A server cabinet intelligent environment monitoring system

Abstract

This invention discloses an intelligent environmental monitoring system for server racks, comprising: an edge perception layer deployed in the server rack for collecting environmental time-series data; a data governance layer including a data verification module, a compression storage module, and a feedback iteration module for anomaly detection, layered compression, and model updates; a collaborative judgment layer including a consensus module for cross-validating anomalous data using the PBFT consensus algorithm; an intelligent control layer including a control module for generating air conditioning control commands using a model predictive control algorithm; and a digital twin layer including a 3D model and a prediction module for predicting environmental change trends and visualizing them. This invention achieves a closed-loop end-to-end system encompassing data verification, lightweight storage, collaborative consensus, intelligent control, and digital twins, significantly reducing false alarm rates, optimizing storage space, improving control accuracy, and enabling predictive maintenance. It can be widely used for data center environmental monitoring.

Description

Technical Field

[0001] This invention belongs to the field of data center environment monitoring technology, and in particular relates to an intelligent environment monitoring system for server racks. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence technologies, data centers are expanding in scale, and the number of server racks is growing exponentially. The stable operation of servers within these racks places extremely high demands on environmental parameters such as temperature, humidity, and airflow, making environmental monitoring systems a critical infrastructure for ensuring data center reliability. Currently, various technical solutions have been developed in the field of server rack environmental monitoring, mainly including independent monitoring systems for single racks, centralized data center monitoring systems, and monitoring systems with basic predictive capabilities. Independent monitoring systems for single racks deploy sensors and local controllers independently in each rack, triggering alarms based on fixed thresholds and using PID algorithms to locally control air conditioning; all sensor data is stored locally. Centralized data center monitoring systems aggregate all sensor data through a local area network to a central server for unified processing. The central server performs anomaly detection and rule-based judgment, centrally issues control commands, and uses a centralized database for storage. Furthermore, some advanced systems add time-series predictive models to the data acquisition, enabling prediction of future temperatures, but the prediction results are only used for alarms.

[0003] However, existing technologies still have many shortcomings in practical applications. Regarding data verification, sensors are prone to drift, aging, and intermittent interruptions during long-term operation. Existing anomaly detection methods often use fixed thresholds or simple filtering, which easily leads to the loss of genuine fault signals, resulting in high false alarm and false negative rates. False alarms cause maintenance personnel to ignore genuine alarms, while false negatives can directly cause server overheating and crashes. In terms of storage optimization, existing technologies generally use full-volume storage. Data centers with thousands of servers generate massive amounts of time-series data every second, not only consuming significant storage space but also resulting in low data retrieval and backtracking efficiency, failing to meet the needs of rapid fault location. Regarding coordination mechanisms, existing solutions are mostly single-point or centralized processing, unable to eliminate the impact of single-rack sensor errors on global judgment. The coupling of hot and cold air currents within the data center means that single-point misjudgments can easily lead to global false alarms and unnecessary adjustments. Regarding control methods, existing technologies suffer from severe response lag. By the time the monitoring system detects a temperature anomaly, a hotspot has already formed, forcing air conditioning to react passively, resulting in significant energy waste and impacting server lifespan. In terms of predictive capabilities, existing solutions are mostly reactive or real-time control measures, only able to react to anomalies that have already occurred, unable to predict future environmental trends, leaving operations and maintenance in a state of constant reactive response. Regarding system closed-loop management, existing technologies are mostly open-loop or semi-open-loop systems, unable to self-optimize based on control effects, and the models become ineffective after long-term operation, making it difficult to achieve continuous and stable environmental monitoring and control. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art, this invention provides an intelligent environment monitoring system for server racks, which solves the problems of inaccurate monitoring, high storage overhead, delayed control response, and lack of predictive maintenance capabilities in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A server rack intelligent environment monitoring system includes:

[0007] The edge sensing layer, deployed in the server rack, includes multimodal sensors for collecting environmental time-series data from the rack.

[0008] The data governance layer, connected to the edge perception layer, includes a data verification module, a compression and storage module, and a feedback and iteration module. The data verification module is used to perform anomaly detection on the environmental time-series data using an isolated forest model, distinguishing between sensor dead data, abnormal data, and normal data. The compression and storage module is used to perform layered compression on the detected data and build an index. The feedback and iteration module is used to obtain the environmental time-series data after the air conditioning control command is executed, and input the executed environmental time-series data into the data verification module to update the detection parameters of the isolated forest model.

[0009] The collaborative analysis layer, connected to the data governance layer, includes a consensus module; the consensus module is used to cross-validate the abnormal data among the nodes of the rack cluster using the PBFT consensus algorithm and output a collaborative judgment result.

[0010] The intelligent control layer, connected to the collaborative judgment layer and the air conditioning system, includes a control module; the control module is used to generate and issue air conditioning control commands based on the collaborative judgment results using a model predictive control algorithm.

[0011] The digital twin layer, connected to the data governance layer, collaborative analysis layer, and intelligent control layer, includes a 3D model and a prediction module. The prediction module is used to predict environmental change trends based on historical and real-time data. The 3D model is used to map the data center environment status and display the prediction results of the environmental change trends.

[0012] Preferably, the isolated forest model in the data verification module includes a drift detection unit; the drift detection unit is used to monitor the distribution shift of environmental time series data, and update the model parameters of the isolated forest model when a distribution shift is detected; the distribution shift is determined by a concept drift detection algorithm.

[0013] Preferably, the compressed storage module includes:

[0014] The first storage area is used to store normal data within a preset first time period;

[0015] The second storage area is used to store normal data that exceeds the preset first time period after differential compression or downsampling compression, and to store a complete copy of the data marked as abnormal.

[0016] An index unit is used to build an index based on a log structure merge tree, the index including a hash index or an inverted index.

[0017] Preferably, the compressed storage module further includes a baseline value update unit; the baseline value update unit is used to update the baseline value of differential compression according to a preset period, the preset period being determined based on the rack load fluctuation characteristics; the rack load fluctuation characteristics include the periodic change pattern of CPU utilization or network throughput obtained through frequency domain analysis.

[0018] Preferably, the consensus module further includes a reputation evaluation unit; the reputation evaluation unit is used to calculate a reputation value based on the online duration, consensus participation, or false alarm rate of the rack nodes; the PBFT consensus algorithm sets weights for node voting based on the reputation value; the reputation value is calculated by weighted summation, and the weight coefficients of the weighted summation include online duration weight, consensus participation weight, and false alarm rate weight, wherein the false alarm rate weight is greater than the online duration weight and the consensus participation weight.

[0019] Preferably, the model predictive control algorithm in the control module is used for:

[0020] The objective function is constructed based on the temperature deviation and energy consumption within a preset time domain.

[0021] Solve the control sequence under preset constraints; and

[0022] The first control command in the control sequence is sent to the air conditioning system as the current control command.

[0023] The objective function is to minimize the sum of the squares of the differences between the predicted temperature and the target temperature and the sum of the estimated energy consumption.

[0024] Preferably, the control module further includes a feedback correction unit; the feedback correction unit is used to acquire the controlled environmental time series data, compare the controlled environmental time series data with the predicted values ​​generated by the model predictive control algorithm, and correct the state variables or model parameters of the prediction model of the model predictive control algorithm according to the comparison result.

[0025] Preferably, the prediction module employs a neural network model based on an attention mechanism; the neural network model based on an attention mechanism is used to determine the thermal correlation between different cabinet locations and predict the temperature distribution within a preset second time period.

[0026] Preferably, the digital twin layer is further used to input the prediction results of the prediction module into the three-dimensional model to generate and display a temperature distribution map within a preset third time period; the preset third time period is at least one preset moment or continuous time interval within the preset second time period; the temperature distribution map is a heat map superimposed on the three-dimensional model.

[0027] Preferably, the digital twin layer further includes a trend inference module; the trend inference module is used to generate a dynamic change sequence based on the prediction results of the prediction module, and continuously display the environmental change process within the preset second time period through the three-dimensional model.

[0028] The technical effects and advantages of the intelligent environmental monitoring system for server racks of this invention are as follows:

[0029] 1. This invention uses an isolated forest model to detect anomalies in environmental time-series data. Combined with a drift detection unit that monitors data distribution shifts in real time and dynamically updates model parameters when conceptual drift is detected, it can accurately distinguish between faulty sensor data, truly abnormal data, and normal data, completely solving noise interference problems such as drift, aging, and momentary interruptions caused by long-term sensor operation. The feedback iteration module sends environmental data after the air conditioning control command is executed back to the data verification module, continuously updating the detection parameters of the isolated forest model. This gives the model adaptive capabilities, enabling it to adapt to normal data distribution drift caused by changes in data center layout and server load fluctuations, avoiding large-scale false alarms.

[0030] 2. This invention employs a layered design of a first storage area and a second storage area. Normal data within a preset first time period is stored in its entirety, while normal data exceeding this time period is differentially compressed or downsampled before storage. Data marked as abnormal is stored as a complete copy, significantly reducing storage overhead while ensuring the integrity of critical data. The baseline value update unit updates the differentially compressed baseline value according to a preset period. This preset period is dynamically adjusted based on the periodic changes in CPU utilization or network throughput obtained through frequency domain analysis, further optimizing compression efficiency. The indexing unit constructs hash indexes and inverted indexes based on a log structure merge tree, supporting second-level data retrieval and backtracking based on timestamps, sensor IDs, or anomaly types.

[0031] 3. This invention uses the PBFT consensus algorithm to perform cross-validation between cabinet cluster nodes on abnormal data uploaded by the data governance layer. Combined with the reputation assessment unit, it calculates the reputation value based on the online duration, consensus participation, or false alarm rate of the cabinet nodes, and sets weights for node voting based on the reputation value (where the false alarm rate weight is greater than the online duration weight and consensus participation weight). This effectively filters out local sensor interference and single-point measurement errors, outputs globally reliable collaborative judgment results, and avoids unnecessary alarms and controls for the entire area caused by false alarms from a single cabinet.

[0032] 4. This invention generates air conditioning control commands based on collaborative decision results using a model predictive control algorithm. It constructs an objective function using temperature deviation and energy consumption within a preset time domain, solves for the optimal control sequence under preset constraints, and issues and executes the first control command, achieving forward-looking and precise regulation. The feedback correction unit acquires the actual environmental data after regulation and compares it with the predicted values ​​generated by the model predictive control algorithm. Based on the comparison results, it corrects the state variables or model parameters of the prediction model, compensating for model mismatch and external disturbances, effectively avoiding over-adjustment or under-adjustment, and achieving dual optimization of temperature control accuracy and energy efficiency.

[0033] 5. This invention uses an attention-based neural network model to determine the thermal correlation between different rack locations, predicts the temperature distribution within a preset second time period, and inputs the prediction results into a 3D model to generate and display a temperature distribution map within a preset third time period. The preset third time period is at least one preset moment or continuous time interval within the preset second time period, and the temperature distribution map is a heatmap superimposed on the 3D model. The trend prediction module generates a dynamic change sequence based on the prediction results, continuously displaying the environmental change process within the preset second time period through the 3D model, enabling maintenance personnel to intuitively view future environmental trends and perform preventative adjustments before server overheating.

[0034] 6. This invention uses a feedback iteration module to send the regulated environmental data back to the data governance layer to update the model parameters, forming a closed-loop self-iterative mechanism of "perception-governance-consensus-regulation-twin". It can automatically run the entire process from data collection, verification, compression, consensus, regulation to prediction and display without human intervention. It can maintain the best monitoring and regulation performance for a long time, adapt to large-scale data center cluster deployment, and significantly reduce manual operation and maintenance costs. Attached Figure Description

[0035] Figure 1 This is an architecture diagram of a server rack intelligent environment monitoring system provided in an embodiment of the present invention.

[0036] Figure 2 This is a structural diagram of the internal modules of the data governance layer provided in an embodiment of the present invention.

[0037] Figure 3The flowchart of the model predictive control algorithm provided in the embodiment of the present invention is shown.

[0038] Figure 4 This is a schematic diagram illustrating the prediction results of the digital twin layer provided in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "includes..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0041] refer to Figure 1-4 This invention provides an intelligent environment monitoring system for server racks, comprising:

[0042] The edge sensing layer, deployed in the server rack, includes multimodal sensors for collecting environmental time-series data from the rack.

[0043] The data governance layer, connected to the edge perception layer, includes a data verification module, a compression and storage module, and a feedback and iteration module. The data verification module is used to perform anomaly detection on the environmental time-series data using an isolated forest model, distinguishing between sensor dead data, abnormal data, and normal data. The compression and storage module is used to perform layered compression on the detected data and build an index. The feedback and iteration module is used to obtain the environmental time-series data after the air conditioning control command is executed, and input the executed environmental time-series data into the data verification module to update the detection parameters of the isolated forest model.

[0044] The collaborative analysis layer, connected to the data governance layer, includes a consensus module; the consensus module is used to cross-validate the abnormal data among the nodes of the rack cluster using the PBFT consensus algorithm and output a collaborative judgment result.

[0045] An intelligent control layer, connected to the collaborative judgment layer and the air conditioning system, includes a control module; the control module is used to generate and issue air conditioning control commands based on the collaborative judgment results using a model predictive control algorithm; and

[0046] The digital twin layer, connected to the data governance layer, collaborative analysis layer, and intelligent control layer, includes a 3D model and a prediction module. The prediction module is used to predict environmental change trends based on historical and real-time data. The 3D model is used to map the data center environment status and display the prediction results of the environmental change trends.

[0047] Example 1

[0048] This embodiment provides an intelligent environment monitoring system for server racks, used for the overall system architecture, and the specific implementation includes:

[0049] Purpose of implementation: This embodiment aims to build a complete multi-layer closed-loop monitoring system to solve the problems of isolated functional modules and inability to flow data in a closed loop in the existing technology, and to realize full-link automation from data acquisition to intelligent control.

[0050] System architecture: such as Figure 1 As shown, this embodiment provides a server rack intelligent environment monitoring system, including an edge perception layer, a data governance layer, a collaborative analysis layer, an intelligent control layer, and a digital twin layer.

[0051] The edge awareness layer is deployed inside each server rack and includes multimodal sensors. These multimodal sensors include at least temperature sensors, humidity sensors, airflow velocity sensors, vibration sensors, smoke sensors, and server power consumption monitoring units, used to collect raw physical quantity data of the environment inside and around the rack from different dimensions. The edge awareness layer collects environmental time-series data at a preset frequency (e.g., 1Hz) and transmits the data stream to the data governance layer in real time.

[0052] The data governance layer is deployed on the edge gateway or near-end server and connected to the edge perception layer via an industrial bus or local area network. The data governance layer includes a data verification module, a compression and storage module, and a feedback iteration module. The data verification module is used to detect and classify anomalies in the raw data; the compression and storage module is used for efficient storage of the detected data; and the feedback iteration module is used to achieve closed-loop self-optimization of the model.

[0053] The collaborative analysis layer is deployed across multiple nodes in a cluster of server racks (typically one node per rack), interconnected via a local area network to form a distributed network. The collaborative analysis layer includes a consensus module for cross-validation of anomalous data across multiple nodes.

[0054] The intelligent control layer is deployed on the central controller in the computer room or on the cloud server, and connects with the collaborative analysis layer and the air conditioning system. The intelligent control layer includes a control module, which is used to generate and issue air conditioning control commands.

[0055] The digital twin layer is deployed in a monitoring center or cloud platform, connecting with the data governance layer, collaborative analysis layer, and intelligent control layer. The digital twin layer includes a 3D model, a prediction module, and a trend projection module, used for visual mapping of environmental conditions and projection of future trends.

[0056] Implementation steps:

[0057] The edge perception layer collects real-time environmental time-series data of the cabinet and transmits it to the data governance layer.

[0058] The data governance layer verifies, compresses, and indexes the data, and uploads abnormal data to the collaborative analysis layer.

[0059] The collaborative judgment layer uses a consensus algorithm to cross-validate abnormal data and outputs collaborative judgment results.

[0060] Based on the collaborative judgment results, the intelligent control layer generates and issues air conditioning control commands through model predictive control algorithms.

[0061] The digital twin layer predicts environmental change trends based on historical and real-time data and visualizes them through a 3D model.

[0062] The feedback iteration module acquires the adjusted environmental data and updates the model parameters of the data verification module, forming a closed loop.

[0063] Implementation Results: This embodiment achieves full-process automation from data acquisition to intelligent control through deep coupling of a five-layer architecture. Seamless data flow between layers avoids information silos and lays the foundation for subsequent fine-tuning of each module.

[0064] Example 2

[0065] This embodiment provides a server rack intelligent environment monitoring system, specifically implementing a data verification module. The specific implementation details include:

[0066] Purpose of implementation: This embodiment aims to solve the drift and noise problems caused by long-term operation of sensors, accurately distinguish between real anomalies and sensor defects, and avoid false alarms and missed alarms.

[0067] System Architecture: The data verification module is deployed within the data governance layer, including the Isolation Forest model and its drift detection unit. The Isolation Forest model adopts a short-period sliding window training mode, which is suitable for edge computing environments with limited resources.

[0068] Implementation steps:

[0069] Acquire the environmental time-series data stream transmitted from the edge perception layer.

[0070] The isolated forest model assigns anomaly scores to each data point and sets high and low thresholds.

[0071] If the abnormal score is higher than the high threshold, it is determined to be sensor defective data, which is discarded directly and a sensor fault alarm is triggered.

[0072] Abnormal scores between the low and high thresholds: These are identified as abnormal data, their features are preserved, and they are uploaded to the collaborative analysis layer.

[0073] Abnormal scores below the low threshold: are judged as normal data and sent to the compressed storage module.

[0074] The drift detection unit monitors the distribution of environmental data in real time. When it detects concept drift (such as a sharp change in server load causing an overall temperature shift), it triggers an incremental update of the model, dynamically adjusting the threshold and tree structure to prevent large-scale false alarms.

[0075] Implementation Results: This embodiment achieves adaptive capability for anomaly detection, with model inference time ≤15ms, anomaly detection accuracy ≥98%, and false positive and false negative rates reduced by more than 90%.

[0076] Example 3

[0077] This embodiment provides a server rack intelligent environment monitoring system for the specific implementation of a compressed storage module, the specific implementation content of which includes:

[0078] Purpose of implementation: This embodiment aims to solve the problem of redundant storage of massive time-series data, significantly reduce storage space usage, and at the same time ensure the integrity of key data and retrieval efficiency.

[0079] System architecture: The compressed storage module includes a first storage area, a second storage area, an index unit, and a baseline value update unit.

[0080] Implementation steps:

[0081] The first storage area stores normal data within a preset first time period (e.g., 24 hours), using a full-volume storage method and supporting high-frequency queries.

[0082] The second storage area stores normal data that exceeds the preset first time period after differential compression or downsampling compression; for data marked as abnormal, a complete copy is retained in the second storage area regardless of the time period to ensure the integrity of fault tracing.

[0083] The baseline value update unit updates the differential compression baseline value according to a preset cycle. The preset cycle is determined based on the rack load fluctuation characteristics: by analyzing the periodic patterns of CPU utilization or network throughput through Fast Fourier Transform (FFT), the update cycle is set to 10 minutes during peak business periods (such as 10:00-12:00 daily) and 1 hour during off-peak periods.

[0084] The index unit builds a hash index (for precise matching based on timestamps and sensor IDs) and an inverted index (for retrieving relevant data locations based on anomaly types) based on the log structure merge tree, supporting second-level data backtracking.

[0085] Implementation results: This embodiment achieves a compression ratio of ≥70%, data retrieval latency of ≤50ms, and a reduction in storage space usage of more than 70%.

[0086] Example 4

[0087] This embodiment provides a server rack intelligent environment monitoring system for the specific implementation of a consensus module, the specific implementation content of which includes:

[0088] Purpose of implementation: This embodiment aims to solve the problem that independent monitoring of a single cabinet is easily affected by single-point errors, and improve the accuracy of cluster environment judgment through distributed consensus.

[0089] System architecture: The consensus module includes the PBFT consensus engine and the reputation evaluation unit.

[0090] Implementation steps:

[0091] Each rack node packages the abnormal data it determines as a request and broadcasts it to its neighboring nodes.

[0092] The credibility assessment unit calculates the credibility value based on the online duration, consensus participation, and false alarm rate of each node. The credibility value is calculated by weighted summation: Credibility value = α × Normalized online duration + β × Consensus participation + γ × (1 - False alarm rate), where γ > α and γ > β, to ensure that the false alarm rate has the highest weight (in this embodiment, α = 0.3, β = 0.2, γ = 0.5).

[0093] In the PBFT consensus process, each node's vote is weighted according to its credibility value, and the total number of votes in favor is accumulated.

[0094] When the weighted number of votes in favor exceeds 2 / 3 of the total weight, the abnormal data is confirmed as a global collaborative judgment result and output to the intelligent control layer.

[0095] Implementation results: This embodiment effectively filters out local sensor interference and single-point errors, with consensus determination time ≤120ms, and significantly improves the reliability of collaborative judgment.

[0096] Example 5

[0097] This embodiment provides an intelligent environment monitoring system for server racks, which is used for the specific implementation of a model predictive control algorithm. The specific implementation content includes:

[0098] Purpose of implementation: This embodiment aims to solve the problems of slow response and energy waste in traditional temperature control, and achieve forward-looking and precise control.

[0099] System architecture: The control module includes an MPC controller and a feedback correction unit.

[0100] Implementation steps:

[0101] Establish a thermodynamic state-space model for the computer room: T(k+1) = A·T(k) + B·u(k) + w(k), where T(k) is the temperature vector of each cabinet at time k, u(k) is the air conditioning control command vector at time k, A and B are system matrices, and w(k) is process noise.

[0102] The MPC controller uses a model to predict cabinet temperature changes within a preset time range (e.g., 15 minutes) based on the current state and future control sequences.

[0103] Construct the objective function: Where N is the prediction time domain, Let Q be the target temperature, and let R be the weight matrix.

[0104] Solve for the optimal control sequence [u(k), u(k+1), u(k+2)] under constraints (such as the lower limit of the supply air temperature and the upper limit of the fan speed).

[0105] The first control command u(k) in the optimal control sequence is sent to the air conditioning system as the current control command.

[0106] The feedback correction unit acquires the actual environmental data T_actual(k+1) after regulation, compares it with the predicted value T(k+1|k) to obtain the error e(k+1), and uses this error to correct the state variables of the prediction model or identify the model parameters online through Kalman filtering.

[0107] Implementation results: This embodiment achieves a temperature control accuracy of ±0.3℃, reduces air conditioning energy consumption by 18%-25%, and effectively avoids over-adjustment / under-adjustment.

[0108] Example 6

[0109] This embodiment provides a server rack intelligent environment monitoring system, specifically for the implementation of a prediction module, the specific implementation content of which includes:

[0110] Purpose of implementation: This embodiment aims to achieve early prediction of environmental trends and upgrade the operation and maintenance mode from passive response to proactive prevention.

[0111] System architecture: The prediction module adopts a Transformer model based on the attention mechanism and is deployed in the digital twin layer.

[0112] Implementation steps:

[0113] Collect historical environmental time-series data (temperature, humidity, load, etc.) from the past 30 days as the training set, take the data from the past 60 minutes as input, and the temperature distribution for the next 30 minutes as output to train the Transformer model.

[0114] The model automatically learns the thermal correlation between different rack locations through a multi-head attention mechanism, capturing spatial coupling characteristics.

[0115] In real-time operation, the prediction module continuously forecasts the temperature distribution for the next 30 minutes based on the latest data.

[0116] The prediction results are input into the 3D model, and the temperature distribution maps at the 10th, 20th, and 30th minutes are overlaid and displayed in the form of a heat map. The trend projection module generates a dynamic change sequence, which continuously displays the environmental change process over the next 30 minutes through the 3D model.

[0117] Implementation Results: This embodiment achieves accurate prediction of temperature distribution over the next 30 minutes, providing a basis for preventative control decisions. Maintenance personnel can intuitively view future thermal changes through a digital twin interface, enabling predictive maintenance.

[0118] Comparative Example 1

[0119] This comparison provides a traditional standalone monitoring system for single cabinets.

[0120] Each rack independently deploys sensors and a local controller, triggering alarms based on fixed thresholds and using a PID algorithm to locally control the air conditioning. Sensor data is stored entirely locally, with no data collaboration mechanism or predictive functionality.

[0121] Comparative Example 2

[0122] This comparison provides a centralized data center monitoring system.

[0123] All sensor data is aggregated via a local area network to a central server for unified processing. The central server performs anomaly detection and rule-based judgment, and centrally issues control commands. It uses a centralized database for storage, lacks a consensus mechanism, and has no predictive capabilities.

[0124] Comparative Example 3

[0125] This comparison provides a monitoring system that only has predictive capabilities.

[0126] A time-series prediction model (such as LSTM) was added to the data collection to predict future temperatures, but the prediction results were only used for alarms and were not linked with the air conditioning control system. The traditional method of controlling the air conditioning was still used.

[0127] Regarding data verification, Comparative Example 1 relies on fixed thresholds or simple filtering, making it prone to false alarms and missed alarms due to sensor drift and noise; Comparative Example 2, while capable of centralized detection, suffers from network latency and cannot adapt to edge environments; Comparative Example 3 employs conventional anomaly detection, lacking adaptive capabilities. Embodiments 1-2 of this invention, through an isolated forest model combined with a drift detection unit, can dynamically identify sensor defects and genuine anomalies, adaptively adjust thresholds, achieve an anomaly detection accuracy of over 98%, and reduce false alarm and missed alarm rates by 90%.

[0128] Regarding storage optimization, Comparative Example 1 uses full storage of raw data, with thousands of server racks generating several TB of data daily, resulting in high storage costs and slow retrieval. Comparative Example 2 uses a centralized database, and query latency reaches minutes after the data volume reaches the PB level. Comparative Example 3 does not perform any special storage optimization. Embodiment 3 of this invention, through a layered design of the first and second storage areas, combined with differential compression, downsampling compression, and log structure merging tree index, achieves a compression ratio of ≥70% and a retrieval latency of ≤50ms, significantly reducing storage and transmission overhead.

[0129] Regarding the collaborative mechanism, Comparative Example 1 and Comparative Example 2 both employ single-point or centralized processing, failing to eliminate the impact of single-rack sensor errors on global judgment. False alarms from a single rack sensor may trigger unnecessary alarms and adjustments across the entire area. Comparative Example 3 also lacks a collaborative mechanism. Embodiment 4 of this invention introduces a lightweight PBFT consensus algorithm and incorporates node reputation evaluation into the consensus process. This enables cross-validation of abnormal data from the rack group, effectively filtering local interference and outputting globally reliable collaborative judgment results.

[0130] Regarding the control method, Comparative Example 1 uses PID control, but the fixed parameters make it difficult to adapt to load changes, and the response lags by 5-10 minutes; Comparative Example 2 is based on simple rule control, which is coarse and cannot cope with complex thermal environments; Comparative Example 3 only predicts and does not automatically control. Example 5 of this invention uses a model predictive control algorithm, which uses the temperature deviation and energy consumption over the next 15 minutes as targets for rolling optimization, and introduces feedback correction to achieve forward-looking and precise control, with a temperature control accuracy of ±0.3℃ and a reduction in air conditioning energy consumption of 18%-25%.

[0131] Regarding predictive capabilities, Comparative Examples 1 and 2 lack predictive functionality and can only passively respond to anomalies that have already occurred. Comparative Example 3, while possessing time-series prediction capabilities, is only used for alarms and is not linked to control systems, thus failing to achieve preventative adjustments. Embodiment 6 of this invention employs a neural network model based on an attention mechanism, capable of learning the thermal correlations between different cabinet locations, predicting the temperature distribution over the next 30 minutes, and dynamically displaying the prediction results in the form of a heatmap through a digital twin layer 3D model, thereby achieving a visualized projection of environmental trends.

[0132] Regarding system closed-loop and operation and maintenance modes, Comparative Example 1 and Comparative Example 2 are both open-loop systems, unable to self-optimize based on the control effect, and the model becomes ineffective after long-term operation; Comparative Example 3 is a semi-open loop, with prediction and control disconnected. This invention uses a feedback iteration module to send the controlled environmental data back to the data governance layer, continuously updating the parameters of the isolated forest model, forming a full-link closed-loop self-iteration of "perception-governance-consensus-control-twin", upgrading the operation and maintenance mode from passive response to proactive prediction and preventive control.

[0133] In summary, this invention has achieved significant breakthroughs over existing technologies in multiple dimensions, including data verification, storage optimization, collaborative mechanisms, control methods, predictive capabilities, and system closed-loop systems. It has solved many technical problems that have long plagued the field and has outstanding substantive features and significant progress.

[0134] This invention provides an intelligent environmental monitoring system for server racks, which can be widely applied to various data centers, cloud computing centers, supercomputing centers, and other scenarios. The system adopts a modular design, allowing for flexible deployment and adaptability to data center environments of different sizes. Through the deep integration of data governance, collaborative consensus, intelligent control, and digital twins, it can significantly improve the intelligence level of data center environment monitoring, reduce operation and maintenance costs, optimize energy consumption, and has good industrial practicality and market promotion value.

[0135] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0136] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A server rack intelligent environment monitoring system, characterized in that, include: The edge sensing layer, deployed in the server rack, includes multimodal sensors for collecting environmental time-series data from the rack. The data governance layer, connected to the edge perception layer, includes a data verification module, a compression and storage module, and a feedback and iteration module. The data verification module is used to perform anomaly detection on the environmental time-series data using an isolated forest model, distinguishing between sensor dead data, abnormal data, and normal data. The compression and storage module is used to perform layered compression on the detected data and build an index. The feedback and iteration module is used to obtain the environmental time-series data after the air conditioning control command is executed, and input the executed environmental time-series data into the data verification module to update the detection parameters of the isolated forest model. The collaborative analysis layer, connected to the data governance layer, includes a consensus module; the consensus module is used to cross-validate the abnormal data among the nodes of the rack cluster using the PBFT consensus algorithm and output a collaborative judgment result. The intelligent control layer, connected to the collaborative judgment layer and the air conditioning system, includes a control module; the control module is used to generate and issue air conditioning control commands based on the collaborative judgment results using a model predictive control algorithm. The digital twin layer, connected to the data governance layer, collaborative analysis layer, and intelligent control layer, includes a 3D model and a prediction module. The prediction module is used to predict environmental change trends based on historical and real-time data. The 3D model is used to map the data center environment status and display the prediction results of the environmental change trends.

2. The server rack intelligent environment monitoring system as described in claim 1, characterized in that, The isolated forest model in the data verification module includes a drift detection unit; the drift detection unit is used to monitor the distribution shift of environmental time series data and update the model parameters of the isolated forest model when a distribution shift is detected; the distribution shift is determined by a concept drift detection algorithm.

3. The server rack intelligent environment monitoring system as described in claim 1, characterized in that, The compressed storage module includes: The first storage area is used to store normal data within a preset first time period; The second storage area is used to store normal data that exceeds the preset first time period after differential compression or downsampling compression, and to store a complete copy of the data marked as abnormal. An index unit is used to build an index based on a log structure merge tree, the index including a hash index or an inverted index.

4. A server rack intelligent environment monitoring system as described in claim 3, characterized in that, The compressed storage module further includes a baseline value update unit; the baseline value update unit is used to update the baseline value of differential compression according to a preset period, the preset period being determined based on the rack load fluctuation characteristics; the rack load fluctuation characteristics include the periodic change pattern of CPU utilization or network throughput obtained through frequency domain analysis.

5. A server rack intelligent environment monitoring system as described in claim 1, characterized in that, The consensus module further includes a reputation assessment unit; the reputation assessment unit is used to calculate a reputation value based on the online duration, consensus participation, or false alarm rate of the rack nodes; the PBFT consensus algorithm sets weights for node voting based on the reputation value; the reputation value is calculated by weighted summation, and the weight coefficients of the weighted summation include online duration weight, consensus participation weight, and false alarm rate weight, wherein the false alarm rate weight is greater than the online duration weight and the consensus participation weight.

6. The server rack intelligent environment monitoring system as described in claim 1, characterized in that, The model prediction control algorithm in the control module is used for: The objective function is constructed based on the temperature deviation and energy consumption within a preset time domain. Solve the control sequence under preset constraints; and The first control command in the control sequence is sent to the air conditioning system as the current control command. The objective function is to minimize the sum of the squares of the differences between the predicted temperature and the target temperature and the sum of the estimated energy consumption.

7. A server rack intelligent environment monitoring system as claimed in claim 6, characterized in that, The control module further includes a feedback correction unit; the feedback correction unit is used to acquire the controlled environmental time series data, compare the controlled environmental time series data with the predicted values ​​generated by the model predictive control algorithm, and correct the state variables or model parameters of the predictive model of the model predictive control algorithm according to the comparison results.

8. A server rack intelligent environment monitoring system as claimed in claim 1, characterized in that, The prediction module employs a neural network model based on an attention mechanism; the neural network model based on an attention mechanism is used to determine the thermal correlation between different cabinet locations and predict the temperature distribution within a preset second time period.

9. A server rack intelligent environment monitoring system as claimed in claim 8, characterized in that, The digital twin layer is also used to input the prediction results of the prediction module into the three-dimensional model to generate and display a temperature distribution map within a preset third time period; the preset third time period is at least one preset moment or continuous time interval within the preset second time period; the temperature distribution map is a heat map superimposed on the three-dimensional model.

10. A server rack intelligent environment monitoring system as claimed in claim 1, characterized in that, The digital twin layer also includes a trend projection module; the trend projection module is used to generate a dynamic change sequence based on the prediction results of the prediction module, and continuously display the environmental change process within the preset second time period through the three-dimensional model.

Images