IDC machine room environment intelligent monitoring method and monitoring system thereof

Abstract

The invention discloses an intelligent monitoring method and a monitoring system for an IDC (Internet Data Center) machine room environment, particularly relates to the field of IDC machine room environment monitoring, is used for solving the problems of multi-parameter relevance and real-time early warning, and aims at solving the problems of multi-parameter relevance and real-time early warning by constructing a high-dimensional topological structure, a Bayesian network model and a dynamic threshold adjustment mechanism of the IDC machine room environment. The relevance and the causal relationship among the multi-dimensional parameters are effectively captured, and the analysis and prediction capability on the complex environment data is remarkably improved. And by utilizing a persistent coherence algorithm and random forest regression, nonlinear correlation of parameters is accurately identified, and the limitation of a traditional method in multi-parameter analysis is overcome. Through Granger causal relationship test, a parameter causal sequence is determined, and a basis is provided for accurate prediction of environmental changes. A dynamic threshold adjustment mechanism flexibly adapts to environment changes, false alarm and missing alarm are reduced, it is ensured that the system rapidly identifies abnormities and triggers alarms in a complex environment, and the intelligence and stability of the system are remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the field of IDC computer room environment monitoring, and more specifically, to an IDC computer room environment intelligent monitoring method and a monitoring system thereof. Background Art

[0002] IDC computer room (Internet Data Center computer room) is a facility dedicated to storing key IT infrastructure such as servers and network equipment. In modern IDC computer room environmental monitoring, with the expansion of data center scale and the increase of system complexity, the monitoring of environmental parameters is no longer limited to the tracking of a single indicator, but requires comprehensive monitoring of multi-dimensional parameters such as temperature, humidity, power load, and air flow speed. There are often complex interactions and correlations between these parameters. Only by accurately identifying and analyzing these correlations can we timely warn of potential system risks and ensure the stable operation of the computer room. However, due to the high dimensionality and diversity of data, existing monitoring systems often face challenges in dealing with multi-parameter associations, and it is difficult to fully capture the potential dependencies and interaction mechanisms between the parameters.

[0003] When dealing with multi-parameter correlations, existing technologies usually rely on static threshold settings and single parameter analysis methods. This results in the system being unable to accurately identify the dynamic correlations between multiple parameters in complex environments, which in turn affects the accuracy of early warnings. For example, during the monitoring process, the potential correlations between multiple parameters are often overlooked, causing the system to issue an alarm only when a certain parameter exceeds the threshold, and failing to predict and identify the overall trend of system changes in advance. This approach is not only prone to false alarms and missed alarms, but can also lead to delayed processing of key issues, and cannot meet the requirements of high stability and high reliability.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiment of the present invention provides an IDC computer room environment intelligent monitoring method and monitoring system thereof, which systematically captures and analyzes the potential correlation and causal relationship between multidimensional parameters by constructing a high-dimensional topological structure of the IDC computer room environment, a Bayesian network model and a dynamic threshold adjustment mechanism, and significantly improves the analysis and prediction capabilities of complex environmental data. By combining the persistent homology algorithm with the random forest regression, the nonlinear correlation of environmental parameters is accurately identified, and its persistence and consistency at different scales are robustly evaluated, overcoming the limitations of traditional monitoring methods in multi-parameter correlation analysis. The Granger causality test further clarifies the causal order between parameters, providing a solid basis for the accurate prediction of environmental changes. The dynamic threshold adjustment mechanism can flexibly adapt to environmental changes in real-time monitoring, improve the sensitivity and reliability of early warning, reduce false alarms and missed alarms, and ensure that the system can quickly identify anomalies and trigger alarms in a complex and changeable environment. It effectively improves the intelligence, accuracy and stability of the IDC computer room environment monitoring system, and provides comprehensive and reliable technical support for optimizing decision-making and risk early warning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Step S1, obtaining multi-dimensional environmental data of the IDC room, performing outlier detection, missing value filling and data smoothing, extracting features through principal component analysis and autoencoder, and converting high-dimensional data into low-dimensional representative features;

[0008] Step S2, based on the extracted multidimensional environmental data, construct a high-dimensional topological structure, apply the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence map and spectrum clustering results, and judge and identify the significant correlation effects between multiple parameters;

[0009] Step S3, after identifying the multi-parameter correlation effect, the parameter dependency model is constructed using the Bayesian network, the potential correlation is revealed through the conditional probability distribution, the nonlinear relationship is captured and the impact is quantified using random forest regression, the causal order between parameters is clarified through the Granger causality test, and the parameter correlation analysis is improved;

[0010] Step S4, based on the conditional probability distribution model of the Bayesian network, the conditional information volume is used to identify key parameters, and a dynamic threshold function is constructed by combining feature importance and causal relationship to monitor in real time and trigger an alarm mechanism.

[0011] In a preferred embodiment, step S2 includes the following contents:

[0012] S2.1, embed the multi-dimensional environmental parameter data of the IDC room into a high-dimensional space to form a point cloud data set. Assume that the point cloud data set is presented in the form of a time series, and let the data vector at each time point be Where t is a moment in the time series, n is the parameter dimension, and the data vectors of all time points form a point cloud x = {x1, x2, ..., x T}, where T is the total number of time points;

[0013] S2.2, on the generated point cloud, apply the Vietoris-Rips complex algorithm to construct a high-dimensional topological complex, set the distance threshold ∈, for any two points x in the point cloud i and x j , if their Euclidean distance is less than or equal to, then connect an edge between these two points;

[0014] S2.3, based on the constructed Vietoris-Rips complex, the homology group H is calculated using the persistent homology algorithm m The persistence of m represents the dimension of the homology group, and we focus on the low-dimensional homology groups of connected components H0, loops H1 and cavities H2. By calculating the persistence of the homology group at different distance thresholds, we get the persistent barcode, where the length of the barcode represents the persistence of the corresponding topological feature at different scales.

[0015] formula:

[0016] in, represents the boundary operator of the m-dimensional chain;

[0017] S2.4, convert the persistence barcode into a multi-scale persistence map, and form a multi-scale persistence map by calculating the persistence at different distance threshold scales; perform multi-scale analysis on all persistence maps, and identify significant association patterns between multiple parameters through decomposition and feature extraction;

[0018] formula:

[0019] Persistence graph, a visual representation of persistent homology at different distance threshold scales;

[0020]

[0021] Multi-scale persistence graph, through persistence analysis at different scales, shows the topological associations between multiple parameters.

[0022] In a preferred embodiment, S2.5, after completing steps S2.1-2.4, determining whether there is a significant multi-parameter correlation effect by analyzing the multi-scale persistence map;

[0023] The length of the persistent barcode is obtained by calculating the difference between the generation and disappearance of the topological feature under different scale distance thresholds; specifically, the generation scale of the topological feature is recorded when it first appears, and the disappearance scale is recorded when the feature disappears. The length of the persistent barcode is the difference between the disappearance scale and the generation scale of the feature.

[0024] The similarity metric of multi-scale persistence graphs is calculated by comparing the structural features of persistence graphs at different scale distance thresholds. At each scale distance threshold, the persistence graph is transformed into a graph structure, where nodes represent topological features and edges represent the associations between features. Then, the Pearson correlation coefficient metric is used to evaluate the structural similarity between persistence graphs at different scale distance thresholds.

[0025] The cohesion of the spectral graph clustering results is determined by measuring the closeness of the nodes within the cluster; first, the standardized graph Laplacian matrix of the parameter association graph is constructed, and the eigenvalue decomposition is performed to decompose the matrix into eigenvectors and eigenvalues. The first rated number of eigenvectors are used as the new feature space for clustering. For each cluster, its internal interconnectivity is calculated, and the average value of the internal interconnectivity of all clusters is taken as the cohesion of the clustering results.

[0026] In a preferred embodiment, the final judgment criterion is expressed by the following formula:

[0027]

[0028] Among them, Result is the judgment result; Significant Association means the existence of multi-parameter association effect; No Significant Association means the absence of multi-parameter association effect;

[0029] L β is the length of the persistent barcode, and the persistence threshold δ p Make comparisons;

[0030] S γ is the similarity measure of the multi-scale persistence graph, and the similarity threshold δ s Make comparisons;

[0031] C θ is the cohesion of the spectral clustering result, and the cohesion threshold δ c Make a comparison.

[0032] In a preferred embodiment, step S3 includes the following contents:

[0033] S3.1, after identifying the existence of multi-parameter correlation effects and completing feature extraction, use the Bayesian network to perform structured modeling on the multi-dimensional environmental parameters of the IDC computer room; the specific steps include, firstly, using the method of maximizing the posterior probability, using the training data to perform structural learning of the Bayesian network, and constructing a directed acyclic graph structure between the parameters, where each node represents an environmental parameter and each edge represents a conditional dependency relationship between two parameters;

[0034] S3.2, based on the Bayesian network structure, random forest regression is further applied to capture the nonlinear relationship between multiple parameters. The specific processing steps are as follows: first, multiple sample subsets and feature subsets are randomly extracted from the environmental parameter data set, and multiple decision trees are constructed based on the randomly extracted multiple sample subsets and feature subsets. Each decision tree is independently generated to predict the value of the target parameter;

[0035] S3.3, use the Granger causality test to further detect the temporal causal relationship between multiple parameters. The processing steps include: first, smoothing the time series data of each environmental parameter to ensure that the data meets the prerequisites of the Granger causality test; then, building a regression model, taking the historical value of a parameter and the current value of other parameters as independent variables, and the current value of the target parameter as the dependent variable, calculating the regression residual and prediction error; then, comparing the prediction accuracy of the regression model that does not contain the historical value of a parameter with the regression model that contains the historical value of the corresponding parameter; if the prediction accuracy is improved to above the corresponding threshold after introducing the historical value of a parameter, it indicates that the historical value of the corresponding parameter has a significant influence and causal relationship on the current value of the target parameter.

[0036] In a preferred embodiment, step S4 includes the following contents:

[0037] S4.1, based on the conditional dependency determined by the Bayesian network, construct the conditional probability distribution model of each environmental parameter, assuming that the environmental parameters are A1, A2, …, A q According to the structure of the Bayesian network, the probability distribution P(A) of each parameter given its parent node is calculated. b ∣Pre(A b )), where Pre(A b ) means A b The parent node set of A1, A2, ..., A q );

[0038]

[0039] S4.2, using the conditional information I(A b ; A c ∣Pre(A c))Measure the information transmission intensity between parameters and identify the key parameters that have the greatest impact on the target parameter. The conditional information is used to evaluate the information dependency between two parameters given other parameters. The parameter with the largest conditional information is selected as the key influencing factor of the target parameter, and the threshold of the parameter with the largest conditional information is dynamically adjusted.

[0040]

[0041] S4.3, after identifying the key parameters, a dynamic threshold adjustment function is constructed, using the feature importance coefficient Imp(A b ) and Granger causality test results, set the dynamic threshold function τ(B) of the target parameter B;

[0042]

[0043] Among them, g(a b ) is a nonlinear transformation function, h(B|Pre(B)) is the predicted value λ of the target parameter under the condition of its previous node b and μ are model coefficients, which determine the weight of the key parameters on the threshold. It represents the number of key influencing parameters, that is, the number of key parameters involved in the calculation when the dynamic threshold of the target parameter B is adjusted.

[0044] In a preferred embodiment, S4.4, during the real-time monitoring process, a dynamic threshold function is used to calculate the real-time threshold of each environmental parameter. The real-time data obtained by the monitoring system is input into the dynamic threshold model to calculate the threshold at the current moment. If the actual monitoring value B obs If the calculated threshold is exceeded, the alarm mechanism is triggered and emergency processing is carried out according to the preset strategy;

[0045]

[0046] Alarm indicates the triggering condition of the alarm status. Whether the alarm is triggered is determined by whether the actual monitoring data exceeds the dynamic threshold.

[0047] Activate means triggering an alarm, that is, when the monitoring data exceeds the dynamic threshold, the system enters the alarm state and executes the corresponding emergency treatment measures;

[0048] Standby indicates the standby state, that is, when the monitoring data is within the dynamic threshold range, the system maintains normal operation and does not trigger an alarm.

[0049] An IDC computer room environment intelligent monitoring system, comprising: a multi-dimensional data acquisition module, a topology structure analysis module, a causal association model module, and a dynamic threshold control module;

[0050] The multi-dimensional data acquisition module acquires the multi-dimensional environmental data of the IDC room, performs outlier detection, missing value filling and data smoothing, extracts features through principal component analysis and autoencoders, converts high-dimensional data into low-dimensional representative features, generates a low-dimensional environmental feature data matrix, and transmits it to the topological structure analysis module;

[0051] The topological structure analysis module constructs a high-dimensional topological structure based on the extracted multi-dimensional environmental data, applies the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence graph and spectral clustering results, judges and identifies the significant correlation effects between multiple parameters, generates the identification results of the multi-parameter correlation effects and their topological structure information, and transmits them to the causal association model module;

[0052] After identifying the multi-parameter correlation effect, the causal association model module uses the Bayesian network to build a parameter dependency model, reveals potential correlations through conditional probability distribution, uses random forest regression to capture nonlinear relationships and quantify the impact, and uses the Granger causality test to clarify the causal order between parameters, improves parameter correlation analysis, generates the dependency structure, nonlinear relationship and causal chain of key parameters, and transmits them to the dynamic threshold control module;

[0053] The dynamic threshold control module is based on the conditional probability distribution model of the Bayesian network. It uses conditional information to identify key parameters, combines feature importance and causal relationships to build a dynamic threshold function, and monitors and triggers the alarm mechanism in real time.

[0054] The technical effects and advantages of an IDC computer room environment intelligent monitoring method and monitoring system of the present invention are as follows:

[0055] 1. The present invention constructs a high-dimensional topological structure of the IDC computer room environment and uses a persistent homology algorithm to accurately capture the potential correlation between multi-dimensional parameters, thereby improving the ability to parse complex environmental data. It is highly scientific and operational, and thus overcomes the limitations of traditional monitoring methods for multi-parameter correlation analysis, which helps to accurately identify system risks in a dynamic environment and ensure the comprehensiveness and stability of the monitoring system. At the same time, combined with persistent barcodes and multi-scale persistence graph analysis methods, it can robustly evaluate the persistence and consistency of parameter associations at different scales, breaking through the shortcomings of single-scale monitoring, and has the advantages of high efficiency and accurate identification. Ultimately, it provides solid theoretical support and technical guarantee for the optimization and decision-making of the IDC computer room environment monitoring system.

[0056] 2. The present invention systematically reveals the conditional dependency and potential correlation between the environmental parameters of the IDC computer room by constructing a Bayesian network model, which makes up for the shortcomings of traditional analysis methods in dealing with complex multi-parameter correlations. The application of random forest regression further captures the nonlinear relationship between environmental parameters, quantifies the impact of each parameter on the target state, and makes the analysis results more comprehensive and accurate. In addition, through the Granger causality test, the causal order between parameters is clarified, and the ability to predict changes in environmental parameters is provided. Combining the above methods, step S3 not only improves the analysis depth of multi-parameter correlation, but also provides solid data support and scientific basis for the optimization of the system and the formulation of early warning mechanisms, significantly enhancing the intelligence and precision level of IDC computer room environmental monitoring.

[0057] 3. The present invention can accurately capture the dependencies between various environmental parameters by constructing a conditional probability distribution model based on a Bayesian network, and use the conditional information to identify the key factors that have the greatest impact on the target parameters, thereby improving the accuracy of threshold adjustment. The dynamic threshold function set in combination with random forest regression and Granger causality test can flexibly adapt to environmental changes in real-time monitoring, making the system early warning more sensitive and reliable. In addition, the real-time calculation and adjustment of dynamic thresholds ensure that in a complex and changeable environment, the system can quickly identify anomalies and trigger alarms, reduce the possibility of false alarms and missed alarms, and ensure the operational safety and stability of the IDC computer room. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of a process of an IDC computer room environment intelligent monitoring method of the present invention;

[0059] Figure 2 The present invention is a structural schematic diagram of an IDC computer room environment intelligent monitoring system. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] Figure 1 The present invention provides an IDC computer room environment intelligent monitoring method, comprising:

[0063] Step S1, obtaining multi-dimensional environmental data of the IDC room, performing outlier detection, missing value filling and data smoothing, extracting features through principal component analysis and autoencoder, and converting high-dimensional data into low-dimensional representative features;

[0064] Step S2, based on the extracted multidimensional environmental data, construct a high-dimensional topological structure, apply the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence map and spectrum clustering results, and judge and identify the significant correlation effects between multiple parameters;

[0065] Step S3, after identifying the multi-parameter correlation effect, the parameter dependency model is constructed using the Bayesian network, the potential correlation is revealed through the conditional probability distribution, the nonlinear relationship is captured and the impact is quantified using random forest regression, the causal order between parameters is clarified through the Granger causality test, and the parameter correlation analysis is improved;

[0066] Step S4, based on the conditional probability distribution model of the Bayesian network, the conditional information volume is used to identify key parameters, and a dynamic threshold function is constructed by combining feature importance and causal relationship to monitor in real time and trigger an alarm mechanism.

[0067] Step S1 includes the following contents:

[0068] First, the multi-dimensional data in the IDC computer room environment monitoring is preliminarily cleaned and preprocessed. The specific steps include outlier detection, missing value filling and data smoothing. In the process of outlier detection, statistical methods or machine learning algorithms are used to identify and mark outliers in the data, and these outliers are deleted, corrected or processed by interpolation according to the actual situation. For missing values ​​in the data set, mean filling, linear interpolation, previous and next value filling or K nearest neighbor interpolation method based on similarity are used to fill them reasonably. In order to reduce random noise in the data, smoothing algorithms such as moving average, exponential smoothing or Gaussian filtering are used to smooth the time series data to obtain a more stable time series.

[0069] After data preprocessing, feature extraction is performed. The steps of feature extraction include principal component analysis and autoencoder methods. Principal component analysis constructs a covariance matrix, decomposes its eigenvalues, selects the eigenvector with the largest explained variance, and forms a new low-dimensional data set, thereby reducing the processed high-dimensional data to a lower dimension and retaining the main information features. The autoencoder constructs a neural network, uses an encoder to map high-dimensional data to a low-dimensional latent space, and reconstructs the original data through a decoder. By minimizing the error between the input data and the reconstructed data, representative low-dimensional features are obtained.

[0070] Through the above preprocessing and feature extraction process, the original multi-dimensional environmental data is converted into low-dimensional representative features, which not only retain the main information of the data but also effectively reduce the data noise, thereby providing a more efficient and accurate data basis for the topological analysis in step S2.

[0071] Step S2 includes the following contents:

[0072] S2.1, embed the multi-dimensional environmental parameter data of the IDC room (such as temperature, humidity, power load, air flow speed, etc.) into a high-dimensional space to form a point cloud data set. Assume that these parameter data are presented in the form of time series, and let the data vector at each time point be Where t is a moment in the time series and n is the parameter dimension. The data vectors of all time points form a point cloud X = {x1, x2, ..., x T}, where T is the total number of time points.

[0073]

[0074] x={x1,x2,…,x T};

[0075] By mapping multi-dimensional environmental parameter data into high-dimensional space, a time series point cloud is formed, which is the basis for establishing a complex correlation structure between parameters. This point cloud data will be used as input for subsequent topological analysis to capture the potential correlation between parameters and data distribution characteristics.

[0076] S2.2, on the generated point cloud, apply the Vietoris-Rips complex algorithm to construct a high-dimensional topological complex. Set the distance threshold ∈, for any two points x in the point cloud i and x j , if their Euclidean distance is less than or equal to, then connect an edge between these two points. Then, this method gradually constructs higher-dimensional simplexes in high-dimensional space, such as triangles, tetrahedrons, etc., to form the Vietoris-Rips complex

[0077] The corresponding Euclidean distance is:

[0078] Vietoris-Rips Complex:

[0079] Among them, σ represents any subset of the point set, and diam(σ) represents the diameter of the subset.

[0080] By constructing the Vietoris-Rips complex, we can capture the high-dimensional structural features in the point cloud data and define the relationship between multi-dimensional parameters. The choice of distance threshold determines the level of detail of the captured topological structure. The smaller the distance threshold, the more fine-grained the structural relationship can be revealed.

[0081] S2.3, based on the constructed Vietoris-Rips complex, the homology group H is calculated using the persistent homology algorithm m m represents the dimension of the homology group, and we focus on low-dimensional homology groups such as H0 (connected components), H1 (loops), and H2 (cavities). By calculating the persistence of the homology group at different distance thresholds, we get the persistence barcode, where the length of the barcode represents the persistence of the corresponding topological feature at different scales.

[0082] formula:

[0083] in, Represents the boundary operator of an m-dimensional chain.

[0084] Persistent homology groups reveal the persistence of topological features of different dimensions in the data space. By analyzing the length and number of persistent barcodes, we can identify the complex correlation structure between environmental parameters. Longer barcodes represent more stable and persistent correlation effects between parameters.

[0085] S2.4, convert the persistence barcode into a multi-scale persistence map, and form a multi-scale persistence map by calculating the persistence at different distance threshold scales. Perform multi-scale analysis on these persistence maps, and identify significant association patterns between multiple parameters through decomposition and feature extraction. Combined with advanced graph theory analysis (such as spectral clustering or multi-graph matching), further quantify the correlation between parameters.

[0086] formula:

[0087] Persistence graph, a visualization of persistent homology at different distance threshold scales.

[0088]

[0089] Multi-scale persistence graph, through persistence analysis at different scales, shows the topological associations between multiple parameters.

[0090] Through Bootstrap robustness analysis, we verify whether the identified multi-parameter correlation structure is stable and reliable. We optimize the parameter setting of topological analysis, improve the accuracy and reliability of correlation identification, and ensure the effectiveness of the final correlation structure in practical applications.

[0091] S2.5, after completing steps S2.1-2.4, determine whether there is a significant multi-parameter association effect by analyzing the multi-scale persistence map.

[0092] In IDC computer room environment monitoring, identifying the correlation between multiple parameters is a key task to ensure system stability and prevent potential failures. The length of the persistent barcode is used to evaluate the persistence of environmental parameters at different scales, revealing the potential correlation between parameters in high-dimensional space. This length indicator can identify those correlations that are stable at multiple scales, thereby eliminating accidental factors. The similarity measure of the multi-scale persistence graph measures the consistency of the multi-parameter correlation structure at multiple scales by comparing the persistence graph structure at different scales, ensuring that the identified correlation is universal and reliable under various operating conditions and avoiding the limitations of a single scale. The cohesion of the spectral clustering results further quantifies the significance of these patterns by analyzing the tightness of the parameter correlation patterns, helping to determine whether there is a structural correlation, ensuring that the analysis results are not only based on the persistence and similarity of the topological features, but also taking into account the actual correlation strength and consistency between the parameters. Combining these three indicators, it is possible to comprehensively and accurately identify the multi-parameter correlation and ensure the effectiveness and stability of the IDC computer room environment monitoring system under complex conditions.

[0093] The length of the persistent barcode is obtained by calculating the difference between the generation and disappearance of the topological feature at different scale distance thresholds. Specifically, the generation scale of the topological feature is recorded when it first appears, and the disappearance scale is recorded when the feature disappears. The length of the persistent barcode is the difference between the disappearance scale and the generation scale of the feature. The length of the persistent barcode reflects the persistence of the topological feature at different scales.

[0094] The similarity metric of multi-scale persistence graphs is calculated by comparing the structural features of persistence graphs at different scale distance thresholds. At each scale distance threshold, the persistence graph is transformed into a graph structure, where nodes represent topological features and edges represent associations between features. Then, metrics such as the Pearson correlation coefficient or the Jaccard similarity coefficient are used to evaluate the structural similarity between persistence graphs at different scale distance thresholds. The higher the similarity metric value, the more consistent the association structure at different scales.

[0095] The cohesion of the spectral graph clustering results is determined by measuring the closeness of the nodes within the cluster. First, the standardized graph Laplacian matrix of the parameter association graph is constructed, and the eigenvalue decomposition is performed to decompose the matrix into eigenvectors and eigenvalues, and the first rated number of eigenvectors are used as the new feature space for clustering. For each cluster, its modularity coefficient or internal interconnectivity is calculated, and the average value of the modularity coefficient or internal interconnectivity of all clusters is used as the cohesion of the clustering result. The higher the cohesion value, the higher the closeness of the clustering result and the strength of the parameter association.

[0096] The final judgment criteria are expressed by the following formula:

[0097]

[0098] Among them, Result is the judgment result; Significant Association means the existence of multi-parameter association effect; No Significant Association means the absence of multi-parameter association effect;

[0099] L β is the length of the persistent barcode, and the persistence threshold δ p Make a comparison.

[0100] S γ is the similarity measure of the multi-scale persistence graph, and the similarity threshold δ s Make a comparison.

[0101] C θ is the cohesion of the spectral clustering result, and the cohesion threshold δ c Make a comparison.

[0102] If the judgment result is that there is a multi-parameter correlation effect, it means that there is a significant interdependence between the multi-dimensional environmental parameters in the IDC room. This correlation indicates that multiple parameters are jointly affecting the environmental state, and there may be potential systemic changes or risks, which require further analysis to optimize the monitoring strategy. On the contrary, if the judgment result is that there is no multi-parameter correlation effect, it means that the environmental parameters are independent of each other, and no significant correlation has been identified. The parameter changes in the monitoring system may occur independently, and there is no need to specially consider the impact of multi-parameter linkage. The system operation is relatively stable.

[0103] The present invention constructs a high-dimensional topological structure of the IDC computer room environment and uses a persistent homology algorithm to accurately capture the potential correlation between multi-dimensional parameters, thereby improving the ability to parse complex environmental data. It is highly scientific and operational, and thus overcomes the limitations of traditional monitoring methods for multi-parameter correlation analysis, which helps to accurately identify system risks in a dynamic environment and ensure the comprehensiveness and stability of the monitoring system. At the same time, combined with persistent barcodes and multi-scale persistence graph analysis methods, it can robustly evaluate the persistence and consistency of parameter associations at different scales, breaking through the shortcomings of single-scale monitoring, and has the advantages of high efficiency and accurate identification. Ultimately, it provides solid theoretical support and technical guarantee for the optimization and decision-making of the IDC computer room environment monitoring system.

[0104] In the environmental monitoring of IDC computer rooms, there are often complex correlations and interdependencies between multidimensional environmental parameters. As environmental conditions change, the correlation structure of these parameters may continue to adjust and evolve. Traditional independent analysis methods are difficult to fully capture these dynamic changes. In order to better understand and predict the interaction between environmental parameters, it is necessary to build a systematic analysis framework that can not only identify the correlation between parameters, but also explore the causal relationship and nonlinear dependency between them. Through the comprehensive application of advanced algorithms such as Bayesian networks, random forest regression, and Granger causality tests, step S3 aims to conduct in-depth correlation analysis of multidimensional environmental parameters in IDC computer rooms, identify potential key factors and causal chains, and thus provide a reliable theoretical basis for subsequent parameter optimization and risk warning.

[0105] Step S3 includes the following contents:

[0106] S3.1, after identifying the existence of multi-parameter correlation effects and completing feature extraction, use the Bayesian network to perform structured modeling on the multi-dimensional environmental parameters of the IDC computer room. The specific steps include, first, using the method of maximizing the posterior probability, using the training data to perform structural learning of the Bayesian network, and constructing a directed acyclic graph structure between the parameters. Each node represents an environmental parameter (such as temperature, humidity, power load, air flow speed, etc.), and each edge represents the conditional dependency relationship between two parameters. By calculating the conditional probability distribution in the Bayesian network, the strength of these dependencies can be quantified, thereby revealing the potential correlation and causal relationship between the parameters. In this process, it is necessary to focus on determining the direct dependency structure between each parameter node to ensure that the network model can truly reflect the interaction between environmental parameters.

[0107] S3.2, based on the Bayesian network structure, random forest regression is further applied to capture the nonlinear relationship between multiple parameters. The specific processing steps are: first, randomly extract multiple sample subsets and feature subsets from the environmental parameter data set, and build multiple decision trees based on the randomly extracted multiple sample subsets and feature subsets. Each decision tree is generated independently to predict the value of the target parameter (such as temperature or humidity). Random forest regression calculates the final prediction result by integrating multiple decision trees, and uses the average or majority voting method to determine the final prediction output. In order to evaluate the influence of each environmental parameter on the target parameter, random forest also outputs the feature importance score of each parameter, which reflects the contribution of the parameter in the prediction. This step can reveal the nonlinear correlation between environmental parameters and quantify the influence of each parameter on the target environmental state.

[0108] S3.3, use the Granger causality test to further detect the temporal causal relationship between multiple parameters. The processing steps include, first, stabilizing the time series data of each environmental parameter to ensure that the data meets the prerequisites of the Granger causality test. Then, build a regression model, take the historical value of a parameter and the current value of other parameters as independent variables, and the current value of the target parameter as the dependent variable, and calculate the regression residual and prediction error. Next, compare the prediction accuracy of the regression model that does not contain the historical value of a parameter with the regression model that contains the historical value of the corresponding parameter. If the historical value of a parameter is introduced, the prediction error of the regression model is greatly reduced, and the prediction accuracy is improved to above the corresponding threshold, indicating that the historical value of the corresponding parameter has a significant influence and causal relationship on the current value of the target parameter. In this way, the Granger causality test can determine the temporal order between the various environmental parameters and their causal influence direction, thereby further deepening the understanding of the parameter association structure.

[0109] Through the conditional dependency modeling of Bayesian networks, the nonlinear correlation analysis of random forest regression, and the causal verification of Granger causality test, the complete correlation structure between the parameters in the IDC computer room environment is obtained, including conditional dependency, nonlinear correlation strength, and time causal order, which can comprehensively and accurately analyze the multi-parameter correlation in the IDC computer room environment. It provides a comprehensive information foundation for understanding and optimizing environmental monitoring, and provides a scientific basis for the precise adjustment of the early warning system and the formulation of optimization strategies.

[0110] The present invention systematically reveals the conditional dependency and potential correlation between the environmental parameters of the IDC computer room by constructing a Bayesian network model, which makes up for the shortcomings of traditional analysis methods in dealing with complex multi-parameter correlations. The application of random forest regression further captures the nonlinear relationship between environmental parameters, quantifies the impact of each parameter on the target state, and makes the analysis results more comprehensive and accurate. In addition, through the Granger causality test, the causal order between parameters is clarified, and the ability to predict changes in environmental parameters is provided. Combining the above methods, step S3 not only improves the analysis depth of multi-parameter correlation, but also provides solid data support and scientific basis for the optimization of the system and the formulation of early warning mechanisms, significantly enhancing the intelligence and precision level of IDC computer room environmental monitoring.

[0111] In the IDC computer room environment monitoring, with the increase of system complexity and data volume, the interdependence and correlation between various environmental parameters are becoming increasingly significant, and the traditional static threshold setting method can no longer effectively cope with complex and changeable environmental changes. The conditional dependency relationship and information transmission intensity between parameters may change dynamically under different environmental conditions, thereby affecting the overall stability and early warning capability of the system. Therefore, it is necessary to introduce a dynamic threshold adjustment method based on Bayesian networks, conditional information volume, and advanced regression models to adapt to the fluctuations of environmental parameters in real time and effectively identify key influencing factors, so as to achieve accurate early warning and timely response to abnormal system conditions. In this context, step S4 is committed to building an intelligent threshold adjustment mechanism to ensure that under the conditions of complex correlation of multiple parameters, the system can quickly respond to environmental changes and ensure the safe and efficient operation of the IDC computer room.

[0112] Step S4 includes the following contents:

[0113] S4.1, based on the conditional dependency determined by the Bayesian network, construct the conditional probability distribution model of each environmental parameter. Let the environmental parameters be A1, A2, …, A q According to the structure of the Bayesian network, the probability distribution P(A) of each parameter given its parent node is calculated. b ∣Pre(A b )), where Pre(A b ) means A b Through these conditional probabilities, the overall joint probability distribution P(A1,A2,…,A q ).

[0114] formula:

[0115] The conditional probability distribution model is used to describe the changing trend of each environmental parameter under the given conditions of other related parameters, providing a probabilistic basis for the dynamic adjustment of subsequent thresholds.

[0116] S4.2, using the conditional information I(A b ; A c ∣Pre(A c ))Measure the information transmission intensity between parameters and identify the key parameters that have the greatest impact on the target parameter. Conditional information is used to evaluate the information dependency between two parameters given other parameters. The parameter with the largest conditional information is selected as the key influencing factor of the target parameter, and the threshold of the parameter with the largest conditional information is dynamically adjusted.

[0117]

[0118] S4.3, after identifying the key parameters, construct a dynamic threshold adjustment function. Use the feature importance coefficient Imp(A b ) and Granger causality test results, set the dynamic threshold function τ(B) of the target parameter B. This function combines the current value a of the key parameter b , feature importance coefficients, and causal order are expressed in the form of nonlinear combinations.

[0119]

[0120] Among them, g(a b ) is a nonlinear transformation function, h(B|Pre(B)) is the predicted value λ of the target parameter under the condition of its previous node b and μ are model coefficients, which determine the weight of the key parameters on the threshold. It represents the number of key influencing parameters, that is, the number of key parameters involved in the calculation when the dynamic threshold of the target parameter B is adjusted.

[0121] S4.4, during real-time monitoring, use the dynamic threshold function to calculate the real-time threshold of each environmental parameter. The real-time data obtained by the monitoring system is input into the dynamic threshold model to calculate the threshold at the current moment. obs If the calculated threshold is exceeded, the alarm mechanism is triggered and emergency processing is carried out according to the preset strategy.

[0122]

[0123] Alarm indicates the triggering condition of the alarm state. Whether to trigger the alarm is determined by whether the actual monitoring data exceeds the dynamic threshold.

[0124] Activate means triggering an alarm, that is, when the monitoring data exceeds the dynamic threshold, the system enters the alarm state and executes corresponding emergency measures.

[0125] Standby indicates the standby state, that is, when the monitoring data is within the dynamic threshold range, the system maintains normal operation and does not trigger an alarm.

[0126] The present invention can accurately capture the dependencies between various environmental parameters by constructing a conditional probability distribution model based on a Bayesian network, and use the conditional information to identify the key factors that have the greatest impact on the target parameters, thereby improving the accuracy of threshold adjustment. The dynamic threshold function set in combination with random forest regression and Granger causality test can flexibly adapt to environmental changes in real-time monitoring, making the system warning more sensitive and reliable. In addition, the real-time calculation and adjustment of dynamic thresholds ensure that in a complex and changeable environment, the system can quickly identify anomalies and trigger alarms, reduce the possibility of false alarms and missed alarms, and ensure the operational safety and stability of the IDC computer room.

[0127] Figure 2 The present invention provides an IDC computer room environment intelligent monitoring system, comprising: a multi-dimensional data acquisition module, a topology structure analysis module, a causal association model module, and a dynamic threshold control module;

[0128] The multi-dimensional data acquisition module acquires the multi-dimensional environmental data of the IDC room, performs outlier detection, missing value filling and data smoothing, extracts features through principal component analysis and autoencoders, converts high-dimensional data into low-dimensional representative features, generates a low-dimensional environmental feature data matrix, and transmits it to the topological structure analysis module;

[0129] The topological structure analysis module constructs a high-dimensional topological structure based on the extracted multi-dimensional environmental data, applies the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence graph and spectral clustering results, judges and identifies the significant correlation effects between multiple parameters, generates the identification results of the multi-parameter correlation effects and their topological structure information, and transmits them to the causal association model module;

[0130] After identifying the multi-parameter correlation effect, the causal association model module uses the Bayesian network to build a parameter dependency model, reveals potential correlations through conditional probability distribution, uses random forest regression to capture nonlinear relationships and quantify the impact, and uses the Granger causality test to clarify the causal order between parameters, improves parameter correlation analysis, generates the dependency structure, nonlinear relationship and causal chain of key parameters, and transmits them to the dynamic threshold control module;

[0131] The dynamic threshold control module is based on the conditional probability distribution model of the Bayesian network. It uses conditional information to identify key parameters, combines feature importance and causal relationships to build a dynamic threshold function, and monitors and triggers the alarm mechanism in real time.

[0132] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0133] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0134] It should be noted that, in this article, if there are relational terms such as first and second, it is only used to distinguish one entity or operation from another entity or operation, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. The above is only a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field of the present application can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the scope of protection of the claims.

Claims

1. An IDC room environment intelligent monitoring method, characterized in that: Includes steps: Step S1, obtaining multi-dimensional environmental data of the IDC room, performing outlier detection, missing value filling and data smoothing, extracting features through principal component analysis and autoencoder, and converting high-dimensional data into low-dimensional representative features; Step S2, based on the extracted multidimensional environmental data, construct a high-dimensional topological structure, apply the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence map and spectrum clustering results, and judge and identify the significant correlation effects between multiple parameters; Step S3, after identifying the multi-parameter correlation effect, the parameter dependency model is constructed using the Bayesian network, the potential correlation is revealed through the conditional probability distribution, the nonlinear relationship is captured and the impact is quantified using random forest regression, the causal order between parameters is clarified through the Granger causality test, and the parameter correlation analysis is improved; Step S4, based on the conditional probability distribution model of the Bayesian network, the conditional information volume is used to identify key parameters, and a dynamic threshold function is constructed by combining feature importance and causal relationship to monitor in real time and trigger an alarm mechanism.

2. The method for intelligently monitoring an IDC computer room environment according to claim 1, characterized in that: Step S2 includes the following contents: S2.1, embed the multi-dimensional environmental parameter data of the IDC room into a high-dimensional space to form a point cloud data set. Assume that the point cloud data set is presented in the form of a time series, and let the data vector at each time point be Where t is a moment in the time series, n is the parameter dimension, and the data vectors of all time points form a point cloud Where T is the total time points; S2.2, on the generated point cloud, apply the Vietoris-Rips complex algorithm to construct a high-dimensional topological complex, set the distance threshold ∈, for any two points x in the point cloud i and x j , if their Euclidean distance is less than or equal to, then connect an edge between these two points; S2.3, based on the constructed Vietoris-Rips complex, the homology group H is calculated using the persistent homology algorithm m The persistence of m represents the dimension of the homology group, and we focus on the low-dimensional homology groups of connected components H0, loops H, and cavities H2. By calculating the persistence of the homology group at different distance thresholds, we get the persistent barcode, where the length of the barcode represents the persistence of the corresponding topological feature at different scales. official: in, Represents the boundary operator of the m-dimensional chain; S2.4, convert the persistence barcode into a multi-scale persistence map, and form a multi-scale persistence map by calculating the persistence at different distance threshold scales; perform multi-scale analysis on all persistence maps, and identify significant association patterns between multiple parameters through decomposition and feature extraction; official: Persistence graph, a visual representation of persistent homology at different distance threshold scales; Multi-scale persistence graph, through persistence analysis at different scales, shows the topological associations between multiple parameters.

3. The method for intelligently monitoring an IDC computer room environment according to claim 2, characterized in that: S2.5, after completing steps S2.1-2.4, determine whether there is a significant multi-parameter association effect by analyzing the multi-scale persistence map; The length of the persistent barcode is obtained by calculating the difference between the generation and disappearance of the topological feature under different scale distance thresholds; specifically, the generation scale of the topological feature is recorded when it first appears, and the disappearance scale is recorded when the feature disappears. The length of the persistent barcode is the difference between the disappearance scale and the generation scale of the feature. The similarity metric of multi-scale persistence graphs is calculated by comparing the structural features of persistence graphs at different scale distance thresholds. At each scale distance threshold, the persistence graph is transformed into a graph structure, where nodes represent topological features and edges represent the associations between features. Then, the Pearson correlation coefficient metric is used to evaluate the structural similarity between persistence graphs at different scale distance thresholds. The cohesion of the spectral graph clustering results is determined by measuring the closeness of the nodes within the cluster; first, the standardized graph Laplacian matrix of the parameter association graph is constructed, and the eigenvalue decomposition is performed to decompose the matrix into eigenvectors and eigenvalues. The first rated number of eigenvectors are used as the new feature space for clustering. For each cluster, its internal interconnectivity is calculated, and the average value of the internal interconnectivity of all clusters is taken as the cohesion of the clustering results.

4. The method for intelligently monitoring an IDC computer room environment according to claim 3, characterized in that: The final judgment criteria are expressed by the following formula: Among them, Result is the judgment result; Significant Association means that there is a multi-parameter association effect; NoSignificant Association means that there is no multi-parameter association effect; L β is the length of the persistent barcode, and the persistence threshold δ p Make comparisons; S γ is the similarity measure of the multi-scale persistence graph, and the similarity threshold δ s Make comparisons; C θ is the cohesion of the spectral clustering result, and the cohesion threshold δ c Make a comparison.

5. The method for intelligently monitoring an IDC computer room environment according to claim 4, characterized in that: Step S3 includes the following contents: S3.1, after identifying the existence of multi-parameter correlation effects and completing feature extraction, use the Bayesian network to perform structured modeling on the multi-dimensional environmental parameters of the IDC computer room; the specific steps include, firstly, using the method of maximizing the posterior probability, using the training data to perform structural learning of the Bayesian network, and constructing a directed acyclic graph structure between the parameters, where each node represents an environmental parameter and each edge represents a conditional dependency relationship between two parameters; S3.2, based on the Bayesian network structure, random forest regression is further applied to capture the nonlinear relationship between multiple parameters. The specific processing steps are as follows: first, multiple sample subsets and feature subsets are randomly extracted from the environmental parameter data set, and multiple decision trees are constructed based on the randomly extracted multiple sample subsets and feature subsets. Each decision tree is independently generated to predict the value of the target parameter; S3.3, use the Granger causality test to further detect the temporal causal relationship between multiple parameters. The processing steps include: first, smoothing the time series data of each environmental parameter to ensure that the data meets the prerequisites of the Granger causality test; then, building a regression model, taking the historical value of a parameter and the current value of other parameters as independent variables, and the current value of the target parameter as the dependent variable, calculating the regression residual and prediction error; then, comparing the prediction accuracy of the regression model that does not contain the historical value of a parameter with the regression model that contains the historical value of the corresponding parameter; if the prediction accuracy is improved to above the corresponding threshold after introducing the historical value of a parameter, it indicates that the historical value of the corresponding parameter has a significant influence and causal relationship on the current value of the target parameter.

6. The method for intelligently monitoring an IDC computer room environment according to claim 5, characterized in that: Step S4 includes the following contents: S4.1, based on the conditional dependency determined by the Bayesian network, construct the conditional probability distribution model of each environmental parameter, assuming that the environmental parameters are A1, A2, …, A q According to the structure of the Bayesian network, the probability distribution P(A) of each parameter given its parent node is calculated. b ∣Pre(A b )), where Pre(A b ) means A b The parent node set of A1, A2, ..., A q ); S4.2, using the conditional information I(A b ; A c ∣Pre(A c ))Measure the information transmission intensity between parameters and identify the key parameters that have the greatest impact on the target parameter. The conditional information is used to evaluate the information dependency between two parameters given other parameters. The parameter with the largest conditional information is selected as the key influencing factor of the target parameter, and the threshold of the parameter with the largest conditional information is dynamically adjusted. S4.3, after identifying the key parameters, a dynamic threshold adjustment function is constructed, using the feature importance coefficient Imp(A b ) and Granger causality test results, set the dynamic threshold function τ(B) of the target parameter B; Among them, g(a b ) is a nonlinear transformation function, h(B|Pre(B)) is the predicted value λ of the target parameter under the condition of its previous node b and μ are model coefficients, which determine the weight of the key parameters on the threshold. It represents the number of key influencing parameters, that is, the number of key parameters involved in the calculation when the dynamic threshold of the target parameter B is adjusted.

7. The method for intelligently monitoring an IDC computer room environment according to claim 6, characterized in that: S4.4, in the real-time monitoring process, the dynamic threshold function is used to calculate the real-time threshold of each environmental parameter. The real-time data obtained by the monitoring system is input into the dynamic threshold model to calculate the threshold at the current moment. If the actual monitoring value B obs If the calculated threshold is exceeded, the alarm mechanism is triggered and emergency processing is carried out according to the preset strategy; Alarm indicates the triggering condition of the alarm status. Whether to trigger the alarm is determined by whether the actual monitoring data exceeds the dynamic threshold. Activate means triggering an alarm, that is, when the monitoring data exceeds the dynamic threshold, the system enters the alarm state and executes the corresponding emergency treatment measures; Standby indicates the standby state, that is, when the monitoring data is within the dynamic threshold range, the system maintains normal operation and does not trigger an alarm.

8. An IDC computer room environment intelligent monitoring system, used to implement an IDC computer room environment intelligent monitoring method according to any one of claims 1 to 7, characterized in that: include: Multidimensional data acquisition module, topological structure analysis module, causal association model module, dynamic threshold control module; The multi-dimensional data acquisition module acquires the multi-dimensional environmental data of the IDC room, performs outlier detection, missing value filling and data smoothing, extracts features through principal component analysis and autoencoders, converts high-dimensional data into low-dimensional representative features, generates a low-dimensional environmental feature data matrix, and transmits it to the topological structure analysis module; The topological structure analysis module constructs a high-dimensional topological structure based on the extracted multi-dimensional environmental data, applies the persistent homology algorithm to calculate and analyze the persistent barcode, multi-scale persistence graph and spectral clustering results, judges and identifies the significant correlation effects between multiple parameters, generates the identification results of the multi-parameter correlation effects and their topological structure information, and transmits them to the causal association model module; After identifying the multi-parameter correlation effect, the causal association model module uses the Bayesian network to build a parameter dependency model, reveals potential correlations through conditional probability distribution, uses random forest regression to capture nonlinear relationships and quantify the impact, and uses the Granger causality test to clarify the causal order between parameters, improves parameter correlation analysis, generates the dependency structure, nonlinear relationship and causal chain of key parameters, and transmits them to the dynamic threshold control module; The dynamic threshold control module is based on the conditional probability distribution model of the Bayesian network. It uses conditional information to identify key parameters, combines feature importance and causal relationships to build a dynamic threshold function, and monitors and triggers the alarm mechanism in real time.

Images