Abnormal data restoration method and device based on self-attention, equipment and medium
By using long-term and short-term neural networks and self-attention codecs to build a multi-channel sequence data reconstruction model in the structural health monitoring system, the problem of poor repair of abnormal monitoring data in the existing technology is solved, and high accuracy and widespread Applicable data repair effects.
Patent Information
- Application Number
- CN202411137551.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-11-26
AI Technical Summary
The existing abnormal monitoring data repair methods, especially statistical methods, lack the effect. Although the machine learning method has good results, its principles are complex and its scope of application is small, it cannot effectively process various abnormal data in massive data in the structural health monitoring system.
The multi-channel sequence data reconstruction model is used based on long-term and short-term neural networks and self-attention encoding decoder. The normal monitoring data is reconstructed by training the model, and the reconstruction data is used instead of the abnormal data until the abnormal data is completely repaired.
It improves the accuracy and operation simplicity of data reconstruction, and is suitable for repairing abnormal monitoring data of various types and large ranges, providing more efficient data repair effects.
Smart Images

Figure CN119025839A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data monitoring technology, and in particular to a method, device, equipment and medium for repairing abnormal data based on self-attention. Background Art
[0002] Structural health monitoring systems are widely used in important structures such as long-span bridges and tall buildings, and collect massive amounts of monitoring data. By mining the potential information in the monitoring data, the structural performance is monitored. However, due to sensor anomalies or equipment transmission failures, the monitoring data inevitably contains various abnormal monitoring data, which seriously hinders the further mining and analysis of the monitoring data, and even leads to incorrect analysis structures. Therefore, it is very necessary to repair the abnormal monitoring data before mining the data.
[0003] There are two main types of abnormal monitoring data repair methods: statistical methods and machine learning methods. The former mainly include probability distribution-based methods and statistical regression-based methods, while the latter mainly include deep learning methods and support vector machines. Although the statistical method is simple in principle, the repair effect is insufficient; although the machine learning method has obvious advantages in the repair effect and repair range of abnormal monitoring data, its principle is complex and its scope of application is small.
[0004] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention
[0005] The present invention provides a method, device, equipment and medium for repairing abnormal monitoring data based on self-attention, thereby effectively solving the problems in the background technology.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is: a method for repairing abnormal data based on self-attention, comprising the following steps: Build a multi-channel sequence data reconstruction model based on long-short term neural network and self-attention encoder-decoder; Obtaining input and output data sets of multi-channel normal monitoring data, and training the data reconstruction model; Reconstructing some abnormal segment monitoring data using the trained data reconstruction model and the previous normal monitoring data; The reconstructed monitoring data is used to replace the corresponding part of the abnormal monitoring data, and the reconstruction and replacement of the abnormal monitoring data are repeated until the section of the abnormal monitoring data is completely repaired.
[0007] Furthermore, the multi-channel sequence data reconstruction model based on the long-term and short-term neural network and the self-attention encoder-decoder includes: Acquire the type of monitoring data to be repaired, and determine the number of sensors of the same type as the characteristic number of input and output data of the sequence data reconstruction model; Through the memory units, hidden units and encoded outputs in the long short-term neural network, it serves as the information transmission channel between the encoder and the decoder; Extract the core information from the monitoring data through the self-attention mechanism.
[0008] Furthermore, the step of obtaining the input and output data sets of the multi-channel normal monitoring data includes: Obtain historical monitoring data and randomly extract multiple normal monitoring data segments; Create input and output data sequence pairs and divide them into training set, validation set and test set.
[0009] Furthermore, when the input-output data sequence pair is produced, multiple normal monitoring data segments are produced using a time window of a fixed length.
[0010] Furthermore, the training of the data reconstruction model comprises the following steps: Training the data reconstruction model using the training set and the square error loss function; Use the validation set to check whether overfitting occurs during the training process; The test set is used to test the generalization performance of the trained data reconstruction model.
[0011] Further, the using the reconstructed normal monitoring data to replace the corresponding part of the abnormal monitoring data includes: The result output by the data reconstruction model includes data of multiple channels, and the reconstructed monitoring data is used to replace the abnormal monitoring data only in the channel where the abnormality occurs; If the length of the reconstructed monitoring data is greater than the length of the abnormal monitoring data, the reconstructed monitoring data is truncated to the length of the abnormal monitoring data; If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstructed monitoring data is used to replace the front part of the abnormal monitoring data.
[0012] Furthermore, the reconstruction and replacement of the repeated abnormal monitoring data includes: If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstruction and replacement process is repeated based on the partially repaired original monitoring data until all the abnormal monitoring data are completely repaired.
[0013] The present invention also includes a self-attention-based abnormal data repair device, using the above method, including: Modeling unit, used to build a multi-channel sequence data reconstruction model based on long-term and short-term neural networks and self-attention encoder-decoders; A training unit, used to obtain input and output data sets of multi-channel normal monitoring data and train the data reconstruction model; A reconstruction unit, used to reconstruct part of the abnormal segment monitoring data using the trained data reconstruction model and the previous normal monitoring data; The repair unit is used to replace the corresponding part of the abnormal monitoring data with the reconstructed monitoring data, and repeat the reconstruction and replacement of the abnormal monitoring data until the section of the abnormal monitoring data is completely repaired.
[0014] The present invention also includes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described above is implemented.
[0015] The present invention also includes a storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described above is implemented.
[0016] The beneficial effects of the present invention are as follows: the present invention extracts the core information of the structure monitoring data through the self-attention encoder-decoder, and considers the correlation of multi-channel data through multi-channel input and output, which improves the accuracy of the reconstructed data in principle, and has the advantages of simple operation and applicability to various types and a wide range of abnormal monitoring data repair. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0018] Figure 1 is a flow chart of the method in Example 1; Figure 2 It is a structural schematic diagram of the device in Example 1; Figure 3 is a flow chart of the abnormal data repair method in Example 2; Figure 4 It is the abnormal data reconstruction model constructed in Example 2; Figure 5 The method for preparing a data set for reconstructing a model in Example 2; Figure 6 is the training result of the reconstruction model in Example 2; Figure 7 and Figure 8 is a comparison diagram of the effects of the original data and the reconstructed data in Example 2; Fig. 9 A schematic diagram of the structure of a computer device. DETAILED DESCRIPTION
[0019] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0020] Embodiment 1: like Figure 1 As shown: A method for repairing abnormal data based on self-attention, comprising the following steps: Build a multi-channel sequence data reconstruction model based on long-short term neural network and self-attention encoder-decoder; Obtain input and output data sets of multi-channel normal monitoring data and train the data reconstruction model; Use the trained data reconstruction model and the previous normal monitoring data to reconstruct some abnormal segment monitoring data; The reconstructed monitoring data is used to replace the corresponding part of the abnormal monitoring data, and the reconstruction and replacement of the abnormal monitoring data are repeated until the section of the abnormal monitoring data is completely repaired.
[0021] The core information of the structure monitoring data is extracted through the self-attention encoder-decoder, and the correlation of multi-channel data is considered through multi-channel input and output, which improves the accuracy of the reconstructed data in principle. It has the advantages of simple operation and applicability to various types and a wide range of abnormal monitoring data repair.
[0022] A multi-channel sequence data reconstruction model is constructed using the long short-term memory network (LSTM) and the self-attention encoder-decoder (Transformer). LSTM is used to capture long-term dependencies in time series data, while the self-attention mechanism is used to mine complex patterns and correlations in the data. The model can process and reconstruct multi-channel monitoring data, thereby improving the accuracy and effect of abnormal data repair.
[0023] Combining the time series modeling capability of LSTM and the powerful feature extraction capability of the self-attention mechanism, the multi-channel sequence data reconstruction model is used to repair abnormal data. Its advantage is that it can effectively capture the long-term dependencies and complex patterns in the data, thereby providing accurate abnormal data repair results.
[0024] In this embodiment, a multi-channel sequence data reconstruction model is built based on a long-term short-term neural network and a self-attention encoder-decoder, including: Obtain the type of monitoring data that needs to be repaired, determine the number of sensors of the same type, and use this as the characteristic number of input and output data of the sequence data reconstruction model; Through the memory units, hidden units and encoded outputs in the long short-term neural network, it serves as the information transmission channel between the encoder and the decoder; Extract the core information from the monitoring data through the self-attention mechanism.
[0025] The memory units and hidden units in the LSTM network are used to capture the time series characteristics of the monitoring data. The LSTM network processes the input data sequence, generates a hidden state (encoded output), and uses it as an information transmission channel between the encoder and the decoder. The hidden state generated by the encoder is combined with the self-attention mechanism to reconstruct the target data sequence. The self-attention mechanism is applied in the decoder to extract the core information in the monitoring data by calculating the correlation between different positions in the input sequence. The self-attention mechanism helps to capture complex patterns and global dependencies, and improve the accuracy and effect of data reconstruction.
[0026] Among them, the input and output data sets for obtaining multi-channel normal monitoring data include: Obtain historical monitoring data and randomly extract multiple normal monitoring data segments; Create input and output data sequence pairs and divide them into training set, validation set and test set.
[0027] When making the input and output data sequence pairs, multiple normal monitoring data segments are made with a fixed-length time window.
[0028] By obtaining the input and output data sets of normal multi-channel monitoring data and using a fixed-length time window to create input and output data sequence pairs, the diversity and reliability of the model training data can be ensured. Reasonable division of the training set, validation set, and test set can effectively improve the generalization ability of the model and the actual application effect.
[0029] In this embodiment, training the data reconstruction model includes the following steps: The data reconstruction model is trained using the training set and the squared error loss function; Use the validation set to see if overfitting occurs during training; Use the test set to test the generalization performance of the model reconstructed from the trained data.
[0030] By using the training set to train the model, monitoring the overfitting on the validation set, and using the test set to evaluate the generalization performance of the model, we can ensure that the data reconstruction model has good performance and practical application effects. This method can not only effectively prevent overfitting, but also verify the performance of the model on unseen data, ensuring its reliability and robustness.
[0031] As a preferred embodiment of the above, using the reconstructed normal monitoring data to replace the corresponding part of the abnormal monitoring data includes: The output of the data reconstruction model includes data from multiple channels, and the reconstructed monitoring data is used to replace the abnormal monitoring data only in the channel where the abnormality occurs; If the length of the reconstructed monitoring data is greater than the length of the abnormal monitoring data, the reconstructed monitoring data is truncated to the length of the abnormal monitoring data; If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstructed monitoring data is used to replace the front part of the abnormal monitoring data.
[0032] Reconstruction and replacement of repeated abnormal monitoring data, including: If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstruction and replacement process is repeated based on the partially repaired original monitoring data until all the abnormal monitoring data are completely repaired.
[0033] Only in the channel where the abnormality occurs, the reconstructed monitoring data is used to replace the abnormal monitoring data, and the data of other normal channels remain unchanged. If the length of the reconstructed data is greater than the length of the abnormal data, the reconstructed monitoring data is truncated to the length of the abnormal monitoring data; if the length of the reconstructed data is less than the length of the abnormal data, the reconstructed monitoring data is used to replace the front part of the abnormal monitoring data; based on the partially repaired original monitoring data, the reconstruction and replacement process is repeated until all abnormal monitoring data are completely repaired. Through the above steps, the reconstructed normal monitoring data can be used to replace the abnormal monitoring data more efficiently, and the reconstruction and replacement process can be repeated when necessary to ensure that all abnormal monitoring data are completely repaired. This method not only improves the accuracy of data repair, but also handles the problem of inconsistency between the length of the reconstructed data and the abnormal data, ensuring the final data integrity and reliability.
[0034] like Figure 2 As shown, this embodiment also includes an abnormal data repair device based on self-attention, using the above method, including: Modeling unit, used to build a multi-channel sequence data reconstruction model based on long-term and short-term neural networks and self-attention encoder-decoders; A training unit, used to obtain input and output data sets of multi-channel normal monitoring data and train a data reconstruction model; A reconstruction unit, used to reconstruct some abnormal segment monitoring data using the trained data reconstruction model and the previous normal monitoring data; The repair unit is used to replace the corresponding part of the abnormal monitoring data with the reconstructed monitoring data, and repeat the reconstruction and replacement of the abnormal monitoring data until the section of the abnormal monitoring data is completely repaired.
[0035] Embodiment 2: like Figure 3 FIG. 1 is a flow chart of a method for repairing abnormal monitoring data based on a self-attention encoder / decoder according to an embodiment of the present invention, and the method comprises the following steps: Step 1: Build a data reconstruction model.
[0036] like Figure 4 As shown, according to the type of monitoring data that needs to be repaired, the number of sensors of the same type in the entire health monitoring system is determined as the feature number of the input data; multiple long short-term neural network units (LSTM) are established as encoders of the input data; the hidden units, memory units and encoded outputs in the long short-term neural network units are passed to the decoder; the decoder is composed of multiple attention mechanisms and long short-term neural network units, which reconstruct the information passed by the encoder into monitoring data.
[0037] Step 2: Establish input and output data sets and complete data reconstruction model training.
[0038] Specifically, multiple segments of normal monitoring data are manually selected from historical monitoring data, and the following methods are used: Figure 5 The time window shown in the figure is truncated by sliding to generate multi-channel input and output data sequence pairs, and the data set composed of the data sequence pairs is divided into training set, validation set and test set in a ratio of 3:1.5:1.5. The training set is used to train the data to rebuild the model, the validation set is used to determine whether the training process is overfitting, and the test set is used to verify the generalization performance of the training results.
[0039] Step 3: Reconstruct abnormal monitoring data.
[0040] Specifically, multi-channel normal monitoring data before the abnormal data occurs is intercepted as an input data sequence, which is input into a trained data reconstruction model to obtain a reconstruction result of part of the abnormal monitoring data.
[0041] Step 4: Replace abnormal data.
[0042] Specifically, the length of the abnormal monitoring data is counted and the length of the abnormal monitoring data is determined. If the length of the reconstructed data is greater than the length of the abnormal data, the reconstructed data is first truncated to the length of the abnormal data and then directly replaces the abnormal monitoring data of the corresponding channel. If the length of the reconstructed data is less than the length of the abnormal monitoring data, the reconstructed data only replaces the front part of the abnormal detection data of the corresponding channel.
[0043] Step 5, loop repair.
[0044] Specifically, if the monitoring data after repair in step 4 still contains abnormal monitoring data, steps 3 and 4 are repeated until the abnormal monitoring data is completely repaired.
[0045] To further illustrate the effectiveness of the present invention, a typical bridge monitoring data case is used to illustrate the specific implementation steps of the abnormal monitoring data repair method based on the self-attention encoder-decoder.
[0046] Step 1: Model Building Taking the displacement monitoring data of a bridge end as an example, the bridge health monitoring system is equipped with 4 displacement sensors at the beam end, so the number of features of the input and output data of the data reconstruction model is 4; the reconstruction model encoder is composed of a long-short-term neural network, which contains only one hidden layer with 300 units; the reconstruction model decoder is composed of a self-attention mechanism and a long-short-term neural network, which contains one hidden layer and a linear activation function layer, and the number of units in the hidden layer is 300.
[0047] Step 2: Dataset creation and training.
[0048] One day of normal monitoring data (86,400 data points / each channel) was selected from the historical monitoring data of the four sensors, and the input and output sequence pairs were intercepted according to two time windows of length 150 and 60 to form the training set, validation set and test set. The above data set was used to complete the training of the data reconstruction model, with 20 training rounds. The training results are shown in the figure below. Figure 6 shown.
[0049] Step 3: Data reconstruction.
[0050] Normal monitoring data with a length of 4200 are randomly selected from 4 sensors, and part of the data of two channels is randomly discarded (the discarded length is 240) to form an abnormal data segment. The normal monitoring data with a length of 150 of the 4 channels before the abnormal data segment is used as input, and 4 abnormal data reconstruction results with a length of 60 are obtained.
[0051] Step 4: Abnormal repair.
[0052] The first 60 data points of the two abnormal data segments are replaced by the reconstruction results of data with a length of 60.
[0053] Step 5: Loop repair.
[0054] Repeat steps 3 and 4 three more times to completely repair the two abnormal data segments. Figure 7 and Figure 8 shown.
[0055] See also Fig. 9 A computer device 400 provided in an embodiment of the present application includes: a processor 410 and a memory 420, wherein the memory 420 stores a computer program executable by the processor 410, and when the computer program is executed by the processor 410, the above method is executed.
[0056] The embodiment of the present application further provides a storage medium 430 on which a computer program is stored. When the computer program is run by the processor 410, the above method is executed.
[0057] Among them, the storage medium 430 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.
[0058] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.
[0059] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0060] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0061] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.
[0062] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.
[0063] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0064] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0065] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A method for repairing abnormal data based on self-attention, It is characterized in that The steps include: Build a multi-channel sequence data reconstruction model based on long-short term neural network and self-attention encoder-decoder; Obtaining input and output data sets of multi-channel normal monitoring data, and training the data reconstruction model; Reconstructing some abnormal segment monitoring data using the trained data reconstruction model and the previous normal monitoring data; The reconstructed monitoring data is used to replace the corresponding part of the abnormal monitoring data, and the reconstruction and replacement of the abnormal monitoring data are repeated until the section of the abnormal monitoring data is completely repaired.
2. The abnormal data repair method based on self-attention according to claim 1, It is characterized in that The multi-channel sequence data reconstruction model based on the long-term short-term neural network and the self-attention encoder-decoder includes: Acquire the type of monitoring data to be repaired, and determine the number of sensors of the same type as the characteristic number of input and output data of the sequence data reconstruction model; Through the memory units, hidden units and encoded outputs in the long short-term neural network, it serves as the information transmission channel between the encoder and the decoder; Extract the core information from the monitoring data through the self-attention mechanism.
3. The abnormal data repair method based on self-attention according to claim 1, It is characterized in that The method of obtaining the input and output data sets of multi-channel normal monitoring data includes: Obtain historical monitoring data and randomly extract multiple normal monitoring data segments; Create input and output data sequence pairs and divide them into training set, validation set and test set.
4. The abnormal data repair method based on self-attention according to claim 3, It is characterized in that When the input-output data sequence pair is produced, multiple normal monitoring data segments are produced using a time window of fixed length.
5. The abnormal data repair method based on self-attention according to claim 3, It is characterized in that The training of the data reconstruction model comprises the following steps: Training the data reconstruction model using the training set and the square error loss function; Use the validation set to check whether overfitting occurs during the training process; The test set is used to test the generalization performance of the trained data reconstruction model.
6. The abnormal data repair method based on self-attention according to claim 1, It is characterized in that The using the reconstructed normal monitoring data to replace the corresponding part of the abnormal monitoring data includes: The result output by the data reconstruction model includes data of multiple channels, and the reconstructed monitoring data is used to replace the abnormal monitoring data only in the channel where the abnormality occurs; If the length of the reconstructed monitoring data is greater than the length of the abnormal monitoring data, the reconstructed monitoring data is truncated to the length of the abnormal monitoring data; If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstructed monitoring data is used to replace the front part of the abnormal monitoring data.
7. The method for repairing abnormal data based on self-attention according to claim 6, It is characterized in that The reconstruction and replacement of the repeated abnormal monitoring data includes: If the length of the reconstructed monitoring data is less than the length of the abnormal monitoring data, the reconstruction and replacement process is repeated based on the partially repaired original monitoring data until all the abnormal monitoring data are completely repaired.
8. An abnormal data repair device based on self-attention, It is characterized in that Use of the method according to any one of claims 1 to 7, comprising: Modeling unit, used to build a multi-channel sequence data reconstruction model based on long-term and short-term neural networks and self-attention encoder-decoders; A training unit, used to obtain input and output data sets of multi-channel normal monitoring data and train the data reconstruction model; A reconstruction unit, used to reconstruct part of the abnormal segment monitoring data using the trained data reconstruction model and the previous normal monitoring data; The repair unit is used to replace the corresponding part of the abnormal monitoring data with the reconstructed monitoring data, and repeat the reconstruction and replacement of the abnormal monitoring data until the section of the abnormal monitoring data is completely repaired.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.
10. A storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Cited By
Pilot attention state data acquisition and processing method and device and related equipment
CN120616532A
