A data center multi-sensor integrated measurement method and device

Abstract

The application discloses a data center multi-sensor integrated measurement method and device, generates a sensor mismatch data pool by collecting heterogeneous performance monitoring sensor data and performing bias intensity grading; extracts a measurement offset feature family by performing feature scanning on the mismatch data, and constructs an offset correlation tensor to locate a key mismatch area; establishes a coordinated monitoring network in the mismatch area, captures a performance interference signal to form an interference calibration map, and identifies an active offset source; performs synchronous coordination by using the active offset source, generates a coordination vector by performing time delay analysis and spatial positioning correction; converts the coordination vector into a measurement probability transfer chain, obtains a superimposed probability state by performing multi-stage probability superposition, determines a measurement decision node by performing probability aggregation analysis; performs probability weight distribution on the decision node to construct a probability configuration network, generates a priority sequence, and outputs a coordination measurement scheme, and the accuracy and reliability of a multi-sensor measurement system are improved.

Description

Technical Field

[0001] This invention relates to the field of data center performance monitoring technology, and in particular to a multi-sensor integrated measurement method and device for data centers. Background Technology

[0002] As the core of information infrastructure, data centers require real-time monitoring of multiple parameters, including the room environment, server hardware status, system performance, and service operation. Currently, data centers commonly deploy environmental monitoring equipment such as temperature and humidity sensors, voltage and current sensors, wind speed sensors, and power consumption sensors. They are also equipped with system performance sensors such as CPU performance monitors, memory monitors, network traffic monitors, and I / O performance monitors. These monitoring devices come from different manufacturers and use different communication protocols and data formats, forming a complex heterogeneous monitoring network.

[0003] However, existing data center monitoring systems suffer from numerous problems. Various monitoring sensors operate independently, lacking unified data standards and coordination mechanisms, leading to issues such as asynchronous timing, spatial mismatch, and inconsistent unit of measurement in the collected data. When measurement results from multiple sensors contradict each other, it is difficult to determine which data is more reliable. Furthermore, the electromagnetic environment generated by densely packed server racks, airflow disturbances caused by air conditioning systems, measurement drift due to equipment aging, performance interference from resource contention between services, response fluctuations caused by network congestion, and performance degradation due to service aging all affect the measurement accuracy of sensors. Traditional monitoring solutions cannot effectively identify and eliminate these interfering factors and lack a quantitative evaluation mechanism for measurement reliability, resulting in frequent false alarms and impacting the operational efficiency of the data center. Summary of the Invention

[0004] This invention discloses a multi-sensor integrated measurement method and apparatus for data centers. The method aims to construct an offset correlation tensor to locate key mismatch regions by classifying deviation intensity and extracting mismatch features from heterogeneous performance monitoring sensor data; establishing a sensor coordination monitoring network to capture performance interference signals and generate an interference calibration map to identify active offset sources; achieving sensor spatial positioning correction using synchronization coordination and time delay analysis; converting the coordination results into a measurement probability transfer chain, and determining measurement decision nodes through multi-level probability superposition and aggregation analysis; finally, generating a priority sequence based on a probability configuration network and outputting an optimized multi-sensor coordinated measurement scheme to achieve high-precision coordinated measurement of heterogeneous sensor networks.

[0005] The first aspect of this invention proposes a multi-sensor integrated measurement method for data centers, comprising the following steps:

[0006] Collect heterogeneous sensor data from each monitoring node in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool.

[0007] The sensor mismatch data pool is scanned to extract the measurement offset feature family. The sensor mismatch data pool and the measurement offset feature family are correlated to obtain the offset correlation tensor. The key mismatch region is located based on the offset correlation tensor.

[0008] A sensor coordination monitoring network is established within the critical mismatch region. Electromagnetic interference signals between sensors are captured through the sensor coordination monitoring network and used as a calibration benchmark to form an interference calibration map. Time-series correlation analysis is performed on the interference calibration map and the measurement offset feature group to obtain active offset sources.

[0009] The active offset source is used to perform synchronous coordination on the key mismatch region to obtain a coordination sequence. The coordination sequence is then subjected to time delay analysis to obtain the spatial propagation time difference. Based on the spatial propagation time difference, sensor spatial positioning correction is performed to generate a coordination vector.

[0010] The coordination vector is converted into a measurement probability transfer chain, and the measurement probability transfer chain is subjected to multi-level probability superposition processing to obtain the superposition probability state. Based on the superposition probability state, probability aggregation analysis is performed to determine the measurement decision node.

[0011] A probability configuration network is constructed by assigning probability weights to the measurement decision nodes, a priority sequence is generated based on the probability configuration network, and a multi-sensor coordinated measurement scheme is output according to the priority sequence.

[0012] A second aspect of the present invention provides a multi-sensor integrated measurement device for data centers, comprising:

[0013] The data acquisition module is used to collect heterogeneous sensor data from various monitoring nodes in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool.

[0014] The offset analysis module is used to perform feature scanning on the sensor mismatch data pool to extract the measurement offset feature family, perform correlation matching between the sensor mismatch data pool and the measurement offset feature family to obtain the offset correlation tensor, and locate the key mismatch region based on the offset correlation tensor.

[0015] The calibration monitoring module is used to establish a sensor coordination monitoring network within the critical mismatch area, capture electromagnetic interference signals between sensors through the sensor coordination monitoring network as a calibration benchmark to form an interference calibration map, and perform time-series correlation analysis on the interference calibration map and the measurement offset feature group to obtain active offset sources.

[0016] The coordination processing module is used to perform synchronous coordination on the key mismatch area using the active offset source to obtain a coordination sequence, perform time delay analysis on the coordination sequence to obtain the spatial propagation time difference, and perform sensor spatial positioning correction based on the spatial propagation time difference to generate a coordination vector.

[0017] The probability calculation module is used to convert the coordination vector into a measurement probability transmission chain, perform multi-level probability superposition processing on the measurement probability transmission chain to obtain the superposition probability state, and perform probability aggregation analysis based on the superposition probability state to determine the measurement decision node.

[0018] The decision output module is used to allocate probability weights to the measurement decision nodes to construct a probability configuration network, generate a priority sequence based on the probability configuration network, and output a multi-sensor coordinated measurement scheme according to the priority sequence.

[0019] The beneficial effects of this invention are reflected in the following points: 1. By classifying and processing the deviation intensity of heterogeneous monitoring sensor data and extracting measurement offset feature groups, a five-level deviation classification strategy is adopted to differentiate sensor data of different accuracy levels, effectively eliminating measurement noise and time synchronization problems from sensors of different sources. Offset feature groups are extracted using feature scanning technology, an offset correlation tensor is constructed, and resonant amplification is performed using an offset resonant cavity, accurately locating key mismatch areas within the data center, solving the technical problem that traditional methods cannot effectively handle multi-source data contradictions. 2. By establishing a sensor coordinated monitoring network to capture performance interference signals, interference signal cancellation window identification and purification gain coefficient evaluation techniques are used to generate an interference calibration map, achieving accurate identification of active offset sources. The offset sources are quantified and classified using the comprehensive scoring mechanism of the activity evaluation framework. Spatial propagation time difference is established and a coordination vector is generated through synchronization coordination and time delay analysis techniques, realizing active identification and systematic correction of performance interference sources. 3. By constructing a measurement probability transfer chain and using probability standing wave technology, stable standing wave nodes are generated by the coherent superposition of antiphase probability pulses and harmonic components. Deterministic probability components are extracted and phase superposition is performed to generate superimposed probability states. By combining the construction of probabilistic potential energy fields and the evaluation of stability index, highly stable measurement decision nodes are selected. A probabilistic configuration network is constructed through a probability weight allocation formula, generating a dynamic priority sequence and outputting an adaptive coordinated measurement scheme. This realizes the technical upgrade of the monitoring system from static configuration to dynamic optimization.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0022] Unless otherwise specified, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0023] Figure 1 This is a flowchart illustrating a multi-sensor integrated measurement method for data centers according to the present invention.

[0024] Figure 2 This is a structural block diagram of a data center multi-sensor integrated measurement device according to the present invention. Detailed Implementation

[0025] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0026] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] The technical solutions of the embodiments of this application will be described below.

[0029] like Figure 1 As shown, this embodiment of the invention provides a multi-sensor integrated measurement method for data centers, including the following steps S110-S160:

[0030] Step S110: Collect heterogeneous sensor data from each monitoring node in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool.

[0031] Specifically, heterogeneous sensor data is collected from various monitoring nodes within the data center. Multiple types of sensors, including temperature sensors, humidity sensors, voltage and current sensors, wind speed sensors, and power consumption sensors, are deployed at key locations within the data center, forming a heterogeneous sensor network. Temperature sensors monitor the ambient temperature of the server room and the server CPU temperature, with a measurement range of 0°C to 80°C and a sampling frequency of 10Hz. Humidity sensors monitor the ambient humidity of the server room, with a measurement range of 0-100%RH and a sampling frequency of 5Hz. Voltage and current sensors monitor the UPS power supply status and server power consumption, with a voltage range of 180V-240V and a current range of 0-50A, and a sampling frequency of 50Hz. Wind speed sensors monitor the airflow velocity of the cooling system, with a measurement range of 0.1-20m / s and a sampling frequency of 20Hz. Power consumption sensors monitor the real-time power consumption of each rack, with a measurement range of 100W-10kW and a sampling frequency of 100Hz. All sensors transmit data through a unified data acquisition network, employing standard Ethernet and SNMP management protocols to ensure the real-time performance and reliability of data transmission. Configure multi-channel synchronous acquisition function in the data acquisition system, and allocate independent acquisition channels and buffer space for each sensor.

[0032] Sensor data is graded by deviation intensity to generate graded deviation queues. The deviation degree between the current measurement value and the historical baseline value of each sensor is calculated using the formula D = |X(t) - X_base| / X_base × 100%, where D is the deviation percentage, X(t) is the measured value at time t, and X_base is the baseline value. The baseline value is obtained by the moving average of the data from the previous 30 minutes, excluding the influence of abnormal abrupt changes. Based on the deviation percentage, the data is divided into five intensity levels: 0-5% deviation is normal, 5-10% is slight, 10-20% is moderate, 20-35% is severe, and greater than 35% is extreme. An independent data queue is created for each deviation level, using a first-in-first-out circular buffer structure with a capacity of 1000 data points. Real-time acquired sensor data is distributed to the corresponding queue according to the calculated deviation level. Each data item includes the sensor ID, timestamp, raw value, deviation value, and deviation level marker. For example, at 15:23:45, the CPU temperature sensor T01 measured 68.5°C, with a reference value of 62.3°C, a deviation rate of 9.9%, and was classified into the slight deviation queue. At the same time, the power consumption sensor P03 measured 3.2kW, with a reference value of 3.0kW, a deviation rate of 6.7%, and was also classified into the slight deviation queue. The real-time data distribution of each queue was statistically analyzed: the normal level queue accounted for an average of 60%, the slight level 25%, the moderate level 10%, the severe level 4%, and the extreme level 1%. When a queue is full, a sliding window method is used to remove the oldest data to maintain dynamic updates to the queues.

[0033] The graded deviation queues are coordinated to obtain a sensor mismatch data pool. Differentiated coordination strategies are employed for the five deviation level queues to achieve graded filtering and fusion of data. Data from the normal level queue is forwarded directly without additional processing, maintaining the original sampling rate and numerical accuracy. The slightly problematic queue uses a simple moving average filter to smooth short-term fluctuations while preserving trend changes. The moderate level queue uses median filtering to effectively remove impulse interference and outliers. The severe level queue undergoes adaptive filtering, reducing noise impact through state prediction and measurement updates. Data from the extreme level queue is marked as anomaly, temporarily isolated, but records are retained for fault analysis. The physical correlation between different sensors is analyzed; rack power consumption should increase accordingly when CPU temperature rises, and cooling system fan speed should increase when ambient temperature rises. Statistical analysis methods are used to calculate the correlation coefficient between sensor pairs to evaluate data consistency and rationality. When physically positively correlated sensor pairs show negative or zero correlation, a mismatch is identified. For example, a server CPU temperature sensor shows a temperature rise to 75°C, but the corresponding power sensor shows a power consumption decrease to 2.1kW, clearly violating the normal pattern of increased power consumption under high CPU load, and is marked as a mismatch. All mismatched sensor data is aggregated into a sensor mismatch data pool. The data pool adopts a multidimensional data table structure, with columns including timestamp, sensor ID, sensor type, raw value, processed value, deviation level, mismatch type, and a list of related sensors.

[0034] Step S120: Perform feature scanning on the sensor mismatch data pool to extract the measurement offset feature family, perform correlation matching between the sensor mismatch data pool and the measurement offset feature family to obtain the offset correlation tensor, and locate the key mismatch region based on the offset correlation tensor.

[0035] In some embodiments, the step of performing feature scanning on the sensor mismatch data pool to extract a measurement offset feature cluster includes: performing phase decomposition on the sensor mismatch data pool to obtain orthogonal components; identifying phase jump points in the orthogonal components to form a phase anomaly sequence; mapping the phase anomaly sequence back to the time domain to extract time domain abrupt change features; and clustering and grouping based on the time domain abrupt change features to generate a measurement offset feature cluster.

[0036] Phase decomposition is performed on the sensor mismatch data pool to obtain orthogonal components. Signal decomposition techniques are used to convert the real-valued signals in the sensor mismatch data pool into composite signals containing phase information. Instantaneous amplitude and instantaneous phase are extracted from the composite signal. The instantaneous amplitude reflects the load changes of the data center equipment, and the instantaneous phase contains the frequency characteristics of the equipment operation. In-phase and quadrature components are separated to achieve orthogonal decomposition of the signal. The quadrature components are particularly sensitive to phase changes and can capture subtle anomalies in the operation of the data center equipment. Normalization is performed on the quadrature components to eliminate the influence of amplitude changes on phase analysis, allowing the analysis to focus on the phase characteristics themselves. Instantaneous frequency changes of the quadrature components are analyzed to identify time periods of abnormal frequency fluctuations. These periods typically correspond to abnormal operation of the data center equipment or environmental interference. For example, after signal decomposition, the quadrature components of the server CPU temperature sensor show sharp jumps at certain times, indicating a sudden change in the temperature control system; the quadrature components of the network traffic sensor show high-frequency oscillations, with oscillation frequencies far exceeding the normal rate of change, suggesting network congestion or attacks. The orthogonal components of all sensors are processed in parallel to generate a multi-channel orthogonal component matrix. The rows of the matrix correspond to different sensors, and the columns correspond to the time sampling sequence. The distribution characteristics of the orthogonal components are statistically analyzed. Under normal circumstances, the orthogonal components exhibit smooth and continuous changes, while the orthogonal components of mismatched data contain multiple discontinuous jumps and abnormal fluctuations, with a significantly increased standard deviation.

[0037] Phase jump points are identified in orthogonal components to form a phase anomaly sequence. The location of phase abrupt changes in orthogonal components is detected through differential operations; a phase jump point is identified when the absolute value of the difference exceeds a set threshold. The time interval distribution of jump points is analyzed, and consecutive jumps with short time intervals are grouped into jump clusters, reflecting concentrated periods of anomaly in data center equipment. Characteristic parameters of each jump cluster are statistically analyzed, including the number of jumps, total duration, average jump amplitude, and jump frequency; these parameters characterize the severity of the equipment anomaly. The periodicity of jumps is detected using an autocorrelation function; a significant peak in the autocorrelation function at a specific delay value indicates a regular anomaly pattern with a corresponding period. For non-periodic jumps, information entropy is used to measure the randomness of the jump distribution; a higher entropy value indicates a more irregular jump. For example, sensors in a data center cooling system detect multiple phase jump points within the observation window; some jumps exhibit quasi-periodic characteristics with relatively fixed jump intervals, corresponding to the cooling cycle. In contrast, the jump points of UPS power supply sensors are completely randomly distributed without a clear temporal pattern. All identified transition points are arranged chronologically to form a phase anomaly sequence. Each element in the sequence records the time and magnitude of the transition. The phase anomaly sequence is assigned a classification label, categorized into periodic, random, and mixed types based on the transition pattern, and into mild, moderate, and severe levels based on the transition intensity.

[0038] Phase anomaly sequences are mapped back to the time domain to extract time-domain abrupt change features. The phase anomaly sequences are converted into time-domain signal representations through inverse phase reconstruction, preserving the time-domain representation of phase jumps. Abrupt change feature points are identified in the reconstructed signal, including rising edges, falling edges, spikes, and step positions, corresponding to actual operational events of data center equipment. The time-domain parameters of the abrupt changes are analyzed: rise time is defined as the transition time from low to high level, fall time is the reverse process, abrupt change amplitude is the absolute value of the signal change, and the abrupt change slope reflects the severity of the change. Abrupt changes are classified according to time-domain parameters: rapid abrupt changes typically correspond to equipment switching actions or fault impacts, while slow abrupt changes correspond to gradual changes in system parameters or the cumulative effects of environmental factors. Energy characteristics of the abrupt change process are extracted, and the signal energy and energy concentration within the abrupt change interval are calculated. High energy concentration indicates severe and localized abrupt changes. For example, a phase jump mapped to the time domain may manifest as a polarity reversal of server load, producing a step-type abrupt change; a linear phase change mapped to the modulation effect of network traffic, producing a frequency-modulated abrupt change feature. The extracted temporal mutation features are organized into feature vectors. Each mutation event is fully described by a multi-dimensional vector, and the vector elements include time location, mutation type, magnitude parameter, and duration.

[0039] Measurement offset feature clusters are generated based on temporal abrupt change features. All extracted temporal abrupt change feature vectors are input into a density clustering method, which automatically identifies cluster centers and boundaries based on the distribution density of feature vectors in multidimensional space. Appropriate neighborhood radii and minimum number of points are set to ensure that the clustering results are neither overly dispersed nor excessively merged. Based on the similarity measure between feature vectors, abrupt changes with similar characteristics are grouped into the same cluster, each cluster representing a typical data center offset pattern. The characteristics of each cluster are statistically analyzed, including the cluster centroid, standard deviation, and number of members; the centroid vector represents the typical characteristics of the cluster. Clustering quality indicators are used to evaluate the clustering effect, ensuring consistency within clusters and diversity between clusters. Several major measurement offset feature clusters are identified, including fast failure cluster, step load cluster, slow drift cluster, periodic fluctuation cluster, and random noise cluster, each with unique data center operating characteristic patterns. Identification rules and discrimination criteria are formulated for each feature cluster to facilitate rapid classification of new data. For example, the fast failure family is characterized by extremely short rise time, short duration, and large amplitude, corresponding to sudden equipment failure; the slow drift family is characterized by slow change, unidirectional change, and long duration, corresponding to equipment aging. A complete library of measurement offset feature families is output, with each family containing detailed feature descriptions, identification templates, and typical samples.

[0040] In some embodiments, the step of performing correlation matching between the sensor mismatch data pool and the measurement offset feature group to obtain the offset correlation tensor includes: identifying data complementarity gaps based on the sensor mismatch data pool; obtaining a compensation sequence of the measurement offset feature group in the data complementarity gaps; converting the compensation sequence into a tensor reconstruction driver; and establishing an offset correlation tensor using the tensor reconstruction driver.

[0041] Identifying data complementarity gaps based on sensor mismatch data pools. The complete time series of the sensor mismatch data pool is scanned to detect data missingness and invalid data distribution for each sensor. A data integrity matrix is ​​constructed to mark data validity; matrix elements are represented binaryly, with valid data marked as one and missing or invalid data marked as zero. Data coverage at each time point is calculated, i.e., the proportion of valid sensors to the total number of sensors; periods with low coverage indicate sparse data. Statistical analysis methods are used to evaluate the correlation between sensor pairs, assessing data redundancy and complementarity. When a sensor's data is missing, the ability of other sensors to provide supplementary information is evaluated; highly correlated sensor pairs exhibit strong complementarity. Continuous data missing intervals are identified, with longer-lasting missing intervals defined as data complementarity gaps. For example, during a certain period, a data center temperature sensor group may partially fail, but the power consumption sensor group may function normally; their physical correlation allows inference of temperature trends from power consumption data. All identified complementarity gaps are statistically analyzed, recording the time range of the gap, the sensors involved, and available complementary sensors.

[0042] Compensation sequences for measurement offset feature families are obtained from data complementarity gaps. For each data complementarity gap, the data characteristics before and after the gap are analyzed to determine the data's changing trend and characteristic pattern. Sequence segments matching the gap boundary characteristics are searched from the measurement offset feature family library, with matching criteria including consistency in feature type, changing trend, and numerical range. The best-matching feature sequence is extracted as a candidate compensation sequence, with a sequence length corresponding to the gap length and including an appropriate transition interval. Parameter adjustments are made to the compensation sequence, including amplitude scaling, phase alignment, and trend correction, to ensure a smooth transition between the compensation sequence and the original data. Interpolation techniques are used to process the boundaries of the compensation sequence to avoid discontinuous jumps at the gap edges. For example, for a data gap in a data center CPU temperature sensor, if there is an upward trend before and after the gap, a corresponding upward sequence is selected from the drift feature family for compensation. A set of compensation sequences is generated, with each compensation scheme including detailed parameter configurations and applicable conditions.

[0043] The compensation sequence is converted into a tensor reconstruction driver. The acquired compensation sequence is encoded into a control instruction set for tensor reconstruction, with each instruction specifying the tensor's filling position and value. A reconstruction driver function is designed to integrate the original data, compensation sequence, and interpolation results, achieving smooth data fusion through weighted combination. Weight coefficients for each data source are determined, with the original data having the highest weight, followed by the compensation sequence, and the interpolated data having the lowest. This weight allocation reflects the differences in data reliability. The driver function is discretized into an operation matrix, with each row corresponding to a specific reconstruction operation, including the target sensor, time interval, data source, and processing parameters. The execution order of reconstruction operations is arranged according to the severity of data missing information, prioritizing completely missing positions and processing partially missing positions later. For example, after executing a data processing instruction, the value at a specific position in the tensor represents the degree of correlation between the corresponding service node and a specific performance characteristic at that moment. A complete driver instruction sequence is generated, supporting both batch and incremental execution modes. The execution conditions and expected effects of each driver instruction are recorded for easy tracking of the reconstruction process.

[0044] Tensor reconstruction is used to establish an offset correlation tensor. A three-dimensional tensor with all zeros is initialized as the basic framework, with the tensor dimensions corresponding to the number of sensors, the number of feature groups, and the time duration. Filling operations are executed sequentially according to the reconstruction driving instructions, updating specific elements in the tensor with each operation. The correlation strength value R = exp(-d² / 2σ²), where d is the feature distance and σ is the bandwidth parameter, achieving distance-based smooth decay. The obtained correlation strength is written to a specified position in the tensor, while smoothing is applied to neighboring positions to maintain local continuity. Trilinear interpolation is used to fill undefined regions in the tensor, ensuring that the tensor has reasonable values ​​at all positions. For example, after executing a certain driving instruction, the value at a specific position in the tensor represents the degree of correlation between the corresponding sensor and a specific feature at that moment. The reconstructed tensor is normalized, including boundary condition processing, outlier correction, and numerical range normalization. Singular value decomposition is used to analyze the structural characteristics of the tensor, extracting the main singular vectors and singular values. Components with a cumulative contribution rate reaching a set threshold are retained to achieve effective tensor compression. The final offset correlation tensor is output, which contains complete correlation information and structural feature description.

[0045] In some embodiments, locating the critical mismatch region based on the offset correlation tensor includes: identifying an oscillation mode in the offset correlation tensor; constructing an offset resonant cavity based on the oscillation mode; performing resonant amplification processing on the offset correlation tensor through the offset resonant cavity to generate an amplified offset signal; and performing spatial anchoring processing on the amplified offset signal to form the critical mismatch region.

[0046] Identify oscillation modes in the offset correlation tensor. Perform spectral analysis on the offset correlation tensor along the time dimension, and extract the amplitude and phase information of each frequency component using Fast Fourier Transform. Identify the dominant frequency components in the spectrum, which correspond to the inherent operating cycle or periodic anomaly sources of data center equipment. Analyze the phase relationship of different sensors at the same frequency; consistent phase indicates a synchronous anomaly mode, while opposite phase indicates an inverse correlation mode. Analyze the time-frequency evolution characteristics of the oscillations using wavelet transform to identify the generation, development, and disappearance processes of the oscillations. Statistically analyze the spatial distribution of oscillation modes to determine which sensors are involved in the oscillations and which remain stable. For example, synchronous oscillations detected by multiple CPU temperature and power consumption sensors in a rack cluster of a data center, with a long duration and gradually increasing amplitude, indicate a possible cooling system failure. Extract characteristic parameters of the oscillation modes, including center frequency, bandwidth, oscillation amplitude, spatial range, and time span. Classify the oscillation modes, distinguishing between local equipment oscillations and global environmental oscillations, and between periodic and transient oscillations.

[0047] An offset resonator is constructed based on oscillation modes. The working principle of the offset resonator is similar to that of a physical resonator, but implemented through a digital algorithm: when the frequency of the input signal matches the natural frequency of the resonator, the signal is significantly amplified; signals deviating from this frequency are attenuated, thus achieving frequency-selective enhancement. The detected main anomalous frequency is set as the resonant frequency of the offset resonator. A smaller damping ratio is chosen to achieve a high quality factor, which is numerically equal to the ratio of the resonant frequency to the half-power bandwidth, determining the selectivity and amplification capability of the resonator. The frequency response characteristics of the offset resonator are determined; the passband width is set according to the distribution range of the oscillation frequencies, and the passband gain is configured according to the amplification requirements of the anomalous signal. A digital implementation scheme for the offset resonator is designed, using digital filtering technology to convert the continuous frequency response into discrete-time processing, and a mathematical model of the resonator is realized through difference equations. The resonator parameters are adjusted to achieve optimal matching with the detected oscillation modes; the matching degree is evaluated through frequency deviation and spatial correlation. For example, to address the 0.05Hz slow oscillation anomaly detected by the power consumption sensor of a rack assembly, an offset resonant cavity with a resonant frequency of 0.05Hz was designed to achieve a 15-fold signal amplification at this frequency, while the amplification factor drops to less than 2 times at 0.1Hz.

[0048] An amplified offset signal is generated by resonant amplification of the offset correlation tensor using an offset resonant cavity. The offset correlation tensor data is input into the resonant cavity, and each data point is frequency-selectively amplified. At the resonant frequency, the signal achieves maximum amplification, with the amplification factor equal to the quality factor; signals deviating from the resonant frequency have a smaller amplification factor. The time series of the tensor is processed point-by-point, and the output signal is the product of the input signal and the frequency response function. Nonlinear effects are considered during processing; when the input signal is too strong, the amplification factor saturates to avoid distortion of the output signal. The changes in each sensor signal after resonant amplification are recorded; strong oscillation signals are significantly enhanced, while weak background noise is relatively suppressed. For example, the temperature oscillation signal of a server rack in a data center is amplified several times after resonant amplification, while the random electromagnetic interference component remains essentially unchanged, resulting in a significantly improved signal-to-noise ratio. An amplified offset signal matrix is ​​generated, with matrix elements representing the amplified signal values ​​of each sensor at each time step.

[0049] Spatial anchoring processing is performed on the amplified offset signals to form critical mismatch regions. The amplified offset signals are mapped from the sensor number space to the actual three-dimensional physical space of the data center, assigning each sensor's signal value its installation coordinates within the server room. Spatial interpolation is performed between sensor locations to generate a continuous offset field distribution; the interpolation method uses inverse distance weighting to ensure a smooth transition. A threshold criterion for mismatch determination is set, adaptively determined based on the statistical characteristics of the offset field, typically the mean plus a certain number of standard deviations. Regions exceeding the threshold in the offset field are identified; these regions correspond to potential mismatch locations within the data center. Connectivity analysis is used to aggregate discrete mismatch points into continuous mismatch regions, considering the spatial adjacency relationships of racks, aisles, and hot / cold aisles. The geometric properties of each mismatch region are calculated, including volume, shape characteristics, centroid location, and principal axis orientation. Critical mismatch regions are screened based on their volume and maximum offset intensity; regions with large volumes and high offset intensity require special attention. For example, a critical mismatch region was identified in the 5th row of racks in Area A of the data center. This region contains multiple abnormal sensors with offset intensities far exceeding other areas, potentially indicating a localized cooling system malfunction.

[0050] Step S130: Establish a sensor coordination monitoring network within the critical mismatch area, capture electromagnetic interference signals between sensors through the sensor coordination monitoring network as a calibration benchmark to form an interference calibration map, and perform time-series correlation analysis on the interference calibration map and measurement offset feature groups to obtain active offset sources.

[0051] Specifically, a coordinated sensor monitoring network is established within the critical mismatch area. Coordinating sensor nodes are added at the center and boundaries of the critical mismatch area, with a regular grid topology to ensure full coverage monitoring of the mismatch area. Each coordination node is equipped with a multimodal sensing unit, including a network latency detector, a CPU load detector, and an I / O performance monitor, enabling synchronous acquisition of different types of performance interference. Nodes are connected via a high-speed network, using dedicated VLANs to isolate monitoring traffic and reduce the impact on the business network during monitoring. The clock synchronization accuracy of all nodes is set to the millisecond level, and the time consistency of data across different nodes is ensured through an NTP clock synchronization mechanism. A hierarchical network architecture is constructed, with bottom-layer nodes responsible for raw performance data acquisition, middle-layer nodes for data aggregation and preprocessing, and top-layer nodes performing coordination control and data distribution. Each node is assigned a unique network address and monitoring priority; high-priority nodes transmit critical performance data first during network congestion. For example, in a rack-level critical mismatch area, nine coordination nodes are deployed to form a 3x3 monitoring grid, with the node spacing determined according to the mismatch intensity distribution, and the node density increasing accordingly at server locations with high mismatch intensity. Configure the node's operating parameters, including sampling frequency, monitoring threshold, data cache size, and alarm triggering conditions. These parameters are optimized based on the performance characteristics of the mismatched area.

[0052] In some embodiments, the step of capturing inter-sensor electromagnetic interference signals through the sensor coordination monitoring network as a calibration reference to form an interference calibration map includes: identifying an interference signal cancellation window from the sensor coordination monitoring network; evaluating a purification gain coefficient based on the interference signal cancellation window; performing interference inverse superposition processing based on the purification gain coefficient to generate a purification interference template; and generating an interference calibration map using the purification interference template.

[0053] Interference signal cancellation windows are identified from the sensor-coordinated monitoring network. In the performance interference signal sequences collected by the coordinated monitoring network, time periods in which interference effects cancel each other out are searched, during which positive and negative performance disturbances can cancel each other out. Reverse interference pairs are identified through performance baseline analysis; when two performance disturbances change in opposite directions and have similar amplitudes, they have cancellation potential. The performance fluctuation variance within different time windows is statistically analyzed, and time points where the variance decreases significantly are identified; these decreases correspond to interference cancellation phenomena. The time-domain waveforms of performance indicators are analyzed to find the moments when performance peaks and troughs align; these moments constitute the center points of the cancellation windows. The effective width of the cancellation window is determined; the window width depends on the duration of the performance disturbance and the system response characteristics. For example, within a 10-second observation window, three effective performance interference cancellation windows are identified, each lasting approximately 2 seconds, with performance fluctuations reduced by more than 60% within each window. The occurrence patterns of cancellation windows are statistically analyzed, including frequency, duration distribution, and window interval characteristics. Parameters for each cancellation window are recorded, including start and end times, main performance indicators, cancellation depth, and interference sources involved in the cancellation. Analyze the factors affecting the cancellation effect, such as the relative location of the interference source, differences in system load, and network topology conditions. Classify the identified cancellation windows according to their cancellation effectiveness, with deeper cancellation windows prioritized for subsequent purification processes.

[0054] The purification gain coefficient is evaluated based on the interference signal cancellation window. Within each identified cancellation window, the fluctuation ratio of the performance indicator before and after purification is calculated, and this ratio is defined as the initial estimate of the purification gain. The purification effect of different performance indicators within the cancellation window is analyzed; the purification gain for CPU load and network latency is usually different. The relationship curve between purification gain and performance indicator type is fitted by regression analysis to obtain the indicator-related gain coefficient function. Considering the server load status factor, the performance interference purification gain varies under different load levels, and a load-gain mapping relationship is established. The time stability of the purification gain is analyzed; a stable gain coefficient indicates reliable purification effect, while fluctuating gain needs adaptive adjustment. For example, within the cancellation window with 80% CPU load, the purification gain coefficient is 3.2, indicating that the performance fluctuation is reduced to 1 / 3.2 of the original; while under dense network I / O, the purification gain coefficient increases to 4.5. The gain coefficients of multiple cancellation windows are statistically averaged to obtain representative purification gain estimates. A purification gain coefficient table is established, organizing the gain data according to dimensions such as performance indicator type, load status, and interference intensity.

[0055] A cleaned interference template is generated by performing inverse superposition processing based on the cleanup gain coefficient. Based on the evaluated cleanup gain coefficient, an inverse performance adjustment signal is designed, with its amplitude and direction precisely adjusted to achieve optimal cancellation. The inverse adjustment signal is generated by suppressing interference by phase reversal of the original performance disturbance signal; the adjustment strength is determined based on the cleanup gain coefficient. The inverse adjustment signal is superimposed on the original performance data, taking into account the effects of system response delay and multi-server interaction. An adaptive algorithm is introduced into the superposition process, with algorithm parameters adjusted according to the real-time cancellation effect to ensure good cleanup performance even when performance characteristics change. Corresponding cleanup templates are generated for different types of performance interference; sudden performance spikes, continuous performance drift, and periodic performance fluctuations require different processing strategies. The results of all cleanup processes are integrated to form a comprehensive cleaned interference template, which includes both time-domain features and performance index characteristics.

[0056] An interference calibration map is generated using a cleanup interference template. The template is applied to the original performance data, and the performance fluctuation distribution after cleanup is statistically analyzed. Performance stability before and after cleanup is compared to generate a cleanup effect distribution map, visually displaying the degree of improvement for each server node. A multi-layered interference calibration map is constructed: the bottom layer displays the original performance fluctuation distribution, the middle layer displays the performance distribution after cleanup, and the top layer displays the cleanup benefits. A color-coding scheme is used to represent different performance stability levels: green indicates high stability, yellow indicates medium stability, and red indicates high fluctuation. Key information is marked on the calibration map, including the location of the main performance interference sources, the area with the best cleanup effect, and areas that still need improvement. For example, the calibration map shows that the performance stability of the southwest corner of the rack improves from red to yellow, indicating a significant cleanup effect; while the northeast corner remains orange, requiring further optimization. Performance metrics of the calibration map are statistically analyzed, including average performance fluctuation level, stability distribution variance, and cleanup coverage. A complete interference calibration map dataset is output, including a static distribution map, a dynamic evolution map, and statistical analysis results.

[0057] Active offset sources are identified through time-series correlation analysis of interference calibration maps and measurement offset feature clusters. The time-series data of the interference calibration maps are compared with the occurrence times of the measurement offset feature clusters. Statistical correlation analysis is used to quantify the correlation between performance interference events and offset features. A peak in the correlation coefficient at a certain time delay indicates a causal relationship. The impact patterns of different types of performance interference on various offset features are analyzed. Sudden load surges typically cause sudden offsets, while continuous resource contention leads to drift offsets. The temporal evolution of the interference-offset correlation is tracked using a sliding window method, constructing a mapping table of interference sources and offset features, recording the type and degree of offset that each interference source may cause. For example, CPU surges generated by the startup of large data tasks are highly correlated with fast pulse offset features, and I / O surges from virtual machine periodic backups are correlated with oscillating offset features. Active offset sources are identified, i.e., those interference sources that frequently generate performance interference and cause offsets. An assessment framework for offset source activity was established, with quantitative judgment criteria: interference sources occurring more than 5 times per hour, with a single incident lasting more than 30 seconds, affecting more than 3 sensors, and with an offset intensity exceeding 200% of the baseline value were marked as candidate active sources. The comprehensive activity score was calculated as A = 0.3 × F + 0.25 × D + 0.25 × S + 0.2 × I, where F is the frequency score, D is the duration score, S is the impact range score, and I is the intensity score, with each item scored on a 0-10 scale. Priority was assigned based on the comprehensive score: 9 points or above was classified as emergency, 7-9 points as important, and 5-7 points as general. Different monitoring strategies and processing times were applied to different levels.

[0058] Step S140: Use the active offset source to perform synchronous coordination on the key mismatch area to obtain the coordination sequence, perform time delay analysis on the coordination sequence to obtain the spatial propagation time difference, and perform sensor spatial positioning correction based on the spatial propagation time difference to generate the coordination vector.

[0059] Specifically, active offset sources are used to perform synchronization coordination in critical mismatch areas to obtain coordination sequences. Based on the priority level of active offset sources, service nodes within the critical mismatch areas are subjected to differentiated coordination processing. For emergency-level active offset sources with a comprehensive score of 9 or higher, a high-priority coordination sequence is immediately initiated, allocating maximum bandwidth and computing resources for synchronization processing. For important-level offset sources with scores of 7-9, a standard coordination sequence is initiated, and coordination optimization is performed while ensuring system stability. For general-level offset sources with scores of 5-7, a background coordination mode is adopted, utilizing system idle time for gradual coordination. Coordination instructions are sent to all service nodes within the mismatch area, including target performance parameters, coordination time windows, and resource allocation schemes. After receiving the coordination instructions, each service node records the time of receipt, execution status, and performance response; this information constitutes the raw coordination data. Timestamp alignment technology is used to organize the response data of different service nodes under a unified time base, eliminating the impact of local clock deviations. The response quality of each service node to the coordination instructions is analyzed; fast response speed and significant performance improvement indicate good coordination results, while delayed response or performance degradation may indicate resource conflicts or configuration problems. Feature parameters are extracted during the coordination process, including command propagation latency, performance improvement, and resource utilization changes. These parameters reflect the propagation characteristics of coordination within the service chain. For example, in a mismatch region containing twelve service nodes, the response time of coordination commands triggered by emergency offset sources differs by several milliseconds to tens of milliseconds across different nodes, with core service nodes responding fastest and edge nodes responding slightly slower. The coordination data of all service nodes is then arranged chronologically to form a coordination sequence.

[0060] In some embodiments, performing delay analysis on the coordinated sequence to obtain the spatial propagation time difference includes: identifying high-precision delay blocks and low-precision delay blocks from the coordinated sequence; performing time compression processing on the high-precision delay blocks to obtain compressed delay data, and performing time expansion processing on the low-precision delay blocks to obtain expanded delay data; generating a sub-precision delay strategy based on the compressed delay data and the expanded delay data; and establishing the spatial propagation time difference according to the sub-precision delay strategy.

[0061] Identify high-precision and low-precision latency blocks from the coordination sequence. Scan each data segment in the coordination sequence for response accuracy and temporal resolution. Data segments with stable response times and high measurement accuracy are marked as high-precision latency blocks. Analyze the consistency of service responses. Services with small response time fluctuations and clear performance improvements can provide high-precision latency measurements; these data are classified as high-precision. Statistically calculate the uncertainty of latency measurements and assess the accuracy level through the standard deviation of multiple measurements. Data blocks with a standard deviation less than a set threshold are considered high-precision. Identify data segments affected by network jitter or resource contention; these segments have lower latency measurement accuracy and are classified as low-precision latency blocks. Analyze the distribution ratio of high-precision and low-precision latency blocks. Typically, high-precision blocks account for 65%, and low-precision blocks account for 35%. For example, the first half of the coordination sequence has stable service responses and latency measurement errors in the millisecond range, belonging to high-precision blocks; the second half is affected by network congestion, and measurement errors reach tens of milliseconds, belonging to low-precision blocks. Each delay block is assigned a precision label and a confidence weight. The weight of high-precision blocks is set to 1.0, and the weight of low-precision blocks is between 0.4 and 0.8 depending on the precision level.

[0062] High-precision delay blocks undergo time compression to obtain compressed delay data, while low-precision delay blocks undergo time expansion to obtain expanded delay data. Time compression is applied to high-precision delay blocks to retain key delay information while reducing data redundancy. Compression methods include extracting key time points, retaining performance change points, and removing stationary segments; the compression ratio is adaptively adjusted based on data characteristics. The equivalent delay after compression is verified to ensure that compression does not compromise the accuracy of delay measurements. Time expansion is performed on low-precision delay blocks, improving time resolution through interpolation and smoothing techniques. Expansion methods include linear interpolation, spline interpolation, and trend fitting; a suitable method is selected to smoothly increase the number of sampling points. Historical performance data is introduced during the expansion process to improve the quality of low-precision data by utilizing historical service response patterns. For example, low-precision data with a sampling rate of 10Hz is expanded to 100Hz through interpolation, improving time resolution tenfold; high-precision data at 100Hz is compressed to 20Hz, retaining key delay measurement features. The information fidelity before and after processing is evaluated to ensure that compression does not excessively lose information and expansion does not introduce false information. Generate compressed latency datasets and expanded latency datasets, which are complementary in terms of time scale.

[0063] A precision-based latency strategy is generated based on compressed and extended latency data. Compressed high-precision latency data is used as a benchmark to determine the accurate value of latency measurements. Extended low-precision latency data is used as supplementary information to fill in gaps in the high-precision data. A precision-based fusion method is designed, assigning different processing priorities and weighting coefficients according to the data precision level. Processing intervals are divided on the time axis: intervals with dense high-precision data employ a fine-grained processing strategy, while intervals dominated by low-precision data employ a coarse-grained processing strategy. A precision switching mechanism is established to smoothly transition at the boundary between high and low precision data, avoiding discontinuities in processing results. For example, in the 0-10 second interval, high-precision data predominates, and the latency resolution is set to 1 millisecond; in the 10-20 second interval, high and low precision data are mixed, and the latency resolution is adjusted to 5 milliseconds; in the 20-30 second interval, low-precision data predominates, and the latency resolution is reduced to 10 milliseconds. The latency measurement error under different precision strategies is calculated, and the strategy combination with the smallest error is selected. A precision-based latency strategy table is generated, specifying the processing methods for different time periods and different service nodes. The strategy parameters are encoded into control vectors to facilitate automated execution.

[0064] Spatial propagation time difference is established based on a precision-based delay strategy. Following this strategy, corresponding processing methods are applied to delay data at different precision levels. The absolute response time difference of each service node relative to the coordinating command initiation point is calculated; the time difference equals the response time minus the command sending time. A response time difference matrix is ​​constructed, where matrix element T[i,j] represents the coordination propagation time difference from service node i to service node j. The time difference matrix is ​​optimized using the symmetry constraint of service dependencies. For bidirectional dependent services, theoretically, T[i,j] should be close to T[j,i]; the actual measured difference is used for error correction. The consistency of the time difference is verified through service link constraints; for any three service nodes, the time difference should satisfy the logical constraints of the propagation path. The time difference is converted into an equivalent service link delay, where the delay equals the time difference multiplied by the service response speed coefficient. The response speed is determined based on the service type and network conditions. For example, database services have slower response speeds, web services have faster response speeds, and caching services have the fastest response speeds. A complete spatial propagation time difference dataset is generated, containing three types of information: original time difference, corrected time difference, and equivalent delay.

[0065] Coordination vectors are generated based on sensor spatial positioning correction using spatial propagation time difference. The measured spatial propagation time difference is used to analyze the response timing patterns of each service node to coordination commands, inferring the actual dependencies between services. When the response time of service A is consistently later than that of service B and the time difference is stable, it indicates that A depends on the processing result of B; when multiple services respond simultaneously, it indicates that they have no direct dependency relationship. The inferred dependencies are compared with the nominal configurations of the services to identify service nodes with configuration deviations. The dependency deviation vector ΔD = D_actual - D_nominal is calculated, where D_actual is the actual dependency and D_nominal is the nominal dependency. Coordination vectors are constructed based on the dependency deviation analysis results. These vectors contain two core parameters: service dependency correction amount and configuration transformation parameters. The service dependency correction amount is used to correct the dependency deviations of each service node, reflecting the position adjustment requirements of each service in the dependency topology. The configuration transformation parameters are used to guide the optimization and adjustment of the system configuration, including key configuration elements such as service priority weights, resource allocation ratios, and routing policy parameters. Numerical encoding organizes dependency corrections and configuration transformation parameters into a unified vector structure, ensuring that the coordination vector can fully describe all information related to service coordination and optimization. For example, a certain API service might have a dependency correction of 0.3 (indicating a need for moderate adjustment), and configuration transformation parameters might include a priority weight of 0.8, a resource allocation ratio of 1.2, and a routing latency threshold of 150ms. The coordination vector is then normalized to ensure that parameters with different dimensions can be processed on a unified scale.

[0066] Step S150: Convert the coordination vector into a measurement probability transfer chain, perform multi-level probability superposition processing on the measurement probability transfer chain to obtain the superposition probability state, and perform probability aggregation analysis based on the superposition probability state to determine the measurement decision node.

[0067] Specifically, the coordination vector is transformed into a measurement probability propagation chain. Service dependency corrections and configuration transformation parameters are extracted from the coordination vector, and these deterministic parameters are converted into probability distributions. The performance uncertainty of each service node is described by a probability distribution, where the mean is the corrected performance and the variance reflects performance stability; a smaller variance indicates more stable performance. Probability propagation relationships between service nodes are constructed. When a performance change in one service affects another, the probability is propagated through statistical correlation analysis, establishing a mapping relationship between the intensity of the impact and the propagation probability. All services are connected into a chain structure according to their dependencies. The performance probability of an upstream service affects the operational reliability of a downstream service, forming a hierarchical propagation of probability. The attenuation characteristics of probability propagation are analyzed. As the propagation depth increases, the determinism of the probability gradually decreases, while the accumulated uncertainty increases. The degree of attenuation is determined by the coupling strength between services. For example, the response probability of Web service W1 is 0.95. After being propagated to database service D1, the probability drops to 0.87, and then further decreases to 0.75 when propagated to storage service S1, demonstrating the attenuation characteristic of probability with the depth of service chain propagation. directional rules for probability propagation are defined, and the propagation direction is determined based on service dependencies. Changes in front-end services affect back-end services, and changes in the data layer affect the application layer, forming a directed propagation chain. A measurement probability propagation chain data structure is generated, containing a node list, dependency connections, and a probability propagation matrix. Each node represents a service, and each edge represents a probability propagation path.

[0068] In some embodiments, the step of performing multi-level probability superposition processing on the measurement probability transmission chain to obtain the superposition probability state includes: monitoring the probability oscillation mode of the measurement probability transmission chain to generate an oscillation trajectory; injecting anti-phase probability at the trough position of the oscillation trajectory to form a probability standing wave; extracting deterministic probability components using the stable nodes of the probability standing wave; and superimposing the deterministic probability components according to phase to generate the superposition probability state.

[0069] The monitoring and measurement of probabilistic oscillation patterns in the probability propagation chain generates oscillation trajectories. The performance probability values ​​of each service node in the probability propagation chain are tracked over time to identify patterns of periodic fluctuations, typically caused by system load cycles or periodic tasks. Statistical analysis methods are used to detect the periodic characteristics of the probability time series; periodic peaks in statistical indicators indicate the presence of probabilistic oscillations, with peak intervals corresponding to the oscillation period. Frequency domain analysis extracts the main periodic components of the oscillations; typical oscillation periods range from 10 to 100 seconds, with different periods corresponding to different business processes. The service propagation characteristics of the oscillations are analyzed; performance fluctuations in certain service nodes trigger latency fluctuations in related service nodes, forming a oscillation propagation pattern. The propagation speed depends on the coupling strength between services. Based on the oscillation pattern monitoring results, oscillation trajectories are constructed, organizing the identified probabilistic oscillations into a continuous trajectory data structure according to the time series. The oscillation trajectory records key oscillation parameters at each moment, including oscillation amplitude, phase position, propagation direction, and attenuation intensity. Connecting time sampling points forms a complete oscillation trajectory curve, which visually displays the dynamic evolution of the probabilistic oscillations over time. Plot a probability-time trajectory graph, with time on the horizontal axis and probability value on the vertical axis. The peaks and troughs of the trajectory curve clearly mark the extreme points of the oscillation.

[0070] For example, injecting an anti-phase probability at the trough of the oscillation trajectory to form a probability standing wave includes: decomposing the oscillation trajectory into a main oscillation component and a harmonic component; generating an anti-phase probability pulse at the trough of the main oscillation component; coherently superimposing the anti-phase probability pulse with the harmonic component to generate a standing wave node; and connecting the standing wave node to form a probability standing wave.

[0071] The oscillation trajectory is decomposed into a principal oscillation component and harmonic components. Frequency domain decomposition is performed on the performance probability oscillation trajectory, breaking down the complex fluctuation pattern into components of different periods, each representing a fluctuation pattern. The periodic component with the largest amplitude is identified as the principal oscillation component, which typically carries most of the fluctuation energy and determines the basic characteristics of the fluctuation. Components with periods that are integer multiples of the principal period are extracted as harmonic components. Harmonics reflect the complex characteristics of the fluctuation, with higher-order harmonics indicating the degree to which the fluctuation deviates from the standard period. The relative intensity and phase relationship of each component are evaluated. The intensity of the principal oscillation component is set as the benchmark, and the harmonic components are represented by relative values. The phase relationship determines the superposition effect of the components. The correlation between the principal oscillation and harmonics is analyzed: in-phase harmonics enhance the fluctuation, out-of-phase harmonics weaken the fluctuation, and quadrature harmonics change the shape of the fluctuation. For example, the principal oscillation period is 60 seconds, the second harmonic period is 30 seconds, and the third harmonic period is 20 seconds. The amplitude and phase of each component together determine the temporal characteristics of the performance fluctuation. The decomposed periodic components retain the complete information of the original oscillation and can be combined to reconstruct the original signal.

[0072] An anti-phase probability pulse is generated at the trough of the main oscillation component. All troughs of the main oscillation component are located, identified through trend analysis to ensure the discovery of true local minima. A compensation pulse is generated at each trough moment; the pulse intensity equals the difference between the probability value at the trough and the average probability, ensuring the pulse intensity matches the fluctuation depth. The pulse phase is set opposite to the main oscillation phase to achieve performance compensation and ensure maximum cancellation with the original fluctuation. The pulse duration is controlled, set as a certain proportion of the oscillation period; too narrow a width results in insufficient response, while too wide a width leads to excessive duration. The pulse intensity curve is adjusted to use smooth intensity changes to reduce the impact of abrupt pulse changes and avoid introducing additional performance disturbances. For example, an anti-phase compensation pulse is generated at a specific performance trough, with pulse parameters precisely designed to ensure optimal matching with the original fluctuation. The time interval of the pulse sequence equals the main oscillation period to maintain synchronization with the fluctuation. The generated anti-phase pulse sequence covers the entire observation time window, providing continuous compensation excitation for the formation of a stable performance point.

[0073] Standing wave nodes are generated by coherently superimposing inverse probability pulses with harmonic components. The generated inverse probability pulse sequence is coherently superimposed with the retained harmonic components, maintaining the temporal relationship of each component during the superposition process. The inverse probability pulse is specifically designed to reverse the load peaks in the data center service chain; when CPU utilization or network latency spikes, the pulse provides a reverse resource allocation probability. Analyzing the probability distribution after superposition reveals that at certain specific moments, the inverse pulses and harmonic components completely cancel each other out, forming a probabilistically stable standing wave node. Numerical analysis determines the precise location of the standing wave node; the node location is determined by the cancellation conditions and is related to the service response cycle and load phase difference. The stability characteristics of the standing wave node are evaluated; stable standing wave nodes are insensitive to small system disturbances and remain unchanged even with network jitter. Performance inversion points are identified among the standing wave nodes; at these inversion points, the inverse pulses and harmonic components are superimposed in phase, and the service performance fluctuation reaches its maximum value. For example, when the inverse probability pulse for web services is superimposed with the harmonic components of database access, a stable standing wave node is formed at a specific moment. The service response probability at the node is fixed at 0.85 and does not change with the load.

[0074] Standing wave nodes are connected to form a probabilistic standing wave (PSW). All identified PSW nodes are connected in chronological order to form a complete PSW structure, exhibiting a regular periodic distribution over time. The PSW reflects the probability distribution pattern of the data center service chain under stable operating conditions, with each PSW node corresponding to a stable moment in service performance. Analyzing the temporal continuity of the PSW reveals smooth transitions between adjacent PSW nodes, with the transition shape conforming to the pattern of service load changes. Characteristic parameters of the PSW are statistically analyzed, including the standing wave period length, number of nodes, and time frequency. These parameters comprehensively describe the temporal characteristics of service chain stability. A temporal distribution graph of the PSW is plotted, with time on the horizontal axis and service response probability on the vertical axis, showing a regular standing wave shape with alternating nodes and sub-nodes. For example, the formed PSW contains 12 PSW nodes and 11 fluctuation points, exhibiting a stable 60-second periodic structure, providing a stable foundation for data center performance prediction. The envelope of the PSW is extracted, reflecting the average temporal distribution of service chain probabilities, removing the influence of short-term fluctuations. Probabilistic standing waves transform time oscillations into stable modes, realizing the conversion from dynamic service performance to a static probabilistic benchmark.

[0075] Deterministic probability components are extracted from stable nodes in probabilistic standing waves (PSWs). Stable nodes with amplitudes consistently zero or close to zero are identified from the PSWs; their probability values ​​are unaffected by service load fluctuations, representing the steady-state operation characteristics of the data center service chain. The mean probability of each stable node is calculated; this mean represents the deterministic probability level at that moment, eliminating the random effects of network jitter and sudden traffic spikes. The deterministic and random components in the probability signal are separated. The deterministic component corresponds to the stable and predictable service response, providing a reliable basis for data center performance measurement. Slowly changing probability trends are extracted through trend analysis, filtering out high-frequency CPU spikes and memory fluctuations, preserving a stable service response probability baseline. Interpolation is performed between stable nodes to reconstruct the deterministic probability distribution of the entire service chain. Smooth interpolation is chosen to ensure the continuity of inter-service dependencies. For example, deterministic probability values ​​are extracted from multiple stable nodes such as web services, database services, and caching services. Through interpolation, a continuous deterministic probability curve is obtained, reflecting the basic response probability distribution of the entire service chain. Analyze the time-varying rate of deterministic components to identify periods with large probability gradients. These periods typically correspond to peak business periods and require close monitoring.

[0076] The deterministic probability components are superimposed in phase to generate a superimposed probability state. The extracted deterministic probability components are then weighted and combined, with different weights assigned based on the business importance of each service component. These weights reflect the contributions of the Web layer, application layer, and data layer components to the overall system state. Phase alignment is performed to unify the time reference points of all components to the data center's standard clock benchmark, eliminating the impact of time deviations between servers and ensuring the correctness of the superposition. The superposition result is calculated using a weighted summation method: adding components with the same direction of response enhances system stability, subtracting components with opposite directions cancels performance conflicts, and weighted components alter the overall service orientation. Phase inconsistency is addressed by performing time calibration when there are response time offsets between different service components, maintaining the response of each component within a reasonable SLA time range. The magnitude and stability of the superimposed probability are calculated. The magnitude, equal to the weighted sum, reflects the overall system responsiveness, while the stability reflects the reliability of the superposition result under varying business loads. For example, multiple deterministic components such as the front-end load balancer, middleware service, and database cluster are weighted and superimposed to obtain a comprehensive probability state that includes the contributions of all critical services. The combined probability is converted into a standard probability representation for practical data center resource allocation and performance decisions. The final superimposed probability state vector is generated, which fully describes the probability configuration and decision basis of the service chain after multi-level processing.

[0077] In some embodiments, the step of performing probabilistic aggregation analysis based on the superimposed probabilistic state to determine the measurement decision node includes: constructing a probabilistic potential energy field based on the superimposed probabilistic state; finding potential energy depressions in the probabilistic potential energy field as probabilistic convergence points; applying perturbation tests to the probabilistic convergence points to evaluate and generate a stability index; and marking convergence points whose stability index exceeds a robustness threshold as measurement decision nodes.

[0078] A probabilistic potential energy field is constructed based on superimposed probabilistic states. This field transforms the probability distribution into a potential energy distribution, with high-probability regions corresponding to low-potential-energy regions and vice versa, forming a spatial structure similar to a physical potential energy field. The superimposed probabilistic states are mapped to a decision weight distribution using the weight function W = k·log(P + ε), where W is the weight, P is the probability, k is the proportionality coefficient, and ε is a small constant to prevent zero values. This mapping transforms high-probability regions into high-weight regions, establishing the numerical foundation of the probabilistic potential energy field. The spatial topological characteristics of the probabilistic potential energy field are analyzed. The direction of the fastest change in potential energy is determined through potential energy gradient calculation; regions with large gradients indicate drastic changes in decision importance. The topological structure of the probabilistic potential energy field is depicted using equipotential energy lines; areas with dense equipotential energy lines indicate drastic changes in potential energy, corresponding to regions of rapid change in decision importance. The concentrated characteristics of the probabilistic potential energy field are analyzed: potential energy valleys correspond to high-probability peaks, i.e., high-importance regions, while potential energy peaks correspond to low-probability valleys, i.e., low-importance regions. The distribution characteristics of the statistical probabilistic potential energy field are used to identify extreme points and saddle points of the potential energy. Extreme points are potential decision locations. For example, the probabilistic potential energy field exhibits multiple local potential energy valleys, each corresponding to a region of high importance, with the deepest potential energy valley having the highest decision value.

[0079] In the probabilistic potential energy field, potential energy depressions are identified as probabilistic convergence points. A weighted clustering method is used to search for local maxima in the weight field, starting from different initial points and moving along the direction of increasing weights until convergence to the weight peak region. Points with zero weight gradients and negative curvature are identified as the centers of weight peaks; these points are local maxima of the weights. The geometric characteristics of each peak region are statistically analyzed, including height, width, and coverage. Height reflects the degree of decision importance, and width reflects the scope of influence. The attraction domain of the peak region is analyzed, and by forward tracing the weight gradient field, it is determined which regions' weights will converge to that peak. Streamline analysis is used to display the convergence path of the weights; streamlines flow from low-weight regions to high-weight peaks, forming convergence patterns. For example, multiple major weight peak regions are identified, each with a clear attraction domain and convergence characteristics. The weight differences between peak regions are calculated; these differences determine the ease with which decisions are transferred between regions, with large differences indicating relative independence. The total weight falling into each peak region is counted as the importance weight of that convergence point. A list of probabilistic convergence points is generated, with each convergence point representing a potential decision location.

[0080] A stability index is generated by applying perturbation tests to probabilistic convergence points. Small-amplitude random perturbations are applied to each probabilistic convergence point, with the perturbation amplitude set at 5% of the weight value and the perturbation direction randomly distributed to simulate the uncertainty and fluctuations in actual decision-making. The response behavior of the convergence points after the perturbation is observed. Stable convergence points quickly recover to their original state, while unstable ones shift or disperse; the response trajectory reflects the stability characteristics of the decision. The recovery time is statistically analyzed, and the recovery time constant is obtained by fitting a recovery curve, with rapid recovery corresponding to a small time constant value. The variance of the convergence point positions is statistically analyzed through multiple perturbation tests; a small variance indicates stable position. Perturbations of different intensities are applied to determine the maximum tolerable perturbation intensity; when the perturbation exceeds this intensity, the convergence point cannot recover to its original state. The stability characteristics of the system are evaluated, and the stability of the system is determined through perturbation response analysis. The stability index S is calculated based on the above test results: S = W1·(recovery speed) + W2·(position stability) + W3·(disturbance resistance) + W4·(system stability), where W1, W2, W3, and W4 are weighting coefficients to ensure a balance of contributions. The stability index is normalized to map the index to the [0,1] interval.

[0081] Convergence points with stability indices exceeding the robustness threshold are marked as measurement decision nodes. The robustness threshold is set as the high quantile of the stability index distribution to ensure that the selected nodes all possess good stability. Convergence points with stability indices higher than the threshold are selected; these points remain stable under various disturbances, ensuring reliable decision results. The spatial distribution of the selected nodes is analyzed to ensure coverage of the main service area and no key locations are overlooked. The minimum distance between decision nodes is calculated to avoid excessive node concentration and maintain appropriate spatial dispersion. Decision weights are assigned to each decision node; the weights are proportional to the stability index, with nodes of higher stability playing a greater role in the decision-making process. For example, decision nodes meeting stability requirements are selected from multiple convergence points, and each node undergoes rigorous stability testing. Functional division of labor is established for decision nodes, with different nodes responsible for different types of decision-making tasks, achieving professional specialization.

[0082] Step S160: Assign probability weights to the measurement decision nodes to construct a probability configuration network, generate a priority sequence based on the probability configuration network, and output a multi-sensor coordinated measurement scheme according to the priority sequence.

[0083] Specifically, a probabilistic configuration network is constructed by assigning probabilistic weights to the measurement decision nodes. The probabilistic state value and stability index of each service node are extracted from the measurement decision nodes, and these two parameters are combined to calculate the node's comprehensive weight. The weight calculation formula is W = αP + βS, where W is the comprehensive weight, P is the probabilistic state value, S is the stability index, and α and β are balance coefficients, typically α = 0.6 and β = 0.4, reflecting the relative importance of service response probability and stability. Decision nodes are connected into a network structure according to service topology location and business dependencies. Bidirectional connections are established between adjacent service nodes, and logical connections are established between business-related nodes. A transfer weight is assigned to each edge in the network; the edge weight is equal to the geometric mean of the weights of the two endpoints, ensuring the smoothness of weight transfer in the service chain. A weight matrix is ​​constructed to represent the connection relationships and weight distribution of the probabilistic configuration network, where matrix element W[i,j] represents the connection weight from service node i to service node j. The degree centrality and betweenness centrality of each node are calculated. Degree centrality reflects the number of direct dependencies of a service node, and betweenness centrality reflects the bridging role of a node in the service chain.

[0084] Priority sequences are generated based on a probabilistic network configuration. All service nodes in the network are sorted according to their comprehensive weights, with the node having the highest weight receiving the highest priority. Dependencies between nodes are considered; if a high-priority service node depends on the output of a low-priority node, the priority of the dependent node is increased. A topology sorting method is used to handle the order constraints of service nodes, ensuring that the priority sequence satisfies all business dependencies. Time constraints are introduced, assigning higher priority to time-sensitive services to prevent SLA timeout failures. Resource requirements for each priority level are calculated, including CPU resources, network bandwidth, and memory consumption, ensuring that high-priority services receive sufficient resources. Dynamic priority adjustment rules are set; if a service node fails to respond multiple times consecutively, its priority is temporarily lowered, allowing other nodes to execute first. For example, in a generated priority sequence [WebGateway, AuthService, UserDB, CacheLayer, LogService, FileStorage, BackupService, MonitorAgent], the WebGateway node is ranked first due to its highest weight and status as the front-end entry point.

[0085] A multi-sensor coordinated measurement scheme is output based on a priority sequence. The priority sequence is converted into a specific service optimization scheduling plan, with each task including execution nodes, performance parameters, and time requirements. A synchronization mechanism for coordinated optimization is designed, where high-priority services initiate performance optimization first, and low-priority services initiate sequentially based on synchronization signals. Monitoring resources are allocated, including monitoring frequency, performance monitoring accuracy, and data transmission bandwidth, prioritizing the resource needs of high-priority services. A data fusion strategy is formulated, where performance data from services of different priorities are weighted and averaged, with higher-priority data having greater weight. Resource conflict resolution rules are set, allocating access rights according to priority when multiple services need to access shared resources simultaneously. For example, the coordinated measurement scheme specifies: Phase 1 (0-30 seconds) performs precise performance monitoring of three high-priority nodes (WebGateway, AuthService, and UserDB) at a monitoring frequency of 100Hz; Phase 2 (30-60 seconds) performs routine performance monitoring of two medium-priority nodes (CacheLayer and LogService) at a monitoring frequency of 50Hz; Phase 3 (60-90 seconds) performs supplementary performance monitoring of the remaining low-priority nodes. Quality indicators and completion standards are set for each optimization phase to ensure that the performance optimization results meet SLA requirements. Develop an execution flowchart for the optimization plan, clearly showing the startup sequence and data flow of each service. Generate executable service configuration scripts, containing configuration commands and synchronization instructions for all services. Output a complete multi-sensor coordinated measurement plan document, including optimization objectives, service configurations, execution steps, data processing flow, and expected results, providing detailed guidance for actual data center operation and maintenance.

[0086] To implement the data center multi-sensor integrated measurement method corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, see [link to documentation]. Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a data center multi-sensor integrated measurement device 200 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The data center multi-sensor integrated measurement device 200 provided in this embodiment includes:

[0087] Data acquisition module 201 is used to collect heterogeneous sensor data from each monitoring node in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool.

[0088] The offset analysis module 202 is used to perform feature scanning on the sensor mismatch data pool to extract the measurement offset feature family, perform correlation matching between the sensor mismatch data pool and the measurement offset feature family to obtain the offset correlation tensor, and locate the key mismatch region based on the offset correlation tensor.

[0089] The calibration monitoring module 203 is used to establish a sensor coordination monitoring network within the critical mismatch area, capture electromagnetic interference signals between sensors through the sensor coordination monitoring network as a calibration benchmark to form an interference calibration map, and perform time-series correlation analysis on the interference calibration map and the measurement offset feature group to obtain active offset sources.

[0090] The coordination processing module 204 is used to perform synchronous coordination on the key mismatch area using the active offset source to obtain a coordination sequence, perform time delay analysis on the coordination sequence to obtain the spatial propagation time difference, and perform sensor spatial positioning correction based on the spatial propagation time difference to generate a coordination vector.

[0091] The probability calculation module 205 is used to convert the coordination vector into a measurement probability transmission chain, perform multi-level probability superposition processing on the measurement probability transmission chain to obtain the superposition probability state, and perform probability aggregation analysis based on the superposition probability state to determine the measurement decision node.

[0092] The decision output module 206 is used to allocate probability weights to the measurement decision nodes to construct a probability configuration network, generate a priority sequence based on the probability configuration network, and output a multi-sensor coordinated measurement scheme according to the priority sequence.

[0093] The aforementioned data center multi-sensor integrated measurement device 200 can implement the data center multi-sensor integrated measurement method of the above-described method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0094] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0095] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A multi-sensor integrated measurement method for data centers, characterized in that, include: Collect heterogeneous sensor data from each monitoring node in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool. The sensor mismatch data pool is scanned to extract measurement offset feature groups. The sensor mismatch data pool and the measurement offset feature groups are correlated to obtain an offset correlation tensor. The offset correlation tensor is a three-dimensional tensor data structure. Its three dimensions correspond to the number of sensors, the number of feature groups, and the time length, respectively. The key mismatch area is located based on the offset correlation tensor. A sensor coordination monitoring network is established within the critical mismatch region. Electromagnetic interference signals between sensors are captured through the sensor coordination monitoring network and used as a calibration benchmark to form an interference calibration map. Time-series correlation analysis is performed on the interference calibration map and the measurement offset feature group to obtain active offset sources. The active offset source is used to perform synchronous coordination on the key mismatch region to obtain a coordination sequence. The coordination sequence is then subjected to time delay analysis to obtain the spatial propagation time difference. Based on the spatial propagation time difference, sensor spatial positioning correction is performed to generate a coordination vector. The coordination vector is converted into a measurement probability transfer chain. The measurement probability transfer chain is a chain data structure containing a node list, dependency connection relationship and probability transfer matrix. Each node represents a service and each edge represents a probability transfer path. The measurement probability transfer chain is subjected to multi-level probability superposition processing to obtain the superposition probability state. Based on the superposition probability state, probability aggregation analysis is performed to determine the measurement decision node. The measurement decision node is a probability convergence point whose stability index exceeds the robustness threshold after disturbance test evaluation. A probability configuration network is constructed by assigning probability weights to the measurement decision nodes, a priority sequence is generated based on the probability configuration network, and a multi-sensor coordinated measurement scheme is output according to the priority sequence.

2. The method according to claim 1, characterized in that, The step of performing feature scanning to extract measurement offset feature families from the sensor mismatch data pool includes: Phase decomposition is performed on the sensor mismatch data pool to obtain orthogonal components; Phase transition points are identified in the orthogonal components to form a phase anomaly sequence; The phase anomaly sequence is mapped back to the time domain to extract time-domain abrupt change features; Based on the temporal abrupt change features, clustering is used to generate measurement offset feature groups.

3. The method according to claim 1, characterized in that, The step of performing correlation matching between the sensor mismatch data pool and the measurement offset feature family to obtain the offset correlation tensor includes: Based on the sensor mismatch data pool, the complementary gap in the data is identified; Obtain the compensation sequence of the measurement offset feature group from the data complementarity gap; Convert the compensation sequence into a tensor reconstruction driver; The offset-related tensor is established using the tensor reconstruction driver.

4. The method according to claim 1, characterized in that, The method of locating the key mismatch region based on the offset correlation tensor includes: Identify oscillation modes in the offset-related tensor; An offset resonant cavity is constructed based on the oscillation mode; The offset correlation tensor is resonantly amplified through the offset resonant cavity to generate an amplified offset signal; The amplified offset signal is spatially anchored to form a critical mismatch region.

5. The method according to claim 1, characterized in that, The step of capturing electromagnetic interference signals between sensors through the sensor coordination monitoring network as a calibration benchmark to form an interference calibration map includes: Identify interference signal cancellation windows from the sensor coordinated monitoring network; The purification gain coefficient is evaluated based on the interference signal cancellation window. A purification interference template is generated by performing inverse superposition of interference based on the purification gain coefficient. An interference calibration map is generated using the purification interference template.

6. The method according to claim 1, characterized in that, The step of performing delay analysis on the coordinated sequence to obtain the spatial propagation time difference includes: Identify high-precision delay blocks and low-precision delay blocks from the coordination sequence; The high-precision delay block is subjected to time compression processing to obtain compressed delay data, and the low-precision delay block is subjected to time expansion processing to obtain expanded delay data. A precision-based delay strategy is generated based on the compressed delay data and the extended delay data. The spatial propagation time difference is established based on the precision delay strategy described above.

7. The method according to claim 1, characterized in that, The step of performing multi-level probability superposition processing on the measurement probability transmission chain to obtain the superposition probability state includes: The probability oscillation mode of the measurement probability transmission chain is monitored to generate an oscillation trajectory; Injecting an inverse probability wave at the trough of the oscillation trajectory forms a probability standing wave; Deterministic probability components are extracted using the stable nodes of the probabilistic standing wave; The deterministic probability components are superimposed according to their phases to generate a superimposed probability state.

8. The method according to claim 1, characterized in that, The step of performing probabilistic aggregation analysis based on the superimposed probability states to determine the measurement decision node includes: Construct a probabilistic potential energy field based on the superimposed probability states; In the probabilistic potential energy field, find potential energy depressions as probability convergence points; A stability index is generated by applying a perturbation test to the probability convergence point. Convergence points where the stability index exceeds the robustness threshold are marked as measurement decision nodes.

9. The method according to claim 7, characterized in that, The step of injecting an inverse probability at the trough of the oscillation trajectory to form a probability standing wave includes: The oscillation trajectory is decomposed into a main oscillation component and harmonic components. An inverted probability pulse is generated at the trough of the main oscillation component; The antiphase probability pulse is coherently superimposed with the harmonic component to generate a standing wave node; The standing wave nodes are connected to form a probabilistic standing wave.

10. A multi-sensor integrated measurement device for a data center, characterized in that, include: The data acquisition module is used to collect heterogeneous sensor data from various monitoring nodes in the data center, classify the sensor data by deviation intensity to generate a graded deviation queue, and perform coordination processing on the graded deviation queue to obtain a sensor mismatch data pool. The offset analysis module is used to perform feature scanning on the sensor mismatch data pool to extract measurement offset feature groups, and to perform correlation matching between the sensor mismatch data pool and the measurement offset feature groups to obtain an offset correlation tensor. The offset correlation tensor is a three-dimensional tensor data structure, and its three dimensions correspond to the number of sensors, the number of feature groups, and the time length, respectively. The key mismatch area is located based on the offset correlation tensor. The calibration monitoring module is used to establish a sensor coordination monitoring network within the critical mismatch area, capture electromagnetic interference signals between sensors through the sensor coordination monitoring network as a calibration benchmark to form an interference calibration map, and perform time-series correlation analysis on the interference calibration map and the measurement offset feature group to obtain active offset sources. The coordination processing module is used to perform synchronous coordination on the key mismatch area using the active offset source to obtain a coordination sequence, perform time delay analysis on the coordination sequence to obtain the spatial propagation time difference, and perform sensor spatial positioning correction based on the spatial propagation time difference to generate a coordination vector. The probability calculation module is used to convert the coordination vector into a measurement probability transfer chain. The measurement probability transfer chain is a chain data structure containing a node list, dependency connection relationship and probability transfer matrix. Each node represents a service and each edge represents a probability transfer path. The measurement probability transfer chain is subjected to multi-level probability superposition processing to obtain the superposition probability state. Based on the superposition probability state, probability aggregation analysis is performed to determine the measurement decision node. The measurement decision node is a probability convergence point whose stability index exceeds the robustness threshold after disturbance testing and evaluation. The decision output module is used to allocate probability weights to the measurement decision nodes to construct a probability configuration network, generate a priority sequence based on the probability configuration network, and output a multi-sensor coordinated measurement scheme according to the priority sequence.

Images