression and remote calibration provided by the embodiment of the present invention;

[0044] Figure 3 It is a schematic block diagram of the computer device provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0046] It should be understood that when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0047] It should also be understood that the terms used in this specification of the present invention are merely for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in this specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0048] It should be further understood that the term "and / or" used in this specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0049] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of the concentrator data compression and remote calibration optimization method provided by an embodiment of the present invention. This concentrator data compression and remote calibration optimization method is applied to a concentrator. This method effectively solves the problems of concentrator data transmission and remote calibration, and significantly improves the performance and intelligence level of the concentrator. First, the original data of the concentrator is acquired and preprocessing operations are performed to obtain an operation result. Then, the operation result is input into a data compression model for data compression. The data compression model extracts shallow features and deep features, fuses them to form a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result. During the remote calibration process, the operation result is input into the remote calibration model, and at the same time, the compression result is used as auxiliary information. The remote calibration model is used to realize remote automatic calibration based on a multi-reference object system and analyze the relationship between different reference objects and calibration parameters to obtain calibration parameters. This method not only improves the data transmission efficiency, reduces the transmission cost and energy consumption, but also realizes the automation and high precision of remote calibration, thus solving the problems faced by the existing concentrator data transmission and remote calibration.

[0050] Figure 1 is a schematic flowchart of the concentrator data compression and remote calibration optimization method provided by an embodiment of the present invention. As Figure 1 shown, this method includes the following steps S110 to S140.

[0051] S110. Acquire the original data of the concentrator and perform preprocessing operations to obtain an operation result.

[0052] In this embodiment, the operation result refers to the original data of the concentrator after preprocessing, which has been cleaned and normalized to improve the quality.

[0053] Specifically, first, the original data of the concentrator is collected, which includes sensor data, device status data, etc. Then, the collected original data is preprocessed, including data cleaning to remove noise and incorrect data points, and normalization to unify the data into a specific range, thereby improving the consistency and stability of the data. Through these preprocessing operations, the obtained operation result will serve as the basis for subsequent feature extraction and fusion steps, providing a high-quality data basis for subsequent feature extraction and fusion steps.

[0054] S120. Input the operation result into a data compression model for data compression to obtain a compression result; wherein, the data compression model first extracts shallow features and deep features from the operation result, fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result.

[0055] In this embodiment, the shallow features include short-term fluctuation patterns in sensor data and instantaneous change features in device status data; the deep features include long-term trends in sensor data and periodic change patterns in device status data.

[0056] The compression result refers to a data representation form with a smaller data volume and key information retained obtained by processing the operation result through a data compression model, which helps reduce the data transmission volume, lower the transmission cost, and energy consumption.

[0057] The data compression model includes a feature extraction network, a feature fusion network, and a compression network, which is trained using the collected data, and the BCDELoss loss function is used to optimize the model.

[0058] The compression network includes a multi-layer perceptron and an autoencoder, which can process structured data and learn the non-linear representation of the data. If the ability to generate data needs to be maintained during the compression process, a generative adversarial network can also be considered.

[0059] Before data transmission, feature extraction and fusion are performed on the original data, and then it is compressed through a data compression model to reduce the data transmission volume. The specific implementation is as follows:

[0060] Shallow feature extraction: The preprocessed original data is subjected to convolution operations and downsampling through a convolutional layer. The convolution operation extracts the spatial features of the data by applying filters (convolution kernels), and the downsampling reduces the size of the feature map and increases the number of channels through the stride parameter, thereby reducing the data spatial dimension and enhancing the feature expression ability. Then, the feature map enters the batch normalization layer to accelerate network convergence and enhance stability, and then passes through the ReLU activation function to increase the network's non-linearity ability, enabling the network to learn more complex data patterns.

[0061] Deep feature extraction: Use a deep neural network structure such as the C2f layer to perform deeper feature extraction on the data. The C2f layer is specifically designed to extract high-level features. The input original information is directly passed to subsequent levels for Concat splicing through residual links, retaining more original information, avoiding information loss and alleviating the vanishing gradient problem, and enhancing the network's deep feature expression ability.

[0062] Feature fusion: Adopt a feature fusion network to fuse shallow features and deep features. First, use upsampling and downsampling techniques to adjust the spatial dimensions of the features to ensure that the sizes of different feature maps are consistent. Then, perform fusion through operations such as splicing and weighted summation to form a comprehensive feature representation.

[0063] Based on the above feature extraction and fusion process, a data compression model is constructed. The model includes a feature extraction network, a feature fusion network, and a compression network. The collected data is used to train the model. The extracted data features are fused and then input into the compression network for compression processing to generate a compressed data representation. During training, the BCDELoss loss function is used to optimize the model to ensure that the compressed data retains the key information of the original data.

[0064] Before data transmission, the original data is first subjected to the above-mentioned feature extraction and fusion, and then compressed through the trained data compression model to remove redundant information, reduce the data transmission volume, improve the transmission efficiency, and reduce the transmission cost and energy consumption.

[0065] In one embodiment, the above step S120 may include steps S121 to S124.

[0066] S121. Use a convolutional layer to perform convolution operations and downsampling on the operation result, extract the local features of the data and the features corresponding to the basic patterns, and enhance the feature expression ability and network performance through the batch normalization layer and the ReLU activation function to obtain shallow features.

[0067] In this embodiment, the shallow features mainly refer to the local features and basic patterns in the data, such as information on the edges, textures, and colors of the data. In concentrator data, the shallow features may include short-term fluctuation patterns in sensor data, instantaneous changes in device status data, etc.

[0068] The preprocessed original data is subjected to convolution operations and downsampling through a convolutional layer. The convolution operation extracts the spatial features of the data by applying filters (convolution kernels). For example, when processing image data, the convolution kernel can extract local features such as edges and textures in the data; when processing time series data, the convolution kernel can extract short-term fluctuation patterns in the data.

[0069] Downsampling reduces the size of the feature map and increases the number of channels through the stride parameter, thereby reducing the data spatial dimension and enhancing the feature expression ability. This step helps to reduce the complexity and computational amount of the data while retaining the key information of the data.

[0070] After that, the feature map enters the batch normalization layer to accelerate network convergence and enhance stability. Batch normalization can ensure that the data received by each layer of the network has the same distribution, thereby improving the training efficiency and model performance.

[0071] Then, the ReLU activation function is used to increase the network's non-linear ability, enabling the network to learn more complex data patterns. The ReLU activation function scales the output value to the range of, which can effectively avoid the problem of gradient disappearance and improve the training speed and performance of the network.

[0072] S122. Use a deep neural network structure to perform deep feature extraction on the operation result, obtain complex patterns and high-level features in the data, and perform secondary processing through residual connections to obtain deep features.

[0073] In this embodiment, deep features mainly refer to complex patterns and high-level features in the data, such as long-term trends, periodic patterns, and high-level features of device operating states in the data. In concentrator data, deep features can include long-term trends in sensor data, periodic change patterns in device status data, etc.

[0074] Use a deep neural network structure such as the C2f layer to perform deeper feature extraction on the data. The C2f layer is specifically designed to extract high-level features. By performing multiple convolution and pooling operations on the data, complex patterns and high-level features in the data are gradually extracted.

[0075] The input original information is directly passed to the subsequent levels for Concat splicing through residual connections, retaining more original information, avoiding information loss and alleviating the problem of gradient disappearance, and enhancing the deep feature expression ability of the network. Residual connections can ensure the efficient transmission of information in the deep network, improving the training efficiency and performance of the model.

[0076] S123. Using a feature fusion network, after spatially adjusting the shallow features and the deep features by using upsampling and downsampling techniques, they are then fused to form a comprehensive representation feature.

[0077] In this embodiment, the comprehensive representation feature refers to a feature representation that is formed after processing the shallow features and the deep features through a feature fusion network and integrates information at different levels. It more comprehensively reflects the characteristics of the data and can provide a richer information basis for subsequent tasks such as remote calibration.

[0078] Using a feature fusion network, the shallow features and the deep features are fused. First, upsampling and downsampling techniques are used to adjust the spatial dimensions of the features to ensure that the sizes of different feature maps are consistent. Upsampling can be achieved through interpolation methods (such as transposed convolution operations) or other magnification techniques to increase the size of the feature map to match the size of other branches or layers; downsampling can be achieved through pooling operations, which are used to reduce the size and computational amount of the feature map while increasing the receptive field of the feature map.

[0079] Then, operations such as concatenation and weighted summation are performed for fusion to form a comprehensive feature representation. The fused feature map can comprehensively consider the multi-scale information of the data and improve the model's ability to understand the data and classification accuracy.

[0080] Using a feature fusion network, the shallow features and the deep features are fused. First, upsampling and downsampling techniques are used to adjust the spatial dimensions of the features to ensure that the sizes of different feature maps are consistent. Upsampling can be achieved through interpolation methods (such as transposed convolution operations) or other magnification techniques to increase the size of the feature map to match the size of other branches or layers; downsampling can be achieved through pooling operations, which are used to reduce the size and computational amount of the feature map while increasing the receptive field of the feature map. In addition, channel adjustment is also required. The number of channels of the feature map is adjusted through convolution operations to make it consistent with the number of channels of other feature maps for effective fusion operations.

[0081] In the feature fusion network, first, the spatial dimensions of the shallow features and the deep features are adjusted. For feature maps with a smaller size, upsampling techniques are used for magnification. Common upsampling methods include nearest neighbor interpolation and bilinear interpolation. Nearest neighbor interpolation increases the size of the feature map by copying the values of neighboring pixels, while bilinear interpolation generates new pixel values by calculating the weighted average of surrounding pixels. In addition, transposed convolution operations (also known as deconvolution) can also be used for upsampling to expand the size of the feature map by learning the reverse convolution process.

[0082] For feature maps with larger sizes, downsampling techniques are used for reduction. Common downsampling methods are pooling operations, such as max pooling and average pooling. Max pooling selects the maximum value within a local region as the pooling result, while average pooling calculates the average value within the local region. Pooling operations can not only reduce the size of the feature map, but also lower the computational complexity and increase the receptive field of the feature map, enabling the model to capture more extensive context information.

[0083] While adjusting the spatial dimensions, channel adjustment is also required to make the number of channels of different feature maps consistent. Channel adjustment is usually achieved through convolution operations. For example, a 1×1 convolutional layer is used to change the number of channels of the feature map. 1×1 convolution can increase or decrease the number of channels of the feature map without changing its spatial dimensions. In this way, feature maps from different sources can be adjusted to the same number of channels, thus preparing for subsequent feature fusion operations.

[0084] Finally, the shallow features and deep features with adjusted spatial dimensions and channel numbers are fused. Common fusion methods include feature concatenation and weighted summation. Feature concatenation connects multiple feature maps along the channel dimension to form a richer feature representation. Weighted summation performs a linear combination of multiple feature maps and emphasizes important feature information by learning weights. The fused feature map contains the comprehensive information of shallow and deep features, can represent the features of the data more comprehensively, and provides more powerful feature support for subsequent classification or regression tasks.

[0085] S124. Compress the comprehensive representation features using a compression network to obtain a compression result.

[0086] Based on the above feature extraction and fusion processes, a data compression model is constructed. The model includes a feature extraction network, a feature fusion network, and a compression network. The feature extraction network is used to extract shallow and deep features of the data; the feature fusion network is used to fuse the extracted features to form a comprehensive feature representation; the compression network is used to compress the fused features to generate a compressed data representation.

[0087] The model is trained using the collected data. The data features are extracted by the feature extraction network, fused by the feature fusion network, and then input into the compression network for compression processing to generate a compressed data representation. During the training process, the BCDE Loss function is used to optimize the model to ensure that the compressed data can retain the key information of the original data to the greatest extent. The BCDE Loss function combines binary cross-entropy loss and Dice loss, can effectively measure the difference between the model output and the true label, and prompts the model to learn to approximate the true label.

[0088] Before data transmission, the original data is first subjected to the above-mentioned feature extraction and fusion, and then compressed by the trained data compression model to remove redundant information, reduce the amount of data transmitted, improve the transmission efficiency, and reduce the transmission cost and energy consumption. The compressed data occupies less bandwidth resources during the transmission process, can complete the transmission task faster, and reduces energy consumption at the same time.

[0089] Through the above process, the data compression model can effectively extract and fuse the key features of the data, generate a compressed data representation, and provide efficient support for subsequent data transmission and processing.

[0090] The operation result is input into the data compression model for data compression. The data compression model first uses the convolutional layer to perform convolution operations and downsampling on the operation result, extracts the local features of the data and the features corresponding to the basic patterns, and enhances the feature expression ability and network performance through the batch normalization layer and the ReLU activation function to obtain shallow features. The shallow features include the short-term fluctuation patterns in the sensor data and the instantaneous change features in the device status data. Then, the deep neural network structure is used to perform deep feature extraction on the operation result, obtain the complex patterns and high-level features in the data, and perform secondary processing through the residual connection to obtain deep features. The deep features include the long-term trends in the sensor data and the periodic change patterns in the device status data. Then, a feature fusion network is adopted to use upsampling and downsampling techniques to adjust the spatial dimensions of the shallow features and the deep features, and then fuse them to form a comprehensive representation feature.

[0091] By inputting the operation result into the data compression model, using the convolutional layer and downsampling to extract shallow features, including short-term fluctuation patterns and instantaneous change features, and enhancing the feature expression ability and network performance through the batch normalization layer and the ReLU activation function; then using the deep neural network to extract deep features, such as long-term trends and periodic change patterns, and processing through the residual connection; finally, adopting the feature fusion network to adjust the spatial dimensions and fuse the shallow and deep features to form a comprehensive representation feature, realizing the efficient compression of data, retaining key information, improving the data transmission efficiency, reducing the transmission cost and energy consumption, and at the same time providing a refined data basis for remote calibration, ensuring the automation and high precision of calibration, thereby enhancing the performance and intelligent level of the concentrator.

[0092] S130: Input the operation result into the remote calibration model, and use the compressed result as auxiliary information for remote calibration to obtain calibration parameters; wherein, the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference object system, and analyzing the relationship between different reference objects and the calibration parameters.

[0093] In this embodiment, the calibration parameters refer to the key values used to describe the performance indicators of the concentrator in different working states, such as the sensitivity of the sensor, the response time of the device, the signal strength, etc. These parameters are crucial for ensuring the normal operation and data accuracy of the concentrator. Through remote calibration, automatic and high-precision adjustment of the concentrator can be achieved, improving its working efficiency and reliability.

[0094] The training process of the remote calibration model includes:

[0095] Use the labeled dataset to train the teacher network, use the teacher network to assign pseudo-labels to the unlabeled dataset to form a pseudo-labeled dataset, and then train the student network together with the labeled dataset.

[0096] The training process of the remote calibration model adopts the knowledge distillation technology, including using the labeled dataset to train the teacher network, using the teacher network to assign pseudo-labels to the unlabeled dataset to form a pseudo-labeled dataset, and then training the student network together with the labeled dataset. First, use the labeled dataset to train the teacher network so that it can extract effective feature representations. Then, use the trained teacher network to predict the unlabeled dataset to generate pseudo-labels. These pseudo-labels, as a form of marking for the unlabeled data, are used for subsequent training of the student network. The student network is jointly trained by combining the labeled dataset and the pseudo-labeled dataset, thus learning a more extensive image representation and enhancing the model's generalization ability to different data distributions. At the same time, various data augmentation techniques, such as cropping, rotation, color enhancement, adding noise, etc., are also used during the training process, enabling the model to handle various changes and noises, and further improving its robustness and generalization ability.

[0097] In one embodiment, the above-mentioned remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference object system, and analyzing the relationship between different reference objects and the calibration parameters, including:

[0098] According to the working environment and data characteristics of the concentrator, select multiple reference objects with different characteristics;

[0099] Use image processing technology and positioning technology means to extract the shape, size, color, and position feature information of the selected reference objects to obtain the feature information of the reference objects;

[0100] With the help of data mining and machine learning algorithms, explore the association between the feature information of different reference objects and the calibration parameters to obtain labeled data;

[0101] Use the labeled data to train the teacher network, generate a pseudo-label extended dataset, and then use the pseudo-label extended dataset to train the student network to construct a remote calibration model based on a multi-reference object system.

[0102] In this embodiment, potential reference object types are determined according to the working environment and data characteristics of the concentrator. For example, sensor nodes, network access points, or other fixed devices can be selected as reference object candidates.

[0103] Considering that different reference objects have different characteristics in terms of shape, size, color, and position, etc., select reference objects with significant features and easy to identify, such as sensor nodes with unique shapes or devices with specific color markings.

[0104] To improve the accuracy and robustness of calibration, multiple reference objects with different characteristics are selected. For example, sensor nodes at different positions, different types of devices, or network access points at different heights can be selected as reference objects.

[0105] Ensure that the selected reference objects are within the data acquisition range of the concentrator, and their characteristics can be stably detected and recognized in the data.

[0106] For each selected reference object, extract its characteristic information. For example, use image processing technology to extract the shape, size, and color characteristics of the reference object; use a positioning system to obtain the position information of the reference object.

[0107] For the shape feature, an edge detection algorithm can be used to identify the contour of the reference object, and its shape parameters (such as the radius of a circle, the length and width of a rectangle, etc.) can be determined through geometric analysis.

[0108] For the size feature, the size information of the reference object can be obtained by measuring its size in the data (such as the number of pixels, the coordinate range, etc.), and it can be converted into the actual physical size.

[0109] For the color feature, color analysis can be performed on the data to extract information such as the color histogram or the main color components of the reference object.

[0110] For the position feature, a positioning system (such as GPS, Wi-Fi positioning, etc.) can be used to obtain the precise position coordinates of the reference object and record them as part of the characteristic information.

[0111] Collect data samples of the concentrator in different calibration states, including data differences before and after calibration and corresponding calibration parameters, etc.

[0112] Preprocess the collected data, such as data cleaning, normalization, etc., to improve the data quality.

[0113] Use data mining and machine learning algorithms to analyze the correlation between the characteristic information of different reference objects and the calibration parameters. For example, correlation analysis can be used to determine the linear or non-linear relationship between the characteristics of the reference object and the calibration parameters.

[0114] Establish a mapping model between features and calibration parameters. For example, algorithms such as linear regression, support vector machines, and neural networks can be used to construct a mapping model based on the analyzed correlation relationship, enabling the model to predict calibration parameters according to the input reference object feature information.

[0115] Use the labeled dataset to train the teacher network, which can extract effective feature representations when processing data.

[0116] Use the trained teacher network to predict the unlabeled dataset, generate pseudo-labels, and form a pseudo-labeled dataset.

[0117] The student network training will use the labeled dataset and the pseudo-labeled dataset. In this way, the student network can perform joint training from labeled and unlabeled data and learn more extensive feature representations.

[0118] During the training process, perform various data augmentation operations on the input data, such as cropping, rotation, color enhancement, adding noise, etc., to enhance the generalization ability of the model.

[0119] Extract feature vectors from the augmented data through the student network and the teacher network respectively, calculate the cross-entropy loss, and prompt the student network to imitate the output of the teacher network to learn better feature representations.

[0120] Use SGD (Stochastic Gradient Descent) to perform backpropagation and parameter update on the student network. At the same time, through the Exponential Moving Average (EMA) method, update the parameters of the student network to the teacher network to maintain the stability and effectiveness of the teacher network parameters.

[0121] Design the structure of the remote calibration model, including the input layer, feature extraction layer, feature fusion layer, output layer, etc.

[0122] The input layer receives the current data from the concentrator, as well as the feature results after feature extraction and fusion and the results after data compression as auxiliary information.

[0123] The feature extraction layer is used to extract key features from the input data, and structures such as convolutional layers and recurrent neural network layers can be used.

[0124] The feature fusion layer fuses the extracted features with the auxiliary information to form a comprehensive feature representation, and operations such as concatenation and weighted summation can be used.

[0125] The output layer outputs calibration parameters, and structures such as fully connected layers or regression layers can be used.

[0126] The remote calibration model is trained using a training set, and the model parameters are adjusted through optimization algorithms (such as SGD, Adam, etc.) so that the model can accurately predict the calibration parameters based on the input data.

[0127] During the training process, methods such as cross-validation are used to evaluate the performance of the model to avoid overfitting and underfitting phenomena.

[0128] According to the verification results, the model structure and parameters are adjusted to optimize the model performance and improve the calibration accuracy and reliability.

[0129] For step S130 of this embodiment, it may include:

[0130] The operation result is input into the remote calibration model to analyze the performance of the concentrator in different working states, identify various characteristic information related to calibration. At the same time, the compression result is input into the remote calibration model as auxiliary information, and the compression result is used to optimize feature matching, filter outliers, and assist the teacher and student networks in learning to obtain calibration parameters.

[0131] The trained remote calibration model is deployed into the concentrator system.

[0132] During the remote calibration process, the currently collected data is input into the remote calibration model, and at the same time, the feature result after feature extraction and fusion and the result after data compression are provided as auxiliary information. The remote calibration model comprehensively considers the input data and auxiliary information, and through internal calculations and analyses of the model, outputs the corresponding calibration parameters to achieve remote automatic calibration.

[0133] Through the above process, a remote calibration model is constructed based on a multi-reference object system, which can make full use of the characteristic information of different reference objects, improve the calibration accuracy and reliability, and provide strong support for the remote calibration of the concentrator.

[0134] Specifically, in the multi-reference object system, each reference object has characteristic information. By comparing the key features in the compression result with the characteristic information of the reference object, the working state of the concentrator and the calibration parameters that need to be adjusted are located.

[0135] During the remote calibration process, the currently collected data is input into the remote calibration model, and at the same time, the operation results (including the feature result after feature extraction and fusion and the result after data compression) in the above compression process are input into the remote calibration model as auxiliary information to obtain more accurate calibration parameters, specifically as follows:

[0136] During remote calibration, in addition to inputting the currently collected data into the remote calibration model, the feature results after feature extraction and fusion and the results after data compression are also input into the model together. These feature results and compression results contain the key features and refined information of the data, which can provide a richer data basis for remote calibration.

[0137] The features extracted during data compression can be used as auxiliary information for remote calibration to help the remote calibration model better understand the current data. For example, the key features of the data identified during the compression process can enhance the remote calibration model's perception of the device state, enabling the model to more accurately identify the operating state and characteristics of the device.

[0138] The remote calibration model is constructed based on a multi-reference object system and performs calibration by analyzing the relationships between different reference objects and calibration parameters. Combining the key features in the compression result with the current data can enable the model to more accurately identify and utilize the features of the reference objects, thereby improving the calibration accuracy.

[0139] The remote calibration model comprehensively considers the information in the current data and the compression result, and through complex calculations and analyses, outputs the corresponding calibration parameters. Since the compression result has removed redundant information and retained the core features of the data, this helps the model to perform calibration more quickly and accurately, improving the calibration efficiency and reliability.

[0140] In this way, inputting the operation result into the remote calibration model and using the compression result as auxiliary information for remote calibration can make full use of the key features in the data and improve the accuracy and reliability of remote calibration.

[0141] Specifically, the training process of the remote calibration model includes training a teacher network using a labeled data set, using the teacher network to assign pseudo-labels to an unlabeled data set to form a pseudo-labeled data set, and then training a student network together with the labeled data set. Specifically, according to the concentrator working environment and data characteristics, multiple reference objects with different features are selected. Using image processing technology and positioning technology means, the shape, size, color, and position feature information of the selected reference objects are extracted to obtain the feature information of the reference objects. With the help of data mining and machine learning algorithms, the associations between the feature information of different reference objects and calibration parameters are mined to obtain labeled data. The teacher network is trained using the labeled data to generate a pseudo-label extended data set, and then the student network is trained with the pseudo-label extended data set to construct a remote calibration model based on a multi-reference object system.

[0142] The operation result is input into the remote calibration model, and the compression result is used as auxiliary information for remote calibration. The remote calibration model is constructed based on a multi-reference object system and performs calibration by analyzing the relationships between different reference objects and calibration parameters. During the remote calibration process, the remote calibration model comprehensively considers the information in the operation result and the compression result. As auxiliary information, the compression result can provide a more refined data basis for the remote calibration model, enabling the model to more quickly and accurately identify and adjust calibration parameters. For example, the feature information in the compression result can enhance the remote calibration model's perception of the device state and help the model more accurately identify the operating state and characteristics of the device. This helps improve the accuracy and reliability of calibration.

[0143] By inputting the operation result into the remote calibration model and using the compression result as auxiliary information, high-precision remote calibration is achieved. Specifically, the remote calibration model is constructed based on a multi-reference object system and can accurately identify and adjust calibration parameters by analyzing the relationships between different reference objects and calibration parameters. In terms of calibration parameters, they are key values used to describe the performance indicators of the concentrator in different working states, such as the sensitivity of the sensor, the response time of the device, etc. These parameters are crucial for ensuring the normal operation and data accuracy of the concentrator. The training process of the remote calibration model involves training a teacher network using a labeled dataset, then using the teacher network to assign pseudo-labels to an unlabeled dataset to form a pseudo-labeled dataset, and then training a student network together with the labeled dataset. In this way, the student network can learn a more extensive feature representation, thereby improving the generalization ability and calibration accuracy of the model. During the actual remote calibration process, the operation result is directly input into the remote calibration model to analyze the performance of the concentrator in different working states and identify various feature information related to calibration. At the same time, the compression result is input into the remote calibration model as auxiliary information, and the key features in the compression result are used to optimize the feature matching process to more accurately locate the working state of the concentrator and the calibration parameters that need to be adjusted. In addition, the compression result is also used to filter out outliers, remove noise and error data points, further improving the accuracy of calibration. Finally, the compression result assists the teacher and student networks in learning, helping the model better understand the data features and improving the learning efficiency. Through these steps, accurate calibration parameters are finally obtained, realizing efficient remote calibration of the concentrator.

[0144] In this embodiment, a teacher network is trained using a labeled dataset. The teacher network can extract effective feature representations when processing images and learn important information and patterns in the data.

[0145] The trained teacher network is used to predict an unlabeled dataset to generate pseudo-labels. These pseudo-labels can be regarded as a form of marking for the unlabeled dataset and are used for subsequent training of the student network.

[0146] The student network training will use the labeled dataset and the pseudo-labeled dataset generated by the teacher network. In this way, the student network can perform joint training from labeled and unlabeled data, and learn a more extensive image representation through pseudo-labels.

[0147] During the training process, various data augmentation operations are performed on the input images, such as cropping (local and global perspectives), color perturbation, Gaussian blur, exposure enhancement, etc. These augmentation operations help increase the robustness and generalization ability of the model to different variant inputs.

[0148] For the images after data augmentation, feature vectors are extracted through the student network and the teacher network respectively. These feature vectors are scaled between 0 and 1 through the softmax function, and their cross-entropy loss is calculated. This process encourages the student network to imitate the output of the teacher network, thus learning better feature representations.

[0149] Use SGD (Stochastic Gradient Descent) to perform backpropagation and parameter update on the student network to minimize the loss function. At the same time, through the Exponential Moving Average (EMA) method, the parameters of the student network are updated into the teacher network to maintain the stability and effectiveness of the teacher network parameters.

[0150] During the remote calibration process, the currently collected data is input into the remote calibration model. The model analyzes and processes the input data based on the relationship between the features learned during the training process and the calibration parameters.

[0151] At the same time, the compression result is provided to the remote calibration model as auxiliary information. The compression result undergoes feature extraction and fusion processing, removing redundant information and retaining the key features of the data. These key features can provide a more refined data basis for remote calibration, helping the model to more accurately identify and adjust the calibration parameters.

[0152] The remote calibration model comprehensively considers the information in the operation result and the compression result, and through complex calculations and analyses, outputs the corresponding calibration parameters. These calibration parameters can accurately reflect the state and performance of the concentrator, providing a reliable basis for remote automatic calibration.

[0153] In this way, inputting the operation result into the remote calibration model and using the compression result as auxiliary information for remote calibration can make full use of the depth information and feature representations in the data, improving the accuracy and reliability of remote calibration. At the same time, the remote calibration model based on the multi-reference object system can avoid the errors and uncertainties that may be brought by a single reference object, further improving the accuracy of the calibration result.

[0154] S140. Output the compression result and the calibration parameters.

[0155] Finally, the compressed result and calibration parameters are output for subsequent monitoring and management.

[0156] The method of this embodiment uses convolutional layers, C2f layers, and SPPF layers for feature extraction, which can extract the most important information from the original data and remove redundant data. These operations can generate feature vectors of fixed length, adapt to input data of different sizes, and maintain the consistency and stability of the output. Through feature extraction and fusion, the key information of the data is concentrated, reducing the amount of data, thereby accelerating the data transmission speed.

[0157] The application of the BCDELoss loss function enables the model to better learn data features during training, improve the compression efficiency, reduce information loss during data transmission, and ensure that the key information of the original data can be accurately restored after compression. This enables the data to be processed and parsed faster during transmission, further improving the transmission efficiency.

[0158] The reduction in the amount of data directly reduces the bandwidth resources required for data transmission. In the Internet of Things environment, bandwidth resources are limited and have a certain cost. Through efficient data compression, the demand for high bandwidth is reduced, thereby reducing the cost of data transmission.

[0159] A smaller amount of data consumes relatively less energy during transmission. Especially in large-scale Internet of Things deployments, a large amount of data transmission consumes a considerable amount of energy. By reducing the amount of data, the present invention helps to reduce the energy consumption of the entire system, which is particularly important for Internet of Things devices that rely on battery power or are energy-constrained.

[0160] The remote calibration model based on the multi-reference object system uses deep learning technology to construct a teacher network and a student network, and improves the generalization ability and performance of the model through methods such as knowledge distillation. This enables the remote calibration process to be carried out without manual intervention, and the concentrator can automatically output the corresponding calibration parameters according to the currently collected data, realizing a remote and automated calibration process, improving the calibration efficiency, and reducing the time and cost of manual operations.

[0161] The application of data augmentation techniques, such as cropping, rotation, color enhancement, etc., increases the diversity of training data, enabling the remote calibration model to adapt to different environments and conditions. This further improves the robustness and adaptability of the model, ensuring accurate remote calibration in various complex actual scenarios.

[0162] By analyzing the relationship between different reference objects and calibration parameters, the calibration parameters of the concentrator are accurately determined. The multi-reference object system provides richer information, avoiding errors and uncertainties that may be brought by a single reference object.

[0163] During the model training process, the labeled dataset and the pseudo-labeled dataset are used for joint training, enabling the student network to learn a more extensive image representation from both labeled and unlabeled data. This training method can improve the prediction accuracy of the model for calibration parameters, making the remote calibration results more accurate and meeting the requirements of high-precision applications.

[0164] In summary, the method of this embodiment can effectively improve the data transmission efficiency, reduce the transmission cost and energy consumption, and at the same time achieve the automation and high precision of remote calibration, thus significantly enhancing the performance and intelligence level of the concentrator and meeting the efficient and accurate requirements for data processing and device management in the Internet of Things environment.

[0165] For example: The original data obtained from the power concentrator may include current, voltage, and power readings from different sensors, as well as device status data (such as device operating status, fault alarms, etc.).

[0166] Remove outliers or incorrect readings, such as extremely large or small values that occasionally appear in a certain sensor. Scale all the data to the same range (such as between 0 and 1) for subsequent feature extraction and model training.

[0167] Use convolutional layers to perform convolutional operations on the normalized current and voltage data to identify short-term fluctuation patterns (such as peak and valley electricity consumption within a day). Then downsample to reduce the feature map size, and use batch normalization layers and ReLU activation functions to enhance the network performance. Use C2f layers to perform deep feature extraction on long-term trends (such as the average load change per day within a week), and use residual links to prevent information loss. Fuse the shallow and deep features into a comprehensive representation feature by concatenation or weighted summation. Input the comprehensive representation feature into a compression network (composed of a multi-layer perceptron and an autoencoder) to generate a compression result. The compression result retains the key information and reduces the bandwidth required for transmission.

[0168] Select multiple reference objects with different characteristics, such as substations A, B, and C as reference points. Use image processing techniques to extract the shape, size, color, and other feature information of these reference objects. Analyze the relationship between these reference object features and calibration parameters (such as sensor sensitivity, response time) through machine learning algorithms.

[0169] First, train the teacher network with the labeled dataset, then use the teacher network to assign pseudo-labels to the unlabeled dataset to form a pseudo-labeled dataset. Finally, train the student network by combining the labeled and pseudo-labeled datasets. In actual operation, input the current data together with the auxiliary information after feature extraction and compression into the remote calibration model, and output accurate calibration parameter adjustment suggestions, such as adjusting the sensitivity of certain sensors to improve the measurement accuracy. Finally, the system will output the compressed data result and the calculated calibration parameters for subsequent monitoring and management decisions.

[0170] In this example, through effective data preprocessing, feature extraction, data compression, and remote calibration, the efficient management and automatic calibration of concentrator data are achieved. This not only improves data transmission efficiency, reduces costs, but also ensures the accuracy and reliability of concentrator operation, thereby enhancing the performance of the entire smart grid system.

[0171] The above-mentioned concentrator data compression and remote calibration optimization method obtains the original concentrator data and performs preprocessing. Subsequently, a specially designed data compression model is used to extract shallow features such as short-term fluctuation patterns and deep features such as long-term trends, and these features are fused to form a comprehensive representation feature. Further, an efficient compression result is obtained through a compression network, thereby significantly improving data transmission efficiency, reducing transmission costs and energy consumption. At the same time, the operation result and the compression result are input into the remote calibration model as auxiliary information. This model is constructed based on the analysis of the relationship between different reference objects and calibration parameters in a multi-reference object system, and can achieve high-precision automatic calculation of calibration parameters. Finally, the output compression result and calibration parameters not only reduce the amount of data, lower the transmission burden, but also realize the automation and precision of the remote calibration process, greatly enhancing the intelligent level of the concentrator in data transmission and performance calibration, and effectively solving the problems of low data transmission efficiency, high cost, and inaccurate remote calibration in the prior art.

[0172] Figure 2 It is a schematic block diagram of a concentrator data compression and remote calibration optimization system 300 provided by an embodiment of the present invention. As Figure 2 shown, corresponding to the above concentrator data compression and remote calibration optimization method, the present invention also provides a concentrator data compression and remote calibration optimization system 300. The concentrator data compression and remote calibration optimization system 300 includes units for executing the above concentrator data compression and remote calibration optimization method, and this system can be configured in the concentrator. Specifically, please refer to Figure 2 , the concentrator data compression and remote calibration optimization system 300 includes an acquisition unit 301, a compression unit 302, a remote calibration unit 303, and an output unit 304.

[0173] An acquisition unit 301, configured to acquire the concentrator raw data and perform preprocessing operations to obtain an operation result; a compression unit 302, configured to input the operation result into a data compression model for data compression to obtain a compression result; wherein, the data compression model first extracts shallow features and deep features of the operation result, and fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result; a remote calibration unit 303, configured to input the operation result into a remote calibration model, and use the compression result as auxiliary information for remote calibration to obtain calibration parameters; wherein, the remote calibration model is constructed by collecting data samples of the concentrator in different calibration states, based on a multi-reference system, and analyzing the relationship between different references and calibration parameters; an output unit 304, configured to output the compression result and the calibration parameters.

[0174] In one embodiment, the compression unit 302 includes:

[0175] A shallow feature extraction subunit, configured to use a convolutional layer to perform convolutional operations and downsampling on the operation result, extract the features corresponding to the local features and basic patterns of the data, and enhance the feature expression ability and network performance through a batch normalization layer and a ReLU activation function to obtain shallow features; a deep feature extraction subunit, configured to use a deep neural network structure to perform deep feature extraction on the operation result, obtain complex patterns and high-level features in the data, and perform secondary processing through a residual link to obtain deep features; a fusion subunit, configured to use a feature fusion network, apply upsampling and downsampling techniques to adjust the spatial dimensions of the shallow features and the deep features, and then perform fusion to form a comprehensive representation feature; a compression subunit, configured to use a compression network to compress the comprehensive representation feature to obtain a compression result.

[0176] In one embodiment, the remote calibration unit 303 is configured to input the operation result into the remote calibration model, analyze the performance of the concentrator in different working states, identify various feature information related to calibration, and at the same time input the compression result as auxiliary information into the remote calibration model, and use the compression result to optimize feature matching, filter outliers and assist the teacher and student networks in learning to obtain calibration parameters.

[0177] In the multi-reference system, each reference has feature information. By comparing the key features in the compression result with the feature information of the reference, the working state of the concentrator and the calibration parameters to be adjusted are located.

[0178] It should be noted that those skilled in the art can clearly understand the specific implementation processes of the above-mentioned concentrator data compression and remote calibration optimization system 300 and each unit. For corresponding descriptions, reference can be made to the foregoing method embodiments. For the sake of convenience and conciseness of description, they will not be elaborated herein.

[0179] The above-mentioned concentrator data compression and remote calibration optimization system 300 can be implemented in the form of a computer program, and this computer program can run on a computer device as Figure 3 shown.

[0180] Please refer to Figure 3 , Figure 3 which is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server. Among them, the server can be an independent server or a server cluster composed of multiple servers.

[0181] Referring to Figure 3 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501. Among them, the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0182] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions. When the program instructions are executed, the processor 502 can be made to execute a concentrator data compression and remote calibration optimization method.

[0183] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0184] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can be made to execute a concentrator data compression and remote calibration optimization method.

[0185] The network interface 505 is used for network communication with other devices. Those skilled in the art can understand that Figure 3 the structure shown in

[0186] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. Among them, the processor 502 is used to run the computer program 5032 stored in the memory to implement all steps of the concentrator data compression and remote calibration optimization method.

[0187] It should be understood that in the embodiments of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0188] Those of ordinary skill in the art can understand that all or part of the processes of the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, and the storage medium is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0189] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes all steps of the concentrator data compression and remote calibration optimization method.

[0190] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, an optical disk, or other computer-readable storage media that can store program codes.

[0191] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0192] In several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0193] The steps in the method embodiments of the present invention can be adjusted, combined, and deleted according to actual needs. The units in the system embodiments of the present invention can be combined, divided, and deleted according to actual needs. In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0194] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention.

[0195] As mentioned above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. Method for optimizing data compression and remote calibration of concentrator, characterized in that include: Obtain the original data of the concentrator and perform preprocessing operations to obtain operation results; The operation result is input into a data compression model for data compression to obtain a compression result; wherein the data compression model first extracts shallow features and deep features from the operation result, and fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs the feature into a compression network for compression processing to obtain a compression result; The operation result is input into a remote calibration model, and the compression result is used as auxiliary information for remote calibration to obtain calibration parameters; wherein the remote calibration model is constructed by collecting data samples of the concentrator under different calibration states, based on a multi-reference system, and by analyzing the relationship between different reference objects and calibration parameters; Outputting the compression result and calibration parameters; The step of inputting the operation result into a data compression model to perform data compression to obtain a compression result includes: The operation result is convolved and downsampled by using a convolution layer to extract local features of the data and features corresponding to the basic pattern, and the feature expression capability and network performance are enhanced by a batch normalization layer and a ReLU activation function to obtain shallow features; Performing deep feature extraction on the operation results using a deep neural network structure to obtain complex patterns and high-level features in the data, and performing secondary processing through residual links to obtain deep features; A feature fusion network is used to adjust the spatial size of the shallow features and the deep features by using upsampling and downsampling techniques, and then the shallow features and the deep features are fused to form a comprehensive representation feature; The comprehensive representation features are compressed using a compression network to obtain a compression result.

2. The concentrator data compression and remote calibration optimization method according to claim 1, characterized in that The shallow features include short-term fluctuation patterns in sensor data and instantaneous change features in device status data; the deep features include long-term trends in sensor data and periodic change patterns in device status data.

3. The concentrator data compression and remote calibration optimization method according to claim 1, wherein The data compression model includes a feature extraction network, a feature fusion network and a compression network, which are trained using the collected data and the BCDELoss loss function is used to optimize the model.

4. The concentrator data compression and remote calibration optimization method according to claim 1, characterized in that The training process of the remote calibration model includes: The teacher network is trained using the labeled dataset, and the teacher network is used to assign pseudo labels to the unlabeled dataset to form a pseudo-labeled dataset, which is then used together with the labeled dataset to train the student network.

5. The concentrator data compression and remote calibration optimization method according to claim 4, wherein The remote calibration model is constructed by collecting data samples of the concentrator in different calibration states and analyzing the relationship between different reference objects and calibration parameters based on a multi-reference object system, including: According to the working environment and data characteristics of the concentrator, multiple reference objects with different characteristics are selected; Use image processing technology and positioning technology to extract the shape, size, color and position feature information of the selected reference object to obtain the feature information of the reference object; With the help of data mining and machine learning algorithms, the association between the characteristic information of different reference objects and the calibration parameters is mined to obtain labeled data; Train a teacher network using the labeled data to generate a pseudo-label extended dataset, and then train a student network with the pseudo-label extended dataset to construct a remote calibration model based on a multi-reference system.

6. The concentrator data compression and remote calibration optimization method according to claim 1, wherein Input the operation result into the remote calibration model, and use the compression result as auxiliary information for remote calibration to obtain calibration parameters, including: Input the operation result into the remote calibration model to analyze the performance of the concentrator in different working states and identify various characteristic information related to calibration. At the same time, input the compression result into the remote calibration model as auxiliary information, and use the compression result to optimize feature matching, filter outliers, and assist the teacher and student networks in learning to obtain calibration parameters.

7. The concentrator data compression and remote calibration optimization method according to claim 6, characterized in that The compression result optimizing feature matching includes: In a multi-reference system, each reference has characteristic information. By comparing the key features in the compression result with the characteristic information of the reference, the working state of the concentrator and the calibration parameters to be adjusted are located.

8. The concentrator data compression and remote calibration optimization system is characterized in that Include: An acquisition unit for acquiring the original data of the concentrator and performing preprocessing operations to obtain an operation result; A compression unit for inputting the operation result into a data compression model for data compression to obtain a compression result. The data compression model first extracts shallow features and deep features from the operation result, fuses the extracted shallow features and deep features to obtain a comprehensive representation feature, and then inputs it into a compression network for compression processing to obtain a compression result; A remote calibration unit for inputting the operation result into the remote calibration model and using the compression result as auxiliary information for remote calibration to obtain calibration parameters. The remote calibration model is constructed by collecting data samples of the concentrator in different calibration states and analyzing the relationship between different references and calibration parameters based on a multi-reference system; An output unit for outputting the compression result and calibration parameters; The compression unit includes: A shallow feature extraction sub-unit for performing convolution operations and downsampling on the operation result using a convolutional layer, extracting the features corresponding to the local features and basic patterns of the data, and enhancing the feature expression ability and network performance through a batch normalization layer and a ReLU activation function to obtain shallow features; A deep feature extraction sub-unit for performing deep feature extraction on the operation result using a deep neural network structure, obtaining complex patterns and high-level features in the data, and performing secondary processing through a residual link to obtain deep features; A fusion sub-unit for using a feature fusion network to adjust the spatial dimensions of the shallow features and the deep features using upsampling and downsampling techniques and then fusing them to form a comprehensive representation feature; A compression sub-unit for compressing the comprehensive representation feature using a compression network to obtain a compression result.

Images