Noise-containing time series missing signal completion method and device based on adaptive feedback enhancement, equipment and storage medium
An adaptive feedback mechanism combining singular value decomposition and multi-scale feature fusion solves the problems of noise suppression and local detail preservation in noisy time-series signal completion, thereby improving the accuracy and stability of signal completion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI UNIV
- Filing Date
- 2026-04-28
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies have insufficient noise suppression capabilities, inadequate restoration of the main structure, and easy loss of local detail information during the completion of noisy time-series signals. They also have poor stability and cannot simultaneously achieve accuracy, stability, and robustness.
By separating effective signals from noise through singular value decomposition, constructing multi-scale features and extracting detailed residual features, and combining an adaptive feedback mechanism for dynamic adjustment, the completion process is optimized.
It improves the accuracy, stability and robustness of the time-series missing signal completion results, enhances noise resistance and local information preservation, and adapts to different noise intensities and disturbance conditions.
Smart Images
Figure CN122432508A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of time-series signal processing technology, and in particular to a method, apparatus, device and storage medium for completing noisy time-series missing signals based on adaptive feedback enhancement. Background Technology
[0002] With the continuous development of data processing and intelligent analysis technologies, time series signal completion for future timeframes has become a crucial research area in data analysis and intelligent processing. In practical applications, time series signals are often affected by random noise, local distortion, non-stationary disturbances, and multi-source interference during acquisition and transmission, resulting in varying degrees of noise contamination in historical observation signals. This weakens the effective structural and dynamic information in the original sequence. For target signals for future timeframes that have not yet been acquired, how to accurately complete them based on noisy historical time series signals has become a key issue affecting the effectiveness of subsequent analysis and applications.
[0003] In existing technologies, one type of method directly uses historical time-series signals to complete future unknown signals. This method is relatively straightforward, but due to the failure to effectively suppress noise disturbances in historical observation signals, it is prone to shifts or distortions in the temporal characteristics used during the completion process, thus reducing the accuracy and stability of the future signal completion results. Another type of method typically preprocesses historical time-series signals using methods such as filtering, denoising, low-rank decomposition, or signal reconstruction, and then completes the future unknown parts based on the processed signal. While this type of method can improve the quality of the input signal to some extent, it usually designs noise suppression, main information recovery, and future signal completion processes separately, lacking a dynamic optimization mechanism around the completion result. This can easily lead to problems such as good overall trend preservation but insufficient compensation for local details, or decreased overall stability after detail enhancement. Furthermore, some existing methods enhance the ability to express information at different temporal levels through multi-scale modeling. However, under noisy historical signal conditions, without an effective feedback enhancement mechanism, it is still difficult to balance the recovery of the main structure, preservation of local details, and optimization of the future signal completion result, resulting in shortcomings in the accuracy, stability, and robustness of the completion results. Therefore, how to provide a noisy time-series missing signal completion method based on adaptive feedback enhancement to improve the accuracy, stability and robustness of the time-series signal completion results at unknown future times has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0004] This application provides a method, apparatus, device, and storage medium for noisy time-series signal completion based on adaptive feedback enhancement. The method involves performing singular value decomposition on a historical noisy time-series signal to separate the effective signal component from the noise interference component, obtaining an initial reconstructed signal. Based on this, multi-scale features are constructed using the initial reconstructed signal, and detailed residual features are extracted by combining the difference information between the original time-series signal and the initial reconstructed signal. Then, an adaptive feedback enhancement mechanism is used to dynamically adjust and optimize the feature representation and completion process, thereby taking into account both the overall trend of the time-series signal and the expression of local detailed information. This addresses the problems of insufficient noise suppression, incomplete restoration of the main structure, easy loss of local detailed information, and poor result stability in existing noisy time-series signal completion processes, improving the accuracy, stability, and robustness of the time-series signal completion results.
[0005] In a first aspect, this application provides a method for completing noisy time-series missing signals based on adaptive feedback enhancement, the method comprising: The original time-series signal is acquired and preprocessed to construct a signal matrix to be processed; the original time-series signal is a sequence of physical quantity measurement values sampled over time by data acquisition devices, sensor nodes, or edge processing units. Singular value decomposition is performed on the signal matrix to be processed. The target retention order is determined according to the singular value energy distribution. The dominant effective component is reconstructed on the signal matrix to be processed to obtain the initial reconstructed signal. Multi-scale features are constructed based on the initial reconstructed signal, and detailed residual features are extracted based on the difference information between the original time series signal and the initial reconstructed signal. The multi-scale features and the detailed residual features are aligned and fused to obtain a unified complete feature representation; Based on the unified completion feature representation, the temporal missing signal completion result at the corresponding time is determined, and the dominant effective component reconstruction parameters and / or feature fusion parameters are adaptively updated according to the stability evaluation result of the temporal missing signal completion result.
[0006] In one possible design, the original time-series signal is acquired and preprocessed to construct the signal matrix to be processed, specifically including: The original time-series signal is preprocessed, and the preprocessing includes at least one of normalization, outlier suppression, dimension alignment, segmentation, block processing, and sliding window reconstruction. Use a length of Step size is The sliding window reconstructs the preprocessed time-series signal, stacking multiple local data segments column-wise to form a signal matrix to be processed. Represent the original time-series signal as a sequence. ,in, Represents the total length of the original timing signal; determines the first... The local data vector corresponding to each window is The signal matrix to be processed [ , , ,…, ],in The number of windows, and ; , and These are elements from the original sequence. , and These are the elements corresponding to the i-th sliding window. , , and These are the vectors obtained from the sliding window, respectively. This is the starting position index of the sliding window; Calculate the environmental complexity index based on the preprocessed signal data. The environmental complexity index It is determined by one or more of the following: local variance, overall fluctuation amplitude, noise estimate, difference between adjacent regions, and rate of change of structural continuity. It is used to characterize the noise intensity, local disturbance degree, temporal continuity change degree, or overall fluctuation level of the current signal.
[0007] In one possible design, singular value decomposition is performed on the signal matrix to be processed, the target retention order is determined based on the singular value energy distribution, and the dominant effective component reconstruction is performed on the signal matrix to be processed to obtain an initial reconstructed signal, specifically including: Signal matrix to be processed Perform singular value decomposition to obtain the left singular vector matrix. Singular value matrix and right singular vector matrix The singular value matrix The diagonal elements are singular values arranged in descending order. , The signal matrix to be processed The upper limit of rank or the upper limit of the number of singular values; Calculate the cumulative energy percentage corresponding to the first k singular values. The calculation formula is: in, For singular value index, The number of singular values participating in the cumulative energy calculation. The total number of singular values. For the first The energy term corresponding to each singular value; According to environmental complexity index Adaptive adjustment of energy threshold ,in Based on the basic energy threshold, This is the adjustment coefficient; Select the one that satisfies The smallest The value is used as the order of retention for the objective. ; From the singular value matrix Before the retention in China Identify the singular values and set the remaining singular values to zero to obtain the truncated singular value matrix. The initial reconstruction matrix is obtained by reconstructing the singular value matrix based on the truncated singular value matrix. ,in and They are respectively with the previous The left and right singular vector matrices corresponding to each singular value; When the signal matrix to be processed When constructing using a sliding window, the initial reconstruction matrix is calculated using either diagonal averaging or window overlap averaging. Convert to one-dimensional initial reconstruction signal .
[0008] In one possible design, multi-scale features are constructed based on the initial reconstructed signal, and detailed residual features are extracted based on the difference information between the original time-series signal and the initial reconstructed signal, specifically including: Based on the initial reconstruction signal Construct multi-scale features, including coarse-scale features. Mesoscale features and fine-scale features At least two of the following: wherein the coarse-scale feature is used to characterize the overall trend of signal change and main structural information, the meso-scale feature is used to characterize the transition relationship and correlation pattern between local time segments, and the fine-scale feature is used to characterize local details, weak changes and sensitive disturbance information; The original timing signal With the initial reconstruction signal Subtracting them yields the residual components. The residual components It reflects the local change information, edge detail information, and weak perturbation information that are not fully preserved in the reconstruction of the original signal by the dominant effective components; Extract detailed residual features based on the residual component R. .
[0009] In one possible design, the multi-scale features and the detailed residual features are aligned and fused to obtain a unified complete feature representation, specifically including: The coarse-scale features are obtained using the following formula. Mesoscale features Fine-scale features and detailed residual features Perform dimension mapping processing separately to obtain aligned feature representations located in the same feature space: , , , in, , , and coarse-scale features Mesoscale features Fine-scale features and detailed residual features Alignment feature representations located in the same feature space , , , These represent the mapping functions for the corresponding feature branches, which are implemented through linear projection, unified embedding, feature encoding, or channel mapping. Calculate the coarse-scale features respectively Mesoscale features Fine-scale features and detailed residual features Quality evaluation value of the corresponding feature branch , , and The quality evaluation value is determined based on at least one of the following: characteristic response intensity, cross-scale consistency, local information density, residual significance, and environmental complexity index. The coarse-scale features are determined based on the quality evaluation values of each feature branch. Mesoscale features Fine-scale features and detailed residual features The corresponding fusion weights , , and And satisfy ; The unified completion feature representation is calculated using the following formula. : When the environmental complexity index exceeds the first preset threshold, the fusion weight corresponding to the coarse-scale features is increased. Fusion weights of and / or mesoscale feature branches When the requirement for preserving local details exceeds the second preset threshold, increase the fusion weight corresponding to the fine-scale features. Fusion weights of and / or detail residual feature branches .
[0010] In one possible design, the temporal missing signal completion result at the corresponding time is determined based on the unified completion feature representation, and the dominant effective component reconstruction parameters and / or feature fusion parameters are adaptively updated based on the stability evaluation result of the temporal missing signal completion result, specifically including: Unify the feature representation The input completion module outputs the completion result of the missing temporal signal at the corresponding time. The completion module is a linear completion model, a recursive network model, a convolutional network model, or a temporal completion model based on an attention mechanism. Constructing stability evaluation indicators The stability evaluation index Determined by at least one of the following evaluation metrics: Completion error is used to characterize the degree of deviation between the completion result and the actual observation result; Cross-scale consistency is used to characterize the degree of coordination between features at different scales; Output fluctuation level, used to characterize the stability of successive batch completion results; Residual sparsity is used to characterize whether detailed residual features are effectively concentrated in local key information regions; Feature distribution stability is used to characterize the degree of distribution change of the unified completion feature representation during continuous processing; When stability evaluation index If the preset conditions are not met, perform at least one of the following parameter update operations: Adjust the reconstructing parameters of the dominant active ingredient, including adjusting the target retention order. Energy thresholds for singular values ; Adjust the fusion parameters, including adjusting the fusion weights corresponding to the multi-scale feature branches; Adjust the detailed compensation parameters, including adjusting the compensation intensity of the detailed residual branches; Adjust the scale division parameters, including adjusting the multi-scale division granularity or sliding window parameters; The parameter update operation is performed once after each round of processing, or uniformly after a predetermined number of batches of processing are completed.
[0011] In one possible design, the method further includes: When stability evaluation index When the completion error exceeds the preset error threshold, the target retention order is increased. When the output fluctuation exceeds the preset fluctuation threshold, increase the fusion weight of the coarse-scale feature and / or meso-scale feature branches; When the residual sparsity is lower than the preset sparsity threshold or the detailed information is insufficient, increase the fusion weight of the detailed residual feature branch and / or the fusion weight corresponding to the fine-scale feature.
[0012] Secondly, this application provides a noisy timing-delay signal completion device based on adaptive feedback enhancement, the device comprising: The data preprocessing module is configured to acquire the raw time-series signal and perform preprocessing to construct a signal matrix to be processed; the raw time-series signal is a sequence of physical quantity measurement values sampled over time by a data acquisition device, sensor node, or edge processing unit. The singular value decomposition and reconstruction module is configured to perform singular value decomposition on the signal matrix to be processed, determine the target retention order based on the singular value energy distribution, and reconstruct the dominant effective components of the signal matrix to be processed to obtain the initial reconstructed signal. The feature construction module is configured to construct multi-scale features based on the initial reconstructed signal and extract detailed residual features based on the difference information between the original time series signal and the initial reconstructed signal. The feature alignment and fusion module is configured to align and fuse the multi-scale features and the detail residual features to obtain a unified complete feature representation. The completion feedback module is configured to determine the completion result of the temporal missing signal at the corresponding time based on the unified completion feature representation, and to adaptively update the reconstruction parameters of the dominant effective component and / or the feature fusion parameters according to the stability evaluation result of the temporal missing signal completion result.
[0013] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the noisy timing missing signal completion method based on adaptive feedback enhancement as described in the first aspect and various possible designs of the first aspect.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the method for completing noisy timing-missing signals based on adaptive feedback enhancement as described in the first aspect and various possible designs of the first aspect.
[0015] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method for completing noisy timing-delayed signals based on adaptive feedback enhancement as described in the first aspect and various possible designs of the first aspect.
[0016] The method, apparatus, device, and storage medium for completing noisy time-series missing signals based on adaptive feedback enhancement provided in this application have at least the following beneficial effects: 1) This application performs singular value decomposition on the signal matrix to be processed, adaptively determines the target retention order based on the energy accumulation ratio of singular values and combined with the environmental complexity index, and then reconstructs the dominant effective components of the original time series signal. This can effectively separate the main structural components and noise disturbance components in the signal, reduce the impact of noise on the subsequent feature modeling and completion process, and provide a stable and effective input basis for the completion of noisy time series missing signals.
[0017] 2) This application extracts detail residual features by utilizing the difference information between the original time series signal and the initial reconstructed signal, and co-models the detail residual features with the main structural features. This can enhance the noise resistance while retaining local change information and weak detail information, avoiding the problem of easy loss of detail information in the prior art, thereby improving the accuracy of the time series missing signal completion result.
[0018] 3) This application constructs multi-scale features including coarse, medium and fine scales, and performs scale alignment and weighted fusion of multi-scale features and detailed residual features. This can simultaneously represent overall trend information, local transition information and detailed change information, reduce the problem of insufficient single-scale representation and scale mismatch caused by direct splicing of multi-scale features, thereby improving the integrity and stability of unified completion feature representation.
[0019] 4) This application introduces the environmental complexity index into the determination process of the dominant effective component reconstruction parameters and the allocation process of multi-scale feature fusion weights, so that the target retention order, singular value energy threshold and fusion weights of each feature branch can be dynamically adjusted with the environmental state, thereby improving the adaptability of the method to different noise intensities and different disturbance conditions.
[0020] 5) This application evaluates the stability of the completion results and their corresponding feature representations, and updates the decomposition and reconstruction parameters and feature fusion parameters based on the evaluation results to form a closed-loop adaptive optimization mechanism, which can effectively improve the stability, reliability and robustness of the completion results of noisy time-series missing signals. Attached Figure Description
[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0022] Figure 1 A flowchart illustrating a method for completing noisy temporal missing signals based on adaptive feedback enhancement, provided in an embodiment of this application; Figure 2 A schematic diagram illustrating the reconstruction of the dominant effective component based on singular value decomposition and environmental complexity adjustment, provided for embodiments of this application; Figure 3 This is a schematic diagram illustrating the multi-scale feature construction and detail residual compensation fusion provided in an embodiment of this application. Figure 4 This is a schematic diagram of the parameter feedback and update closed loop provided in the embodiments of this application; Figure 5 This is a structural diagram of a noisy timing missing signal completion device based on adaptive feedback enhancement provided in an embodiment of this application.
[0023] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation
[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0025] The collection, storage, use, processing, transmission, provision, and disclosure of relevant data and information in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0026] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0027] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0028] This application proposes a method for completing noisy temporal missing signals based on adaptive feedback enhancement, such as... Figure 1 As shown, the method for completing noisy time-series missing signals based on adaptive feedback enhancement can be implemented through the following steps S10-S50.
[0029] S10: Acquire the original time series signal and perform preprocessing to construct the signal matrix to be processed; the original time series signal is a sequence of physical quantity measurement values sampled over time by data acquisition equipment, sensor nodes or edge processing units.
[0030] It should be noted that the original time-series signal is a sequence of physical quantity measurements sampled over time, acquired by data acquisition equipment, sensor nodes, or edge processing units. Specifically, it can be a sequence of physical quantity measurements sampled over time, such as vibration signals in equipment operation status monitoring, temperature or pressure signals in industrial process control, current or voltage signals in power systems, sound waves or wind speed signals in environmental monitoring, electrocardiogram or electroencephalogram signals in the biomedical field, and echo signals in communication or radar systems. For example, in the scenario of fault diagnosis of industrial rotating machinery, the method of this application can be applied to complete missing data of noisy vibration signals acquired by accelerometers installed on bearings or gearboxes, to recover signal interruptions caused by occasional sensor failures or packet loss, thereby providing complete time-series data for subsequent fault feature extraction and lifespan prediction. As another example, in the scenario of environmental monitoring in wireless sensor networks, the method of this application can be applied to complete noisy and incomplete time-series data generated by temperature or humidity sensor nodes due to battery power fluctuations, channel interference, etc., thereby supporting regional climate change trend analysis and early warning of abnormal events.
[0031] In some implementations, step S10 may be specifically implemented through steps S101-S103.
[0032] S101: Obtain the original timing signal.
[0033] The original timing signal described in this embodiment may originate from data acquisition equipment, data receiving terminal, sensor node, edge processing unit or server cache unit, etc., and the specific source is not limited.
[0034] S102: Preprocess the original time-series signal to reduce the impact of abnormal noise, invalid redundancy, and data representation differences on subsequent signal decomposition, feature modeling, and completion processing. The preprocessing includes at least one or more of the following operations: 1) Normalize the original time series signal to reduce the impact of differences in amplitude range between different samples.
[0035] 2) Suppress or remove outliers to reduce the impact of impulse interference or mutation values on subsequent decomposition results.
[0036] 3) Align the data dimensions to ensure that signals from different batches or sources have a unified representation.
[0037] 4) The original time series signal is segmented, divided into blocks, or processed by sliding windows to construct a signal matrix suitable for decomposition analysis and completion modeling.
[0038] S103: If the original time-series signal is a one-dimensional time-series sequence, a length of [length missing] can be used. Step size is The data is reconstructed using a sliding window approach, and multiple local data fragments are stacked column-wise to form the signal matrix to be processed. Let the original time-series signal be represented as a sequence. ,in, This represents the total length of the original timing signal; then the first... The local data vector corresponding to each window is The signal matrix to be processed [ , , ,…, ],in The number of windows, and ; , and These are elements from the original sequence. , and These are the elements corresponding to the i-th sliding window. , , and These are the vectors obtained from the sliding window, respectively. This is the starting position index of the sliding window. If the original data is a two-dimensional matrix, the signal matrix to be processed can be constructed using a block sampling method; if the original data is multi-dimensional, it can be expanded first, and then the signal matrix to be processed can be constructed according to preset rules.
[0039] Through the above steps S101-S103, the original time series signal can be converted into a unified input form suitable for singular value decomposition, multi-scale modeling, and feedback regulation.
[0040] In some implementations, environmental complexity indices can also be calculated based on the preprocessed signal data. This is used to characterize the degree of complexity of the interference environment in which the current signal is located. The environmental complexity index... It can be determined by one or more of the following: local variance, overall fluctuation amplitude, noise estimate, difference between adjacent regions, and rate of change of structural continuity.
[0041] S20: Perform singular value decomposition on the signal matrix to be processed, determine the target retention order based on the singular value energy distribution, and reconstruct the dominant effective components of the signal matrix to be processed to obtain the initial reconstructed signal.
[0042] In some implementation methods, please refer to Figure 2 As shown, step S20 performs singular value decomposition on the signal matrix to be processed and reconstructs the dominant effective components to obtain the initial reconstructed signal. The specific process includes the following steps S201-S205.
[0043] S201: Process the signal matrix obtained in step S10 Perform singular value decomposition to obtain the left singular vector matrix. Singular value matrix and right singular vector matrix ,Right now Among them, the singular value matrix The diagonal elements are singular values arranged in descending order. , The signal matrix to be processed The upper limit of rank or the upper limit of the number of singular values.
[0044] In order to extract the dominant effective component from the original time-series signal and suppress complex noise interference, the target preservation order is determined through the following steps S202 and S203. .
[0045] S202: Calculate the cumulative energy percentage corresponding to the first k singular values. The calculation formula is: in, For singular value index, The number of singular values participating in the cumulative energy calculation. The total number of singular values. For the first The energy term corresponding to each singular value.
[0046] S203: Determine the initial retention order based on the relationship between the cumulative energy percentage and the preset energy threshold.
[0047] In some implementations, the preset energy threshold is based on an environmental complexity index. Adaptive adjustment. This can be achieved in the following ways: ,in, Based on the basic energy threshold, For adjustment coefficients, As an indicator of environmental complexity, and After normalization, the value is in the interval [0,1].
[0048] Choose the smallest that satisfies the following formula The order of retention as the objective : .
[0049] S204: Obtaining the target retention order Then, from the singular value matrix Before the retention in China The singular values are 1, and the remaining singular values are set to zero to obtain the truncated singular value matrix. Based on the truncated singular value matrix, the dominant effective component is reconstructed from the signal matrix to be processed to obtain the initial reconstructed matrix. : .in, and They respectively represent the previous The singular vector matrix corresponding to each singular value.
[0050] S205: When the signal matrix to be processed When constructed using a sliding window, the initial reconstruction matrix can be further refined. Perform inverse mapping recovery, such as by diagonal averaging, window overlap averaging, or block recovery, to convert it into the initial reconstruction signal. .
[0051] Step S20 separates and characterizes the dominant effective components and noise disturbance components in the original time series signal, thereby extracting relatively stable main structure information and reducing the impact of random noise, local interference and unstable disturbances on the subsequent feature modeling and completion process.
[0052] S30: Construct multi-scale features based on the initial reconstructed signal, and extract detailed residual features based on the difference information between the original time series signal and the initial reconstructed signal.
[0053] In some implementations, such as Figure 3 As shown, step S30 can be implemented through the following steps S301 and S302.
[0054] S301: Obtaining the initial reconstruction signal Subsequently, multi-scale features are constructed based on the initial reconstructed signal to characterize the main structural information and change features of the signal at different observation scales.
[0055] In this embodiment, at least two different scales can be set, such as coarse scale, medium scale, and fine scale. For different scales, different window lengths, block sizes, sampling granularities, or local region partitioning strategies can be used to extract candidate features at the corresponding scale from the initial reconstructed signal. Wherein: Coarse-scale features are used to characterize the overall trend of signal changes, global profile, or long-term change information. Mesoscale features are used to characterize the transitional relationships between local regions and changes in the mid-level structure; Fine-scale features are used to characterize local details, subtle changes, or short-term dynamic features.
[0056] In one specific implementation, the initial reconstructed signal can be divided into blocks or reassembled according to different scales, and then input into a feature extraction network for encoding representation to construct feature branches at the corresponding scales, thereby obtaining coarse-scale features. Mesoscale features and fine-scale features .
[0057] S302: Based on the original timing signal With the initial reconstruction signal Extracting detailed residual features from the differences between them .
[0058] To compensate for the potential weakening of local detail information during the reconstruction of the dominant effective component, further steps are performed in step S302 based on the original time-series signal. With the initial reconstruction signal Extracting detailed residual features from the differences between them .
[0059] Specifically, the original timing signal With the initial reconstruction signal Subtracting them yields the residual components. The residual components This primarily reflects local variation information, edge detail information, and weak perturbation information that were not fully preserved in the reconstruction of the original signal from the dominant effective components. Based on these residual components... Further extraction yields detailed residual features. .
[0060] In one specific implementation, the multi-scale features and detailed residual features are constructed using a dual-branch approach, where the main structure branch processes the initial reconstructed signal, and the detailed compensation branch processes the residual components. This approach maintains a stable representation of the main structural information while compensating for potential loss of local detail information and subtle changes during the reconstruction process.
[0061] S40: Align and fuse multi-scale features and detail residual features to obtain a unified complete feature representation.
[0062] In some implementations, such as Figure 3 As shown, step S40 can be implemented through the following steps S401-S403.
[0063] Since features extracted at different scales and detailed residual features differ in dimensionality, scale space, and expression intensity, they need to be uniformly mapped and scale-aligned. Therefore, step S401 is executed: for coarse-scale features... Mesoscale features Fine-scale features and detailed residual features Perform dimension mapping processing separately to obtain aligned feature representations located in the same feature space. , , , .in, , , , These represent the mapping functions for the corresponding branches. The mapping processing performed by these mapping functions can be achieved through linear projection, unified embedding, feature encoding, channel mapping, or a combination thereof.
[0064] S402: After scale alignment is completed, multiple feature branches are fused.
[0065] In one implementation, each coarse-scale feature is calculated separately. Mesoscale features Fine-scale features and detailed residual features Corresponding quality evaluation value , , and The quality evaluation value can be determined based on at least one of the following: feature response intensity, cross-scale consistency, local information density, residual significance, and environmental complexity index. Coarse-scale features are determined based on the quality evaluation values of each feature branch. Mesoscale features Fine-scale features and detailed residual features The corresponding fusion weights , , and And satisfy .
[0066] S403: Calculate the unified completion feature representation The calculation formula can be expressed as: .
[0067] In a preferred embodiment, when the environmental complexity is high, the fusion weight of the coarse-scale feature and main structure feature branches is appropriately increased to enhance the overall noise resistance; when the detail changes are significant or the local preservation requirement is high, the fusion weight of the fine-scale feature and detail residual feature branches is appropriately increased to enhance the local detail expression capability.
[0068] In another embodiment, the fusion process may also include cross-scale consistency constraints, that is, before fusion, the similarity or correlation between features at different scales is evaluated, and noisy feature branches with poor consistency are suppressed, so as to reduce the propagation of invalid information in the fusion process.
[0069] Step S40 can unify candidate features from different scales and sources into a complete feature representation with higher stability and integrity.
[0070] S50: Determine the time-series missing signal completion result based on the unified completion feature representation, and adaptively update the dominant effective component reconstruction parameters and / or feature fusion parameters according to the stability evaluation results of the time-series missing signal completion result.
[0071] In some implementation methods, please refer to Figure 4 As shown, step S50, which outputs the time-series missing signal completion result based on the unified completion feature representation and performs adaptive feedback update based on stability evaluation, is specifically implemented through the following steps S501-S503.
[0072] S501: Convert the unified completion feature representation obtained in step S40 into... The input completion module outputs the completion result of the temporal missing signal at the corresponding time. The completion module is a temporal completion model used to generate the completion result of the target time based on a unified completion feature representation. Specifically, it can be a linear completion model, a recursive network model, a convolutional network model, an attention-based completion model, or other temporal modeling models, and the specific form is not limited.
[0073] S502: To further improve the adaptability of the method in this application under complex interference environments, a stability evaluation is performed on the time-series missing signal completion results and their corresponding completion feature representations, and the dominant effective component reconstruction parameters and / or multi-scale fusion parameters are updated based on the evaluation results. Specifically, a stability evaluation index can be constructed. The stability evaluation index It can be determined by at least one of the following evaluation metrics: 1) Completion error, used to characterize the degree of deviation between the completion result and the actual observation result; 2) Cross-scale consistency, used to characterize the degree of coordination between features at different scales; 3) Output fluctuation level, used to characterize the stability of successive batch completion results; 4) Residual sparsity, used to characterize whether detailed residual features are effectively concentrated in local key information regions; 5) Feature distribution stability, used to characterize the degree of distribution change of the unified completion feature representation during continuous processing.
[0074] S503: When the stability evaluation index fails to meet the preset conditions, parameter updates can be performed. The parameter updates include at least one of the following: 1) Adjust the reconstruction parameters of the dominant effective components, including adjusting the target retention order and the singular value energy threshold; 2) Adjust the fusion parameters, including adjusting the fusion weights corresponding to the multi-scale feature branches; 3) Adjust the detailed compensation parameters, including adjusting the compensation intensity of the detailed residual branches; 4) Adjust the scale division parameters, including adjusting the multi-scale division granularity or window parameters.
[0075] For example, when the completion error is large, the target retention order can be appropriately increased; when the output results fluctuate greatly, the fusion ratio of the main structure branches can be increased; when the detailed information is insufficient, the weight of the detailed residual branches can be increased.
[0076] In some implementations, the feedback update in step S50 can be performed once after each round of processing, or it can be performed uniformly after a predetermined number of batches of processing are completed, in order to adapt to the computing resource requirements and real-time requirements of different application scenarios.
[0077] Through step S50, while outputting the result of completing the missing timing signal, the key parameters in the aforementioned processing process can be adjusted in a closed loop, thereby further enhancing the stability and robustness of the present invention in continuous processing scenarios and dynamic environments.
[0078] This application also provides a noisy timing-delay signal completion device based on adaptive feedback enhancement, used to implement the method described in any of the above embodiments, such as... Figure 5 As shown, the noisy timing missing signal completion device based on adaptive feedback enhancement includes: The data preprocessing module 501 is configured to acquire the raw time-series signal and perform preprocessing to construct a signal matrix to be processed; the raw time-series signal is a sequence of physical quantity measurement values sampled over time by a data acquisition device, sensor node, or edge processing unit. The singular value decomposition reconstruction module 502 is configured to perform singular value decomposition on the signal matrix to be processed, determine the target retention order based on the singular value energy distribution, and reconstruct the dominant effective components of the signal matrix to be processed to obtain an initial reconstructed signal. The feature construction module 503 is configured to construct multi-scale features based on the initial reconstructed signal and extract detailed residual features based on the difference information between the original time series signal and the initial reconstructed signal. The feature alignment and fusion module 504 is configured to align and fuse the multi-scale features and the detail residual features to obtain a unified complete feature representation. The completion feedback module 505 is configured to determine the completion result of the temporal missing signal at the corresponding time based on the unified completion feature representation, and to adaptively update the dominant effective component reconstruction parameters and / or feature fusion parameters according to the stability evaluation result of the temporal missing signal completion result.
[0079] This application provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus.
[0080] The processor executes computer execution instructions stored in memory, causing the processor to perform the scheme in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0081] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.
[0082] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.
[0083] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the above-described embodiment of the noisy timing missing signal completion method based on adaptive feedback enhancement.
[0084] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the noisy timing missing signal completion method based on adaptive feedback enhancement in the above embodiments.
[0085] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.
[0086] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.
[0087] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.
[0088] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.
[0089] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.
[0090] The memory may include high-speed RAM, and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.
[0091] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Architecture (EISA) buses, etc. Buses can be categorized into address buses, data buses, control buses, etc.
[0092] The aforementioned storage medium 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 read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.
[0093] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.
[0094] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method for completing noisy temporally missing signals based on adaptive feedback enhancement, characterized in that, The method includes: The original time-series signal is acquired and preprocessed to construct a signal matrix to be processed; the original time-series signal is a sequence of physical quantity measurement values sampled over time by data acquisition devices, sensor nodes, or edge processing units. Singular value decomposition is performed on the signal matrix to be processed. The target retention order is determined according to the singular value energy distribution. The dominant effective component is reconstructed on the signal matrix to be processed to obtain the initial reconstructed signal. Multi-scale features are constructed based on the initial reconstructed signal, and detailed residual features are extracted based on the difference information between the original time series signal and the initial reconstructed signal. The multi-scale features and the detailed residual features are aligned and fused to obtain a unified complete feature representation; Based on the unified completion feature representation, the temporal missing signal completion result at the corresponding time is determined, and the dominant effective component reconstruction parameters and / or feature fusion parameters are adaptively updated according to the stability evaluation result of the temporal missing signal completion result.
2. The method according to claim 1, characterized in that, The original time-series signal is acquired and preprocessed to construct the signal matrix to be processed, specifically including: The original time-series signal is preprocessed, and the preprocessing includes at least one of normalization, outlier suppression, dimension alignment, segmentation, block processing, and sliding window reconstruction. Use a length of Step size is The sliding window reconstructs the preprocessed time-series signal, stacking multiple local data segments column-wise to form a signal matrix to be processed. Represent the original time-series signal as a sequence. ,in, Represents the total length of the original timing signal; determines the first... The local data vector corresponding to each window is The signal matrix to be processed [ , , ,…, ],in The number of windows, and ; , and These are elements from the original sequence. , and These are elements in the original sequence corresponding to the i-th sliding window. , , and These are the vectors obtained from the sliding window, respectively. This is the starting position index of the sliding window; Calculate the environmental complexity index based on the preprocessed signal data. The environmental complexity index It is determined by one or more of the following: local variance, overall fluctuation amplitude, noise estimate, difference between adjacent regions, and rate of change of structural continuity. It is used to characterize the noise intensity, local disturbance degree, temporal continuity change degree, or overall fluctuation level of the current signal.
3. The method according to claim 1, characterized in that, Singular value decomposition is performed on the signal matrix to be processed. The target retention order is determined based on the singular value energy distribution. The dominant effective component is then reconstructed from the signal matrix to be processed to obtain the initial reconstructed signal, specifically including: Signal matrix to be processed Perform singular value decomposition to obtain the left singular vector matrix. Singular value matrix and right singular vector matrix The singular value matrix The diagonal elements are singular values arranged in descending order. , The signal matrix to be processed The upper limit of rank or the upper limit of the number of singular values; Calculate the cumulative energy percentage corresponding to the first k singular values. The calculation formula is: in, For singular value index, The number of singular values participating in the cumulative energy calculation. The total number of singular values. For the first The energy term corresponding to each singular value; According to environmental complexity index Adaptive adjustment of energy threshold ,in Based on the basic energy threshold, This is the adjustment coefficient; Select the one that satisfies The smallest The value is used as the order of retention for the objective. ; From the singular value matrix Before the retention in China Identify the singular values and set the remaining singular values to zero to obtain the truncated singular value matrix. The initial reconstruction matrix is obtained by reconstructing the singular value matrix based on the truncated singular value matrix. ,in and They are respectively with the previous The left and right singular vector matrices corresponding to each singular value; When the signal matrix to be processed When constructing using a sliding window, the initial reconstruction matrix is calculated using either diagonal averaging or window overlap averaging. Convert to one-dimensional initial reconstruction signal .
4. The method according to claim 1, characterized in that, Multi-scale features are constructed based on the initial reconstructed signal, and detailed residual features are extracted based on the difference information between the original time-series signal and the initial reconstructed signal, specifically including: Based on the initial reconstruction signal Construct multi-scale features, including coarse-scale features. Mesoscale features and fine-scale features At least two of the following: wherein the coarse-scale feature is used to characterize the overall trend of signal change and main structural information, the meso-scale feature is used to characterize the transition relationship and correlation pattern between local time segments, and the fine-scale feature is used to characterize local details, weak changes and sensitive disturbance information; The original timing signal With the initial reconstruction signal Subtracting them yields the residual components. The residual components It reflects the local change information, edge detail information, and weak perturbation information that are not fully preserved in the reconstruction of the original signal by the dominant effective components; Extract detailed residual features based on the residual component R. .
5. The method according to claim 4, characterized in that, Aligning and fusing the multi-scale features and the detailed residual features to obtain a unified complete feature representation, specifically including: The coarse-scale features are obtained using the following formula. Mesoscale features Fine-scale features and detailed residual features Perform dimension mapping processing separately to obtain aligned feature representations located in the same feature space: ， ， ， in, , , and coarse-scale features Mesoscale features Fine-scale features and detailed residual features Alignment feature representations located in the same feature space , , , These represent the mapping functions for the corresponding feature branches, which are implemented through linear projection, unified embedding, feature encoding, or channel mapping. Calculate the coarse-scale features respectively Mesoscale features Fine-scale features and detailed residual features Quality evaluation value of the corresponding feature branch , , and The quality evaluation value is determined based on at least one of the following: characteristic response intensity, cross-scale consistency, local information density, residual significance, and environmental complexity index. The coarse-scale features are determined based on the quality evaluation values of each feature branch. Mesoscale features Fine-scale features and detailed residual features The corresponding fusion weights , , and And satisfy ; The unified completion feature representation is calculated using the following formula. : When the environmental complexity index exceeds the first preset threshold, the fusion weight corresponding to the coarse-scale features is increased. Fusion weights of and / or mesoscale feature branches When the requirement for preserving local details exceeds the second preset threshold, increase the fusion weight corresponding to the fine-scale features. Fusion weights of and / or detail residual feature branches .
6. The method according to claim 1, characterized in that, Based on the unified completion feature representation, the temporal missing signal completion result at the corresponding time is determined, and based on the stability evaluation result of the temporal missing signal completion result, the dominant effective component reconstruction parameters and / or feature fusion parameters are adaptively updated, specifically including: Unify the feature representation The input completion module outputs the completion result of the missing temporal signal at the corresponding time. The completion module is a linear completion model, a recursive network model, a convolutional network model, or a temporal completion model based on an attention mechanism. Constructing stability evaluation indicators The stability evaluation index Determined by at least one of the following evaluation metrics: Completion error is used to characterize the degree of deviation between the completion result and the actual observation result; Cross-scale consistency is used to characterize the degree of coordination between features at different scales; Output fluctuation level, used to characterize the stability of successive batch completion results; Residual sparsity is used to characterize whether detailed residual features are effectively concentrated in local key information regions; Feature distribution stability is used to characterize the degree of distribution change of the unified completion feature representation during continuous processing; When stability evaluation index If the preset conditions are not met, perform at least one of the following parameter update operations: Adjust the reconstructing parameters of the dominant active ingredient, including adjusting the target retention order. Energy thresholds for singular values ; Adjust the fusion parameters, including adjusting the fusion weights corresponding to the multi-scale feature branches; Adjust the detailed compensation parameters, including adjusting the compensation intensity of the detailed residual branches; Adjust the scale division parameters, including adjusting the multi-scale division granularity or sliding window parameters; The parameter update operation is performed once after each round of processing, or uniformly after a predetermined number of batches of processing are completed.
7. The method according to claim 6, characterized in that, The method further includes: When stability evaluation index When the completion error exceeds the preset error threshold, the target retention order is increased. When the output fluctuation exceeds the preset fluctuation threshold, increase the fusion weight of the coarse-scale feature and / or meso-scale feature branches; When the residual sparsity is lower than the preset sparsity threshold or the detailed information is insufficient, increase the fusion weight of the detailed residual feature branch and / or the fusion weight corresponding to the fine-scale feature.
8. A device for completing noisy, time-series missing signals based on adaptive feedback enhancement, characterized in that, The device includes: The data preprocessing module is configured to acquire the raw time-series signal and perform preprocessing to construct a signal matrix to be processed; the raw time-series signal is a sequence of physical quantity measurement values sampled over time by a data acquisition device, sensor node, or edge processing unit. The singular value decomposition and reconstruction module is configured to perform singular value decomposition on the signal matrix to be processed, determine the target retention order based on the singular value energy distribution, and reconstruct the dominant effective components of the signal matrix to be processed to obtain the initial reconstructed signal. The feature construction module is configured to construct multi-scale features based on the initial reconstructed signal and extract detailed residual features based on the difference information between the original time series signal and the initial reconstructed signal. The feature alignment and fusion module is configured to align and fuse the multi-scale features and the detail residual features to obtain a unified complete feature representation. The completion feedback module is configured to determine the completion result of the temporal missing signal at the corresponding time based on the unified completion feature representation, and to adaptively update the reconstruction parameters of the dominant effective component and / or the feature fusion parameters according to the stability evaluation result of the temporal missing signal completion result.
9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the noisy timing missing signal completion method based on adaptive feedback enhancement as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the noisy timing missing signal completion method based on adaptive feedback enhancement as described in any one of claims 1-7.