Methods, systems, devices and media for fault diagnosis of sensors in computer room air conditioning systems

By denoising the sensor dataset of the smart data center air conditioning system and performing two diagnostic tests, and using ELM and GMM models, the problem of low accuracy in sensor fault diagnosis was solved, enabling earlier and more accurate fault detection.

CN117329642BActive Publication Date: 2026-11-17CHINA TELECOM CORP LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311193260.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-15
Publication Date
2026-11-17
Estimated Expiration
2043-09-15

AI Technical Summary

Technical Problem

Existing fault diagnosis methods for sensors in smart data center air conditioning systems have low accuracy in detecting hidden faults and fail to effectively handle data from multiple sensors, affecting the normal operation of the system.

Method used

The sensor dataset of the smart data center air conditioning system was acquired, and after denoising, it was input into the Extreme Learning Machine (ELM) and Gaussian Mixture Model (GMM) for two diagnostic tests. The improved Empirical Mode Decomposition (ICEEMDAN) and Extreme Learning Machine (ELM) models were combined to select diagnostic results with different failure rates.

Benefits of technology

This improves the accuracy and practicality of sensor fault diagnosis, enabling the earlier detection of hidden faults and reducing interference with the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117329642B_ABST
    Figure CN117329642B_ABST
Patent Text Reader

Abstract

This application discloses a method, system, device, and storage medium for sensor fault diagnosis in a data center air conditioning system. The method includes the following steps: acquiring a first sensor dataset of a smart data center air conditioning system; the first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring; denoising the first sensor dataset to obtain a target intrinsic mode component function set and a target combined intrinsic mode function set; inputting the target intrinsic mode component function set and the target combined intrinsic mode function set into a first processing model for a first diagnosis to obtain a first diagnosis result and a first grouped dataset; inputting the first grouped dataset into a second processing model for a second diagnosis to obtain a second diagnosis result; the fault rates of the first diagnosis result and the second diagnosis result are different. This method can improve the accuracy of fault diagnosis methods. This application can be widely applied in the field of data processing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system, device and storage medium for diagnosing sensor faults in a computer room air conditioning system. Background Technology

[0002] In related technologies, current methods for diagnosing hidden sensor faults in smart data center air conditioning systems are primarily data-driven. However, smart data center air conditioning systems operate under consistently stable cooling loads. Therefore, data-driven methods are more suitable for diagnosing hidden sensor faults. Linear sensor fault diagnosis methods often only consider faults with higher severity. Before these models detect the fault, the hidden fault may have already caused fluctuations in the thermal and humidity environment, interfering with the normal operation of the data center system. Furthermore, air conditioning system sensors include not only temperature sensors but also pressure sensors and other types of sensors. Consequently, the accuracy of current methods for diagnosing hidden sensor faults in smart data center air conditioning systems is low, and there is a significant gap between them and practical applications. Therefore, there are still technical problems that need to be solved in related technologies. Summary of the Invention

[0003] The purpose of this application is to at least partially solve one of the technical problems existing in the prior art.

[0004] Therefore, one objective of this application is to provide a method, system, device, and storage medium for diagnosing sensor faults in a computer room air conditioning system, which can improve the accuracy of fault diagnosis.

[0005] To achieve the above technical objectives, the technical solution adopted in this application includes: a sensor fault diagnosis method for a data center air conditioning system, comprising: acquiring a first sensor dataset of a smart data center air conditioning system; the first sensor dataset being used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring; performing noise reduction processing on the first sensor dataset to obtain a target intrinsic modal component function set and a target combined intrinsic modal function set; inputting the target intrinsic modal component function set and the target combined intrinsic modal function set into a first processing model for a first diagnosis to obtain a first diagnosis result and a first grouped dataset; inputting the first grouped dataset into a second processing model for a second diagnosis to obtain a second diagnosis result; the first diagnosis result being used to characterize a diagnosis result where the sensor failure rate is greater than or equal to a preset threshold and the second diagnosis result being used to characterize a diagnosis result where the sensor failure rate is less than a preset threshold.

[0006] In addition, the method for diagnosing sensor faults in a computer room air conditioning system according to the above embodiments of the present invention may also have the following additional technical features: Furthermore, in this embodiment of the application, the first sensor dataset is decomposed to obtain a first intrinsic mode component function set; The first intrinsic mode component function set is averaged to obtain the second intrinsic mode component function set; The first residual is obtained based on the second set of intrinsic mode component functions; Adding positive and negative paired second and third Gaussian white noise to the first residual yields the third set of intrinsic mode component functions; The fourth eigenmode component function set is obtained by averaging the third eigenmode component function set. The second residual is obtained based on the fourth intrinsic mode component function set; The first residual is updated to the second residual, and the first residual is added with positive and negative pairs of second Gaussian white noise and third Gaussian white noise to obtain the third intrinsic mode component function set. This process continues until the second residual is a monotonic function, thus obtaining the fifth intrinsic mode component function set. The fifth intrinsic mode component function set is filtered to obtain the intrinsic mode component function set and the combined intrinsic mode function set.

[0007] Furthermore, in this embodiment of the application, the method further includes training the first processing model, wherein training the first processing model includes: Acquire historical operational data from the sensors; Based on the historical operation data, determine the output matrix of the hidden layer nodes in the model; Based on the output matrix and the preset expected output matrix, the target output weights of the first processing model are obtained; The target output weights are used as the output weights of the first processing model after training.

[0008] Furthermore, in this embodiment of the application, the method further includes training the second processing model, wherein the second processing model is a Gaussian model, and the step of training the second processing model specifically includes: Obtain the initialization parameters, Gaussian contribution rate, and Gaussian mean of the Gaussian model; Based on the initialization parameters, the log-likelihood expectation function of the Gaussian model is constructed; the log-likelihood expectation function is:

[0009] in Let Yq be the log-likelihood expectation function, θ be each element of the training set, β be the randomly initialized likelihood function parameters, and β be the initial parameters of the Gaussian model. For the new input parameters; The first probability; When the random initialization likelihood function parameter is an invisible variable and the probability distribution is the initialization parameter, the first probability is determined according to the Bayesian probability formula; The new input parameters are adjusted based on the initialization parameters, the Gaussian contribution rate, and the Gaussian mean until the log-likelihood expectation function reaches its maximum value.

[0010] Furthermore, in this embodiment of the application, the step of filtering the fifth intrinsic mode component function set to obtain the intrinsic mode component function set and the combined intrinsic mode function set specifically includes: Calculate the contribution coefficient of each component function in the fifth intrinsic mode component function set and calculate the contribution coefficient threshold; The intrinsic mode component function set is defined as all component functions whose contribution coefficients are greater than or equal to the contribution coefficient threshold. The set of combined intrinsic modulo functions is defined as the threshold of all contribution coefficients that are less than a preset threshold.

[0011] Furthermore, in this embodiment of the application, the step of calculating the contribution coefficient threshold specifically includes: Determine the standard noise deviation based on the first intrinsic mode component function set; The contribution coefficient threshold is calculated based on the standard noise deviation.

[0012] Furthermore, in this embodiment of the application, the step of decomposing the first sensor dataset to obtain the first intrinsic mode component function set specifically includes: Add the first Gaussian white noise to the first sensor dataset to obtain the sensor data to be decomposed; The sensor data to be decomposed is subjected to EMD decomposition to obtain the first intrinsic mode component function set.

[0013] On the other hand, embodiments of this application also provide a sensor fault diagnosis system for a computer room air conditioning system, including: The acquisition unit is used to acquire the first sensor dataset of the smart data center air conditioning system; the first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring. The first processing unit is used to perform noise reduction processing on the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set. The second processing unit is used to input the target intrinsic modal component function set and the target combined intrinsic modal function set into the first processing model to perform a diagnosis, and obtain the first diagnosis result and the first grouped dataset. The third processing unit is used to input the first grouped dataset into the second processing model for secondary diagnosis to obtain a second diagnostic result; the failure rates of the first diagnostic result and the second diagnostic result are different.

[0014] On the other hand, this application also provides a sensor fault diagnosis device for a computer room air conditioning system, comprising: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a sensor fault diagnosis method for a computer room air conditioning system as described in any one of the inventions.

[0015] In addition, this application also provides a storage medium storing processor-executable instructions, which, when executed by a processor, are used to perform a sensor fault diagnosis method for a computer room air conditioning system as described in any of the preceding claims.

[0016] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application: This application can acquire real-time operating data from all sensors in a smart data center air conditioning system; denoise the real-time operating data of all sensors to obtain a target intrinsic modal component function set and a target combined intrinsic modal function set; then perform two diagnostic tests on the target intrinsic modal component function set and the target combined intrinsic modal function set to obtain two diagnostic results with different failure rates. This application considers not only temperature data but also data from various other sensors during fault diagnosis. By performing two fault diagnoses on all sensor data, this application can improve the practicality and accuracy of the fault diagnosis method. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the steps of a sensor fault diagnosis method for a computer room air conditioning system in a specific embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the steps of denoising the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set in a specific embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the steps of training the first processing model in a specific embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the steps of training the second processing model in a specific embodiment of the present invention; Figure 5This is a schematic diagram illustrating the steps of filtering the fifth intrinsic mode component function set to obtain the intrinsic mode component function set and combining the intrinsic mode function set in a specific embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the steps for calculating the contribution coefficient threshold in a specific embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the steps of decomposing the first sensor dataset to obtain the first intrinsic mode component function set in a specific embodiment of the present invention; Figure 8 This is a fault diagnosis accuracy map of the test set using the ICEEMDAN-ELM-GMM fault diagnosis method in this invention. Figure 9 Diagnostic accuracy map of the test set using the CEEMDAN-ELM fault diagnosis method in the prior art; Figure 10 Diagnostic accuracy map of the test set using the ICEEMDAN-ELM fault diagnosis method in the prior art; Figure 11 This is a comparison diagram of ablation experiments using the fault diagnosis model in this invention; Figure 12 This is a schematic diagram of the structure of a sensor fault diagnosis system for a computer room air conditioning system in a specific embodiment of the present invention; Figure 13 This is a schematic diagram of the structure of a sensor fault diagnosis device for a computer room air conditioning system in a specific embodiment of the present invention. Detailed Implementation

[0018] The following detailed description, in conjunction with the accompanying drawings, illustrates the principles and processes of the sensor fault diagnosis method, system, device, and storage medium for the computer room air conditioning system according to embodiments of the present invention.

[0019] First, let me explain the terms used in this application: 1) ELM (Extreme Learning Machine) = Extreme Learning Machine; 2) GMM (Gaussian Mixture Model) = Gaussian Mixture Model; 3) ICEEMDAN (Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) = An improved adaptive noise complete ensemble empirical mode decomposition model; 4) IMF (Intrinsic Mode Functions) = Intrinsic Mode Component Functions; 5) CIMF = Combinatorial intrinsic mode function; 6) EMD (Empirical Mode Decomposition).

[0020] The following reference Figure 1 This invention discloses a method for diagnosing sensor faults in a computer room air conditioning system. The method may include, but is not limited to, steps S101-S104. S101. Obtain the first sensor dataset of the smart data center air conditioning system; the first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring.

[0021] It should be noted that, in some embodiments of this application, the first sensor dataset can be a collection of data from various different sensors. The sensors can be categorized as follows: chilled water supply temperature sensor, chilled water return temperature sensor, chilled water supply-return temperature difference sensor, chilled water flow sensor, and supply-return pressure difference sensor, etc. The sensors mainly include three types: temperature, flow rate, and pressure sensors, and the data from these three types of sensors can have different data characteristics. Each sensor data point can be data where time and the corresponding temperature, flow rate, or pressure satisfy a certain functional relationship.

[0022] In some feasible embodiments of this application, the processor can establish a wired, wireless, or communication connection with the host computer or hardware device where the database is located. After establishing the connection, the processor can directly obtain real-time operating data of all sensors in the smart data center air conditioning system from the database.

[0023] It is understood that the above-mentioned wired connection methods may include the connection between the mobile device and the processing module, the connection between the processing module and the hardware device, and other wired connections between the processing module and other known or future-developed devices; while the above-mentioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (Ultra Wide Band) connection, and other known or future-developed wireless connection methods.

[0024] S102. Denoise the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set.

[0025] It should be noted that the target intrinsic modal component function set can be a set of intrinsic modal component functions after denoising each sensor data; while the target combined intrinsic modal function set can be a set of intrinsic modal component functions after denoising. The functions of the two sets contribute differently to fault diagnosis.

[0026] In some feasible embodiments of this application, the processor can perform noise reduction processing on the first sensor dataset obtained from the database to obtain a set of target intrinsic mode component functions corresponding to a portion of the first sensor data and a set of target combined intrinsic mode functions corresponding to another portion.

[0027] Understandably, the processor can preprocess the first sensor dataset obtained from the database before performing noise reduction processing. The purpose of preprocessing is to remove erroneous data, such as missing data or missing data from a sensor at a certain time period.

[0028] S103. Input the target intrinsic modal component function set and the target combined intrinsic modal function set into the first processing model to perform a diagnosis, and obtain the first diagnosis result and the first grouped dataset.

[0029] It should be noted that the first processing model in this application is a pre-trained processing model, which can be an ELM model. The first diagnostic result can be a diagnostic result with a failure rate greater than or equal to a certain threshold, and the first group dataset can be a set of grouped data with a failure rate less than a certain threshold.

[0030] In some feasible embodiments of this application, the processor can simultaneously input the target intrinsic modal component function set and the target combined intrinsic modal function set into the first processing model for a diagnosis. After processing by the built-in algorithm of the model, the first diagnosis result and the first grouped dataset can be obtained.

[0031] Understandably, the model's built-in algorithm can be trained using historical data from sensors; this historical data can be obtained from a database or manually entered.

[0032] S104. Input the first grouped dataset into the second processing model for secondary diagnosis to obtain the second diagnostic result; the first diagnostic result is used to characterize the diagnostic result when the sensor failure rate is greater than or equal to the preset threshold and the second diagnostic result is used to characterize the diagnostic result when the sensor failure rate is less than the preset threshold.

[0033] It should be noted that the second diagnostic result can be a failure rate below a certain threshold. Furthermore, the second processing model for the second diagnosis can differ from the first processing model.

[0034] In some feasible embodiments of this application, the processor can input the first grouped dataset into the second processing model for secondary diagnosis. After processing by the built-in algorithm of the model, the second diagnostic result can be obtained. The first diagnostic result is used to characterize the diagnostic result where the sensor failure rate is greater than or equal to a preset threshold, and the second diagnostic result is used to characterize the diagnostic result where the sensor failure rate is less than the preset threshold.

[0035] Understandably, the preset threshold can be any value, and users can adjust it according to specific circumstances. The built-in algorithm of the second processing model can also be trained using historical data from the sensor; this historical data can be obtained from a database or manually input.

[0036] Furthermore, referring to Figure 2 , Figure 2 This is a schematic diagram illustrating the steps of denoising the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set. Figure 2 The step of denoising the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set may include, but is not limited to, steps S201-S208.

[0037] S201. Decompose the first sensor dataset to obtain the first intrinsic mode component function set; S202. The first intrinsic mode component function set is averaged to obtain the second intrinsic mode component function set. S203. Obtain the first residual based on the second eigenmode component function set; S204. Add positive and negative paired second Gaussian white noise and third Gaussian white noise to the first residual to obtain the third intrinsic mode component function set; S205. Average the third intrinsic mode component function set to obtain the fourth intrinsic mode component function set; S206. Based on the fourth eigenmode component function set, obtain the second residual; S207. Update the first residual to the second residual and return to execute the step of adding positive and negative paired second Gaussian white noise and third Gaussian white noise to the first residual to obtain the third intrinsic mode component function set, until the second residual is a monotonic function to obtain the fifth intrinsic mode component function set. S208. Filter the fifth intrinsic mode component function set to obtain the intrinsic mode component function set and the combined intrinsic mode function set.

[0038] It is understandable that the first intrinsic mode component (IMF) set can be the IMF set obtained after the first decomposition of the sensor dataset. The second IMF set can be the average value of the IMFs after the first decomposition. The third IMF set can be the IMF set obtained after the second decomposition of the sensor dataset. The fourth IMF set can be the average value of the IMFs after the second decomposition. The first residual can be the residual calculated using the average value of the IMFs after the first decomposition. The second residual can be the residual calculated using the average value of the IMFs after the second decomposition. The second and third Gaussian white noises can be Gaussian white noises added to the calculation of the third IMF set. Both the second and third Gaussian white noises can be preset white noises, and the relationship between the second and third Gaussian white noises is that they are positive and negative pairs.

[0039] In some feasible embodiments of this application, the processor can decompose the first sensor dataset using formula (1) to obtain the first intrinsic mode component function set; wherein formula (1) is: (1) In formula (1) Let i be the i-th IMF obtained after EMD decomposition. The i-th IMF obtained from the CEEMDAN decomposition is usually taken as the average value. , It is a Gaussian white noise signal. The number of times white noise is added. A standard table for white noise. The data is from the first sensor; then, the first intrinsic mode component function set is averaged to obtain the second intrinsic mode component function set; specifically, refer to formula (2): (2) in, The first intrinsic mode component function set, where N is the number of functions in the function set. For summation operations; Then the processor can obtain the first residual based on the second intrinsic mode component function set; specifically, refer to formula (3): (3) Where r1(t) is the first residual and y(t) is the first sensor data. The second intrinsic mode component function set is then added to the first residual by the processor, which adds positive and negative pairs of second and third Gaussian white noise to obtain the third intrinsic mode component function set. It can be understood that the specific addition method can refer to formula (1), where t in r1(t) is replaced by the second and third Gaussian white noise.

[0040] The fourth eigenmode component function set is obtained by averaging the third eigenmode component function set. You can refer to formula (4): (4) in Let N be the set of third intrinsic mode component functions, and N be the number of functions in the set. For summation operations. Then the processor can obtain the second residual based on the fourth intrinsic mode component function set; specifically, refer to formula (5): (5) Where r1(t) is the first residual and r2(t) is the second residual. This is the fourth intrinsic mode component function set. Then the processor can repeat the steps of formula (4) and formula (5) until the second residual r2(t) is a monotonic function, thus obtaining the fifth intrinsic mode component function set; finally, the processor can filter the fifth intrinsic mode component function set according to its contribution rate to fault diagnosis, thus obtaining two different contribution rate intrinsic mode component function sets for fault diagnosis and a combined intrinsic mode function set.

[0041] Furthermore, the fault diagnosis method also includes training the first processing model. (Refer to...) Figure 3 , Figure 3 This is a schematic diagram illustrating the steps involved in training the first processing model. Figure 3 The step of training the first processing model may specifically include, but is not limited to, steps S301-S304.

[0042] S301. Obtain historical operating data of the sensor; S302. Based on historical running data, determine the output matrix of the hidden layer nodes in the model; S303. Based on the output matrix and the preset expected output matrix, obtain the target output weights of the first processing model; S304. Use the target output weights as the output weights of the first processing model after training.

[0043] Specifically, in some embodiments of this application, it is now assumed that there are N arbitrary samples. , Here, n and m are the input layer dimension and output layer dimension, respectively. The number of hidden layer nodes is set to K, and the specific single-hidden-layer feedforward neural network is represented as follows: (6) and These are the output weights and input weights, respectively. For hidden layer bias; This is the activation function.

[0044] To minimize output error, the above formula can be expressed as: (7) Simplified representation: (8) H is the output matrix of the hidden layer node; v is the output weight matrix; T is the expected output matrix; m is the number of output variables; k is 1, 2, 3, ..., m.

[0045] Since the input weights and hidden layer biases of ELM are randomly set, the output matrix of the hidden layer, denoted by H, can be obtained after these two parameters are randomly determined. + The matrix is ​​transposed, and the target output weight v is obtained using the following equation (9) to complete the training. After obtaining the output weights, the features of the sensor's operating data are passed to the trained ELM model to obtain the corresponding diagnostic results. Specifically, firstly, the features of the test data are multiplied by the weights of the hidden layer, and an activation function is applied to calculate the output of the hidden layer. Then, the output of the hidden layer is multiplied by the calculated output weights to obtain the final model output. Finally, based on the output of the ELM model, the sensor dataset input to the model can be divided into different fault diagnosis results.

[0046] (9) The ELM in this embodiment has high learning efficiency and strong generalization ability. Therefore, this technology selects Extreme Learning Machine as the sensor fault pre-diagnosis module of the smart computer room air conditioning system. Through this module, faults with low fault degree are pre-grouped, and faults with high fault degree are directly output.

[0047] Furthermore, the fault diagnosis method of this application may also include training a second processing model, wherein the second processing model is a Gaussian model. (See reference...) Figure 4 , Figure 4 This is a schematic diagram illustrating the steps involved in training the second processing model. Figure 4 The step of training the second processing model may include, but is not limited to, steps S401-S404.

[0048] S401. Obtain the initialization parameters, Gaussian contribution rate, and Gaussian mean of the Gaussian model; S402. Based on the initialization parameters, construct the log-likelihood expectation function of the Gaussian model; the log-likelihood expectation function is: (10) in Let Yq be the log-likelihood expectation function, θ be each element of the training set, β be the randomly initialized likelihood function parameters, and β be the initial parameters of the Gaussian model. For the new input parameters; The first probability; S403. When the likelihood function parameters are randomly initialized as invisible variables and the probability distribution is the initialization parameter, the first probability is determined according to the Bayesian probability formula. S404. Adjust the new input parameters according to the initialization parameters, Gaussian contribution rate, and Gaussian mean until the log-likelihood expectation function reaches its maximum value.

[0049] In some feasible embodiments of this application, the GMM may contain N Gaussian functions. The Gaussian contribution rate is... The Gaussian mean is μ, and the covariance matrix is ​​Σ, as shown in the following equation: (11) In formula (12), P represents the probability of the Gaussian mixture model, and g represents the Gaussian function.

[0050] (12) (13) The initialization parameters of the Gaussian Mixture Model (GMM) are constructed from data collected during normal operation and faulty operation. The Gaussian Mixture Model is established based on expectation maximization.

[0051] (14) The length of the training set Y={y1, y2,... yQ} is set to Q.

[0052] New parameters β is trained together with the initial parameter β, as shown below: (15) Used for training until the model converges.

[0053] The specific steps are as follows: Define Q(β, β), Q(β, The function β) represents the log-likelihood expectation of the training data, as shown in Equation (16): (16) Step 1: When the variable is invisible and the probability distribution is β, use Bayes' probability formula to calculate the probability that the training data belongs to the hypothesized hidden state. ,in Let be the first probability determined by Bayes' probability formula.

[0054] 2-Step: When Calculate the parameter values ​​when the partial derivatives of the R function for parameters μ and β are 0. .

[0055] Repeat steps 1 and 2, while re-evaluating the GMM parameters, until the likelihood function is found. Stop iterating when the maximum value is reached.

[0056] Furthermore, referring to Figure 5 , Figure 5 This is a schematic diagram illustrating the steps in this application embodiment to filter the fifth intrinsic mode component function set, obtain the intrinsic mode component function set, and combine the intrinsic mode function sets. Figure 5 In the process, the step of filtering the fifth intrinsic mode component function set to obtain the intrinsic mode component function set and the combined intrinsic mode function set may include, but is not limited to, steps S501-S503.

[0057] S501. Calculate the contribution coefficient of each component function in the fifth intrinsic mode component function set and calculate the contribution coefficient threshold. S502. The intrinsic mode component function set is the set of all component functions whose contribution coefficients are greater than or equal to the contribution coefficient threshold. S503. Use the threshold values ​​of all contribution coefficients that are less than a preset threshold as the set of combined intrinsic modulo functions.

[0058] In some feasible embodiments of this application, the contribution coefficient of each component function can be obtained by the following formula: (17) (18) (19) Where P(X(t)) is the probability of each sensor dataset, PC k For component functions C k Contribution coefficient, PC k(t) represents the contribution of each component function. The contribution coefficient threshold can be calculated using different formulas. When all contribution coefficients of a component function are greater than or equal to the contribution coefficient threshold, the component function is a function of the intrinsic modal component function set. When all contribution coefficients of a component function are less than the contribution coefficient threshold of a preset threshold, the component function is a combined intrinsic modal function set.

[0059] Furthermore, referring to Figure 6 , Figure 6 This is a schematic diagram illustrating the steps for calculating the contribution coefficient threshold in an embodiment of this application. Figure 6 The step of calculating the contribution coefficient threshold may specifically include, but is not limited to, steps S601-S602.

[0060] S601. Determine the standard noise deviation based on the first intrinsic mode component function set; S602. Determine and calculate the contribution coefficient threshold based on the standard noise deviation.

[0061] In some feasible embodiments of this application, the contribution coefficient threshold can be calculated with reference to the following formula: (20) (twenty one) W(i) represents the standard noise deviation, while P(i) represents the threshold value of the obtained contribution coefficient. First eigenmode component function set.

[0062] Furthermore, referring to Figure 7 , Figure 7 This is a schematic diagram illustrating the steps of decomposing the first sensor dataset to obtain the first set of intrinsic mode component functions. Figure 7 The step of decomposing the first sensor dataset to obtain the first intrinsic mode component function set may include, but is not limited to, steps S701-S702.

[0063] S702. Add the first Gaussian white noise to the first sensor dataset to obtain the sensor data to be decomposed; S701. Perform EMD decomposition on the sensor data to be decomposed to obtain the first intrinsic mode component function set.

[0064] In some embodiments of this application, the processor can add Gaussian white noise. New signal received ,in Then, the new signal is decomposed into EMD to obtain the first-order IMF, denoted as C1. The specific calculation formula can be found in formula (22).

[0065] (twenty two) Let i be the i-th IMF obtained after EMD decomposition. The i-th IMF obtained from the CEEMDAN decomposition is usually taken as the average value. , It is a Gaussian white noise signal. The number of times white noise is added. A standard table for white noise. The signal to be decomposed.

[0066] The specific calculation principle of this application is explained below with reference to the accompanying drawings: First, collect the operation data of the smart data center sensors, preprocess the collected historical sensor data, and remove erroneous data with missing information or other problems. Secondly, the processed data is denoised using the improved Complete Set Empirical Mode Decomposition (ICEEMDAN) method to remove high-frequency noise caused by sensor errors, resulting in a denoised smart data center sensor dataset. Specifically, Empirical Mode Decomposition (EMD) is a data denoising method that can effectively evaluate signals. Any signal can be decomposed into a set of Intrinsic Mode Functions (IMFs) and residuals using EMD. EMD was proposed for decomposing time series data, but it suffers from severe mode aliasing. To address this issue, Integrated Empirical Mode Decomposition (EEMD) was proposed, which averages the results of multiple EMD operations on the original time series with added Gaussian white noise. However, EEMD introduces a new problem: the signal recovered from the components may contain residual noise. To further improve the performance of EEMD, Fully Integrated Empirical Mode Decomposition (CEEMDAN) was proposed. This method incorporates adaptive white noise at each stage of decomposition, improving the reconstruction quality of the original signal and providing better IMF spectrum separation. Compared to EEMD, CEEMDAN reduces the number of filtering iterations and reconstruction errors, thus reducing computational cost. The CEEMDAN decomposition steps are as follows: Step 1: Add Gaussian white noise New signal received ,in The new signal is decomposed by EMD, and the first-order IMF is represented as C1.

[0067]

[0068] Let i be the i-th IMF obtained after EMD decomposition. The i-th IMF obtained from the CEEMDAN decomposition is usually taken as the average value. , It is a Gaussian white noise signal. The number of times white noise is added. A standard table for white noise. The signal to be decomposed is the historical data of the sensor.

[0069] Step 2: Perform an overall average of the generated N modal components to obtain the first IMF component of the CEEMDAN decomposition:

[0070] Step 3: Calculate the residual after removing the first modal component:

[0071] Step 4: In A new signal is obtained by adding paired positive and negative Gaussian white noise. The new signal is then used as a carrier for EMD decomposition to obtain the first-order modal components. Therefore, the second IMF of the CEEMDAN decomposition can be obtained:

[0072] Step 5: Calculate the residual after removing the second modal component:

[0073] Step 6: Repeat the above steps until the obtained residual signal is a monotonic function and cannot be further decomposed; the algorithm ends at this point. At this point, K represents the number of IMFs, and the original signal... Decomposed into:

[0074] However, the data processed by the CEEMDAN method still contains a significant amount of noise, which can affect the diagnosis of hidden faults. Therefore, this technique improves the CEEMDAN method to address this issue. Since the contribution coefficients of high-frequency components are relatively small, their impact on the classification of the fault diagnosis model is minimal, and they can be recombined into a new Combined Intrinsic Mode Function (CIMF). This technique uses the contribution coefficient as an indicator to evaluate the importance of each component to the original sequence. A general threshold formula is used to solve the method, yielding a threshold β. When PCk > P(i), it indicates that the corresponding IMF has a stable frequency and possesses a large amount of effective information; therefore, this IMF should be retained. Conversely, components below the threshold are merged into a single CIMF to reduce their impact on the fault diagnosis model. M represents the number of decomposed signals.

[0075]

[0076]

[0077] W(i) represents the standard noise deviation, while P(i) represents the obtained threshold.

[0078]

[0079]

[0080]

[0081] Then, the dataset is input into the Extreme Learning Machine (ELM) and Gaussian Mixture Model (GMM) for fault diagnosis. Since the fault severity of hidden faults is low, it is difficult for a single model to diagnose them. Therefore, the sensor fault data is input into the ELM model for fault diagnosis. The diagnostic results corresponding to the operating data of sensors with a failure rate of more than 8% are output, and the operating data of sensors with a failure rate of less than 8% are grouped together and input into the GMM model for further diagnosis. Specifically, two models can be used: one is the Extreme Learning Model, where all layers are fully connected, which improves the parameter update rate and speeds up the computation.

[0082] Now assume there are N arbitrary samples , , Here, n and m are the input and output layer dimensions, respectively. The number of hidden layer nodes is set to K, and the specific single-hidden-layer feedforward neural network is represented as follows:

[0083] and These are the output and input weights, respectively. For hidden layer bias; This is the activation function.

[0084] To minimize output error, the above formula can be expressed as:

[0085] Simplified representation:

[0086]

[0087] H is the output matrix of the hidden layer node; v is the output weight matrix; T is the expected output matrix; m is the output layer dimension m; k is 1, 2, 3, ..., m.

[0088] Due to the input weights of ELM Hidden layer bias It is randomly set. After two parameters are randomly determined, the output matrix of the hidden layer can be obtained, denoted by H. + The output weights v are then calculated using the following formula, which is the transpose of H, to complete the training. After obtaining the output weights, the processor can pass the features of the test data to the trained ELM model. Specifically, first, the features of the test data are multiplied by the weights of the hidden layer, and an activation function is applied to calculate the output of the hidden layer. Then, the output of the hidden layer is multiplied by the calculated output weights to obtain the final model output.

[0089] Classification results: Finally, based on the output of the ELM model, the samples can be classified into different categories.

[0090]

[0091] ELM has high learning efficiency and strong generalization ability. Therefore, this technology selects Extreme Learning Machine as the sensor fault pre-diagnosis module of the smart computer room air conditioning system. Through this module, faults with low fault severity are pre-grouped, and faults with high fault severity are directly output. This model mainly outputs sensor data diagnosis results for faults with a fault rate of more than 8% and outputs grouped data for faults with a fault rate of less than 8%.

[0092] Another model is the Gaussian model, which contains N Gaussian functions. The Gaussian contribution rate is... The Gaussian mean is μ, and the covariance matrix is ​​Σ.

[0093]

[0094] In the following formula, P represents the probability of the Gaussian mixture model, and g represents the Gaussian function.

[0095]

[0096]

[0097] The initialization parameters of the Gaussian Mixture Model (GMM) are constructed from data collected during normal operation and faulty operation. The Gaussian Mixture Model is established based on expectation maximization.

[0098]

[0099] The length of the training set Y={y1, y2,... yQ} is set to Q.

[0100] New parameters β is trained together with the initial parameter β, as shown below.

[0101]

[0102] β is used for training until the model converges.

[0103] The specific steps are as follows: Define Q(β, β), Q(β, The function β) represents the expected log-likelihood of the training data, as shown in the following formula:

[0104] Step 1: When θ is an invisible variable and the probability distribution is β, use Bayes' probability formula to calculate the probability that the training data belongs to the hypothetical hidden state θ.

[0105] 2-Step: When Calculate the parameter values ​​when the partial derivatives of the R function for the parameters μ and β are 0; Repeat steps 1 and 2, while re-evaluating the GMM parameters. Stop iterating when the maximum likelihood function is found. After obtaining the maximum likelihood function, the Gaussian model is trained. Input the group data with a failure rate below 8% from the Extreme Learning Machine output into the GMM model for re-diagnosis, and obtain a diagnosis result with a failure rate below 8% as the corresponding diagnosis result.

[0106] Finally, the fault diagnosis results are output. This method improves the success rate of diagnosing hidden faults through a hierarchical diagnosis strategy. The accuracy of fault diagnosis in this application can be referenced. Figure 8 by Figure 11 ,exist Figure 8 as well as Figure 11 From this, it can be seen that compared to Figure 9 The test set diagnostic method using the CEEMDAN-ELM fault diagnosis method and Figure 10 The accuracy of the ICEEMDAN-ELM fault diagnosis method improves significantly with the increase of the number of samples.

[0107] In addition, refer to Figure 12 ,and Figure 1Corresponding to the method described above, this application also provides a sensor fault diagnosis system for a data center air conditioning system. The system may include: an acquisition unit 1001, a first processing unit 1002, a second processing unit 1003, and a third processing unit 1004. The acquisition unit 1001 can be used to acquire a first sensor dataset of a smart data center air conditioning system; the first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring; the first processing unit 1002 can be used to perform denoising processing on the first sensor dataset to obtain a target intrinsic mode component function set and a target combined intrinsic mode function set; the second processing unit 1003 can be used to input the target intrinsic mode component function set and the target combined intrinsic mode function set into a first processing model for a first diagnosis to obtain a first diagnosis result and a first grouped dataset; the third processing unit 1004 can be used to input the first grouped dataset into a second processing model for a second diagnosis to obtain a second diagnosis result; the failure rates of the first diagnosis result and the second diagnosis result are different.

[0108] It should be noted that the content of the above-described embodiments of the data center air conditioning system sensor fault diagnosis method is applicable to the embodiments of this data center air conditioning system sensor fault diagnosis system. The specific functions implemented by the embodiments of this data center air conditioning system sensor fault diagnosis system are the same as those of the above-described embodiments of the data center air conditioning system sensor fault diagnosis method, and the beneficial effects achieved are also the same as those achieved by the above-described embodiments of the data center air conditioning system sensor fault diagnosis method.

[0109] and Figure 1 Corresponding to the method described herein, embodiments of this application also provide a sensor fault diagnosis device for a computer room air conditioning system, the specific structure of which can be referred to Figure 13 ,include: At least one processor 1011; At least one memory 1012 is used to store at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the sensor fault diagnosis method for the computer room air conditioning system.

[0110] It should be noted that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0111] and Figure 1 Corresponding to the method described above, this application also provides a storage medium storing processor-executable instructions, which, when executed by the processor, are used to perform the aforementioned sensor fault diagnosis method for a computer room air conditioning system.

[0112] It should be noted that the content of the above embodiments of the data center air conditioning system sensor fault diagnosis method is applicable to this storage medium embodiment. The specific functions implemented by this storage medium embodiment are the same as those of the above embodiments of the data center air conditioning system sensor fault diagnosis method, and the beneficial effects achieved are also the same as those achieved by the above embodiments of the data center air conditioning system sensor fault diagnosis method.

[0113] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0114] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0115] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several programs to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable programs for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can retrieve and execute a program from or in conjunction with such a program execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with a program execution system, apparatus, or device.

[0117] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0118] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0119] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0120] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0121] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for diagnosing sensor faults in a computer room air conditioning system, characterized in that, include: Obtain the first sensor dataset of the smart data center air conditioning system; The first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring; The first sensor dataset is denoised to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set; The target intrinsic modal component function set and the target combined intrinsic modal function set are input into the first processing model for a diagnosis to obtain a first diagnosis result and a first grouped dataset. The first diagnosis result is a diagnosis result with a failure rate greater than or equal to a preset threshold. The first grouped dataset is a grouped data set with a failure rate less than a certain threshold. The failure rate is used to characterize the degree of failure. The first grouped dataset is input into the second processing model for secondary diagnosis to obtain a second diagnostic result; the second diagnostic result is used to characterize the diagnostic result that the sensor failure rate is less than a preset threshold.

2. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 1, characterized in that, The step of denoising the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set specifically includes: The first sensor dataset is decomposed to obtain the first intrinsic mode component function set; The first intrinsic mode component function set is averaged to obtain the second intrinsic mode component function set; The first residual is obtained based on the second set of intrinsic mode component functions; Adding positive and negative paired second and third Gaussian white noise to the first residual yields the third set of intrinsic mode component functions; The fourth eigenmode component function set is obtained by averaging the third eigenmode component function set. The second residual is obtained based on the fourth intrinsic mode component function set; The first residual is updated to the second residual and the process is returned to execute the step of adding positive and negative paired second Gaussian white noise and third Gaussian white noise to the first residual to obtain the third intrinsic mode component function set, until the second residual is a monotonic function to obtain the fifth intrinsic mode component function set; The fifth intrinsic mode component function set is filtered to obtain the intrinsic mode component function set and the combined intrinsic mode function set.

3. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 1, characterized in that, The method further includes training the first processing model, wherein training the first processing model includes: Acquire historical operational data from the sensors; Based on the historical operation data, determine the output matrix of the hidden layer nodes in the model; Based on the output matrix and the preset expected output matrix, the target output weights of the first processing model are obtained; The target output weights are used as the output weights of the first processing model after training.

4. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 1, characterized in that, The method further includes training the second processing model, which is a Gaussian model. The step of training the second processing model specifically includes: Obtain the initialization parameters, Gaussian contribution rate, and Gaussian mean of the Gaussian model; Based on the initialization parameters, the log-likelihood expectation function of the Gaussian model is constructed; the log-likelihood expectation function is: in Let Yq be the log-likelihood expectation function, θ be each element of the training set, β be the randomly initialized likelihood function parameters, and β be the initial parameters of the Gaussian model. For the new input parameters; The first probability; When the random initialization likelihood function parameter is an invisible variable and the probability distribution is the initialization parameter, the first probability is determined according to the Bayesian probability formula; The new input parameters are adjusted based on the initialization parameters, the Gaussian contribution rate, and the Gaussian mean until the log-likelihood expectation function reaches its maximum value.

5. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 2, characterized in that, The step of filtering the fifth intrinsic mode component function set to obtain the intrinsic mode component function set and the combined intrinsic mode function set specifically includes: Calculate the contribution coefficient of each component function in the fifth intrinsic mode component function set and calculate the contribution coefficient threshold; The intrinsic mode component function set is defined as all component functions whose contribution coefficients are greater than or equal to the contribution coefficient threshold. The set of combined intrinsic modulo functions is defined as the threshold of all contribution coefficients that are less than a preset threshold.

6. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 5, characterized in that, The step of calculating the contribution coefficient threshold specifically includes: Determine the standard noise deviation based on the first intrinsic mode component function set; The contribution coefficient threshold is calculated based on the standard noise deviation.

7. The method for diagnosing sensor faults in a computer room air conditioning system according to claim 2, characterized in that, The step of decomposing the first sensor dataset to obtain the first intrinsic mode component function set specifically includes: Add the first Gaussian white noise to the first sensor dataset to obtain the sensor data to be decomposed; The sensor data to be decomposed is subjected to EMD decomposition to obtain the first intrinsic mode component function set.

8. A sensor fault diagnosis system for a computer room air conditioning system, characterized in that, include: The acquisition unit is used to acquire the first sensor dataset of the smart data center air conditioning system. The first sensor dataset is used to characterize the real-time operating data of all sensors in the smart data center air conditioning system under real-time monitoring; The first processing unit is used to perform noise reduction processing on the first sensor dataset to obtain the target intrinsic mode component function set and the target combined intrinsic mode function set. The second processing unit is used to input the target intrinsic modal component function set and the target combined intrinsic modal function set into the first processing model for a diagnosis to obtain a first diagnosis result and a first grouped dataset. The first diagnosis result is a diagnosis result with a failure rate greater than or equal to a preset threshold. The first grouped dataset is a grouped data set with a failure rate less than a certain threshold. The failure rate is used to characterize the degree of failure. The third processing unit is used to input the first grouped dataset into the second processing model for secondary diagnosis to obtain the second diagnostic result. The second diagnostic result is used to characterize the diagnostic result when the sensor failure rate is less than a preset threshold.

9. A sensor fault diagnosis device for a computer room air conditioning system, characterized in that... include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the sensor fault diagnosis method for a computer room air conditioning system as described in any one of claims 1-7.

10. A storage medium storing processor-executable instructions, characterized in that, The processor-executable instructions, when executed by the processor, are used to perform the sensor fault diagnosis method for a computer room air conditioning system as described in any one of claims 1-7.

Citation Information

Patent Citations

  • An equipment fault diagnosis method and device based on cloud service

    CN109711062A

  • Aero-engine control system sensor fault diagnosis method based on AANN network group

    CN111474919A