Industrial thermal management cloud native management method and platform

Abstract

The application provides an industrial thermal management cloud native management method and platform. The method comprises the following steps: obtaining the temperature, load and task deployment data of an industrial node, constructing a node thermal task mapping matrix and a thermal resource characterization result, and then generating a task migration arrangement strategy and performing closed-loop management. The problem of the split of traditional thermal management and cloud native scheduling is solved, the global optimal scheduling of computing power and thermal energy is realized, local thermal aggregation intensification or thermal risk diffusion is effectively avoided, and the stability and safety of the industrial site are ensured.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation, and in particular relates to a cloud-native management method and platform for industrial thermal management. Background Technology

[0002] Industrial thermal management is not simply a matter of temperature monitoring or localized heat dissipation control. Especially with the widespread adoption of edge computing nodes, industrial control servers, cooling units, heat exchange units, and containerized business components in modern industrial environments, thermal status has evolved from an operational indicator specific to individual devices to a systemic constraint spanning the device, control, edge, and platform layers. Existing, relatively similar technological approaches can generally be divided into two categories: one is traditional industrial thermal management solutions, primarily relying on temperature sensors, threshold alarms, fan start / stop, cooling loop adjustment, or localized load reduction to monitor and control single devices or areas; the other is cloud platform or cloud-native management solutions oriented towards computing resources, mainly focusing on container scheduling, service orchestration, and elastic management of standard resources such as CPU, memory, network, and storage. Both types of solutions have their uses in their respective scenarios, but once they enter the complex scenario of an industrial thermal management cloud-native platform, the shortcomings of existing technologies become acutely apparent. The problem with traditional thermal management solutions is that they often understand thermal risk as a local phenomenon after the temperature at a certain point exceeds a threshold. They tend to "detect and then deal with" rather than reflect the coupling, propagation, accumulation process, and time-delay characteristics of heat between industrial objects. They also cannot explain whether the temperature rise at a certain node comes from its own load, heat transfer from adjacent devices, cooling capacity decay, or changes in environmental conditions. Therefore, they are difficult to support unified platform-level management and cannot directly serve container deployment, task migration, service orchestration, and resource isolation.

[0003] The problem with existing cloud-native management solutions is that they are designed for inherently discrete and easily quantifiable digital resources. Industrial thermal states, however, are inherently continuous, coupled, and possess inertial and propagational physical quantities, which cannot be directly equated with ordinary resource indicators and incorporated into the scheduler. Using only instantaneous temperature values ​​or fixed temperature thresholds as the basis for scheduling fails to accurately reflect the true thermal carrying capacity of nodes, nor does it reflect the evolution of thermal risks under the combined effects of multiple devices, tasks, and cooling links within the same area. This easily leads to a disconnect between platform-side scheduling decisions and the actual thermal state, resulting in problems such as increased local heat accumulation after scheduling, the spread of thermal risks from one node to adjacent nodes, and a disconnect between cooling actions and task adjustments. Furthermore, industrial scenarios have higher requirements for stability, security, and interpretability. When making migration, load reduction, isolation, or linkage control actions, the platform cannot simply provide results; it also needs to establish a mechanism that can transform complex thermal states into platform-identifiable, computable, and executable management constraints, elevating thermal management from device-level response to platform-level coordination. Therefore, the key shortcoming of existing technologies is not just that "temperature prediction is not accurate enough" or "there are not enough scheduling rules", but that there is a lack of a technical approach that is oriented towards industrial objects, adapted to cloud-native platforms, and can uniformly express thermal states and drive unified management. Summary of the Invention

[0004] This invention discloses a cloud-native management method and platform for industrial thermal management to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, a first aspect of the present invention provides an industrial thermal management cloud-native management method, the method comprising: S1: Acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data, and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result; S2: Based on the node integrated thermal task mapping result and the task thermal mapping value in the node thermal task mapping matrix, calculate the dispersion of the task thermal mapping value within the node and generate the node thermal distribution coefficient; correct the node integrated thermal task mapping result based on the node thermal distribution coefficient and generate the node thermal resource characterization result. S3: Sort the industrial nodes according to the node thermal resource characterization results and select the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix to generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node; S4: Based on the cloud-native thermal management orchestration results, control the migration of cloud-native tasks between industrial nodes; obtain the execution flag of task migration; calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

[0006] Furthermore, the acquisition of node temperature data, node load data, and cloud-native task deployment data of industrial nodes specifically includes: Temperature sampling values ​​of the nodes are obtained through temperature acquisition devices on the industrial nodes; Obtain node load status values ​​through an agent program deployed on the industrial nodes; The deployment mapping between task instances and nodes can be obtained through the scheduling interface of the cloud-native scheduling system.

[0007] Furthermore, the node temperature data and the node load data are normalized, specifically including: An operating range table is established for each industrial node within the platform. The operating range table records the upper and lower limits of stable operation of temperature and load for each node. The node temperature sampling value and the node load sampling value are mapped to the corresponding operating interval table, and converted into node temperature status value and node load status value in a uniform proportional space.

[0008] Furthermore, the calculation of the node hot task mapping matrix specifically includes: The task instance is bound to the node temperature status value and node load status value of the node it belongs to; Based on preset weighting parameters, the node temperature status value and the node load status value are weighted and combined to generate a task-level thermal mapping value. The task-level hot mapping values ​​of each task instance are organized by node to form a node hot task mapping matrix.

[0009] Furthermore, the degree of dispersion of the task heat mapping values ​​within the computing node specifically includes: Calculate the average value of all task-level heatmaps within the node; Calculate the sum of the absolute values ​​of the differences between each task-level heat mapping value and the average value; The sum of the absolute values ​​of the differences is normalized with the node comprehensive thermal task mapping result to generate the node thermal distribution coefficient.

[0010] Furthermore, the generation of cloud-native thermal management orchestration results, which include source nodes, task instances, and target nodes, specifically includes: The node thermal resource characterization results are used as global risk items, and the task-level thermal mapping values ​​are used as local contribution items to calculate task migration driving values. The migration adaptation result is calculated based on the task migration driving value, the thermal resource characterization result of the candidate target node, and the thermal resource difference between the source node and the target node. A task migration list is generated based on the task migration driver value and the migration adaptation result.

[0011] Furthermore, the step of ranking the industrial nodes based on the node thermal resource characterization results specifically includes: Establish a node thermal resource status table and sort the node thermal resource characterization results from smallest to largest. Write the industrial nodes that are ranked higher into the priority carrying set and the industrial nodes that are ranked lower into the restricted carrying set. A list of nodes to be processed is generated based on the restricted bearer set.

[0012] Further, the step of calculating and updating the node thermal management result after execution based on the execution flag, the node hot task mapping matrix, and the node thermal resource characterization result specifically includes: For the source node, subtract the task-level heat mapping value of the migrated task from the heat resource characterization result before its execution to obtain the heat management result after the source node is executed. For the target node, the thermal management result after execution is obtained by subtracting the task-level thermal mapping value of the outgoing task from the thermal resource characterization result before execution, and then adding the product of the task-level thermal mapping value of the incoming task and the incoming thermal absorption coefficient of the target node.

[0013] Furthermore, the control cloud-native task is migrated between industrial nodes, specifically including: For containerized task instances, update the node affinity constraints of the task instance through the cloud-native scheduling interface and trigger rescheduling; For industrial edge service replicas, a new replica is started on the target node through the service orchestration interface, and the original replica on the source node is terminated after the new replica is ready.

[0014] In a second aspect of the invention, an industrial thermal management cloud-native management platform is provided, the platform comprising: The data acquisition module is used to acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data; and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result. The characterization generation module is used to calculate the dispersion of task heat mapping values ​​within a node based on the node integrated heat task mapping result and the task heat mapping values ​​in the node heat task mapping matrix, and generate a node heat distribution coefficient; and to correct the node integrated heat task mapping result based on the node heat distribution coefficient, thereby generating a node heat resource characterization result. The orchestration generation module is used to sort industrial nodes according to the node thermal resource characterization results and filter out the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix and generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node; The execution control module is used to control the migration of cloud-native tasks between industrial nodes according to the cloud-native thermal management orchestration results; obtain the execution flag of task migration; and calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

[0015] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides a cloud-native management method and platform for industrial thermal management. By transforming industrial thermal states into management objects recognizable by the cloud-native platform, it achieves a leap from single-device-level temperature control to unified platform-level orchestration. Its core advantage lies in resolving the disconnect between traditional thermal management and cloud-native scheduling. Utilizing node thermal task mapping and thermal resource characterization technologies, it deeply integrates physical thermal risks (temperature, load) with logical task distribution (containers, services), constructing a continuous management chain encompassing thermal state perception, thermal constraint formation, and closed-loop handling. This not only overcomes the lag inherent in relying solely on threshold alarms but also predicts the evolution trend of thermal risks during task migration and deployment based on thermal distribution coefficients and migration driving values. This effectively avoids the exacerbation of localized heat accumulation or the spread of thermal risks caused by simply pursuing balanced computing resources, thereby achieving globally optimal scheduling of computing power and thermal energy while ensuring the stability and safety of the industrial site. Attached Figure Description

[0016] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0017] Figure 1 This is a flowchart of a cloud-native management method for industrial thermal management according to the present invention.

[0018] Figure 2 This is a framework diagram of an industrial thermal management cloud-native management platform according to the present invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] In one or more embodiments, such as Figure 1 As shown, a cloud-native management method for industrial thermal management is disclosed, the method comprising the following: S1: Acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data, and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result.

[0021] Specifically, this step establishes a unified mapping relationship between the thermal state of industrial nodes and the deployment status of cloud-native tasks. In actual industrial settings, thermal management monitoring typically occurs at the node level, while task scheduling usually occurs at the platform level. If these are simply stored side-by-side, it becomes difficult to directly determine the current thermal environment of a particular task and to establish a basis for thermal management at the task level. Therefore, this step uses node temperature data, node load data, and task deployment data to create an industrial node thermal task mapping result within the platform. This allows each task to be linked to the current thermal state of its node, further generating a comprehensive thermal stress result at the node level. This allows subsequent steps to directly focus on the "correspondence between the tasks currently being carried by the node and the thermal stress," rather than reassembling the underlying data. Inputs include node temperature data, node load data, and task deployment data. Node temperature data is provided by temperature acquisition devices deployed on industrial nodes. Common implementations include temperature probes inside the cabinet, thermal elements on the device's motherboard, or temperature monitoring modules built into the node. The acquisition results are uploaded to the platform via an edge gateway, and the platform receives the temperature value at the corresponding time according to the node identifier. Node load data is read by the agent program deployed on the node. The agent program can directly call the operating system kernel interface, container runtime monitoring interface, or industrial control system open interface to obtain the current node load status value and report it to the platform at fixed intervals. Task deployment data comes from the cloud-native scheduling system. The platform obtains the deployment correspondence between task instances and nodes through the scheduling interface and organizes it into a deployment identifier indicating whether the task resides on a specified node. For example, in an industrial edge site, node A is currently carrying visual inspection tasks and energy consumption analysis tasks. The edge gateway uploads the temperature sampling value of node A, the node agent reports the load status of node A, and the scheduling system returns a record that both tasks are deployed on node A. Then the platform can obtain the thermal status, load status, and task deployment relationship of node A at the same time and proceed with subsequent mapping processing.

[0022] The platform first aligns the three types of inputs by node identifier, forming node records under the same sampling period. The alignment method here is to use the node as the primary index, writing the temperature sample value, load sample value, and the set of task instances currently deployed on that node into the same node operation record. After alignment, the platform performs state-based processing on the node temperature and load data. Specifically, an operation interval table can be established for each node. This table is generated by the platform in the initial stage of operation based on historical operation data, recording the stable upper and lower bounds of the node's temperature and load. When the current sample value enters the platform, it is converted into a proportional state value according to the corresponding operation interval table for that node, resulting in the node temperature state value and the node load state value. After this processing, both types of state values ​​fall within a unified proportional space, facilitating combination within the same relational expression. This intra-node interval-based approach is adopted because the thermal and load characteristics of industrial nodes often differ. For example, some nodes have strong heat dissipation structures, while others bear high-computation workloads for extended periods. Directly comparing raw values ​​across nodes can easily amplify these structural differences. A proportional state based on each node's own operation interval is more suitable for unified platform management.

[0023] After obtaining the node temperature and load status values, the platform reads the set of task instances on that node and binds each task instance to the node's current state, forming a task-level thermal mapping value. This mapping value is calculated using the following formula: ; in, Represents a node On the task The heat mapping value; Indicates task At the node The deployment identifier on the node is derived from the deployment relationship returned by the cloud-native scheduling system, and its value is 1 when the task resides on this node. Represents a node The temperature status value is obtained by converting the temperature sampling value of the node through the node's operating interval table; Represents a node The load status value is obtained by converting the load sample value of the node through the node operation interval table; The weighting parameters for temperature and load states in the thermal mapping are written in the platform configuration file. In heat-sensitive scenarios, the weighting of temperature can be appropriately increased; in scenarios with large task fluctuations, the weighting of load states can be appropriately increased. The purpose of this relationship is to directly project the thermal state of the node layer to the task layer. In this way, the platform not only knows that a node is currently under high thermal stress, but also understands that every task on that node is currently under the same thermal environment constraints. Taking an industrial quality inspection edge node as an example, if node A's temperature state value is 0.72 and its load state value is 0.58, the configuration parameters... If we take 0.6, and both the visual inspection task and the data aggregation task are deployed on this node, then the heat mapping value for both tasks can be calculated as 0.664. This value is not used as an isolated monitoring quantity within the platform, but rather as a task-level basic quantity for subsequent thermal resource characterization.

[0024] Subsequently, the platform aggregates the task-level hot mapping values ​​by node, generating a node-wide comprehensive hot task mapping result: ; in, Represents a node The comprehensive hot task mapping results, summed within the range of currently deployed nodes. The collection of all task instances on; For the nodes obtained above On the task The heat mapping value. This summary result reflects the node's thermal status, node load status, and task aggregation degree. For example, if two nodes have similar temperature status values, but one node has deployed more tasks, then the overall heat task mapping result of that node will be higher, and the platform will identify it as a higher priority object to manage in subsequent processing. In this way, the platform forms not a simple temperature ranking, nor a simple load ranking, but a node thermal stress result that has integrated task deployment relationships.

[0025] This step outputs the node hot task mapping matrix. Node-wide integrated hot task mapping results The node thermal task mapping matrix characterizes the thermal correlation state of each task on its respective node, while the node comprehensive thermal task mapping result characterizes the current overall thermal pressure level of each industrial node. These two results will then proceed to the next step. Used to determine the state of thermal resources at the node level This step is used to maintain the correspondence between tasks and node thermal states. Through this step, the platform organizes node temperature data, node load data, and task deployment data into a unified node thermal task mapping result, enabling thermal states to enter the cloud-native management chain in a task-aware form for the first time. When subsequent steps generate thermal resource representations based on this, they can retain both the thermal stress results at the node level and the correspondence between the task level and the node thermal environment, thus ensuring that thermal management orchestration is based on the actual deployment state.

[0026] S2: Based on the node integrated thermal task mapping result and the task thermal mapping value in the node thermal task mapping matrix, calculate the dispersion of the task thermal mapping value within the node and generate the node thermal distribution coefficient; correct the node integrated thermal task mapping result based on the node thermal distribution coefficient and generate the node thermal resource characterization result.

[0027] Specifically, a node hot task mapping matrix has already been formed in the previous stage. Node-wide integrated hot task mapping results Building upon this foundation, this step further structures and compresses the node's hot state, enabling it to directly enter the cloud-native scheduling decision-making process. In the previous stage, It is obtained by combining the node temperature status and the node load status. For all The linear summation result within the nodes is essentially derived from the weighted linear superposition model, which corresponds to the basic form of multi-factor load superposition in engineering. Building upon this, this step introduces the factor of "intra-node distribution structure" to further derive the original results, ensuring that nodal thermal resources reflect not only the total amount but also structural differences.

[0028] In specific processing, first at each node Internally, the heat mapping values ​​for all tasks. Statistical analysis is performed. This statistical process is based on the definition of the average value in classical statistics, and it iterates through the task set within each node to obtain the node's average heatmap value. The introduction of this average value stems from the fundamental statistical principle that "the sample mean is used to characterize the overall level," and its physical meaning lies in representing the average thermal stress experienced by a unit task within a node. Subsequently, the deviation within a node is constructed based on the average value. This process originates from the definition of absolute deviation in statistics, used to characterize the dispersion of the sample distribution. Through the analysis of each task... and By taking the absolute value of the difference and summing them, we can obtain the total discrete quantity of the heat mapping value within the node.

[0029] Based on this, the discrete quantity is mapped to the node-integrated thermal task result. Normalized combinations are performed to form the node heat distribution coefficient. For nodes that currently have no tasks deployed, it is directly recorded as... For nodes currently having deployed tasks, calculate using the following formula: ; This formula is derived from a combination of absolute deviation and linear normalization. The numerator is derived from the form of mean absolute deviation in statistics, and the denominator... The result, derived from the linear superposition of the previous stage, is used to unify the scale of the deviation. Because... and All are dimensionless proportional values, and their differences are also dimensionless quantities. The numerator is the sum of dimensionless quantities, and the denominator is... Since it is also a dimensionless quantity, the entire expression maintains consistency. The numerical meaning of this coefficient lies in the fact that when all tasks within a node... When they get close, the molecules are smaller. Lower; when there are large differences between tasks, the molecule increases. The rise in temperature indicates whether the thermal pressure within the node is concentrated or dispersed; when the node is not currently deploying any tasks, The nodal heat distribution coefficient is directly set to 0, and is naturally obtained from subsequent formulas. .

[0030] In obtaining Then, the node integrated hot task mapping results from the previous stage are analyzed. Structural corrections are performed to obtain the nodal thermal resource characterization results. This correction is derived from the scaled-up model in control engineering, which involves superimposing an adjustment term related to the structural state onto the baseline quantity. Its calculation form is as follows: ; in, Represents a node Thermal resource characterization results, Indicates the overall thermal pressure of the node. This represents the heat distribution structure within the node. To adjust the parameters. This formula can be understood as introducing an amplification factor related to the distribution state on the basis of the original thermal pressure. When the thermal pressure is concentrated within the node, Increase, make Compared to Amplified; when the thermal pressure distribution is uniform, Smaller near .because It is a dimensionless quantity. It is a dimensionless quantity. Since it is a proportional parameter, the entire term inside the parentheses is a dimensionless proportional term, multiplied by... It will remain consistent thereafter.

[0031] From the perspective of the formula derivation relationship Depends on and ,and And from Therefore, it is calculated that Completely derived from the output of the previous stage and Derived from; Then directly from and Calculations show that the formation from The continuous derivation chain.

[0032] In the actual calculation process, we will use a specific node as an example for illustration. Suppose that a certain industrial node has three tasks, and the results obtained in the previous stage... The values ​​are 0.62, 0.64, and 0.66 respectively, then the average heat mapping value of the node is... The arithmetic mean of the three is 0.64. At this point, the deviations of each task from the mean are 0.02, 0, and 0.02, respectively, with a sum of deviations of 0.04. If the calculation obtained in the previous stage... The sum of the three is 1.92, then the nodal heat distribution coefficient is... Dividing 0.04 by 1.92 gives approximately 0.0208. When adjusting the parameter... When the value is 0.5, the nodal thermal resource characterization results The value is 1.92 multiplied by (1+0.5×0.0208), which is approximately 1.94. At this point, the correction range is relatively small, reflecting that the thermal pressure distribution inside the node is relatively balanced.

[0033] Consider another node, its If the values ​​are 0.30, 0.55, and 1.05, then the average value is... The value is 0.63, corresponding to deviations of 0.33, 0.08, and 0.42, with a sum of deviations of 0.83. At this point... If it is 1.90, then Dividing 0.83 by 1.90 gives approximately 0.4368. When If we still take 0.5, The product is 1.90 multiplied by (1 + 0.5 × 0.4368), which is approximately 2.31. It can be seen that, under similar total thermal pressure conditions, the final result is significantly different due to the substantial differences in thermal mapping values ​​between tasks. The amplification is significant, which means that this node should be prioritized in subsequent scheduling.

[0034] Through the above calculation process, the nodal thermal resource characterization results can be obtained. At the same time, the original node hot task mapping matrix is ​​retained. .in, Used to describe the overall thermal resource stress of a node, while Maintain the correspondence between tasks and node hot states. In subsequent steps, This will be used as the basis for node selection. This will be used to determine the adjustment order of tasks within a node, thereby realizing a continuous processing chain from node-level judgment to task-level execution.

[0035] S3: Sort the industrial nodes according to the node thermal resource characterization results and select the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix to generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node.

[0036] Specifically, this step builds upon the node thermal resource characterization results already generated in the previous stage. and node hot task mapping matrix This generates cloud-native thermal management orchestration results based on the previous stage. The overall thermal stress of the node and the thermal distribution state of the tasks inside the node have been compressed into a unified node-level state quantity, which is suitable for answering the question "which nodes should be prioritized to limit their continued task carrying". This preserves the correspondence between task instances and their respective node thermal environments, making it suitable for answering the question, "Which tasks within the same node are more suitable for priority adjustment?" This step uses these two types of results as the sole state input, combining node-level thermal resource constraints and task-level thermal correlation strength into migration-driven results and migration-in adaptation results, further forming an orchestration record of "source node - task instance - target node." In the practical scenario of an industrial thermal management cloud-native platform, the processing object in this step is not an abstract resource vector, but rather specific task instances on industrial edge nodes and industrial server nodes. Therefore, the orchestration results can be directly executed by a container scheduler or service orchestrator.

[0037] At the beginning of each scheduling cycle, the platform reads the full data from the state storage module. Establish a node thermal resource status table; simultaneously read the full data from the task status table. The task instances are then grouped into task sets within a node based on their node identifiers. The scheduling controller then performs a two-layer filtering process: the first layer filters by node dimension... The hot resources are sorted from smallest to largest. Nodes ranked higher are written to the priority carrying set, and nodes ranked lower are written to the limited carrying set. The second layer, within each node of the limited carrying set, is based on... Task instances are sorted from highest to lowest priority to form a task hot priority sequence. The resulting node set and task sequence correspond to "which node to process first" and "which task to process first within that node," respectively. In scenarios such as industrial visual inspection, industrial quality inspection, and industrial edge computing, this two-layer processing has direct engineering significance: node-level sorting is responsible for alleviating overall heat accumulation, while task-level sorting is responsible for shortening the time for a single adjustment to produce an effect.

[0038] To unify the two layers of filtering into a single computable task migration-driven result, this step first constructs the task migration-driven value. Its original form originates from the risk-weighted product model in scheduling theory: when the overall risk of a task's node is higher, and the stronger the association between the task and the high-heat environment, the task should have a higher migration priority. Based on this classic idea, the node-level thermal resource state is taken as the global risk term, and the task-level heat mapping value is taken as the local contribution term, to obtain the task migration driving value: ; in, Represents a node On the task Migration-driven values; Represents a node On the task The heat mapping value is obtained from the previous stage based on the node's thermal state, node load state, and task deployment relationship. Represents a node The thermal resource characterization results were obtained in the previous stage based on the overall thermal pressure of the nodes and the internal thermal distribution structure of the nodes. The product relationship in this formula is not a simple addition, but rather uses multiplication to couple the "local thermal pressure of the task" and the "overall thermal pressure of the node" together, so that when either of them increases, They will all increase synchronously. For example, in industrial edge detection scenarios, if the node... of For a certain image reasoning task on a node, the value is 2.10. If it is 0.92, then the task's The value is 1.932; if the log aggregation task is on the same node If it is 0.38, then its The value is 0.798. This allows the platform to prioritize image inference tasks within the same node, as these tasks are more sensitive to thermal stress release.

[0039] In obtaining Afterwards, the platform continues to construct migration adaptation results for each candidate target node. The design of this relation is based on the "benefit minus penalty" principle in resource allocation. The benefit component considers two factors simultaneously: first, whether the task itself is worth migrating, i.e. The size of the target node; secondly, whether the target node is currently suitable as a migration location. In this scheme, the target node adaptability is determined by... This indicates that the term increases when the target node's thermal resources are more abundant and decreases when the target node's thermal resources are scarcer. Simultaneously, a thermal drop suppression term is introduced to distinguish whether "although the target node is available, the difference in thermal resources between it and the source node is sufficient to form effective heat release." This suppression term is constructed using the absolute value of the difference in thermal resources between the source and target nodes; the suppression is stronger when the difference is small and weaker when the difference is large. This yields the migration adaptation results: ; in, Indicates the node On the task Migrate to node The migration and adaptation results; The aforementioned task migration driver value; Indicates candidate target node The thermal resource characterization results are derived from the node state storage of the previous stage; Indicates the source node The results of thermal resource characterization; This represents the thermal drop adjustment parameter, written by the scheduling strategy configuration service, used to adjust the platform's sensitivity to the thermal difference between source and target nodes. The first part of the formula is the target node acceptance benefit term, reflecting the positive benefit of "worthwhile tasks entering nodes with ample thermal resources"; the second part is the thermal drop suppression term, reflecting the industrial field rule that "when the thermal difference between source and target nodes is too small, the migration effect is limited." Because... , and They are all proportional state variables in the same state space. As a proportional adjustment, the entire relation remains consistent in the numerical space.

[0040] This is illustrated using a practical calculation process at an industrial edge site. Let's assume the thermal resource characterization results of source node A are... The thermal resource characterization result of candidate target node B is 2.20. The heatmap value for a certain visual reasoning task on node A is 0.70. The scheduling parameter is 0.90. Let's take 0.20. First, substitute it into the first formula to get the task migration driving value. Then, substituting into the second formula, the target node acceptance benefit term is... The thermal drop suppression term is Therefore, the migration adaptation result If another candidate target node C If the value is 1.15, then the target node's acceptance revenue item is... The thermal drop suppression term is The final migration and adaptation results The platform directly selects node B as the target node for this visual inference task because this node has a higher migration and adaptation result. When extending this process to all tasks within a single node, the scheduling controller will prioritize selecting... For larger tasks, select from the candidate target nodes. The largest node generates a list of task migrations sequentially.

[0041] When deployed on the platform, the scheduling controller records all generated triples as orchestration items of "task instance identifier - source node identifier - target node identifier," along with node-level status tags. For containerized task instances, the controller updates node affinity constraints and triggers rescheduling through the scheduling interface; for industrial edge service replicas, the controller adjusts the replica distribution strategy and issues replacement requests through the service orchestration interface. The final orchestration result consists of two parts: a node-level hot management orchestration result, used to identify which nodes are in a priority carrying state and which nodes are in a limited carrying state; and a task-level migration list, used to identify the source and target nodes for each task to be adjusted. The former is directly derived from the full dataset. The sorting results, the latter being directly derived from and The results are calculated layer by layer. Through this process, the node thermal resource characterization results and task thermal mapping results formed in the previous stage are completely transformed into thermal management orchestration results that can be executed by the cloud-native platform, enabling industrial thermal management to move from state identification to task deployment and reorganization.

[0042] S4: Based on the cloud-native thermal management orchestration results, control the migration of cloud-native tasks between industrial nodes; obtain the execution flag of task migration; calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

[0043] Specifically, after generating the thermal management orchestration results in the previous stage, this step translates those results into the actual thermal management outcomes after execution. The output of the previous stage consists of two parts: one is the node-level orchestration results, identifying which nodes enter the limited-capacity state and which enter the priority-capacity state; the other is the task migration list, identifying the source node and target node for each task instance. Each record in the task migration list retains the corresponding task's thermal mapping value. Source node thermal resource characterization results and the thermal resource characterization results of the target node These quantities are not re-collected, but rather directly derived from the state snapshots attached to the orchestration results generated in the previous stage. Therefore, they can be directly invoked in this step. In the execution scenario of the industrial thermal management cloud-native platform, the core issue addressed in this step is not the judgment of "whether to migrate," but rather "how to update and transform the node's thermal occupancy state into a new execution result after the migration action is actually completed." This process allows the entire solution to move from "thermal resource state-driven orchestration" to "orchestration actions changing the node's thermal state," thus forming an execution result chain that can continuously run within the platform.

[0044] At the execution level, the scheduling controller first reads the task migration list line by line and converts each migration record into a specific scheduling operation. For containerized task instances, the scheduling controller updates the node constraints of the task instance through the cloud-native scheduling interface, writes the target node identifier into the task instance's scheduling rules, and triggers a task rescheduling request. For industrial edge service replicas, the scheduling controller starts a new replica instance on the target node through the service orchestration interface. After the replica enters the ready state, it terminates the original replica on the source node. Throughout the execution process, the platform maintains an execution event association table to record two types of events based on the task instance identifier: one is the event when the target node instance enters the running state, and the other is the event when the source node instance exits the running state. When these two types of events occur simultaneously within the same execution window, the platform marks the migration action as completed and writes an execution flag into the migration list. .in, Indicates task From the source node Migrate to target node The execution flag is set to 1 when the target node starts successfully and the source node instance has been released; otherwise, it is set to 0. This flag is jointly generated by the orchestration system event stream and the node runtime event stream. The former comes from the task status interface of the cloud-native scheduling system, and the latter comes from the instance runtime status records reported by the node agent. Through this event association method, the execution status formed by the platform is not simply "command issued," but rather an execution result synchronized with the actual changes in the node's load.

[0045] After obtaining the execution flags for all migration actions, the platform begins generating the post-execution node hot management results. The calculation of these results is based on the principle of flow conservation in scheduling theory, meaning that the state of a node after execution is equal to its pre-execution state minus the hot occupancy of tasks that have been migrated out, plus the new hot occupancy created by the tasks that have been migrated in at that node. The migration-out portion directly uses the task hot mapping values ​​retained from the previous stage. Because when a task is released from the source node, the source node directly reduces the heat-related share originally occupied by that task; the inbound portion adopts a form where "the task's heat mapping value and the target node's heat resource state work together," since the actual heat occupation of the target node is different when the same task enters a target node with different heat resource states. Based on this idea, the target node's heat resource state is transformed into a bounded saturation coefficient. This ensures that the node's thermal resources approach 1 when they are scarce and approach 0 when they are plentiful, thus yielding the node thermal management result after execution: ; in, Represents a node Thermal management results after this round of scheduling and execution; Represents a node The thermal resource characterization results prior to execution, this value comes from the node-level orchestration results of the previous stage; Indicates task From the source node Migrate to target node The execution flag, which is generated by associating scheduling system events and node execution events; Represents a node On the task The hot mapping value is derived from the node hot task mapping result formed in the first stage and is continuously attached to the task migration record in the second and third stages. Indicates task From the source node Migrate to target node Execution marker; This indicates that the task is in the source node. The heat mapping value on; Represents a node The influx heat absorption coefficient is directly derived from the node's thermal resource state before execution. Its construction form originates from the saturation factor expression in control theory, used to map the node's thermal resource state to the interval between 0 and 1. The entire formula can be understood in three parts: the first part... The first part is the baseline state before node execution; the second part is the hot occupancy released by the actual migration-out task in this round; and the third part is the new hot occupancy formed at the target node by the actual migration-in task in this round. Since all terms in the formula are in a unified state space, this update result can be directly used as the state input for the next round of hot management.

[0046] This formula is a continuation of the previous stage. The previous stage was achieved through... The node hot resource status is determined, and the task migration list determines which tasks should migrate out of which nodes and into which target nodes; this step then introduces actual execution flags based on this. The process transforms "planned migration" into "completed migration," and then updates the node status based on the release amount of the source node and the absorption amount of the target node after the migration. In other words, the previous stage answered "how should it be adjusted," while this step answers "what state the system is in after the adjustment."

[0047] This can be further illustrated by examining the execution process at an industrial edge computing site. Assume that node A has the following thermal resource characterization results before execution. The heat map value of node B before execution is 2.20. Node B's heat resource characterization result before execution is 0.70. Node A has a visual reasoning task with a heat map value of 2.20. The value is 0.90. In the previous stage, this task was added to the task migration list from node A to node B. During this step, the scheduling controller launches a new instance of the visual reasoning task on node B and confirms that the instance is in running state through the orchestration system status interface; subsequently, the original instance is terminated on node A, and the resources are confirmed to have been released through the node agent. If this visual reasoning task is recorded as a task... Then, the execution flag for this migration action should be written as... For node A, there is no migration task in this round, therefore the thermal management result after its execution is as follows: For node B, its migration heat absorption coefficient is: Therefore, the additional heat occupation caused by relocating to this task is approximately Then the thermal management result after node B is executed is The calculation process shows that the heat utilization of node A has decreased significantly, while the heat utilization of node B has increased but remains within a manageable range. The overall system heat distribution has shifted from a single-point clustering state to a more balanced distribution. If other migration actions occur within the same cycle, the platform will process all actions that meet the requirements in the same manner. The migration records are accumulated and added into the formula to form the final execution result of each node at the end of this round.

[0048] In actual deployment, the platform will calculate the... Write the execution results to the execution result storage module, and synchronously write the task deployment relationship after execution to the task status table. The execution result storage module contains... This is used to characterize the changes in heat utilization of each node after the completion of this round of orchestration actions. The latest task deployment relationship in the task status table serves as the deployment basis for regenerating the node hot task mapping relationship in subsequent cycles. The next cycle is regenerated based on the updated task deployment relationship and the newly collected node temperature and node load data. , and In this way, the execution results not only record the actual effect of the current round of choreography, but also provide a continuous operational basis for subsequent cycles.

[0049] This step outputs two results, one of which is the node's thermal management result. Secondly, there is the task deployment relationship after execution. The former characterizes the actual hot occupancy status of each node after the completion of this round of hot management actions, while the latter characterizes the actual residence location of the task instance after execution. These two outputs respectively support the node status update and task deployment update levels, giving the entire method a foundation for continuously running execution results within the platform.

[0050] In one or more embodiments, such as Figure 2 As shown, an industrial thermal management cloud-native management platform is disclosed, the platform comprising: The data acquisition module is used to acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data; and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result. The characterization generation module is used to calculate the dispersion of task heat mapping values ​​within a node based on the node integrated heat task mapping result and the task heat mapping values ​​in the node heat task mapping matrix, and generate a node heat distribution coefficient; and to correct the node integrated heat task mapping result based on the node heat distribution coefficient, thereby generating a node heat resource characterization result. The orchestration generation module is used to sort industrial nodes according to the node thermal resource characterization results and filter out the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix and generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node; The execution control module is used to control the migration of cloud-native tasks between industrial nodes according to the cloud-native thermal management orchestration results; obtain the execution flag of task migration; and calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

[0051] It is worth noting that the specific workflow of the industrial thermal management cloud-native management platform provided in this embodiment of the invention is the same as that of the industrial thermal management cloud-native management method described in the above embodiment, and will not be repeated here.

[0052] This invention also provides an industrial thermal management cloud-native management device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiment of an industrial thermal management cloud-native management method, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above platform embodiments.

[0053] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the industrial thermal management cloud-native management device.

[0054] The aforementioned industrial thermal management cloud-native management device can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. This device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the device may also include input / output devices, network access devices, buses, etc.

[0055] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the industrial thermal management cloud-native management device, connecting all parts of the device via various interfaces and lines.

[0056] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the industrial thermal management cloud-native management device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating platform, at least one application program required for a function, etc.; the data storage area may store data created according to the operation of the controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0057] If the integrated module of the industrial thermal management cloud-native management device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0058] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0059] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A cloud-native management method for industrial thermal management, characterized in that, The method includes: S1: Acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data, and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result; S2: Based on the node integrated thermal task mapping result and the task thermal mapping value in the node thermal task mapping matrix, calculate the dispersion of the task thermal mapping value within the node and generate the node thermal distribution coefficient; correct the node integrated thermal task mapping result based on the node thermal distribution coefficient and generate the node thermal resource characterization result. S3: Sort the industrial nodes according to the node thermal resource characterization results and select the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix to generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node; S4: Based on the cloud-native thermal management orchestration results, control the migration of cloud-native tasks between industrial nodes; obtain the execution flag of task migration; calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

2. The cloud-native management method for industrial thermal management according to claim 1, characterized in that, The acquisition of node temperature data, node load data, and cloud-native task deployment data of industrial nodes specifically includes: Temperature sampling values ​​of the nodes are obtained through temperature acquisition devices on the industrial nodes; Obtain node load status values ​​through an agent program deployed on the industrial nodes; The deployment mapping between task instances and nodes can be obtained through the scheduling interface of the cloud-native scheduling system.

3. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The node temperature data and the node load data are normalized, specifically including: An operating range table is established for each industrial node within the platform. The operating range table records the upper and lower limits of stable operation of temperature and load for each node. The node temperature sampling value and the node load sampling value are mapped to the corresponding operating interval table, and converted into node temperature status value and node load status value in a uniform proportional space.

4. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The calculation of the node hot task mapping matrix specifically includes: The task instance is bound to the node temperature status value and node load status value of the node it belongs to; Based on preset weighting parameters, the node temperature status value and the node load status value are weighted and combined to generate a task-level thermal mapping value. The task-level hot mapping values ​​of each task instance are organized by node to form a node hot task mapping matrix.

5. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The degree of dispersion of the task heat mapping values ​​within the computing node specifically includes: Calculate the average value of all task-level heatmaps within the node; Calculate the sum of the absolute values ​​of the differences between each task-level heat mapping value and the average value; The sum of the absolute values ​​of the differences is normalized with the node comprehensive thermal task mapping result to generate the node thermal distribution coefficient.

6. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The generation of cloud-native thermal management orchestration results, which includes source nodes, task instances, and target nodes, specifically includes: The node thermal resource characterization results are used as global risk items, and the task-level thermal mapping values ​​are used as local contribution items to calculate task migration driving values. The migration adaptation result is calculated based on the task migration driving value, the thermal resource characterization result of the candidate target node, and the thermal resource difference between the source node and the target node. A task migration list is generated based on the task migration driver value and the migration adaptation result.

7. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The process of ranking industrial nodes based on the node thermal resource characterization results specifically includes: Establish a node thermal resource status table and sort the node thermal resource characterization results from smallest to largest. Write the industrial nodes that are ranked higher into the priority carrying set and the industrial nodes that are ranked lower into the restricted carrying set. A list of nodes to be processed is generated based on the restricted bearer set.

8. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The step of calculating and updating the node thermal management result after execution based on the execution flag, the node hot task mapping matrix, and the node thermal resource characterization result specifically includes: For the source node, subtract the task-level heat mapping value of the migrated task from the heat resource characterization result before its execution to obtain the heat management result after the source node is executed. For the target node, the thermal management result after execution is obtained by subtracting the task-level thermal mapping value of the outgoing task from the thermal resource characterization result before execution, and then adding the product of the task-level thermal mapping value of the incoming task and the incoming thermal absorption coefficient of the target node.

9. The industrial thermal management cloud-native management method according to claim 1, characterized in that, The control cloud-native task is migrated between industrial nodes, specifically including: For containerized task instances, update the node affinity constraints of the task instance through the cloud-native scheduling interface and trigger rescheduling; For industrial edge service replicas, a new replica is started on the target node through the service orchestration interface, and the original replica on the source node is terminated after the new replica is ready.

10. An industrial thermal management cloud-native management platform, characterized in that, The platform includes: The data acquisition module is used to acquire node temperature data, node load data, and cloud-native task deployment data of industrial nodes; normalize the node temperature data and node load data to obtain node temperature status values ​​and node load status values; calculate and generate a node thermal task mapping matrix based on the node temperature status values, node load status values, and cloud-native task deployment data; and summarize the data in the node thermal task mapping matrix to generate a node comprehensive thermal task mapping result. The characterization generation module is used to calculate the dispersion of task heat mapping values ​​within a node based on the node integrated heat task mapping result and the task heat mapping values ​​in the node heat task mapping matrix, and generate a node heat distribution coefficient; and to correct the node integrated heat task mapping result based on the node heat distribution coefficient, thereby generating a node heat resource characterization result. The orchestration generation module is used to sort industrial nodes according to the node thermal resource characterization results and filter out the restricted carrying set; sort the tasks in each node in the restricted carrying set according to the node thermal task mapping matrix and generate a task thermal priority sequence; calculate the task migration driving value and migration adaptation result according to the node thermal resource characterization results and the task thermal mapping value in the node thermal task mapping matrix, and generate a cloud-native thermal management orchestration result containing the source node, task instance and target node; The execution control module is used to control the migration of cloud-native tasks between industrial nodes according to the cloud-native thermal management orchestration results; obtain the execution flag of task migration; and calculate and update the node thermal management results after execution based on the execution flag, the node thermal task mapping matrix, and the node thermal resource characterization results.

Images