Real-time monitoring method and device for cloud computing server room and medium

Abstract

The embodiment of the invention discloses a real-time monitoring method and equipment for a cloud computing server room and a medium, relates to the technical field of equipment supervision, and is used for solving the problems of multiple blind areas and low reliability and real-time performance in the existing monitoring. The method comprises the following steps: collecting multi-source data in real time through a heterogeneous sensor network deployed in a cloud computing server room, and preprocessing the multi-source data based on edge computing nodes to obtain multi-source data features; performing alignment processing on the multi-source data features based on a preset space-time calibration algorithm so as to construct a global dynamic topology matrix of the cloud computing server room based on a preset machine room digital twin model and the aligned multi-source data features; based on the historical operation data and the global dynamic topology matrix of the cloud computing server room, constructing a multi-dimensional prediction map of the cloud computing server room; and inputting the multi-dimensional prediction map into a preset space-time diagram convolutional neural network model to identify a current abnormal event of the cloud computing server room and carry out early warning.

Description

Technical Field

[0001] This specification relates to the technical field of equipment supervision, and particularly to a real-time monitoring method, device, and medium for a cloud computing server computer room. Background Art

[0002] With the rapid development of information technology, cloud computing data centers, as the core infrastructure for supporting massive data storage and ultra-large-scale computing, have been widely used. Currently, in a cloud computing server computer room, various complex operating environment factors can not only significantly affect the operating efficiency of servers, reduce computing performance, but also may cause problems such as overheating and damage of equipment, short-circuit faults, etc. Moreover, it may even cause fire hazards due to local overheating, resulting in significant economic losses and business interruptions. Therefore, the real-time monitoring of cloud computing server computer rooms has become one of the important links in daily operation and maintenance.

[0003] Traditional computer room monitoring systems mostly adopt limited monitoring points, can only collect a few types of environmental parameters, and have limited data processing capabilities. They often can only perform simple threshold alarms and are difficult to accurately and real-time perceive and analyze the global operating state of the computer room. In addition, server devices in the computer room are densely arranged, there are many power devices and the load changes frequently. Traditional monitoring methods are difficult to capture local subtle abnormal changes. Especially when facing complex features such as sound pattern signals generated by small animal activities and weak vibrations of early equipment failures, it is often difficult to detect, resulting in many blind spots in the monitoring of the computer room operating state, greatly increasing the difficulty and risk of computer room operation and maintenance. Summary of the Invention

[0004] To solve the above technical problems, one or more embodiments of this specification provide a real-time monitoring method, device, and medium for a cloud computing server computer room.

[0005] One or more embodiments of this specification adopt the following technical solutions: One or more embodiments of this specification provide a real-time monitoring method for a cloud computing server computer room, the method including: Real-time collecting multi-source data through a heterogeneous sensor network deployed in the cloud computing server computer room, and preprocessing the multi-source data based on an edge computing node to obtain multi-source data features; Performing alignment processing on the multi-source data features based on a preset spatio-temporal calibration algorithm, and constructing a global dynamic topology matrix of the cloud computing server computer room based on a preset digital twin model of the computer room and the aligned multi-source data features; Constructing a multi-dimensional prediction map of the cloud computing server computer room based on the historical operation data of the cloud computing server computer room and the global dynamic topology matrix; Input the multi-dimensional prediction map into a pre-set spatio-temporal graph convolutional neural network model to identify the current abnormal event in the cloud computing server room; Based on the type and level of the current abnormal event, trigger the corresponding response mechanism for early warning.

[0006] Optionally, in one or more embodiments of this specification, before the multi-source data is collected in real time through the heterogeneous sensor network deployed in the cloud computing server room, the method further includes: According to the device models of the components in the cloud computing server room, determine the deployment condition data and monitoring condition data corresponding to each component, and based on the device models of various types of sensors, determine the performance data corresponding to each type of sensor; Based on the deployment condition data and monitoring condition data corresponding to each component, determine the physical space constraint condition data of the cloud computing server room, and based on the physical coordinates of each component and the physical space constraint condition data, determine the sensor type corresponding to each component; According to the monitoring performance data corresponding to each sensor type and the monitoring performance data, determine the deployment objective functions corresponding to different types of heterogeneous sensors, and optimize the deployment objective functions based on a pre-set gradient descent strategy to obtain the deployment positions of each heterogeneous sensor; According to the deployment positions of each heterogeneous sensor, establish the heterogeneous sensor network corresponding to the cloud computing server room.

[0007] Optionally, in one or more embodiments of this specification, preprocess the multi-source data based on an edge computing node to obtain multi-source data features, specifically including: Receive the multi-source data collected by the heterogeneous sensor network, and convert each multi-source data into a to-be-processed multi-source data in a unified format based on the protocol adapter corresponding to each edge computing node; wherein, the multi-source data includes: temperature, humidity, infrared thermal imaging, device vibration, current fluctuation, voiceprint signal; Perform data cleaning on the to-be-processed multi-source data, and perform normalization processing on the to-be-processed multi-source data after cleaning to convert the to-be-processed multi-source data after cleaning into standard multi-source data; Based on the data types corresponding to each standard multi-source data, determine the corresponding feature extraction process to obtain multi-source data features.

[0008] Optionally, in one or more embodiments of this specification, perform spatio-temporal alignment processing on the multi-source data features based on a pre-set spatio-temporal calibration algorithm, specifically including: Determine the time window corresponding to the multi-source data features based on the timestamp information corresponding to each of the multi-source data features and the preset sampling constraint data; wherein, the preset sampling constraint data is determined based on the historical event duration in the cloud computing server computer room; Slide and compare the time windows with the highest similarity of each multi-source data feature in adjacent time windows to determine the time offset between each multi-source data feature based on the time window with the highest similarity; Group the heterogeneous servers based on the physical space constraint data of the cloud computing server computer room and the deployment locations of the heterogeneous sensors in the heterogeneous sensor network, so as to use the multi-source data feature groups corresponding to each group of heterogeneous servers as the multi-source data feature groups to be aligned; Obtain the spatial feature points corresponding to each multi-source data feature in each multi-source data feature group to be aligned, and determine the spatial offset of each multi-source data feature based on the difference between the spatial feature points; Perform spatio-temporal alignment processing on the multi-source data features based on the time offset and the spatial offset to obtain the aligned multi-source data features.

[0009] Optionally, in one or more embodiments of this specification, based on the preset digital twin model of the computer room and the aligned multi-source data features, construct the global dynamic topology matrix of the cloud computing server computer room, which specifically includes: Obtain the components in the digital twin model of the computer room and the spatial position relationships corresponding to the components, and construct the static connection topology between the components based on the spatial position relationships and the association relationships between the components; wherein, the association relationships include: subordination relationships, connection relationships; Based on the corresponding relationship between the heterogeneous sensors corresponding to the aligned multi-source data features and the components, bind the aligned multi-source data features to the components to determine one or more aligned multi-source data features corresponding to each component; Map the one or more aligned multi-source data features corresponding to each component to the dynamic attribute fields of the component; Based on the dynamic attribute fields, quantify the dynamic strength of the association relationships between the components in real time, and adjust the weights of the edges in the static connection topology based on the dynamic strength to construct the global dynamic topology matrix of the cloud computing server computer room.

[0010] Optionally, in one or more embodiments of this specification, based on the historical operation data of the cloud computing server computer room and the global dynamic topology matrix, construct the multi-dimensional prediction map of the cloud computing server computer room, which specifically includes: Collect the historical operation data of the cloud computing server room to obtain the timestamp sequence and energy consumption phase offset corresponding to each historical start-stop event in the cloud computing server room based on the historical operation data; Based on the timestamp sequence and energy consumption phase offset corresponding to each historical start-stop event, determine the energy propagation path corresponding to each historical start-stop event in each operation stage; Perform sliding cutting on the global dynamic topology matrix based on a preset sliding time window to obtain a sequence of temporal topology snapshots; wherein, the preset sliding time window is determined based on the duration of the historical start-stop events; Based on the real-time state corresponding to the sequence of temporal topology snapshots and the energy propagation path, calculate the energy propagation weights corresponding to each connection relationship in the global dynamic topology matrix; Aggregate the energy propagation weights and the connection relationships within each sliding time window to obtain a multi-dimensional prediction map of the cloud computing server room.

[0011] Optionally, in one or more embodiments of this specification, input the multi-dimensional prediction map into a preset spatio-temporal graph convolutional neural network model to identify the current abnormal event in the cloud computing server room, specifically including: Obtain an initial spatio-temporal graph convolutional neural network model with a spatial convolutional layer and a temporal convolutional layer; Train the initial spatio-temporal graph convolutional neural network model based on existing labeled samples to obtain a preset spatio-temporal graph convolutional neural network model; Input the multi-dimensional prediction map into the preset spatio-temporal graph convolutional neural network model and output the current abnormal event in the cloud computing server room.

[0012] Optionally, in one or more embodiments of this specification, based on the type and level of the current abnormal event, trigger a corresponding response mechanism for early warning, specifically including: According to the historical operation data of the cloud computing server room, determine the historical abnormal event types and historical event levels of the cloud computing server room, and configure response mechanisms corresponding to each historical abnormal event type and historical event level; Obtain the corresponding response mechanism according to the type and level of the current abnormal event, and determine whether the current abnormal event has associated devices based on the response mechanism; If not, generate a corresponding early warning signal based on the preset template corresponding to the response mechanism and transmit it to the corresponding receiving object for early warning; If so, control the associated devices to perform an associated response.

[0013] One or more embodiments of this specification provide a real-time monitoring device for a cloud computing server computer room. The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any one of the above methods.

[0014] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and the computer-executable instructions are configured to: be capable of executing any one of the above methods.

[0015] One or more of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: By deploying a heterogeneous sensor network to collect multi-source data in real time, various operating environment factors and equipment status information in the computer room can be comprehensively perceived. And preprocessing the multi-source data based on edge computing nodes to obtain multi-source data features can effectively reduce the data transmission pressure and improve the data processing efficiency. Using a preset spatio-temporal calibration algorithm to perform spatio-temporal alignment processing on the multi-source data features can solve the problem of inconsistency in timestamp and spatial position of data collected by different sensors, and ensure the accuracy and relevance of the data. Based on the historical operation data of the cloud computing server computer room and the global dynamic topology matrix, a multi-dimensional prediction map is constructed to predict the operation trend of the computer room from different angles, which can efficiently identify possible abnormal events in the cloud computing server computer room and improve the accuracy and timeliness of anomaly detection. According to the type and level of the abnormal event, the corresponding response mechanism is triggered for early warning, reducing the impact of the abnormal event on the operation of the computer room. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings: Figure 1 It is a flowchart of a real-time monitoring method for a cloud computing server computer room provided by an embodiment of this specification; Figure 2 It is a structural diagram of a real-time monitoring device for a cloud computing server computer room provided by an embodiment of this specification; Figure 3Schematic structural diagram of a non-volatile storage medium provided by an embodiment of this specification. Detailed implementation manners

[0017] An embodiment of this specification provides a real-time monitoring method, device, and medium for a cloud computing server room.

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only some of the embodiments of this specification, rather than all the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0019] As Figure 1 shown, an embodiment of this specification provides a flowchart of a real-time monitoring method for a cloud computing server room. It can be seen from Figure 1 that in one or more embodiments of this specification, a real-time monitoring method for a cloud computing server room includes the following steps: S101: Real-time collect multi-source data through a heterogeneous sensor network deployed in the cloud computing server room, and preprocess the multi-source data based on an edge computing node to obtain multi-source data features.

[0020] To achieve full-domain data coverage in the cloud computing server room and thus achieve precise monitoring of the cloud computing server room, in the embodiments of this specification, a network composed of multiple types of sensors is deployed inside the cloud computing server room, covering server devices, infrastructure such as power supplies, air conditioners, fire protection systems, and other device components set in the cloud computing server room. Thus, multi-source data in the cloud computing server room is real-time collected according to the heterogeneous sensor network. Since the types of heterogeneous sensors are diverse and the data formats such as analog signals, digital signals, image streams, log texts, etc. are also different, in order to achieve unified analysis of multi-source data and avoid the problem of data islands, the multi-source data will be preprocessed based on an edge computing node to obtain multi-source data features.

[0021] Further, in one or more embodiments of this specification, before real-time collecting multi-source data through a heterogeneous sensor network deployed in the cloud computing server room, the method further includes the following steps: According to the device models of each component in the cloud computing server computer room, determine the deployment condition data and monitoring condition data corresponding to each component, and based on the device models of various types of sensors, determine the performance data corresponding to various types of sensors. Among them, it should be noted that the deployment condition data are the physical attributes and installation requirements of each component. For example, high-density servers need to be deployed close to the air outlet of the air conditioner; while the monitoring condition data clarify the parameters that need to be monitored for each component and information such as the monitoring frequency. At this time, by combining the deployment condition data and monitoring condition data corresponding to the components, the physical space constraint condition data of the cloud computing server computer room can be analyzed. For example: temperature and humidity sensors need to be deployed near the precision air conditioner, but strong electromagnetic interference areas need to be avoided. Therefore, the safety distance threshold, signal coverage range, anti-interference intensity, etc. all belong to the physical space constraint condition data. At this time, according to the physical coordinates of each component and the physical space constraint condition data, the sensor type corresponding to each component can be determined. It can be understood that the physical coordinates of each component can be matched with the physical space constraint condition data to determine the physical space constraint condition data corresponding to each physical coordinate, and then according to the physical space constraint condition data corresponding to each physical coordinate, the sensor type corresponding to each component can be determined. Then, according to the monitoring performance data corresponding to each sensor type and the monitoring performance data, determine the deployment objective function corresponding to different types of heterogeneous sensors. For example, when the coverage integrity objective of ensuring that all key components in the computer room are within the effective monitoring range of the sensors, the corresponding constraint condition is the physical effective monitoring range of the sensors; the monitoring accuracy objective related to the sensor type and deployment distance; the deployment cost objective including the sensor hardware cost, wiring cost, maintenance cost, etc. The deployment objective function corresponding to the heterogeneous sensors is obtained by weighting each objective with its corresponding weight coefficient. Then, optimize the deployment objective function based on the preset gradient descent strategy to obtain the deployment positions of various heterogeneous sensors. It can be understood that the preset gradient descent strategy is the current gradient calculation method and will not be elaborated here. Then, according to the deployment positions of various heterogeneous sensors, establish the heterogeneous sensor network corresponding to the cloud computing server computer room.

[0022] In this process, the sensor type is accurately selected according to the component monitoring requirements, which can avoid the monitoring blind spots or redundancy problems caused by blind deployment. In the past, when deploying sensors in the computer room, due to the lack of full consideration of the physical space constraints, the position selection of sensors was unreasonable, resulting in problems such as signal occlusion and interference. However, in the embodiments of this specification, by combining the physical space constraints to select sensors, signal interference and equipment failures can be reduced. By combining the deployment condition data, monitoring condition data, and sensor performance data, various factors such as the physical space constraints, monitoring performance, and deployment costs are comprehensively considered, enabling the optimization of sensor deployment and achieving a balance in multiple aspects such as the performance and cost of the monitoring system in the computer room. According to the actual situations of the layouts, components, and sensors of different computer rooms, the optimization of the sensor deployment positions is flexibly carried out, improving the applicability to cloud computing server computer rooms of various scales and types.

[0023] Specifically, in one or more embodiments of this specification, the multi-source data is preprocessed based on the edge computing nodes to obtain multi-source data features, which specifically include: Since the multi-source data comes from different locations and different devices in the computer room and has diversity and complexity, in the embodiments of this specification, the multi-source data collected by the heterogeneous sensor network is received, and each multi-source data is converted into the multi-source data to be processed in a unified format based on the protocol adapter corresponding to each edge computing node; among them, it should be noted that the multi-source data includes: temperature, humidity, infrared thermal imaging, equipment vibration, current fluctuation, voiceprint signal, etc. By converting the multi-source data collected by the heterogeneous sensor network into a unified format through the protocol adapter, the problem of inconsistent data formats of different sensors is solved, enabling various data to be processed under a unified framework, improving the compatibility and integration of the system, and facilitating subsequent unified management and analysis. Then, the multi-source data to be processed is subjected to data cleaning, and the normalized processing is performed on the cleaned multi-source data to be processed to convert the cleaned multi-source data to be processed into standard multi-source data, thereby avoiding problems caused by inconsistent data formats. Based on the data types corresponding to each standard multi-source data, the corresponding feature extraction process is determined to obtain the multi-source data features. Determining the corresponding feature extraction process according to different data types can more accurately extract the features that are important for analysis and decision-making, improving the effectiveness and representativeness of the features.

[0024] S102: Align the multi-source data features based on the preset spatio-temporal calibration algorithm, and construct the global dynamic topology matrix of the cloud computing server computer room based on the preset digital twin model of the computer room and the aligned multi-source data features.

[0025] In traditional multi-source data processing, due to differences in the acquisition time and spatial location of different sensors, data cannot be directly corresponded and associated in terms of time and space, which brings difficulties to data analysis and fusion. Therefore, to solve this problem, in the embodiments of this specification, the multi-source data features will be aligned according to a preset spatio-temporal calibration algorithm, and then a global dynamic topology matrix of the cloud computing server room will be constructed based on the preset digital twin model of the computer room and the aligned multi-source data features. By constructing the global dynamic topology matrix, the information islands between components are broken, various information in the computer room is integrated into a unified map, realizing the interconnection, interoperability and collaborative management of the global information in the computer room, which helps to improve the operation efficiency and management level of the computer room, and timely discover potential problems and make optimization adjustments.

[0026] Specifically, in one or more embodiments of this specification, the spatio-temporal alignment processing of multi-source data features based on a preset spatio-temporal calibration algorithm specifically includes: First, based on the timestamp information corresponding to each multi-source data feature and the preset sampling constraint data, determine the time window corresponding to the multi-source data feature; wherein, the preset sampling constraint data is determined based on the duration of historical events in the cloud computing server room, that is, the mean value of the statistical historical event duration is used as the length of the time window. Then, slide and compare the time windows with the highest similarity of each multi-source data feature within adjacent time windows, so as to determine the time offset between each multi-source data feature according to the time window with the highest similarity. From the spatial perspective, in order to put spatially related data together, the heterogeneous servers will be grouped according to the physical space constraint data of the cloud computing server room and the deployment locations of each heterogeneous sensor in the heterogeneous sensor network, and then the multi-source data feature groups corresponding to each group of heterogeneous servers will be used as the multi-source data feature groups to be aligned. Then, in order to determine the spatial differences between different data and provide a basis for subsequent spatial alignment, the spatial feature points corresponding to each multi-source data feature within each multi-source data feature group to be aligned will be obtained, and the spatial offset of each multi-source data feature will be determined according to the difference between each spatial feature point. Finally, through the time offset and the spatial offset, the spatio-temporal alignment processing of the multi-source data features is performed to obtain the aligned multi-source data features.

[0027] In this process, through steps such as time window division, time offset determination, and space offset determination, data is calibrated from both time and space dimensions to make data from different sources consistent in space-time. Through dynamic time windows and similarity matching, data with different sampling frequencies and asynchronization is aligned, avoiding analysis biases caused by time misalignment. Through grouping and spatial feature point analysis, errors caused by differences in sensor deployment positions are eliminated, ensuring the consistency of data from the same source in space. Through the combined correction of time offset and space offset, multi-source data is strictly aligned in the space-time dimension, providing high-quality input for subsequent construction of the global dynamic topology matrix and abnormal event recognition.

[0028] Specifically, in one or more embodiments of this specification, based on the preset digital twin model of the computer room and the characteristics of the aligned multi-source data, a global dynamic topology matrix of the cloud computing server room is constructed, specifically including: Extract the spatial position relationships corresponding to components such as servers, switches, and air conditioners from the digital twin model of the computer room. Then, based on the position relationships and the association relationships between components, such as subordination relationships and connection relationships, a static connection topology between components is constructed. That is, with components as nodes and position relationships and association relationships as edges, a static connection topology is constructed. Here, it should be noted that: the subordination relationship can be that a server belongs to a certain cabinet, and the association relationship can be that a switch is connected to a server through a network cable, and an air conditioner is associated with a cabinet through an air duct. Then, based on the correspondence between the heterogeneous sensors corresponding to the characteristics of the aligned multi-source data and the components, the characteristics of the aligned multi-source data are bound to the components to determine one or more characteristics of the aligned multi-source data corresponding to each component. It can be understood that each component can be bound to the data characteristics corresponding to multiple sensors. For example, a server is simultaneously associated with data such as temperature, CPU load, and network traffic. Then, map one or more characteristics of the aligned multi-source data corresponding to each component to the dynamic attribute fields of the components. For example, convert the multi-source data characteristics of temperature value 25°C and CPU utilization rate 80% into dynamic attribute fields such as "temperature=25°C" and "cpu_usage=80". Then, according to the dynamic attribute fields, the dynamic strength of the association relationships between the components is quantified in real time. For example, the connection strength between a server and a switch can be measured by the size of the network traffic. The greater the traffic, the higher the strength. Then, based on the dynamic strength, adjust the weights of the edges in the above static connection topology. For example, the initial weight is 1, and when the traffic increases, the weight increases to 1.5. After dynamically adjusting the weights of the edges in the static topology based on the association strength, a weighted dynamic topology matrix is formed to obtain the global dynamic topology matrix of the cloud computing server room.

[0029] S103: Based on the historical operation data of the cloud computing server room and the global dynamic topology matrix, construct a multi-dimensional prediction map of the cloud computing server room.

[0030] To facilitate the early warning of events such as equipment failures and load peaks, in the embodiments of this specification, a multi-dimensional prediction map of the cloud computing server room is constructed through the historical operation data of the cloud computing server room and the global dynamic topology matrix. In this process, a digital twin prediction model of the cloud computing server room is constructed through historical data-driven dynamic modeling and multi-dimensional correlation analysis, solving the problems existing in the traditional monitoring system in monitoring and analysis, and improving the reliability of subsequent early warnings.

[0031] Specifically, in one or more embodiments of this specification, based on the historical operation data of the cloud computing server room and the global dynamic topology matrix, constructing a multi-dimensional prediction map of the cloud computing server room specifically includes: Collect the historical operation data of the cloud computing server room, so as to obtain the timestamp sequence and energy consumption phase offset corresponding to each historical start-stop event in the cloud computing server room according to the historical operation data. It can be understood that historical start-stop events can be identified by keyword matching or threshold detection, and the corresponding timestamps and energy consumption data are associated. The energy consumption phase offset refers to the delay or lead relationship between the device start-stop moment and the energy consumption fluctuation on the time axis, which can be determined based on the phase difference between the calculated start-stop event time series and the energy consumption series. Then, based on the timestamp sequence and energy consumption phase offset corresponding to each historical start-stop event, determine the energy propagation path corresponding to each historical start-stop event in each operation stage, that is, based on the energy consumption phase offset of each component corresponding to the time series, determine the causal relationship of energy consumption changes, and thus construct a directed graph to represent the energy transfer path. Then, perform a sliding cut on the global dynamic topology matrix according to the preset sliding time window to obtain a sequence of time-series topology snapshots; among them, it should be noted that the preset sliding time window is determined based on the duration of the historical start-stop event, and the window size needs to cover the duration of the historical start-stop event. For example, a server usually takes 1-5 minutes to start, and the window is set to 10 minutes, while the sliding step is less than the window size to ensure data overlap and time-series continuity. Then, according to the real-time state and energy propagation path corresponding to the obtained sequence of time-series topology snapshots, calculate the energy propagation weight corresponding to each connection relationship in the global dynamic topology matrix. It should be noted that the weight factors involved in calculating the energy propagation weight include: physical distance: the weight of adjacent devices is higher, such as the heat conduction weight between servers in the same cabinet > across cabinets, energy consumption contribution: calculate the influence degree of the start-stop of a certain device on the energy consumption of other devices through regression analysis. For example, if the start of device A increases the power of device B by 10%, the weight is 0.1, historical frequency: the weight of devices with frequent linkage is higher, such as the start-stop association frequency of the air conditioner and the server fan). Then, aggregate the energy propagation weight and connection relationship within each sliding time window to obtain a multi-dimensional prediction map of the cloud computing server room including the time dimension, space dimension, and energy dimension.

[0032] During this process, the energy conduction sequence and intensity between components are accurately identified through the energy consumption phase offset and the timestamp sequence, avoiding the limitation of traditional topologies that only reflect physical connections. The energy propagation weights and connection relationships are aggregated within each sliding time window, and finally a multi-dimensional prediction map including the time dimension, space dimension, and energy dimension is obtained. This multi-dimensional fusion method can comprehensively and three-dimensionally reflect the operating state of the computer room and the energy consumption propagation situation, providing a more comprehensive perspective for the comprehensive evaluation and optimization of the computer room.

[0033] S104: Input the multi-dimensional prediction map into a pre-set spatio-temporal graph convolutional neural network model to identify the current abnormal events in the cloud computing server computer room.

[0034] After obtaining the multi-dimensional prediction map in the above steps, the multi-dimensional prediction map is input into a pre-set spatio-temporal convolutional neural network model to identify the current abnormal events in the cloud computing server computer room. Specifically, in one or more embodiments of this specification, inputting the multi-dimensional prediction map into a pre-set spatio-temporal graph convolutional neural network model to identify the current abnormal events in the cloud computing server computer room specifically includes the following process: First, in traditional monitoring methods, there is often a single focus on time series while ignoring spatial associations, or a single focus on spatial topologies while ignoring dynamic time series, resulting in missed detections or false detections. Therefore, in the embodiments of this specification, an initial spatio-temporal graph convolutional neural network model with a spatial convolutional layer and a time convolutional layer is obtained. It can be understood that based on spatial convolution, the physical or logical dependencies between devices can be captured, and through time convolution, the dynamic evolution process of anomalies can be captured. Then, the initial spatio-temporal graph convolutional neural network model is trained according to existing labeled samples to obtain a pre-set spatio-temporal graph convolutional neural network model. The multi-dimensional prediction map is input into the pre-set spatio-temporal graph convolutional neural network model, and the current abnormal events in the cloud computing server computer room are output.

[0035] S105: Based on the type and level of the current abnormal event, trigger the corresponding response mechanism for early warning.

[0036] After obtaining the current abnormal events identified in the above step S104, in order to enable timely early warning responses, the corresponding response mechanism will be triggered for early warning according to the type and level of the current abnormal event, so as to implement hierarchical classification processing of abnormal events of different types and severities, ensure that the countermeasures taken are targeted and effective, notify the operation and maintenance personnel in a timely manner for processing, and reduce the impact of abnormal events on the operation of the computer room.

[0037] Specifically, in one or more embodiments of this specification, based on the type and level of the current abnormal event, a corresponding response mechanism is triggered for early warning, which specifically includes: determining the historical abnormal event types and historical event levels of the cloud computing server room according to the historical operation data of the cloud computing server room, and configuring response mechanisms corresponding to each historical abnormal event type and historical event level. Then, according to the type and level of the current abnormal event, the corresponding response mechanism is determined to determine whether the current abnormal event has associated devices based on the response mechanism. If not, a corresponding early warning signal is generated based on the preset template corresponding to the response mechanism and transmitted to the corresponding receiving object for early warning. If so, the associated devices are controlled to perform an associated response. For example, when the server overheats, the air conditioner can be started simultaneously, the fan speed can be adjusted, and the server load can be scheduled to comprehensively and quickly solve the overheating problem. In this process, response rules are defined in advance through historical data to avoid the blindness of real-time decision-making, ensure the optimal processing method for abnormal events of different types and severities, control the associated devices to perform automated responses without manual intervention, quickly handle abnormal events, reduce the risk of manual response delay or error, improve the operation and maintenance efficiency, and effectively handle complex abnormal events by coordinating the collaborative work of devices. By accurately identifying abnormal events and triggering corresponding response mechanisms, hierarchical response is achieved, effectively controlling risks and reducing the impact of abnormal events on the operation of the computer room.

[0038] As Figure 2 shown, an embodiment of this specification provides a structural schematic diagram of a real-time monitoring device for a cloud computing server room. It can be seen that Figure 2 in one or more embodiments of this specification, a real-time monitoring device for a cloud computing server room includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to: execute the method described in any of the above.

[0039] As Figure 3 shown, an embodiment of this specification provides a structural schematic diagram of a non-volatile storage medium. It can be seen that Figure 3 in one or more embodiments of this specification, a non-volatile storage medium stores computer-executable instructions, and the computer-executable instructions can: execute the method described in any of the above.

[0040] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the embodiments of devices, equipment, and non-volatile computer storage media, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the corresponding parts of the method embodiments for the relevant content.

[0041] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order from that in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0042] The above is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to one or more embodiments of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included within the scope of the claims of this specification.

Claims

1. A real-time monitoring method for a cloud computing server computer room, characterized in that, The method includes: Collecting multi-source data in real time through a heterogeneous sensor network deployed in the cloud computing server computer room, and preprocessing the multi-source data based on edge computing nodes to obtain multi-source data features; Performing alignment processing on the multi-source data features based on a preset spatio-temporal calibration algorithm, and constructing a global dynamic topology matrix of the cloud computing server computer room based on the preset digital twin model of the computer room and the aligned multi-source data features; Constructing a multi-dimensional prediction map of the cloud computing server computer room based on the historical operation data of the cloud computing server computer room and the global dynamic topology matrix; Inputting the multi-dimensional prediction map into a preset spatio-temporal graph convolutional neural network model to identify the current abnormal events in the cloud computing server computer room; Triggering a corresponding response mechanism for early warning based on the type and level of the current abnormal event.

2. The real-time monitoring method of a cloud computing server computer room according to claim 1, characterized in that Before collecting multi-source data in real time through a heterogeneous sensor network deployed in the cloud computing server computer room, the method further includes: Determining the deployment condition data and monitoring condition data corresponding to each component according to the device models of the components in the cloud computing server computer room, and determining the performance data corresponding to each type of sensor based on the device models of each type of sensor; Determining the physical space constraint condition data of the cloud computing server computer room based on the deployment condition data and monitoring condition data corresponding to each component, and determining the sensor type corresponding to each component according to the physical coordinates of each component and the physical space constraint condition data; Determining the deployment objective functions corresponding to different types of heterogeneous sensors according to the monitoring performance data corresponding to each sensor type and the monitoring performance data, and optimizing the deployment objective functions based on a preset gradient descent strategy to obtain the deployment positions of each heterogeneous sensor; Establishing a heterogeneous sensor network corresponding to the cloud computing server computer room according to the deployment positions of each heterogeneous sensor.

3. The real-time monitoring method of a cloud computing server room according to claim 1, characterized in that Preprocessing the multi-source data based on edge computing nodes to obtain multi-source data features, specifically including: Receiving the multi-source data collected by the heterogeneous sensor network, and converting each multi-source data into a uniformly formatted multi-source data to be processed based on the protocol adapter corresponding to each edge computing node; wherein, the multi-source data includes: temperature, humidity, infrared thermal imaging, device vibration, current fluctuation, voiceprint signal; Performing data cleaning on the multi-source data to be processed, and performing normalization processing on the cleaned multi-source data to be processed to convert the cleaned multi-source data to be processed into standard multi-source data; Determining a corresponding feature extraction process based on the data type corresponding to each standard multi-source data to obtain multi-source data features.

4. The real-time monitoring method of a cloud computing server room according to claim 2, characterized in that, Performing spatio-temporal alignment processing on the multi-source data features based on a preset spatio-temporal calibration algorithm, specifically including: Determining the time window corresponding to the multi-source data features based on the timestamp information corresponding to each multi-source data feature and the preset sampling constraint data; wherein, the preset sampling constraint data is determined based on the historical event duration in the cloud computing server computer room; Sliding and comparing time windows with the highest similarity of each multi-source data feature within adjacent time windows to determine the time offset between each multi-source data feature based on the time window with the highest similarity; Grouping the heterogeneous servers based on the physical space constraint data of the cloud computing server room and the deployment locations of each heterogeneous sensor in the heterogeneous sensor network, so as to use the multi-source data feature groups corresponding to each group of heterogeneous servers as the multi-source data feature groups to be aligned; Obtaining the spatial feature points corresponding to each multi-source data feature in each multi-source data feature group to be aligned, and determining the spatial offset of each multi-source data feature according to the difference between each spatial feature point; Performing spatio-temporal alignment processing on the multi-source data features based on the time offset and the spatial offset to obtain the aligned multi-source data features; 5. The real-time monitoring method of a cloud computing server computer room according to claim 1, characterized in that Based on a preset digital twin model of the computer room and the aligned multi-source data features, constructing a global dynamic topology matrix of the cloud computing server room, specifically including: Obtaining the components in the digital twin model of the computer room and the spatial position relationships corresponding to each component, and constructing a static connection topology between each component based on the spatial position relationships and the association relationships between each component; where the association relationships include: subordination relationships, connection relationships; Based on the corresponding relationship between the heterogeneous sensors corresponding to the aligned multi-source data features and the components, binding the aligned multi-source data features with the components to determine one or more aligned multi-source data features corresponding to each component; Mapping one or more aligned multi-source data features corresponding to each component to the dynamic attribute fields of the component; Based on the dynamic attribute fields, quantifying the dynamic strength of the association relationships between each component in real time, and adjusting the weights of each edge in the static connection topology based on the dynamic strength to construct the global dynamic topology matrix of the cloud computing server room; 6. The real-time monitoring method of a cloud computing server room according to claim 1, characterized in that, Based on the historical operation data of the cloud computing server room and the global dynamic topology matrix, constructing a multi-dimensional prediction map of the cloud computing server room, specifically including: Collecting the historical operation data of the cloud computing server room to obtain the time stamp sequence and energy consumption phase offset corresponding to each historical start-stop event in the cloud computing server room based on the historical operation data; Based on the time stamp sequence and energy consumption phase offset corresponding to each historical start-stop event, determining the energy propagation path corresponding to each historical start-stop event in each operation stage; Performing sliding cutting on the global dynamic topology matrix based on a preset sliding time window to obtain a sequence of time series topology snapshots; where the preset sliding time window is determined based on the duration of the historical start-stop event; Calculating the energy propagation weights corresponding to each connection relationship in the global dynamic topology matrix based on the real-time state corresponding to the sequence of time series topology snapshots and the energy propagation path; Aggregate the energy propagation weight and the connection relationship within each of the sliding time windows to obtain a multi-dimensional prediction map of the cloud computing server computer room.

7. A real-time monitoring method for a cloud computing server room according to claim 1, characterized in that, Input the multi-dimensional prediction map into a pre-set spatio-temporal graph convolutional neural network model to identify the current abnormal event in the cloud computing server computer room, specifically including: Obtain an initial spatio-temporal graph convolutional neural network model with a spatial convolutional layer and a temporal convolutional layer; Train the initial spatio-temporal graph convolutional neural network model based on existing labeled samples to obtain a pre-set spatio-temporal graph convolutional neural network model; Input the multi-dimensional prediction map into the pre-set spatio-temporal graph convolutional neural network model and output the current abnormal event in the cloud computing server computer room.

8. The real-time monitoring method of a cloud computing server room according to claim 1, characterized in that Based on the type and level of the current abnormal event, trigger a corresponding response mechanism for early warning, specifically including: According to the historical operation data of the cloud computing server computer room, determine the historical abnormal event types and historical event levels of the cloud computing server computer room, and configure response mechanisms corresponding to each of the historical abnormal event types and historical event levels; Obtain the corresponding response mechanism according to the type and level of the current abnormal event, and determine whether the current abnormal event has associated devices based on the response mechanism; If not, generate a corresponding early warning signal based on the pre-set template corresponding to the response mechanism and transmit it to the corresponding receiving object for early warning; If so, control the associated devices to perform an associated response.

9. A real-time monitoring device for a cloud computing server room, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute the method according to any one of claims 1-8 above.

10. A non-volatile storage medium stores computer-executable instructions, characterized in that, The computer-executable instructions can: execute the method according to any one of claims 1-8 above.

Images