IDC machine room operation and maintenance visual monitoring method and system

Abstract

The application is suitable for the technical field of machine room operation and maintenance monitoring, and particularly relates to an IDC machine room operation and maintenance visual monitoring method and system, which comprises the following steps: counting the business attributes and physical topology relationship of machine room nodes, constructing a machine room node topology network, and performing mutual detection based on the machine room node topology network; constructing an anomaly detection model and training the same; collecting the environment data and operation state data of each machine room, and constructing a change distribution prediction curve; extracting prediction parameters based on the change distribution prediction curve, integrating the prediction parameters, introducing the same into the optimized anomaly detection model, performing visual processing, and outputting a visual picture. The application generates a semi-transparent color temperature layer with a risk mark based on an anomaly propagation heat map as a visual base map, realizes seamless, accurate and intuitive positioning from visual early warning information to specific fault equipment nodes, and greatly shortens the time for operation and maintenance personnel to diagnose the root cause and respond to anomalies.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer room operation and maintenance monitoring, and in particular relates to an IDC computer room operation and maintenance visualization monitoring method and system. Background Art

[0002] Visualized monitoring of computer room operations and maintenance is a management method that uses a graphical interface to display the operating status and environmental parameters of various equipment within the computer room. It integrates real-time data collection, processing, and analysis, visually displaying the performance indicators of hardware facilities such as servers, network equipment, and storage systems, as well as environmental conditions such as temperature and humidity. This monitoring method not only improves troubleshooting efficiency but also helps administrators quickly understand the overall health of the system, enabling effective resource planning and risk management. Through visualization tools, comprehensive insights into computer room operations are achieved, supporting decision-making and ensuring business continuity and stability.

[0003] In the existing computer room operation and maintenance process, the computer room operation and maintenance data is collected and used as the basis for judging whether the computer room operation is normal. However, it is difficult to predict the operating status of the computer room, and abnormal operation of the computer room is prone to occur. Summary of the Invention

[0004] The purpose of the present invention is to provide an IDC computer room operation and maintenance visualization monitoring method and system, aiming to solve the problem that in the existing computer room operation and maintenance process, by collecting computer room operation and maintenance data and using the computer room operation and maintenance data as the basis for judging whether the computer room operation is normal, it is difficult to predict the computer room operation status and is prone to abnormal operation of the computer room.

[0005] The present invention is implemented as follows: a method for visually monitoring IDC computer room operation and maintenance, the method comprising:

[0006] Collect statistics on the business attributes and physical topology relationships of computer room nodes, build a computer room node topology network, and conduct mutual detection based on the computer room node topology network;

[0007] Obtain historical operating data from each computer room, build an anomaly detection model, and train it based on the historical operating data of the computer room to obtain an optimized anomaly detection model;

[0008] Sensors are used to collect environmental data and operating status data from each computer room, obtain multiple evaluation parameters, and construct a change distribution prediction curve for each evaluation parameter;

[0009] The prediction parameters are extracted based on the change distribution prediction curve, integrated, and imported into the optimized anomaly detection model to output the detection results. The evaluation parameters and detection results are visualized and a visualization screen is output.

[0010] Another object of the present invention is to provide an IDC computer room operation and maintenance visual monitoring system, the system comprising:

[0011] The data detection module is used to collect statistics on the business attributes and physical topology relationships of the computer room nodes, build a computer room node topology network, and perform mutual detection based on the computer room node topology network;

[0012] The model training module is used to obtain the historical operation data of each computer room, build an anomaly detection model, and train it based on the historical operation data of the computer room to obtain an optimized anomaly detection model;

[0013] The parameter prediction module is used to collect environmental data and operating status data of each computer room through sensors, obtain multiple evaluation parameters, and construct a change distribution prediction curve for each evaluation parameter;

[0014] The prediction and detection module is used to extract prediction parameters based on the change distribution prediction curve, integrate the prediction parameters, import them into the optimized anomaly detection model to output the detection results, visualize the evaluation parameters and detection results, and output the visualization picture.

[0015] The beneficial effects of the present invention are:

[0016] 1. This paper constructs an anomaly detection model input layer consisting of a primary channel (a fully connected layer processes temperature and humidity) and a secondary channel (a convolutional layer processes CPU and network latency). It also uses topological constraints to force adjacent nodes to meet reconstruction error standards. This, combined with backtracking to verify the model's sensitivity through historical anomaly event fragments, significantly improves the model's feature extraction capabilities and anomaly detection accuracy for multi-dimensional heterogeneous operating data in computer rooms (environmental parameters and equipment indicators), effectively addressing the problem of misjudgment caused by differences in data dimensions and insufficient utilization of physical topological correlations.

[0017] 2. The present invention assigns topological weights to the parameter coordinates of core network nodes and edge nodes, and performs drift correction based on real-time environmental data (such as the temperature change rate exceeding a threshold increases the density of fitting points, and humidity mutation triggers segmented reconstruction). It generates a prediction curve that forcibly associates the temperature gradient difference between adjacent cabinets with the synchronization rate of equipment delay fluctuations. This ensures that parameter change predictions strictly follow the physical topology of the computer room and the interaction rules between the equipment, greatly overcoming the risk of prediction distortion caused by pure mathematical fitting methods being detached from actual physical constraints.

[0018] 3. Based on the abnormal propagation heat map, the present invention generates a semi-transparent color temperature layer with risk markers as a visualization base map, implements curve rendering with dual-channel visual coding of line type (indicating credibility level) and line width (indicating rating strength), and establishes a linkage mechanism in which clicking on a predicted curve segment triggers the highlighting and flashing of the associated physical device in the three-dimensional topology map. Historical abnormal pulse markers are superimposed for spatiotemporal comparison, achieving seamless, accurate, and intuitive positioning from visual warning information to specific faulty equipment nodes, greatly shortening the time it takes for operation and maintenance personnel to diagnose the root cause and respond to abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of a method for visually monitoring IDC computer room operation and maintenance provided by an embodiment of the present invention;

[0020] Figure 2 An architectural diagram of an IDC computer room operation and maintenance visualization monitoring system provided by an embodiment of the present invention;

[0021] Figure 3 An architectural diagram of a model training module provided in an embodiment of the present invention;

[0022] Figure 4 An architectural diagram of a parameter prediction module provided by an embodiment of the present invention;

[0023] Figure 5 This is an architectural diagram of a prediction and detection module provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] like Figure 1 FIG. 1 is a flowchart of a method for visually monitoring IDC computer room operation and maintenance provided by an embodiment of the present invention, the method comprising:

[0026] S100: Count the service attributes and physical topology relationships of the computer room nodes, build a computer room node topology network, and perform mutual detection based on the computer room node topology network.

[0027] In this step, the business attributes and physical topology relationships of the computer room nodes are counted. Different computer rooms have different business types and their data types are also different. Multiple sensors are set up in the computer room to monitor the environmental data in the computer room, such as temperature sensors (cabinets, air conditioning outlet / return vents and rooms), humidity sensors, airflow sensors, water immersion sensors, smoke sensors, access control sensors, noise sensors and vibration sensors. The operating parameters of the equipment in the computer room include the operating indicators of servers / network devices / storage devices (CPU utilization, memory utilization, disk I / O, disk space, network traffic, network card error rate, power status and fan speed), UPS status (input / output voltage and current, battery status, load rate), precision air conditioning operating parameters (set temperature, supply air temperature, return air temperature, compressor status, fan speed, refrigerant pressure) and PDU current, voltage and power, etc., which are reported by the equipment itself. At the same time, two business nodes are randomly selected based on the computer room node topology network to detect each other, perform intercommunication and mutual PING detection, and control all business nodes to send ICMP detection packets to each other, so as to obtain delay parameters and packet loss parameters between computer room nodes.

[0028] S200, obtaining historical operation data of each computer room, building an anomaly detection model, and training the model based on the historical operation data of the computer room to obtain an optimized anomaly detection model.

[0029] In this step, the historical operation data of each computer room is obtained. During the historical operation of the computer room, the operation data of the computer room is monitored, including the environmental parameters and operation parameters of the computer room. A data set is constructed by manual labeling, and an anomaly detection model is constructed. The anomaly detection model includes an input layer, an encoder, a decoder, a loss function and an anomaly judgment layer, wherein the input layer is used to receive prediction parameters, the encoder is used to compress high-dimensional input data into a low-dimensional latent space, which is composed of multiple fully connected layers, the encoder outputs a low-dimensional latent vector, and the decoder outputs reconstructed data based on the low-dimensional latent vector. The loss function is used to evaluate the difference between the original data and the reconstructed data. The anomaly judgment layer is used to perform anomaly judgment after the model training is completed. The anomaly detection model is trained and verified through the data set, so that the anomaly detection model is used to perform anomaly judgment on the collected data.

[0030] S300 , collecting environmental data and operating status data of each computer room through sensors, obtaining multiple evaluation parameters, and constructing a change distribution prediction curve for each evaluation parameter.

[0031] In this step, the environmental data and operating status data of each computer room are collected through sensors. The environmental data and operating status data collected at this time are real-time data. Data collection is performed at preset time intervals, and each parameter is used as an evaluation parameter, such as temperature, humidity, delay, and packet loss rate. The collected continuous data is marked in a two-dimensional coordinate system, and a prediction function for each evaluation parameter is obtained by function fitting. The change distribution prediction curve of the corresponding evaluation parameter is generated by the prediction function. The change distribution prediction curve records the predicted data of the parameter in a short period of time.

[0032] S400, extracting prediction parameters based on the change distribution prediction curve, integrating the prediction parameters, importing them into the optimized anomaly detection model to output the detection results, visualizing the evaluation parameters and the detection results, and outputting a visualization screen.

[0033] In this step, prediction parameters are extracted based on the change distribution prediction curve, the sampling point time value is determined according to the required prediction time, the corresponding change distribution prediction curve is queried based on the sampling point time value, and the prediction parameters of the corresponding time are obtained. The prediction parameters at the same time are merged together and imported into the optimized anomaly detection model as a set of data to be predicted, so as to output the detection results, and the change distribution prediction curve of each evaluation parameter is visualized. The change distribution prediction curve is controlled based on the detection results, such as changing its color according to whether the parameter is normal or not. When the temperature exceeds the preset value, the corresponding warning color is displayed and an alarm is issued.

[0034] As a preferred embodiment of the present invention, the steps of obtaining historical operation data of each computer room, building an anomaly detection model, and training the model based on the historical operation data of the computer room specifically include:

[0035] S201, obtaining historical operation data of the computer room, where the historical operation data of the computer room includes environmental data and equipment operation index data.

[0036] In this step, the historical operation data of the computer room is obtained, and the historical operation data within the corresponding time length is retrieved according to the amount of data required for training. The historical operation data is divided according to the preset time intervals, and each moment corresponds to a set of historical operation data.

[0037] S202: Clean the historical operation data, fill in missing values, remove outliers, perform timestamp alignment and normalization, and construct a data set. The data set is divided into a training data set and a validation data set.

[0038] S203, constructing an anomaly detection model. The anomaly detection model uses an unsupervised learning autoencoder and is trained and verified using a data set.

[0039] In this step, the historical operation data is cleaned. Missing values, duplicate values ​​and outliers in the historical operation data are identified by data cleaning. Forward filling or backward filling strategies are used to supplement the missing values, and outliers and duplicate values ​​are removed. Timestamp alignment is used to solve the problem of inconsistent timestamps of data points from different devices or sensors, and they are unified to the same time point or time period to facilitate association analysis. Normalization is used to scale indicators of different dimensions and orders of magnitude to a unified numerical range, eliminating dimensional effects, accelerating model convergence, and improving model accuracy. Manual labeling is performed to construct corresponding training data sets and verification data sets, and the anomaly detection model is trained and verified using the training data sets and verification data sets until the anomaly detection success rate of the anomaly detection model reaches the preset value.

[0040] It should be noted that the anomaly detection model includes an input layer, an encoder, a decoder, a loss function and an anomaly judgment layer, wherein the input layer is used to receive prediction parameters, the encoder is used to compress high-dimensional input data into a low-dimensional latent space, which is composed of multiple fully connected layers. The encoder outputs a low-dimensional latent vector, and the decoder outputs reconstructed data based on the low-dimensional latent vector. The loss function is used to evaluate the difference between the original data and the reconstructed data. The anomaly judgment layer is used to perform anomaly judgment after the model training is completed; the input layer is divided into a main channel and a sub-channel. The main channel is processed by the fully connected layer to process temperature / humidity, and the sub-channel is processed by the convolutional layer to process CPU / network delay.

[0041] As a preferred embodiment of the present invention, the step of training and validating the anomaly detection model using the data set specifically includes:

[0042] S20321, configure the input layer. The input layer includes a main channel and a sub-channel. The main channel is composed of a fully connected layer, and the sub-channel is composed of a convolutional layer.

[0043] Among them, the main channel is used to process environmental parameter time series data, and the secondary channel is used to process equipment indicator time series data;

[0044] S20322, associating the historical operating data of the computer room according to the topological node position, mapping the environmental parameters to the primary channel to form primary channel time series data, and mapping the equipment indicators to the secondary channel to form secondary channel time series data; combining the primary channel time series data and the secondary channel time series data with the training data set to generate a training matrix; wherein the vertical dimension of the training matrix corresponds to the node attribute, and the horizontal dimension corresponds to the time dimension;

[0045] S20323, using an unsupervised learning mechanism, randomly split the samples in the training matrix into segments A and B. Using the decoder of the anomaly detection model, reconstruct the feature distribution of segment B based on segment A, and calculate the reconstruction error between the original feature distribution of segment B and the reconstructed feature distribution of segment B using a loss function. During training, topological constraints are imposed to minimize the difference in reconstruction errors between adjacent nodes, thereby obtaining a preliminarily trained anomaly detection model.

[0046] S20324: Randomly block the operating status data of some nodes on the validation set and use the preliminarily trained anomaly detection model to reconstruct the status of the blocked nodes. Calculate the reconstruction accuracy based on the delay indicator of mutual detection between nodes. Dynamically adjust the verification threshold based on the reconstruction accuracy to establish a topology-aware verification rule base.

[0047] S20325: Based on the topology-aware verification rule base, the verification set is classified by business attributes. Based on the classification type, the anomaly judgment layer of the anomaly detection model is used to calculate the reconstruction error of key device indicators in various business scenarios, and the delay reconstruction accuracy is verified to generate a scenario verification report.

[0048] S20326: Extract data segments from periods of known historical abnormal events and input them into the preliminarily trained anomaly detection model. Generate a latent space feature representation through the encoder. Use the anomaly judgment layer to capture feature mutation points in the latent space feature representation and calculate the matching delay between them and the actual anomaly timestamp to obtain a retrospective detection result.

[0049] S20327, based on the scenario verification report and the retrospective detection results, optimize the anomaly detection model after preliminary training, enhance the encoder's feature extraction capability for core network nodes, increase the decoder's reconstruction weight for high business value nodes, and update the anomaly detection model parameters; after the anomaly detection model parameters are updated, use the topology-aware verification rule library to test the topology reconstruction capability, and use the business scenario verification set to evaluate the special performance, and combine the retrospective detection results to verify the historical anomaly sensitivity, and finally obtain the optimized anomaly detection model.

[0050] In this embodiment, the present invention constructs a dual-channel anomaly detection model (the main channel fully connected layer processes temperature and humidity data, and the secondary channel convolutional layer processes equipment performance indicators) and implants a topological constraint mechanism to force the reconstruction error difference of adjacent nodes to meet the standard. Combined with the historical abnormal event fragments to backtrack and verify the model sensitivity, the present invention solves the false alarm and missed alarm problems caused by traditional detection methods that ignore the physical correlation of equipment and the heterogeneous characteristics of environment-equipment indicators.

[0051] As a preferred embodiment of the present invention, the step of collecting environmental data and operating status data of each computer room through sensors to obtain multiple evaluation parameters and constructing a change distribution prediction curve for each evaluation parameter specifically includes:

[0052] S301, regularly acquiring detection data from sensors in each computer room and operating status data of the computer room through a data collector to obtain multiple evaluation parameters.

[0053] In this step, by regularly accessing each sensor, the corresponding environmental data and the operating status data of the computer room can be obtained. The above data is collected in real time, and the detection values ​​of multiple evaluation parameters are obtained. For example, at time A, the temperature of the computer room is T1, the humidity is W1, and the power of the computer room is P1. Each evaluation parameter corresponds to a value.

[0054] S302 , constructing a two-dimensional coordinate system, marking the values ​​of the same evaluation parameter at different times in the two-dimensional coordinate system in chronological order, and obtaining a plurality of parameter coordinates.

[0055] In this step, a two-dimensional coordinate system is constructed, in which the horizontal axis is the time value, the numerical values ​​of the evaluation parameters are normalized, and then marked in the two-dimensional coordinate system, thereby forming multiple coordinate points in the two-dimensional coordinate system.

[0056] S303 : determining a corresponding fitting function based on the multiple parameter coordinates, thereby converting the coordinate points in the two-dimensional coordinate system into a change distribution prediction curve.

[0057] As a preferred embodiment of the present invention, the step of determining a corresponding fitting function based on multiple parameter coordinates to convert coordinate points in a two-dimensional coordinate system into a change distribution prediction curve specifically includes:

[0058] S3031, obtaining the physical topology relationship of the computer room nodes, and assigning weights to the core network node coordinates and the parameter coordinates of the edge nodes, to generate a parameter coordinate set with topological weights;

[0059] S3032: Based on the parameter coordinate set with topological weights, the fluctuation pattern of similar topological structures in the historical operation data of the computer room is retrieved, and the morphological similarity between the current coordinate trend and the historical data of the same period is compared to obtain a selected confidence historical reference curve template;

[0060] S3033: Automatically select a fitting strategy based on the selected confidence historical reference curve template to generate a fitting strategy that is suitable for the current business scenario. The automatically selected fitting strategy includes: linear fitting for steady-state fluctuations, spline fitting for oscillatory fluctuations, and segmented fitting for step fluctuations.

[0061] S3034 , based on the service attributes and the fitting strategy adapted to the current service scenario, calculate the node CPU parameter fitting error tolerance and the node disk parameter tolerance according to the service type to obtain a service-aware fitting parameter set;

[0062] S3035: Acquire real-time environmental data, apply a service-aware fitting parameter set, and use the real-time environmental data as a correction factor. When the temperature change rate exceeds a preset temperature change rate threshold, the fitting point density is automatically increased. When the humidity suddenly changes and exceeds a preset humidity threshold, segmented reconstruction is triggered, ultimately generating a drift-corrected dynamic function.

[0063] S3036, based on the drift-corrected dynamic function, enforces topology constraints and associates the temperature difference between adjacent cabinets with the device delay fluctuation synchronization rate to obtain a prediction curve that conforms to physical laws;

[0064] S3037, obtain real-time data, superimpose the prediction curve that conforms to the physical laws with the real-time data and display it. When the deviation between the predicted value of the prediction curve that conforms to the physical laws and the actual value of the real-time data continues to exceed the preset deviation threshold, the abnormal area is automatically marked, and the screened confidence historical reference curve template is retroactively corrected to obtain the corrected historical reference curve template; and prediction is performed based on the corrected historical reference curve template to obtain the final prediction curve that conforms to the physical laws, and the final prediction curve that conforms to the physical laws is used as the change distribution prediction curve.

[0065] In this embodiment, the present invention assigns physical weights to topological node coordinates, retrieves historical fluctuation patterns to screen high-confidence reference curves, dynamically selects fitting strategies (steady-state linear, oscillating spline, step segmentation) that are adapted to business scenarios, and introduces real-time environmental data as a drift correction factor (temperature change rate triggers fitting point encryption, humidity mutation drives segment reconstruction), forcibly associating cabinet temperature gradient differences with device delay fluctuation synchronization rates, thereby overcoming the prediction distortion defect caused by mathematical fitting deviating from physical laws.

[0066] As a preferred embodiment of the present invention, the steps of extracting prediction parameters based on the change distribution prediction curve, integrating the prediction parameters, importing them into the optimized anomaly detection model, outputting the detection results, visualizing the evaluation parameters and the detection results, and outputting the visualization screen specifically include:

[0067] S401 , extracting prediction parameters from each change distribution prediction curve according to a preset extraction interval, and taking the prediction parameters extracted at the same time as a group of data to be predicted.

[0068] In this step, the predicted parameters are extracted from each change distribution prediction curve according to a preset extraction interval, for example, 10 seconds as a time interval, the predicted parameters corresponding to each time point in the next 10 minutes are extracted, and the predicted parameters at the same time are divided into one to-be-predicted data.

[0069] S402, the to-be-predicted data is input into the optimized anomaly detection model, the to-be-predicted data corresponding to each time in the future is analyzed by the optimized anomaly detection model, and a corresponding detection result is generated.

[0070] S403, the change distribution prediction curve corresponding to the evaluation parameter is visually output, and the style of the output change distribution prediction curve is controlled according to the detection result, and a visual picture is output.

[0071] As a preferred embodiment of the present application, the step of inputting the to-be-predicted data into the optimized anomaly detection model, analyzing the to-be-predicted data corresponding to each time in the future by the optimized anomaly detection model, and generating a corresponding detection result, specifically includes:

[0072] S4021, the physical topology relationship of the machine room node is obtained, and the to-be-predicted data is divided according to the topology group, the core network node group is assigned a high-priority processing queue, the edge node group is assigned a regular queue, and a to-be-predicted data set of the topology group is generated;

[0073] S4022, according to the to-be-predicted data set of the topology group, the abnormal mode of the similar topology group in the historical operation data of the machine room is searched, the similarity of the precursor characteristics of the current to-be-predicted data and the historical abnormal events is compared, and the matching confidence score of each group is obtained;

[0074] S4023, based on the matching confidence score of each group, a dynamic weight is assigned to each layer of the anomaly detection model, the feature extraction strength of the confidence corresponding group of the input layer is increased, the judgment threshold of the abnormal judgment layer for the low confidence group is reduced, and a topology-aware model weight configuration is generated;

[0075] S4024, the topology-aware model weight configuration is applied to construct a space-time correlation matrix in the encoder latent space, longitudinally correlate the historical states of the nodes in the same group, and laterally correlate the real-time interactions of the devices in the current group, and form a feature matrix with space-time weights.

[0076] S4025, based on the feature matrix with space-time weights, in combination with the topology network relationship, when a node detects a potential anomaly, the influence probability of the associated devices is predicted along the physical connection path, and the risk coefficient of the abnormal propagation of the core switch to the server group is calculated, to output an abnormal propagation heat map.

[0077] S4026, performing three-level verification by the anomaly judgment layer according to the anomaly propagation heat map, and generating a detection result with a confidence rating;

[0078] The three-level verification includes single-device reconstruction error verification, intra-group node consistency verification, and historical pattern backtracking verification.

[0079] S4027, obtaining a real-time monitoring picture, superimposing the detection result with the confidence rating on the real-time monitoring picture, allowing an operation and maintenance personnel to confirm and feed back the high-confidence anomaly, and automatically correcting the historical matching confidence rating and the model weight configuration of the topology awareness of each group to obtain a final detection result.

[0080] In the embodiment, the application calculates the group matching confidence rating by dividing the to-be-predicted data (core node high-priority queue and edge node regular queue) according to the topology group, dynamically allocates the model layer weight (enhances high-confidence group feature extraction in the input layer and reduces the low-confidence group determination threshold in the anomaly judgment layer), constructs a space-time correlation matrix (vertically correlates the historical state and horizontally correlates the real-time interaction of devices), outputs the anomaly propagation heat map, and performs three-level verification (single-device reconstruction, group consistency, and historical pattern backtracking) combined with operation and maintenance feedback to correct the confidence rating, thereby solving the problems of response lag of the traditional detection method to key nodes and insufficient group correlation anomaly recognition.

[0081] As a preferred embodiment of the application, the step of visualizing the change distribution prediction curve corresponding to the evaluation parameter and controlling the style of the output change distribution prediction curve according to the detection result to output a visualized picture specifically includes:

[0082] S4031, extracting the core node coordinates of the high-risk area based on the anomaly propagation heat map, marking the prediction curve segment corresponding to the evaluation parameter as a red highlighted focus area in the visualized picture, and generating a prediction curve framework with a risk marker;

[0083] S4032, converting the anomaly propagation heat map into a semi-transparent color temperature layer using the prediction curve framework with the risk marker, and superimposing the semi-transparent color temperature layer on the prediction curve background area to form a visualized base map with heat background fusion; wherein the semi-transparent color temperature layer includes a red high-risk area and a light yellow low-risk area.

[0084] S4033, implementing double-channel visual coding on the visualized base map with heat background fusion based on the matching confidence rating of each group to generate a dynamically adaptive curve rendering scheme; wherein the double-channel visual coding includes a line type channel and a line width channel, the line type channel represents the confidence rating level, and the line width channel represents the rating intensity.

[0085] S4034: Apply a dynamic and adaptive curve rendering solution and establish a click response mechanism. When a user selects a specific predicted curve segment, the corresponding physical device node is triggered to highlight and flash in the 3D topology map, thereby obtaining a visual linkage result between the predicted data and the physical device.

[0086] S4035: Based on the visual linkage results between the prediction data and the physical device, the modified historical reference curve template is called, a semi-transparent historical trajectory is superimposed below the current prediction curve, and the time deviation between the historical anomaly point and the current prediction point is marked with a pulse marker to construct a spatiotemporal comparative analysis layer;

[0087] S4036: Threshold alarms are generated based on the spatiotemporal comparative analysis layer, triggering a comprehensive alarm response. The threshold alarms include: when the predicted value continuously exceeds a preset deviation threshold, the curve shape is switched to a sawtooth shape, a verification pop-up window is displayed, and an audible and visual alarm device is activated.

[0088] S4037,captures the record of the comprehensive alarm response.,When the operation and maintenance personnel manually adjust the alarm style,,the screening strategy of the confidence history reference curve template and the,calculation parameters of the matching confidence score are automatically,reverse-optimized to finally obtain the final visualization picture that has been,manually calibrated.

[0089] In this embodiment, the present invention extracts high-risk coordinates based on the abnormal propagation heat map, marks the predicted curve segment as a red focus area, converts the heat map into a semi-transparent color temperature background map (red area is high risk and light yellow area is low risk), implements dual-channel visual encoding (line type maps credibility level and line width maps risk intensity), establishes a prediction curve click-triggered three-dimensional device highlight linkage mechanism, superimposes semi-transparent historical trajectories and annotates time deviations with pulse markers, and triggers multimodal responses through jagged curve deformation, pop-up verification, and sound and light alarms, and captures manually adjusted records to reversely optimize the historical template screening strategy, thereby solving the defects of disconnection between visual warning and physical equipment positioning and the failure of dynamic integration of historical experience.

[0090] like Figure 2 As shown, an IDC computer room operation and maintenance visualization monitoring system provided by an embodiment of the present invention includes:

[0091] The data detection module 100 is used to collect statistics on the service attributes and physical topological relationships of the computer room nodes, build a computer room node topology network, and perform mutual detection based on the computer room node topology network.

[0092] In the system, the data detection module 100 counts the business attributes and physical topology relationship of the machine room nodes. Different machine rooms have different business types, and their data types are also different. Multiple sensors are arranged in the machine room to monitor the environmental data in the machine room, such as temperature sensors (cabinets, air conditioner outlet / inlet and room), humidity sensors, air flow sensors, water immersion sensors, smoke sensors, access control sensors, noise sensors and vibration sensors, etc. The running parameters of the equipment in the machine room include the running indexes of servers / network devices / storage devices (CPU utilization, memory utilization, disk I / O, disk space, network traffic, network card error rate, power supply state and fan speed), UPS state (input / output voltage and current, battery state, load rate), precise air conditioner running parameters (set temperature, supply air temperature, return air temperature, compressor state, fan speed, refrigerant pressure) and PDU current, voltage and power, etc. Then, the equipment reports the data by itself, and at the same time, two business nodes are randomly selected based on the topology network of the machine room nodes to detect each other, perform mutual communication and mutual PING detection, control all business nodes to send ICMP detection packets to each other, so as to obtain the delay parameters and packet loss parameters between the machine room nodes.

[0093] The model training module 200 is configured to obtain historical running data of each machine room, construct an anomaly detection model, and train the machine room based on the historical running data.

[0094] In the system, the model training module 200 obtains the historical running data of each machine room, monitors the running data of the machine room during the historical running process of the machine room, including the environmental parameters and running parameters of the machine room, constructs a data set by means of artificial marking, constructs an anomaly detection model, and the anomaly detection model includes an input layer, an encoder, a decoder, a loss function and an anomaly judgment layer. The input layer is used to receive the prediction parameters, the encoder is used to compress the high-dimensional input data into a low-dimensional latent space, which is composed of multiple fully connected layers, the encoder outputs a low-dimensional latent vector, the decoder outputs reconstructed data based on the low-dimensional latent vector, the loss function is used to evaluate the difference between the original data and the reconstructed data, and the anomaly judgment layer is used to perform anomaly judgment after the model training is completed. The anomaly detection model is trained and verified by the data set, so as to use the anomaly detection model to perform anomaly determination on the collected data.

[0095] The parameter prediction module 300 is configured to collect the environmental data and the running state data of each machine room through sensors, obtain multiple evaluation parameters, and construct a change distribution prediction curve for each evaluation parameter.

[0096] The prediction detection module 400 is used to extract prediction parameters based on the change distribution prediction curve, integrate the prediction parameters, import them into the optimized anomaly detection model, output the detection results, visualize the evaluation parameters and the detection results, and output a visualization picture.

[0097] like Figure 3 As shown, as a preferred embodiment of the present invention, the model training module 200 includes:

[0098] The historical operation data acquisition unit 201 is used to acquire the historical operation data of the computer room, where the historical operation data includes historical environment data and historical operation data.

[0099] In this module, the historical operation data acquisition unit 201 acquires the historical operation data of the computer room, retrieves the historical operation data within the corresponding time length according to the amount of data required for training, divides the historical operation data according to the preset time intervals, and each moment corresponds to a set of historical operation data.

[0100] The data set division unit 202 is used to perform data cleaning on the historical operation data, fill in missing values, remove outliers, perform timestamp alignment and normalization processing, and construct a data set. The data set is divided into a training data set and a validation data set.

[0101] The model construction unit 203 is used to construct an anomaly detection model. The anomaly detection model uses an unsupervised learning autoencoder and is trained and verified using a data set.

[0102] In this module, data cleaning is performed on historical operation data. Missing values, duplicate values ​​and outliers in the historical operation data are identified through data cleaning. Forward filling or backward filling strategies are used to supplement missing values, and outliers and duplicate values ​​are removed. Timestamp alignment is performed to solve the problem of inconsistent timestamps of data points from different devices or sensors, and they are unified to the same time point or time period to facilitate association analysis. Normalization is used to scale indicators of different dimensions and orders of magnitude to a unified numerical range, eliminating dimensional effects, accelerating model convergence, and improving model accuracy. Through manual labeling, corresponding training data sets and verification data sets are constructed, and the anomaly detection model is trained and verified using the training data sets and verification data sets until the anomaly detection success rate of the anomaly detection model reaches the preset value.

[0103] like Figure 4 As shown, as a preferred embodiment of the present invention, the parameter prediction module 300 includes:

[0104] The data acquisition unit 301 is used to periodically acquire detection data from sensors in each computer room and operating status data of the computer room through a data collector to obtain multiple evaluation parameters.

[0105] In this module, the data acquisition unit 301 can obtain corresponding environmental data and the operating status data of the computer room by regularly accessing various sensors. The above data is collected in real time, and the detection values ​​of multiple evaluation parameters are obtained. For example, at time A, the temperature of the computer room is T1, the humidity is W1, and the power of the computer room is P1. Each evaluation parameter corresponds to a value.

[0106] The parameter coordinate construction unit 302 is used to construct a two-dimensional coordinate system, and mark the values ​​of the same evaluation parameter at different times in the two-dimensional coordinate system in chronological order to obtain multiple parameter coordinates.

[0107] In this module, the parameter coordinate construction unit 302 constructs a two-dimensional coordinate system, in which the horizontal coordinate is the time value, normalizes the numerical value of the evaluation parameter, and then marks it in the two-dimensional coordinate system, thereby forming multiple coordinate points in the two-dimensional coordinate system.

[0108] The prediction curve construction unit 303 is used to determine a corresponding fitting function based on multiple parameter coordinates, so as to convert the coordinate points in the two-dimensional coordinate system into a change distribution prediction curve.

[0109] like Figure 5 As shown, as a preferred embodiment of the present invention, the prediction detection module 400 includes:

[0110] The prediction parameter extraction unit 401 is used to extract prediction parameters from each change distribution prediction curve according to a preset extraction interval, and take the prediction parameters extracted at the same time as a group of data to be predicted.

[0111] In this module, the prediction parameter extraction unit 401 extracts prediction parameters from each change distribution prediction curve according to a preset extraction interval, such as 10 seconds as a time interval, and extracts the prediction parameters corresponding to each time point in the next 10 minutes. The prediction parameters at the same time are divided into one data to be predicted.

[0112] The anomaly detection unit 402 is used to import the data to be predicted into the optimized anomaly detection model, analyze the data to be predicted corresponding to each future moment through the optimized anomaly detection model, and generate corresponding detection results.

[0113] The data visualization unit 403 is used to visualize the change distribution prediction curve corresponding to the evaluation parameters, control the style of the output change distribution prediction curve according to the detection results, and output a visualization picture.

[0114] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A visual monitoring method for IDC computer room operation and maintenance, characterized in that: The method comprises: Collect statistics on the business attributes and physical topology relationships of computer room nodes, build a computer room node topology network, and conduct mutual detection based on the computer room node topology network; Obtain historical operating data from each computer room, build an anomaly detection model, and train it based on the historical operating data of the computer room to obtain an optimized anomaly detection model; Sensors are used to collect environmental data and operating status data from each computer room, obtain multiple evaluation parameters, and construct a change distribution prediction curve for each evaluation parameter; Extract prediction parameters based on the change distribution prediction curve, integrate the prediction parameters, import them into the optimized anomaly detection model to output the detection results, visualize the evaluation parameters and detection results, and output a visualization screen; Build an anomaly detection model. The anomaly detection model uses an unsupervised learning encoder and is trained and verified using a dataset. The steps of training and validating the anomaly detection model using the dataset specifically include: Configure the input layer, which includes a main channel and a sub-channel. The main channel is composed of a fully connected layer, and the sub-channel is composed of a convolutional layer. Correlating the historical operation data of the computer room according to the topological node position, mapping the environmental parameters to the main channel to form the main channel time series data, and mapping the equipment indicators to the secondary channel to form the secondary channel time series data; combining the main channel time series data and the secondary channel time series data with the training data set to generate a training matrix; Using an unsupervised learning mechanism, the samples in the training matrix are randomly divided into segments A and B. The decoder of the anomaly detection model is used to reconstruct the feature distribution of segment B based on segment A. The reconstruction error between the original feature distribution of segment B and the reconstructed feature distribution of segment B is calculated using a loss function. Topological constraints are imposed during training to minimize the difference in reconstruction errors between adjacent nodes, thus obtaining a preliminarily trained anomaly detection model. On the validation set, the operating status data of some nodes is randomly blocked, and the status of the blocked nodes is reconstructed using the preliminarily trained anomaly detection model. The reconstruction accuracy is calculated based on the delay indicator of mutual detection between nodes. The verification threshold is dynamically adjusted according to the reconstruction accuracy to establish a topology-aware verification rule base. Based on the topology-aware verification rule library, the verification set is classified by business attributes. Based on the classification type, the anomaly detection model's anomaly judgment layer calculates the reconstruction error of key device indicators in various business scenarios and verifies the delay reconstruction accuracy to generate a scenario verification report. Extract data segments from periods of known historical anomaly events and input them into the preliminarily trained anomaly detection model. The encoder generates a latent space feature representation. The anomaly judgment layer captures the feature mutation points of the latent space feature representation and calculates the matching delay between them and the actual anomaly timestamp to obtain the retrospective detection results. Based on the scenario verification report and retrospective detection results, the anomaly detection model after preliminary training is optimized, the encoder's feature extraction capability for core network nodes is enhanced, and the decoder's reconstruction weight for high-business-value nodes is increased to update the anomaly detection model parameters. After the anomaly detection model parameters are updated, the topology-aware verification rule library is used to test the topology reconstruction capability, and the business scenario verification set is used to evaluate the specific performance. The retrospective detection results are combined to verify the historical anomaly sensitivity, and finally the optimized anomaly detection model is obtained.

2. The IDC room operation and maintenance visual monitoring method according to claim 1, characterized in that: The steps of obtaining historical operation data of each computer room, building an anomaly detection model, and training the model based on the historical operation data of the computer room specifically include: Obtain historical operating data of the computer room, including environmental data and equipment operating indicator data; The historical operation data is cleaned, missing values ​​are filled, outliers are removed, timestamp alignment and normalization are performed, and a data set is constructed. The data set is divided into a training data set and a validation data set.

3. The IDC computer room operation and maintenance visual monitoring method according to claim 1, characterized in that: The step of collecting environmental data and operating status data of each computer room through sensors to obtain multiple evaluation parameters and constructing a change distribution prediction curve for each evaluation parameter specifically includes: The data collector regularly acquires the detection data of sensors in each computer room and the operating status data of the computer room to obtain multiple evaluation parameters; Construct a two-dimensional coordinate system, mark the values ​​of the same evaluation parameter at different times in the two-dimensional coordinate system in chronological order, and obtain multiple parameter coordinates; A corresponding fitting function is determined based on multiple parameter coordinates, thereby converting the coordinate points in the two-dimensional coordinate system into a change distribution prediction curve.

4. The IDC room operation and maintenance visual monitoring method according to claim 3, characterized in that: The step of determining a corresponding fitting function based on multiple parameter coordinates to convert the coordinate points in the two-dimensional coordinate system into a change distribution prediction curve specifically includes: Obtain the physical topology of the computer room nodes, assign weights to the core network node coordinates and the parameter coordinates of the edge nodes, and generate a parameter coordinate set with topological weights; Based on a parameter coordinate set with topological weights, the fluctuation pattern of similar topological structures in the historical operation data of the computer room is retrieved. The morphological similarity between the current coordinate trend and the historical data of the same period is compared to obtain a selected confidence historical reference curve template. Based on the filtered confidence historical reference curve template, select a fitting strategy and generate a fitting strategy that adapts to the current business scenario; Based on the business attributes and the fitting strategy adapted to the current business scenario, the node CPU parameter fitting error tolerance and node disk parameter tolerance are calculated according to the business type to obtain a business-aware fitting parameter set. Real-time environmental data is acquired, and a business-aware fitting parameter set is applied. This data is used as a correction factor. When the temperature change rate exceeds a preset temperature change rate threshold, the density of fitting points is automatically increased. When the humidity suddenly changes and exceeds a preset humidity threshold, segmented reconstruction is triggered, ultimately generating a drift-corrected dynamic function. Based on the drift-corrected dynamic function, the topology constraints are enforced, and the temperature difference between adjacent cabinets is correlated with the device delay fluctuation synchronization rate to obtain a prediction curve that conforms to physical laws. Real-time data is acquired, and a prediction curve that conforms to physical laws is superimposed on the real-time data for display. When the deviation between the predicted value of the prediction curve that conforms to physical laws and the actual value of the real-time data continues to exceed the preset deviation threshold, the abnormal area is automatically marked, and the filtered confidence historical reference curve template is retroactively corrected to obtain a corrected historical reference curve template; and prediction is performed based on the corrected historical reference curve template to obtain the final prediction curve that conforms to physical laws, and the final prediction curve that conforms to physical laws is used as the change distribution prediction curve.

5. The IDC computer room operation and maintenance visual monitoring method according to claim 4, characterized in that: The steps of extracting prediction parameters based on the change distribution prediction curve, integrating the prediction parameters, importing them into the optimized anomaly detection model, outputting the detection results, visualizing the evaluation parameters and the detection results, and outputting a visualization screen specifically include: Extract prediction parameters from each change distribution prediction curve according to a preset extraction interval, and use the prediction parameters extracted at the same time as a group of data to be predicted; Import the data to be predicted into the optimized anomaly detection model, analyze the data to be predicted corresponding to each moment in the future through the optimized anomaly detection model, and generate corresponding detection results; The change distribution prediction curve corresponding to the evaluation parameters is visualized and output, and the style of the output change distribution prediction curve is controlled according to the detection results to output a visualization picture.

6. The IDC room operation and maintenance visual monitoring method according to claim 5, characterized in that: The step of importing the data to be predicted into the optimized anomaly detection model, analyzing the data to be predicted corresponding to each future moment by the optimized anomaly detection model, and generating corresponding detection results specifically includes: Obtain the physical topology of the nodes in the computer room and divide the data to be predicted into topological groups. Assign priority processing queues to the core network node group and regular queues to the edge node group to generate a topologically grouped data set to be predicted. Based on the topologically grouped dataset to be predicted, the abnormal patterns of similar topological groups in the historical operation data of the computer room are retrieved. The similarity between the current data to be predicted and the precursor features of historical abnormal events is compared to obtain the matching confidence score of each group. Based on the matching confidence score of each group, dynamic weights are assigned to each layer of the anomaly detection model. The input layer increases the feature extraction strength for groups corresponding to the confidence interval, and the anomaly judgment layer lowers the judgment threshold for low-confidence groups, generating a topology-aware model weight configuration. Applying topology-aware model weight configuration, a spatiotemporal association matrix is ​​constructed in the encoder's latent space. This vertically associates the historical states of nodes in the same group and horizontally associates the real-time interactions of devices within the current group, forming a feature matrix with spatiotemporal weights. Based on a feature matrix with spatiotemporal weights and combined with topological network relationships, when a node detects a potential anomaly, it predicts the impact probability of associated devices along the physical connection path and calculates the risk coefficient of anomaly propagation from the core switch to the server group to output an anomaly propagation heat map. Based on the anomaly propagation heat map, a three-level verification is performed through the anomaly judgment layer to generate detection results with credibility ratings; Real-time monitoring footage is obtained and the detection results with credibility ratings are superimposed on the real-time monitoring footage, allowing operations personnel to provide confirmation feedback on high-credibility anomalies and automatically correct the historical matching confidence scores and topology-aware model weight configurations of each group to obtain the final detection results.

7. The IDC room operation and maintenance visual monitoring method according to claim 6, characterized in that: The step of visually outputting the change distribution prediction curve corresponding to the evaluation parameter, controlling the style of the output change distribution prediction curve according to the detection result, and outputting the visualization picture specifically includes: Based on the anomaly propagation heat map, the core node coordinates of the high-risk area are extracted, and the prediction curve segments corresponding to the evaluation parameters are marked as red highlighted focus areas in the visualization screen to generate a prediction curve framework with risk markings; Using the prediction curve framework with risk markers, the abnormal propagation heat map is converted into a semi-transparent color temperature layer and superimposed on the background area of ​​the prediction curve to form a visual base map fused with the thermal background; Based on the matching confidence score of each group, a dual-channel visual encoding is implemented on the visualization base map fused with the thermal background to generate a dynamic and adaptive curve rendering scheme; A dynamic and adaptive curve rendering solution is applied, and a click response mechanism is established. When the user selects a specific predicted curve segment, the corresponding physical device node is triggered to highlight and flash in the 3D topology map, obtaining a visual linkage result between the predicted data and the physical device. Based on the visual linkage results of the predicted data and the physical device, the revised historical reference curve template is called, and a semi-transparent historical trajectory is superimposed below the current predicted curve. The time deviation between the historical anomaly point and the current predicted point is marked with pulse markers to construct a spatiotemporal comparative analysis layer. Combined with the spatiotemporal comparative analysis layer to generate threshold alarms and trigger comprehensive alarm responses; Capture records of comprehensive alarm responses. When the operation and maintenance personnel manually adjust the alarm style, the screening strategy of the confidence history reference curve template and the calculation parameters of the matching confidence score are automatically reverse optimized, and finally the final visualization picture that has been manually calibrated is obtained.

8. An IDC computer room operation and maintenance visual monitoring system, characterized in that: The system applies the IDC computer room operation and maintenance visualization monitoring method according to any one of claims 1 to 7, and the system includes: The data detection module is used to collect statistics on the business attributes and physical topology relationships of the computer room nodes, build a computer room node topology network, and perform mutual detection based on the computer room node topology network; The model training module is used to obtain the historical operation data of each computer room, build an anomaly detection model, and train it based on the historical operation data of the computer room to obtain an optimized anomaly detection model; The parameter prediction module is used to collect environmental data and operating status data of each computer room through sensors, obtain multiple evaluation parameters, and construct a change distribution prediction curve for each evaluation parameter; The prediction and detection module is used to extract prediction parameters based on the change distribution prediction curve, integrate the prediction parameters, import them into the optimized anomaly detection model to output the detection results, visualize the evaluation parameters and detection results, and output the visualization picture.

Images