Search

CN-121522104-B - Intelligent data center air quality early warning method and system based on sensor network

CN121522104BCN 121522104 BCN121522104 BCN 121522104BCN-121522104-B

Abstract

The invention relates to the technical field of sensor networks and Internet of things, in particular to a data center air quality intelligent early warning method and system based on a sensor network; the method comprises the steps of constructing a self-organizing collaborative network by deploying multi-type sensor nodes in a key area of a data center, collecting and correcting multidimensional air data in real time, improving data reliability by means of dynamic weight fusion and local anomaly identification among the nodes, establishing an air parameter and machine room structure association model by means of space mapping and time sequence analysis, identifying microscale diffusion trend, constructing a self-adapting dynamic threshold mechanism, combining historical statistics and real-time feedback to achieve threshold rolling optimization, generating a hierarchical alarm strategy based on multilevel early warning judgment and triggering conditions, and continuously and automatically optimizing early warning precision and response efficiency by means of closed loop feedback. The invention realizes the omnibearing, intelligent and highly reliable early warning and regulation of the air quality of the data center.

Inventors

  • SHEN LI
  • ZHANG ZHONGPING

Assignees

  • 北京致控远科技有限公司

Dates

Publication Date
20260508
Application Date
20251202

Claims (8)

  1. 1. The intelligent data center air quality early warning method based on the sensor network is characterized by comprising the following specific implementation steps of: S1, deploying multi-type sensor nodes in a key area of a data center, constructing a self-organizing collaborative network, collecting multi-dimensional air data in real time, and carrying out node end correction and local anomaly identification on the collected data; S2, improving data reliability through inter-node dynamic weight fusion and local anomaly identification, and establishing an air parameter and machine room structure association model by utilizing space mapping and time sequence analysis, wherein the identification of the microscale diffusion trend specifically comprises the following steps: mapping correction parameters of each node to cabinet coordinates, an air supply and exhaust path and a cold channel structure of a data center, and constructing a dynamic corresponding frame of a multidimensional air state and a spatial layout, wherein the dynamic corresponding frame specifically comprises the following steps: A1, mapping correction parameters of each node to cabinet coordinates, an air supply and exhaust path and a cold channel structure of a data center, and constructing a dynamic corresponding frame of a multidimensional air state and a spatial layout, wherein the dynamic corresponding frame specifically comprises the following steps: a2, taking out the corrected multidimensional vector and the local anomaly flag from the node data packet, and calculating a node and neighborhood consistency residual error; a3, calculating an initial credibility score, and performing local smoothing by using the adjacency right between nodes to obtain a final credibility score; forming an observation weight matrix based on the adjacency weights and the credibility among the nodes, and carrying out space and time sequence analysis by combining drawing Laplace smoothing and time sequence smoothing; analyzing the change rate, change direction and local disturbance persistence of the air parameters in a continuous time window sliding mode, identifying microscale diffusion or aggregation trend in a space distribution frame, The method comprises the following steps: b1, based on inter-node adjacency rights Combining with credibility to form an observation weight matrix : ; ; B2, definition map Laplace Solving a space smoothing minimization problem: ; B3, introducing an exponential weighting history item to each grid point, and performing time sequence smoothing: ; Wherein, the Representing an observation weight matrix, namely a diagonal matrix, L representing a graph Laplace matrix, D representing a degree matrix, and the diagonal matrix, the elements of which are defined as I.e. the sum of the connected weights of node i and all neighbors, W represents a weighted adjacency matrix between nodes, each element being ; Representing a spatially smoothed intensity coefficient, controlling a trade-off between an observation fit term and a graph smoothing term; Representing vectorized observation data; A spatial estimation solution vector representing a graph laplace smoothing; Representing time smoothing coefficients for estimating the current time instant History estimation Combining; indicating that node i is at time The confidence score of the neighborhood smoothed; Coupling evaluation is carried out on external factors of cabinet thermal load change, air duct resistance and adjacent region equipment operation fluctuation and air quality trend by constructing a trans-regional factor chain, and potential driving forces of abnormal formation and diffusion are quantified, specifically: And C1, dividing the nodes into three types according to the credibility and the original label: Trusted set : ; Suspicious set : ; Low trust set : ; C2, for each sensor Calculating a difference from the neighborhood smoothing estimation: ; If it is Judging that the real burst is abnormal, otherwise, judging that the sensor is abnormal or noise; c3, for each spatial point, using a weighted Huber loss to calculate a fusion value z: ; ; The implementation adopts iterative weighted least square solution, and the initial value of the weight is used ; C4, if a certain i is judged to be a real burst, the fusion result at the point allows higher sensitivity; Wherein, the And Representing a set confidence threshold; a gap metric representing the node observations and the fusion estimate; Representing adaptive consistency thresholds for comparison Whether or not it is significant; z represents a fusion estimate, i.e. a robust fusion value obtained by minimizing the Huber loss for observation at a certain spatial or lattice point; For switching points, i.e. threshold values, when residual errors Using a quadratic penalty, otherwise linearly growing; representing an adaptive threshold, dependent on a historical fluctuation standard deviation; A local abnormality flag, 1 for abnormality, 0 for normal; based on disturbance trend and factor risk quantification result, outputting comprehensive credibility and possible cause of air quality abnormality by fusing probability model; s3, constructing a self-adaptive dynamic threshold mechanism, and realizing threshold rolling optimization by combining historical statistics and real-time feedback; And S4, generating a hierarchical alarm strategy based on the multi-stage early warning judgment and the triggering condition, and continuously and self-optimizing early warning precision and response efficiency through closed loop feedback.
  2. 2. The intelligent data center air quality early warning method based on the sensor network according to claim 1, wherein the step S1 specifically comprises: a temperature and humidity sensor, a particulate matter sensor, a gas concentration sensor and a wind speed and wind direction sensor are arranged in a machine room cold channel, a machine room hot channel, a machine cabinet channel, an air conditioner air outlet and an underfloor return air area; Constructing a single-node air state vector, wherein the vector comprises ambient temperature, relative humidity, particle concentration, gas concentration, air flow speed and wind direction; Constructing a node self-organizing and cooperative network, forming a self-organizing network by each node through low-power wireless communication, dynamically calculating cooperative weights according to the distance between the nodes and the health state, and carrying out local data fusion and abnormal primary identification; The node collects air state vectors according to the self-adaptive sampling frequency, and adds a unique identifier and a time stamp, and a key area is collected at high frequency and is complemented by adjacent nodes; And the acquired data are subjected to drift correction, filtering, delay compensation and wind flow smoothing at the node end, and meanwhile, the preliminary abnormality early warning of the node end is supported.
  3. 3. The intelligent data center air quality early warning method based on the sensor network according to claim 2, wherein the construction of the node self-organizing and collaboration network comprises the following steps: Before each sensor node cooperates with the adjacent nodes in real time, calculating a cooperation weight, wherein the weight comprehensively considers the distance between the nodes, the health state of the nodes and the reliability of the data; Each node selects an optimal neighbor set according to the cooperative weight, and performs weighted fusion on the data of the neighbor nodes to generate a local cooperative air state vector; The network performs dynamic topology maintenance according to the health state of the nodes and the environmental change, including failure node rejection, new node addition self-adaptive reconstruction neighbor set and weight.
  4. 4. The intelligent data center air quality early warning method based on the sensor network according to claim 3, wherein the step S3 specifically comprises: calculating the average value and standard deviation of the areas according to the historical fusion index sequence and the abnormal event, and generating an initial threshold value of each area through weighted combination; The contribution weight of each component to the threshold value is corrected in real time by utilizing the uncertainty of the region, and the stability is ensured by combining the historical weight; according to the adjusted weight and the fusion index, rolling and updating the threshold value, and amplifying and correcting the emergency; and counting historical false alarm and false alarm, and adjusting the rolling update rate and the safety coefficient through false alarm and false alarm indexes to realize closed loop optimization.
  5. 5. The intelligent early warning method for the air quality of the data center based on the sensor network according to claim 4 is characterized in that the average value and the standard deviation of the areas are calculated according to the historical fusion index sequence and the abnormal event, and the initial threshold value of each area is generated through weighted combination, and the method specifically comprises the following steps: Selecting a fusion index sequence and a corresponding abnormal event label in a history period, and calculating statistical distribution, including mean value and standard deviation, of each region; defining an initial threshold value based on a weighted combination of the mean value and the standard deviation, and considering a safety coefficient; and generating an initial threshold value for the multidimensional component by adopting a weighted combination mode, wherein the component weight is based on historical statistics and risk correlation.
  6. 6. The intelligent data center air quality early warning method based on the sensor network according to claim 5, wherein the step S4 specifically comprises: the environmental state is divided into normal, mild, moderate and severe early warning grades by comparing the regional index with a dynamic threshold and combining with an uncertainty correction rule; By setting a time accumulation coefficient and a neighborhood synergistic factor, the continuity and the space consistency of the early warning trigger are judged, and false alarm caused by short-time fluctuation or local misinformation is avoided; Matching execution actions of different levels according to the early warning level and the triggering result, and dynamically adjusting the strategy strength according to the regional importance and the real-time state; Based on the actual alarm effect, the false alarm and missing alarm condition and the air quality recovery condition, the threshold interval, the triggering parameter and the alarm priority are automatically optimized.
  7. 7. The intelligent data center air quality early warning method based on the sensor network according to claim 6, wherein the judgment of persistence and spatial consistency of early warning triggering is performed by setting a time accumulation coefficient and a neighborhood synergistic factor, specifically comprising: introducing a time accumulation factor, and calculating the proportion of the past sliding window meeting the threshold exceeding condition; introducing a neighborhood synergistic factor, and calculating the proportion of the existence of the super-threshold region in the neighborhood at the current moment; and setting a triggering condition by combining time accumulation and neighborhood cooperation, and triggering early warning when the continuous judgment or neighborhood resonance meets the requirement.
  8. 8. A data center air quality intelligent early warning system based on a sensor network, which is used for executing the data center air quality intelligent early warning method based on the sensor network as claimed in any one of claims 1 to 7, and is characterized by comprising the following steps: the sensor node and self-organizing network module is responsible for actual physical perception and reliable acquisition, comprises a multi-type sensor node and a low-power wireless self-organizing gateway, wherein the node has the functions of local self-detection, health scoring, positioning identification, adjustable sampling rate and node end correction, and supports inter-node collaborative weight calculation and redundant acquisition strategies; The edge preprocessing and multidimensional fusion module is deployed on an edge server, receives the corrected multidimensional state and the local anomaly marker, completes reliability assessment, neighborhood reliability propagation, graph time sequence weighted interpolation and robust fusion, and outputs fusion estimation and region level indexes which are continuous in space and have uncertainty quantification; The self-adaptive threshold and prediction decision module builds and evolves a dynamic threshold by using a regional index, uncertainty and historical event knowledge base, and comprises a component weight self-adaptive device, a threshold rolling updater, a burst amplifying and buffering strategy and a short-term trend prediction engine based on a time sequence model, so as to realize verification feedback and online learning mechanism of the threshold; The multi-stage early warning and intelligent warning executing module is responsible for converting threshold decision into executable operation and maintenance actions, and comprises an early warning grade judging device, a time accumulation and neighborhood cooperative trigger, a warning strategy library, a warning issuing and executing interface and a closed loop feedback collecting unit.

Description

Intelligent data center air quality early warning method and system based on sensor network Technical Field The invention relates to the technical field of sensor networks and Internet of things, in particular to a data center air quality intelligent early warning method and system based on a sensor network. Background In recent years, as the power density of a server is continuously improved, the non-uniformity of cabinet-level microenvironment airflow and pollutant distribution is increasingly remarkable, and higher requirements are put on the refinement and instantaneity of air quality monitoring. The invention discloses a real-time intelligent indoor air quality monitoring system, which is disclosed in China patent application with publication number of CN109596492A and comprises a sensor sensing subsystem, a computer analysis terminal and an intelligent terminal, wherein the sensor sensing subsystem is used for collecting indoor air quality information and comprises a wireless sensor network constructed by a sink node and a plurality of sensor nodes deployed in the indoor, the sensor nodes collect air quality information of monitoring positions, the sink node is mainly used for converging the air quality information collected by each sensor node and sending the air quality information to the computer analysis terminal, and the computer analysis terminal is connected with the intelligent terminals so as to send the received air quality information to the intelligent terminals. Meanwhile, with the progress of sensor technology, edge calculation and intelligent analysis methods, the construction of an air quality early warning system capable of sensing in real time, intelligent analysis and autonomous decision-making has become an important research direction for improving the operation and maintenance intelligence level of a data center. Disclosure of Invention The invention aims to solve the problems in the background art and provides a data center air quality intelligent early warning method and system based on a sensor network. The technical scheme of the invention is that the intelligent data center air quality early warning method based on the sensor network comprises the following concrete implementation steps: S1, deploying multi-type sensor nodes in a key area of a data center, constructing a self-organizing collaborative network, collecting multi-dimensional air data in real time, and carrying out node end correction and local anomaly identification on the collected data; S2, improving data reliability through dynamic weight fusion and local anomaly identification among nodes, establishing an air parameter and machine room structure association model through space mapping and time sequence analysis, and identifying microscale diffusion trend; s3, constructing a self-adaptive dynamic threshold mechanism, and realizing threshold rolling optimization by combining historical statistics and real-time feedback; And S4, generating a hierarchical alarm strategy based on the multi-stage early warning judgment and the triggering condition, and continuously and self-optimizing early warning precision and response efficiency through closed loop feedback. Preferably, step S1 specifically includes: a temperature and humidity sensor, a particulate matter sensor, a gas concentration sensor and a wind speed and wind direction sensor are arranged in a machine room cold channel, a machine room hot channel, a machine cabinet channel, an air conditioner air outlet and an underfloor return air area; Constructing a single-node air state vector, wherein the vector comprises ambient temperature, relative humidity, particle concentration, gas concentration, air flow speed and wind direction; Constructing a node self-organizing and cooperative network, forming a self-organizing network by each node through low-power wireless communication, dynamically calculating cooperative weights according to the distance between the nodes and the health state, and carrying out local data fusion and abnormal primary identification; The node collects air state vectors according to the self-adaptive sampling frequency, and adds a unique identifier and a time stamp, and a key area is collected at high frequency and is complemented by adjacent nodes; And the acquired data are subjected to drift correction, filtering, delay compensation and wind flow smoothing at the node end, and meanwhile, the preliminary abnormality early warning of the node end is supported. Preferably, the node self-organizing and cooperative network is constructed, which specifically comprises: Before each sensor node cooperates with the adjacent nodes in real time, calculating a cooperation weight, wherein the weight comprehensively considers the distance between the nodes, the health state of the nodes and the reliability of the data; Each node selects an optimal neighbor set according to the cooperative weight, and performs weighted fusion on the data of the neighb