Supply chain iot data fusion and real-time decision system based on distributed computing power

Abstract

The application relates to the field of computing power scheduling and optimization, and discloses a supply chain Internet of Things data fusion and real-time decision system based on distributed computing power, which comprises the following steps: acquiring operation state information of a data center and related facilities; performing time synchronization, abnormality detection, multi-source heterogeneous data fusion and standardization processing on collected data; quantifying computing resources, cooling capacity, power limitation and service quality of each data center; jointly solving multi-data center task allocation, computing power migration, node power configuration, cooling strategy and vehicle or warehouse operation scheduling; adopting a rolling execution strategy to generate current period control quantity and convert the same into server resource allocation, task migration, cooling adjustment and power allocation instructions; issuing the instructions to a computing cluster scheduling system and a basic facility control system to generate an execution result; monitoring the execution result, business indexes, energy consumption and equipment states; updating an optimization model and a scheduling strategy; and providing the updated model and strategy to a scheduling optimization module. The application has the advantages of improving energy utilization efficiency and computing power utilization rate.

Description

Technical Field

[0001] This invention relates to the field of computing power scheduling and optimization, specifically to a supply chain IoT data fusion and real-time decision-making system based on distributed computing power. Background Technology

[0002] With the rapid growth of cloud computing, large-scale artificial intelligence training, and scientific computing, the computing power of data centers continues to expand, significantly increasing the power consumption of computing and cooling equipment. This has led to increasing pressure on energy costs and carbon emissions. Currently, data center energy efficiency optimization mainly relies on single-room temperature control strategies, power limiting measures, and empirical rule adjustments, while reducing energy consumption by improving equipment energy efficiency ratios and using natural cooling sources. However, under high-density server clusters and time-varying business loads, traditional static configuration and distributed control methods are insufficient to meet real-time and security requirements, easily leading to problems such as localized overload, cooling redundancy, and resource idleness. Furthermore, the deployment of multiple data centers across regions presents challenges related to energy prices, Differences exist in climate conditions and power supply structures, leading to increasingly complex demands for load dispatching and power allocation. Existing systems generally lack a unified coordination mechanism for multi-center resources, power constraints, temperature control capabilities, and service quality. There is also a lack of linkage between IT resource management and infrastructure control, making it difficult to form a closed loop for energy efficiency optimization. Furthermore, the electricity market is gradually introducing time-of-use pricing, carbon emission factors, and renewable energy access policies, but existing technologies have failed to fully utilize these signals in business dispatching to achieve comprehensive cost reduction and emission reduction. Therefore, it is essential to design a supply chain IoT data fusion and real-time decision-making system based on distributed computing power to improve energy utilization efficiency and computing power utilization. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a supply chain IoT data fusion and real-time decision-making system based on distributed computing power, which has the advantages of improving energy efficiency and computing power utilization, and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goals of improving energy efficiency and computing power utilization, this invention provides the following technical solution: a supply chain IoT data fusion and real-time decision-making system based on distributed computing power, comprising: Status Acquisition Module: Acquires operational status information of the data center and related facilities, including server resource utilization, task load, temperature and cooling parameters, power supply and load status, carbon emission factors and energy price signals, and performs time synchronization, anomaly detection, multi-source heterogeneous data fusion and standardization processing on the acquired data; Resource Modeling Module: Receives data output from the Status Acquisition Module, quantifies the computing resources, cooling capacity, power limitations, and service quality of each data center, and predicts the available resource range, cooling capacity boundaries, and energy consumption characteristics based on environmental conditions and task requirements, generating analysis results; Scheduling optimization module: Based on the analysis results, it jointly solves the multi-data center task allocation, computing power migration, node power configuration, cooling strategy and vehicle or warehouse operation scheduling, and generates the control quantity for the current time period by adopting a rolling execution strategy. The strategy distribution module converts the control quantities for the current time period into server resource allocation, task migration, cooling adjustment, and power allocation instructions, and distributes them to the computing cluster scheduling system and infrastructure control system to generate execution results. Operational feedback module: Based on the execution results, monitor the execution results, business indicators, energy consumption and equipment status, update and optimize the model and scheduling strategy, and provide the updated model and strategy to the scheduling optimization module.

[0005] Preferably, the process of performing time synchronization, anomaly detection, multi-source heterogeneous data fusion, and standardization on the collected data is as follows: The collected data is uniformly calibrated according to the timestamp, and abnormal data points are detected, removed, or corrected. Combine data from different sources to perform feature mapping and consistency matching; A unified feature output is generated through standardization processing.

[0006] Preferably, the process of quantifying the computing resources, cooling capacity, power limitations, and service quality of each data center is as follows: Receive feature data output from the status acquisition module after time synchronization, anomaly detection, multi-source fusion and standardization processing; Analyze the server resources, cooling system capacity, and power constraints of each data center; The available computing power, cooling capacity, and power limits are quantified, and node service quality indicators are calculated to form a resource capacity table.

[0007] Preferably, the process for generating analysis results is as follows: Based on the resource capacity table, a basic quantitative analysis is conducted on the available computing power, cooling capacity, and power limit of each data center, and the service capacity and energy consumption response characteristics of each node under different load levels are calculated. By combining historical load patterns, environmental data, and future task requirements, we can predict the resource utilization, cooling capacity limits, and power consumption of each data center in the future period, and assess potential bottleneck nodes and load fluctuation risks. The prediction results are combined with the capacity constraints, service quality requirements and energy cost factors of each data center to generate analysis results that can directly support scheduling optimization.

[0008] Preferably, the process of jointly solving for multi-datacenter task allocation, computing power migration, node power configuration, cooling strategies, and vehicle or warehouse operation scheduling is as follows: Based on the comprehensive analysis results, combined with the capacity constraints, service quality requirements, energy costs and carbon emission indicators of each data center, multi-objective coordination and optimization calculations are performed to generate scheduling and resource allocation strategies. The task allocation, computing power migration, node power configuration, and cooling strategy are jointly solved. During the optimization process, cross-warehouse and vehicle operation scheduling constraints are also taken into consideration. Preliminary control variables are generated using a rolling programming solution method.

[0009] Preferably, the process of generating the control quantity for the current time period is as follows: Based on the initial control volume, the executable parts within the current time window are screened and finely adjusted; By combining real-time collected status data and environmental conditions, the task allocation, computing power migration, power allocation and cooling strategies are dynamically corrected; Based on the rolling update mechanism, the executable control instructions for the current time period are calculated, and the possible adjustment needs for the next time window are predicted to generate the control quantity for the current time period.

[0010] Preferably, the process of converting commands into server resource allocation, task migration, cooling adjustment, and power allocation instructions is as follows: Based on the control variables for the current time period, the task allocation, computing power migration, node power configuration, and cooling strategies involved in the control variables are analyzed. Identify the target node, task type, and resource requirements corresponding to each scheduling instruction; Considering the execution capabilities, network communication latency, power and cooling constraints, service quality indicators, and security constraints of each data center and node, abstract control variables are mapped to specific executable operation instructions.

[0011] Preferably, the process for generating the execution result is as follows: Based on executable operation instructions, the instructions are decomposed, packaged, and formatted in accordance with the interface specifications of the data center scheduling system and infrastructure control system; The command is sent to the control modules of each data center and related facilities through the communication network. The control system executes the task load adjustment, virtual machine migration, cooling system adjustment and node power allocation according to the command. During execution, the execution status, resource utilization, task completion status, and abnormal events of each node are collected in real time. The data collected during the execution process is locally aggregated and preliminarily analyzed to generate an execution result report.

[0012] Preferably, the process for monitoring execution results, business indicators, energy consumption, and equipment status is as follows: Based on the execution result report, the execution results are analyzed and interpreted. Calculate task execution deviations, the difference between actual and planned energy consumption, and abnormal equipment status indicators; The analysis results are comprehensively evaluated with business performance indicators, energy consumption data, and equipment status to generate feedback information that can be optimized.

[0013] Preferably, the process for monitoring execution results, business indicators, energy consumption, and equipment status is as follows: Based on the execution result report, the execution results are analyzed and interpreted. Calculate task execution deviations, the difference between actual and planned energy consumption, and abnormal equipment status indicators; The analysis results are comprehensively evaluated with business performance indicators, energy consumption data, and equipment status to generate feedback information that can be optimized.

[0014] Preferably, the process of providing the updated model and strategy to the scheduling optimization module is as follows: Based on the feedback information that can be optimized, various indicators are classified, processed, and summarized for analysis; Calculate the resource utilization deviation, power consumption difference, cooling capacity fluctuation and service quality deviation of each data center, and perform trend analysis and anomaly verification in combination with historical operating data; Based on the analysis results, the resource capacity parameters, node power constraints, cooling strategies, and multi-objective optimization weights of each data center are dynamically adjusted. The optimization model is calibrated by adjusting the prediction model coefficients, updating the rolling scheduling window parameters, and weighting the multi-objective optimization algorithm. The updated optimization model and scheduling strategy are output to the scheduling optimization module to form a rolling optimization mechanism for the next time window.

[0015] Compared with existing technologies, this invention provides a supply chain IoT data fusion and real-time decision-making system based on distributed computing power, which has the following beneficial effects: This invention constructs a closed-loop control process through status acquisition, resource modeling, scheduling optimization, strategy issuance, and operational feedback to achieve real-time coordination and dynamic management of computing resources and energy systems in multi-datacenter scenarios within a supply chain environment. The system collects and predicts operational parameters such as server load, cooling status, power supply and demand, and carbon factor, forming an analysis of resource capacity and energy consumption characteristics. Combined with a rolling scheduling strategy, it achieves task allocation, computing power migration, node power configuration, and cooling adjustment. By continuously sensing and responding to load changes and energy price signals, the system can improve computing power capacity and energy efficiency under business demands and service quality constraints, reducing resource idleness and energy waste, and achieving refined control of computing power supply and energy consumption allocation. Furthermore, the execution feedback mechanism corrects operational deviations and equipment status in real time, ensuring that the scheduling strategy continuously adapts to actual operating conditions and energy conditions, enhancing system stability and adaptability, and supporting the dynamic response and efficient operation of supply chain businesses. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, the supply chain IoT data fusion and real-time decision-making system based on distributed computing power includes: Status Acquisition Module: Acquires operational status information of the data center and related facilities, including server resource utilization, task load, temperature and cooling parameters, power supply and load status, carbon emission factors and energy price signals, and performs time synchronization, anomaly detection, multi-source heterogeneous data fusion and standardization processing on the acquired data; The process of time synchronization, anomaly detection, multi-source heterogeneous data fusion, and standardization of the acquired data in the status acquisition module is as follows: The collected data is uniformly calibrated according to the timestamp, and abnormal data points are detected, removed, or corrected. The timestamp format in the data stream is parsed, and the time standards of different systems are uniformly converted to Coordinated Universal Time (UTC) to ensure that all types of data are aligned under the same time base. Based on a preset time window, data collected from different sources are aggregated and interpolated. For data with inconsistent sampling frequencies, missing values ​​are filled in using linear or spline interpolation to form a continuous time series. The baseline deviation value is calculated for each data point using the moving median and median absolute deviation; if the value exceeds a set threshold, it is marked as an anomaly. For detected anomalous data points, corrections are made based on linear interpolation or nearest neighbor values ​​of the upstream and downstream continuous time series to ensure the continuity and integrity of the sequence. This process performs independent detection on key parameters such as current, voltage, temperature, wind speed, load utilization, and node frequency to ensure that different types of data are not affected by instantaneous anomalies before entering feature construction.

[0019] Combine data from different sources to perform feature mapping and consistency matching; A unified data model is established for each type of data, including attributes such as IT operation profile, node utilization, rack temperature, cooling power, energy price, grid marginal emission factor, and network latency. Through field mapping and unit conversion, the characteristics of different sources are normalized to the same dimension. For example, power is unified to kilowatts, temperature to degrees Celsius, and latency to milliseconds. Then, the data are associated according to the correspondence between nodes and time windows to form a unified feature set across domains.

[0020] A unified feature output is generated through standardization processing; The fused feature set is standardized to generate a unified output for scheduling and optimization. Each feature is standardized by subtracting the moving median and dividing by the absolute deviation of the median, so that features with different dimensions and amplitude ranges can participate in the calculation on the same scale. The standardization process takes into account the moving window length and update frequency, so that the feature values ​​can be dynamically adjusted with the real-time changes of the data stream, while maintaining historical trends and short-term fluctuation information. The standardized output forms a multi-dimensional feature matrix.

[0021] Resource Modeling Module: Receives data output from the Status Acquisition Module, quantifies the computing resources, cooling capacity, power limitations, and service quality of each data center, and predicts the available resource range, cooling capacity boundaries, and energy consumption characteristics based on environmental conditions and task requirements, generating analysis results; The process of quantifying the computing resources, cooling capacity, power limitations, and quality of service for each data center in the resource modeling module is as follows: Receive feature data output from the status acquisition module after time synchronization, anomaly detection, multi-source fusion and standardization processing; The data includes server node utilization, task queue length, CPU or GPU or memory usage, cooling system airflow, chilled water temperature, liquid cooling circuit status, grid power usage, and service indicators such as response time, task latency, and error rate. The receiving process parses the data in the correct format, stores the feature matrices of different centers and nodes in a unified local cache, and establishes a time index to ensure that subsequent analysis and quantification processes are based on data input within the same time window.

[0022] Analyze the server resources, cooling system capacity, and power constraints of each data center; Resource analysis is performed on the server clusters of each data center, including the number of computing cores, CPU or GPU frequency and power consumption curves, memory capacity, and node status. The analysis steps involve statistically analyzing the historical load and current utilization of nodes to calculate schedulable computing power, the number of parallel jobs, and peak load capacity. Simultaneously, the hardware status and availability of each node are recorded. For heterogeneous clusters, a unified computing power conversion method is used to convert the computing power of different types of processors into a unified computing power unit. The cooling system and power constraints of each data center are analyzed, obtaining the rated cooling capacity, real-time load, and temperature distribution of the air-cooled, liquid-cooled, and chilled water circulation systems. The maximum server power and temperature redundancy that each cooling loop can support are calculated, generating a cooling capacity matrix. Power constraint analysis includes total power distribution capacity, power limits for each power supply loop, and instantaneous load allocation capacity. Combined with current grid power and demand response constraints, the available power range is quantified. The analysis process uses historical measurement curves and real-time sensor data from the data center for modeling to ensure dynamic matching between power constraints and cooling capacity.

[0023] Quantify available computing power, cooling capacity and power limits, and calculate node service quality indicators to form a resource capacity table. Service quality metrics are calculated based on node task queues, response latency, and task completion rate, including percentile response time, task timeout rate, number of failure retries, and system availability. Metrics are calculated separately for tasks of different priorities, and then summarized into a service quality matrix by data center, rack, and node dimensions. Finally, the available computing power, cooling capacity, power limit, and service quality metrics of the servers are integrated into a resource capacity table for each data center. The table is arranged by node, rack, server room, and data center hierarchy, and records the current value, historical reference value, and timestamp information of each capability.

[0024] The process of generating analysis results in the resource modeling module is as follows: Based on the resource capacity table, a basic quantitative analysis is conducted on the available computing power, cooling capacity, and power limit of each data center, and the service capacity and energy consumption response characteristics of each node under different load levels are calculated. The server processing capacity of each node is discretized according to CPU, GPU, and memory utilization to form multiple load tiers. Secondly, the available cooling capacity under each load tier is calculated by combining the cooling system capacity and real-time environmental parameters such as data center temperature, humidity, and airflow. Thirdly, the node power usage under each load tier is calculated according to power limit constraints and real-time power consumption data. Finally, the load tiers, available computing power, cooling capacity, and power consumption are integrated to form a node load-service capacity and energy consumption response matrix.

[0025] By combining historical load patterns, environmental data, and future task requirements, we can predict the resource utilization, cooling capacity limits, and power consumption of each data center in the future period, and assess potential bottleneck nodes and load fluctuation risks. A short-term forecasting model is established based on historical task completion records, job types, load fluctuation patterns, and temperature change trends to predict the node load demand in future periods. Secondly, the cooling capacity boundary is predicted by combining the historical power consumption curve of the cooling system with the environmental parameter change patterns, and the temperature safety constraints of nodes under different task combinations are calculated. Thirdly, the power consumption of each node in future periods is estimated by referring to power supply and power limits. Finally, the predicted load, cooling capacity, and power consumption are combined to form a future resource state matrix.

[0026] The prediction results are combined with the capacity constraints, service quality requirements and energy cost factors of each data center to generate analysis results that can directly support scheduling optimization. The feasibility of resource allocation is verified by comparing the predicted load, cooling capacity boundaries, and power consumption of each node with the actual capacity constraints. Secondly, service quality indicators, such as task response time, completion rate, and latency distribution, are used to calculate the node service capacity score under different scheduling schemes. Thirdly, energy cost information, including electricity prices, carbon emission costs, and renewable energy output, is incorporated into the calculation to form a multi-dimensional optimization index table. Finally, the node capacity, service quality, energy consumption, and cost information are integrated and output to form a data structure that can be directly input into the scheduling optimization algorithm, including a node state matrix, load prediction matrix, cooling boundary matrix, and power consumption matrix, providing complete data support for real-time scheduling and multi-objective optimization.

[0027] Scheduling optimization module: Based on the analysis results, it jointly solves the multi-data center task allocation, computing power migration, node power configuration, cooling strategy and vehicle or warehouse operation scheduling, and generates the control quantity for the current time period by adopting a rolling execution strategy. The scheduling optimization module performs a joint solution for multi-datacenter task allocation, computing power migration, node power configuration, cooling strategies, and vehicle or warehouse operation scheduling as follows: Based on the comprehensive analysis results, combined with the capacity constraints, service quality requirements, energy costs and carbon emission indicators of each data center, multi-objective coordination and optimization calculations are performed to generate scheduling and resource allocation strategies. Based on the comprehensive analysis results, the available computing power, cooling capacity, power limit, and historical load prediction matrix of each data center are used as inputs. Service quality indicators, energy cost information, and carbon emission calculation boundaries are also incorporated to construct a multi-objective optimization model. This model quantifies the priority, time window, and constraints of transportation tasks, computing tasks, and warehousing operations to form a unified objective function. By establishing a set of constraints, including node computing power boundaries, cooling capacity constraints, power caps, service level indicators, and time and capacity constraints for cross-warehouse and vehicle operations, the model achieves joint solutions for task allocation, computing power migration, node power configuration, and cooling strategies.

[0028] The task allocation, computing power migration, node power configuration, and cooling strategy are jointly solved. Each task is discretized according to its computing resource requirements, execution time, latency sensitivity, and priority to form a task queue. Next, the feasibility and power consumption cost of different nodes executing tasks are calculated using the node capability matrix and cooling boundary matrix. Then, the computing power migration scheme between nodes is solved by integer programming or heuristic optimization algorithm, and the power allocation and cooling load corresponding to each scheme are calculated. Finally, a joint scheduling matrix is ​​generated.

[0029] During the optimization process, cross-warehouse and vehicle operation scheduling constraints are also taken into consideration. A time-series model is established for picking, replenishment, and loading / unloading operations within the warehouse, considering task duration, priority, personnel and equipment availability, and calculating the impact of node load on the takt time of warehousing operations. A capacity constraint and route optimization model is established for vehicle scheduling, combining vehicle energy consumption curves and carbon emission coefficients to calculate the impact of task allocation on transportation energy consumption and emissions. In the optimization solution, warehousing operation time-series and vehicle route constraints are used as additional constraints of the joint objective function, so that task allocation, computing power migration, node power and cooling strategies form a coordinated and consistent scheduling strategy with warehousing and transportation operations.

[0030] Preliminary control variables are generated using a rolling planning solution method. The scheduling cycle is divided into continuous time windows. For each window, the executable preliminary control variables are calculated, including task allocation nodes, migration paths, power limits, and cooling modes. Constraints are filtered for each control variable to eliminate operations that exceed node capabilities, cooling boundaries, or power caps. The preliminary control variables are then input into the rolling planning framework to generate the preliminary control variables.

[0031] The process of generating the control quantity for the current time period in the scheduling optimization module is as follows: Based on the initial control volume, the executable parts within the current time window are screened and finely adjusted; For each task allocation, computing power migration scheme, node power configuration, and cooling strategy, an executability determination is made, including whether the current node load is within the allowable range, whether the cooling system has the execution capability, whether the power usage is below the maximum limit, and whether the task time window covers the current time period. For control quantities that are not executable or partially executable, they are adjusted in segments according to priority and node capability, task nodes are reassigned or migration paths are modified, and the initial control quantities are corrected into a set of fully executable strategies within the current time window, forming an operable control candidate set.

[0032] By combining real-time collected status data and environmental conditions, the task allocation, computing power migration, power allocation and cooling strategies are dynamically corrected; The system acquires real-time information on node computing power usage, temperature, power consumption, cooling system load, and job execution progress, and matches this information with the control candidate set item by item. For situations where a node's load exceeds its limit, its cooling capacity is insufficient, or its power is close to its limit, the system adjusts task allocation, computing power migration, and power configuration in real time, including migrating some tasks to backup nodes, adjusting the power allocation ratio, switching cooling modes, or delaying low-priority jobs. At the same time, the system adapts cooling strategies and power usage based on environmental conditions such as external temperature, air circulation, or energy price fluctuations, so that the dynamically corrected control quantities can be directly executed within safe and constrained ranges.

[0033] Based on the rolling update mechanism, the executable control instructions for the current time period are calculated, and the possible adjustment needs for the next time window are predicted to generate the control quantity for the current time period. For each node, task, and job, the execution sequence for the current time period is calculated, including task start time, migration path, power usage curve, and cooling equipment on / off status. At the same time, the load change, node available computing power, and cooling capacity are estimated using a predictive model for the next time window, and possible adjustment needs are marked to provide a reference for subsequent rolling updates. Finally, the control quantities for the current time period are output, including node-level control instructions, job-level execution order, and cooling strategy, ensuring that each control instruction can be directly issued and executed within the constraints of time, power, and resources.

[0034] The strategy distribution module converts the control quantities for the current time period into server resource allocation, task migration, cooling adjustment, and power allocation instructions, and distributes them to the computing cluster scheduling system and infrastructure control system to generate execution results. The process of converting the policy distribution module into server resource allocation, task migration, cooling adjustment, and power allocation instructions is as follows: Based on the control variables for the current time period, the task allocation, computing power migration, node power configuration, and cooling strategies involved in the control variables are analyzed. The system reads the execution object, execution order, and resource requirements of each control instruction in the control variables, including task ID, compute node ID, target node ID, estimated runtime, required computing power, power usage plan, and cooling mode. Each instruction is then decomposed into a structured form to generate a task list, node load plan, power allocation table, and cooling control table, providing a data foundation for subsequent instruction mapping. During the parsing process, the system performs preliminary verification of the instruction's validity by combining node status and real-time environmental data, eliminating invalid or unexecutable instructions.

[0035] Identify the target node, task type, and resource requirements corresponding to each scheduling instruction; The task types are categorized into computational tasks, storage tasks, data migration tasks, and hybrid jobs. Target nodes are matched for each task, including available computing power, memory capacity, storage I / O capabilities, and network bandwidth. The power budget, cooling load, and execution time window required for each task are calculated. A complete mapping table is formed between task types, target nodes, and resource requirements to ensure that each instruction logically corresponds to physical resources.

[0036] Considering the execution capabilities, network communication latency, power and cooling constraints, service quality indicators and security constraints of each data center and node, the abstract control variables are mapped to specific executable operation instructions; By leveraging node load and resource status, task allocation instructions are transformed into node-level execution commands, including task initiation, migration paths, resource locking, and priority settings. Computing power migration schemes are mapped into data transmission operation instructions and computation switching commands, taking into account network latency and data consistency verification. Power allocation strategies are mapped into node power management commands, generating fan adjustment, liquid cooling pump speed, or chiller start / stop commands based on cooling system load and temperature thresholds. Simultaneously, task execution order constraints and access control are generated based on service quality indicators and security policies. The final set of commands can be directly distributed to servers and cooling equipment, achieving a closed-loop mapping from control quantities to physical execution.

[0037] The process of generating execution results in the strategy distribution module is as follows: Based on executable operation instructions, the instructions are decomposed, packaged, and formatted in accordance with the interface specifications of the data center scheduling system and infrastructure control system; Each operation instruction is decomposed according to the target node and task type, including sub-instructions for task initiation, virtual machine migration, power adjustment, and cooling regulation; then it is packaged according to the interface specifications of each system, including command header information, task identifier, execution priority, resource allocation parameters, and timestamp; finally, the instruction content is formatted to generate a standard message structure that can be sent through communication protocols, ensuring that the data center scheduling system and infrastructure control system can correctly parse and execute the instructions.

[0038] The command is sent to the control modules of each data center and related facilities through the communication network. The control system executes the task load adjustment, virtual machine migration, cooling system adjustment and node power allocation according to the command. Using the data center's internal network and control system interface, instructions are sent to each server node, virtual machine management module, and cooling equipment controller in batch or real-time stream. After receiving the instructions, the control system executes operations according to the node's capabilities and resource constraints, including starting or migrating computing tasks, adjusting server power settings, and starting or adjusting the cooling equipment's operating mode. During execution, the control system monitors the task status and resource allocation to ensure that instructions are completed in the expected order, with the required accuracy, and within the expected time window.

[0039] During execution, the execution status, resource utilization, task completion status, and abnormal events of each node are collected in real time. The node monitoring agent collects metrics such as CPU utilization, memory usage, storage I / O, network bandwidth, virtual machine migration progress, power consumption, and cooling equipment status; it also monitors task completion, including task startup, runtime, migration completion markers, and task failure or delay events; and it stores the collected data in a structured manner using timestamps and node identifiers to provide accurate data support for performance evaluation and anomaly analysis.

[0040] The data collected during the execution process is locally aggregated and preliminarily analyzed to generate an execution result report; The system performs statistical calculations on resource utilization, task completion, power consumption, and cooling load data collected from each node, generating time series, node load distribution, and power consumption curves. It detects abnormal events or deviations during execution, marking task delays, migration failures, or resource overruns. Based on a predefined analysis model, it calculates the execution efficiency indicators of each node and the data center as a whole, generating standardized reports including task completion rate, node resource utilization, power consumption statistics, and cooling equipment response status, providing basic data for upper-level scheduling optimization and control adjustments.

[0041] Operational feedback module: Based on the execution results, monitor the execution results, business indicators, energy consumption and equipment status, update and optimize the model and scheduling strategy, and provide the updated model and strategy to the scheduling optimization module.

[0042] The process of monitoring execution results, business metrics, energy consumption, and equipment status in the operation feedback module is as follows: Based on the execution result report, the execution results are analyzed and interpreted. The task completion records, resource utilization data, power consumption, and cooling equipment status in the execution results report are stored in a structured manner according to nodes, timestamps, and task types. Then, the task execution status is analyzed, including matching the task start, running, migration completion time, and execution duration, comparing the planned task schedule with the execution time, and identifying task delays, migration failures, or early completion. At the same time, the node resource usage, virtual machine allocation, and cooling equipment status are parsed to provide the raw data foundation for subsequent deviation calculations.

[0043] Calculate task execution deviations, the difference between actual and planned energy consumption, and abnormal equipment status indicators; Calculate the execution deviation for each task, including the difference between task completion time and planned time, migration delay, and resource allocation deviation; statistically analyze the difference between actual and planned energy consumption, stratified by server node, data center, and task type, including CPU power consumption, memory power consumption, storage and network power consumption, and cooling system energy consumption; simultaneously, monitor and calculate equipment status indicators, including excessive server load, abnormal temperature, abnormal operation of fans and cooling pumps, and power supply fluctuations, and identify abnormal states through indicator quantification and threshold comparison.

[0044] The analysis results are comprehensively evaluated with business performance indicators, energy consumption data and equipment status to generate feedback information that can be optimized. By comparing and analyzing task completion rate, latency metrics, node resource utilization, and business performance metrics such as throughput and request response time, performance bottlenecks and resource-insufficient areas are identified. Combined with energy consumption deviations and equipment status anomalies, operating costs, energy consumption levels, and potential equipment risks are assessed. Finally, structured feedback information is generated, including key anomaly metrics for each node and data center, energy consumption deviations, task latency statistics, and optimization suggestion data tables.

[0045] The process by which the updated model and strategy are provided to the scheduling optimization module in the runtime feedback module is as follows: Based on the feedback information that can be optimized, various indicators are classified, processed, and summarized for analysis; Data on task completion deviation, node resource utilization, power consumption, cooling system status, and service quality deviation are stored in a multi-dimensional structure according to data center, server node, time window, and task type; similar indicators are aggregated and calculated, including mean, variance, maximum and minimum values, and percentile statistics.

[0046] Calculate the resource utilization deviation, power consumption difference, cooling capacity fluctuation and service quality deviation of each data center, and perform trend analysis and anomaly verification in combination with historical operating data; Calculate the deviation between the actual and planned resource utilization rates of each node and data center in the current period, including CPU, memory, storage, and network bandwidth usage; statistically analyze the deviation between actual power consumption and cooling system usage and design capacity to identify potential overload or inefficient areas; compare the deviation results with historical data sequences, and use moving average or weighted regression methods to analyze trends and detect long-term anomalies or periodic fluctuations.

[0047] Based on the analysis results, the resource capacity parameters, node power constraints, cooling strategies, and multi-objective optimization weights of each data center are dynamically adjusted. Based on the results of deviation and trend analysis, the available computing power, power limits, cooling capacity boundaries, and load allocation ratios for each data center were redefined; the node power allocation strategy was adjusted to ensure that the power of high-load nodes meets performance requirements, while controlling the overall energy consumption within the allowable range; and the multi-objective optimization weights, including task latency, energy cost, and equipment lifespan weights, were reallocated according to task type and service quality indicators.

[0048] The optimization model is calibrated by adjusting the prediction model coefficients, updating the rolling scheduling window parameters, and weighting the multi-objective optimization algorithm. The coefficients of the resource utilization prediction model are regressed and corrected based on historical operating data and the latest feedback results to ensure that the model output matches the actual operating status; the rolling scheduling window length and time step are updated to match the current load fluctuations and task execution rhythm; the objective function weights of the multi-objective optimization algorithm are adjusted according to the optimization results and deviation analysis so that task allocation, power regulation and cooling strategies reflect the latest operating constraints in the optimization calculation.

[0049] The updated optimization model and scheduling strategy are output to the scheduling optimization module to form a rolling optimization mechanism for the next time window; The revised parameters and strategies are encapsulated in a standardized interface format, including node resource constraints, power allocation strategies, cooling strategies, and multi-objective weight settings. They are then passed to the computing engine through the scheduling optimization module interface for use in the next round of rolling planning. At the same time, log records are generated, including parameter modification history, strategy adjustment basis, and timestamps, for subsequent tracking and model iteration verification.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A supply chain IoT data fusion and real-time decision-making system based on distributed computing power, characterized in that: Includes the following steps: Status Acquisition Module: Acquires operational status information of the data center and related facilities, including server resource utilization, task load, temperature and cooling parameters, power supply and load status, carbon emission factors and energy price signals, and performs time synchronization, anomaly detection, multi-source heterogeneous data fusion and standardization processing on the acquired data; Resource Modeling Module: Receives data output from the Status Acquisition Module, quantifies the computing resources, cooling capacity, power limitations, and service quality of each data center, and predicts the available resource range, cooling capacity boundaries, and energy consumption characteristics based on environmental conditions and task requirements, generating analysis results; Scheduling optimization module: Based on the analysis results, it jointly solves the multi-data center task allocation, computing power migration, node power configuration, cooling strategy and vehicle or warehouse operation scheduling, and generates the control quantity for the current time period by adopting a rolling execution strategy. The strategy distribution module converts the control quantities for the current time period into server resource allocation, task migration, cooling adjustment, and power allocation instructions, and distributes them to the computing cluster scheduling system and infrastructure control system to generate execution results. Operational feedback module: Based on the execution results, monitor the execution results, business indicators, energy consumption and equipment status, update and optimize the model and scheduling strategy, and provide the updated model and strategy to the scheduling optimization module.

2. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 1, characterized in that, The process of time synchronization, anomaly detection, multi-source heterogeneous data fusion, and standardization of the collected data is as follows: The collected data is uniformly calibrated according to the timestamp, and abnormal data points are detected, removed, or corrected. Combine data from different sources to perform feature mapping and consistency matching; A unified feature output is generated through standardization processing.

3. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 2, characterized in that, The process of quantifying the computing resources, cooling capacity, power limitations, and service quality of each data center is as follows: Receive feature data output from the status acquisition module after time synchronization, anomaly detection, multi-source fusion and standardization processing; Analyze the server resources, cooling system capacity, and power constraints of each data center; The available computing power, cooling capacity, and power limits are quantified, and node service quality indicators are calculated to form a resource capacity table.

4. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 3, characterized in that, The process of generating analysis results is as follows: Based on the resource capacity table, a basic quantitative analysis is conducted on the available computing power, cooling capacity, and power limit of each data center, and the service capacity and energy consumption response characteristics of each node under different load levels are calculated. By combining historical load patterns, environmental data, and future task requirements, we can predict the resource utilization, cooling capacity limits, and power consumption of each data center in the future period, and assess potential bottleneck nodes and load fluctuation risks. The prediction results are combined with the capacity constraints, service quality requirements and energy cost factors of each data center to generate analysis results that can directly support scheduling optimization.

5. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 4, characterized in that, The process of jointly solving multi-datacenter task allocation, computing power migration, node power configuration, cooling strategy, and vehicle or warehouse operation scheduling is as follows: Based on the comprehensive analysis results, combined with the capacity constraints, service quality requirements, energy costs and carbon emission indicators of each data center, multi-objective coordination and optimization calculations are performed to generate scheduling and resource allocation strategies. The task allocation, computing power migration, node power configuration, and cooling strategy are jointly solved. During the optimization process, cross-warehouse and vehicle operation scheduling constraints are also taken into consideration. Preliminary control variables are generated using a rolling programming solution method.

6. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 5, characterized in that, The process of generating the control quantity for the current time period is as follows: Based on the initial control volume, the executable parts within the current time window are screened and finely adjusted; By combining real-time collected status data and environmental conditions, the task allocation, computing power migration, power allocation and cooling strategies are dynamically corrected; Based on the rolling update mechanism, the executable control instructions for the current time period are calculated, and the possible adjustment needs for the next time window are predicted to generate the control quantity for the current time period.

7. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 6, characterized in that, The process of converting commands into server resource allocation, task migration, cooling adjustment, and power allocation instructions is as follows: Based on the control variables for the current time period, the task allocation, computing power migration, node power configuration, and cooling strategies involved in the control variables are analyzed. Identify the target node, task type, and resource requirements corresponding to each scheduling instruction; Considering the execution capabilities, network communication latency, power and cooling constraints, service quality indicators, and security constraints of each data center and node, abstract control variables are mapped to specific executable operation instructions.

8. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 7, characterized in that, The process of generating the execution result is as follows: Based on executable operation instructions, the instructions are decomposed, packaged, and formatted in accordance with the interface specifications of the data center scheduling system and infrastructure control system; The command is sent to the control modules of each data center and related facilities through the communication network. The control system executes the task load adjustment, virtual machine migration, cooling system adjustment and node power allocation according to the command. During execution, the execution status, resource utilization, task completion status, and abnormal events of each node are collected in real time. The data collected during the execution process is locally aggregated and preliminarily analyzed to generate an execution result report.

9. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 8, characterized in that, The process of monitoring execution results, business indicators, energy consumption, and equipment status is as follows: Based on the execution result report, the execution results are analyzed and interpreted. Calculate task execution deviations, the difference between actual and planned energy consumption, and abnormal equipment status indicators; The analysis results are comprehensively evaluated with business performance indicators, energy consumption data, and equipment status to generate feedback information that can be optimized.

10. The supply chain IoT data fusion and real-time decision-making system based on distributed computing power according to claim 9, characterized in that, The process of providing the updated model and strategy to the scheduling optimization module is as follows: Based on the feedback information that can be optimized, various indicators are classified, processed, and summarized for analysis; Calculate the resource utilization deviation, power consumption difference, cooling capacity fluctuation and service quality deviation of each data center, and perform trend analysis and anomaly verification in combination with historical operating data; Based on the analysis results, the resource capacity parameters, node power constraints, cooling strategies, and multi-objective optimization weights of each data center are dynamically adjusted. The optimization model is calibrated by adjusting the prediction model coefficients, updating the rolling scheduling window parameters, and weighting the multi-objective optimization algorithm. The updated optimization model and scheduling strategy are output to the scheduling optimization module to form a rolling optimization mechanism for the next time window.

Images