A prefabricated data center fault alarm method and system based on an algorithm model

Abstract

The present application relates to the technical field of data processing, and more particularly to a prefabricated data center fault alarm method and system based on an algorithm model. The method comprises: obtaining multi-dimensional real-time running data of a target device and neighborhood associated devices in a prefabricated data center, and constructing a device space topology; calculating a normalized space potential energy weight based on rated power, physical distance and air flow resistance coefficient, and determining a dynamic coupling conduction coefficient in combination with a Pearson correlation coefficient and a waveform cross-correlation lag time; calculating an external interference combined field strength based on an abnormal deviation amount and the conduction coefficient; using an LSTM prediction value, introducing an adaptive interference correction gain, subtracting the field strength influence, obtaining an essential residual error, and realizing fault alarm of the prefabricated data center. By constructing the device space topology and the potential energy weight, calculating the dynamic coupling conduction coefficient, evaluating the external interference combined field strength, introducing the adaptive correction gain, and eliminating the interference, the accuracy of the fault alarm is improved.

Description

A method and system for fault alarm in prefabricated data centers based on algorithm models. Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a prefabricated data center fault alarm method and system based on an algorithm model. Background Technology

[0002] With the development of cloud computing and edge computing, prefabricated data centers have become an important form of next-generation computing infrastructure due to their modular design, convenient deployment, short construction cycle, and high energy efficiency. However, this highly integrated design concept also brings unique operation and maintenance challenges. Inside the containers or modules of prefabricated data centers, the physical space is extremely compact, and the physical density of equipment such as servers, in-row air conditioners, and power distribution units is much higher than that of traditional computer rooms. The high-density layout leads to a strong physical coupling effect between devices. For example, local hot spots of high-density computing nodes can easily spread through narrow airflow channels to form a heat island effect, affecting downstream equipment; the mechanical vibration of cooling fans can also be easily transmitted to sensitive storage devices through the rigidly connected rack frame.

[0003] Currently, in equipment health management, Long Short-Term Memory (LSTM) networks are typically used to perform time-series analysis on equipment operating data. The residual between predicted and actual values ​​is calculated to determine whether the equipment is in a potential failure state, i.e., a progressive failure. However, existing LSTM monitoring methods are usually based on the assumption of independent and identically distributed systems, treating each device as an isolated individual for univariate or independent multivariate analysis, ignoring the complex spatial coupling relationships between devices.

[0004] In the tightly coupled environment of prefabricated data centers, when neighboring equipment malfunctions, such as a sudden temperature rise due to a load surge or increased vibration caused by a fan failure, interference signals are transmitted through physical media, such as air and rack frames, and superimposed on the target equipment. This causes the sensor readings of the target equipment to deviate from the normal range for a short period of time. At this time, traditional LSTM models cannot distinguish whether this deviation is due to internal degradation of the equipment itself or passive interference from the external environment. As a result, such passive deviations caused by external coupling are misjudged as potential fault states of the target equipment itself. This not only leads to a large number of false alarms, causing maintenance personnel to desensitize alarms, but also masks the real source of the fault, making it impossible for the maintenance system to accurately locate risks, and thus affecting the accuracy of fault alarms in prefabricated data centers. Summary of the Invention

[0005] To address the problem that traditional LSTM cannot distinguish between device degradation and external coupling interference, resulting in a high false alarm rate and affecting the accuracy of fault alarms in prefabricated data centers, this invention provides a fault alarm method and system for prefabricated data centers based on an algorithm model.

[0006] In a first aspect, the present invention provides a prefabricated data center fault alarm method based on an algorithm model, employing the following technical solution:

[0007] A fault alarm method for prefabricated data centers based on an algorithm model includes: acquiring real-time operational data containing multiple dimensions of a target device and its neighboring associated devices within the prefabricated data center, and constructing a device spatial topology based on the physical layout information of the prefabricated data center; calculating the normalized spatial potential energy weight of each associated device to the target device based on the rated power of each associated device, the physical distance between each associated device and the target device, and the airflow impedance coefficient; calculating the dynamic coupling transmission coefficient of each associated device to the target device in each dimension based on the normalized spatial potential energy weight, the Pearson correlation coefficient of each associated device to the target device in each dimension, and the waveform cross-correlation lag time; calculating the external interference synthetic field strength experienced by the target device in each dimension at the current moment based on the abnormal deviation of each associated device in each dimension and the dynamic coupling transmission coefficient; using a long short-term memory network for modeling to obtain the predicted value of the target device in each dimension, and introducing an adaptive interference correction gain to subtract the influence of the external interference synthetic field strength from the original prediction residual to calculate the essential residual of the target device in each dimension at the current moment; and realizing fault alarm for the prefabricated data center based on the essential residual.

[0008] This invention achieves the evaluation and modeling of physical coupling relationships between devices by constructing a device spatial topology and normalized spatial potential energy weights, more accurately reflecting the intensity of spatial influence between devices and providing a reliable topological foundation for subsequent coupling transmission analysis. By calculating dynamic coupling transmission coefficients, and comprehensively considering spatial potential energy weights, correlation coefficients, and time lag effects, it achieves accurate evaluation of multi-dimensional coupling transmission, more accurately reflecting the real-time influence of neighboring devices on the target device. By calculating the synthetic field strength of external interference, it achieves a comprehensive evaluation of the external coupling interference experienced by the target device, thereby more accurately separating the effects of external interference and internal degradation. By introducing adaptive interference correction gain, the influence of external interference is subtracted from the original prediction residual to calculate the essential residual, effectively eliminating false alarms caused by external coupling interference and improving the accuracy and reliability of fault alarms.

[0009] Furthermore, the multiple dimensions include temperature, vibration, voltage, and current dimensions.

[0010] Furthermore, the real-time running data is data that has been processed by Z-score standardization.

[0011] Further, the calculation of the normalized spatial potential energy weights of each associated device to the target device includes: calculating the original physical potential energy of each associated device to the target device, and then performing a process on the set of the original physical potential energy. After normalization, the normalized spatial potential energy weights of each associated device relative to the target device are obtained. The original physical potential energy satisfies:

[0012] In the formula, For associated devices For target equipment The original physical potential energy, For associated devices Rated power, For associated devices With target equipment The Euclidean distance between them For associated devices With target equipment The airflow resistance coefficient between them.

[0013] This invention enables a scientific assessment of the intensity of influence between devices by constructing a physical potential energy model that includes power and distance square terms. The introduction of the airflow impedance coefficient more accurately reflects the obstruction effect of the airflow channel on heat conduction. Normalization ensures the rationality and interpretability of the weight distribution, providing a reliable potential weight basis for the calculation of dynamic coupling transmission coefficients.

[0014] Furthermore, the range of the airflow impedance coefficient is: in response to the associated device Located in the target device When the upstream cold channel is connected, the range of the airflow resistance coefficient is: Otherwise, the airflow resistance coefficient is 0.

[0015] Furthermore, the dynamic coupling transmission coefficient satisfies:

[0016] In the formula, In the first Related devices in each dimension For target equipment The dynamic coupling transmission coefficient, For associated devices For target equipment Normalized spatial potential weights, For associated devices With target equipment Between the first Pearson correlation coefficients in each dimension It is a natural constant. The time decay constant, For associated devices With target equipment Between the first Waveform cross-correlation lag time in each dimension It is the absolute value symbol.

[0017] This invention achieves a comprehensive evaluation of dynamic coupling transmission coefficients by constructing a product model that includes normalized spatial potential energy weights, absolute values ​​of correlation coefficients, and time lag exponential terms. It comprehensively considers spatial influence intensity, linear correlation, and time lag effects. The exponential decay term accurately reflects the impact of time lag on coupling transmission, thereby enabling accurate evaluation of the real-time coupling transmission strength of neighboring devices to the target device.

[0018] Furthermore, the combined field strength of the external interference satisfies:

[0019] In the formula, For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension For target equipment The set consisting of all associated devices within the neighborhood of . In the first Related devices in each dimension For target equipment The dynamic coupling transmission coefficient, For the device associated at the current moment In the Values ​​of each dimension For all time-related devices within the current time period. In the The mean of each dimension, It is the hyperbolic tangent function. It is a minimum value function. For symbolic functions, It is the absolute value symbol.

[0020] This invention achieves a comprehensive assessment of the combined field strength of external interference by constructing a weighted sum including dynamic coupling conduction coefficients and hyperbolic tangent functions. The minimum function limits the influence amplitude of abnormal deviations, and the hyperbolic tangent function achieves nonlinear compression of the deviation amount, thereby more accurately reflecting the comprehensive interference influence of neighboring devices on the target device and providing a reliable basis for interference compensation for the calculation of essential residuals.

[0021] Furthermore, the essential residual satisfies:

[0022] In the formula, For the target device at the current moment In the The essential residuals in each dimension For the target device at the current moment In the Values ​​of each dimension For the target device at the current moment No. Predicted values ​​for each dimension The adaptive interference correction gain is obtained by minimizing the essential residual through the historical validation set. For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension It is the absolute value symbol.

[0023] This invention achieves a scientific assessment of the intrinsic residual by constructing a difference model that includes a prediction residual and an external interference compensation term. The adaptive interference correction gain ensures the accuracy of interference compensation, and the absolute value function guarantees the non-negativity of the residual. This effectively separates the effects of internal degradation and external interference, thereby improving the accuracy of fault identification.

[0024] Furthermore, the method for implementing fault alarms in prefabricated data centers includes: responding to a situation where the essential residual of a target device in any dimension exceeds a preset threshold across multiple consecutive time points including the current time, determining that the target device has a fault risk, and issuing an early warning, thereby completing the fault alarm for prefabricated data centers based on an algorithm model.

[0025] Secondly, this invention provides a prefabricated data center fault alarm system based on an algorithm model, employing the following technical solution:

[0026] A prefabricated data center fault alarm system based on an algorithm model includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned prefabricated data center fault alarm method based on an algorithm model.

[0027] By adopting the above technical solution, a computer program is generated from the above-mentioned prefabricated data center fault alarm method based on an algorithm model, and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, making them convenient to use.

[0028] The present invention has the following technical effects:

[0029] (1) In view of the problem that traditional LSTM models analyze devices in isolation and cannot distinguish between external coupling interference and device degradation, this invention constructs a device spatial topology, calculates normalized spatial potential energy weights by combining the physical characteristics of associated devices such as rated power, physical distance, and airflow impedance coefficient, and then integrates Pearson correlation coefficient and waveform cross-correlation lag time to quantify dynamic coupling transmission coefficient, so as to more accurately characterize the interference transmission intensity and lag characteristics of associated devices to target devices. Finally, the comprehensive external interference borne by the target device is evaluated by the external interference synthetic field strength. Adaptive interference correction gain is introduced to remove the interference effect from the original prediction residual. The resulting essential residual only reflects the internal degradation state of the device itself, avoiding misjudging the abnormality of neighboring devices, such as temperature interference caused by sudden load increase or vibration transmission caused by fan failure, as the target device's own fault, reducing the false alarm rate, and effectively solving the problem of alarm desensitization by maintenance personnel and the masking of fault sources.

[0030] (2) Breaking through the limitations of the traditional model's independent and identically distributed assumption, it fully adapts to the high integration and strong physical coupling of prefabricated data centers. The calculation of the dynamic coupling transmission coefficient not only considers the spatial relationship between devices, but also integrates multi-dimensional physical and data characteristics such as power, airflow impedance, correlation coefficient, and lag time, so as to achieve accurate evaluation of complex coupling effects such as heat conduction and vibration conduction. Compared with traditional interference modeling, the coupling relationship calculation of this invention is more in line with the actual physical environment of prefabricated data centers, so as to accurately capture the transmission law of external interference, make the calculation of the essential residual have scene adaptability, and ensure the accuracy of fault identification.

[0031] (3) Traditional methods often misjudge external interference as a fault in the target device, which can lead maintenance personnel to focus on the interfered device rather than the real source of the fault, increasing the cost of troubleshooting. In the process of calculating the composite field strength of external interference, this invention clarifies the interference contribution of each associated device to the target device through the dynamic coupling transmission coefficient. When the target device shows an abnormal prompt, the abnormal deviation and coupling transmission coefficient of each associated device can be traced back to quickly locate the source device that caused the interference. For example, if a server load suddenly increases and the temperature of the surrounding devices becomes abnormal, it can help maintenance personnel directly lock the core of the fault, avoid blind troubleshooting, shorten the fault location time, and improve the maintenance response efficiency and fault handling accuracy of the prefabricated data center.

[0032] (4) By reducing false alarms and accurately locating fault sources, the ineffective workload and troubleshooting costs of operation and maintenance personnel are reduced, and unnecessary downtime maintenance caused by false alarms is avoided. At the same time, timely identification of real equipment faults and fault sources can prevent the spread of faults. For example, if the vibration caused by the fan failure of a certain equipment is transmitted to the storage device, it may cause damage to the storage device if it is not dealt with in time. This reduces the impact of faults on the computing power service of the data center and provides key technical support for the efficient operation and maintenance of prefabricated data centers. It not only ensures the operational reliability of computing power infrastructure, but also reduces the overall operation and maintenance costs. Attached Figure Description

[0033] Figure 1 is a flowchart of a prefabricated data center fault alarm method based on an algorithm model according to an embodiment of the present invention.

[0034] Figure 2 is a schematic diagram of the external interference synthetic field strength distribution in a prefabricated data center fault alarm method based on an algorithm model according to an embodiment of the present invention.

[0035] Figure 3 is a schematic diagram comparing the original residual and the essential residual of traditional LSTM in a prefabricated data center fault alarm method based on an algorithm model according to an embodiment of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] This invention discloses a prefabricated data center fault alarm method based on an algorithm model, referring to Figure 1, including steps S1-S6:

[0038] S1: Obtain real-time operational data containing multiple dimensions of the target device and its neighboring devices within the prefabricated data center, and construct the device spatial topology based on the physical layout information of the prefabricated data center.

[0039] Specifically, a sensor array deployed within a prefabricated data center acquires real-time operational data from the target device and its neighboring devices, encompassing multiple dimensions including temperature, vibration, voltage, and current. Simultaneously, based on the physical layout information of the prefabricated data center, such as rack coordinates, device spacing, and airflow direction, a device spatial topology is constructed to determine the target device's neighborhood, including physical distance. The set of all related devices within the meter. In order to eliminate the differences in different physical dimensions, the operating data obtained in this step is data that has been processed by Z-score standardization. Specifically, it is based on the statistical characteristics of the device health benchmark period, including the mean and standard deviation, and converts the original collected data into dimensionless standardized data. All subsequent calculations are based on this standardized data. The health benchmark period is a period of multiple consecutive days, such as 15 days, during which the device has operated without generating any fault alarms.

[0040] S2: Based on the rated power of each associated device, the physical distance between each associated device and the target device, and the airflow impedance coefficient, calculate the normalized spatial potential energy weight of each associated device to the target device.

[0041] It should be noted that physical distance alone cannot fully characterize the interference potential between devices. The rated power of different devices determines the upper limit of the energy they can generate in terms of heat or vibration, while the airflow direction determines the medium path for heat transfer. Therefore, this step aims to combine the energy source strength, i.e., rated power, and the transmission path attenuation, i.e., distance and airflow factor, to construct a static potential energy model that reflects the inherent ability of physical interference. This model is used to measure the theoretically likely magnitude of the impact that related devices in the neighborhood can have on the target device.

[0042] Specifically, the calculation of the normalized spatial potential energy weights of each associated device with respect to the target device includes:

[0043] Calculate the original physical potential energy of each associated device relative to the target device, wherein the original physical potential energy satisfies:

[0044] ;

[0045] In the formula, For associated devices For target equipment The original physical potential energy, measured in watts per square meter. For associated devices Rated power, in watts. For associated devices With target equipment The Euclidean distance between them, in meters. For associated devices With target equipment The airflow resistance coefficient between them is dimensionless and obtained through a preset table in CFD simulation, responding to the associated equipment. Located in the target device When the upstream cold channel is connected, the range of the airflow resistance coefficient is: Otherwise, the airflow resistance coefficient is 0.

[0046] in, This indicates the energy level of the interference source device; the higher the value, the stronger the interference. The stronger the potential energy that generates heat accumulation or mechanical vibration during operation, the better. The larger the value, the more it indicates its affinity for the target device. The higher the potential interference energy, the better. The impedance factor represents the physical conduction path. A larger value indicates a greater distance between devices or difficulty in conduction due to obstruction caused by airflow headwinds. The smaller the value, the more severe the attenuation of interference during transmission.

[0047] Then, the set of original physical potential energy is subjected to... Normalization is performed to obtain the normalized spatial potential energy weights of each associated device to the target device, ensuring that the sum of the weights is 1, so as to accurately reflect the relative contribution of each associated device to the physical interference of the target device in the neighborhood.

[0048] S3: Based on the normalized spatial potential weights, the Pearson correlation coefficients of each associated device and the target device in each dimension, and the waveform cross-correlation lag time, calculate the dynamic coupling transmission coefficients of each associated device to the target device in each dimension.

[0049] It should be noted that spatial potential energy only represents a physical possibility, not a factual reality at the current moment. Even if two devices are physically close, there is no substantial interference if their operating states are completely unrelated. Therefore, this step introduces Pearson correlation coefficient and time lag analysis to combine static physical connections with dynamic time-series waveforms.

[0050] Specifically, the dynamic coupling conduction coefficient satisfies:

[0051] ;

[0052] In the formula, In the first Related devices in each dimension For target equipment The dynamic coupling conduction coefficient is dimensionless. For associated devices For target equipment The normalized spatial potential weights are dimensionless. For associated devices With target equipment Between the first The Pearson correlation coefficient in each dimension is dimensionless. It is a natural constant. The time decay constant is expressed in units of 1. For example, the range of values ​​is Used for punishment Excessively long spurious correlations For associated devices With target equipment Between the first The waveform cross-correlation lag time in each dimension, in units of It is obtained through cross-correlation function. It is the absolute value symbol.

[0053] in, The effectiveness of handling causal relationships indicates that interference transmission should be instantaneous or have a short time lag. If the correlation peak between two devices occurs after an extremely long lag time, this correlation is more likely to be some long-term environmental commonality, such as diurnal temperature variation, rather than direct physical coupling between the devices. As the coefficient increases, the value of the coefficient decreases rapidly, thereby suppressing spurious correlation interference with long time lag.

[0054] S4: Based on the abnormal deviation of each associated device in each dimension and the dynamic coupling transmission coefficient, calculate the combined external interference field strength that the target device experiences in each dimension at the current moment.

[0055] It should be noted that the target device is often surrounded by multiple related devices simultaneously, and the external interference it experiences is the result of multiple sources superimposed. Furthermore, the minor fluctuations of the related devices are considered normal noise, and only when they deviate significantly do they constitute interference. Therefore, this step calculates a comprehensive composite field strength of external interference by superimposing the weighted abnormal deviations of all effective neighboring devices and introducing truncation protection.

[0056] Specifically, the combined field strength of the external interference satisfies:

[0057] ;

[0058] In the formula, For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension, dimensionless. For target equipment The set consisting of all associated devices within the neighborhood of . In the first Related devices in each dimension For target equipment The dynamic coupling conduction coefficient is dimensionless. For the device associated at the current moment In the Values ​​in each dimension, dimensionless. For all time-related devices within the current time period. In the The mean of each dimension, dimensionless. It is the hyperbolic tangent function. It is a minimum value function. For symbolic functions, It is the absolute value symbol.

[0059] in, Indicates the target device The summation of all associated devices demonstrates the superposition of interference. Indicates a single associated device The contributing interference component, i.e. the current anomaly level of the associated device, after being nonlinearly limited, is multiplied by its dynamic coupling transmission coefficient to the target device, thereby quantifying the specific interference intensity transmitted from the associated device to the target device at the current moment.

[0060] As shown in Figure 2, Figure 2 is a time-series curve of the combined field strength of external interference experienced by the target device at the current moment and at continuous historical moments under a certain dimension, such as temperature. The horizontal axis represents continuous time sampling points, reflecting the time process, such as each sampling point corresponding to 1 second; the vertical axis represents the quantified value of the combined field strength of external interference. Positive values ​​indicate that abnormal deviations of associated devices cause positive interference to be experienced by the target device, such as increased temperature and enhanced vibration, while negative values ​​indicate reverse interference, such as decreased temperature and weakened vibration. The larger the absolute value of the amplitude, the stronger the external coupling interference experienced by the target device. Figure 2 intuitively shows the superposition effect of abnormal interference from multiple associated devices in the time dimension. For example, the peak of the curve corresponds to the simultaneous obvious abnormality of multiple associated devices at a certain moment, or the dominant interference generated by a single strong interference source device, such as a sudden increase in the load of a high-power server, which provides a basis for subsequent calculation of the essential residual.

[0061] S5: Model using a long short-term memory network to obtain the predicted value of the target device in each dimension, and introduce an adaptive interference correction gain to subtract the influence of the external interference synthetic field strength from the original prediction residual, and calculate the essential residual of the target device in each dimension at the current time.

[0062] It should be noted that the original prediction residuals of the LSTM model contain information from both the device's own faults and external environmental interference. Directly using the original residuals for threshold determination will lead to a high false alarm rate. Therefore, the core idea of ​​this step is differential purification, which uses external interference to synthesize the field strength and subtracts the deviation caused by external factors from the original residuals. The corrected residuals can more purely reflect the device's own health status.

[0063] Specifically, the essential residual satisfies:

[0064] ;

[0065] In the formula, For the target device at the current moment In the The essential residuals in each dimension are dimensionless. For the target device at the current moment In the Values ​​in each dimension, dimensionless. For the target device at the current moment No. Predicted values ​​in each dimension, dimensionless. The adaptive disturbance correction gain, obtained by minimizing the essential residual through a historical validation set, is dimensionless and aims to address the problem that traditional fixed-parameter methods cannot adapt to environmental changes. The variance of the essential residual is minimized on the historical validation set, which must satisfy the following conditions: sampled from the device health baseline period, containing at least 5 active failure events of neighboring devices, and a 30-minute stabilization period after each failure event; For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension, dimensionless. It is the absolute value symbol.

[0066] As shown in Figure 3, Figure 3 is a time-series comparison curve of the original residual of the traditional LSTM model, the essential residual calculated by this invention, and the preset alarm threshold for the same target device in a certain dimension. The horizontal axis represents continuous time sampling points, consistent with the time dimension of Figure 2, ensuring the time-series correspondence between interference and residuals; the vertical axis represents the residual value, reflecting the degree of deviation between the actual value and the predicted value of the device's operating data. As can be seen from the figure, the original residual curve of the traditional LSTM is significantly affected by external interference. In Figure 2, there are obvious fluctuations in the time period corresponding to the peak of the interference field strength, which repeatedly approaches or exceeds the alarm threshold, indicating that the traditional method has the risk of false alarms. The essential residual curve of this solution removes the influence of external interference through adaptive interference correction gain, and the curve is more stable overall. It only shows a continuous rise and exceeds the alarm threshold when the device itself undergoes real degradation. Finally, the alarm trigger flag is only triggered when the essential residual of this solution exceeds the threshold for multiple consecutive moments, intuitively verifying that this invention can effectively filter false deviations caused by external interference, respond only to the device's own faults, and reduce the false alarm rate.

[0067] S6: Based on the aforementioned essential residual, fault alarms for prefabricated data centers are implemented.

[0068] Specifically, the method for implementing fault alarms in prefabricated data centers includes:

[0069] In response to multiple consecutive time points including the current time, such as when the essential residual of the target device in any dimension is greater than a preset threshold for five consecutive time points, such as... Threshold, The standard deviation of the health baseline period in this dimension is used to determine the risk of failure in the target equipment and issue an early warning, thus completing the fault alarm of the prefabricated data center based on the algorithm model.

[0070] This invention also discloses a prefabricated data center fault alarm system based on an algorithm model, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a prefabricated data center fault alarm method based on an algorithm model according to the present invention.

[0071] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0072] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A prefabricated data center fault alarm method based on an algorithm model, characterized in that, include: This process involves acquiring real-time operational data across multiple dimensions of the target device and its neighboring devices within a prefabricated data center, and constructing a spatial topology for the devices based on the physical layout information of the prefabricated data center. Based on the rated power of each associated device, the physical distance between each associated device and the target device, and the airflow impedance coefficient, the normalized spatial potential energy weights of each associated device relative to the target device are calculated. This includes calculating the original physical potential energy of each associated device relative to the target device, and then performing a process to refactor the set of original physical potential energy. After normalization, the normalized spatial potential energy weights of each associated device relative to the target device are obtained. The original physical potential energy satisfies: ； For associated devices For target equipment The original physical potential energy, For associated devices Rated power, For associated devices With target equipment The Euclidean distance between them For associated devices With target equipment The airflow impedance coefficient between them; based on the normalized spatial potential energy weight, the Pearson correlation coefficient of each associated device and the target device in each dimension, and the waveform cross-correlation lag time, the dynamic coupling transmission coefficient of each associated device to the target device in each dimension is calculated; based on the abnormal deviation of each associated device in each dimension and the dynamic coupling transmission coefficient, the external interference synthetic field strength experienced by the target device in each dimension at the current moment is calculated; a long short-term memory network is used for modeling to obtain the predicted value of the target device in each dimension, and an adaptive interference correction gain is introduced to subtract the influence of the external interference synthetic field strength from the original prediction residual to calculate the essential residual of the target device in each dimension at the current moment; based on the essential residual, fault alarm of the prefabricated data center is realized; the dynamic coupling transmission coefficient satisfies: ； In the first Related devices in each dimension For target equipment The dynamic coupling transmission coefficient, For associated devices For target equipment Normalized spatial potential weights, For associated devices With target equipment Between the first Pearson correlation coefficients in each dimension It is a natural constant. The time decay constant, For associated devices With target equipment Between the first Waveform cross-correlation lag time in each dimension The sign is absolute; the combined field strength of external disturbances satisfies: ； For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension For target equipment The set consisting of all associated devices within the neighborhood of . In the first Related devices in each dimension For target equipment The dynamic coupling transmission coefficient, For the device associated at the current moment In the Values ​​of each dimension For all time-related devices within the current time period. In the The mean of each dimension, It is the hyperbolic tangent function. It is a minimum value function. It is a symbolic function.

2. The prefabricated data center fault alarm method based on an algorithm model according to claim 1, characterized in that, The multiple dimensions include temperature, vibration, voltage, and current.

3. The prefabricated data center fault alarm method based on an algorithm model according to claim 1, characterized in that, The real-time running data is data that has been processed by Z-score standardization.

4. The prefabricated data center fault alarm method based on an algorithm model according to claim 1, characterized in that, The range of values ​​for the airflow impedance coefficient is: in response to the associated equipment Located in the target device When the upstream cold aisle is in use, the range of the airflow resistance coefficient is: Otherwise, the airflow resistance coefficient is 0.

5. The prefabricated data center fault alarm method based on an algorithm model according to claim 1, characterized in that, The essential residual satisfies: In the formula, For the target device at the current moment In the The essential residuals in each dimension For the target device at the current moment In the Values ​​of each dimension For the target device at the current moment The Predicted values ​​for each dimension The adaptive interference correction gain is obtained by minimizing the essential residual through the historical validation set. For the target device at the current moment In the The combined field strength of external disturbances experienced in each dimension It is the absolute value symbol.

6. The prefabricated data center fault alarm method based on an algorithm model according to claim 1, characterized in that, The method for implementing fault alarms in prefabricated data centers includes: responding to a situation where the essential residual of a target device in any dimension exceeds a preset threshold across multiple consecutive time points including the current time, determining that the target device has a fault risk, and issuing an early warning, thereby completing the fault alarm for prefabricated data centers based on an algorithm model.

7. A prefabricated data center fault alarm system based on an algorithm model, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a prefabricated data center fault alarm method based on an algorithm model according to any one of claims 1-6.

Images