Server inspection method and system based on IT operation and maintenance platform

Abstract

The invention relates to the technical field of equipment operation and maintenance, and discloses a server inspection method and system based on an IT operation and maintenance platform, and the method comprises the steps: collecting an original environment parameter data set, and generating a continuous environment parameter distribution diagram; identifying a temperature and humidity abnormal region by using K-means clustering, calculating a variation gradient and mapping the variation gradient into variability intensity, and extracting high-intensity point locations for weighted fusion to obtain a priority inspection target point set; searching a low-cost path in an environment grid map fusing obstacles, terrains and airflow retardation by using an A-Star algorithm, comparing environment parameter fluctuation difference values in real time, re-clustering overproof nodes, updating the map and connection cost, and regenerating an inspection path; and fusing variation gradient data to evaluate coverage, filling path gaps through linear interpolation and dynamically adjusting filling parameters until a preset coverage evaluation rule is met, and finally forming an optimized inspection path. According to the method, the problem of insufficient routing inspection coverage integrity in the prior art can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment operation and maintenance, and in particular to a server inspection method and system of an IT operation and maintenance platform. BACKGROUND

[0002] At present, with the rapid development of industrial data mining and big data analysis technology, as the core carrier of high-density computing environment, the internal temperature and humidity, airflow and other environmental parameters of the server room show significant three-dimensional spatial heterogeneity. In order to realize the reliable operation of the equipment, high-frequency and full-coverage environmental inspection of the room space is required to identify potential abnormal areas such as local overheating and airflow blockage in time, which puts higher technical requirements on the dynamic and accurate planning of the inspection path.

[0003] In one prior art, the server room inspection path usually adopts manual experience or static preset route. First, a number of risk points are determined according to the discrete monitoring results of the fixed sensor array, and then a one-time inspection route is generated by using the shortest path algorithm and is periodically repeated. For example, when the temperature of a certain fixed sensor exceeds the limit, the corresponding coordinates are added to the inspection list, and the inspection is performed in turn according to the predetermined order. However, this process does not integrate the spatial gradient changes of the three-dimensional airflow field and real-time environmental parameters of the server room, resulting in that the newly added hot spots or airflow dead angles cannot be timely included in the inspection range, and the path coverage degree decreases rapidly with the change of the environment.

[0004] Due to the lack of spatial gradient coordination mechanism between the static path and the real-time environmental parameters, the prior art cannot dynamically compensate for the inspection coverage gap when the environmental parameters change suddenly, resulting in an increase in the missing inspection rate of high-risk areas. Therefore, there is a problem of insufficient inspection coverage integrity in the prior art. SUMMARY

[0005] The present application provides a server inspection method and device based on an IT operation and maintenance platform, electronic equipment and storage medium, to solve the problem of insufficient inspection coverage integrity in the prior art.

[0006] In the first aspect, in order to solve the above technical problems, the present application provides a server inspection method based on an IT operation and maintenance platform, comprising: obtaining an original environmental parameter data set; performing grid interpolation processing according to the original environmental parameter data set to obtain an environmental parameter distribution map; performing K-means clustering processing according to the environmental parameter distribution map to obtain a high-risk area list; performing variation gradient calculation according to the high-risk area list to obtain a variation intensity map; performing high-intensity point extraction and weighted average fusion according to the variation intensity map to obtain a priority inspection target point set; According to the priority inspection target point set, an A-Star algorithm is adopted for path planning and path cost calculation to obtain an initial inspection path sequence; According to the initial inspection path sequence and the original environment parameter data set, path node dynamic adjustment is performed to obtain an updated inspection path; According to the updated inspection path and the mutation gradient data of the mutation intensity mapping, path coverage preliminary evaluation is performed to obtain a path coverage preliminary evaluation result; According to the path coverage preliminary evaluation result, linear interpolation filling is performed to obtain a final optimized inspection path.

[0007] In a second aspect, the present application provides a server inspection device based on an IT operation and maintenance platform, comprising: A data acquisition module is configured to acquire an original environment parameter data set; A parameter distribution map generation module is configured to perform grid interpolation processing according to the original environment parameter data set to obtain an environment parameter distribution map; A high-risk area division module is configured to perform K-means clustering processing according to the environment parameter distribution map to obtain a high-risk area list; A mutation gradient calculation module is configured to perform mutation gradient calculation according to the high-risk area list to obtain a mutation intensity mapping; A priority inspection target determination module is configured to perform high-intensity point extraction and weighted average fusion according to the mutation intensity mapping to obtain a priority inspection target point set; An initial path planning module is configured to perform path planning and path cost calculation according to the priority inspection target point set by adopting an A-Star algorithm to obtain an initial inspection path sequence; A path updating module is configured to perform path node dynamic adjustment according to the initial inspection path sequence and the original environment parameter data set to obtain an updated inspection path; A path coverage evaluation module is configured to perform path coverage preliminary evaluation according to the updated inspection path and the mutation gradient data of the mutation intensity mapping to obtain a path coverage preliminary evaluation result; A final path determination module is configured to perform linear interpolation filling according to the path coverage preliminary evaluation result to obtain a final optimized inspection path.

[0008] In a third aspect, the present application further provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the server inspection method based on the IT operation and maintenance platform according to any one of the above aspects when executing the computer program.

[0009] In a fourth aspect, the present application also provides a computer readable storage medium comprising a stored computer program, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform any of the above-mentioned server inspection methods based on an IT operation and maintenance platform.

[0010] Compared with the prior art, the present application has the following beneficial effects: (1) The present application performs gridding interpolation and Kriging anomaly correction on the original environmental parameter data set returned by the discrete measurement points, forming a continuous and smooth environmental parameter distribution map, so that the temperature, humidity and air flow speed remain complete and consistent in the three-dimensional space, providing reliable input for subsequent clustering and gradient calculation, and improving the robustness of high-risk area identification.

[0011] (2) The present application performs K-means clustering based on the environmental parameter distribution map, marks abnormal clusters in combination with preset temperature and humidity thresholds, and then generates a high-risk area list through structured storage, so that the spatial range and attribute information of the abnormal area are synchronously presented, making it easy for managers to intuitively grasp the hotspot distribution and provide a basis for dynamic scheduling.

[0012] (3) The present application constructs a digital machine room model inside the high-risk area and integrates real-time airflow simulation, calculates the temperature and humidity variation gradient using the finite volume method, forms a variation intensity map, and makes the influence of airflow resistance on temperature and humidity distribution be included in the risk assessment, thereby enhancing the adaptability to potential hazards in complex airflow environment.

[0013] (4) The present application obtains a set of priority inspection target points through high-intensity point extraction and weighted average fusion, and then searches for a priority path in an environmental grid map that fuses obstacle, terrain and airflow resistance data using the A-Star algorithm, so that the path cost reflects both spatial distance and ventilation resistance, reducing the moving time and energy consumption in the inspection process.

[0014] (5) The present application compares the real-time sensor feedback data with the original environmental parameters to obtain the environmental parameter fluctuation difference, re-clusters the nodes that exceed the standard and recalculates the connection cost locally, and then generates a new inspection path sequence according to a preset sorting rule, so that the path can be adjusted in real time as the environment changes, improving the timeliness and flexibility of the inspection path.

[0015] (6) The present application fuses the predicted value of the real-time environmental parameter and the variation gradient data, extracts high-intensity points again and calculates the coverage rate, fills the gaps in the path through linear interpolation and dynamically adjusts the filling parameters until the preset coverage evaluation rule is met, forming the final optimized inspection path, so that the high-risk points are continuously included in the inspection field, and the coverage blind area problem caused by environmental mutation is alleviated. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a server inspection method flow diagram based on an IT operation and maintenance platform provided by the first embodiment of the present application; Figure 2 is a server inspection system structure diagram based on an IT operation and maintenance platform provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0018] With reference to Figure 1 The first embodiment of the present application provides a server inspection method based on an IT operation and maintenance platform, comprising the following steps: S11, obtaining an original environment parameter data set, wherein the original environment parameter data set contains original environment parameters; S12, performing grid interpolation processing according to the original environment parameter data set to obtain an environment parameter distribution map; S13, performing K-means clustering processing according to the environment parameter distribution map to obtain a high-risk area list; S14, performing mutation gradient calculation according to the high-risk area list to obtain a mutation intensity map; S15, performing high-intensity point extraction and weighted average fusion according to the mutation intensity map to obtain a priority inspection target point set; S16, performing path planning and path cost calculation using an A-Star algorithm according to the priority inspection target point set to obtain an initial inspection path sequence; S17, performing path node dynamic adjustment according to the initial inspection path sequence and the original environment parameter data set to obtain an updated inspection path; S18, performing path coverage preliminary evaluation according to the updated inspection path and the mutation gradient data of the mutation intensity map to obtain a path coverage preliminary evaluation result; S19, performing linear interpolation filling according to the path coverage preliminary evaluation result to obtain a final optimized inspection path.

[0019] In step S11, the original environment parameter data set is obtained.

[0020] It is worth mentioning that the system obtains three types of data in parallel through the sensor network in the three-dimensional space of the machine room, the BIM interface and the dynamic environment monitoring interface, as the original input for subsequent calculation. The original environmental parameter data set is composed of discrete measurement points, including temperature, humidity, air flow speed and air pressure difference, each with information such as sampling time; the collection frequency is temperature and humidity every 30 seconds, air flow speed and air pressure difference every 10 seconds.

[0021] In step S12, according to the original environmental parameter data set, a gridding interpolation process is performed to obtain an environmental parameter distribution map, including: Performing a gridding interpolation algorithm on the original environmental parameter data set to generate spatially continuous distribution data; When the spatially continuous distribution data has an abnormal value, the Kriging interpolation method is used to correct the abnormal value to obtain smooth distribution data; According to the smooth distribution data, an environmental parameter distribution map in the three-dimensional space is generated.

[0022] It is worth mentioning that the original environmental parameter data set is composed of discrete measurement points returned by temperature, humidity and air flow speed sensors arranged in the three-dimensional space of the machine room every minute, each measurement point with x, y, z coordinates and a time stamp. The system takes these discrete points as the source, performs a gridding interpolation algorithm, and outputs spatially continuous distribution data covering the entire machine room volume. The method uses three-dimensional inverse distance weighted interpolation to first establish a regular grid with a step size of 0.5m in the outer range of the machine room, with the total number of nodes equal to the product of the number of long, wide and high grids; for any node, collect adjacent measurement points within a search radius of 2.5m, if there are less than 5 valid measurement points, the radius is expanded to 3.5m, and if there are still less than 5, mark the node as a to-be-supplemented empty value. The interpolation weight is calculated by the inverse square of the distance, that is: Where wi is the calculated interpolation weight of measurement point i, di is the Euclidean distance value (in meters) from measurement point i to the node, wi is dimensionless; the node estimated value is equal to the sum of the product of each measurement value and wi divided by the sum of wi, temperature and humidity are directly substituted, air pressure difference is directly substituted as a scalar; air flow speed is first taken as a scalar rate by taking its vector module, and then interpolated, thus obtaining the continuous estimated value of each node to form the spatially continuous distribution data.

[0023] When the spatial continuous distribution data has outliers, the Kriging interpolation method is used to correct the outliers to obtain smooth distribution data. The input is the spatial continuous distribution data, and the output is the smooth distribution data. The method is as follows: (take temperature value as an example, and the rest of the data are processed similarly), first, traverse all grid nodes. If the absolute value of the difference between the temperature value of a node and the average value of the nodes in the same depth neighborhood is greater than a set threshold, the node is determined to be an outlier. The threshold setting method is that the system calculates the average value μ and the standard deviation σ of the temperature values of all nodes in the same depth layer, and sets the threshold as μ ± 3σ. When the absolute value of the difference between the temperature value of a node and μ is greater than 3σ, the node is determined to be an outlier. The dimension remains the same as the temperature; a search window is established within 1m in the horizontal and vertical directions with the abnormal node as the center, all effective nodes within the window are collected as sample points, the semi-variance between the sample points and the abnormal node is calculated, and the range, sill value and nugget value are fitted according to the spherical model to form the variogram function; the weight coefficient of each sample point to the estimated abnormal node is calculated using the variogram function, the weight satisfies the unbiasedness and the minimum estimation variance condition, the weight and the sample observation value are linearly combined to obtain the corrected value of the abnormal node, and the unit remains the same as the original dimension; the corrected value replaces the abnormal value, and the smooth distribution data is written into the spatial continuous distribution data to form the smooth distribution data.

[0024] The smooth distribution data includes continuous estimated values of temperature, humidity and air flow velocity of each grid node and corresponding three-dimensional coordinates. The system generates an environmental parameter distribution map in three-dimensional space based on this, and outputs a visual and queryable field map. First, a regular grid consistent with the outline of the machine room is established, with a node spacing of 0.5 meters. Each node stores three physical quantities: temperature, humidity and air flow velocity, forming a multi-component volume data; then each component is assigned a color scale. The temperature uses a blue-green-yellow-red gradient color band, the humidity uses a green-yellow-red gradient color band, and the air flow velocity uses a gray scale. The two ends of the color band correspond to the minimum and maximum values of the physical quantity, respectively. The node value is linearly mapped to the RGB component and written into the voxel attribute. At the same time, isosurfaces are generated. Temperature is extracted every 2°C, humidity is extracted every 5% RH, and air flow velocity is extracted every 0.1m / s. The isosurfaces are expressed as a collection of triangles for fast rendering and cross-section observation. To support slice browsing, the system provides three orthogonal cross-sections: XY, XZ and YZ. The user inputs any coordinate axis position, and the color cloud chart on the cross-section is displayed immediately. The cross-section data is sampled from the voxel through trilinear interpolation to ensure continuous and no step.

[0025] In step S13, according to the environmental parameter distribution map, K-means clustering processing is performed to obtain a high-risk area list, including: Temperature and humidity data are extracted from the environmental parameter distribution map, and after data cleaning and formatting, a preprocessed data set is obtained. K-means clustering is performed on the preprocessed dataset, and the clusters are divided into multiple clusters according to temperature and humidity characteristics in three-dimensional space to generate a regional grouping dataset. Clusters in the regional grouping dataset whose temperature exceeds a preset temperature threshold or whose humidity value exceeds a preset humidity threshold are marked as abnormal regions to form an abnormal region dataset. The abnormal region dataset is stored in a structured manner to generate a high-risk region list containing coordinates and corresponding temperature and humidity information.

[0026] The environmental parameter distribution map is generated by color mapping and isosurface extraction from smooth distribution data, containing node temperature, humidity, air flow speed, and three-dimensional coordinates. The system extracts temperature and humidity scalar fields from the map, and outputs the preprocessed dataset for K-means clustering after data cleaning and formatting. Specifically, first, read the three-dimensional grid definition of the environmental parameter distribution map, which is composed of regularly arranged cubic voxels with a side length of 0.5 meters, covering the entire space of the machine room. Define each voxel unit as an initial spatial region, which is uniquely identified by the three-dimensional coordinates of its center point. The collection of all voxel units constitutes the spatial region division dataset. Then, traverse each voxel unit in the spatial region division dataset, and based on the center point coordinates of the voxel, retrieve and extract the temperature and humidity values of the corresponding grid nodes from the environmental parameter distribution map, discard the air flow speed, and retain the x, y, and z coordinates to form an original extraction table containing node number, coordinates, temperature, and humidity, with units of °C and percentage, and consistent dimensions with the sensor. In the data cleaning stage, perform 3σ test on the temperature column, and replace the outliers with the median value of the eight neighbors in the same layer. Perform boundary test on the humidity column, and truncate the outliers to 0 or 100. Missing nodes are filled with bilinear interpolation in the same layer. In the formatting stage, temperature and humidity are kept to one decimal place, coordinates are kept to three decimal places, all fields are converted to floating point type, a timestamp field is added to record the time of extraction, the table is sorted in ascending order of node number, and a two-dimensional table is formed, with each row representing a spatial point and each column representing a temperature or humidity value.

[0027] K-means clustering is adopted on the pretreated dataset. When clustering, only clusters are divided in the two-dimensional feature space of temperature and humidity, and the specific operation of eliminating the dimension effect is as follows: first, the average value μT and the standard deviation σT of the temperature column are calculated, and the average value μH and the standard deviation σH of the humidity column are calculated. The temperature value of each node is divided by σT after being subtracted by μT, and the humidity value is divided by σH after being subtracted by μH, to obtain the dimensionless standardized temperature and standardized humidity. Then, the standardized temperature and standardized humidity form a two-dimensional feature vector, the kmeans++ strategy is used to select the initial cluster center, the number of clusters k is set based on the clustering target, k=4 in this embodiment, corresponding to four typical abnormal types of high temperature and high humidity, high temperature and low humidity, low temperature and high humidity, and low temperature and low humidity; when iterating, the Euclidean distance of the node to the cluster center is calculated, the node is assigned to the nearest cluster, and the standardized temperature mean and the standardized humidity mean in the cluster are taken as the new cluster center coordinates when updating the cluster center. Repeat until the cluster center change amount meets the temperature less than 0.1 and the humidity less than 0.1 at the same time, or the maximum iteration number (such as 200 times) is reached, and the iteration is terminated; the system adds a cluster label to each node, the label value is 0 to k-1, forms a regional grouping dataset, retains the original coordinates and temperature and humidity values, and adds a cluster number field, which is used for subsequent threshold comparison.

[0028] In an implementation manner, forming the abnormal region dataset can be that the system reads the regional grouping dataset cluster by cluster, first calculates the arithmetic mean of the temperature of all nodes in the same cluster to obtain the cluster average temperature, then calculates the arithmetic mean of the humidity of all nodes in the same cluster to obtain the cluster average humidity, and the preset temperature threshold and the preset humidity threshold are input by the user during system initialization. In this embodiment, the preset temperature threshold is set to 28°C, and the preset humidity threshold is set to 70%RH. The setting is based on the upper limit value allowed by the equipment inlet side in the computer room design specification, and the dimension is consistent with the node record. The comparison logic adopts an or condition, that is, the cluster average temperature is greater than the preset temperature threshold, or the cluster average humidity is greater than the preset humidity threshold. If any condition is met, the cluster as a whole is marked as an abnormal region to avoid missing local overheating or high-humidity corners. The marking method is to append an abnormal flag field at the end of the regional grouping dataset, the abnormal flag is set to 1, and the normal cluster is set to 0, to form the abnormal region dataset. The dataset retains the original node coordinates, temperature, humidity, and cluster label, and adds an abnormal flag, which is used for subsequent structured storage and high-risk region list generation.

[0029] The system performs structured storage on the abnormal area data set to generate a high-risk area list containing coordinates and corresponding temperature and humidity information. First, the system traverses the abnormal area data set, filters all records with abnormal flag equal to 1, reads node number, three-dimensional coordinates, temperature, and humidity row by row, writes the coordinates of all nodes in the same abnormal cluster into a record, and appends the average temperature and average humidity of the cluster to form a row entry. The entry field order is cluster number, node coordinate sequence, average temperature, and average humidity. Fixed delimiters are used between fields, and commas are used to separate points in the coordinate sequence. The xyz three components are separated by spaces to form a text line. The storage medium is a local relational database table, the table name is associated with the inspection date, and the cluster number is set as the primary key to ensure that there is only one record for the same cluster.

[0030] In step S14, according to the high-risk area list, the variation gradient calculation is performed to obtain the variation intensity map, including: According to the high-risk area list, the abnormal area extraction is performed to obtain the to-be-calculated area; According to the to-be-calculated area, the model is constructed to obtain the digital machine room model, and the digital machine room model contains temperature distribution and humidity distribution; According to the digital machine room model, the air flow rate scalar field in the smooth distribution data is extracted to form an air flow distribution data set; According to the temperature and humidity scalar fields in the digital machine room model, the variation gradient between adjacent voxels is calculated respectively to obtain a variation gradient data set; If the variation gradient in the variation gradient data set exceeds the preset gradient threshold, it is marked as high risk, and the gradient value exceeding the threshold is normalized and mapped to the machine room three-dimensional grid to obtain the variation intensity map.

[0031] In an implementation manner, the abnormal area extraction can be that the system reads the high-risk area list, parses the node coordinate sequence row by row, calculates the maximum and minimum values of x, y, and z respectively, forms an axis-aligned bounding box, expands the box by 0.25 m (1 times grid step) to obtain the to-be-calculated area, and the fields are xMin, xMax, yMin, yMax, zMin, and zMax, with a unit of meters. Subsequent gradient calculation is only performed within the box to avoid full-machine-room redundancy and effectively reduce the calculation difficulty.

[0032] The digital machine room model construction operation can be based on the to-be-calculated area bounding box, and the space in the box is divided by 0.25 m step to generate a regular voxel grid. Each voxel center coordinate corresponds to the S12 smooth distribution data node one by one. The temperature and humidity values in the smooth distribution data that fall within the box are directly written into the voxel attributes. The temperature attribute field is in Celsius, and the humidity attribute field is in %RH. The geometry and field quantity jointly constitute the digital machine room model for gradient operation.

[0033] After the model is built, the existing air flow rate is directly extracted: for each voxel center, the nearest neighbor node is retrieved in the air flow rate scalar field output in S12 (search radius 0.5 m, if there are multiple nodes, the inverse distance weighted average is used, and the weight is set to 1 divided by the square of the distance), the air flow rate scalar value of the voxel center is obtained, the unit is meter per second, and the air flow distribution dataset is written, the fields are x, y, z coordinates, air flow rate, for subsequent gradient calculation.

[0034] The system reads the temperature and humidity values from the voxel attributes, and uses the central difference method: the values of the adjacent nodes in the ±x, ±y, ±z directions are taken for the internal voxels, the difference is divided by twice the grid step (0.5 m), and the temperature gradient component and the humidity gradient component are obtained, and the boundary voxels are filled with the first order upwind; then the temperature gradient size and the humidity gradient size are calculated, that is, the square sum of each component is taken and the square root is taken, to form the variation gradient dataset, the fields are node coordinates, temperature gradient size (degree Celsius per meter), humidity gradient size (%RH per meter), and the dimensions are consistent with the physical quantities.

[0035] The variation intensity mapping generation can be that the system traverses the variation gradient dataset, compares the temperature gradient size and the humidity gradient size with the user preset gradient threshold value (designed based on thermodynamic simulation and industry specifications, equipment manufacturer's specification book, the purpose is to maintain the stability requirements of the computer room thermal environment and the normal operation requirements of the equipment, in this example, the temperature is set to 2°C / m, and the humidity is set to 5%RH / m), and uses the or logic, that is, any one exceeds to mark high risk; for the high-risk node, the gradient value is subtracted from the threshold value, and then divided by the maximum value of the difference between the gradient value of all voxels in the current to-be-calculated region and the gradient threshold value, to obtain a 0-1 normalized intensity, multiplied by 255 and taken as an integer, written into the corresponding voxel of the computer room three-dimensional grid, to form a variation intensity map, and the larger the voxel value, the higher the risk, for subsequent high-intensity point extraction.

[0036] Exemplarily, the to-be-calculated region bounding box xMin=8.0, xMax=12.0, yMin=4.0, yMax=6.0, zMin=2.0, zMax=2.5, after subdivision, 16*8*2=256 effective voxels are obtained; at the voxel center x10.0 y5.0 z2.25, the smooth distribution data gives the temperature 27.2°C, the humidity 68%RH, and the air flow rate 0.48m / s, the central difference gives the temperature gradient size 0.4°C / m, the humidity gradient size 6%RH / m, and the humidity gradient is greater than the preset threshold value 5%RH / m, the node is marked as high risk, the normalized intensity is (6-5) / 10=0.1, the mapping intensity value is 25, and the voxel is written, and the variation intensity mapping is completed.

[0037] In step S15, according to the variation intensity map, high-intensity point extraction and weighted average fusion are performed to obtain a priority inspection target point set, including: extracting points with intensity values higher than a preset intensity threshold from the variability intensity map to generate a high-intensity point sequence; calculating the spatial distribution density of the high-intensity point sequence, and if the spatial distribution density is higher than a preset density threshold, marking it as a high-density area to obtain a high-density point dataset; According to the high-density point dataset, the point coordinates are fused to obtain the center coordinate dataset of the high-density area; According to the center coordinate dataset, priority sorting is performed to determine the priority inspection target point set.

[0038] The system performs threshold scanning on the variability intensity map to extract points with intensity values higher than a preset intensity threshold from all voxels to generate a high-intensity point sequence. The input is the mapped machine room three-dimensional grid, and each voxel saves the normalized intensity value from 0 to 255. The preset intensity threshold is determined by the system taking the 80th percentile of the normalized intensity value of the whole field after mapping, and the user can adjust the percentile ratio during initialization. The specific operation is as follows: the system traverses the grid in the order of x, y, and z, reads the intensity value of each voxel, and if the intensity value is greater than the preset intensity threshold, the voxel center coordinates and intensity value are written into a temporary list. After traversal, arrange in descending order of intensity value to form a high-intensity point sequence. The sequence field is x, y, z coordinates and intensity value, the coordinate unit is meter, and the intensity value is dimensionless. The effect is that only the risk prominent voxels are retained, and the subsequent weighted average and coordinate fusion steps can directly read this sequence, reducing the amount of calculation. For example, the voxel x10.0 y5.0 z2.25 has an intensity value of 25, which is lower than the preset intensity threshold of 200, and is not included. The voxel x10.5 y5.0 z2.25 has an intensity value of 210, which is higher than the preset intensity threshold, and the point coordinates and intensity are written into the high-intensity point sequence. The sequence is used for calculating the center of the high-density area in the subsequent steps, and the extraction is completed.

[0039] The system takes the high-intensity point sequence as input, calculates the spatial distribution density, and marks the area with a density higher than the preset density threshold as a high-density area, and outputs the high-density point data set. The specific process is as follows: first, the system reads each point coordinate in the high-intensity point sequence, and calculates the spatial point density using the sliding cube window method. The specific operation is as follows: based on the three-dimensional space grid of the machine room, a cube with a side length of 1.0 meters is defined as a sliding window; in turn, the center of the window is aligned with each grid node, and the number of high-intensity points falling within the window range is calculated; the number is divided by the window volume (1 cubic meter) to obtain the spatial distribution density at the center position of the window, with the dimension of per cubic meter. Subsequently, the system compares the calculated density value with the preset density threshold, which is set to 8 points per cubic meter in this embodiment. If the spatial distribution density of a certain area is higher than the preset density threshold, the area is marked as a high-density area, and the corresponding point and its density value are written into the high-density point data set. The data set fields include point coordinates and density values, with the coordinate unit being meters and the density value unit being per cubic meter. The effect is that the risk hotspots in the aggregation state in space are retained, and isolated high points are removed, providing concentrated samples for subsequent weighted average fusion.

[0040] wherein the preset density threshold is set according to the minimum point density statistical value of the high-risk area confirmed in the historical inspection record of the machine room, and the 75th percentile is set (based on the normal operation requirement of the equipment, to ensure that potential high-risk areas with a slightly lower density than the minimum point density statistical value of the high-risk area can also be identified, in practice, the 75th percentile can be modified according to the normal operation requirement of the equipment), and the user can adjust it according to the management requirement during initialization; in this embodiment, 8 points per cubic meter is used to identify spatially aggregated risk hotspots. If the kernel density estimate value of a certain area is higher than the threshold, it is marked as a high-density area, and the corresponding point and density value are written into the high-density point data set for subsequent weighted average fusion.

[0041] The system takes the high-density point data set as input, and performs coordinate fusion on the points marked as high-density areas, and outputs the center coordinate data set of the high-density area. The specific operation is as follows: the system reads the high-density point data set, and groups the points belonging to the same high-density area according to the area identifier; the weighted average fusion is performed on the point coordinates in each group, and the weighted weight is set as the proportion of the intensity value of the corresponding point to the total intensity value in the group, to ensure that the high-risk point has a larger weight; the weighted average calculation is performed on x, y and z coordinates respectively, to obtain the fused center point coordinates, with the same dimension as the original coordinates, and the unit is meter; the center point coordinates of all high-density areas are written into the center coordinate data set, and the data set fields include area number, center coordinate x, center coordinate y, and center coordinate z, which are used for subsequent priority sorting.

[0042] The system takes a central coordinate data set as input, prioritizes, and outputs a priority inspection target point set. The specific operation is as follows: the system reads the central coordinate data set to obtain the central point coordinates of each high-density area and the corresponding area number; the inspection optimization rule can use a weighted scoring method, and the scoring weight distribution can be 40% for area density, 40% for area strength, and 20% for device criticality, wherein the area density is obtained by normalizing the number of point positions in the high-density point position data set in the area, the area strength is obtained by normalizing the sum of point position strength values in the area, and the device criticality is assigned by the user according to the importance level of the cabinet or device where the central point is located, with levels 1 to 5, with 5 being the highest; the system calculates the weighted score for each central point, which is equal to the area density multiplied by 0.4 plus the area strength multiplied by 0.4 plus the device criticality multiplied by 0.2, and the score is dimensionless; all central points are sorted in descending order of score to form a priority inspection target point set, and the point set field includes the sorting sequence number, area number, central coordinate x, central coordinate y, central coordinate z, and score, which is used for subsequent path planning. The area density normalization method is to divide the number of point positions in the area by the total number of point positions in the field to obtain a dimensionless value of 0-1; the area strength normalization method is to divide the sum of point position strength values in the area by the total sum of field strength to obtain a dimensionless value of 0-1.

[0043] In step S16, according to the priority inspection target point set, an A-Star algorithm is used for path planning and path cost calculation to obtain an initial inspection path sequence, including: Obtaining starting position coordinates, obstacle data, terrain data, and airflow resistance data; According to the starting position coordinates, the obstacle data, and the terrain data, a map is constructed to obtain an environmental grid map; An A-Star algorithm is used to search for connection routes from the starting position to each priority inspection target point set on the environmental grid map to generate a preliminary path data set; The spatial distance of each route in the preliminary path data set and the weighted cost of the airflow resistance data are calculated, and the routes with a weighted cost lower than a preset cost threshold are marked as priority paths to obtain a priority path subset; According to the spatial distance of each route in the preliminary path data set and the weighted cost, the point set in the priority path subset is prioritized, and the path order is dynamically adjusted and updated to generate an initial inspection path sequence.

[0044] In an implementation, the system obtains the starting position coordinates, obstacle data, terrain data and airflow resistance data at one time through local mapping combined with real-time sensor collection, ensuring that the subsequent A-Star algorithm searches for a path under real constraints. The input end calls the existing three-dimensional laser scanning point cloud, dynamic environment database and UWB positioning interface in the machine room; the output is four types of boundary files in a unified coordinate system, which are directly read by the map construction module. The starting position coordinates are provided by the UWB positioning base station deployed at the entrance of the machine room. The positioning tag worn by the operation and maintenance personnel or the inspection robot broadcasts the ranging frame after being powered on. The base station calculates the coordinates of the tag through the time difference of arrival, with an accuracy of 0.1 m. The system takes the first online coordinates as the starting position coordinates, with a unit of meters, and the origin is at the lower left corner of the machine room, aligned with the BIM model. The obstacle data is derived from the pre-completed three-dimensional laser scanning point cloud. The scanner is set every 2 m along the passage. After obtaining the millimeter-level point cloud, the ground point classification and noise removal are performed in the matching software. Then the alpha-shape algorithm is used to extract the convex hull to generate triangular facets. The facet vertex coordinates are the obstacle contour, with a data format of XYZ, including entities such as pillars, cabinets, cable bridges, temporary maintenance barriers, etc. The system reads the facet set to obtain the obstacle data. The terrain data is obtained by analyzing the building information model in IFC format: the floor, space and wall elements in the model have been modeled according to the local coordinate system. The system extracts the geometric vertices of these elements to form triangular meshes of walls, raised floors, suspended ceilings and inclined floors. The mesh vertex coordinates are the terrain data, which are used for elevation difference and slope calculation in the A-Star algorithm. The specific way to obtain the airflow resistance coefficient is as follows: the system sets micro-pressure difference-wind speed integrated sensors at each floor air outlet and passage intersection, and uploads the instantaneous wind speed v and pressure difference ΔP in real time; taking the reference wind speed v_ref (given by the machine room air conditioning design document, unit: m / s) under the design working condition as the reference, the local resistance coefficient is calculated as β = 1-v / v_ref, with a value range of 0-1 and dimensionless; then β is written into a voxel grid with a resolution of 0.5 m according to the nearest neighbor principle to form the airflow resistance data, which is used for path cost calculation.

[0045] The system takes the starting position coordinates as the reference, fuses the obstacle data and the terrain data, and constructs an environment grid map for A-Star search. The input includes the starting position coordinates, the obstacle triangle facet set, and the terrain triangle facet set, and the output is a regular voxel grid, each voxel recording the passing generation value. The specific operation is as follows: the system first establishes a regular grid covering the machine room range, the grid step is 0.5 meters, the grid origin is aligned with the starting position coordinates, and the node number in x, y, and z directions is obtained by dividing the machine room outline size by the step length and rounding up, forming a structured voxel array. Then each voxel center point is traversed, and the relationship between the point and the obstacle facet is judged. If the shortest distance from the center point to the facet is less than 0.25 meters, the voxel is marked as an obstacle, and the passing generation value is set to infinity, otherwise, the terrain judgment is entered. In the terrain judgment, the system calculates the maximum elevation difference Δz (unit: m) of the voxel center and its 8-neighborhood, and uses the exponential penalty formula: to obtain the terrain generation value; when Δz≤0.05m, gterrain≈1, when Δz=0.2m, gterrain≈1.8, and when Δz=0.5m, gterrain≈4.5, realizing the continuous difference of gentle-steep slope. The airflow resistance generation value gflow=1+β, β is the real-time resistance coefficient (0-1), which is linearly mapped to the interval [1, 2]. The final passing generation value: has a value range of about 2-6.5, and the larger the coefficient represents the greater the walking or flying resistance. The generation values of all voxels are written into a three-dimensional array to form an environment grid map, and the array index corresponds to the voxel coordinates one by one for A-Star algorithm query.

[0046] The system takes the environment grid map as the search space, and uses the A-Star algorithm to find the shortest connection route from the starting position to each priority inspection target point set in the three-dimensional voxel network, generating a preliminary path data set. The specific operation is as follows: the system first pre-computes the jump point table based on the regular six-neighborhood, and when encountering the same generation value voxel continuously in the straight line direction, it directly jumps over until it encounters an obstacle, a boundary or a generation value change position to generate a candidate node. Once the table is queried, it replaces the original multi-step expansion, which can significantly reduce the node expansion amount and reduce the calculation requirement. The open list uses a binary minimum heap, and the cumulative passing generation cost and three-dimensional Euclidean distance are normalized by the maximum value and added to obtain the key value; only the jump point and the target neighborhood perform standard g value update, and the rest of the voxels are marked as continuous same cost area and are not put into the heap, and the cycle continues until the current node is within 0.5m voxel step length of any priority inspection target, which is recognized as reached, and the three-dimensional path point sequence is obtained by backtracking along the parent node. The path sequence set of all target points forms the preliminary path data set, with fields of target point number, path number, x, y, and z, for subsequent cost calculation and optimal path screening.

[0047] With the preliminary path data set as input, the system performs a normalized weighted cost calculation on each route, then compares it with a preset cost threshold, marks the priority path, and obtains a priority path subset. The specific operation is as follows: the system first establishes a normalized basic quantity for all candidate paths from the same starting point; the spatial distance part calculates the Euclidean distance cumulative sum D of each path point sequence, with the dimension of meters, and takes the maximum distance Dmax in the group, and the normalized distance d = D / Dmax, with the value range of 0-1; the airflow resistance part reads the resistance coefficient β of the path passing through the voxel, calculates its arithmetic mean value B, with the dimension of 0-1, takes the maximum resistance Bmax in the group, and the normalized resistance b = B / Bmax, with the value range of 0-1; both normalization operations are derived from the same batch of path statistical extreme values, avoiding dimensional differences between different starting point groups. The weighted cost C adopts the user-set normalized weight: in this example, the spatial distance is 0.4 and the airflow resistance is 0.6, reflecting that the airflow resistance has a greater impact on the time consumption of movement; the calculation formula is Cn = 0.4·d + 0.6·b, with the result being dimensionless and the value range being 0-1. The preset cost threshold Cnth takes 80% quantile in the same normalization group, which can retain the "head" paths with lower comprehensive cost and shorter overall time consumption, while eliminating the tail high-cost routes, so that the subsequent sorting and splicing links are concentrated in the high-quality solution space, reducing redundant calculations; the quantile ratio can be adjusted by the user in the initialization configuration file, supporting a 60%-90% interval sliding to adapt to different machine room sizes or emergency inspection needs. In this example, Cnth = 0.72, with the same dimension as Cn; if Cn is less than Cth, the preferred flag 1 is appended to the end of the path record, otherwise 0 is appended, and after all the marking is completed, the paths with the preferred flag equal to 1 are filtered to form the priority path subset, which has the same fields as the preliminary path data set and is used for subsequent sorting. For example, after normalization of the 5 candidate paths from the starting point to target A, a certain path has d = 0.35, b = 0.80, and Cn = 0.4 x 0.35 + 0.6 x 0.80 = 0.62, which is less than the threshold 0.72, and is marked as a priority path. The normalization weighting and filtering are completed.

[0048] The system takes the priority path subset as input, sorts the point set according to priority by combining spatial distance and weighted cost, and generates an initial inspection path sequence by dynamically adjusting and updating the path order. The input is the path record with the marked preferred flag, and the output is the complete inspection route arranged according to priority. The specific operation is as follows: the system first groups the priority path subset according to the target point number, and each group contains multiple priority paths from the starting position to the same priority inspection target; for each path set, the normalized spatial distance and the normalized weighted cost are calculated, and the normalization method is the maximum value normalization in the group, resulting in a dimensionless value of 0 to 1; then the weighted scoring method is used, and the weight setting is based on the "balanced time consumption" principle, with a spatial distance of 50% and a weighted cost of 50% in this example, so that the path is neither too long nor too resistant. The priority score of each path is calculated, and the score S is: In the formula, the score S is dimensionless, and the larger the value is, the better the path is; Cn is a normalized spatial distance, equal to the current path distance divided by the maximum distance in the same group, dimensionless; b is a normalized weighted cost, equal to the current path weighted cost divided by the maximum weighted cost in the same group, dimensionless. The first path in each group is arranged in descending order of the score, and is taken as the entering route of the target to form a target point order; if there are multiple target points, the system uses the nearest neighbor insertion method to dynamically adjust the order: starting from the starting position, the point with the highest path score and the smallest spatial distance in the unvisited target is selected as the next station, the visited set is updated, and the process is repeated until all target points are included. Finally, the three-dimensional point sequence of all selected paths is spliced in the access order to generate an initial inspection path sequence, and the sequence field is the path number, x, y and z, which is used for subsequent node dynamic adjustment.

[0049] In step S17, according to the initial inspection path sequence and the original environment parameter data set, the path node dynamic adjustment is performed to obtain an updated inspection path, including: obtaining real-time sensor feedback data; comparing the real-time sensor feedback data and the original environment parameters in the original environment parameter data set according to the initial inspection path sequence to obtain an environment parameter fluctuation difference value; if the environment parameter fluctuation difference value exceeds a preset fluctuation threshold, re-grouping the path nodes by K-means clustering to determine a set of path nodes to be adjusted; updating the environment grid map according to the current obstacle distribution and air flow resistance data, and re-calculating the connection cost between nodes in the set of path nodes to obtain an updated path node cost table; regenerating an inspection path sequence according to the updated path node cost table and combining a preset path sorting rule to obtain an updated inspection path.

[0050] The real-time sensor feedback data is obtained by a dynamic host through Modbus TCP protocol polling, including environmental quantity instantaneous value, device operating parameters and obstacle information; the data packet has a time stamp and a delay compensation value, and the system takes the latest valid record as the current environment state for synchronization with the static model.

[0051] The system takes the initial inspection path sequence as the main line, compares the real-time sensor feedback data with the original value of the corresponding node in the original environment parameter data set point by point, and obtains the fluctuation difference value. The input is the initial inspection path sequence, the original environment parameter data set and the real-time sensor feedback data; the output is the fluctuation difference value table for subsequent threshold judgment, and the specific operation is as follows. The system first parses the initial inspection path sequence, extracts the three-dimensional coordinates of each path node, and retrieves the nearest grid node in the original environment parameter data set with the coordinates, reads the original temperature, original humidity and original air flow speed of the node; at the same time, the current temperature, current humidity and current air flow speed at the same coordinates are obtained through the real-time sensor feedback data, if the real-time data is missing, the adjacent sensor is supplemented by tri-linear interpolation to ensure the same dimension. Then, the system calculates the difference value of the corresponding parameters, i.e. the fluctuation difference value, wherein the temperature difference value is equal to the current temperature minus the original temperature, the humidity difference value is equal to the current humidity minus the original humidity, and the air flow speed difference value is equal to the current air flow speed minus the original air flow speed, with the same unit and original dimension. The system writes these difference values into the fluctuation difference value table, and the table fields include path sequence number, node coordinates, temperature difference value, humidity difference value and air flow speed difference value.

[0052] The system reads the fluctuation difference value table, normalizes the over-standard nodes, and forms the path node set to be adjusted. The specific operation is as follows: the preset fluctuation threshold is written into the configuration file by the user during initialization, and in this example, the temperature is 2°C, the humidity is 3%, and the air flow is 0.2m / s, with the same dimension as the corresponding difference value; the system traverses the fluctuation difference value table, takes the absolute value of the temperature difference value, humidity difference value and air flow speed difference value of each record, and appends the over-standard flag 1 if any value exceeds the threshold, otherwise appends 0, to form the fluctuation difference value table with flag. In order to eliminate the dimensional difference, the system performs maximum value normalization on the over-standard nodes: within the same batch path set, the maximum temperature difference value Tmax, the maximum humidity difference value Tmax and the maximum air flow speed difference value Tmax are extracted respectively, the node coordinates x, y and z are normalized with the space span Xmax, Ymax and Zmax of the group, and the dimensionless five-dimensional feature vector [x / Xmax, y / Ymax, z / Zmax, AT / ATmax, AH / AHmax, AV / AVmax] is obtained. K-means clustering is adopted, the cluster number k=2 (high fluctuation and low fluctuation), and the iteration termination condition is that the normalized cluster center change is less than 0.05 and the normalized difference value vector change is less than 0.05, which ensures convergence and dimensional consistency. After clustering, the original path sequence number, normalized coordinates and normalized difference value vector of the nodes belonging to the high fluctuation cluster after normalization are written into the path node set to be adjusted, which is used for subsequent map updating.

[0053] The system takes the current obstacle distribution and airflow resistance data as input, incrementally updates the environmental grid map, and recalculates the connection cost between each node in the set of path nodes to be adjusted, to obtain an updated path node cost table. The specific operation is as follows: the system first reads the triangle facet set of the current obstacle distribution, which is generated by the laser radar point cloud after alpha-shape convex hull extraction and compared with the original map obstacle layer by difference; the voxels occupied by the new facets are marked as obstacles, and the voxels corresponding to the facets are deleted to restore the passable state, and the obstacle marking is derived from the difference between the vertex coordinates of the facets in the last scanning period and the current period. The airflow resistance data is uploaded in real time by the micro-pressure difference-wind speed integrated sensor, and the system maps the latest resistance coefficient β to the corresponding voxel with a 0.5-meter grid step; if the β change of a voxel exceeds 0.1, the passable value of the voxel is recalculated, and the recalculation formula is that the passable value is equal to the terrain base value plus ten times β, wherein the terrain base value is 1 on flat ground, 2 in the region with an elevation difference greater than 0.2 meters, and β is a dimensionless coefficient between 0 and 1. The calculation result is rounded and written into the voxel attribute, keeping the same dimension. For each node in the set of path nodes to be adjusted, the system queries the updated voxel passable value within a 6-neighborhood range centered on the node; the movement cost is equal to the target voxel passable value multiplied by the grid step 0.5 meters, with a unit of meters; if the neighborhood voxel is marked as an obstacle, its movement cost is set to infinity, otherwise the calculated value is retained. The system uses Dijkstra local recalculation: taking the current node as the source, iteratively updating the g value in the updated map until the new g value of all nodes in the set converges, to obtain the updated path node cost table.

[0054] The system takes the updated path node cost table as input, and uses the normalized weighted scoring method to regenerate the inspection path sequence to obtain the updated inspection path. The specific operation is as follows: the system reads the updated path node cost table (including node number, three-dimensional coordinates and cumulative cost g); the pre-set path sorting rule is written into the configuration file by the user during initialization, and the weighted scoring method is used in this embodiment, with the weight distribution being 60% for movement cost and 40% for spatial distance. The movement cost accounts for 60% to reflect the dominant influence of airflow and terrain resistance on time consumption, and the spatial distance accounts for 40% to retain the geometric shortest tendency, both of which are normalized to the same extreme value to avoid dimensional differences; the weight proportion is adjustable by the user in the range of 60%-80%, meeting the preferences of different machine rooms for "resistance" or "short distance". First, normalize the maximum value of all nodes in the same recalculation group: the normalized value of the movement cost g is g / gmax, and the normalized value of the spatial distance d is d / dmax, both of which are dimensionless; the weighted score S: The score range is 0-1, and the smaller the score, the lower the comprehensive cost. The system takes the starting position as the root and arranges the updated nodes in ascending order of S; each node is connected in turn, and the path point interval is kept at a grid step of 0.5 m. The coordinates of the vertices of the broken line are written into the updated inspection path sequence, and the sequence field is the path number, x, y, and z, which is used for subsequent coverage evaluation. For example, in a certain group, gmax=12 m, dmax=6 m, a certain node g=6 m, d=3 m, and after normalization, S=0.6*0.5+0.4*0.5=0.5. After arranging in ascending order of S, the node is located in the middle of the sequence, and the system connects in turn to complete the path regeneration.

[0055] In step S18, according to the updated inspection path and the variation gradient data of the variation intensity map, a preliminary evaluation of the path coverage is performed to obtain a preliminary evaluation result of the path coverage, including: According to the real-time sensor feedback data, the environmental parameter prediction value and the spatial coordinates on the updated inspection path are extracted to obtain dynamic distribution data of the environmental parameters; The dynamic distribution data of the environmental parameters and the variation gradient data of the variation intensity map are used to calculate the risk value. If the calculated risk value exceeds the preset risk threshold, a high-risk area is marked, and a risk value data set is generated; According to the risk value data set, high-intensity point positions are extracted to obtain a high-intensity point position sequence; According to the risk value data set, high-intensity point positions are extracted to obtain a high-intensity point position sequence;

[0056] According to the real-time sensor feedback data, the environmental parameter prediction value and the spatial coordinates on the updated inspection path are extracted to obtain dynamic distribution data of the environmental parameters. The system uses the three-dimensional node sequence of the updated inspection path as an index to search for the instantaneous environmental parameters of the same or nearest neighbor coordinate points in the real-time sensor feedback data. The search radius is set to a grid step of 0.5 m. If there are multiple sensors within this radius, the inverse distance weighted average is used, and the weighted average weight coefficient is equal to 1 divided by the square of the distance between the measuring point and the node, to ensure that the near point contributes more. For missing coordinates, linear interpolation is used to fill in the front and back points of the path, so that each path node obtains the current temperature, current humidity, and current air flow velocity, which are consistent with the original dimensions. Then, the system writes the coordinates of each node and the corresponding current environmental parameters into the dynamic distribution data table, and the table field is the path number, x, y, z, current temperature, current humidity, and current air flow velocity, which is used for subsequent gradient fusion.

[0057] The operation of risk value calculation of the dynamic distribution data of environmental parameters and the variation gradient data of variation intensity mapping is as follows: the system reads the current temperature and current humidity of each path node in the dynamic distribution data table, and queries the variation gradient data set at the same coordinate to obtain the temperature gradient size and humidity gradient size of the node, with dimensions of ℃ / m and %RH / m respectively; in order to eliminate the dimensional difference, the system performs maximum normalization on the two gradients respectively: in the same batch path set, the maximum temperature gradient Tgrad_max and the maximum humidity gradient Hgrad_max are taken, the temperature gradient normalized value t_grad = temperature gradient size / Tgrad_max is calculated, and the humidity gradient normalized value h_grad = humidity gradient size / Hgrad_max is calculated, both of which are dimensionless.

[0058] The risk value R is linearly combined by the user set weight: in this example, the temperature gradient normalized value is 0.6 and the humidity gradient normalized value is 0.4, which reflects the higher sensitivity of temperature gradient to overheating risk; the calculation formula R = 0.6·t_grad + 0.4·h_grad is obtained, which is dimensionless and has a value range of 0-1. The preset risk threshold Rth is taken as the 80% quantile in the same batch path set, and in this example Rth = 0.72, which is dimensionless; if R is greater than Rth, the node is marked as high risk and written into the risk value data set, with fields of node coordinates, risk value R and high risk flag.

[0059] Among them, the preset risk threshold Rth is set according to the following method: in the same batch path set, the system will statistically analyze the risk value distribution corresponding to the high risk nodes confirmed in the same period (or the previous week) in history, and take the 80% quantile value as Rth, so as to ensure that the marked area covers most of the positions that "really need intervention" in the past, while eliminating low confidence samples at the tail; this proportion can be slid by the user in the range of 60%-90% to balance sensitivity and inspection load.

[0060] For example, in a batch path, Tgrad_max = 4 ℃ / m, Hgrad_max = 10 %RH / m, the temperature gradient size of node A is 3.2 ℃ / m, the humidity gradient size is 8 %RH / m, after normalization t_grad = 0.8, h_grad = 0.8, risk value R = 0.6 x 0.8 + 0.4 x 0.8 = 0.8, which is greater than the threshold value 0.72, the node is marked as high risk and written into the risk value data set, and the risk value calculation and marking are completed.

[0061] The input for the high-intensity point extraction operation is a risk value dataset that has completed risk value R calculation and 80% quantile threshold Rth setting (Rth = 0.72 in this example) within the same batch path set. The system traverses all nodes, and if the node risk value R is greater than Rth, the node coordinates and risk value are written to a temporary list. After traversal, the sequence is arranged in descending order of risk value to form a high-intensity point sequence, with the sequence fields being x, y, z coordinates and risk value, with coordinate units in meters and risk value being dimensionless.

[0062] In combination with the preset coverage evaluation rule, the system calculates the coverage rate of the updated inspection path on the high-intensity point sequence to obtain the preliminary path coverage evaluation result. The specific operation is as follows: within the same batch path set, the system calculates the Euclidean distance between the nodes of the high-intensity point sequence and the nearest path point, and takes the 80% quantile of the distance set as the preset coverage distance threshold Dth to ensure that the covered area covers most of the previously confirmed intervention locations. Users can adjust the interval of 60%-90% to balance sensitivity and inspection load, and in this example, 0.75m, which is dimensionless. The Euclidean distance between each point in the high-intensity point sequence and the nearest path point on the updated inspection path is calculated. If the distance is ≤Dth, the point is determined to be covered, and the coverage flag is set to 1, otherwise it is set to 0. The coverage flags of all points are accumulated to obtain the number of covered points, which is divided by the total number of high-intensity point sequences to obtain the coverage rate, which is dimensionless. The coverage rate is written into the path coverage preliminary evaluation result, with fields for the total number of high-intensity points, the number of covered points, and the coverage rate for subsequent gap filling judgment. For example, within a batch path set, the high-intensity point sequence has 100 points and an 80% quantile distance of 0.75m. The updated inspection path contains 150 path points. After traversal, 82 points have a nearest distance ≤0.75m from the path, with a coverage rate of 82 / 100=0.82, which is greater than the preset coverage target of 80%. The system writes this coverage rate into the evaluation result, completing the preliminary coverage evaluation.

[0063] In step S19, linear interpolation filling is performed according to the path coverage preliminary evaluation result to obtain the final optimized inspection path, including: Filling the path gaps in the path coverage preliminary evaluation result to generate a filled path point sequence; Associating the filled path point sequence with the variability intensity mapping variability gradient data. If the association result shows a density anomaly, a set of density anomaly points is extracted, and the filling parameters are dynamically adjusted to obtain an adjusted filling parameter set; According to the adjusted filling parameter set, in combination with the dynamic distribution data of the environmental parameters, the inspection path is regenerated, and it is determined whether its coverage meets the preset coverage evaluation rule. If it does, it is taken as the final optimized inspection path.

[0064] The path gap in the preliminary evaluation result of path coverage is filled to generate a filled path point sequence. The specific operation is that the system reads the preliminary evaluation result of path coverage, extracts the three-dimensional node sequence of the updated inspection path, and calculates the Euclidean distance between adjacent nodes. If the distance is greater than 0.5 meters of the grid step, intermediate points are inserted between the two points in a linear interpolation manner, the number of insertion is equal to the distance divided by the step and rounded up, and the coordinates of the inserted points are obtained by adding the step increment in the vector direction to the starting point coordinates, and the increment component is equal to the terminal point coordinates minus the starting point coordinates and divided by the number of insertion, ensuring that the x, y, z directions are incremented synchronously. The temperature, humidity, and air flow velocity of the inserted points are obtained by linear interpolation of the corresponding environmental parameters of the starting point and the terminal point, forming a continuous path point sequence, and the sequence field is x, y, z coordinates, temperature, humidity, and air flow velocity, which is used for subsequent density correlation.

[0065] The operation of correlating the filled path point sequence with the variation gradient data of the variation intensity map is as follows: the system reads the temperature gradient size and humidity gradient size in the variation intensity map for each path point in the filled path point sequence with the same coordinates, with dimensions of ℃ / m and percentage per meter, respectively. Calculate the average gradient in the neighborhood of the point, and the neighborhood radius is 0.5 meters of the grid step. If the average is greater than the preset gradient density threshold, it is determined to be a density anomaly. The preset gradient density threshold is set according to the 80% quantile of the historical variation gradient distribution in the region to be calculated: the system reads the temperature gradient and humidity gradient of all voxels in the region, respectively, and calculates the cumulative distribution of the temperature gradient and the cumulative distribution of the humidity gradient. Take the 80% quantile value of the temperature gradient as the temperature side threshold, and take the 80% quantile value of the humidity gradient as the humidity side threshold. In this example, it corresponds to 2 ℃ / m or 5 percentage per meter, using or logic. The marked point is written into the density anomaly point set, and the set field is the coordinate, the average temperature gradient, and the average humidity gradient. Then the system dynamically adjusts the filling parameters, reduces the linear interpolation step to 0.5 times the original step, i.e. 0.25 meters, and recalculates the number of inserted points, which is equal to the distance divided by the new step and rounded up, to obtain the adjusted filling parameter set, with the set field being the new step and the new insertion number, which is used for subsequent denser path generation.

[0066] The system takes the adjusted filling parameter set as input, re-plans the inspection path and judges the coverage. The specific operation is as follows: the system reads the new step (0.25 meters) and the number of insertions, and re-executes linear interpolation in the original path gap to generate a denser path point sequence; then take this sequence as a new candidate inspection trajectory, and re-calculate its nearest neighbor distance with the high intensity point sequence. If the new trajectory coverage rate is ≥90% of the target, the iteration is immediately terminated, and the current sequence is written into the final optimized inspection path. If it is not satisfied, continue to reduce the step to 0.125 meters and repeat the interpolation and calculation until the target coverage rate is met, and the final optimized inspection path is obtained.

[0067] In summary, the application discloses a server inspection method based on an IT operation and maintenance platform, which solves the problem of insufficient inspection coverage integrity in the prior art by collecting environmental parameters in real time, identifying abnormal areas, dynamically planning a path, and evaluating coverage.

[0068] Referring to Figure 2 The second embodiment of the application provides a server inspection device based on an IT operation and maintenance platform, comprising: A data acquisition module is configured to acquire an original environmental parameter data set. A parameter distribution map generation module is configured to perform gridding interpolation processing based on the original environmental parameter data set to obtain an environmental parameter distribution map. A high-risk area division module is configured to perform K-means clustering processing based on the environmental parameter distribution map to obtain a high-risk area list. A mutation gradient calculation module is configured to perform mutation gradient calculation based on the high-risk area list to obtain a mutation intensity map. A priority inspection target determination module is configured to perform high-intensity point extraction and weighted average fusion based on the mutation intensity map to obtain a priority inspection target point set. An initial path planning module is configured to perform path planning using an A-Star algorithm and path cost calculation based on the priority inspection target point set to obtain an initial inspection path sequence. A path update module is configured to perform path node dynamic adjustment based on the initial inspection path sequence and the original environmental parameter data set to obtain an updated inspection path. A path coverage evaluation module is configured to perform preliminary path coverage evaluation based on the updated inspection path and mutation gradient data of the mutation intensity map to obtain a preliminary path coverage evaluation result. A final path determination module is configured to perform linear interpolation filling based on the preliminary path coverage evaluation result to obtain a final optimized inspection path.

[0069] It should be noted that the server inspection device based on the IT operation and maintenance platform provided by the embodiments of the application is used to perform all process steps of the server inspection method based on the IT operation and maintenance platform of the above-mentioned embodiments, and the working principles and advantages of the two are one-to-one corresponding, and thus will not be repeated.

[0070] The embodiments of the application also provide an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, a server inspection program based on an IT operation and maintenance platform. The processor implements the steps in each of the server inspection method embodiments based on the IT operation and maintenance platform when executing the computer program, for exampleFigure 1 The processor executes the computer program to implement the functions of the modules in the above-mentioned device embodiments, such as the data acquisition module.

[0071] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0072] The electronic device can be a computing device such as a desktop computer, a notebook computer, a palm computer, and a smart tablet. The electronic device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0073] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, and connects all parts of the electronic device through various interfaces and lines.

[0074] The memory can be used to store the computer program or modules, and the processor realizes various functions of the electronic device by running or executing the computer program or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0075] The modules integrated in the electronic device can be stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0076] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0077] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A server inspection method based on an IT operations and maintenance platform, characterized in that, include: Obtain the original environmental parameter dataset; Based on the original environmental parameter dataset, a gridded interpolation process is performed to obtain an environmental parameter distribution map; Based on the environmental parameter distribution map, K-means clustering was performed to obtain a list of high-risk areas; Based on the list of high-risk areas, the variation gradient is calculated to obtain the variation intensity mapping; Based on the aforementioned variability intensity mapping, high-intensity points are extracted and weighted averages are fused to obtain a priority inspection target point set; Based on the set of priority inspection target points, the A-Star algorithm is used to plan the path and calculate the path cost to obtain the initial inspection path sequence. Based on the initial inspection path sequence and the original environmental parameter dataset, the path nodes are dynamically adjusted to obtain the updated inspection path; Based on the updated inspection path and the variation gradient data of the variation intensity mapping, a preliminary assessment of path coverage is performed to obtain the preliminary assessment result of path coverage. Based on the preliminary assessment results of the path coverage, linear interpolation is performed to fill in the gaps and obtain the final optimized inspection path.

2. The server inspection method based on an IT operation and maintenance platform according to claim 1, characterized in that, The step of performing gridded interpolation processing based on the original environmental parameter dataset to obtain an environmental parameter distribution map includes: A gridded interpolation algorithm is performed on the original environmental parameter dataset to generate spatially continuous distributed data; When outliers exist in the spatially continuous distributed data, Kriging interpolation is used to correct the outliers to obtain smoothed distributed data. Based on the smoothed distribution data, an environmental parameter distribution map in three-dimensional space is generated.

3. The server inspection method based on an IT operation and maintenance platform according to claim 1, characterized in that, The process of performing K-means clustering based on the environmental parameter distribution map to obtain a list of high-risk areas includes: Temperature and humidity data are extracted from the environmental parameter distribution map, and a preprocessed dataset is obtained after data cleaning and formatting. The preprocessed dataset is subjected to K-means clustering, which divides it into multiple clusters in three-dimensional space according to temperature and humidity features, generating a regional grouped dataset. Clusters in the grouped regional dataset whose temperature exceeds a preset temperature threshold or whose humidity exceeds a preset humidity threshold are marked as abnormal regions, forming an abnormal region dataset. The dataset of the abnormal areas is stored in a structured manner to generate a list of high-risk areas containing coordinates and corresponding temperature and humidity information.

4. The server inspection method based on an IT operation and maintenance platform according to claim 1, characterized in that, The step of calculating the mutation gradient based on the list of high-risk areas to obtain the variability intensity mapping includes: Based on the list of high-risk areas, abnormal areas are extracted to obtain the area to be calculated; Based on the area to be calculated, a model is constructed to obtain a digital data center model, which includes temperature distribution and humidity distribution. Based on the digital data center model, the airflow velocity scalar field is extracted from the smoothed distribution data to form an airflow distribution dataset. Based on the temperature and humidity scalar fields within the digital data center model, the variation gradient between adjacent voxels is calculated to obtain the variation gradient dataset. If the mutated gradient in the mutated gradient dataset exceeds a preset gradient threshold, it is marked as high risk, and the gradient values ​​exceeding the threshold are normalized and mapped to the three-dimensional mesh of the computer room to obtain the mutability intensity mapping.

5. A server inspection method based on an IT operations and maintenance platform according to claim 1, characterized in that, The step of extracting high-intensity points and performing weighted average fusion based on the variability intensity mapping to obtain a priority inspection target point set includes: Points with intensity values ​​higher than a preset intensity threshold are extracted from the variability intensity map to generate a high-intensity point sequence; Calculate the spatial distribution density of the high-intensity point sequence. If the spatial distribution density is higher than a preset density threshold, it is marked as a high-density region, and a high-density point dataset is obtained. Based on the high-density point dataset, point coordinate fusion is performed to obtain the center coordinate dataset of the high-density area; Based on the central coordinate dataset, priority sorting is performed to determine the set of priority inspection target points.

6. A server inspection method based on an IT operations and maintenance platform according to claim 1, characterized in that, The step of using the A-Star algorithm to plan paths and calculate path costs based on the priority inspection target point set to obtain an initial inspection path sequence includes: Acquire the starting position coordinates, obstacle data, terrain data, and airflow obstruction data; Based on the starting position coordinates, the obstacle data, and the terrain data, a map is constructed to obtain an environmental grid map; The A-Star algorithm is used to search for connecting routes from the starting position to each set of priority inspection target points on the environmental grid map to generate a preliminary path dataset. Calculate the spatial distance of each route in the preliminary path dataset and the weighted cost of the airflow obstruction data, and mark the routes with the weighted cost below a preset cost threshold as priority paths to obtain a subset of priority paths; Based on the spatial distance of each route in the preliminary path dataset and the weighted cost, the point set in the priority path subset is prioritized and the path order is dynamically adjusted and updated to generate an initial inspection path sequence.

7. A server inspection method based on an IT operations and maintenance platform according to claim 6, characterized in that, The step of dynamically adjusting path nodes based on the initial inspection path sequence and the original environmental parameter dataset to obtain an updated inspection path includes: Acquire real-time sensor feedback data; Based on the initial inspection path sequence, the environmental parameter fluctuation difference is obtained by comparing the real-time sensor feedback data with the original environmental parameters in the original environmental parameter dataset. If the fluctuation difference of the environmental parameters exceeds the preset fluctuation threshold, K-means clustering is used to regroup the path nodes to determine the set of path nodes to be adjusted. Based on the obstacle distribution and airflow obstruction data, the environmental grid map is updated and the connection cost between each node in the path node set is recalculated to obtain the updated path node cost table. Based on the updated path node cost table and the preset path sorting rules, the inspection path sequence is regenerated to obtain the updated inspection path.

8. A server inspection method based on an IT operations and maintenance platform according to claim 7, characterized in that, The preliminary path coverage assessment is performed based on the updated inspection path and the variation gradient data mapped by the variability intensity, resulting in a preliminary path coverage assessment result, including: Based on the real-time sensor feedback data, the predicted values ​​and spatial coordinates of environmental parameters are extracted on the updated inspection path to obtain the dynamic distribution data of environmental parameters. The dynamic distribution data of the environmental parameters and the variation gradient data mapped by the variability intensity are used to calculate the risk value. If the calculated risk value exceeds the preset risk threshold, a high-risk area is marked and a risk value dataset is generated. Based on the risk value dataset, high-intensity locations are extracted to obtain a high-intensity location sequence; Based on the preset coverage evaluation rules, the coverage rate of the updated inspection path to the high-intensity point sequence is calculated to obtain the preliminary assessment result of path coverage.

9. A server inspection method based on an IT operations and maintenance platform according to claim 8, characterized in that, The step of performing linear interpolation to fill in the path coverage based on the preliminary assessment results to obtain the final optimized inspection path includes: Fill the path gaps in the preliminary path coverage assessment results to generate a filled path point sequence; The filled path point sequence is associated with the variation gradient data of the variation intensity mapping. If the association result shows density anomaly, the set of density anomaly points is extracted, and the filling parameters are dynamically adjusted to obtain the adjusted filling parameter set. Based on the adjusted set of filling parameters and the dynamic distribution data of the environmental parameters, the inspection path is regenerated, and it is determined whether its coverage meets the preset coverage evaluation rules. If it does, it is used as the final optimized inspection path.

10. A server inspection system based on an IT operations and maintenance platform, characterized in that, include: The data acquisition module is used to acquire the raw environmental parameter dataset; The parameter distribution map generation module is used to perform gridded interpolation processing based on the original environmental parameter dataset to obtain an environmental parameter distribution map; The high-risk area delineation module is used to perform K-means clustering based on the environmental parameter distribution map to obtain a list of high-risk areas; The mutation gradient calculation module is used to calculate the mutation gradient based on the list of high-risk areas to obtain the variability intensity mapping. The priority inspection target determination module is used to extract high-intensity points and perform weighted average fusion based on the variability intensity mapping to obtain a set of priority inspection target points. The initial path planning module is used to perform path planning and path cost calculation based on the set of priority inspection target points, and obtain the initial inspection path sequence. The path update module is used to dynamically adjust the path nodes based on the initial inspection path sequence and the original environmental parameter dataset to obtain an updated inspection path. The path coverage assessment module is used to perform a preliminary assessment of path coverage based on the updated inspection path and the variation gradient data mapped by the variability intensity, and obtain a preliminary assessment result of path coverage. The final path determination module is used to perform linear interpolation filling based on the preliminary assessment results of the path coverage to obtain the final optimized inspection path.

Images