Edge computing data co-processing device

Abstract

The invention relates to the field of edge computing data co-processing, and discloses an edge computing data co-processing device, which comprises the following steps: collecting heterogeneous sensing data from a plurality of edge nodes, and constructing a correlation representation relationship; acquiring local clock drift, communication queuing delay, sampling jitter and load fluctuation states of each edge node, mapping each node state to a corresponding object and a corresponding signal in combination with the association representation relationship, and generating corresponding offset parameter representation; performing non-rigid time alignment and local feature reconstruction on the multi-source data corresponding to the same object, and generating a unified joint expression fragment on the premise of keeping the continuity of an original signal; performing cross-source consistency constraint analysis on the joint expression fragment, predicting potential degradation or failure risk, and generating a collaborative evaluation result; and performing joint adjustment on the participation weight, the calculation priority, the data forwarding path and the local parameter of the edge node to form a cooperative processing self-evolution mechanism. The method has the advantage of improving credibility and stability.

Description

Technical Field

[0001] This invention relates to the field of edge computing data collaborative processing, specifically to an edge computing data collaborative processing device. Background Technology

[0002] With the development of the Industrial Internet and the Internet of Things, in scenarios such as distributed production line monitoring, smart substation operation and maintenance, and urban infrastructure sensing, multiple edge nodes are usually deployed on-site to process data from different sensors locally to reduce the load on the central system and reduce response latency. However, most existing systems adopt the approach of "each node processes independently and summarizes periodically". When multiple nodes sense the same object at the same time, due to inconsistencies in sampling frequency, time base and local processing strategy, the data often lacks strict alignment and semantic correspondence. Under network jitter, clock drift or local load changes, this offset will be amplified in multi-source fusion or collaborative control, leading to judgment contradictions or control conflicts, reducing the overall reliability and stability of the system. Therefore, existing technologies lack a unified data alignment and collaborative processing mechanism for the same object, making it difficult to meet the application requirements with high real-time and consistency requirements. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an edge computing data collaborative processing device that has the advantages of improved reliability and stability, and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goals of improving reliability and stability, this invention provides the following technical solution: an edge computing data collaborative processing device, comprising: Data acquisition module: Collects heterogeneous sensor data from multiple edge nodes, and performs object-level merging based on the physical relationship, spatial topology relationship and signal coupling characteristics of the monitored objects to construct an association representation relationship that represents the mapping relationship between objects, signals and nodes; State mapping module: acquires the local clock drift, communication queuing delay, sampling jitter and load fluctuation status of each edge node, maps the state of each node to the corresponding object and signal by combining the correlation representation relationship, and jointly estimates the relative offset of multi-source data in the time dimension and feature dimension to generate the corresponding offset parameter representation. Feature reconstruction module: Based on the corresponding offset parameters, non-rigid time alignment and local feature reconstruction are performed on the multi-source data corresponding to the same object to generate a unified joint expression fragment while maintaining the continuity of the original signal; Consistency Analysis Module: Performs cross-source consistency constraint analysis on joint expression fragments, combines consistency indices and offset parameters to characterize the time change trend, predicts potential degradation or failure risks, and generates a collaborative assessment result that includes the current deviation status and future risk level; The strategy evolution module: Based on the collaborative evaluation results, it jointly adjusts the participation weight, computation priority, data forwarding path and local parameters of edge nodes, and updates the offset parameter representation, consistency constraint parameters and strategy adjustment rules in combination with the operation feedback, forming a collaborative processing self-evolution mechanism driven by consistency and prediction.

[0005] Preferably, the process of constructing the association representation relationship between the representation object, signal, and node is as follows: Data streams are collected by sensors deployed at each edge node and aggregated according to the physical connections and topological adjacency relationships between objects; The signals collected by each node are classified and mapped using the signal coupling index, generating a mapping matrix of objects, signals and nodes; By establishing a preliminary object-level data index through data stream identification and timestamp management, an associated representation relationship is formed that represents the mapping relationship between objects, signals, and nodes.

[0006] Preferably, the process of mapping the state of each node to the corresponding object and signal by combining the association representation relationship is as follows: In the data acquisition module of each edge node, the local system time offset, task queue queuing delay, sampling interval jitter and current processing load index of the node are read in real time to form a multi-dimensional node status parameter set. Map the set of node state parameters to the associated representation relationship to generate a multi-dimensional data input table corresponding to objects, nodes and parameters; The parameter differences in the multidimensional data input table are normalized and standardized to form a unified parameter set.

[0007] Preferably, the process of generating the corresponding offset parameter representation is as follows: Based on the parameter set, the multidimensional data input table corresponding to the object, node and parameter is used as the basis for joint analysis; The offset of each object in the time dimension and feature dimension of the multi-source data at different nodes is measured and calculated, and iterative adjustments are made. The offset information obtained from the joint estimation is integrated to form the corresponding offset parameter representation.

[0008] Preferably, the process of performing non-rigid time alignment and local feature reconstruction on multi-source data corresponding to the same object is as follows: Based on the offset parameter characterization, the offset of multi-source data collected by each object at different edge nodes in the time dimension and feature dimension is corrected point by point; During the alignment process, missing sampling points within the data segment are interpolated, abnormal abrupt changes are identified and corrected, and the signal is smoothed by combining local statistical features. The aligned and reconstructed data segments are subjected to segmented feature extraction and standardization, including amplitude normalization, frequency domain feature mapping and local temporal pattern coding, and each segment is uniformly mapped to an object-level joint representation structure.

[0009] Preferably, the process of generating a unified joint expression fragment while maintaining the continuity of the original signal is as follows: Based on the data segments after non-rigid time alignment and local feature reconstruction, the multi-source data of each object is subjected to continuity verification and filtering to correct local signal jumps and noise interference. The processed data is spliced ​​and fused in the time and feature dimensions, integrating the local feature vectors and time sequence patterns of each node, and mapped to the object-level joint representation structure; During the mapping and fusion process, the weights are dynamically adjusted based on the offset parameter. Weighted mapping and local smoothing are applied to high offset or local abnormal regions to generate a unified object-level joint expression fragment.

[0010] Preferably, the process of performing cross-source consistency constraint analysis on jointly expressed fragments is as follows: Based on object-level joint expression fragments, the time series and feature vectors of multi-source data are mapped to the cross-source consistency analysis structure; By comparing the joint expression fragments with historical steady-state models, physical consistency models, and prior behavioral templates, deviation patterns and potential anomalous regions are identified. Calculate the multi-source data correlation, consistency index and offset difference between each object and each node to form a cross-source consistency matrix; Based on the evolution trend of the cross-source consistency matrix and the offset difference over time, the current consistency status, existing abnormal segments, and abnormality score of each object are output.

[0011] Preferably, the process for generating a collaborative assessment result that includes the current deviation status and the future risk level is as follows: Based on the current consistency status, abnormal segments, and abnormality score of each object, combined with the historical change trajectory represented by the offset parameter, the object-level risk value is calculated. The calculated risk values ​​are classified and determined, and the current deviation status and potential future abnormal risks are marked as different risk levels. The collaborative assessment results are mapped to the corresponding edge nodes and monitoring objects to form a final collaborative assessment result that includes the current deviation status and future risk level.

[0012] Preferably, the process of jointly adjusting the participation weight, calculation priority, data forwarding path, and local parameters of edge nodes is as follows: Based on the current deviation status and future risk level in the final collaborative evaluation results, the participation weight of each node in data processing and forwarding is redistributed; By combining the object-level risk values ​​and deviation trends in the collaborative assessment results, the calculation priority is adjusted and the data forwarding path is optimized; The local parameters of each node are adaptively adjusted, including sampling frequency, cache allocation, and task scheduling strategy.

[0013] Preferably, the process of forming a collaborative processing self-evolution mechanism driven by consistency and prediction is as follows: The system collects the execution status and feedback data of each node in real time, including the node's participation weight, calculation priority, data forwarding path, local parameter status, processing delay, offset correction results, and changes in consistency indicators. Based on the collected feedback data and adaptive adjustment results, the offset parameter characterization, consistency constraint parameters and policy adjustment rules are dynamically updated, and the participation weight, calculation priority and data forwarding strategy of each node are further corrected. Through continuous iterative feedback update cycles, a collaborative processing self-evolution mechanism is formed, driven by consistency constraints and risk prediction.

[0014] Compared with the prior art, the present invention provides an edge computing data collaborative processing device, which has the following beneficial effects: This invention utilizes a data acquisition module to perform object-level merging of heterogeneous sensor data from multiple edge nodes, constructing an association representation relationship that maps objects, signals, and nodes. This enables structured management of multi-source information on the same physical object. A state mapping module, considering local clock drift, communication delay, sampling jitter, and load fluctuations of each node, jointly estimates the relative offsets of multi-source data in the time and feature dimensions, forming offset parameter representations to provide accurate references for subsequent feature reconstruction. The feature reconstruction module performs non-rigid alignment and local reconstruction of the multi-source data based on the offset parameters, generating continuous and unified joint expression fragments and simultaneously generating accompanying information reflecting reliability and uncertainty. A consistency analysis module performs cross-source consistency constraint analysis on the joint expression fragments. By identifying subsets of representations that deviate from the main trend, predicting potential degradation or failure risks, and forming collaborative evaluation results, the strategy evolution module dynamically adjusts the participation weights, computation priorities, data forwarding paths, and local model parameters of edge nodes based on the collaborative evaluation results. Combined with operational feedback, it performs closed-loop updates on offset parameters, consistency constraints, and strategy rules, forming a self-evolving collaborative processing mechanism. As a result, this device can significantly improve the accuracy and consistency of multi-source data fusion, reduce the interference of cross-node information conflicts on state judgment and collaborative control, and improve the overall stability, reliability, and response speed of the system. At the same time, by predicting potential degradation and failure risks, it can achieve intelligent and forward-looking management of edge node operation and maintenance and collaborative strategies, meeting the application requirements of high real-time and high consistency scenarios. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, an edge computing data collaborative processing device includes: Data acquisition module: Collects heterogeneous sensing data from multiple edge nodes, and performs object-level merging based on the physical relationships, spatial topological relationships and signal coupling characteristics of the monitored objects to construct an association representation relationship that represents the mapping relationship between objects, signals and nodes.

[0018] The process of constructing the association representation relationship between the representation object, signal, and node in the data acquisition module is as follows: Data streams are collected by sensors deployed at each edge node and aggregated according to the physical connections and topological adjacency relationships between objects; Multiple types of sensors adapted to the monitored object are deployed on multiple edge nodes, including temperature sensors, vibration sensors, current sensors, and pressure sensors. The sensor deployment scheme is determined based on the physical size of the object, signal attenuation characteristics, and node coverage. The spacing between adjacent sensors can be set from 10cm to 1m according to the object characteristics to ensure sufficient data coverage and reasonable redundancy in key areas. The sensors acquire the corresponding data streams in real time, and perform preliminary buffering and preprocessing through the edge nodes. Preprocessing includes low-pass filtering to remove high-frequency noise, outlier removal using the 3σ principle, and normalization to unify the dimensions of different signal types. The data acquisition frequency can be set from 10Hz to 100Hz according to the dynamic characteristics of the object, and a time synchronization mechanism ensures that the data acquisition frequency of each node is consistent with the data acquisition frequency of the monitored object. According to the alignment, a continuous object-level time-series data stream is formed. Based on the physical connection relationships between the monitored objects, such as mechanical connections, electrical connections, or fluid pipeline connections, and the network topology adjacency relationships of edge nodes, the sensor data of the same object are aggregated. This includes constructing an object topology graph, where vertices represent nodes and edges represent physical or signal associations between objects. Through adjacency matrices or graph traversal algorithms, multi-node data of the same object are synchronously aggregated within a specified time window. Weighted average or principal component analysis methods are used to integrate cross-node signals to ensure data integrity and representativeness. During the data aggregation process, delay compensation and conflict handling strategies can be introduced, such as using predictive interpolation for data with delays exceeding a threshold, to maintain the continuity and temporal consistency of object-level data.

[0019] The signals collected by each node are classified and mapped using the signal coupling index, generating a mapping matrix of objects, signals and nodes; For the aggregated object-level data, the coupling degree between the signals acquired by each node is calculated. Coupling degree indicators can include mutual information, Pearson correlation coefficient, covariance, energy coupling degree, etc. Based on the coupling degree, highly correlated signals are grouped into the same category and mapped to the corresponding objects and nodes. An initial coupling threshold is set and dynamically and adaptively adjusted based on historical data. The mapping strategy includes prioritizing mapping highly coupled signals to main nodes and reserving low-coupling signals on auxiliary nodes to provide redundancy. A mapping matrix between objects, signals, and nodes is generated. The matrix records the object to which each signal belongs, the acquisition node, and the coupling degree information, providing a basic data structure for anomaly detection, control optimization, and node collaboration.

[0020] By establishing a preliminary object-level data index through data stream identification and timestamp management, an associated representation relationship is formed that represents the mapping relationship between objects, signals, and nodes; Each acquired data stream is uniquely identified, including object ID, signal ID, and node ID, and accompanied by an acquisition timestamp and data quality identifier. The index structure adopts a triplet format: object ID-signal ID-node ID, and is stored in a distributed time-series database or graph database. It supports batch queries and incremental updates based on time windows. The index management mechanism enables rapid location of the signal status of any object at a specific point in time. It supports historical data tracing, incremental writing, and data stream quality assessment, providing a structured and operable data management foundation for constructing association representation relationships. After completing data aggregation, signal mapping, and index management, the relationships between objects, signals, and nodes are integrated to form association representation relationships. The mapping matrix can be combined with the object-level index to generate a queryable association representation structure, such as a three-dimensional matrix or graph database. Each object corresponds to its acquired signal and the node it belongs to, while retaining signal coupling information and time index. The association representation relationship supports rapid querying, automatic updating of the mapping relationship when a node or signal is added, and retention of historical version information to support time-series analysis and anomaly backtracking.

[0021] State mapping module: acquires the local clock drift, communication queuing delay, sampling jitter and load fluctuation status of each edge node, maps the state of each node to the corresponding object and signal by combining the correlation representation relationship, and jointly estimates the relative offset of multi-source data in the time dimension and feature dimension to generate the corresponding offset parameter representation.

[0022] The process of mapping the state of each node to its corresponding object and signal by combining the association representation relationship in the state mapping module is as follows: In the data acquisition module of each edge node, the local system time offset, task queue queuing delay, sampling interval jitter and current processing load index of the node are read in real time to form a multi-dimensional node status parameter set. In the data acquisition module of each edge node, multi-dimensional status parameters of the node are read in real time, including local system time offset, task queue queuing delay, sampling interval jitter, and current processing load indicators. Node status parameter acquisition can be achieved through an embedded monitoring module. Each indicator is acquired according to a preset sampling frequency, such as 1Hz to 10Hz. The time offset is calculated by periodically synchronizing the node clock with a standard time source. The task queue queuing delay can be calculated by recording the timestamps of the head and tail tasks in the queue. The sampling interval jitter is obtained by statistically analyzing the difference between consecutive sampling intervals. The processing load indicators can be obtained through CPU and memory utilization. All node status data is temporarily stored in the edge node memory cache and appended with timestamps and node IDs to form a multi-dimensional node status parameter set, providing a real-time data foundation for mapping.

[0023] Map the set of node state parameters to the associated representation relationship to generate a multi-dimensional data input table corresponding to objects, nodes and parameters; The multidimensional node state parameter set is mapped to the previously constructed association representation relationship, and each node state parameter is mapped to the object and signal it manages. The mapping can be completed by querying the association representation matrix or graph database. Each node state parameter is queried for the corresponding object and signal ID through the node ID, and an object-node-parameter ternary is generated. Priority is given to matching the main node signal. When mapping highly coupled signals, redundant node states are retained as auxiliary information to prevent node anomalies. The table generated by the mapping result is called the multidimensional data input table corresponding to objects, nodes and parameters, which supports fast indexing, batch querying and incremental updates.

[0024] The parameter differences in the multidimensional data input table are normalized and standardized to form a unified parameter set; The various node state parameters in the multidimensional data input table are normalized and standardized to form a unified parameter set. Time offset and sampling jitter can be linearly normalized to map the parameters to the 0-1 interval. Queue queuing delay and processing load indicators can be standardized using Z-score to eliminate the influence of different dimensions and highlight abnormal fluctuation characteristics. Normalization and standardization can be completed at edge nodes or central processing units. The processing results retain object ID, signal ID, and node ID, enabling comparison and analysis of multidimensional parameters at the same scale. The unified parameter set can be used for both object-level state estimation and... It can provide an input basis for dynamic scheduling, anomaly detection, and predictive analysis. After normalization and standardization, the state parameters of each node are summarized by object and signal to generate an object-level multidimensional parameter set. The triplet data can be grouped and summarized according to object ID to calculate the object-level average state value, maximum and minimum value, and standard deviation. At the same time, the state of each node is retained as redundant information to support backtracking and anomaly analysis. The resulting object-level parameter set can be directly used for task scheduling, signal correction, anomaly detection, and system optimization control in edge computing, realizing effective mapping and operable utilization of node state to object and signal.

[0025] The process of generating the corresponding offset parameter representation in the state mapping module is as follows: Based on the parameter set, the multidimensional data input table corresponding to the object, node and parameter is used as the basis for joint analysis; The multidimensional data input table, which includes the normalized and standardized states of each node, is used as the basis for joint analysis. The data table is organized as a three-dimensional matrix, with dimensions of object ID, signal ID, and node state parameters. Each element in the matrix records the normalized node parameter value and the acquisition timestamp. Normalization uses a linear mapping to the 0-1 interval, and standardization uses the Z-score method with a mean of 0 and a standard deviation of 1. It supports simultaneous access to the state data of any object at different nodes, provides a unified interface for offset measurement and joint estimation, ensures data integrity and indexability, and supports system analysis in the time and feature dimensions.

[0026] The offset of each object in the time dimension and feature dimension of the multi-source data at different nodes is measured and calculated, and iterative adjustments are made. For each object, multi-source data collected from different nodes is used to measure and calculate offsets in both the time and feature dimensions. In the time dimension, signal delay differences are calculated using cross-correlation functions or mutual information methods to measure the signal time offset between nodes. The sampling window length is 5-10 seconds, dynamically adjustable to accommodate different sampling rates. In the feature dimension, differences in node state parameters, such as sampling interval jitter, task queue delay, and processing load differences, are evaluated. The offset calculation employs an iterative optimization method, gradually adjusting the offset value through weighted least squares or recursive filtering. The iteration terminates when the offset change is less than 0.01ms or the maximum number of iterations (50) is reached. Abnormal nodes or data are removed using the 3σ principle or empirical threshold to ensure the robustness and accuracy of the offset estimation. The initially measured offset values ​​are then input into the iterative optimization model for adjustment to reduce offset errors between objects and nodes. Iterative optimization is achieved through recursive least squares, Kalman filtering, or gradient descent methods. After each iteration, the offset results are compared with the original multidimensional parameter table to ensure that the adjustment does not introduce abnormal fluctuations. The time series continuity is maintained during the iteration process, and weighted fusion is performed using redundant information from multiple nodes. The fusion weights are dynamically allocated according to the historical reliability of the nodes and the signal coupling degree to improve the reliability and interpretability of the offset parameters.

[0027] The offset information obtained from the joint estimation is integrated to form the corresponding offset parameter representation; The offset values ​​of each object at different nodes and different signals are organized into a three-dimensional matrix according to object ID, signal ID, and node ID. Each matrix element records the offset and calculation accuracy index. It supports fast indexing and batch querying. The information of objects, nodes, signals and offsets can also be stored as a graph database structure, where objects, nodes and signals are nodes, and offsets and related information are attributes of edges, realizing flexible relationship query and analysis. The offset parameter characterization can be used to correct object-level data, time synchronization, signal fusion and edge node collaborative control, and retain historical offset records for anomaly backtracking and dynamic optimization, thus forming a structured and operable offset parameter system.

[0028] Feature reconstruction module: Based on the corresponding offset parameters, non-rigid time alignment and local feature reconstruction are performed on the multi-source data corresponding to the same object to generate a unified joint expression fragment while maintaining the continuity of the original signal.

[0029] The process of performing non-rigid time alignment and local feature reconstruction on multi-source data corresponding to the same object in the feature reconstruction module is as follows: Based on the offset parameter characterization, the offset of multi-source data collected by each object at different edge nodes in the time dimension and feature dimension is corrected point by point; The time axis of each signal is adjusted point by point according to the offset parameter to eliminate the time difference between nodes. At the same time, the sampling value is corrected according to the offset in the feature dimension to ensure the correspondence and synchronization of data of each node on the same object. This ensures that multi-source signals can be directly fused and analyzed in time and feature space, laying the foundation for the construction of joint representation.

[0030] During the alignment process, missing sampling points within the data segment are interpolated, abnormal abrupt changes are identified and corrected, and the signal is smoothed by combining local statistical features. During time alignment, missing sampling points within the data segment are filled in. Linear interpolation or cubic spline interpolation algorithms are used to calculate the value of the missing point as a weighted average of the nearest valid sampling points, or cubic spline fitting is used to fit the continuous segment to generate the interpolated value. Interpolation markers are recorded to facilitate the distinction between true sampling and compensated sampling during analysis. Interpolation ensures signal continuity and avoids the impact of missing points on feature calculation and pattern extraction. There may be anomalous abrupt changes in the aligned signal, such as outliers caused by transient interference or sampling errors. A sliding window statistical method is used to detect anomalies. The window size is generally 5 to 20 sampling points. The mean μ and standard deviation σ within the window are calculated. Sampling points that deviate from μ ± 3σ are marked as anomalies. Anomalies can be corrected by local weighted smoothing. The outlier value is replaced by the weighted average of the nearest valid points or local polynomial fitting, while the correction marker is retained for anomaly backtracking analysis.

[0031] The aligned and reconstructed data segments are subjected to segmented feature extraction and standardization, including amplitude normalization, frequency domain feature mapping and local temporal pattern coding, and each segment is uniformly mapped to an object-level joint representation structure. After alignment and correction, the data undergoes segmented feature extraction and standardization. Segmentation methods can employ fixed time windows or sliding segmentation with a fixed number of sampling points. Feature extraction includes the following: amplitude normalization, mapping the signal to the 0-1 interval to ensure comparability between signals at different nodes; frequency domain feature mapping, extracting local spectral energy distribution through Fast Fourier Transform, such as the amplitude of the dominant frequency component and the frequency band energy ratio; and local time-series pattern encoding, using a sliding window to generate time-series segment vectors, including statistical features such as mean, variance, peak value, skewness, and local trends. Standardization can employ Z-score normalization or min-max normalization to ensure that all feature dimensions are on the same order of magnitude. The feature vectors generated from each segment are mapped to an object-level joint representation structure. The joint representation can be in the form of a three-dimensional matrix or a graph database. In a three-dimensional matrix, the dimensions are object ID × signal ID × time segment index, and each element stores the corresponding feature vector and quality index. In a graph database, objects, nodes, and signals are nodes, and segmented feature vectors and offset information are used as edge attributes, enabling complex queries and analysis of relationships between nodes.

[0032] The process of generating a unified joint expression fragment in the feature reconstruction module while maintaining the continuity of the original signal is as follows: Based on the data segments after non-rigid time alignment and local feature reconstruction, the multi-source data of each object is subjected to continuity verification and filtering to correct local signal jumps and noise interference. For each signal, a sliding window detection is performed to check whether the difference between adjacent sampling points exceeds a preset threshold, such as an amplitude change greater than 3 times the standard deviation or 5% of the dynamic range. Local jumps or anomalies are identified. For detected anomalies, local weighted smoothing or median filtering can be used for correction, while retaining anomaly markers. This includes low-pass or band-pass filtering of signal noise to remove high-frequency interference and system noise, ensuring the overall continuity and analyzability of the signal.

[0033] The processed data is spliced ​​and fused in the time and feature dimensions, integrating the local feature vectors and time sequence patterns of each node, and mapped to the object-level joint representation structure; Based on the offset parameter characterization, the time of each node signal is corrected to a unified reference axis, and missing sampling points are interpolated to ensure the continuity of the time series. For feature vectors of the same type, such as amplitude, frequency domain energy, and local time series statistics, a weighted average is used for fusion. The weights are dynamically adjusted based on node signal quality, historical offset, and anomaly correction. Simultaneously, the differences in feature magnitude are standardized. The fused signal is segmented by a sliding window, and local time series patterns are extracted, including statistical features such as mean, variance, peak value, skewness, and trend, forming a time index mapping. Finally, the feature vectors of each segment are integrated into an object-level joint representation structure, which can be a three-dimensional matrix (object ID × signal ID × time segment), with each element containing a feature vector and quality index, or a graph database, where objects, nodes, and signals are nodes, and offset information and fused features are edge attributes. This achieves unified representation of multi-source data, continuity preservation, and signal quality tracking, and can be directly used for object-level analysis, anomaly detection, signal fusion, and collaborative control of edge nodes.

[0034] During the mapping and fusion process, the weights are dynamically adjusted based on the offset parameter characterization. Weighted mapping and local smoothing are applied to high offset or local abnormal regions to generate a unified object-level joint expression fragment. For nodes with large offsets or frequent historical anomalies, their signal weights are reduced, while nodes with small offsets and high signal quality are given higher weights to ensure that the contribution of key node data to the joint representation is maximized. In the time dimension, the signals of each node are weighted and interpolated to eliminate discontinuities caused by local missing values ​​or time misalignments. In the feature dimension, feature vectors of the same type are fused by weighted averaging, and the fusion result is locally smoothed, for example, by using cubic spline smoothing or local weighted regression to correct abrupt changes and high-frequency noise. The processed continuous feature segments are mapped to an object-level joint representation structure, which can be in the form of a three-dimensional matrix (object ID × signal ID × time segment), with each element storing feature vectors and quality indicators, or in the form of a graph database, where objects, nodes, and signals are nodes, and offset information and fusion features are edge attributes, forming a unified object-level joint representation segment.

[0035] Consistency Analysis Module: Performs cross-source consistency constraint analysis on joint expression fragments, combines consistency indices and offset parameters to characterize the time change trend, predicts potential degradation or failure risks, and generates a collaborative assessment result that includes the current deviation status and future risk level.

[0036] The process of performing cross-source consistency constraint analysis on jointly expressed fragments in the consistency analysis module is as follows: Based on object-level joint expression fragments, the time series and feature vectors of multi-source data are mapped to the cross-source consistency analysis structure; Based on the object-level joint expression fragment, the feature vector of each time segment in the object-level joint expression fragment is indexed according to the node and signal type, and mapped to a unified analysis structure, such as a three-dimensional matrix or a multi-dimensional tensor, with dimensions of object ID × signal ID × time segment, respectively. This allows simultaneous access to the time series and feature information of any object at different nodes, providing a unified interface for consistency calculation and ensuring the comparability and traceability of cross-source data.

[0037] By comparing the joint expression fragments with historical steady-state models, physical consistency models, and prior behavioral templates, deviation patterns and potential anomalous regions are identified. Historical stable state models can be generated from the statistical features of object-level joint expression fragments collected over a long period of time. Physical consistency models can be constructed through physical connection relationships or signal coupling laws between objects. Prior behavior templates can be derived from normal operating patterns defined by experts. Comparison methods include calculating the mean deviation, variance change, and frequency domain feature differences of signals to identify time periods and local abnormal regions that deviate from the normal pattern and to label potential abnormal fragments.

[0038] Calculate the multi-source data correlation, consistency index and offset difference between each object and each node to form a cross-source consistency matrix; The consistency of signals between different nodes is quantified using indicators such as correlation coefficient, mutual information, covariance, or energy coupling. The deviation of feature vectors in the time dimension and feature dimension is measured to form an offset difference matrix. The calculation results between each object, node, and signal are integrated into a cross-source consistency matrix to record the consistency level of multi-source data of each object in a structured manner.

[0039] Based on the evolution trend of the cross-source consistency matrix and the offset difference over time, output the current consistency status of each object, the existing abnormal segments, and the degree of abnormality score; Time segments with cross-source consistency indices below a preset threshold or with significant deviations are marked. Anomaly scores are calculated for the marked segments based on the consistency deviation magnitude, duration, and deviation trend. The output includes the current consistency status of each object, the start and end times of the abnormal segment, and the anomaly score. The abnormal information can also be associated with node location and signal type for subsequent fault location, early warning, or optimized control.

[0040] The process of generating a collaborative assessment result that includes the current deviation status and future risk level in the consistency analysis module is as follows: Based on the current consistency status, abnormal segments, and abnormality score of each object, combined with the historical change trajectory represented by the offset parameter, the object-level risk value is calculated. For each object, the current consistency status, abnormal segments, and abnormality scores generated by cross-source consistency analysis are obtained. At the same time, the historical change trajectory represented by the offset parameters of the object is extracted, including time series offset, number of anomaly corrections, and node contribution weights. Based on this data, each object is quantitatively evaluated through risk calculation models, such as weighted scoring models, Bayesian update models, or time series prediction models, to generate object-level risk values. The risk values ​​comprehensively reflect the current deviation status and the probability of future potential anomalies, providing a numerical basis for risk classification.

[0041] The calculated risk values ​​are classified and determined, and the current deviation status and potential future abnormal risks are marked as different risk levels. Based on preset or dynamic thresholds, risk values ​​are classified into low, medium, high, or four or more levels. The risk level is adjusted by combining the duration, deviation magnitude, and historical deviation patterns of the abnormal segment. This ensures that the classification reflects both the current degree of deviation and potential future risk trends. The classification results can be used to quickly identify high-risk objects and key monitoring nodes, providing a reference for anomaly response and decision-making.

[0042] The collaborative assessment results are mapped to the corresponding edge nodes and monitoring objects to form a final collaborative assessment result that includes the current deviation status and future risk level. An object-node mapping table is established, with each row recording the object ID, corresponding acquisition node ID, risk value, current deviation status, abnormal section, and abnormal score. For objects involving the collaboration of multiple nodes, the risks between nodes are weighted and summarized based on node signal quality, historical offset, and abnormal frequency to form a comprehensive risk index. The final collaborative evaluation result includes the current deviation status, abnormal section, and future risk level of each object, and is also associated with specific edge nodes, supporting real-time early warning, resource scheduling, and collaborative control of the edge computing system.

[0043] The strategy evolution module: Based on the collaborative evaluation results, it jointly adjusts the participation weight, computation priority, data forwarding path and local parameters of edge nodes, and updates the offset parameter representation, consistency constraint parameters and strategy adjustment rules in combination with the operation feedback, forming a collaborative processing self-evolution mechanism driven by consistency and prediction.

[0044] The process of jointly adjusting the participation weights, computation priorities, data forwarding paths, and local parameters of edge nodes in the strategy evolution module is as follows: Based on the current deviation status and future risk level in the final collaborative evaluation results, the participation weight of each node in data processing and forwarding is redistributed; For each edge node, the current deviation status, abnormal segment, and future risk level of its corresponding object are obtained. Based on this information, the participation weight of the node in data processing and forwarding is redistributed. For nodes that are responsible for monitoring high-risk objects or frequently experiencing anomalies, their participation weight is increased to ensure that critical data is processed first. For nodes with low deviation and low risk, their participation weight is reduced to save computing and communication resources. Node weights can be dynamically updated locally on the node or on the edge controller, and the history of weight changes is recorded for optimization and analysis.

[0045] By combining the object-level risk values ​​and deviation trends in the collaborative assessment results, the calculation priority is adjusted and the data forwarding path is optimized; By combining object-level risk values ​​and deviation trends, the task computation priority of each node is dynamically adjusted. Signal processing tasks for high-risk objects are prioritized, while tasks for low-risk objects can be delayed or processed in batches. At the same time, data forwarding paths are optimized based on data flow between nodes and network topology. For example, by using shortest path first, load balancing, or risk-weighted routing strategies, critical data is quickly transmitted to aggregation or control nodes. Path optimization takes into account node load, bandwidth limitations, and task latency to ensure that the data flow of the entire edge network is efficient and reliable.

[0046] Adaptive adjustment of local parameters for each node, including sampling frequency, cache allocation, and task scheduling strategy; The local parameters of each node are dynamically adjusted to coordinate with weight allocation and priority adjustment. This includes adjusting the sampling frequency, increasing the sampling rate of high-risk objects and decreasing the sampling rate of low-risk objects based on the object risk level, optimizing cache allocation, prioritizing the storage of high-weight tasks or key data fragments in the fast access cache to ensure processing efficiency, and adjusting task scheduling strategies. By combining the current node load, task priority, and collaborative scheduling rules, local computing resources are flexibly arranged to achieve load balancing and response acceleration. The adjustment of node local parameters can be executed in real time or updated periodically and linked with the collaborative evaluation results to achieve adaptive collaborative control of edge nodes.

[0047] The process by which a cooperative self-evolution mechanism driven by consistency and prediction is formed in the strategy evolution module is as follows: The system collects the execution status and feedback data of each node in real time, including the node's participation weight, calculation priority, data forwarding path, local parameter status, processing delay, offset correction results, and changes in consistency indicators. During the operation of edge nodes, the execution status and feedback information of the nodes are collected in real time, including the node's participation weight, calculation priority, data forwarding path, local parameters such as sampling frequency, cache allocation, task scheduling strategy status, as well as processing latency, offset correction results, number of anomaly corrections, and changes in consistency indicators. The data can be periodically uploaded through the local cache of the edge nodes or transmitted to the collaborative control module in real time through the message bus. By summarizing and structurally storing the multi-dimensional feedback data, such as in the form of an object-node-time three-dimensional matrix or graph database, a comprehensive state basis is provided for the self-evolution mechanism.

[0048] Based on the collected feedback data and adaptive adjustment results, the offset parameter characterization, consistency constraint parameters and policy adjustment rules are dynamically updated, and the participation weight, calculation priority and data forwarding strategy of each node are further corrected. The effects of node offset correction and changes in consistency indicators are analyzed. Offset parameters are adjusted to improve data accuracy in time and feature dimensions. Constraint parameters are corrected based on the effect of consistency constraint execution, such as adjusting cross-source consistency thresholds and weight allocation coefficients. Policy adjustment rules are optimized, including participation weight update strategies, calculation priority allocation rules, and data forwarding path optimization strategies, so that the system can allocate resources and schedule tasks more accurately in the next iteration. Incremental calculation and sliding window analysis methods are used in the update process to ensure real-time response and reduce computational overhead.

[0049] Through a continuous iterative feedback update cycle, a collaborative processing self-evolution mechanism driven by consistency constraints and risk prediction is formed. The dynamically updated parameters and strategies are then applied to each node for execution, resulting in new node states and data processing results. Feedback is continuously collected, and through a continuous iterative feedback update cycle, the system achieves self-adaptation and self-evolution. The processing weight of high-risk objects or nodes with severe deviations is automatically increased, and data forwarding paths and task priorities are dynamically adjusted according to real-time risk and consistency indicators. At the same time, the offset parameter representation and consistency constraint rules are optimized with historical feedback, enabling the system to achieve efficient collaboration under different time and load conditions. This self-evolution mechanism is driven by consistency constraints and risk prediction, enabling edge nodes to maintain an adaptive and optimal state in multi-source data processing, anomaly correction, and collaborative control.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An edge computing data collaborative processing device, characterized in that, include: Data acquisition module: Collects heterogeneous sensor data from multiple edge nodes, and performs object-level merging based on the physical relationship, spatial topology relationship and signal coupling characteristics of the monitored objects to construct an association representation relationship that represents the mapping relationship between objects, signals and nodes; State mapping module: acquires the local clock drift, communication queuing delay, sampling jitter and load fluctuation status of each edge node, maps the state of each node to the corresponding object and signal by combining the correlation representation relationship, and jointly estimates the relative offset of multi-source data in the time dimension and feature dimension to generate the corresponding offset parameter representation. Feature reconstruction module: Based on the corresponding offset parameters, non-rigid time alignment and local feature reconstruction are performed on the multi-source data corresponding to the same object to generate a unified joint expression fragment while maintaining the continuity of the original signal; Consistency Analysis Module: Performs cross-source consistency constraint analysis on joint expression fragments, combines consistency indices and offset parameters to characterize the time change trend, predicts potential degradation or failure risks, and generates a collaborative assessment result that includes the current deviation status and future risk level; The strategy evolution module: Based on the collaborative evaluation results, it jointly adjusts the participation weight, computation priority, data forwarding path and local parameters of edge nodes, and updates the offset parameter representation, consistency constraint parameters and strategy adjustment rules in combination with the operation feedback, forming a collaborative processing self-evolution mechanism driven by consistency and prediction.

2. The edge computing data collaborative processing device according to claim 1, characterized in that, The process of constructing the associative representation relationship between representation objects, signals, and nodes is as follows: Data streams are collected by sensors deployed at each edge node and aggregated according to the physical connections and topological adjacency relationships between objects; The signals collected by each node are classified and mapped using the signal coupling index, generating a mapping matrix of objects, signals and nodes; By establishing a preliminary object-level data index through data stream identification and timestamp management, an associated representation relationship is formed that represents the mapping relationship between objects, signals, and nodes.

3. The edge computing data collaborative processing device according to claim 2, characterized in that, The process of mapping the state of each node to the corresponding object and signal by combining the association representation relationship is as follows: In the data acquisition module of each edge node, the local system time offset, task queue queuing delay, sampling interval jitter and current processing load index of the node are read in real time to form a multi-dimensional node status parameter set. Map the set of node state parameters to the associated representation relationship to generate a multi-dimensional data input table corresponding to objects, nodes and parameters; The parameter differences in the multidimensional data input table are normalized and standardized to form a unified parameter set.

4. The edge computing data collaborative processing device according to claim 3, characterized in that, The process of generating the corresponding offset parameter representation is as follows: Based on the parameter set, the multidimensional data input table corresponding to the object, node and parameter is used as the basis for joint analysis; The offset of each object in the time dimension and feature dimension of the multi-source data at different nodes is measured and calculated, and iterative adjustments are made. The offset information obtained from the joint estimation is integrated to form the corresponding offset parameter representation.

5. The edge computing data collaborative processing device according to claim 4, characterized in that, The process of performing non-rigid time alignment and local feature reconstruction on multi-source data corresponding to the same object is as follows: Based on the offset parameter characterization, the offset of multi-source data collected by each object at different edge nodes in the time dimension and feature dimension is corrected point by point; During the alignment process, missing sampling points within the data segment are interpolated, abnormal abrupt changes are identified and corrected, and the signal is smoothed by combining local statistical features. The aligned and reconstructed data segments are subjected to segmented feature extraction and standardization, including amplitude normalization, frequency domain feature mapping and local temporal pattern coding, and each segment is uniformly mapped to an object-level joint representation structure.

6. The edge computing data collaborative processing device according to claim 5, characterized in that, The process of generating a unified joint expression fragment while maintaining the continuity of the original signal is as follows: Based on the data segments after non-rigid time alignment and local feature reconstruction, the multi-source data of each object is subjected to continuity verification and filtering to correct local signal jumps and noise interference. The processed data is spliced ​​and fused in the time and feature dimensions, integrating the local feature vectors and time sequence patterns of each node, and mapped to the object-level joint representation structure; During the mapping and fusion process, the weights are dynamically adjusted based on the offset parameter. Weighted mapping and local smoothing are applied to high offset or local abnormal regions to generate a unified object-level joint expression fragment.

7. The edge computing data collaborative processing device according to claim 6, characterized in that, The process of performing cross-source consistency constraint analysis on jointly expressed fragments is as follows: Based on object-level joint expression fragments, the time series and feature vectors of multi-source data are mapped to the cross-source consistency analysis structure; By comparing the joint expression fragments with historical steady-state models, physical consistency models, and prior behavioral templates, deviation patterns and potential anomalous regions are identified. Calculate the multi-source data correlation, consistency index and offset difference between each object and each node to form a cross-source consistency matrix; Based on the evolution trend of the cross-source consistency matrix and the offset difference over time, the current consistency status, existing abnormal segments, and abnormality score of each object are output.

8. The edge computing data collaborative processing device according to claim 7, characterized in that, The process of generating a collaborative assessment result that includes the current deviation status and the future risk level is as follows: Based on the current consistency status, abnormal segments, and abnormality score of each object, combined with the historical change trajectory represented by the offset parameter, the object-level risk value is calculated. The calculated risk values ​​are classified and determined, and the current deviation status and potential future abnormal risks are marked as different risk levels. The collaborative assessment results are mapped to the corresponding edge nodes and monitoring objects to form a final collaborative assessment result that includes the current deviation status and future risk level.

9. The edge computing data collaborative processing device according to claim 8, characterized in that, The process of jointly adjusting the participation weights, computation priorities, data forwarding paths, and local parameters of edge nodes is as follows: Based on the current deviation status and future risk level in the final collaborative evaluation results, the participation weight of each node in data processing and forwarding is redistributed; By combining the object-level risk values ​​and deviation trends in the collaborative assessment results, the calculation priority is adjusted and the data forwarding path is optimized; The local parameters of each node are adaptively adjusted, including sampling frequency, cache allocation, and task scheduling strategy.

10. An edge computing data collaborative processing device according to claim 9, characterized in that, The process of forming a collaborative processing self-evolution mechanism driven by consistency and prediction is as follows: The system collects the execution status and feedback data of each node in real time, including the node's participation weight, calculation priority, data forwarding path, local parameter status, processing delay, offset correction results, and changes in consistency indicators. Based on the collected feedback data and adaptive adjustment results, the offset parameter characterization, consistency constraint parameters and policy adjustment rules are dynamically updated, and the participation weight, calculation priority and data forwarding strategy of each node are further corrected. Through continuous iterative feedback update cycles, a collaborative processing self-evolution mechanism is formed, driven by consistency constraints and risk prediction.

Images