Liquid cooling CDU distribution early warning method, system and device based on intelligent monitoring

Abstract

This application relates to the field of monitoring and early warning technology, and provides a method, system, and device for early warning of liquid-cooled CDU allocation based on intelligent monitoring. It collects flow and temperature data from each branch of the CDU. A global probability distribution is obtained by dividing the flow data into intervals and performing frequency statistics. The marginal probability distribution of each branch is obtained by separately statistically analyzing the flow data. A coupling index characterizing the flow coupling relationship between branches is calculated based on the global and marginal probability distributions. An allocation entropy index is calculated based on the global probability distribution. The allocation entropy index and coupling index are compared with their respective historical benchmarks to obtain a comprehensive early warning index. The current early warning level is determined by combining the above information. When the early warning level reaches a preset level, an early warning notification is output to the operation and maintenance platform and management terminal. This completes the entire processing flow from perception, analysis, judgment to feedback, improving the efficiency and accuracy of early warning for abnormal liquid-cooled CDU allocation.

Description

Technical Field

[0001] This application relates to the field of monitoring and early warning technology, and more specifically, to a method, system and device for early warning of liquid-cooled CDU allocation based on intelligent monitoring. Background Technology

[0002] Cooling Distribution Units (CDUs) in liquid cooling systems are core heat dissipation devices for AI data centers and high-performance computing clusters. Their secondary side typically has dozens of parallel branches, each leading to different server racks. Due to differences in pipe diameter, length, and valve opening, uneven flow distribution among the branches has become a common problem affecting heat dissipation efficiency. Existing monitoring solutions mainly rely on installing flow meters on each branch to obtain individual flow values. An alarm is triggered when the absolute flow value of a particular branch falls below a preset threshold, or maintenance personnel periodically compare the flow data of each branch for manual judgment.

[0003] However, single-branch threshold alarms can only detect problems when the flow rate has deviated significantly from the normal value. At this time, the chip being served may have been operating at high temperature for a long time and cannot issue an early warning in the early stage of the distribution imbalance. At the same time, each branch is monitored as an independent entity, ignoring the overall uniformity of flow distribution between branches and the flow coupling relationship between adjacent branches. It is difficult to distinguish between systemic flow redistribution and local branch failure, resulting in delayed warnings and a high false alarm rate. Summary of the Invention

[0004] This application provides a method, system, and device for early warning of liquid-cooled CDU allocation based on intelligent monitoring, thereby solving the problem of low efficiency and accuracy of early warning of liquid-cooled CDU allocation in the prior art.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to one aspect of this application, a liquid-cooled CDU allocation early warning method based on intelligent monitoring is provided, comprising: collecting flow signals and temperature signals of each branch in the cooling capacity distribution unit (CDU); generating flow data and temperature data respectively after signal processing and storing them in a data buffer; extracting the flow data of each branch from the data buffer based on a preset time window; obtaining a global probability distribution characterizing the overall flow distribution characteristics of the CDU by dividing the flow data into intervals and performing frequency statistics; obtaining the marginal probability distribution of each branch by separately statistically analyzing the flow data of each branch; and calculating an allocation entropy index to characterize the uniformity of the overall flow distribution of the CDU based on the global probability distribution. The system calculates coupling indices to characterize the flow coupling relationship between branches based on the global probability distribution and the marginal probability distribution. It then compares the allocation entropy index and the coupling index with their respective historical benchmarks and performs a fusion calculation based on the relative relationship between the allocation entropy index and the coupling index to obtain a comprehensive early warning index. Based on the deviation of the current comprehensive early warning index from the historical normal baseline, the dynamic change trend of the comprehensive early warning index within the time window, and the temperature data, the current early warning level is determined. When the early warning level reaches a preset level, an early warning notification is output to the operation and maintenance platform and management terminal, and the current operating data and early warning results are stored in the historical database.

[0007] In this application, based on the aforementioned scheme, the process of acquiring flow and temperature signals from each branch of the cooling capacity distribution unit (CDU), generating flow and temperature data respectively after signal processing, and storing them in a data buffer includes: acquiring analog signals corresponding to flow velocity through flow sensors at the inlet of each branch of the cooling capacity distribution unit (CDU), acquiring analog signals corresponding to temperature through temperature sensors at the inlet of each branch or the inlet / outlet of the cold plate, and transmitting the analog signals to the analog input channel of the data acquisition terminal; synchronously sampling and performing analog-to-digital conversion on the analog signals in the analog input channel to generate an original digital sequence; filtering the original digital sequence to generate filtered digital quantities, converting the filtered digital quantities into engineering physical quantities according to the calibration coefficients of each analog input channel, as flow and temperature data; and storing the flow and temperature data in a circular queue in the data buffer.

[0008] In this application, based on the aforementioned scheme, the step of extracting traffic data of each branch from the data buffer based on a preset time window, obtaining a global probability distribution characterizing the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics, and obtaining the marginal probability distribution of each branch by separately statistically analyzing the traffic data of each branch, includes: extracting traffic values ​​of each branch from the data buffer based on a preset time window to construct a traffic matrix; dividing the traffic values ​​in the traffic matrix into several intervals according to their numerical values; counting the first frequency of the traffic values ​​falling into each interval, dividing the first frequency by the total number of traffic values ​​to obtain the interval probability of the interval, and combining the interval probabilities of each interval to obtain a global probability distribution characterizing the overall traffic distribution of the CDU; extracting the traffic value sequence of each branch from the traffic matrix, counting the second frequency of the traffic values ​​in the traffic value sequence falling into the interval, and dividing the second frequency by the number of sampling points of the branch to obtain the marginal probability distribution of the branch.

[0009] In this application, based on the aforementioned scheme, the step of calculating the allocation entropy index to characterize the uniformity of overall CDU flow distribution according to the global probability distribution, and calculating the coupling index to characterize the flow coupling relationship between branches according to the global probability distribution and the marginal probability distribution, includes: calculating information entropy based on the interval probabilities in the global probability distribution, determining a probability deviation factor based on the degree of deviation of the interval probabilities from the average probability, and superimposing the probability deviation factor on the information entropy to obtain the allocation entropy index; for any two branches, generating a joint probability distribution based on the joint frequency of the flow values ​​of the two branches falling into each interval combination at each sampling time within the same time window, and generating mutual information between the two branches by combining the marginal probability distributions of the two branches; calculating the mutual information between the branch and other branches, and weighting and summing the mutual information based on the physical pipeline distance between the branches to obtain the spatial coupling degree of the branch relative to other branches; and averaging the spatial coupling degrees of all branches to obtain the coupling index to characterize the flow coupling relationship between branches.

[0010] In this application, based on the aforementioned scheme, the step of comparing the allocation entropy index and the coupling index with their respective historical benchmarks, and then performing a fusion calculation based on the relative relationship between the allocation entropy index and the coupling index to obtain a comprehensive early warning index, includes: obtaining the first mean and first standard deviation of the allocation entropy index under the historical benchmark state; obtaining the second mean and second standard deviation of the coupling index under the historical benchmark state; subtracting the first mean from the allocation entropy index and dividing by the first standard deviation to obtain a standardized allocation entropy deviation value; subtracting the second mean from the coupling index and dividing by the second standard deviation to obtain a standardized coupling deviation value; calculating the ratio of the allocation entropy index to the coupling index as a dynamic ratio term; and weighted summing the standardized allocation entropy deviation value, the standardized coupling deviation value, and the dynamic ratio term to obtain a comprehensive early warning index.

[0011] In this application, based on the aforementioned scheme, determining the current warning level according to the deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data includes: obtaining the third mean of the comprehensive warning index under the historical baseline state; subtracting the third mean from the comprehensive warning index to obtain the deviation; calculating the difference between the comprehensive warning indexes between multiple adjacent time windows and taking the average value to obtain the average rate of change of the comprehensive warning index within the most recent multiple time windows; setting a deviation threshold range and a rate of change threshold range based on the statistical quantiles of the distribution of the comprehensive warning index under the historical baseline state; and determining the current warning level according to the deviation threshold range into which the deviation falls, the rate of change threshold range into which the average rate of change falls, and the temperature data.

[0012] In this application, based on the aforementioned scheme, the step of outputting an early warning notification to the operation and maintenance platform and management terminal when the early warning level reaches a preset level, and storing the current operating data and early warning results in the historical database, includes: when the early warning level reaches a preset level, calculating the change in coupling indicators of each branch between adjacent time windows, and determining the branch that contributes the most to the change in coupling indicators as the target location result based on the proportion of the change in each branch to the total change in all branches; pushing the current early warning level, comprehensive early warning index, and the target location result to the operation and maintenance management platform, and visually displaying them on the monitoring interface using color indicators and trend charts; generating an early warning notification, and sending the early warning notification to the management terminal through a preset communication channel.

[0013] According to one aspect of this application, a liquid-cooled CDU distribution early warning system based on intelligent monitoring is provided, comprising: The acquisition module is used to acquire the flow and temperature signals of each branch in the cooling capacity distribution unit (CDU), and after signal processing, generate flow data and temperature data respectively and store them in the data buffer area. The extraction module is used to extract traffic data of each branch from the data buffer based on a preset time window, and obtain a global probability distribution that characterizes the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics. The marginal probability distribution of each branch is obtained by separately statistically analyzing the traffic data of each branch. The index module is used to calculate the allocation entropy index, which characterizes the uniformity of the overall traffic distribution of the CDU, based on the global probability distribution; and to calculate the coupling index, which characterizes the traffic coupling relationship between branches, based on the global probability distribution and the marginal probability distribution. The allocation entropy index and the coupling index are compared with their respective historical benchmarks, and the relative relationship between the allocation entropy index and the coupling index is combined for fusion calculation to obtain a comprehensive early warning index. The level module is used to determine the current warning level based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data. The early warning module is used to output early warning notifications to the operation and maintenance platform and management terminal when the early warning level reaches the preset level, and to store the current operation data and early warning results in the historical database.

[0014] According to one aspect of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the liquid-cooled CDU allocation early warning method based on intelligent monitoring as described in the above embodiments.

[0015] According to one aspect of this application, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the intelligent monitoring-based liquid-cooled CDU allocation early warning method as described in the above embodiments.

[0016] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the intelligent monitoring-based liquid-cooled CDU allocation early warning method provided in the various optional implementations described above.

[0017] The main differences and technical effects of the technical solution of this application compared with the prior art are as follows: This technical solution provides a synchronous and accurate raw data foundation for subsequent analysis by collecting and processing flow and temperature signals. By constructing a global probability distribution and marginal probability distributions for each branch, the raw flow data is unified into the same probability space for quantification, eliminating the incomparability caused by differences in numerical scales between branches, and ensuring that subsequent statistical characteristics accurately reflect the overall distribution status of the system. Based on this, the system state is characterized from two orthogonal dimensions—uniformity and synergy—using distribution and coupling indicators respectively. These indicators are then compared with historical benchmarks and fused into a comprehensive early warning index, enabling the system to detect early signs of distribution imbalance. By analyzing the deviation and trend of the comprehensive early warning index, combined with cross-validation of temperature data, a multi-level early warning system with both foresight and specificity is achieved. Finally, information is output and data is stored when an early warning is triggered, completing the entire processing from perception, analysis, judgment to feedback, improving the efficiency and accuracy of early warning for abnormal distribution in liquid-cooled CDUs.

[0018] 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

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 The flowchart of a liquid-cooled CDU allocation early warning method based on intelligent monitoring is illustrated in one embodiment of this application.

[0021] Figure 2 The flowchart illustrating the determination of the warning level is shown in one embodiment of this application.

[0022] Figure 3 The illustration shows a schematic diagram of a liquid-cooled CDU distribution early warning system based on intelligent monitoring in one embodiment of this application.

[0023] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0024] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, fully conveying the concept of the exemplary embodiments to those skilled in the art.

[0025] It should be noted that the data acquisition or information collection in this embodiment is performed after authorization by the user or the object of collection, and its process and purpose strictly follow the relevant regulations.

[0026] The block diagrams shown in the attached figures are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more hardware modules composed of smart chips, smart integrated circuits, or application-specific integrated circuits (ASICs), or in different network and / or processor devices and / or microcontroller devices.

[0027] The implementation details of the technical solution of this application are described below: Figure 1 A flowchart of a liquid-cooled CDU allocation early warning method based on intelligent monitoring according to an embodiment of this application is shown. (Refer to...) Figure 1 As shown, the intelligent monitoring-based liquid-cooled CDU allocation early warning method includes at least steps S110 to S150, which are described in detail below: S110 collects the flow and temperature signals of each branch in the cooling capacity distribution unit (CDU), and generates flow and temperature data respectively after signal processing and stores them in the data buffer area.

[0028] In this embodiment, the flow sensors at the inlet of each branch of the cooling capacity distribution unit (CDU) output analog electrical signals proportional to the instantaneous flow rate. Simultaneously, temperature sensors at each branch inlet or cold plate inlet / outlet convert the coolant temperature into resistance changes, which are then converted into standard analog signals by a transmitter. These two types of analog signals are transmitted in parallel through shielded cables to the analog input channel of the data acquisition terminal.

[0029] In this embodiment, the acquisition terminal has a built-in multi-channel analog-to-digital converter. A hardware timer generates strictly equal-interval sampling trigger pulses to perform sampling and analog-to-digital conversion on all channels, latching the converted values ​​of all channels to facilitate strict alignment of data from each branch on the time axis. The original digital sequence is then preprocessed to eliminate measurement errors introduced by electromagnetic interference and sensor thermal noise. Based on the calibration coefficients of each channel, the filtered digital code values ​​are converted into engineering physical quantities to obtain the current flow rate and temperature values ​​of each branch. Finally, the data, along with the timestamp, is written to a circular queue in memory.

[0030] In one embodiment of this application, the flow rate and temperature signals of each branch in the cooling capacity distribution unit (CDU) are collected, and after signal processing, flow rate data and temperature data are generated and stored in a data buffer, including: The flow rate is collected by the flow sensor at the inlet of each branch in the cold distribution unit (CDU), and the temperature is collected by the temperature sensor at the inlet of each branch or the inlet and outlet of the cold plate. The analog signals are then transmitted to the analog input channel of the data acquisition terminal. The analog signals in the analog input channel are synchronously sampled and converted from analog to digital to generate the original digital sequence; The original digital sequence is filtered to generate filtered digital quantities. Based on the calibration coefficients of each analog input channel, the filtered digital quantities are converted into engineering physical quantities, which are used as flow data and temperature data. The flow rate data and temperature data are stored in a circular queue in the data cache area.

[0031] In this embodiment, based on the principle of turbine rotation frequency measurement or ultrasonic time-of-flight velocity measurement, sensors are installed at the inlet and outlet of each branch in the Cooling Distribution Unit (CDU). Specifically, a flow sensor continuously outputs an analog electrical signal proportional to the instantaneous flow velocity passing through the sensor, while a platinum resistance temperature sensor installed at each branch outlet converts the coolant temperature into a change in resistance value, which is then converted into a standard analog electrical signal by a transmitter. The analog signal is transmitted in parallel to the analog input channel of the data acquisition terminal via a shielded twisted-pair cable.

[0032] In this embodiment, the data acquisition terminal is an industrial-grade embedded smart gateway with a built-in multi-channel high-precision analog-to-digital converter. Each channel is equipped with an independent anti-aliasing low-pass filter to suppress high-frequency noise. The acquisition terminal runs a real-time operating system, which generates strictly equal-interval sampling trigger pulses through a hardware timer. It performs synchronous sampling on all channels at a preset sampling period. When the timer count overflows and an interrupt occurs, the firmware driver reads the conversion completion register of each channel's analog-to-digital converter in sequence and latches the values ​​of all channels. This helps to strictly align the flow and temperature signals of each branch on the time axis, eliminates the phase difference introduced by time-division sampling, and generates the original digital sequence.

[0033] Next, the original digital sequence is preprocessed. First, a median filtering operation is performed, which sorts the multiple raw values ​​continuously acquired by each channel and takes the median value to remove isolated pulse noise caused by electromagnetic interference. Then, a first-order hysteresis filter is performed, which smooths the random fluctuations caused by the sensor's own thermal noise and fluid turbulence by weighting the current raw value with the previous filtered value according to a preset ratio, thereby effectively improving the signal-to-noise ratio of the signal.

[0034] After filtering, the preprocessing module converts the filtered digital code values ​​into engineering physical quantities with clear dimensions based on the calibration coefficients of each channel, including the sensitivity coefficients obtained through factory verification and the offsets obtained from on-site zero-point and full-scale calibration. For flow signals, the voltage or current values ​​are converted into volumetric flow rates per unit time. For temperature signals, the resistance values ​​are converted into degrees Celsius based on the inverse fitting polynomial of the platinum resistance calibration table, thereby obtaining the flow and temperature data of each branch at the current moment.

[0035] Finally, the current flow and temperature data of each branch, along with the timestamp, are assembled into a complete data record. This record is then written directly to the circular queue in the data buffer at the position pointed to by the current write pointer. The capacity of the circular queue is pre-configured to accommodate at least the data volume of multiple consecutive sliding windows. The write pointer increments with each sampling period. When the pointer reaches the end of the queue, it automatically returns to the beginning position and overwrites the oldest data record, thus maintaining a fixed-length sequence of the most recent historical data. This allows subsequent time-window-based data extraction operations to quickly access continuous data within a specified time period without locking, contributing to the real-time performance and efficiency of data access.

[0036] As an example, in this embodiment, the time window length can be set to 60 seconds, the sliding step size to 10 seconds, and the sampling period to 1 second. Then, a single window contains data from 60 sampling times, and the circular queue capacity is configured to at least 3600 records, which can store data from the most recent 60 windows to support multi-window trend analysis.

[0037] The above process, by synchronously acquiring flow and temperature signals from each branch and performing filtering, noise reduction, and calibration during signal conditioning, transforms the raw analog signals into engineering data with uniform dimensions, which is then stored in a buffer. This helps ensure strict alignment of the raw data on the time axis and the accuracy of physical quantities for subsequent analysis. The circular queue storage structure allows historical data to be continuously updated in fixed-length scrolling fashion, providing an efficient memory access basis for sliding window extraction operations. This avoids data fragmentation and redundant I / O overhead, contributing to improved real-time performance and reliability of the early warning system from the data source.

[0038] S120, based on a preset time window, extract traffic data of each branch from the data buffer, and obtain a global probability distribution characterizing the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics, and obtain the marginal probability distribution of each branch by separately statistically analyzing the traffic data of each branch.

[0039] In this embodiment, at each monitoring time, the flow values ​​of all branches within the past preset window are retrieved from the data buffer using a preset time window (fixed sliding step size), constructing a two-dimensional flow matrix. The observed values ​​in the matrix are divided into multiple intervals with equal width according to their numerical range. The frequency of the observed values ​​in each interval is counted and normalized to obtain a global probability distribution characterizing the overall flow distribution of all branches. Simultaneously, for each branch, the flow sequence of that branch is extracted from the same matrix. Using the same interval boundaries as the global division, the flow sequence of each branch is counted and normalized to obtain an independent marginal probability distribution for each branch. This provides input data in probability space for subsequent calculations of statistical dependencies between branches and overall distribution uniformity.

[0040] In one embodiment of this application, based on a preset time window, traffic data for each branch is extracted from the data buffer. By dividing the traffic data into intervals and performing frequency statistics, a global probability distribution characterizing the overall traffic distribution of the CDU is obtained. The marginal probability distribution of each branch is obtained by separately statistically analyzing the traffic data for each branch, including: Based on a preset time window, the flow values ​​of each branch are extracted from the data cache to construct a flow matrix; The flow values ​​in the flow matrix are divided into several intervals according to their numerical values; The first frequency of the flow value falling into each interval is counted, and the first frequency is divided by the total number of flow values ​​to obtain the interval probability. The interval probabilities of each interval are combined to obtain the global probability distribution characterizing the overall flow distribution of the CDU. Extract the flow value sequence of each branch from the flow matrix, count the second frequency of the flow value in the flow value sequence falling into the interval, divide the second frequency by the number of sampling points of the branch, and obtain the marginal probability distribution of the branch.

[0041] In this embodiment, at each monitoring moment, a window extraction operation is triggered with a fixed sliding step size to retrieve the flow values ​​of all branches within a preset time window at each sampling moment from the circular queue in the data buffer. The retrieval operation directly accesses the specified address range of the circular queue through a memory pointer, and organizes all flow records within the hit time period by using the branch number as the row index and the time order as the column index to construct a flow matrix.

[0042] Specifically, the rows of the flow matrix correspond to the flow time series of a branch at all sampling times within the window, and the columns of the flow matrix correspond to the flow cross-sectional data of all branches at the same sampling time, providing a unified data view for subsequent global statistics and independent branch statistics.

[0043] Next, for the constructed flow matrix, all the flow values ​​it contains are expanded into a one-dimensional sequence, and the minimum and maximum values ​​in the sequence are determined to define the overall numerical range. Based on the preset number of interval divisions, the numerical range is divided into multiple continuous and non-overlapping intervals of equal width. Each interval has the same width, which is equal to the numerical range divided by the number of interval divisions. This helps to ensure that each flow value uniquely falls into one of the intervals, establishing a unified discretization mapping rule for subsequent frequency statistics.

[0044] Specifically, in this embodiment, the number of interval divisions K is determined based on the total number of branches N, and can be taken as K = round(log2N) + 1 to ensure that the number of intervals is adaptive in systems with different numbers of branches. For a typical CDU system with N=32, K=6 can be taken.

[0045] Next, the frequency of the flow value falling into each interval is counted. All flow values ​​in the entire flow matrix are traversed. For each observation, the interval to which the value belongs is determined and the frequency counter of the corresponding interval is incremented once. After all observations have been traversed, the frequency vector of each interval is obtained, which is used as the first frequency.

[0046] Next, the first frequency of each interval is divided by the total number of all traffic values ​​to perform normalization, resulting in a global probability distribution. Each component in the global probability distribution represents the proportion of all traffic values ​​of all branches within the current time window that fall within the corresponding traffic interval range, thus characterizing the concentration or dispersion of traffic values ​​in the CDU secondary branch network as a whole.

[0047] Meanwhile, the marginal probability distribution of each branch is statistically analyzed separately. The entire row of data corresponding to that branch is extracted from the flow matrix to obtain the flow value sequence for all sampling times within the window. Using the same interval boundaries and number of intervals as the global partition, the second frequency of the branch's flow value falling into each interval is counted. This second frequency is then normalized by dividing by the total number of sampling points for that branch, yielding the marginal probability distribution of that branch.

[0048] It should be noted that the above process adopts the same interval division method as the global process, which helps to define the probability distribution of different branches under the same set of numerical interval systems, so that the subsequent cross-branch mutual information calculation has a unified symbol table basis.

[0049] The above process, by discretizing all flow values ​​of each branch within the same time window into a unified numerical interval system, eliminates the inconsistency in numerical scales caused by differences in sensor range or installation location. This allows the originally scattered branch flow data to be quantitatively compared within a unified probability space. The construction of the global probability distribution characterizes the degree of concentration or dispersion of flow values ​​across all branches at the overall level, while the independent statistics of the marginal probability distributions of each branch preserve the fluctuation characteristics of each individual branch. The synchronous output of both provides a structured and computable probability input for subsequent comprehensive analysis of the allocation status from the two dimensions of overall uniformity and local correlation. At the same time, the window sliding mechanism effectively filters out instantaneous impulse noise, allowing statistical characteristics to focus on the macroscopic behavioral patterns of the system.

[0050] S130, calculate the allocation entropy index to characterize the uniformity of overall CDU traffic distribution based on the global probability distribution, calculate the coupling index to characterize the traffic coupling relationship between branches based on the global probability distribution and the marginal probability distribution, compare the allocation entropy index and the coupling index with their respective historical benchmarks, and perform a fusion calculation based on the relative relationship between the allocation entropy index and the coupling index to obtain a comprehensive early warning index.

[0051] In this embodiment, based on the probability values ​​of each interval in the global probability distribution, a probability deviation factor that is positively correlated with the degree of deviation of the probability of each interval from the average probability is superimposed on the information entropy to calculate the allocation entropy index, which characterizes the uniformity of the overall traffic distribution of the CDU.

[0052] Meanwhile, for any two branches, the joint frequency of their flow values ​​falling into each interval combination at the same sampling time is counted to construct a joint probability distribution. The mutual information between branches is calculated by combining the marginal probability distribution of each branch. Then, the mutual information is weighted and summed according to the physical pipeline distance between branches and averaged to obtain a coupling index that characterizes the flow coupling relationship between branches.

[0053] Subsequently, the mean of each in the historical baseline state is subtracted from the distribution entropy index and the coupling index, and then divided by the corresponding standard deviation to obtain two standardized deviation values. At the same time, the ratio of the distribution entropy index to the coupling index is calculated as a dynamic ratio term. The two standardized deviation values ​​and the dynamic ratio term are added together according to their respective weights to obtain the comprehensive early warning index.

[0054] It should be noted that the allocation entropy index in this application is a one-dimensional real-valued index constructed based on the global probability distribution to quantify the uniformity of the flow distribution of each branch on the secondary side of the CDU across a numerical interval. In the ideal case of complete uniformity, the probability of each interval is equal, and the allocation entropy index takes the theoretical maximum value; when the flow concentrates in certain intervals, the allocation entropy index decreases, and the magnitude of the decrease reflects the degree of flow concentration.

[0055] It should be noted that the coupling index in this application quantifies the statistical coordination of flow changes among all branches by calculating the average value of the weighted coupling degree of each branch (i.e., the sum of mutual information between branches weighted by pipeline distance). The larger the coupling index, the more synchronized the flow changes among branches are; a significant decrease in the coupling index indicates that the coordination relationship between branches has been disrupted, possibly due to a local fault in one or more branches.

[0056] In one embodiment of this application, an allocation entropy index characterizing the uniformity of overall CDU traffic distribution is calculated based on the global probability distribution, and a coupling index characterizing the inter-branch traffic coupling relationship is calculated based on the global probability distribution and the marginal probability distribution, including: Information entropy is calculated based on the interval probabilities in the global probability distribution. A probability deviation factor is determined based on the degree to which the interval probabilities deviate from the average probability. The probability deviation factor is then superimposed on the information entropy to obtain the allocation entropy index. For any two branches, a joint probability distribution is generated based on the joint frequency of the flow values ​​of the two branches falling into each interval combination at each sampling time within the same time window. The mutual information between the two branches is then generated by combining the marginal probability distributions of the two branches. Calculate the mutual information between the branch and other branches, and perform a weighted summation of the mutual information based on the physical pipeline distance between the branches to obtain the spatial coupling degree of the branch relative to other branches; The average spatial coupling degree of all branches is taken to obtain a coupling index used to characterize the flow coupling relationship between branches.

[0057] First, the information entropy is calculated based on the probabilities of each interval in the global probability distribution, and the deviation of each interval probability from the average probability is also calculated. Divide the deviation by the standard deviation of the global probability distribution. The standardized deviation is obtained. Adaptive weighting coefficients are determined based on the relative magnitude of the total number of branches and the number of intervals. The standardized deviation is squared and multiplied by the adaptive weighting coefficients to construct the probability deviation factor. A probability deviation factor is superimposed on the information entropy, and the result of adding 1 to the probability deviation factor is used as the multiplier factor. Multiply by the corresponding interval probability and its logarithm Then, sum them up and take the negative value to obtain the distribution entropy index. The above calculation of the distribution entropy index... The process can be described as follows: Where j and K represent the identifier and total number of intervals, respectively. This represents the probability that all flow values ​​of all branches fall into the j-th interval. = N / (N+K) represents the adaptive weighting coefficient, which is used to control the strength of the penalty term, where N is the total number of branches; represents the standard deviation of the average probability and the global probability distribution, respectively.

[0058] It should be noted that when the interval probability is zero, the entire term is set to zero because the limit value converges to zero; when the probability is a very small positive floating-point number, the lower limit of the logarithmic calculation is used to prevent overflow to negative infinity and ensure numerical stability.

[0059] This application introduces a penalty factor that is positively correlated with the degree of deviation of the probability from the average. When the flow is concentrated in a few intervals, the penalty factor significantly amplifies the entropy value, making the distribution entropy index more sensitive to uneven states and enabling it to generate a detectable numerical response at the initial stage of distribution imbalance. The distribution entropy index obtained through the above calculation is used to quantify the uniformity of the distribution of all branch flow values ​​within the current window across the numerical intervals. The larger the value, the more dispersed (uniform) the flow is distributed across the intervals; the smaller the value, the more concentrated the flow is in a few intervals (uneven).

[0060] Next, for any two branches, the flow value sequences of these two branches at each sampling time within the same time window are extracted from the flow matrix, forming multiple flow tuples at multiple sampling times. For each tuple, according to a globally unified interval partitioning scheme, the interval number to which the flow value of the first branch belongs and the interval number to which the flow value of the second branch belongs are determined, forming an interval number combination. All sampling times are traversed, and the joint frequency of each interval number combination is counted. Each joint frequency is divided by the total number of sampling points to obtain the joint probability distribution of these two branches, where each element represents the probability that the flow values ​​of the two branches simultaneously fall into their respective intervals. Subsequently, combined with the marginal probability distribution of each branch, the mutual information between these two branches is calculated to quantify the statistical dependence of the flow changes of the two branches.

[0061] Then, for each target branch, the system reads the preset physical topology distance matrix, obtains the pipeline distance between the target branch and other branches in the network, and calculates the global average value of the pipeline distance between all branches.

[0062] Specifically, for a branch formed by the target branch and any other branch, the mutual information between these two branches will be... Multiplied by an exponential decay factor The exponential decay factor has a base of the natural exponent e and is based on the pipe distance between the two branches. The ratio to the global average distance D is calculated using the negative exponent. The closer the pipe distance, the closer the exponential decay factor is to one, indicating a higher contribution weight of mutual information; the farther the pipe distance, the closer the exponential decay factor is to zero, indicating a lower contribution weight of mutual information. This weighting method is consistent with the physical characteristic of stronger flow coupling between adjacent branches in a fluid network. The weighted mutual information of the target branch and all other branches is accumulated to obtain the spatial coupling degree of that branch relative to all other branches. for: in, This represents the mutual information between branch i and branch j. The distance between branch i and branch j in the physical topology is represented by , and D represents the average distance between all branches, i.e., the global average distance.

[0063] After calculating the spatial coupling degree for each branch, the spatial coupling degree value for each branch is obtained. These values ​​are summed and divided by the total number of branches to obtain the average spatial coupling degree of the entire branch network. This average coupling degree serves as a coupling index characterizing the flow coupling relationship among all branches. This coupling index reflects the overall strength of the statistical dependence between the flow of each branch at a global level. The coupling index is larger when the degree of synchronous fluctuation between branches is high, and decreases accordingly when branches are independent or the coupling relationship is broken. The coupling index and the allocation entropy index describe the flow distribution state on the secondary side of the CDU from different dimensions. The allocation entropy index focuses on the concentration or dispersion of flow values ​​in the numerical domain, while the coupling index focuses on the cooperative change relationship between branches. The two complement each other to provide a comprehensive characterization of the flow distribution system state.

[0064] In one embodiment of this application, the allocation entropy index and the coupling index are compared with their respective historical benchmarks, and a comprehensive early warning index is obtained by combining the relative relationship between the allocation entropy index and the coupling index, including: Obtain the first mean and first standard deviation of the allocation entropy index under historical benchmark conditions, and obtain the second mean and second standard deviation of the coupling index under historical benchmark conditions; Subtracting the first mean from the distribution entropy index and then dividing by the first standard deviation yields the standardized distribution entropy deviation; subtracting the second mean from the coupling index and then dividing by the second standard deviation yields the standardized coupling deviation. Calculate the ratio of the distribution entropy index to the coupling index, and use it as a dynamic ratio term; The standardized allocation entropy deviation, the standardized coupling deviation, and the dynamic ratio term are weighted and summed to obtain the comprehensive early warning index.

[0065] In this embodiment, the first mean and first standard deviation of the distribution entropy index under historical baseline conditions, and the second mean and second standard deviation of the coupling index under historical baseline conditions are read from the database. These historical baseline values ​​are established in the initial stage after the system is first put into operation by statistically analyzing the distribution entropy index sequence and the coupling index sequence calculated over multiple consecutive time windows. Specifically, the first mean is the arithmetic mean of the distribution entropy index sequence, and the first standard deviation is a measure of the dispersion of the sequence relative to the mean. Similarly, the second mean and second standard deviation correspond to the coupling index sequence. These baseline values ​​characterize the typical level and normal fluctuation range of traffic distribution uniformity and inter-branch coupling relationship under healthy operating conditions.

[0066] It should be noted that, in this embodiment, the historical baseline state is defined as the period during which the system operates continuously for no less than 24 hours under stable conditions, and the flow fluctuation of each branch does not exceed ±10% of its mean, and the outlet temperature fluctuation does not exceed ±2℃. Within this period, after removing windows where data is abnormal due to external disturbances (such as cabinet maintenance or valve operation), the index sequences of the remaining windows are statistically analyzed to obtain the mean and standard deviation of each index as the historical baseline.

[0067] Simultaneously, the first mean is subtracted from the allocation entropy index calculated in the current window, and the difference is divided by the first standard deviation to obtain the standardized allocation entropy deviation value. This standardization process eliminates the influence of the allocation entropy index's own dimensions and normal fluctuation range, allowing the deviation value to measure the degree to which the current allocation uniformity deviates from the historical typical level, using historical standard deviations as the unit. When this value is positive, it indicates that the current allocation entropy is higher than the historical average level, meaning that the distribution of traffic among branches is more dispersed than historically normal; when this value is negative, it indicates that the current allocation entropy is lower than the historical average level, meaning that traffic distribution tends to be concentrated. Similarly, the second mean is subtracted from the coupling index calculated in the current window, and the difference is divided by the second standard deviation to obtain the standardized coupling deviation value, which is used to measure the direction and magnitude of the deviation of the current coupling strength between branches relative to the historical typical level.

[0068] Optionally, in practical applications, a comprehensive judgment can be made based on the statistical distribution of distribution entropy under normal operating conditions, combined with coupling indicators.

[0069] Simultaneously, the ratio of the current window allocation entropy index to the coupling index is calculated as a dynamic ratio term. This dynamic ratio term captures the relative relationship between the two core indicators. When allocation entropy increases (allocation becomes more uneven) while the coupling index decreases (inter-branch coordination weakens), this ratio increases; conversely, when allocation entropy decreases while the coupling index increases, this ratio decreases. The dynamic ratio term has a different physical meaning from the aforementioned two standardized deviation values: the standardized deviation values ​​measure the absolute deviation of each indicator from its respective historical benchmark, while the dynamic ratio term reflects the inherent proportional relationship between the two indicators. Together, they constitute complementary information, supporting a comprehensive assessment of the system state.

[0070] Subsequently, the standardized allocation entropy deviation, standardized coupling deviation, and dynamic ratio term are weighted and summed to obtain the comprehensive early warning index. This comprehensive early warning index integrates information from three dimensions: the degree of degradation of allocation uniformity, the degree of abnormality in inter-branch coupling, and the dynamic ratio relationship between the two indicators. It provides a single numerical value to comprehensively and quantitatively characterize the overall health status of the CDU secondary-side flow allocation system.

[0071] The above process, through the dual calculation of allocation and coupling indices, captures system state information in two orthogonal dimensions: the uniformity of flow distribution and the cooperative relationship between branches. The allocation index, by superimposing a probability deviation factor positively correlated with the degree of probability deviation, ensures a significant response when flow concentrates in a few branches, addressing the insufficient sensitivity of conventional statistics in extremely uneven conditions. The coupling index, by introducing distance-weighted mutual information calculations, integrates the physical topology of the fluid network into the measurement of statistical dependencies, making the calculation results more closely reflect actual hydraulic coupling characteristics. Standardizing both indices relative to their respective historical benchmarks before fusing them with the dynamic ratio term not only eliminates the incomparability caused by differences in dimensions and scale between different systems but also amplifies the effect through the ratio term when both indices deteriorate in the same direction. This allows the comprehensive early warning index to keenly reflect early signs of the distribution state shifting from equilibrium to imbalance.

[0072] S140, determine the current warning level based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data.

[0073] In this embodiment, the mean value of the comprehensive warning index under historical baseline conditions is obtained. The deviation is obtained by subtracting the mean value from the current comprehensive warning index. At the same time, the comprehensive warning index records of multiple consecutive time windows are extracted, and the average value of the difference between adjacent windows is calculated as the average rate of change to characterize the dynamic change trend. Then, based on the deviation threshold interval into which the deviation falls and the rate of change threshold interval into which the average rate of change falls, the current warning level is determined according to the preset multi-level judgment rules. The deviation threshold interval and the rate of change threshold interval are both preset based on the statistical quantile of the distribution of the comprehensive warning index under historical baseline conditions. Temperature data is used as auxiliary verification information in the judgment process to cross-verify whether the abnormal flow distribution is accompanied by a significant change in thermal state, thereby enhancing the reliability of the judgment results.

[0074] like Figure 2 As shown, in one embodiment of this application, the current warning level is determined based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data, including: S210, obtain the third average value of the comprehensive early warning index under the historical benchmark state, and subtract the third average value from the comprehensive early warning index to obtain the deviation; S220, calculate the difference between the comprehensive early warning index between multiple adjacent time windows and take the average value to obtain the average rate of change of the comprehensive early warning index in the most recent multiple time windows; S230 sets the deviation threshold range and the rate of change threshold range based on the distribution statistical quantile of the comprehensive early warning index under historical baseline conditions. S240, determine the current warning level based on the deviation threshold range into which the deviation amount falls, the change rate threshold range into which the average change rate falls, and the temperature data.

[0075] In this embodiment, the third mean of the comprehensive early warning index under historical baseline conditions is read from the system's persistent storage. The third mean is obtained by taking the arithmetic mean of the comprehensive early warning index sequence over multiple consecutive time windows in the initial stage, representing the typical central position of the comprehensive early warning index of the system in a healthy state.

[0076] Then, the third mean is subtracted from the comprehensive early warning index calculated at the current moment to obtain the deviation. The deviation is a signed real number. A positive value indicates that the current comprehensive early warning index is higher than the historical average level, indicating that the system state is deteriorating. A negative value indicates that the current comprehensive early warning index is lower than the historical average level, indicating that the system state is better than the historical norm. The absolute value of the deviation reflects the severity of the current state deviating from the historical typical level.

[0077] Simultaneously, comprehensive early warning index records for multiple consecutive time windows are extracted from the comprehensive early warning index time-series buffer. These records are arranged in descending chronological order, with each record corresponding to a comprehensive early warning index value at a sliding window moment. Following the oldest to the newest chronological order, the differences between adjacent windows are calculated sequentially, yielding multiple first-order difference values. Each difference value reflects the amount and direction of change in the comprehensive early warning index within the adjacent window interval; positive values ​​indicate an increase in the comprehensive early warning index, while negative values ​​indicate a decrease. The sum of all difference values ​​is then divided by the total number of differences to obtain the average rate of change of the comprehensive early warning index over the most recent multiple time windows. This average rate of change, with each window interval as the time unit, characterizes the average speed and trend of change of the comprehensive early warning index within the recent time range.

[0078] Based on the statistical quantiles of the comprehensive early warning index distribution under historical baseline conditions, multi-level deviation threshold intervals and multi-level change rate threshold intervals are pre-set. By sorting the comprehensive early warning index samples in the initial operation phase from smallest to largest value, the values ​​corresponding to different percentiles are taken as the boundaries of each threshold level. Based on the deviation threshold interval into which the deviation amount falls, the change rate threshold interval into which the average change rate falls, and the synchronously collected temperature data, the current early warning level is comprehensively determined: the deviation amount is used to measure the degree of static deviation, the average change rate is used to capture the speed of dynamic deterioration, and the temperature data, as a physical quantity independent of flow rate, is used to confirm the existence of a real thermal anomaly by checking whether the outlet temperature corresponding to the abnormal branch rises synchronously when the deviation amount or change rate reaches the critical interval, thus avoiding false triggering caused by flow sensor drift. The three pieces of information are mapped to discrete levels of normal, level three early warning, level two early warning, or level one early warning through a preset judgment rule table.

[0079] Specifically, candidate warning levels are initially determined based on the deviation threshold range into which the deviation amount falls and the change rate threshold range into which the average change rate falls. Then, the outlet temperature data of the candidate abnormal branch is obtained. If the outlet temperature data does not rise synchronously, the candidate warning level is reduced by one level or kept normal. If the outlet temperature data rises synchronously, the candidate warning level is confirmed, and the final warning level is determined accordingly.

[0080] The above process, by simultaneously examining the static deviation of the comprehensive early warning index relative to historical benchmarks and the dynamic rate of change within multiple recent windows, achieves a dual judgment of the amplitude and trend of the system state. The static deviation identifies the absolute degree to which the current state deviates from a healthy level, while the dynamic rate of change captures the trend information of an accelerating deterioration. The combined judgment of these two factors allows the early warning system to issue forward-looking signals when the state is slowly drifting but has not yet reached a threshold, and to respond quickly when sudden deterioration occurs. The introduction of temperature data as a cross-validation of an independent physical quantity effectively eliminates false deviations caused by flow sensor drift or interference, improving the specificity and reliability of early warning judgments and avoiding the operational burden caused by misjudgments based on a single flow dimension.

[0081] S150: When the warning level reaches the preset level, it outputs a warning notification to the operation and maintenance platform and management terminal, and stores the current operation data and warning results in the historical database.

[0082] In this embodiment, when the obtained warning level reaches the preset level, the warning notification output process is triggered, and the warning level and comprehensive warning index are pushed to the operation and maintenance management platform for visualization. At the same time, a notification message containing a warning summary is sent to the management terminal through the communication channel, and the complete operation results such as traffic data, temperature data, comprehensive warning index and warning level of the current time window are written into the historical database for subsequent threshold adaptive calibration and model iterative optimization.

[0083] In one embodiment of this application, when the warning level reaches a preset level, a warning notification is output to the operation and maintenance platform and management terminal, and the current operating data and warning results are stored in the historical database, including: When the warning level reaches the preset level, the change in the coupling index of each branch between adjacent time windows is calculated. Based on the proportion of the change of each branch to the total change of all branches, the branch that contributes the most to the change of the coupling index is determined as the target location result. The current warning level, comprehensive warning index, and target location results are pushed to the operation and maintenance management platform and displayed visually on the monitoring interface using color-coded indicators and trend charts. A warning notification is generated and sent to the management terminal through a preset communication channel.

[0084] When the warning level reaches a preset level, the abnormal branch location subprocess is triggered. For each branch, the difference in its coupling index between the current time window and the previous adjacent time window is calculated to obtain the inter-window change in the spatial coupling degree of that branch. This difference reflects the drift of the coupling strength between that branch and all other branches in the network relative to the previous window. The absolute values ​​of the changes of all branches are summed to obtain the absolute total of the changes of all branches. Then, the absolute change of each branch is divided by this absolute total to obtain the contribution weight of each branch to the average spatial coupling degree change of the entire network. The sum of these weights is one, representing the share of each branch in the overall change of coupling relationship. The branches are sorted from largest to smallest according to their contribution weights. The top few branches with the cumulative contribution reaching a preset proportion are selected as candidate abnormal branches. These branches are then further confirmed by the corresponding exit temperature data. If the exit temperature of the candidate abnormal branch increases significantly, the branch is finally determined as the target location result, and the branch information most likely to have physical anomalies is output simultaneously when the warning is triggered.

[0085] The current warning level, comprehensive warning index value, and target location result are assembled into a warning notification and pushed to the designated data tag on the operation and maintenance management platform via industrial communication protocol. The operation and maintenance management platform dynamically renders the monitoring interface based on the received data tag values: the main monitoring panel displays the comprehensive warning index as a dashboard or numerical value, with the background color automatically switching to the corresponding warning color according to the warning level. Simultaneously, the trend chart sub-panel plots the historical change curve of the comprehensive warning index over a recent period, highlighting the current trigger point with a flashing marker to assist operation and maintenance personnel in intuitively assessing the severity and evolution direction of the anomaly. The warning report summary information is presented in a pop-up window on the monitoring interface, including the warning level, trigger time, comprehensive warning index value, anomaly branch number, and its exit temperature, providing operation and maintenance personnel with immediate visual decision-making support.

[0086] Subsequently, the warning notification is sent to the management terminal through communication channels, including calling the SMS gateway interface to send a simplified warning summary information to the management terminal mobile terminal, and calling the email service interface to send an email containing a complete warning report text and a structured data table to the email address.

[0087] Optionally, the complete operational results, including traffic data, temperature data, spatial coupling degree sequences of each branch, comprehensive warning index, deviation, average rate of change, and warning level for the current time window, are assembled with the warning results into a single data record. This record is then persistently stored in a time-series database via a database connection pool. The labeled data accumulated in the historical database is used for subsequent offline batch processing tasks. For example, the comprehensive warning index distribution for normal windows in historical data is periodically recalculated, and the deviation and rate of change thresholds are dynamically updated. This allows the warning judgment logic to adaptively follow the long-term, slow changes in the system, preventing fixed thresholds from drifting and failing in the later stages of the equipment's lifespan.

[0088] The above process, by automatically calculating the contribution of changes in coupling indicators of each branch when an early warning is triggered, enables rapid location of abnormal branches. This avoids the inefficient blind troubleshooting of maintenance personnel checking dozens of branches one by one, allowing them to directly check relevant branches based on the location results. Multi-channel output of early warning notifications, visual rendering of the monitoring interface, and real-time notifications from management terminals facilitate seamless information transfer from the system to personnel, transforming abstract analysis results into warning signals with clear operational guidance. Persistent storage of historical data establishes a self-learning processing mechanism for the system. Accumulated operational records provide a data foundation for the adaptive calibration of subsequent judgment thresholds, enabling the sensitivity of the early warning system to dynamically adjust according to long-term trends such as equipment aging and environmental changes, avoiding performance degradation that occurs at the end of the lifecycle of fixed thresholds.

[0089] This technical solution provides a synchronous and accurate raw data foundation for subsequent analysis by collecting and processing flow and temperature signals. By constructing a global probability distribution and marginal probability distributions for each branch, the raw flow data is unified into the same probability space for quantification, eliminating the incomparability caused by differences in numerical scales between branches, and ensuring that subsequent statistical characteristics accurately reflect the overall distribution status of the system. Based on this, the system state is characterized from two orthogonal dimensions—uniformity and synergy—using distribution and coupling indicators respectively. These indicators are then compared with historical benchmarks and fused into a comprehensive early warning index, enabling the system to detect early signs of distribution imbalance. By analyzing the deviation and trend of the comprehensive early warning index, combined with cross-validation of temperature data, a multi-level early warning system with both foresight and specificity is achieved. Finally, information is output and data is stored when an early warning is triggered, completing the entire processing from perception, analysis, judgment to feedback, improving the efficiency and accuracy of early warning for abnormal distribution in liquid-cooled CDUs.

[0090] The following describes embodiments of the intelligent monitoring-based liquid-cooled CDU allocation early warning system of this application, which can be used to execute the intelligent monitoring-based liquid-cooled CDU allocation early warning method in the above embodiments of this application. It is understood that the intelligent monitoring-based liquid-cooled CDU allocation early warning system can be a computer program (including program code) running on a computer device. For example, the intelligent monitoring-based liquid-cooled CDU allocation early warning system can be configured as industrial application software or industrial control management software to realize industrial cloud computing of industrial big data generated during the production process through an industrial cloud platform. The intelligent monitoring-based liquid-cooled CDU allocation early warning system can be used to execute the corresponding steps in the method provided in the embodiments of this application. For details not disclosed in the embodiments of the intelligent monitoring-based liquid-cooled CDU allocation early warning system of this application, please refer to the embodiments of the intelligent monitoring-based liquid-cooled CDU allocation early warning method of this application described above.

[0091] Figure 3 A block diagram of a liquid-cooled CDU distribution early warning system based on intelligent monitoring according to an embodiment of this application is shown.

[0092] Reference Figure 3 As shown, an embodiment of the liquid-cooled CDU distribution early warning system based on intelligent monitoring according to this application includes: The acquisition module 310 is used to acquire the flow and temperature signals of each branch in the cooling capacity distribution unit (CDU), and after signal processing, generate flow data and temperature data respectively and store them in the data buffer area. The extraction module 320 is used to extract traffic data of each branch from the data buffer based on a preset time window, and obtain a global probability distribution that characterizes the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics, and to obtain the marginal probability distribution of each branch by performing separate statistics on the traffic data of each branch. The index module 330 is used to calculate the allocation entropy index, which characterizes the uniformity of the overall traffic distribution of the CDU, based on the global probability distribution; calculate the coupling index, which characterizes the traffic coupling relationship between branches, based on the global probability distribution and the marginal probability distribution; compare the allocation entropy index and the coupling index with their respective historical benchmarks; and perform a fusion calculation based on the relative relationship between the allocation entropy index and the coupling index to obtain a comprehensive early warning index. The rating module 340 is used to determine the current warning level based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data. The early warning module 350 is used to output an early warning notification to the operation and maintenance platform and management terminal when the early warning level reaches the preset level, and to store the current operation data and early warning results in the historical database.

[0093] In this application, based on the aforementioned scheme, the process of acquiring flow and temperature signals from each branch of the cooling capacity distribution unit (CDU), generating flow and temperature data respectively after signal processing, and storing them in a data buffer includes: acquiring analog signals corresponding to flow velocity through flow sensors at the inlet of each branch of the cooling capacity distribution unit (CDU), acquiring analog signals corresponding to temperature through temperature sensors at the inlet of each branch or the inlet / outlet of the cold plate, and transmitting the analog signals to the analog input channel of the data acquisition terminal; synchronously sampling and performing analog-to-digital conversion on the analog signals in the analog input channel to generate an original digital sequence; filtering the original digital sequence to generate filtered digital quantities, converting the filtered digital quantities into engineering physical quantities according to the calibration coefficients of each analog input channel, as flow and temperature data; and storing the flow and temperature data in a circular queue in the data buffer.

[0094] In this application, based on the aforementioned scheme, the step of extracting traffic data of each branch from the data buffer based on a preset time window, obtaining a global probability distribution characterizing the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics, and obtaining the marginal probability distribution of each branch by separately statistically analyzing the traffic data of each branch, includes: extracting traffic values ​​of each branch from the data buffer based on a preset time window to construct a traffic matrix; dividing the traffic values ​​in the traffic matrix into several intervals according to their numerical values; counting the first frequency of the traffic values ​​falling into each interval, dividing the first frequency by the total number of traffic values ​​to obtain the interval probability of the interval, and combining the interval probabilities of each interval to obtain a global probability distribution characterizing the overall traffic distribution of the CDU; extracting the traffic value sequence of each branch from the traffic matrix, counting the second frequency of the traffic values ​​in the traffic value sequence falling into the interval, and dividing the second frequency by the number of sampling points of the branch to obtain the marginal probability distribution of the branch.

[0095] In this application, based on the aforementioned scheme, the step of calculating the allocation entropy index to characterize the uniformity of overall CDU flow distribution according to the global probability distribution, and calculating the coupling index to characterize the flow coupling relationship between branches according to the global probability distribution and the marginal probability distribution, includes: calculating information entropy based on the interval probabilities in the global probability distribution, determining a probability deviation factor based on the degree of deviation of the interval probabilities from the average probability, and superimposing the probability deviation factor on the information entropy to obtain the allocation entropy index; for any two branches, generating a joint probability distribution based on the joint frequency of the flow values ​​of the two branches falling into each interval combination at each sampling time within the same time window, and generating mutual information between the two branches by combining the marginal probability distributions of the two branches; calculating the mutual information between the branch and other branches, and weighting and summing the mutual information based on the physical pipeline distance between the branches to obtain the spatial coupling degree of the branch relative to other branches; and averaging the spatial coupling degrees of all branches to obtain the coupling index to characterize the flow coupling relationship between branches.

[0096] In this application, based on the aforementioned scheme, the step of comparing the allocation entropy index and the coupling index with their respective historical benchmarks, and then performing a fusion calculation based on the relative relationship between the allocation entropy index and the coupling index to obtain a comprehensive early warning index, includes: obtaining the first mean and first standard deviation of the allocation entropy index under the historical benchmark state; obtaining the second mean and second standard deviation of the coupling index under the historical benchmark state; subtracting the first mean from the allocation entropy index and dividing by the first standard deviation to obtain a standardized allocation entropy deviation value; subtracting the second mean from the coupling index and dividing by the second standard deviation to obtain a standardized coupling deviation value; calculating the ratio of the allocation entropy index to the coupling index as a dynamic ratio term; and weighted summing the standardized allocation entropy deviation value, the standardized coupling deviation value, and the dynamic ratio term to obtain a comprehensive early warning index.

[0097] In this application, based on the aforementioned scheme, determining the current warning level according to the deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data includes: obtaining the third mean of the comprehensive warning index under the historical baseline state; subtracting the third mean from the comprehensive warning index to obtain the deviation; calculating the difference between the comprehensive warning indexes between multiple adjacent time windows and taking the average value to obtain the average rate of change of the comprehensive warning index within the most recent multiple time windows; setting a deviation threshold range and a rate of change threshold range based on the statistical quantiles of the distribution of the comprehensive warning index under the historical baseline state; and determining the current warning level according to the deviation threshold range into which the deviation falls, the rate of change threshold range into which the average rate of change falls, and the temperature data.

[0098] In this application, based on the aforementioned scheme, the step of outputting an early warning notification to the operation and maintenance platform and management terminal when the early warning level reaches a preset level, and storing the current operating data and early warning results in the historical database, includes: when the early warning level reaches a preset level, calculating the change in coupling indicators of each branch between adjacent time windows, and determining the branch that contributes the most to the change in coupling indicators as the target location result based on the proportion of the change in each branch to the total change in all branches; pushing the current early warning level, comprehensive early warning index, and the target location result to the operation and maintenance management platform, and visually displaying them on the monitoring interface using color indicators and trend charts; generating an early warning notification, and sending the early warning notification to the management terminal through a preset communication channel.

[0099] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0100] In this embodiment, the computer system includes a central processing unit 401, which can perform various appropriate actions and processes based on programs stored in read-only memory 402 or programs loaded from storage section 408 into random access memory 403, such as executing the intelligent monitoring-based liquid-cooled CDU allocation early warning method described in the above embodiment. The random access memory 403 also stores various programs and data required for system operation, thereby realizing big data storage and big data management. The central processing unit 401, read-only memory 402, and random access memory 403 are interconnected via bus 404. Input / output interface 405 is also connected to bus 404.

[0101] The following components are connected to the input / output interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0102] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit 401, it performs various functions defined in the system of this application.

[0103] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable medium can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Computer programs contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0104] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0105] One aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0106] On the other hand, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the intelligent monitoring-based liquid-cooled CDU allocation early warning method described in the above embodiments.

[0107] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

Claims

1. A liquid-cooled CDU allocation early warning method based on intelligent monitoring, characterized in that, include: The flow and temperature signals of each branch in the cooling capacity distribution unit (CDU) are collected, and after signal processing, flow data and temperature data are generated and stored in the data buffer area. Based on a preset time window, traffic data of each branch is extracted from the data buffer. By dividing the traffic data into intervals and performing frequency statistics, a global probability distribution characterizing the overall traffic distribution of the CDU is obtained. The marginal probability distribution of each branch is obtained by separately statistically analyzing the traffic data of each branch. The allocation entropy index, used to characterize the uniformity of overall CDU traffic distribution, is calculated based on the global probability distribution. The coupling index, used to characterize the traffic coupling relationship between branches, is calculated based on the global probability distribution and the marginal probability distribution. The allocation entropy index and the coupling index are compared with their respective historical benchmarks, and the comprehensive early warning index is obtained by combining the relative relationship between the allocation entropy index and the coupling index. The current warning level is determined based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data. When the warning level reaches the preset level, a warning notification is sent to the operation and maintenance platform and management terminal, and the current operation data and warning results are stored in the historical database.

2. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 1, characterized in that, The flow and temperature signals of each branch in the cooling capacity distribution unit (CDU) are collected, processed, and then used to generate flow and temperature data, which are stored in the data buffer. The flow rate is collected by the flow sensor at the inlet of each branch in the cold distribution unit (CDU), and the temperature is collected by the temperature sensor at the inlet or outlet of the cold plate. The analog signals are then transmitted to the analog input channel of the data acquisition terminal. The analog signals in the analog input channel are synchronously sampled and converted from analog to digital to generate the original digital sequence; The original digital sequence is filtered to generate filtered digital quantities. Based on the calibration coefficients of each analog input channel, the filtered digital quantities are converted into engineering physical quantities, which are used as flow data and temperature data. The flow rate data and temperature data are stored in a circular queue in the data cache area.

3. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 1, characterized in that, Based on a preset time window, traffic data for each branch is extracted from the data buffer. By dividing the traffic data into intervals and performing frequency statistics, a global probability distribution characterizing the overall traffic distribution of the CDU is obtained. The marginal probability distribution for each branch is obtained by separately statistically analyzing its traffic data, including: Based on a preset time window, the flow values ​​of each branch are extracted from the data cache to construct a flow matrix; The flow values ​​in the flow matrix are divided into several intervals according to their numerical values; The first frequency of the flow value falling into each interval is counted, and the first frequency is divided by the total number of flow values ​​to obtain the interval probability. The interval probabilities of each interval are combined to obtain the global probability distribution characterizing the overall flow distribution of the CDU. Extract the flow value sequence of each branch from the flow matrix, count the second frequency of the flow value in the flow value sequence falling into the interval, divide the second frequency by the number of sampling points of the branch, and obtain the marginal probability distribution of the branch.

4. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 3, characterized in that, The allocation entropy index, used to characterize the uniformity of overall CDU traffic distribution, is calculated based on the global probability distribution. The coupling index, used to characterize the traffic coupling relationship between branches, is calculated based on the global probability distribution and the marginal probability distribution, including: Information entropy is calculated based on the interval probabilities in the global probability distribution. A probability deviation factor is determined based on the degree to which the interval probabilities deviate from the average probability. The probability deviation factor is then superimposed on the information entropy to obtain the allocation entropy index. The average probability represents the average value of all interval probabilities. For any two branches, a joint probability distribution is generated based on the joint frequency of the flow values ​​of the two branches falling into each interval combination at each sampling time within the same time window. The mutual information between the two branches is then generated by combining the marginal probability distributions of the two branches. Calculate the mutual information between the branch and other branches, and perform a weighted summation of the mutual information based on the physical pipeline distance between the branches to obtain the spatial coupling degree of the branch relative to other branches; The average spatial coupling degree of all branches is taken to obtain a coupling index used to characterize the flow coupling relationship between branches.

5. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 1, characterized in that, The allocation entropy index and the coupling index are compared with their respective historical benchmarks, and the relative relationship between the allocation entropy index and the coupling index is combined for fusion calculation to obtain a comprehensive early warning index, including: Obtain the first mean and first standard deviation of the allocation entropy index under historical benchmark conditions, and obtain the second mean and second standard deviation of the coupling index under historical benchmark conditions; Subtracting the first mean from the distribution entropy index and then dividing by the first standard deviation yields the standardized distribution entropy deviation; subtracting the second mean from the coupling index and then dividing by the second standard deviation yields the standardized coupling deviation. Calculate the ratio of the distribution entropy index to the coupling index, and use it as a dynamic ratio term; The standardized allocation entropy deviation, the standardized coupling deviation, and the dynamic ratio term are weighted and summed to obtain the comprehensive early warning index.

6. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 1, characterized in that, Based on the deviation of the current comprehensive warning index from the historical normal baseline, the dynamic trend of the comprehensive warning index within the time window, and the temperature data, the current warning level is determined, including: Obtain the third mean of the comprehensive early warning index under historical baseline conditions, and subtract the third mean from the comprehensive early warning index to obtain the deviation. Calculate the difference between the comprehensive early warning indexes between multiple adjacent time windows and take the average value to obtain the average rate of change of the comprehensive early warning indexes in the most recent multiple time windows; Based on the distribution statistical quantiles of the comprehensive early warning index under historical baseline conditions, deviation threshold intervals and change rate threshold intervals are set. The current warning level is determined based on the deviation threshold range into which the deviation amount falls, the change rate threshold range into which the average change rate falls, and the temperature data.

7. The liquid-cooled CDU allocation early warning method based on intelligent monitoring according to claim 1, characterized in that, When the warning level reaches the preset level, a warning notification is output to the operation and maintenance platform and management terminal, and the current operating data and warning results are stored in the historical database, including: When the warning level reaches the preset level, the change in the coupling index of each branch between adjacent time windows is calculated. Based on the proportion of the change of each branch to the total change of all branches, the branch that contributes the most to the change of the coupling index is determined as the target location result. The current warning level, comprehensive warning index, and target location results are pushed to the operation and maintenance management platform and displayed visually on the monitoring interface using color-coded indicators and trend charts. A warning notification is generated and sent to the management terminal through a preset communication channel.

8. A liquid-cooled CDU distribution early warning system based on intelligent monitoring, characterized in that, include: The acquisition module is used to acquire the flow and temperature signals of each branch in the cooling capacity distribution unit (CDU), and after signal processing, generate flow data and temperature data respectively and store them in the data buffer area. The extraction module is used to extract traffic data of each branch from the data buffer based on a preset time window, and obtain a global probability distribution that characterizes the overall traffic distribution of the CDU by dividing the traffic data into intervals and performing frequency statistics. The marginal probability distribution of each branch is obtained by separately statistically analyzing the traffic data of each branch. The index module is used to calculate the allocation entropy index, which characterizes the uniformity of the overall traffic distribution of the CDU, based on the global probability distribution; and to calculate the coupling index, which characterizes the traffic coupling relationship between branches, based on the global probability distribution and the marginal probability distribution. The allocation entropy index and the coupling index are compared with their respective historical benchmarks, and the relative relationship between the allocation entropy index and the coupling index is combined for fusion calculation to obtain a comprehensive early warning index. The level module is used to determine the current warning level based on the degree of deviation of the current comprehensive warning index from the historical normal baseline, the dynamic change trend of the comprehensive warning index within the time window, and the temperature data. The early warning module is used to output early warning notifications to the operation and maintenance platform and management terminal when the early warning level reaches the preset level, and to store the current operation data and early warning results in the historical database.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the liquid-cooled CDU allocation early warning method based on intelligent monitoring as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the liquid-cooled CDU allocation early warning method based on intelligent monitoring as described in any one of claims 1 to 7.

Images