A Zero-Intrusion Computational Data Collaborative Sensing and Time Alignment Method

By non-intrusive collection of computing power events and resource loads, combined with cross-node communication protocols and timing compensation, a collaborative alignment sequence is generated, which solves the timing misalignment problem in collaborative perception of computing power data and realizes high-precision global perception of computing power networks.

CN123122859APending Publication Date: 2026-11-17GUIZHOU POWER GRID CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610985141.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing methods for collaborative sensing and time alignment of computing and power data are difficult to achieve the accuracy of global sensing state of computing power networks in complex computing and power interaction scenarios, and are prone to temporal misalignment and feature coupling shift.

Method used

By deploying lightweight probes to non-intrusively collect task lifecycle events and resource loads, a global tracking identifier is generated. Combined with cross-node communication protocols and full-link call topology, physical power consumption data is distributed to determine the synchronization deviation of cross-domain temporal nonlinear events and perform timing compensation to generate a collaborative alignment sequence. Finally, computing and power fusion feature extraction and compensation parameter iterative updates are performed.

Benefits of technology

It improves the accuracy of global perception of computing power networks in complex interactive environments, ensures strict spatiotemporal mapping and consistency between computing power task scheduling and hardware energy consumption response, eliminates timing misalignment, and provides high-confidence timing alignment support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122859A_ABST
    Figure CN123122859A_ABST
Patent Text Reader

Abstract

This application provides a zero-intrusion computing power data collaborative perception and time alignment method, relating to the field of data processing technology. It generates a global tracking identifier based on task lifecycle events, transmits the global tracking identifier through a cross-node communication protocol to obtain the full-link call topology, and distributes physical power consumption data according to the resource occupancy ratio in the full-link call topology to obtain a node power consumption sequence. It determines the event synchronization deviation due to cross-domain temporal nonlinearity by the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence, and corrects the same-frequency computing power segments in the node power consumption sequence based on the event synchronization deviation to obtain a collaborative alignment sequence. The compensation parameters for computing power time alignment are iteratively updated based on computing power fusion characteristics. This application can achieve accurate collaborative updates of zero-intrusion computing power data collaborative perception and time alignment, thereby improving the accuracy of the global perception state of the computing power network after complex computing power interactions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a zero-intrusion computational data collaborative sensing and time alignment method. Background Technology

[0002] Data processing refers to the collection, cleaning, transformation, and structured analysis of computing load logs generated when computing nodes execute tasks and power consumption monitoring data collected in real time by underlying physical devices in the computing network operating environment. In the scenario of computing-power collaborative sensing, the core of data processing lies in performing multi-dimensional feature extraction and spatiotemporal correlation analysis on discrete computing logic states and continuous physical energy consumption signals. The aim is to deeply explore the inherent mapping law between computing task scheduling and physical hardware energy consumption response, thereby providing standardized and high-confidence underlying data support for the fine-grained scheduling of computing resources, energy efficiency assessment, and accurate construction of computing-power fusion characteristics.

[0003] However, current methods for collaborative sensing and time alignment of computing power data rely solely on coarse-grained timestamp matching and stateless unidirectional data stitching for data alignment and feature extraction. They fail to incorporate nonlinear mapping of state transitions at computing power segment boundaries, cross-node temporal rearrangement, quantification of spatiotemporal coupling correlation of computing power features, deep feature fusion deduction, and dynamic iteration of compensation parameters. This makes it difficult to directionally correct timing misalignments and feature coupling offsets under complex computing power scheduling, and to ensure the consistency of global timing and feature representation in collaborative sensing. This easily leads to problems such as overlapping and disordered timing during data alignment, causing feature loss of synchronization, and cross-node sensing violating strict spatiotemporal coupling dependencies. Therefore, achieving accurate collaborative updates for zero-intrusion collaborative sensing and time alignment of computing power data to improve the accuracy of the global perception state of the computing power network after complex computing power interactions is a key challenge facing the industry. Summary of the Invention

[0004] This application provides a zero-intrusion computing power data collaborative sensing and time alignment method, which can achieve accurate collaborative updates of zero-intrusion computing power data collaborative sensing and time alignment, so as to improve the accuracy of the global perception state of the computing power network after complex computing power interactions.

[0005] Firstly, this application provides a zero-intrusion computational data collaborative sensing and time alignment method, wherein the method: Based on the deployed lightweight probe, non-intrusive collection of task lifecycle events and resource load is performed to obtain computing power event sequences; A global tracking identifier is generated based on the task lifecycle events. The global tracking identifier is transmitted through a cross-node communication protocol to obtain the end-to-end call topology. The physical power consumption data is distributed according to the resource occupancy ratio in the end-to-end call topology to obtain the node power consumption sequence. The event synchronization deviation of cross-domain temporal nonlinearity is determined by the state transition point of the computing power event sequence and the power consumption mutation point of the node power consumption sequence, and the same-frequency computing power segment in the node power consumption sequence is corrected according to the event synchronization deviation to obtain the cooperative alignment sequence. The collaborative alignment sequence is fused and deduced to obtain computational-electrical fusion characteristics, and the compensation parameters for computational-electrical timing alignment are iteratively updated based on the computational-electrical fusion characteristics.

[0006] In this embodiment, the non-intrusive collection of task lifecycle events and resource load to obtain the computing power event sequence specifically includes: Non-intrusive capture of task state transition signals and underlying resource indicators to obtain raw perception data; The raw sensing data is analyzed for features to separate task lifecycle events and resource load within the corresponding execution time window; The task lifecycle events and the resource load are spliced ​​together in time sequence to obtain a computing power event sequence.

[0007] In this embodiment, generating a global tracking identifier based on the task lifecycle event specifically includes: Extract the task creation trigger signal from the task lifecycle events to generate network request features; The global tracking identifier is determined based on the network request characteristics.

[0008] In this embodiment, the global tracking identifier is transmitted through a cross-node communication protocol to obtain the end-to-end call topology, specifically including: Intercept packets along the network stack transmit / receive path of the cross-node communication protocol, and non-intrusively inject the global tracking identifier into the extended field of the protocol header; Based on the protocol parsing results of the extended fields during cross-node transmission, record the call offset between the source node and the destination node; Based on the call offset, the distributed call fragments are aggregated into a directed graph to generate a full-link call topology.

[0009] In this embodiment, the physical power consumption data is distributed according to the resource occupancy ratio in the end-to-end call topology to obtain the node power consumption sequence, which specifically includes: The instantaneous power consumption allocation value is determined by combining the total node power corresponding to the physical power consumption data with the resource occupancy ratio. The instantaneous power consumption allocation value is time-aligned based on the timestamp of the full-link call topology, and combined with the physical power consumption data, a node power consumption sequence is generated.

[0010] In this embodiment, determining the event synchronization deviation of cross-domain temporal nonlinearity through the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence specifically includes: Extract the logical trigger time of the state transition point and the physical response time of the power consumption mutation point to determine the transient delay difference; Based on the transient delay difference, a cross-domain node clock drift sequence is constructed, and a piecewise polynomial is used to fit the clock drift sequence to obtain a timing compensation curve. The event synchronization deviation is determined based on the timing compensation curve.

[0011] In this embodiment, the timing compensation curve is obtained by fitting the clock drift sequence using a piecewise polynomial, specifically including: The clock drift sequence is divided into multiple steady-state intervals, and polynomial regression is performed on each steady-state interval to obtain local component curves; Boundary fusion is performed on the curves of each local component to obtain the time-series compensation curve.

[0012] In this embodiment, the process of correcting the same-frequency computing power segments in the node power consumption sequence based on the event synchronization deviation to obtain the cooperative alignment sequence specifically includes: Based on the basic execution frequency of the computing power task, spectral features are extracted from the node power consumption sequence to locate the computing power segment with the same frequency, and the timing compensation offset is determined in combination with the event synchronization deviation. The timing compensation offset is applied to the corresponding sampling point to perform timing translation, thereby obtaining the event transition characteristics; The collaborative alignment sequence is determined based on the event transition characteristics.

[0013] In this embodiment, determining the cooperative alignment sequence based on the event transition characteristics specifically includes: Extract the state transition parameters of the event transition features at the intersection of computing power segments, and determine the event transition alignment mapping relationship; Based on the event transition alignment mapping relationship, cross-node timing rearrangement is performed on the power consumption correction subsequence of each node to generate a cooperative alignment sequence.

[0014] In this embodiment, the fusion deduction of the cooperative alignment sequence to obtain the computational-electronic fusion features specifically includes: The computing load and power consumption fluctuation features are extracted from the cooperative alignment sequence, a computing power consumption feature matrix is ​​constructed, and the spatiotemporal coupling correlation between the computing load and the power consumption fluctuation features is determined. Based on the spatiotemporal coupling correlation degree, the computational power feature matrix is ​​fused and deduced to obtain the computational power fusion feature.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: Based on the deployed lightweight probe, non-intrusive collection of task lifecycle events and resource load is performed to obtain a computing power event sequence. A global tracking identifier is generated based on the task lifecycle events, and the global tracking identifier is transmitted through a cross-node communication protocol to obtain the end-to-end call topology. The physical power consumption data is distributed according to the resource occupancy ratio in the end-to-end call topology to obtain a node power consumption sequence. The event synchronization deviation of cross-domain temporal nonlinearity is determined by the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence. The same-frequency computing power segments in the node power consumption sequence are corrected according to the event synchronization deviation to obtain a collaborative alignment sequence. The collaborative alignment sequence is fused and deduced to obtain computing power fusion characteristics. The compensation parameters of computing power timing alignment are iteratively updated according to the computing power fusion characteristics.

[0016] Therefore, in this application, the compensation parameters for computing power timing alignment can be iteratively updated based on the computing power fusion characteristics. Specifically, by determining the node power consumption sequence, a high-fidelity dynamic evolution characterization of physical energy consumption in the continuous time dimension can be obtained. This reveals the transient power response law and steady-state energy consumption baseline drift characteristics of the underlying hardware under complex computing power scheduling excitation. This provides a physically consistent original signal benchmark for subsequent removal of environmental noise interference and extraction of computing power coupling characteristics, effectively avoiding the loss of energy consumption information caused by discrete sampling, and ensuring that the global perception model accurately depicts and quantifies the real operating state of physical devices in a multi-dimensional spatiotemporal scale. By determining the cooperative alignment sequence, a strict spatiotemporal mapping relationship between computing power logic events and physical energy consumption signals under a unified time base can be obtained. This eliminates timing misalignment and phase delay caused by heterogeneous clock sources and sampling frequency differences, ensuring strict causal consistency of computing and power fusion characteristics on the time axis. This enables the global state perception to accurately track the transient coupling trajectory of computing power task scheduling and hardware energy consumption fluctuations, thereby providing high-confidence timing alignment support for the dynamic iteration of compensation parameters and improving the accuracy of global state perception of the computing power network in complex interactive environments.

[0017] In summary, the technical solution adopted in this application can achieve precise collaborative updates of zero-intrusion computing power data collaborative perception and time alignment, thereby improving the accuracy of the global perception state of the computing power network after complex computing power interactions. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an exemplary flowchart of a zero-intrusion computer data collaborative sensing and time alignment method provided in this application; Figure 2 This is a flowchart illustrating the process of determining the cooperative alignment sequence provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a zero-intrusion computing power data collaborative perception and time alignment method. Its core is based on a deployed lightweight probe to non-intrusively collect task lifecycle events and resource loads, obtaining a computing power event sequence. A global tracking identifier is generated based on the task lifecycle events, and this identifier is transmitted via a cross-node communication protocol to obtain a full-link call topology. Physical power consumption data is distributed according to the resource occupancy ratio in the full-link call topology to obtain a node power consumption sequence. Cross-domain temporal nonlinearity event synchronization deviations are determined by the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence. The same-frequency computing power segments in the node power consumption sequence are corrected based on these event synchronization deviations to obtain a collaborative alignment sequence. The collaborative alignment sequence is fused and extrapolated to obtain computing power fusion characteristics, and the compensation parameters for computing power time alignment are iteratively updated based on these characteristics.

[0022] To better understand the above technical solutions, a detailed description of the technical solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a zero-intrusion computational data collaborative sensing and time alignment method according to an embodiment of this application. The method includes the following steps: In step S1, based on the deployed lightweight probe, non-intrusive collection of task lifecycle events and resource load is performed to obtain a computing power event sequence.

[0023] In practical implementation, the lightweight probe, once deployed, can be implemented in the following way: In the operating system of the computing node, probe code is injected through eBPF technology or dynamic link library. The probe registers callback hooks in the process scheduling interface, passively listens for lifecycle events such as task creation, execution, and blocking, and reads the CPU cycle counter to obtain a high-precision timestamp. At the same time, the probe reads the system status file and hardware management interface at a fixed sampling period to extract the real-time utilization of CPU, memory, and GPU. The collected data is packaged and written into a lock-free circular buffer. When the buffer is full, it is reported asynchronously in batches. The entire process does not require modification of the business source code, achieving non-intrusive awareness with extremely low overhead. This will not be elaborated further here.

[0024] It should be noted that, in this application, the lightweight probe refers to a software monitoring module deployed on computing nodes to collect task events and load data with extremely low overhead and non-intrusiveness.

[0025] In this embodiment, the non-intrusive collection of task lifecycle events and resource load to obtain the computing power event sequence can be achieved through the following steps: Non-intrusive capture of task state transition signals and underlying resource indicators to obtain raw perception data; The raw sensing data is analyzed for features to separate task lifecycle events and resource load within the corresponding execution time window; The task lifecycle events and the resource load are spliced ​​together in time sequence to obtain a computing power event sequence.

[0026] In practical implementation, firstly, an extended Berkeley packet filter technique is used to attach a tracing program at the kernel process scheduling function. When the task state changes, a hook function is triggered to capture the task state transition signal and record the hardware timestamp. Simultaneously, by reading the system virtual file system and hardware management interface, low-level resource indicators such as processor cycle count and memory page table status are obtained at a fixed sampling period. Both are packaged and written into a lock-free circular buffer to form raw sensing data. Then, the raw sensing data is read from the circular buffer and split according to the type identifier field in the data header. State transition data is mapped to task lifecycle events containing task identifiers and state types. For indicator data, a sliding window mean filtering algorithm is used to remove high-frequency noise in the low-level resource indicators, and the resource utilization within the execution time window is calculated, thereby separating the smoothed resource load. Finally, the occurrence timestamp of the task lifecycle event and the sampling timestamp of the resource load are extracted. Using the timestamp of the task lifecycle event as the baseline anchor point, a nearest neighbor time matching algorithm is used to find the sampling point with the smallest time difference in the resource load. The corresponding resource load value is appended to the attribute field of the task lifecycle event. The concatenated records are arranged in ascending order of time to generate a computing power event sequence.

[0027] It should be noted that, in this application, "task lifecycle event" refers to the various state transition nodes used to identify the computing task from creation, scheduling, execution to termination; "resource load" refers to the actual consumption of hardware computing resources by the computing task during execution; "non-intrusive acquisition" refers to the passive acquisition of system operating status data without modifying the business application source code or interfering with the normal business logic execution; "task state transition signal" is the instantaneous trigger pulse used to accurately capture the state switching of the computing task at the operating system scheduling level; "low-level resource indicators" refers to the physical operating status and utilization value of the hardware device at a specific sampling moment; "raw sensing data" refers to the unprocessed state signals and hardware indicators captured by the unified encapsulation probe; "feature parsing" is used to extract and classify state events and quantitative indicators with clear business semantics from the mixed raw data stream; and "computing power event sequence" is used to organize the task status and resource consumption in an orderly manner according to time sequence, providing a structured computing power domain time series input for cross-domain time sequence alignment and computing-power fusion simulation.

[0028] In step S2, a global tracking identifier is generated based on the task lifecycle event. The global tracking identifier is transmitted through a cross-node communication protocol to obtain the full-link call topology. The physical power consumption data is distributed according to the resource occupancy ratio in the full-link call topology to obtain the node power consumption sequence.

[0029] In this embodiment, the global tracking identifier can be generated based on the task lifecycle event in the following manner: Extract the task creation trigger signal from the task lifecycle events to generate network request features; The global tracking identifier is determined based on the network request characteristics.

[0030] In practical implementation, firstly, the type identifier field in the task lifecycle event set is traversed, and records with the status of task initialization are selected to extract the task creation trigger signal. The network socket descriptor bound to the task is obtained through system calls, and the network protocol type, destination Internet Protocol address, and port number are read. A message digest algorithm is used to perform a hash operation on the request payload to obtain a payload digest value. The above network attribute information is concatenated in a preset order to generate a network request feature. Then, the machine identifier and monotonically increasing sequence number of the current computing node are obtained, and a nanosecond-level timestamp is obtained by reading the hardware timestamp counter. The network request feature, machine identifier, sequence number, and timestamp are concatenated by bit operations with a fixed bit width to generate a 64-bit long integer value. Cyclic redundancy check is performed on this value to obtain a checksum, which is then appended to the end to determine the global tracking identifier containing routing features and time information.

[0031] It should be noted that in this application, the task creation trigger signal is a low-level system call interruption or hook callback event used to accurately capture the instantiation of a new computing task in a distributed computing power system; the network request feature is network layer attribute information such as protocol type, target address, and payload digest that characterizes the newly created task during cross-node communication; and the global tracking identifier is a unified digital credential that uniquely marks the identity of a complete business request flowing among all participating nodes in the distributed computing power cluster.

[0032] In this embodiment, the global tracking identifier is transmitted through a cross-node communication protocol to obtain the end-to-end call topology, which can be achieved through the following steps: Intercept packets along the network stack transmit / receive path of the cross-node communication protocol, and non-intrusively inject the global tracking identifier into the extended field of the protocol header; Based on the protocol parsing results of the extended fields during cross-node transmission, record the call offset between the source node and the destination node; Based on the call offset, the distributed call fragments are aggregated into a directed graph to generate a full-link call topology.

[0033] In practical implementation, firstly, an interceptor is attached to the transport layer send hook of the operating system's network protocol stack using extended Berkeley packet filtering technology. When an outbound packet is captured, the interceptor reads the global tracking identifier cached in the current thread context, serializes it into a byte stream, and writes it into a predefined custom metadata key-value pair in the application layer protocol header, completing the non-intrusive injection of the extended field. Then, the destination node intercepts the inbound packet at the receive hook of the network protocol stack, parses the protocol header to extract the global tracking identifier from the extended field to generate the protocol parsing result, and simultaneously reads the timestamp of the embedded packet from the source node and the current hardware timestamp of the destination node, calculates the difference between the two, and combines it with the current call stack depth information to record the call offset between the source and destination nodes. Finally, the topology aggregation service collects the distributed call fragments containing call offsets reported by each node and groups them using the global tracking identifier as the aggregation key. Within the group, directed edges are constructed using the source and destination nodes as vertices and the logical level in the call offset as the edge weight. A topological sorting algorithm is used to connect and detect loops in the directed edges, eliminating redundant paths. Finally, the directed graph is aggregated to generate a structured full-link call topology.

[0034] It should be noted that, in this application, the cross-node communication protocol refers to the communication rules that regulate data exchange and remote procedure calls between different computing nodes in a distributed computing power system; the network stack send / receive path refers to the logical processing channel in the operating system's network protocol stack where data packets are encapsulated and sent downwards from the application layer or decapsulated and received upwards from the network interface; the extended field refers to the data area reserved or defined in the header of the standard communication protocol; cross-node transmission refers to the complete flow process of a data packet from the computing node initiating the call to the computing node receiving the call via the physical network link; the protocol parsing result refers to the structured control information extracted by the receiving end after decapsulating the header of the arriving packet; and the source node refers to the node that identifies the node in the distributed call link. The starting computing device is the one that initiates a remote service request or data transmission task; the destination node is the terminal computing device that receives the remote service request and executes the specific computing task in the distributed call chain; the call offset is the relative difference between the source node initiating the call and the destination node receiving the call in a time series or logical hierarchy; the distributed call fragment is the record of the local interaction behavior and context information between the source and destination ends during a single cross-node remote call; directed graph aggregation is a graph theory operation that connects and merges multiple discrete local call relationships according to the call direction and logical order; the full-link call topology is a directed graph structure that depicts all nodes and their mutual call dependencies that a global business request flows through in the distributed computing power cluster.

[0035] In this embodiment, the physical power consumption data is distributed according to the resource occupancy ratio in the end-to-end call topology to obtain the node power consumption sequence, which can be achieved by the following steps: The instantaneous power consumption allocation value is determined by combining the total node power corresponding to the physical power consumption data with the resource occupancy ratio. The instantaneous power consumption allocation value is time-aligned based on the timestamp of the full-link call topology, and combined with the physical power consumption data, a node power consumption sequence is generated.

[0036] In practice, the process begins by extracting the resource occupancy ratio of each task within a node from the end-to-end call topology. The total node power is obtained by reading physical power consumption data collected from smart meters. A weighted proportional allocation model is then used to multiply the total node power by the resource occupancy ratio of a specific task, calculating the instantaneous power consumption allocation value for that task. The resource occupancy ratio is obtained by a weighted sum of processor cycle utilization and memory bandwidth utilization, with weight coefficients ranging from zero to one and summing to one. This weighted proportional allocation model, calibrated through historical benchmark tests, ensures that the allocation value is strictly proportional to the actual resource consumption. Next, the timestamps of task lifecycle events in the end-to-end call topology are extracted. A time grid is constructed based on the fixed sampling period of the physical power consumption data. A linear interpolation algorithm is used to map the instantaneous power consumption allocation value to the most recent sampling time point to complete timing alignment. Subsequently, the aligned instantaneous power consumption allocation value is cross-validated with the physical power consumption data at the corresponding time point. If the deviation exceeds a preset threshold, recalibration is triggered. Finally, the node power consumption sequence is generated by arranging the data in ascending order of time.

[0037] It should be noted that, in this application, the resource occupancy ratio refers to the weight of the actual consumption of underlying hardware computing resources by a specific computing task in a single node or cross-node link; physical power consumption data is the total amount of electrical energy or real-time power actually consumed by the hardware device of the computing node within a specific time window; instantaneous power consumption allocation value refers to the amount of electrical energy that a specific computing task should share in a very short time interval; time alignment refers to eliminating the clock deviation and sampling frequency difference between the computing event timestamp and the physical power consumption sampling timestamp; node power consumption sequence refers to recording the power consumption of each task on a single computing node in chronological order.

[0038] In step S3, the event synchronization deviation of cross-domain temporal nonlinearity is determined by the state transition point of the computing power event sequence and the power consumption mutation point of the node power consumption sequence, and the same-frequency computing power segment in the node power consumption sequence is corrected according to the event synchronization deviation to obtain the cooperative alignment sequence.

[0039] In this embodiment, determining the event synchronization deviation of cross-domain temporal nonlinearity through the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence can be achieved by the following steps: Extract the logical trigger time of the state transition point and the physical response time of the power consumption mutation point to determine the transient delay difference; Based on the transient delay difference, a cross-domain node clock drift sequence is constructed, and a piecewise polynomial is used to fit the clock drift sequence to obtain a timing compensation curve. The event synchronization deviation is determined based on the timing compensation curve.

[0040] In practical implementation, firstly, the computing power event sequence is traversed to identify state transition points and their timestamps are read as logical trigger times. First-order difference operations are applied to the node power consumption sequence, and data points with difference values ​​greater than a preset power consumption step threshold are selected as power consumption mutation points, and their timestamps are read as physical response times. The physical response time is subtracted from the logical trigger time to calculate the transient delay difference, thereby quantifying the actual lag time from logical trigger to physical response for a single event. Then, multiple transient delay differences are arranged in ascending order of logical trigger times to construct a cross-domain node clock drift sequence. The sequence is fitted using a piecewise polynomial with the least squares method, using a fixed time window as the segment boundary. The coefficients of a cubic polynomial are solved within each window. To ensure continuity at the window boundaries, a constraint of equal first-order derivatives is introduced for smooth transition, ultimately generating a continuous timing compensation curve to characterize the evolution of clock drift. Finally, the logical trigger time of each state transition point in the computing power event sequence is extracted and substituted into the polynomial equation of the timing compensation curve to calculate the theoretical clock drift compensation amount at that time. The theoretical clock drift compensation amount is then subtracted from the actual observed transient delay difference to obtain the residual time misalignment amount. The root mean square error of the residual time misalignment amount at all state transition points is calculated to finally determine the event synchronization deviation that characterizes the overall alignment accuracy.

[0041] It should be noted that, in this application, a state transition point refers to a discrete-time location that identifies a state switch of a computing task at the operating system or application layer; a power consumption mutation point refers to a physical time location that characterizes a significant step change in power consumption caused by drastic load changes in hardware devices; cross-domain timing nonlinearity refers to describing the non-constant proportional mapping relationship between the logical time in the computing power domain and the hardware time in the physical domain due to clock frequency differences and transmission delays; the logical trigger moment refers to the absolute time value recorded when the software issues a state switch command or captures a state change signal; the physical response moment refers to the absolute time value recorded when the hardware sensor detects a step change in power consumption caused by the state switch; and the transient delay difference is... The absolute time difference between the logical trigger moment and the physical response moment during a single state transition is quantified; the cross-domain node clock drift sequence refers to the data set reflecting the relative clock offset evolution between the computing power node and the physical acquisition device, which is recorded in chronological order as multiple transient delay differences; the piecewise polynomial refers to the mathematical model used to divide a complex nonlinear time series into multiple locally linear and low-order nonlinear intervals and approximate them using polynomial functions respectively; the timing compensation curve is a function graph characterizing the amount of time deviation compensation that changes continuously with time between the computing power domain and the physical domain; the event synchronization deviation refers to the residual time misalignment between the computing power event and the physical power consumption data after initial alignment.

[0042] In addition, in this embodiment, the timing compensation curve is obtained by fitting the clock drift sequence with a piecewise polynomial using the following steps: The clock drift sequence is divided into multiple steady-state intervals, and polynomial regression is performed on each steady-state interval to obtain local component curves; Boundary fusion is performed on the curves of each local component to obtain the time-series compensation curve.

[0043] In practice, the process begins by calculating the sliding window variance of the clock drift sequence. When the variance falls below a set threshold, it is marked as a steady-state interval. For each steady-state interval's discrete data points, a design matrix containing linear to cubic time variables is constructed. The normal equation of this design matrix is ​​solved using the least squares method to calculate the polynomial coefficients. These coefficients are then substituted into the cubic polynomial equation to generate local component curves characterizing the clock drift evolution within that interval. Next, the timestamps at the boundaries of adjacent steady-state intervals are extracted as fusion nodes. Overlapping windows are set on both sides of the fusion nodes, and the function values ​​of two adjacent local component curves within these windows are calculated. A cubic spline interpolation algorithm is used, with the continuity of the function values ​​and first derivatives as constraints, to construct a system of equations. The interpolation coefficients are then solved, and the obtained coefficients are used to replace the original function expression at the boundary, completing the boundary fusion of each local component curve. Finally, a globally smooth timing compensation curve is output.

[0044] It should be noted that, in this application, a steady-state interval refers to a time segment in which the variance of data fluctuation in a clock drift sequence is below a preset threshold and the trend of change remains relatively consistent; multinomial regression refers to a statistical modeling method that approximates the change law of the dependent variable by constructing a high-power combination of independent variables; local component curves refer to independent mathematical function images that reflect the change law of clock drift over time within a single steady-state interval; boundary fusion refers to a smoothing operation that eliminates the numerical jumps and derivative discontinuities caused by adjacent local component curves at the interval boundaries.

[0045] Preferably, in this embodiment, the same-frequency computing power segment in the node power consumption sequence is corrected according to the event synchronization deviation to obtain a cooperative alignment sequence, with reference to... Figure 2 As shown in the figure, this is a flowchart illustrating the process of determining the cooperative alignment sequence in some embodiments of this application. In this embodiment, determining the cooperative alignment sequence can be achieved using the following steps: In step S31, based on the basic execution frequency of the computing power task, spectral features are extracted from the node power consumption sequence to locate the same frequency computing power segment, and the timing compensation offset is determined in combination with the event synchronization deviation. In step S32, the timing compensation offset is applied to the corresponding sampling point to perform timing translation, thereby obtaining event transition features; In step S33, a cooperative alignment sequence is determined based on the event transition characteristics.

[0046] In practical implementation, firstly, the base execution frequency of the computing power task is extracted. Spectral features are extracted from the node power consumption sequence using Fast Fourier Transform, and frequency bands matching the base execution frequency are selected to locate the same-frequency computing power segment. The event synchronization deviation within the corresponding time window of this segment is read and input into a linear mapping model, multiplied by a sampling period conversion coefficient, to calculate the timing compensation offset in units of sampling points. The conversion coefficient ranges from zero to one and is calibrated by the ratio of the hardware sampling rate to the computing power clock frequency. Then, each sampling point within the same-frequency computing power segment is traversed, its original time index is read, and the timing compensation offset is added to calculate the new time index after translation. A cubic convolution interpolation algorithm is used, employing the power consumption values ​​of four adjacent original sampling points around the new time index for weighted calculation, to reconstruct the continuous power consumption signal after translation. The signal gradient change rate and curvature at the state transition boundary are extracted and encapsulated as event transition features to characterize the edge morphology after translation. Finally, the gradient change rate in the event transition features is read. If it exceeds the preset smoothing threshold, a Gaussian filtering operation is triggered to smooth the power consumption signal at the boundary, eliminating high-frequency glitches and noise introduced by time shift. The processed power consumption signal is bound to the original computing power event timestamp one by one to construct triplet data containing time, power consumption and task status. All triplets are arranged in ascending order of time to finally generate a structured collaborative alignment sequence.

[0047] It should be noted that in this application, the basic execution frequency refers to the inherent periodic frequency of the continuous operation of the computing power task; the same-frequency computing power segment refers to the power data segment used to filter out the power consumption sequence that is consistent with the execution frequency of a specific computing power task; the timing compensation offset refers to the precise time adjustment value used to correct the residual time misalignment between the computing power event and the physical power consumption sampling point; timing shift refers to the mathematical operation of moving the position of discrete data sampling points as a whole or locally on the time axis; the event transition feature is a characterization of the smooth transition form and gradient change law of the power consumption data at the boundary of the computing power task state switching after timing shift; and the cooperative alignment sequence is a structured data set that records the strict synchronization of computing power events and physical power consumption on the time axis after timing correction and feature smoothing.

[0048] In addition, in this embodiment, determining the cooperative alignment sequence based on the event transition characteristics can be achieved using the following steps: Extract the state transition parameters of the event transition features at the intersection of computing power segments, and determine the event transition alignment mapping relationship; Based on the event transition alignment mapping relationship, cross-node timing rearrangement is performed on the power consumption correction subsequence of each node to generate a cooperative alignment sequence.

[0049] In specific implementation, firstly, at the boundary of the computing power segment, the first and second derivatives of the event transition characteristics are calculated as state transition parameters to quantify the load step amplitude and rate. A sample set including logic switching and physical response time is constructed. A support vector regression model is used, with the state transition parameters as input and the time difference as output for training. The input is mapped to a high-dimensional space through a radial basis function to solve for the regression coefficients, generating an event transition alignment mapping relationship. Then, the local timestamps of the power consumption correction subsequences of each node are read and substituted into the event transition alignment mapping relationship to calculate the global standard time coordinates. For conflicts between data from multiple nodes under the same global coordinates, a merge sort algorithm is used to compare data priorities. High-priority node data is written to the target array first, and low-priority data is postponed. All subsequences are traversed to complete the reordering and splicing under the global coordinates, generating a cooperative alignment sequence.

[0050] It should be noted that, in this application, the boundary of computing power segments refers to the critical time position that marks the handover of control and release of resources between two adjacent independent computing tasks on the time axis; the state transition parameter is a physical index that quantifies the magnitude and rate of the step change in hardware load at the moment of handover of computing tasks; the event transition alignment mapping relationship is a mathematical function that characterizes the nonlinear correspondence between the logical switching time of computing power tasks and the physical power consumption response time; the node power consumption correction subsequence is a set of discrete power consumption data recorded within a single computing node after local time-series shifting and error compensation; cross-node time-series rearrangement is the operation of reordering and splicing the power consumption correction subsequences distributed on different physical nodes according to a unified global time base.

[0051] In step S4, the cooperative alignment sequence is fused and deduced to obtain the computing and power fusion features, and the compensation parameters of computing and power timing alignment are iteratively updated based on the computing and power fusion features.

[0052] In this embodiment, the fusion deduction of the cooperative alignment sequence to obtain the computational-electronic fusion features can be achieved through the following steps: The computing load and power consumption fluctuation features are extracted from the cooperative alignment sequence, a computing power consumption feature matrix is ​​constructed, and the spatiotemporal coupling correlation between the computing load and the power consumption fluctuation features is determined. Based on the spatiotemporal coupling correlation degree, the computational power feature matrix is ​​fused and deduced to obtain the computational power fusion feature.

[0053] In specific implementation, firstly, the collaborative alignment sequence is traversed, and the weighted sum of processor utilization and memory occupancy is extracted as the computing load. The sliding window variance of the power consumption signal is calculated as the power consumption fluctuation feature. The computing power feature matrix is ​​constructed by concatenating the data row by row over time. The Pearson correlation coefficient is used to calculate the linear correlation value between the two in the time dimension, and the spatial attenuation weight is calculated by combining the node topological distance. The spatiotemporal coupling correlation degree is determined by multiplying the time correlation value with the spatial weight. Then, the spatiotemporal coupling correlation degree is input into the Softmax function for normalization to generate a feature fusion weight vector. A weighted linear combination is performed on the computing power feature matrix to obtain the initial fusion vector, which is input into a multilayer perceptron model containing two fully connected layers and a ReLU activation function. The output of the hidden layer nodes is calculated through forward propagation, and after dimensionality reduction mapping of the output layer, the computing power fusion features representing the computing power collaborative state are finally extracted.

[0054] It should be noted that, in this application, fusion simulation refers to the operation of deep feature interaction and joint computation of multi-dimensional computing power and power consumption data under a unified spatiotemporal framework; computing power load is the proportion of logical resources consumed by computing nodes in processing computing tasks within a specific time window; power consumption fluctuation characteristics describe the dynamic fluctuation law and transient extreme value distribution of power consumption of hardware devices during operation; computing power feature matrix refers to a two-dimensional data array formed by structuring computing power load and power consumption fluctuation characteristics on a unified time axis; spatiotemporal coupling correlation degree refers to a comprehensive statistical index that quantifies the mutual influence and synergistic change intensity between the logical state of computing power and the evolution of physical energy consumption; computing power fusion characteristics are high-dimensional joint representation vectors reflecting the logical execution state of computing tasks and the energy consumption response state of physical hardware.

[0055] In specific implementation, in this application, the compensation parameters for computational timing alignment that are iteratively updated based on the computational-electrical fusion features can be determined in the following way: First, the computational-electrical fusion features are input into a preset mean squared error loss function to calculate the feature deviation value in the current state. The partial derivative of the feature deviation value with respect to the compensation parameters for the current computational-electrical timing alignment is calculated using the backpropagation algorithm to obtain the gradient vector representing the optimization direction. The gradient vector is multiplied by a preset learning rate to obtain the parameter update step size. This step size is subtracted from the compensation parameters for the current computational-electrical timing alignment to complete a single iteration. Then, it is determined whether the feature deviation value is lower than a preset convergence threshold. If not, the above process is repeated until the deviation value converges, and the final optimized compensation parameters for computational-electrical timing alignment are output. Further details are omitted here.

[0056] Therefore, in this application, the compensation parameters for computing power timing alignment can be iteratively updated based on the computing power fusion characteristics. Specifically, by determining the node power consumption sequence, a high-fidelity dynamic evolution characterization of physical energy consumption in the continuous time dimension can be obtained. This reveals the transient power response law and steady-state energy consumption baseline drift characteristics of the underlying hardware under complex computing power scheduling excitation. This provides a physically consistent original signal benchmark for subsequent removal of environmental noise interference and extraction of computing power coupling characteristics, effectively avoiding the loss of energy consumption information caused by discrete sampling, and ensuring that the global perception model accurately depicts and quantifies the real operating state of physical devices in a multi-dimensional spatiotemporal scale. By determining the cooperative alignment sequence, a strict spatiotemporal mapping relationship between computing power logic events and physical energy consumption signals under a unified time base can be obtained. This eliminates timing misalignment and phase delay caused by heterogeneous clock sources and sampling frequency differences, ensuring strict causal consistency of computing and power fusion characteristics on the time axis. This enables the global state perception to accurately track the transient coupling trajectory of computing power task scheduling and hardware energy consumption fluctuations, thereby providing high-confidence timing alignment support for the dynamic iteration of compensation parameters and improving the accuracy of global state perception of the computing power network in complex interactive environments.

[0057] In summary, the technical solution adopted in this application can achieve precise collaborative updates of zero-intrusion computing power data collaborative perception and time alignment, thereby improving the accuracy of the global perception state of the computing power network after complex computing power interactions.

[0058] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0060] It should also be noted that 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. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A zero-intrusion collaborative sensing and time alignment method for computer data, characterized in that, The method includes: Based on the deployed lightweight probe, non-intrusive collection of task lifecycle events and resource load is performed to obtain computing power event sequences; A global tracking identifier is generated based on the task lifecycle events. The global tracking identifier is transmitted through a cross-node communication protocol to obtain the end-to-end call topology. The physical power consumption data is distributed according to the resource occupancy ratio in the end-to-end call topology to obtain the node power consumption sequence. The event synchronization deviation of cross-domain temporal nonlinearity is determined by the state transition point of the computing power event sequence and the power consumption mutation point of the node power consumption sequence, and the same-frequency computing power segment in the node power consumption sequence is corrected according to the event synchronization deviation to obtain the cooperative alignment sequence. The collaborative alignment sequence is fused and deduced to obtain computational-electrical fusion characteristics, and the compensation parameters for computational-electrical timing alignment are iteratively updated based on the computational-electrical fusion characteristics.

2. The zero-intrusion computer data collaborative sensing and time alignment method as described in claim 1, characterized in that, Non-intrusive collection of task lifecycle events and resource load yields a computing power event sequence, specifically including: Non-intrusive capture of task state transition signals and underlying resource indicators to obtain raw perception data; The raw sensing data is analyzed for features to separate task lifecycle events and resource load within the corresponding execution time window; The task lifecycle events and the resource load are spliced ​​together in time sequence to obtain a computing power event sequence.

3. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 1, characterized in that, Generating a global tracking identifier based on the task lifecycle events specifically includes: Extract the task creation trigger signal from the task lifecycle events to generate network request features; The global tracking identifier is determined based on the network request characteristics.

4. The zero-intrusion computer data collaborative sensing and time alignment method as described in claim 1, characterized in that, The global tracking identifier is transmitted via a cross-node communication protocol to obtain the full-link call topology, which specifically includes: Intercept packets along the network stack transmit / receive path of the cross-node communication protocol, and non-intrusively inject the global tracking identifier into the extended field of the protocol header; Based on the protocol parsing results of the extended fields during cross-node transmission, record the call offset between the source node and the destination node; Based on the call offset, the distributed call fragments are aggregated into a directed graph to generate a full-link call topology.

5. The zero-intrusion computer data collaborative sensing and time alignment method as described in claim 1, characterized in that, The physical power consumption data is distributed according to the resource occupancy ratio in the full-link call topology, resulting in a node power consumption sequence that specifically includes: The instantaneous power consumption allocation value is determined by combining the total node power corresponding to the physical power consumption data with the resource occupancy ratio. The instantaneous power consumption allocation value is time-aligned based on the timestamp of the full-link call topology, and combined with the physical power consumption data, a node power consumption sequence is generated.

6. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 1, characterized in that, Determining the cross-domain temporal nonlinearity event synchronization deviation by using the state transition points of the computing power event sequence and the power consumption mutation points of the node power consumption sequence specifically includes: Extract the logical trigger time of the state transition point and the physical response time of the power consumption mutation point to determine the transient delay difference; Based on the transient delay difference, a cross-domain node clock drift sequence is constructed, and a piecewise polynomial is used to fit the clock drift sequence to obtain a timing compensation curve. The event synchronization deviation is determined based on the timing compensation curve.

7. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 6, characterized in that, The timing compensation curve is obtained by fitting the clock drift sequence using a piecewise polynomial, specifically including: The clock drift sequence is divided into multiple steady-state intervals, and polynomial regression is performed on each steady-state interval to obtain local component curves; Boundary fusion is performed on the curves of each local component to obtain the time-series compensation curve.

8. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 1, characterized in that, The co-alignment sequence is obtained by correcting the same-frequency computing power segments in the node power consumption sequence based on the event synchronization deviation, specifically including: Based on the basic execution frequency of the computing power task, spectral features are extracted from the node power consumption sequence to locate the computing power segment with the same frequency, and the timing compensation offset is determined in combination with the event synchronization deviation. The timing compensation offset is applied to the corresponding sampling point to perform timing translation, thereby obtaining the event transition characteristics; The collaborative alignment sequence is determined based on the event transition characteristics.

9. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 8, characterized in that, Determining the cooperative alignment sequence based on the event transition characteristics specifically includes: Extract the state transition parameters of the event transition features at the intersection of computing power segments, and determine the event transition alignment mapping relationship; Based on the event transition alignment mapping relationship, cross-node timing rearrangement is performed on the power consumption correction subsequence of each node to generate a cooperative alignment sequence.

10. The zero-intrusion computational data collaborative sensing and time alignment method as described in claim 1, characterized in that, The fusion and deduction of the cooperative alignment sequence yields the following specific computational-electronic fusion features: The computing load and power consumption fluctuation features are extracted from the cooperative alignment sequence, a computing power consumption feature matrix is ​​constructed, and the spatiotemporal coupling correlation between the computing load and the power consumption fluctuation features is determined. Based on the spatiotemporal coupling correlation degree, the computational power feature matrix is ​​fused and deduced to obtain the computational power fusion feature.