A big data-based asset management method and system

Abstract

This invention relates to the field of intelligent asset supervision technology, specifically to an asset management method and system based on big data, comprising the following steps: collecting time-series data of operational status and environmental stress parameters using a sensor array; constructing a multidimensional feature correlation matrix based on covariance analysis and calculating a dynamic loss index; spatially mapping this index with the task load vector to be assigned to generate an asset-task matching score; and finally comparing the score with a dynamic judgment benchmark value to output an adaptive allocation instruction. In this invention, by capturing the multidimensional characteristic fluctuations of physical entities in real time, a nonlinear loss assessment system reflecting the actual load pressure is established, solving the problem of delayed asset status perception caused by the reliance on static ledgers in traditional methods. The dynamic adaptation algorithm accurately calculates the fit between asset capabilities and task requirements, maximizing resource allocation efficiency while ensuring equipment operational safety, effectively extending the service life of high-value assets, and reducing operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of intelligent asset supervision technology, and in particular to an asset management method and system based on big data. Background Technology

[0002] The field of intelligent asset monitoring technology involves the status monitoring, value assessment, and resource allocation of high-value physical entities throughout their entire lifecycle. It typically encompasses equipment operation data collection, quantitative health analysis, and automated scheduling management based on business needs. Traditional asset management methods, on the other hand, rely primarily on manual data entry or barcode scanning to establish static asset ledgers. Asset status is updated through regular physical inventory checks and financial depreciation algorithms, and passive maintenance or disposal is performed when equipment experiences significant malfunctions or reaches its preset service life.

[0003] Existing technologies rely on static ledgers and periodic inventory checks, which cannot detect the real performance degradation of physical entities under complex operating conditions in real time. This leads to a serious disconnect between the book value of assets and their actual health status. The lack of a dynamic assessment mechanism based on real-time load pressure results in high-value equipment being idle under low load or overloaded in a sub-healthy state. It is impossible to establish an accurate quantitative mapping between asset capabilities and task requirements, making it difficult to cope with frequently changing business scheduling needs, ultimately leading to misallocation of resources and out-of-control operation and maintenance costs. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an asset management method and system based on big data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an asset management method based on big data, comprising the following steps: S1: Collect time-series data of operational status and environmental stress parameters using sensing devices deployed at key nodes, and perform time-domain alignment processing on the operational status time-series data and environmental stress parameters based on a timestamp synchronization protocol to extract the asset panoramic feature set; S2: Input the asset panoramic feature set into the nonlinear lifetime decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to calculate the dynamic loss index. S3: Obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the adaptation distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data. S4: Compare the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset preferred range conditions, generate an adaptive allocation instruction to drive the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

[0006] As a further aspect of the present invention, the extraction process of the asset panoramic feature set specifically includes: S11: The instantaneous vibration amplitude sequence of the equipment and the environmental temperature and humidity change sequence are synchronously collected by the high-frequency vibration sensor and the temperature and humidity transmitter. The heterogeneous data stream is labeled at the nanosecond level using the network time protocol, and abnormal data frames with a signal-to-noise ratio lower than the preset threshold are removed to generate the original state data stream. S12: The discrete points in the original state data stream are resampled using the Lagrange interpolation method, and the aligned multi-source data is mapped to a unified time axis coordinate system to generate a synchronization state matrix. S13: Perform Fast Fourier Transform and Wavelet Packet Decomposition on the synchronization state matrix to extract time-domain statistical features and frequency-domain energy spectrum features respectively. Use principal component analysis algorithm to perform dimensionality reduction and fusion processing on the extracted mixed features to generate the asset panoramic feature set.

[0007] As a further aspect of the present invention, the calculation process of the dynamic loss index specifically includes: S21: Import the asset panoramic feature set into a preset nonlinear lifetime decay model, use the Pearson correlation coefficient algorithm to calculate the mutual influence weights between parameters of each dimension in the feature set, identify key coupling factors, and generate a parameter coupling weight vector. S22: Based on the parameter coupling weight vector, the feature elements in the asset panoramic feature set are weighted and recombined to construct a multidimensional feature association matrix; S23: Set a time window of preset length, perform continuous definite integral operation on the trace of the multidimensional feature correlation matrix within the time window, calculate the cumulative fatigue damage of the equipment in the recent operating cycle, and generate the dynamic loss index after normalizing the cumulative fatigue damage.

[0008] As a further aspect of the present invention, the process of generating the asset task matching score and the dynamic judgment benchmark value specifically includes: S31: Parse the received task scheduling request, extract the computing resource consumption, I / O throughput requirements and expected duration of the task, and construct the load vector of the task to be assigned. S32: Project the task load vector to be assigned to the same dimensionless feature space as the dynamic loss index using the range transformation method, and calculate the fitting distance between the projected vector and the dynamic loss index by calling the preset weighted Euclidean distance formula, and generate the asset task matching score. S33: Retrieve historical matching score data of the asset when it successfully performed similar tasks in the past preset period, calculate the weighted moving average and standard deviation of the historical matching score data, set the upper and lower floating thresholds based on the confidence interval theory of normal distribution, and generate the dynamic judgment benchmark value.

[0009] As a further aspect of the present invention, the process of generating and executing the adaptive allocation instruction specifically includes: S41: Calculate the difference ratio between the asset task matching score and the dynamic judgment benchmark value, and determine whether the difference ratio falls within the preset preferred range; S42: Based on the task priority level and the remaining maintenance window time of the asset, match the load balancing strategy from the preset strategy library, determine the asset start time, operating power mode and network bandwidth allocation quota, and generate the adaptive allocation instruction. S43: The adaptive dispatching command is sent to the underlying physical actuator via the industrial bus protocol, driving the actuator to adjust the physical connection status and operating parameters of the equipment, and after execution, a closed-loop confirmation signal is sent back to the management system.

[0010] As a further aspect of the present invention, the resampling process in S12 specifically includes: Obtain two adjacent sampling time points and their corresponding state values ​​in the original state data stream, and calculate the time difference ratio between the time point to be interpolated and the sampling time points before and after. A cubic spline interpolation function model is constructed, and the time difference ratio is substituted into the interpolation function model to calculate the estimated state values ​​of the points to be interpolated. The least squares method is used to smooth and correct the estimated results to generate a resampled data sequence. The resampled data sequences are merged into column vectors according to a unified timestamp index to construct the synchronization state matrix.

[0011] As a further aspect of the present invention, the specific logic of the integration operation in S23 is as follows: Obtain the multidimensional feature correlation matrix sequence and the preset material fatigue attenuation coefficient within the current time window, and construct a numerical integral equation; The area under the eigenvalue change curve of the matrix within the sliding window is calculated using the trapezoidal rule. The calculated area value is then multiplied by the material fatigue attenuation coefficient to obtain the incremental loss value within a single sliding window cycle. The incremental loss values ​​generated by all sliding window cycles before the current moment are summed, and an environmental stress correction factor is introduced to perform nonlinear compensation on the summation result to generate the dynamic loss index.

[0012] As a further aspect of the present invention, the formula for calculating the adaptation distance in step S32 is specifically as follows: ; in, This represents the asset-task matching score. Represents the total number of feature dimensions. Representing the The importance weight coefficients of each dimension The load vector of the task to be assigned represents the first... Normalized values ​​in each dimension The dynamic loss index represents the first... Mapping component values ​​in each dimension.

[0013] As a further aspect of the present invention, the process of matching the optimal load balancing strategy in S42 specifically includes: Obtain the predicted remaining lifetime of all available assets in the system at the current moment and their real-time energy efficiency indicators, and construct a global resource status topology map; The particle swarm optimization algorithm is used to perform multi-objective optimization search in the global resource state topology graph to determine the allocation path combination; The parameter set corresponding to the allocation path combination is mapped to a specific control instruction code, and the control instruction code is embedded into the data packet of the adaptive allocation instruction.

[0014] A big data-based asset management system, the system being used to implement the aforementioned big data-based asset management method, the system comprising: The panoramic feature extraction module is used to collect operational status time-series data and environmental stress parameters using sensing devices deployed at key nodes, and to perform time-domain alignment processing on the operational status time-series data and environmental stress parameters based on a timestamp synchronization protocol to extract the asset panoramic feature set. The loss index calculation module is used to input the asset panoramic feature set into the nonlinear lifetime decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to calculate the dynamic loss index. The task suitability assessment module is used to obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the suitability distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data. The adaptive allocation module is used to compare the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset preferred range conditions, an adaptive allocation instruction is generated to drive the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by capturing the multidimensional fluctuation characteristics of physical entities in real time during operation, a dynamic evaluation model reflecting the nonlinear coupling relationship of various parameters is constructed. Covariance analysis and time sliding window algorithm are used to accurately calculate the real-time wear and tear of the equipment, solving the problems of delayed asset status perception and value assessment distortion in the traditional static management model. A matching algorithm based on spatial mapping is adopted to quantify the fit between asset capabilities and task load, and adaptive scheduling is achieved based on dynamic benchmark values. While avoiding equipment overload damage, the utilization rate of idle resources is significantly improved, realizing intelligent closed-loop management and control of the entire asset life cycle. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the asset management method based on big data according to the present invention. Figure 2 This is a flowchart of the asset panoramic feature set extraction process of the present invention; Figure 3 This is a flowchart of the dynamic loss index calculation process of the present invention; Figure 4 This is a flowchart of the asset task matching score and dynamic judgment benchmark value generation process of the present invention; Figure 5 This is a flowchart of the adaptive allocation instruction generation and execution process of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0018] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0019] Please see Figure 1 and Figure 2 This invention provides a technical solution: an asset management method based on big data, comprising the following steps: S1: Collect time-series data of operational status and environmental stress parameters using sensing devices deployed at key nodes. Perform time-domain alignment processing on the time-series data of operational status and environmental stress parameters based on a timestamp synchronization protocol to extract a panoramic feature set of the asset.

[0020] Among them, the operational status time series data refers to the status monitoring data continuously collected in chronological order during the operation of the asset, including at least one of vibration amplitude sequence, rotational speed sequence, current sequence, and power consumption sequence; environmental stress parameters refer to external environmental parameters that affect the operational reliability of the asset, including at least one of temperature, humidity, air pressure, and electromagnetic interference intensity.

[0021] The extraction process of the asset panorama feature set specifically includes: S11: The instantaneous vibration amplitude sequence of the equipment and the environmental temperature and humidity change sequence are synchronously collected by the high-frequency vibration sensor and the temperature and humidity transmitter. The heterogeneous data stream is labeled at the nanosecond level using the network time protocol, and abnormal data frames with a signal-to-noise ratio lower than the preset threshold are removed to generate the original state data stream.

[0022] S12: The Lagrange interpolation method is used to resample the discrete points in the original state data stream, and the aligned multi-source data is mapped to a unified time axis coordinate system to generate a synchronization state matrix.

[0023] The resampling process specifically includes: Obtain two adjacent sampling time points and their corresponding state values ​​in the original state data stream, and calculate the time difference ratio between the time point to be interpolated and the sampling time points before and after; A cubic spline interpolation function model is constructed, and the time difference ratio is substituted into the interpolation function model to calculate the estimated state values ​​of the points to be interpolated. The least squares method is then used to smooth and correct the estimated results, generating a resampled data sequence. The resampled data sequences are merged into column vectors according to a unified timestamp index to construct a synchronization state matrix.

[0024] S13: Perform Fast Fourier Transform and Wavelet Packet Decomposition on the synchronization state matrix to extract time-domain statistical features and frequency-domain energy spectrum features, respectively. Use principal component analysis algorithm to perform dimensionality reduction and fusion processing on the extracted mixed features to generate a panoramic feature set of assets.

[0025] Industrial-grade piezoelectric triaxial vibration accelerometers were deployed at key stress points in the server rack's fan bearings and hard drive array brackets. High-precision temperature and humidity transmitters were also installed at the cold aisle air inlets and hot aisle air outlets of the rack. The vibration sensor sampling frequency was set to 25600 Hz to meet the Nyquist sampling theorem's requirement for capturing high-frequency fault characteristics, and the temperature and humidity transmitter sampling interval was set to 1 second. The data acquisition card was activated, reading the sensor's raw voltage signal through a 4-20 mA analog current loop and converting it to a digital signal via an analog-to-digital converter. A Network Time Protocol (NTP) client was invoked to send a synchronization request message to the Stratum Level 1 core time server, recording the message's timestamp. Server receives timestamps Server sends timestamp and the client received timestamp The network round-trip delay and local clock deviation are calculated using formulas, and the local crystal oscillator clock of the data acquisition card is calibrated at the nanosecond level to ensure that the time stamp error of all sensor data is controlled within 10 microseconds. The calibrated time stamps are read, and the vibration waveform data and temperature and humidity data are encapsulated into a data frame with a unified 64-bit UNIX timestamp header.

[0026] The aforementioned Network Time Protocol (NTP) is a network protocol used to synchronize computer time in a computer network. It achieves high-precision time synchronization by exchanging timestamped data packets between the client and the server and calculating network latency and clock offset.

[0027] For the acquired raw signals, a statistical signal-to-noise ratio (SNR) analysis is performed. A signal window with 1024 sampling points is selected, and the root mean square value of the signal amplitude within the window is calculated as the estimated signal power. Simultaneously, the high-frequency noise component of the signal is extracted through wavelet transform, and its variance is calculated as the estimated noise power. The ratio of signal power to noise power is converted to decibels, and the SNR threshold is set to 25 dB. If the calculated SNR of the current data frame is lower than 25 dB, it is determined that the data frame is affected by strong electromagnetic interference or poor sensor contact, and the data frame is directly discarded and an anomaly log is recorded; if the SNR is higher than or equal to 25 dB, the data frame is retained. The retained data frames are parsed to extract the vibration acceleration amplitude (unit: g) and temperature and humidity values ​​(unit: degrees Celsius / percentage), forming a raw state data stream arranged in chronological order. For example, a vibration amplitude of 0.05g was collected at time 1672531200.001 and a vibration amplitude of 0.052g was collected at time 1672531200.002, which constitutes a time series input.

[0028] The signal-to-noise ratio (SNR) mentioned above refers to the ratio of signal power to noise power, usually expressed in decibels (dB). It is used to measure the quality of a signal; the higher the value, the clearer the signal and the less it is affected by noise.

[0029] A uniform resampling frequency of 1000 Hz is set, meaning a standard timestamp anchor point is generated every 1 millisecond. The original state data stream is traversed to find the previous actual sampling point closest to the current standard anchor point. With the next actual sampling point Assume the standard anchor time is... The previous sampling point had a time interval of 100 milliseconds and a value of 0.5; the next sampling point had a time interval of 105 milliseconds and a value of 0.6. Calculate the time difference ratio. .like If it is 102 milliseconds, then .

[0030] Two sampling points before and after the interpolation point are selected, for a total of four points, to construct a system of cubic spline polynomial equations. Solving the system of equations yields the coefficients of the cubic polynomial. Substituting the calculated time difference ratio into the polynomial, the vibration amplitude and temperature / humidity values ​​of the interpolation point are calculated. To eliminate the Runge phenomenon introduced by interpolation, ten consecutive resampling points, including the interpolation point, are selected to construct an observation matrix and a design matrix. Using the least squares method, a smooth curve with the minimum sum of squared errors is found to approximate these interpolation points, and the corresponding values ​​on the smooth curve replace the original interpolation results. For example, if the interpolated vibration value is 0.54g, it is adjusted to 0.538g after least squares correction to remove abrupt noise.

[0031] The cubic spline interpolation mentioned above refers to an interpolation method that uses piecewise cubic polynomials to smoothly connect discrete data points. It ensures the continuity of the first and second derivatives of the interpolation function at the connection points, thus generating curves with high smoothness.

[0032] Establish a dimension as An empty matrix, in which The length of the time series (e.g., 10,000 rows, representing 10 seconds of data). The number of feature channels is specified (e.g., 3 columns, representing X-axis vibration, temperature, and humidity, respectively). The corrected resampled vibration sequence is filled into the first column, the temperature sequence into the second column, and the humidity sequence into the third column, ensuring that each row of data strictly corresponds to the same millisecond time.

[0033] For the vibration column vector of the synchronization state matrix, Fast Fourier Transform (FFT) is applied. A Hanning window is used to window and truncate the signal to reduce spectral leakage. The frequencies and amplitudes of the first five main peaks in the spectrum are calculated to extract the "dominant frequency energy" feature. Simultaneously, a 3-level wavelet packet decomposition is performed on the signal using the db4 wavelet basis, calculating the energy proportion of the eight frequency band nodes in the third level to generate a frequency domain energy spectrum feature vector. In the time domain, the mean, variance, kurtosis, and skewness of the signal are calculated. The time-domain and frequency-domain features are concatenated to form a high-dimensional feature vector (e.g., 128 dimensions). Principal Component Analysis (PCA) is called to calculate the covariance matrix of the high-dimensional feature matrix and perform eigenvalue decomposition. The eigenvalues ​​are sorted in descending order, and the cumulative variance contribution rate is calculated. The top eigenvalues ​​with a cumulative contribution rate of 95% are selected. The first six principal components are used as features after dimensionality reduction. These six principal components constitute the asset panorama feature set, which is used for subsequent lifetime prediction.

[0034] Principal component analysis (PCA) is a statistical method that uses orthogonal transformation to convert a set of potentially correlated variables into a set of linearly uncorrelated variables (called principal components), thereby reducing data dimensionality while retaining most of the data information.

[0035] Table 1 shows some typical data segments after sensor acquisition and preprocessing, used to verify the effectiveness of data alignment and feature extraction.

[0036] Table 1 Sensor Acquisition and Preprocessing Feature Data ; The results show that the high-precision acquisition and processing procedures described above can transform heterogeneous physical signals into standard feature vectors that computers can understand. Furthermore, the corrected data is closer to the actual operating state of the equipment and eliminates environmental noise interference.

[0037] Please see Figure 1 and Figure 3 S2: Input the asset panoramic feature set into the nonlinear lifetime decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to solve the dynamic loss index.

[0038] The calculation process for the dynamic loss index specifically includes: S21: Import the asset panoramic feature set into the preset nonlinear lifetime decay model, use the Pearson correlation coefficient algorithm to calculate the mutual influence weights between the parameters of each dimension in the feature set, identify key coupling factors, and generate parameter coupling weight vectors.

[0039] S22: Based on the parameter coupling weight vector, the feature elements in the asset panorama feature set are weighted and recombined to construct a multi-dimensional feature association matrix.

[0040] S23: Set a time window of preset length, perform continuous definite integral operation on the trace of the multidimensional feature correlation matrix within the time window, calculate the cumulative fatigue damage of the equipment in the recent operating cycle, and generate a dynamic loss index after normalizing the cumulative fatigue damage.

[0041] The cumulative fatigue damage is obtained by integrating the trace of the multidimensional feature correlation matrix within the time window, specifically including: obtaining the matrix trace corresponding to each time step within the time window. Based on material fatigue attenuation coefficient Construct the integral equation: ; The integral equation is discretely solved using the trapezoidal integral method to obtain the incremental loss value within the current time window; then the incremental loss values ​​of multiple time windows are accumulated and nonlinear compensation is performed in combination with the environmental stress correction factor to generate the cumulative fatigue damage.

[0042] The specific logic for integral operations is as follows: Obtain the multidimensional feature correlation matrix sequence within the current time window and the preset material fatigue attenuation coefficient, and construct a numerical integral equation; The area under the eigenvalue change curve of the matrix within the sliding window is calculated using the trapezoidal rule. The calculated area value is then multiplied by the material fatigue attenuation coefficient to obtain the incremental loss value within a single sliding window cycle. The incremental loss values ​​generated by all sliding window cycles before the current moment are summed up, and an environmental stress correction factor is introduced to perform nonlinear compensation on the summation result to generate a dynamic loss index.

[0043] Read the asset panorama feature set generated in step S1, assuming it contains 6 principal component feature vectors. Select feature data from the most recent 1000 historical time steps to construct a sample matrix. For each pair of feature vectors (e.g., ... and The linear correlation coefficient is calculated using the Pearson correlation coefficient formula. The absolute values ​​of all calculated correlation coefficients are then summed row-wise. For example, The sum of the absolute values ​​of the correlation coefficients with the other five characteristics is 3.5. The sum is 1.2. This sum is used as the original weight for each feature dimension. The original weights are then normalized: the weight of each feature is divided by the sum of all feature weights to obtain the parameter coupling weight vector. If... The weighting is 0.45, indicating that It is most tightly coupled with other parameters and is a key coupling factor affecting lifespan.

[0044] The Pearson correlation coefficient mentioned above is a statistic used to measure the degree of linear correlation between two variables X and Y. Its value ranges from -1 to 1, and the larger the absolute value, the stronger the correlation.

[0045] The calculated weight vector is used to perform a weighted transformation on the original feature set. Let the feature set matrix be... The weight vector is Construct a diagonal matrix. Its diagonal elements are the weight vector. The corresponding value. Perform matrix multiplication. This yields the weighted feature matrix. Further calculation of the autocorrelation matrix of the weighted feature matrix is ​​then performed. The matrix This is the multidimensional feature correlation matrix. The diagonal elements in the matrix represent the energy of each feature after considering the coupling weights, and the off-diagonal elements represent the weighted covariance between features.

[0046] The time window length is set to 60 seconds (i.e., containing 60,000 millisecond-level sampling points). The multidimensional feature correlation matrix corresponding to each time step within the past 60 seconds is extracted from the cache. The trace of the matrix at each time step is calculated, which is the sum of the diagonal elements of the matrix; this value represents the total stress energy level of the system at the current time step. Time steps are recorded. The trace of the matrix is Read the fatigue degradation coefficient of the equipment from the materials database. For example, for silicon-based chip materials, Set as Construct the integral equation: Simultaneously, the same integration operation is performed on each component on the diagonal of the matrix to generate component loss values ​​for each dimension, providing data support for subsequent multidimensional mapping.

[0047] The integral above is discretized using the composite trapezoidal rule. The 60-second time window is divided into... a small interval (e.g.) (seconds). For the first For each interval, calculate the area of ​​the infinitesimal element: The total area is obtained by summing the areas of all infinitesimal elements. Assume the calculated total area is 50,000 units. Multiply this area by the fatigue attenuation coefficient: This value of 0.6 represents the incremental loss value within the 60-second sliding window.

[0048] Initialize cumulative loss variable (Initial value is 0). After each sliding window calculation, the incremental loss value is accumulated to... For example, the previous moment The initial value is 15.4, which accumulates to 16.0. Read the current ambient temperature. (e.g., 50 degrees Celsius, i.e., 323.15 Kelvin) and reference temperature (25 degrees Celsius, or 298.15 Kelvin). The environmental stress correction factor is calculated based on the Arrheniuse equation. Set activation energy The Boltzmann constant is 0.7 electron volts. for Electron volts / Kelvin. Calculation formula, exponent term: Calculate the correction factor: Calculate the final dynamic loss index: .like The current dynamic loss index is 131.2. Finally, this index is divided by the theoretical maximum lifespan loss threshold of the device (e.g., 1000), and normalized to a decimal between 0 and 1 (0.1312). Simultaneously, using the component loss values ​​of each dimension and the same correction logic, a multi-dimensional loss vector is generated and converted into a health mapping vector. (For example (This is for use in subsequent steps.)

[0049] The Arrheniuse equation is an empirical formula that describes the relationship between the rate constant of a chemical reaction and temperature. In reliability engineering, it is often used to describe the accelerated aging of electronic components at different temperatures.

[0050] Please see Figure 1 and Figure 4 S3: Obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the adaptation distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data.

[0051] The process of generating asset-task matching scores and dynamic judgment benchmarks specifically includes: S31: Parse the received task scheduling request, extract the computing resource consumption, I / O throughput requirements, and expected duration of the task, and construct the load vector of the task to be assigned.

[0052] S32: The range transformation method is used to project the task load vector to be assigned to the same dimensionless feature space as the dynamic loss index. The preset weighted Euclidean distance formula is called to calculate the fitting distance between the projected vector and the dynamic loss index, and an asset-task matching score is generated.

[0053] The formula for calculating the adaptation distance is as follows: ; in, This represents the asset-task matching score. Represents the total number of feature dimensions. Representing the The importance weight coefficients of each dimension The vector representing the load of the task to be assigned is in the th position. Normalized values ​​in each dimension The dynamic loss index represents the first Mapping component values ​​in each dimension.

[0054] S33: Retrieve historical matching score data of the asset when it successfully performed similar tasks in the past preset period, calculate the weighted moving average and standard deviation of the historical matching score data, set the upper and lower floating thresholds based on the confidence interval theory of normal distribution, and generate a dynamic judgment benchmark value.

[0055] Listen to the job scheduling system's API interface and intercept task request packets awaiting assignment. Parse the JSON-formatted task description file and extract key fields. For example, extract the following: CPU requirement is 8 cores with an estimated utilization of 80% (quantized as 80%), memory requirement is 16GB (quantized as 16), I / O read / write requests are 500MB / s (quantized as 500), and the estimated task duration is 2 hours (quantized as 7200 seconds). Construct the original load vector. .

[0056] Preset maximum and minimum limits for each dimension: CPU [0,100], Memory [0,64], I / O [0,1000], Time [0,86400]. Utilize the range transformation formula. right Perform normalization. CPU dimension: Memory dimension: = I / O dimension: Time dimension: The normalized load vector is obtained. Obtain the health mapping component of the current asset in the corresponding dimension. (Generated based on step S2). Assume The value is The higher the value, the healthier and more resilient that dimension is. Weighting coefficients are set. The weights are pre-set based on the task type (computation-intensive).

[0057] The range transformation method mentioned above refers to a linear transformation method that maps data to a specific interval (usually 0 to 1) to eliminate the scale difference between data of different dimensions and make them comparable.

[0058] The fitting distance is calculated using the following weighted Euclidean distance formula: ; in, This represents the asset-task matching score. The total number of feature dimensions (4 in this example). Representing the The importance weight coefficients of each dimension The vector representing the load of the task to be assigned is in the th position. Normalized values ​​in each dimension The dynamic loss index represents the first Mapping component values ​​in each dimension.

[0059] The obtained parameters (value is) ) and parameters (value is) Substituting into the above formula, we obtain the final result. The calculation process is as follows: Dimension 1 (CPU): Dimension 2 (Memory): The third dimension (I / O): 4th Dimension (Time): Summation: . √ root: The result indicates that the fit distance between the asset and the task is 0.398. Since this formula calculates "distance," a smaller value indicates a higher degree of matching (better match).

[0060] The database was queried to retrieve matching records of similar (computation-intensive) tasks that the asset had performed within the past 30 days without failure. Table 2 lists some of the historical matching data.

[0061] Table 2. Historical Task Matching Score Statistics ; Extract the S-value sequence from Table 2 There are a total of 100 samples. Calculate the arithmetic mean. Assuming Calculate the standard deviation. Assuming Based on a 95% confidence level, a Z-score of 1.96 was selected. The upper limit of the dynamic judgment benchmark was calculated: Calculate the lower limit of the dynamic judgment benchmark value: Set the dynamic judgment benchmark value to an interval. This range means that if the matching distance S of a new task falls within this range, then the asset is considered to be in its "comfort zone" in handling the task, and the risk is controllable.

[0062] The confidence interval theory mentioned above refers to the range of a population parameter estimated using sample statistics in statistics. The probability that the interval contains the population parameter is called the confidence level (e.g., 95%).

[0063] Please see Figure 1 and Figure 5 S4: Compare the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset preferred range conditions, generate an adaptive allocation instruction to drive the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

[0064] The process of generating and executing adaptive allocation instructions specifically includes: S41: Calculate the difference ratio between the asset task matching score and the dynamic judgment benchmark value, and determine whether the difference ratio falls within the preset preferred range.

[0065] S42: Based on the task priority level and the remaining maintenance window time of the asset, match the load balancing strategy from the preset strategy library, determine the asset start time, operating power mode and network bandwidth allocation quota, and generate adaptive allocation instructions.

[0066] The process of matching the optimal load balancing strategy specifically includes: Obtain the predicted remaining lifetime of all available assets in the system at the current moment and their real-time energy efficiency indicators, and construct a global resource status topology map; The particle swarm optimization algorithm is used to perform multi-objective optimization search in the global resource state topology graph to determine the allocation path combination; The parameter set corresponding to the allocation path combination is mapped to a specific control instruction code, and the control instruction code is embedded into the data packet of the adaptive allocation instruction.

[0067] S43: The adaptive dispatching command is sent to the underlying physical actuator through the industrial bus protocol, which drives the actuator to adjust the physical connection status and operating parameters of the equipment, and sends a closed-loop confirmation signal back to the management system after the execution is completed.

[0068] Read the current task matching distance calculated in step S3. and the dynamic determination of the benchmark value range First, determine Is it within the range? The conditions are met. Further calculation of the difference ratio is performed to assess the degree of deviation of the matched point relative to the mean: A result of 0.7 indicates that the current workload is slightly above the historical average (positive deviation), but still within one standard deviation. The preset optimal interval condition is: Because 0.7 is located at If the matching degree is determined to be "preferred", the system decides to assign the task to that asset.

[0069] The task priority is set to "P1 (High Priority)," and the asset has 48 hours until the next scheduled maintenance window. A search of the policy library yields a match for the "High Performance - Compact" load balancing policy. Based on the policy definition: 1. Startup Time: Set to "Immediate Startup" (0ms delay). 2. Operating Power Mode: Set to "Performance" mode, allowing CPU turbo boost up to 3.5GHz and fan speed set to 80% duty cycle. 3. Network Bandwidth: Allocate a dedicated Virtual LAN (VLAN) channel, guaranteeing a minimum bandwidth of 1Gbps. These parameters are then encoded to generate adaptive allocation instructions.

[0070] Iterate through all 100 server assets within the data center. Using the dynamic depreciation index from step S2, deduce the predicted remaining lifetime (RUL). For example, asset A has an RUL of 5000 hours, and asset B has an RUL of 200 hours. Simultaneously, read real-time energy efficiency (FLOPS / Watt) from the intelligent PDU (Power Distribution Unit). Construct a directed weighted topology graph where nodes represent assets, edges represent network links, and the weight of each node is determined by a vector consisting of (RUL, energy efficiency).

[0071] Initialize a population of 50 particles. Each particle's position vector represents a task allocation scheme (i.e., which tasks are assigned to which nodes). Define the fitness function: ,in The goal is to minimize total energy consumption and the RUL (Residual Energy Loss) of the occupied assets. Algorithm parameters are set as follows: maximum number of iterations 100, inertia weight... Individual learning factor Social learning factor Iteration begins: In the first iteration, the optimal fitness value is 0.85. In the 50th iteration, the optimal fitness value converges to 0.32. In the 100th iteration, the globally optimal position vector, i.e., the optimal combination of assignment paths, is output. For example, it determines to assign the current task to blade server number 5 in rack 3.

[0072] in, This represents the fitness function value of the particle swarm optimization algorithm; and These are the weighting coefficients; This represents the remaining lifetime value consumed during task execution. This indicates the energy cost associated with executing the task;

[0073] The aforementioned Particle Swarm Optimization (PSO) algorithm is a global optimization algorithm based on swarm intelligence. It simulates the foraging behavior of bird flocks and finds the optimal solution in the solution space through cooperation and information sharing among particles.

[0074] Convert the optimized parameters into hexadecimal control codes. Target IP address: 192.168.1.105->0xC0A80169. Power mode: Performance->0x01. Fan speed: 80%->0x50. Build the instruction packet payload: HEAD(0xAA)+ADDR(0xC0A80169)+PWR(0x01)+FAN(0x50)+CHKSUM(0x5E)+TAIL(0x55).

[0075] The management node establishes a socket connection with the underlying PLC or BMC (Baseboard Management Controller) via the Modbus TCP / IP protocol. The generated instruction packets are written to holding registers at addresses 40001 to 40005. The underlying actuator (such as the motherboard controller of a server) triggers a hardware interrupt upon detecting a change in the register value. The actions are as follows: 1. Adjust the CPU voltage regulator module (VRM) voltage to 1.2V. 2. Increase the fan speed to 3000 RPM via the PWM signal pin. 3. Configure the network card QoS queue. The execution time is approximately 200 milliseconds. Upon completion, the underlying actuator sets the status register address 40010 to 1 (Done). After the management system reads that 40010 is set to 1, it confirms that the task environment is ready and officially begins transmitting the task data stream. This closed-loop feedback mechanism ensures the physical implementation of the dispatch instructions.

[0076] A big data-based asset management system, used to execute the aforementioned big data-based asset management method, the system comprising: The panoramic feature extraction module is used to collect operational status time-series data and environmental stress parameters using sensing devices deployed at key nodes. Based on the timestamp synchronization protocol, the operational status time-series data and environmental stress parameters are processed in the time domain to extract the panoramic feature set of the asset. The loss index calculation module is used to input the asset panoramic feature set into the nonlinear life decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to calculate the dynamic loss index. The task suitability assessment module is used to obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the suitability distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data. The adaptive allocation module compares the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset optimal range conditions, an adaptive allocation instruction is generated, which drives the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

[0077] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the protection scope defined by the technical solution of the present invention.

Claims

1. A big data based asset management method, characterized by, Includes the following steps: S1: Collect time-series data of operational status and environmental stress parameters using sensing devices deployed at key nodes, and perform time-domain alignment processing on the operational status time-series data and environmental stress parameters based on a timestamp synchronization protocol to extract the asset panoramic feature set; S2: Input the asset panoramic feature set into the nonlinear lifetime decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to calculate the dynamic loss index. S3: Obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the adaptation distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data. S4: Compare the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset preferred range conditions, generate an adaptive allocation instruction to drive the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

2. The big data based asset management method according to claim 1, wherein, The extraction process of the asset panoramic feature set specifically includes: S11: The instantaneous vibration amplitude sequence of the equipment and the environmental temperature and humidity change sequence are synchronously collected by the high-frequency vibration sensor and the temperature and humidity transmitter. The heterogeneous data stream is labeled at the nanosecond level using the network time protocol, and abnormal data frames with a signal-to-noise ratio lower than the preset threshold are removed to generate the original state data stream. S12: The discrete points in the original state data stream are resampled using the Lagrange interpolation method, and the aligned multi-source data is mapped to a unified time axis coordinate system to generate a synchronization state matrix. S13: Perform Fast Fourier Transform and Wavelet Packet Decomposition on the synchronization state matrix to extract time-domain statistical features and frequency-domain energy spectrum features respectively. Use principal component analysis algorithm to perform dimensionality reduction and fusion processing on the extracted mixed features to generate the asset panoramic feature set.

3. The big data based asset management method of claim 1, wherein, The calculation process of the dynamic loss index specifically includes: S21: Import the asset panoramic feature set into a preset nonlinear lifetime decay model, use the Pearson correlation coefficient algorithm to calculate the mutual influence weights between parameters of each dimension in the feature set, identify key coupling factors, and generate a parameter coupling weight vector. S22: Based on the parameter coupling weight vector, the feature elements in the asset panoramic feature set are weighted and recombined to construct a multi-dimensional feature association matrix; S23: Set a time window of preset length, perform continuous definite integral operation on the trace of the multidimensional feature correlation matrix within the time window, calculate the cumulative fatigue damage of the equipment in the recent operating cycle, and generate the dynamic loss index after normalizing the cumulative fatigue damage.

4. The big data based asset management method of claim 1, wherein, The process of generating the asset task matching score and the dynamic judgment benchmark value specifically includes: S31: Parse the received task scheduling request, extract the computing resource consumption, I / O throughput requirements and expected duration of the task, and construct the load vector of the task to be assigned. S32: Project the task load vector to be assigned to the same dimensionless feature space as the dynamic loss index using the range transformation method, and calculate the fitting distance between the projected vector and the dynamic loss index by calling the preset weighted Euclidean distance formula, and generate the asset task matching score. S33: Retrieve historical matching score data of the asset when it successfully performed similar tasks in the past preset period, calculate the weighted moving average and standard deviation of the historical matching score data, set the upper and lower floating thresholds based on the confidence interval theory of normal distribution, and generate the dynamic judgment benchmark value.

5. The asset management method based on big data according to claim 1, characterized in that, The process of generating and executing the adaptive allocation instruction specifically includes: S41: Calculate the difference ratio between the asset task matching score and the dynamic judgment benchmark value, and determine whether the difference ratio falls within the preset preferred range; S42: Based on the task priority level and the remaining maintenance window time of the asset, match the load balancing strategy from the preset strategy library, determine the asset start time, operating power mode and network bandwidth allocation quota, and generate the adaptive allocation instruction. S43: The adaptive dispatching command is sent to the underlying physical actuator via the industrial bus protocol, driving the actuator to adjust the physical connection status and operating parameters of the equipment, and after execution, a closed-loop confirmation signal is sent back to the management system.

6. The asset management method based on big data according to claim 2, characterized in that, The resampling process in S12 specifically includes: Obtain two adjacent sampling time points and their corresponding state values ​​in the original state data stream, and calculate the time difference ratio between the time point to be interpolated and the sampling time points before and after. A cubic spline interpolation function model is constructed, and the time difference ratio is substituted into the interpolation function model to calculate the estimated state values ​​of the points to be interpolated. The least squares method is then used to smooth and correct the estimated results, thereby generating a resampled data sequence. The resampled data sequences are merged into column vectors according to a unified timestamp index to construct the synchronization state matrix.

7. The asset management method based on big data according to claim 3, characterized in that, The specific logic for the integral operation in S23 is as follows: Obtain the multidimensional feature correlation matrix sequence and the preset material fatigue attenuation coefficient within the current time window, and construct a numerical integral equation; The area under the eigenvalue change curve of the matrix within the sliding window is calculated using the trapezoidal rule. The calculated area value is then multiplied by the material fatigue attenuation coefficient to obtain the incremental loss value within a single sliding window cycle. The incremental loss values ​​generated by all sliding window cycles before the current moment are summed, and an environmental stress correction factor is introduced to perform nonlinear compensation on the summation result to generate the dynamic loss index.

8. The asset management method based on big data according to claim 4, characterized in that, The formula for calculating the adaptation distance in S32 is as follows: ； in, This represents the asset-task matching score. Represents the total number of feature dimensions. Representing the The importance weight coefficients of each dimension The load vector of the task to be assigned represents the first... Normalized values ​​in each dimension The dynamic loss index represents the first... Mapping component values ​​in each dimension.

9. The asset management method based on big data according to claim 5, characterized in that, The process of matching the optimal load balancing strategy in S42 specifically includes: Obtain the predicted remaining lifetime of all available assets in the system at the current moment and their real-time energy efficiency indicators, and construct a global resource status topology map; The particle swarm optimization algorithm is used to perform multi-objective optimization search in the global resource state topology graph to determine the allocation path combination; The parameter set corresponding to the allocation path combination is mapped to a specific control instruction code, and the control instruction code is embedded into the data packet of the adaptive allocation instruction.

10. A big data-based asset management system, characterized in that, The system is used to implement the asset management method based on big data as described in any one of claims 1-9, and the system comprises: The panoramic feature extraction module is used to collect operational status time-series data and environmental stress parameters using sensing devices deployed at key nodes, and to perform time-domain alignment processing on the operational status time-series data and environmental stress parameters based on a timestamp synchronization protocol to extract the asset panoramic feature set. The loss index calculation module is used to input the asset panoramic feature set into the nonlinear lifetime decay model, analyze the parameter coupling relationship based on covariance analysis logic and construct a multidimensional feature correlation matrix, and use the time sliding window algorithm to perform integral operation on the multidimensional feature correlation matrix to calculate the dynamic loss index. The task suitability assessment module is used to obtain the task load vector to be assigned, map the task load vector to be assigned to the dimension space where the dynamic loss index is located, calculate the suitability distance between the task load vector to be assigned and the dynamic loss index based on the weighted Euclidean distance formula, generate an asset task matching score, and generate a dynamic judgment benchmark value based on historical operation data. The adaptive allocation module is used to compare the asset task matching score with the dynamic judgment benchmark value. If the asset task matching score meets the preset preferred range conditions, an adaptive allocation instruction is generated to drive the underlying physical execution mechanism to automatically complete the asset allocation according to the adaptive allocation instruction.

Images