An all-intelligent analog load distributed cooperative control method

Abstract

The present application relates to the technical field of cooperative control, in particular to a full-intelligent analog load distributed cooperative control method, comprising the following steps: acquiring node load data, dynamically predicting and adjusting distribution, cooperatively scheduling and optimizing control, and generating an intelligent load control scheme.In the present application, by real-time extraction of load state information and standardized verification, the running state of each node has a unified measurement basis, combined with node processing capacity and throughput performance to carry out task distribution, realize task scheduling and dynamic matching of node performance, realize predictive load regulation by combining historical data trend prediction with current state, avoid task delay and node overload risk, use real-time state feedback to continuously optimize task distribution strategy, improve resource use efficiency, strengthen the cooperative relationship between nodes, guarantee the sensitivity and consistency of system scheduling under the environment of dynamic load change, support the full-process adaptability and cooperative stability of task scheduling in a multi-node system.

Description

Technical Field

[0001] This invention relates to the field of collaborative control technology, and in particular to a fully intelligent analog load distributed collaborative control method. Background Technology

[0002] The field of collaborative control technology involves information interaction and coordinated operation among individuals in multi-node systems. Its core aspects include state information sharing, collaborative generation of control commands, and decision-making consistency among multiple stakeholders. It primarily achieves overall system function coordination and optimization through the design of distributed control strategies to ensure that control units in the distributed system work collaboratively to achieve overall operational goals. Among these, the load-distributed collaborative control method refers to coordinating load allocation behavior among nodes in typical application scenarios such as multi-node power systems or smart manufacturing networks. This is done by pre-setting static load allocation rules, employing time-triggered data transmission mechanisms, or using task scheduling strategies based on node number order. Control tasks are typically allocated and executed based on static weight configuration between nodes, distance-priority sorting, or cyclic scheduling tables.

[0003] In existing load control processes, task scheduling often relies on preset static load rules and numbering order. In scenarios where node load fluctuates frequently or resource status changes rapidly, it is difficult to respond in a timely manner to the actual carrying capacity of each node, which can easily lead to resource allocation imbalance and scheduling result deviation. Static configuration methods lack awareness of the real-time operating status of nodes. Cyclic or distance-based scheduling strategies can easily lead to task backlog or resource idleness under conditions of uneven number of nodes or significant performance differences. Especially in smart manufacturing and power load networks, this often causes problems such as local system congestion and overall performance degradation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a fully intelligent analog load distributed collaborative control method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a fully intelligent analog load distributed collaborative control method, comprising the following steps:

[0006] S1: Acquire and monitor the real-time load status data of each control node, perform data cleaning and standardization, verify the accuracy and consistency of the load data, and generate node load monitoring data.

[0007] S2: Based on the node load monitoring data, synchronously acquire node processing capacity scores and task processing throughput parameters, analyze the load balancing status between nodes, and automatically adjust task allocation to obtain a load allocation strategy.

[0008] S3: Statistically analyze the historical load data and load change trends of nodes, use time series forecasting to analyze future load fluctuations, combine real-time load data of nodes, adjust the load allocation strategy of each node in advance, and generate load fluctuation prediction and pre-adjustment plan.

[0009] S4: Based on the load fluctuation prediction and pre-adjustment scheme and the node load monitoring data, dynamically adjust the load allocation, adaptively adjust for overload and resource waste, and generate an adaptive load adjustment strategy.

[0010] S5: Based on the node load monitoring data and the adaptive adjustment strategy, coordinate the task allocation and resource scheduling among the nodes, execute fully intelligent simulation management of load allocation scheduling, and obtain a fully intelligent simulated load distributed collaborative control scheme.

[0011] As a further aspect of the present invention, the node load monitoring data includes load health assessment tags, node resource consumption levels, and system response status records; the load allocation strategy includes resource scheduling priority levels, task transfer weights, and node adaptation coefficients; the load fluctuation prediction and pre-adjustment scheme includes future load peak positions and trend change confidence intervals; the adaptive load adjustment strategy includes task migration threshold setting results, dynamic rebalancing mechanisms, and resource idle detection rules; and the fully intelligent simulated load distributed collaborative control scheme includes intelligent coordination control information, a global view of task scheduling, and node collaborative optimization records.

[0012] As a further aspect of the present invention, the specific steps for acquiring the node load monitoring data are as follows:

[0013] S111: Acquire and monitor the raw load data of each control node, including CPU load, memory usage, and ready queue length; record the load sampling value sequence of each control node within a unit of time; complete the preliminary grouping and alignment of data according to node number and timestamp order to obtain the raw dataset of control node load.

[0014] S112: Based on the original dataset of the control node load, perform outlier removal operations on the CPU load, memory usage and ready queue length along the time axis, perform range consistency judgment and filtering on the fluctuation amplitude between adjacent time points of nodes, remove data that deviates from the range of the node group, and obtain a node load cleaned data set.

[0015] S113: Based on the node load cleaning data set, normalize it according to a unified time axis, rely on time series consistency verification, identify the information distribution characteristics and fluctuation status in the load sequence, and obtain node load monitoring data.

[0016] As a further aspect of the present invention, the specific steps for obtaining the load allocation strategy are as follows:

[0017] S211: Based on the node load monitoring data, obtain the processing capability score corresponding to each node, match it according to the node number, construct the corresponding mapping between the node number and the processing capability score, synchronously collect the number of tasks processed by the node and the task execution completion time within a unit time, calculate the task processing throughput and form a two-dimensional mapping table with the processing capability score, and establish a node performance parameter set.

[0018] S212: Based on the node performance parameter set, extract the processing capability score and task processing throughput of each node, perform load balancing status analysis between nodes according to the load status of each node at the current moment and the load difference between nodes, calculate and obtain the load balancing index between nodes, judge the combination between each node, and establish load balancing measurement results.

[0019] S213: Based on the load balancing measurement results, using the current load value of each node and the corresponding processing capacity score as the judgment benchmark, identify the node group whose current load deviation exceeds the load deviation threshold, and synchronously redistribute tasks to obtain the load distribution strategy.

[0020] As a further aspect of the present invention, the specific steps for obtaining the load fluctuation prediction and pre-adjustment scheme are as follows:

[0021] S311: Obtain historical load data and historical time series records of each node, organize the original load logs in a structured manner according to the node number and timestamp, form a load trajectory sequence by combining the continuous time slice load values ​​of the same node, construct a node load change trend sequence based on equal time interval segmentation, smooth the trend sequence and detect interval changes, identify the directional changes of node load status over time, and generate node load trend features.

[0022] S312: Based on the node load trend characteristics, combined with the load change amplitude sequence over a continuous period, intraday periodic fluctuation information, mutation frequency records and load peak records, the future load status is inferred, the possible load offset direction and degree of change of each node in the future are identified, and a node load prediction difference sequence is established.

[0023] S313: Based on the node load prediction difference sequence, extract the real-time load value and prediction deviation of each node at the current moment, determine the risk of abnormal fluctuation in the current load status, combine the node processing capacity score and historical scheduling response record, adjust the task allocation ratio of the node in advance, update the task assignment mapping relationship, and obtain the load fluctuation prediction and pre-adjustment scheme.

[0024] As a further aspect of the present invention, the specific steps for obtaining the adaptive load adjustment strategy are as follows:

[0025] S411: Based on the load fluctuation prediction and pre-adjustment scheme and the node load monitoring data, construct a simulation input data structure according to the combination of task load index and node capability index, simulate the load distribution state after the task migrates between different nodes in sequence, compare the output of the task distribution under each combination, identify and output the load distribution response trend characteristics under the current state.

[0026] S412: Based on the load distribution response trend characteristics, extract the average resource utilization rate of each node under each round of task simulation configuration, compare it with the set overload threshold and resource waste threshold, determine overloaded and resource waste nodes, perform abnormal screening on the current task configuration status of related nodes, and statistically analyze the proportion and distribution of abnormal nodes in the overall node set to obtain a set of abnormal resource scheduling nodes.

[0027] S413: Based on the set of abnormal resource scheduling nodes, extract the task processing records and parameter information of unassigned tasks from the abnormal nodes, determine whether the current task configuration causes abnormal deviation in resource utilization efficiency, evaluate the matching rationality of the current task on the node, remove and reconstruct configurations with low matching between task execution and node performance, update the task scheduling mapping, and generate an adaptive load adjustment strategy.

[0028] As a further aspect of the present invention, the specific steps for obtaining the fully intelligent simulated load distributed collaborative control scheme are as follows:

[0029] S511: Based on the node load monitoring data and the adaptive adjustment strategy, construct a set of scheduling information structures for each node, extract load status information and resource utilization based on the real-time performance of the nodes during the scheduling process, analyze the distribution of scheduling capabilities of each node in the current system, define the scheduling resource status mapping range, and establish a scheduling resource status mapping matrix.

[0030] S512: Based on the resource utilization of each node in the scheduling resource status mapping matrix, adjust the scheduling coordination relationship between nodes, analyze the resource utilization level and the priority distribution pattern of scheduling tasks, identify the load transfer relationship between nodes whose resource utilization exceeds the scheduling threshold and nodes in the scheduling buffer, allocate the task scheduling amount of each node, and obtain the node task allocation adjustment result.

[0031] S513: Based on the node task allocation adjustment results, determine the collaborative relationship between each node, compare and analyze the resource utilization level, task distribution status and scheduling tendency between nodes, identify node pairs with stable collaborative relationships, screen node groups with collaborative control foundation, construct scheduling link structure, and obtain a fully intelligent simulated load distributed collaborative control scheme.

[0032] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0033] In this invention, by extracting load status information in real time and performing standardized verification, the operating status of each node has a unified measurement basis. Task allocation is carried out in combination with node processing capacity and throughput performance to achieve dynamic matching between task scheduling and node performance. Predictive load control is achieved by combining historical data trend prediction with the current status to avoid the risk of task delay and node overload. The task allocation strategy is continuously optimized by real-time status feedback to improve resource utilization efficiency, strengthen the collaborative relationship between nodes, ensure the sensitivity and consistency of system scheduling in dynamic load change environment, support the full-process adaptability and collaborative stability of task scheduling in multi-node systems, and enhance the transparency, predictability and self-optimization capability of system operating status. Attached Figure Description

[0034] Figure 1 This is a flowchart of the main steps of the present invention;

[0035] Figure 2 This is a flowchart of the node load monitoring data acquisition process of the present invention;

[0036] Figure 3 This is a flowchart illustrating the load distribution strategy acquisition process of the present invention.

[0037] Figure 4 This is a flowchart of the load fluctuation prediction and pre-adjustment scheme of the present invention;

[0038] Figure 5 This is a flowchart illustrating the adaptive load adjustment strategy acquisition process of the present invention.

[0039] Figure 6 This is a flowchart illustrating the fully intelligent simulated load distributed collaborative control scheme of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0042] Please see Figure 1 A fully intelligent analog load distributed collaborative control method includes the following steps:

[0043] S1: Acquire and monitor real-time load status data of each control node, including CPU load, memory usage, and ready queue length (operating system performance monitoring standard indicator (number of ready tasks in / proc / schedstat in Linux kernel, reflecting the degree of processing congestion), perform data cleaning and standardization, verify the accuracy and consistency of load data, and generate node load monitoring data.

[0044] S2: Based on node load monitoring data, synchronously acquire node processing capability scores (CPU performance benchmark test scores released by the Standard Performance Evaluation Organization, used to quantify node processing capability) and task processing throughput parameters. Analyze the load balancing status between nodes through intelligent simulation and automatically adjust task allocation, matching the load allocation with the real-time status of each node to obtain a load allocation strategy.

[0045] S3: Statistically analyze the historical load data and load change trends of nodes, use time series forecasting to analyze future load fluctuations, combine real-time load data of nodes, adjust the load allocation strategy of each node in advance, and generate load fluctuation prediction and pre-adjustment plan.

[0046] S4: Based on the load fluctuation prediction and pre-adjustment scheme and node load monitoring data, dynamically adjust the load distribution through intelligent simulation, monitor the collaborative working status between nodes, adaptively adjust for overload (overload exceeding the 80% CPU utilization threshold) and resource waste (resource waste below 20% resource utilization), perform real-time optimization of task scheduling between nodes, confirm that the load distribution strategy matches the actual load situation, and generate an adaptive load adjustment strategy.

[0047] S5: Based on node load monitoring data and adaptive adjustment strategies, the task allocation and resource scheduling between nodes are optimized through intelligent simulation collaborative control, and the fully intelligent simulation management of load allocation and scheduling is executed to obtain a fully intelligent simulated load distributed collaborative control scheme.

[0048] Node load monitoring data includes load health assessment labels, node resource consumption levels, and system response status records. Load allocation strategies include resource scheduling priority levels, task transfer weights, and node adaptation coefficients. Load fluctuation prediction and pre-adjustment schemes include future load peak positions and trend change confidence intervals. Adaptive load adjustment strategies include task migration threshold setting results, dynamic rebalancing mechanisms, and resource idle detection rules. The fully intelligent simulated load distributed collaborative control scheme includes intelligent coordination and control information, a global view of task scheduling, and node collaborative optimization records.

[0049] Please see Figure 2 The specific steps of S1 are as follows:

[0050] S111: Acquire and monitor the raw load data of each control node, including CPU load, memory usage, and ready queue length; record the load sampling value sequence of each control node within a unit of time; complete the preliminary grouping and alignment of data according to node number and timestamp order to obtain the raw dataset of control node load.

[0051] Acquire and monitor raw load data for each control node, including CPU load, memory usage, and ready queue length. During implementation, data collection must be performed using node ID and timestamp as unique indices. Data from all nodes should be collected synchronously at each time point to avoid data distortion caused by cross-time period mixing. The main data sources are the ` / proc / stat` file (CPU usage), ` / proc / meminfo` file (memory usage), and ` / proc / schedstat` file (scheduling queue length) under the Linux system. After collection, the data is stored sequentially in a raw table, which must cover the CPU usage of each node at the specified time point. Data such as load percentage, memory utilization percentage, and ready queue length can be collected every 10 or 30 seconds to ensure continuous recording and coverage. For example, at 10:00:00, the status data of Node-1, Node-2, and Node-3 are collected, recording CPU load as 76.5%, 82.3%, and 91.0%, memory utilization as 68.2%, 72.5%, and 88.1%, and ready queue lengths as 5, 8, and 14, respectively. After collection, the batch of data is classified and sorted according to timestamp and node number to establish a standardized storage format and form the basic original dataset, as shown in Table 1, to obtain the original dataset of control node load.

[0052] Table 1 Node Monitoring Data

[0053] Node number CPU load (%) Memory utilization (%) Ready queue length Timestamp Node-1 76.5 68.2 5 10:00:00 Node-2 82.3 72.5 8 10:00:00 Node-3 91.0 88.1 14 10:00:00

[0054] As shown in Table 1, the raw data of the three nodes at the same time point have been collected and organized for subsequent cleaning and detection operations.

[0055] S112: Based on the original dataset of control node load, outlier removal operations are performed on CPU load, memory usage and ready queue length along the time axis. Consistency judgment is performed on the range of fluctuation amplitude between adjacent time points of nodes. Data that deviates from the range of the node group is removed to obtain the node load cleaned data set.

[0056] Based on the original dataset of control node load, CPU load, memory utilization, and ready queue length need to be cleaned at each timestamp. The cleaning process first calculates the central tendency and dispersion of various indicators to identify outlier ranges. For example, in Table 1, the CPU load values ​​are 76.5%, 82.3%, and 91.0%, with a mean of 83.27% and a standard deviation of 7.25%. The set threshold rule is that the deviation from the mean should not exceed twice the standard deviation, resulting in a calculated range of 68.77% to 97.77%. The data from all three nodes fall within this range, so all are retained. If, at the same time point, a node's CPU load is 45%, it needs to be removed because it is below the lower threshold of 68.77%. To remove this data, the memory usage rate was processed in the same way. The mean was 76.27%, the standard deviation was 10.23%, and the range was 55.81% to 96.73%. The memory usage rates of the three nodes in Table 1 were 68.2%, 72.5%, and 88.1%, respectively. All of them were within this range and were retained. Data with a usage rate above 95% was removed because it exceeded the upper limit. The ready queue lengths were 5, 8, and 14, with a mean of 9.0 and a standard deviation of 4.58, ranging from -0.16 to 18.16. All three data were within the range and were retained. Therefore, after filtering, a cleaned data set was obtained. This set removed points that deviated too much or had unreasonable values, resulting in a cleaned node load data set.

[0057] S113: Based on the node load cleaning data set, normalize it according to a unified time axis, rely on time series consistency verification, identify the information distribution characteristics and fluctuation status in the load series, and obtain node load monitoring data.

[0058] Based on the node load cleansing dataset, it is necessary to standardize the data of each node at the same timestamp to ensure that different indicators are comparable under the same scale. The standardization method uses the minimum and maximum values ​​of each indicator at that timestamp to determine their position, constructing a normalization table. Then, CPU load, memory utilization, and ready queue length are used as three-dimensional combined features for time series consistency detection. For example, in Table 1, the minimum CPU load is 76.5% and the maximum is 91.0%. Therefore, the CPU load position of Node-1 is marked as 0, Node-3 as 1, and Node-2, which is in between, is marked as 0.39. Similarly, memory utilization and ready queue length are also standardized. The process involves determining the three-dimensional combined feature sequence of each node and checking its continuity over time. For example, if the ready queue length of Node-1 gradually increases from 5 to 6 and then to 7 between 10:00:00 and 10:00:30, the CPU load gradually increases from 76.5% to 79% and then to 81%, and the memory utilization rate increases from 68.2% to 70% and then to 71%, then it is determined to be a stable sequence. If there is a sudden change in value, such as the queue length suddenly increasing from 5 to 15, it exceeds the allowable fluctuation range of 80% in a single step, triggering an anomaly marker. The data at that moment is removed as an unstable point or archived separately. Finally, after consistency verification and archiving, a complete node load monitoring record is generated, and node load monitoring data is obtained.

[0059] Please see Figure 3 The specific steps of S2 are as follows:

[0060] S211: Based on node load monitoring data, obtain the processing capability score corresponding to each node, match it according to the node number, construct the corresponding mapping between node number and processing capability score, synchronously collect the number of tasks processed by the node and the task execution completion time within a unit time, calculate the task processing throughput and form a two-dimensional mapping table with the processing capability score, and establish a set of node performance parameters.

[0061] Based on node load monitoring data, the processing capacity score corresponding to each node is obtained. During implementation, the unique node number is used as an index to allocate the SPECint score value extracted from the standard performance evaluation organization database to the corresponding node. For example, the processing capacity scores of Node-1 to Node-4 are 1800, 2400, 2100, and 2600, respectively. Then, the number of tasks processed by each node per unit time is collected, the number of tasks completed within a 10-second time window is counted, and the average task processing time is recorded. Taking Node-2 as an example, if 150 tasks are completed within 10 seconds, the average processing time is approximately 0.067 seconds / task. This task processing volume reflects the node throughput. Next, the throughput parameters and processing capacity scores of each node are merged, and a two-dimensional mapping table is constructed. In this way, a mapping reference structure between task capacity and processing performance is established. At the same time, real-time node load data is introduced to enable cross-item comparison in subsequent processing. Finally, a complete set of node performance parameters is established, as shown in Table 2.

[0062] Table 2 Node Performance Indicators

[0063] Node number Processing ability score Number of tasks per unit time Average load (%) Node-1 1800 120 70 Node-2 2400 150 85 Node-3 2100 135 60 Node-4 2600 160 90

[0064] As shown in Table 2, there are differences in the processing capacity and task throughput of different nodes under the same collection period. These differences can be used for comparative analysis in subsequent balanced scheduling to establish a set of node performance parameters.

[0065] S212: Based on the node performance parameter set, extract the processing capacity score and task processing throughput of each node. Based on the current load status of each node and the load differences between nodes, perform load balancing analysis between nodes using the following formula:

[0066]

[0067] The load balancing index between nodes is calculated, the combinations of nodes are judged, and a load balancing metric is established; where B i,j L represents the load balancing index value of node i relative to node j. i L j L k S represents the current load values ​​of nodes i, j, and k. i S j P represents the processing capability score of nodes i and j. i P j This represents the ratio of the unit load task processing capacity of nodes i and j. This represents the sum of the load differences between node i and all N nodes;

[0068] The system retrieves the node performance parameter set, extracting the processing capacity score and task throughput parameters for each node. In practice, using the node ID as the primary key, it obtains the performance score, the number of tasks processed per unit time, and the average load value of the node. All nodes are then paired to construct a load difference distribution matrix between nodes. A difference analysis is then performed using the node capacity score and the ratio of processing capacity per unit time. For example, between Node-2 and Node-3, the load difference is 85% - 60% = 25%, the score difference is 2400 - 2100 = 300, and the ratio of processing capacity per unit time is:

[0069]

[0070]

[0071] Substitute into the formula to calculate, where:

[0072] |L2-L3|=|85-60|=25;

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] therefore:

[0079]

[0080] Next, calculate the average load difference between node 2 and all nodes:

[0081]

[0082]

[0083] Finally obtained

[0084] B 2,3 =0.01111+15=15.01111;

[0085] This value is the load balancing index between Node-2 and Node-3, representing the relative difference in their load and capacity. By calculating the load balancing index between each pair of nodes in the same way as above, a complete load balancing metric can be established.

[0086] The load balancing index is a numerical indicator used to quantify the degree of matching between the load status and processing capacity of two nodes at a specific moment. Its core significance lies in comprehensively reflecting the coordination of load distribution and the coupling difference of capacity structure between nodes. This index normalizes the load difference between nodes and adjusts the capacity by combining the processing capacity score of the two nodes with the ratio of unit load processing intensity. It further adds the load difference of the current node relative to the average load of the entire cluster to construct a composite index that reflects both the degree of local imbalance and the degree of global load detachment. The larger the value, the higher the degree of load imbalance between nodes and the worse the capacity matching. This index can be used as a basis for judging task migration, load adjustment and node scheduling decisions, and can be used to identify node combinations that are not suitable to continue to undertake the current task.

[0087] The formula is based on a comprehensive quantitative requirement of the coupling degree between load differences and processing capabilities among nodes. It reflects the strength of the scheduling balance relationship among nodes through weighted and normalized calculations of multiple parameters. The first term... The numerator |L is used to measure the normalized deviation of the load difference between two nodes under their capacity structure. i -L j | represents the absolute value of the load difference, indicating the direct difference in task processing pressure between two nodes at the current moment. The denominator is the square root of the sum of the squares of the ratios of the node's average processing capacity score to the unit load task processing intensity, serving as a normalization and capability coupling weight adjustment. The sum of the squares and the square root of the result form a Euclidean distance, representing the aggregated performance of multi-dimensional capabilities under a single index. This design ensures that high-capacity nodes are not overestimated even with large load differences. The second term... This is the average difference between node i and all other nodes in terms of load dimension, reflecting the degree of deviation of node i in the whole system. The sum of the two parts can comprehensively reflect the load coordination status between nodes and the whole system. The overall structure ensures that it considers both the direct coordination between two nodes and the state deviation of nodes relative to the overall environment, thereby accurately quantifying their priority in participating in task migration or load adjustment.

[0088] S213: Based on the load balancing measurement results, using the current load value of each node and the corresponding processing capacity score as the judgment benchmark, identify the node group whose current load deviation exceeds the load deviation threshold (20%), and synchronously redistribute tasks to obtain the load distribution strategy.

[0089] Based on the load balancing metrics, the actual load value and processing capacity score of each node are first extracted to construct a node status table. This table is used to compare and determine whether there is a deviation in the current task allocation status. A 20% deviation threshold is used to identify load deviations. This threshold is derived from the general safety range provisions for resource load mutation response control in CPU resource scheduling research. The judgment method is to compare the difference between the actual load value of the node and the baseline load value. If it exceeds ±20%, it is considered an overload or underload state. For example, if the system baseline load is 70%, the scheduling mechanism is triggered when the node load exceeds 84% ​​or falls below 56%. For nodes within the abnormal range... For each node, its current task list and processing capacity score are extracted. Combined with the node's average task processing time, a portion of tasks are selected to be transferred to other low-load nodes. During the task transfer process, the difference in processing capacity scores between the target nodes must be limited to within 10%. For example, if the difference in processing capacity scores between Node-1 and Node-3 is 1800 and 2100, respectively, which is 16.7%, it does not meet the requirements for direct transfer. Nodes with similar processing capacity scores should be selected first to redistribute the scheduled tasks. Finally, after the task migration between nodes is completed, a task deployment status that matches the processing capacity of all tasks in the current time period is formed, resulting in the load distribution strategy.

[0090] Please see Figure 4 The specific steps of S3 are as follows:

[0091] S311: Obtain historical load data and historical time series records of each node, organize the original load logs in a structured manner according to the node number and timestamp, form a load trajectory sequence by combining the continuous time slice load values ​​of the same node, construct a node load change trend sequence based on equal time interval segmentation, smooth the trend sequence and detect interval changes, identify the directional changes of node load status over time, and generate node load trend features.

[0092] Historical load data and time-series records for each node are acquired. During execution, load-related parameters such as CPU utilization and memory usage are first extracted from the node's runtime logs and recorded at 5-minute intervals. These are then categorized and sorted by node number and timestamp. Subsequently, the load values ​​from consecutive time slices are combined to form a time-series trajectory. Using data from each node over a 7-day period as a range, the average load value, maximum load value, and number of load spikes for all sampling points within that period are calculated. For example, the load record for Node-2 shows an average load of 75.2% over 7 days, with a maximum peak of 9... 8%, with 26 load mutations occurring daily. The criterion for a mutation is that the load value change between two adjacent sampling points exceeds 10%. If the load values ​​at sampling times of 10:00 and 10:05 are 76% and 87% respectively on a certain day, it is recorded as a mutation event. After the statistics are completed, the various indicators are summarized to generate a node load trend data table, as shown in Table 10. In trend detection, the direction of the difference between adjacent time points is used to determine the increasing or decreasing interval. If three consecutive sampling points show an increasing trend, it is marked as an upward segment. Finally, the load change trend sequence is constructed by combining the overall trend segments to obtain the node load trend characteristics.

[0093] Table 3 Historical Load Fluctuation of Nodes

[0094]

[0095]

[0096] As shown in Table 3, the four nodes exhibit different characteristics in terms of load intensity and fluctuation frequency over seven consecutive days, providing a structured input basis for subsequent forecasting.

[0097] S312: Based on the node load trend characteristics, combined with the load change amplitude sequence over a continuous period, intraday periodic fluctuation information, mutation frequency records and load peak records, the future load status is inferred, the possible load shift direction and degree of change of each node in the future are identified, and a node load prediction difference sequence is established.

[0098] Based on the node load trend characteristics, load change data for each node within a time period is retrieved. By extracting the maximum daily fluctuation amplitude, average range, and frequency of sudden changes for each node, a load trajectory change rate record set is constructed. Simultaneously, the change trends of typical periods such as morning peak, midday stability, and evening decline are marked in the time series, forming a multi-period node load change feature group. On this basis, combined with the real-time load data of the current time node, the continuation direction of the short-term trend trajectory is compared with the current state. If there is a short-term upward trend but the real-time value begins to decline, it is marked as a trend deviation node. Then, the difference between the predicted load and the real-time load of each node at the current moment is extracted to establish a set of differences between the future load state and the current state of each node. Nodes with an error greater than ±15% will be included in the prediction deviation queue. This difference can be used to identify the degree of deviation between the prediction accuracy and the actual operating state of the system. Finally, it is integrated into a node prediction fluctuation list, generating a node load prediction difference sequence.

[0099] S313: Based on the node load prediction difference sequence, extract the real-time load value and prediction deviation of each node at the current moment, determine the risk of abnormal fluctuation in the current load status, combine the node processing capacity score and historical scheduling response record, adjust the task allocation ratio of the node in advance, update the task assignment mapping relationship, and obtain the load fluctuation prediction and pre-adjustment plan.

[0100] The node load prediction difference sequence is invoked. For the set of nodes with a deviation greater than 15%, it is determined whether there is a risk of task overload or load offset in the current period. The task stability is judged by the combination of the node's real-time load value and the prediction deviation. If the current value is 80% and the prediction is 62%, the deviation is 18%, and the task allocation strategy needs to be adjusted in advance. At the same time, the historical scheduling records of the node are retrieved to identify the past task recovery success rate, average task execution time and scheduling response interval to evaluate its scheduling capacity. During the task adjustment process, the proportion of assignments with task execution time exceeding the historical average is reduced first. The task allocation table is regenerated, and the range of target nodes for task migration is limited based on the node capability score. If the historical task execution time deviation of the target node is less than 5% and the difference in processing capability score is no more than 10%, it is included in the scheduling candidate range. A new round of task allocation scheme is generated according to the current task pool status. Finally, the task adjustment configuration across time periods and across nodes is summarized to obtain the load fluctuation prediction and pre-adjustment scheme.

[0101] Please see Figure 5 The specific steps of S4 are as follows:

[0102] S411: Based on the load fluctuation prediction and pre-adjustment scheme and node load monitoring data, construct the simulation input data structure according to the combination of task load index and node capability index, simulate the load distribution state after the task migrates between different nodes in turn, compare the output of the task distribution under each combination, identify and output the load distribution response trend characteristics under the current state.

[0103] Based on the load fluctuation prediction and pre-adjustment scheme and node load monitoring data, the predicted task volume, real-time CPU load percentage, processing capacity score, and average task time of each node in the current period are collected. When constructing the simulation input structure, the number of tasks expected to be received by each node is first extracted from the system scheduling queue, such as 320 tasks expected for Node-2. Then, the current CPU utilization value of the node is recorded in real time, for example, 85.3% for Node-2. Combined with the processing capacity score (e.g., 9.2) and the average task time (e.g., 1.6 seconds), the above data is constructed into a structured input row, forming the following input information:

[0104] Table 4 Node Operation Status Input Table

[0105]

[0106]

[0107] When constructing the simulation system, the task migration range was set to change within 20% of the task volume in the task pool. Based on the input data in the table above, the load transfer simulation between nodes was carried out. For example, the load of Node-4 decreased from 92.4% to 75.2%, while that of Node-1 increased from 68.5% to 77.6%. After the load redistribution, the CPU utilization of each node tended to be balanced. By comparing the fluctuation range of the CPU utilization of each node and the average time taken to complete the task before and after the migration, the trend of response performance change caused by the load change was analyzed. Then, a collaborative feature set of task migration and resource response was constructed, and the resource absorption capacity characteristics of each node in the migration round were extracted. Finally, the load distribution response trend characteristics were established.

[0108] S412: Based on the load distribution response trend characteristics, extract the average resource utilization rate of each node under each round of task simulation configuration, compare it with the set overload threshold of 80% and resource waste threshold of 20%, identify overloaded and resource waste nodes, perform anomaly screening on the current task configuration status of related nodes, and statistically analyze the proportion and distribution of abnormal nodes in the overall node set to obtain the set of resource abnormal scheduling nodes.

[0109] Based on the load distribution response trend characteristics, the resource utilization under the node simulation configuration is screened and analyzed to determine whether there is any abnormal resource usage in the current configuration. During the operation, the system resource overload threshold is set to 80% CPU utilization and the resource waste threshold is set to 20% CPU utilization. The values ​​of each node in the table above are judged in turn. For example, Node-2's current CPU is 85.3%, which is higher than 80%, so it is identified as an overloaded node. Node-3 is 45.9%, which does not meet the waste judgment condition, so it is skipped. Then, the task density index is calculated based on the task volume divided by the average time. For example, the task density of Node-4 is 271. The highest value among the four nodes is 43 tasks / second (380 ÷ 1.4). Combined with its CPU utilization of 92.4%, it is judged to be a scheduling abnormal node. At the same time, although its task density is high, the average time has not been significantly reduced, indicating a processing bottleneck. Therefore, this node is further filtered into the abnormal scheduling candidate set. On this basis, the ratio between the processing capacity score and the task response latency is introduced for verification. If the ratio is lower than the set reference lower limit, such as 6.0, it is considered an inefficient node. Node-3 has a score of 6.4 and a time of 2.5 seconds, which is low response efficiency, resulting in a low resource utilization state. It is also included in the candidate set, and finally the set of resource abnormal scheduling nodes is obtained.

[0110] S413: Based on the set of abnormal resource scheduling nodes, extract the task processing records and parameter information of unassigned tasks of abnormal nodes, determine whether the current task configuration causes abnormal deviation of resource utilization efficiency, evaluate the matching rationality of the current task on the node, remove and reconstruct configurations with low matching between task execution and node performance, update task scheduling mapping, and generate adaptive load adjustment strategy.

[0111] The system calls upon the set of abnormal scheduling nodes, and compares the current task configuration of abnormal nodes such as Node-2, Node-3, and Node-4 with the unassigned task status in the system task pool. It extracts their task density per unit time and compares it with the system average density value, such as 215 tasks / second. If the deviation rate is greater than 15%, the node is marked as an abnormal scheduling node. For example, Node-4's density is 271.43 tasks / second, which deviates from the average by 26.2%, and is marked as a deviation node. In subsequent operations, the task priority information within the node is extracted. For example, Node-4's task pool contains 88 tasks with priority P=3, and its node processing capacity score is 9.5, which is lower than the system's set high-priority task matching score limit of 8. By combining .0, it was confirmed that there was a matching degree deviation in the task of this node. The matching degree was calculated to be 0.316, which is less than 0.6. Therefore, this type of task was removed and reassigned to nodes with higher scores or lower loads in the task pool. For example, Node-1 has a low load and a matching degree of 0.643, making it suitable to receive some tasks. After the task migration, the load status of all nodes was recalculated. When the load of Node-2 dropped to 76.4%, Node-4 dropped to 78.1%, Node-1 rose to 72.3%, and Node-3 remained stable in the 60.1% range, the resource utilization of all nodes in the system fell back to the set range of 20% to 80%, forming an effective node task balancing configuration, and finally obtaining the adaptive load adjustment strategy.

[0112] Please see Figure 6 The specific steps of S5 are as follows:

[0113] S511: Based on node load monitoring data and adaptive adjustment strategies, construct a set of scheduling information structures for each node, extract load status information and resource utilization based on the real-time performance of nodes during the scheduling process, analyze the distribution of scheduling capabilities of each node in the current system, define the range of scheduling resource status mapping, and establish a scheduling resource status mapping matrix.

[0114] Based on node load monitoring data and adaptive adjustment strategies, all scheduling nodes in the current system are first identified and their parameters extracted. For each node, the statistical values ​​of its task scheduling quantity, current resource utilization ratio, load balance status, and historical task priority need to be obtained. In the specific implementation process, it is assumed that nodes 1 to 4 each undertake part of the task scheduling. The system extracts the above four types of parameters from the scheduling center module and stores them in a unified structure. In the actual sampling process, the node scheduling characteristics are shown in Table 5:

[0115] Table 5 Node Scheduling Feature Input Table

[0116] Node number Number of tasks scheduled Resource utilization ratio Load balancing status Average task priority Node-1 340 0.72 0.69 2.6 Node-2 295 0.65 0.74 2.7 Node-3 310 0.60 0.66 3.0 Node-4 355 0.77 0.79 2.5

[0117] As shown in Table 5, the number of task scheduling reflects the current load distribution scale of each node, the resource occupancy ratio is used to judge the current resource usage tension of the node, and a resource occupancy ratio exceeding 0.75 is set as resource tension, the load balance status represents the distribution balance of tasks in the time dimension, if this value exceeds 0.75, the task scheduling tends to be concentrated, and the average task priority is used to characterize the importance of the task. For example, the task priority value of Node-3 is 3.0, which is higher than other nodes. This node is suitable as the scheduling receiver of priority tasks. By eliminating nodes with high occupancy, heavy load, and low priority, such as Node-4, as scheduling migration targets, the optimal scheduling resource structure is screened, and finally the scheduling resource status mapping matrix is ​​established.

[0118] S512: Based on the resource utilization of each node in the scheduling resource status mapping matrix, adjust the scheduling coordination relationship between nodes, analyze the resource utilization level and the priority distribution pattern of scheduling tasks, identify the load transfer relationship between nodes whose resource utilization exceeds the scheduling threshold and nodes in the scheduling buffer, allocate the task scheduling amount of each node, and obtain the node task allocation adjustment results.

[0119] After invoking the scheduling resource status mapping matrix, the system identifies the distribution characteristics of each node in terms of resource usage and task scheduling priority. It then makes initial allocation decisions for nodes with high resource usage but low task priority. For example, Node-4 has a resource usage ratio of 0.77 and an average priority of 2.5, both at or below the threshold. Allocation of some low-priority tasks can alleviate its resource pressure. Considering that Node-4 currently has 355 scheduled tasks, the system sets the allocation ratio to 12% of the tasks scheduled on that node, approximately 43 tasks. The receiving node is selected as N, which has relatively abundant resources and high task priority. Node-3 has a resource utilization ratio of 0.60 and an average priority of 3.0. After tasks are added, its resource utilization ratio will increase slightly, but it will still be within the safe range of below 0.75 allowed by the system. The number of tasks added is controlled to 100% of the number of tasks removed. The system synchronously corrects the task scheduling quantity and load index of the two nodes. After the update, the number of task scheduling for Node-4 is reduced to 312, and the number of task scheduling for Node-C is increased to 353. The resource utilization ratio is slightly increased to 0.69 based on the average of the added tasks. A node status update dataset is formed to obtain the node task allocation adjustment results.

[0120] S513: Based on the node task allocation adjustment results, determine the collaborative relationship between each node, compare and analyze the resource utilization level, task distribution status and scheduling tendency between nodes, identify node pairs with stable collaborative relationships, screen node groups with collaborative control foundation, construct scheduling link structure, and obtain a fully intelligent simulated load distributed collaborative control scheme.

[0121] After calling the node task allocation adjustment results, the system determines the collaborative relationship between nodes based on the updated task scheduling quantity and resource usage level. For example, if node 4 schedules out 43 tasks and node 3 schedules in 43 tasks, the resource load relationship shifts. Based on the task changes, priority change trends, and resource occupancy offset values ​​between nodes, the system performs scheduling stability analysis. If the changes in various parameters are within the system's set stable range, i.e., the task scheduling quantity fluctuation does not exceed ±15%, the resource occupancy ratio does not exceed ±0.1, and the task priority offset value does not exceed ±0.3, then a controllable collaborative relationship is considered to have been formed between nodes. Nodes 1 and 2 are marked as stable nodes because their task scheduling and resource status have not fluctuated significantly. Based on this, the system establishes a scheduling link graph. In the graph, the edge weight represents the task flow frequency, and the node represents the scheduling endpoint, forming several scheduling subgraphs. Each subgraph represents a load balancing block, and a fully intelligent simulated load distributed collaborative control scheme is generated.

[0122] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fully intelligent analog load distributed cooperative control method, characterized in that, Includes the following steps: S1: Acquire and monitor the real-time load status data of each control node, perform data cleaning and standardization, verify the accuracy and consistency of the load data, and generate control node load monitoring data. S2: Based on the load monitoring data of the control nodes, synchronously obtain the control node processing capacity score and task processing throughput parameters, analyze the load balancing status between control nodes, and automatically adjust the task allocation to obtain the load allocation strategy. S3: Statistically analyze the historical load data and load change trends of control nodes, use time series forecasting to analyze future load fluctuations, combine real-time load data of control nodes, adjust the load allocation strategy of each control node in advance, and generate load fluctuation prediction and pre-adjustment schemes. S4: Based on the load fluctuation prediction and pre-adjustment scheme and the load monitoring data of the control node, dynamically adjust the load allocation, adaptively adjust for overload and resource waste, and generate an adaptive load adjustment strategy. S5: Based on the load monitoring data of the control nodes and the adaptive load adjustment strategy, coordinate the task allocation and resource scheduling between each control node, execute fully intelligent simulation management of load allocation scheduling, and obtain a fully intelligent simulated load distributed collaborative control scheme. The fully intelligent simulated load distributed collaborative control scheme includes intelligent coordination control information, a global view of task scheduling, and control node collaborative optimization records. The specific steps for obtaining the load fluctuation prediction and pre-adjustment scheme are as follows: S311: Obtain historical load data and historical time series records of each control node, organize the original load logs in a structured manner according to the control node number and timestamp, form a load trajectory sequence by combining the continuous time slice load values ​​of the same control node, construct a control node load change trend sequence based on equal time interval segmentation, perform smoothing processing and interval change detection on the trend sequence, identify the directional change of control node load status over time, and generate control node load trend features. S312: Based on the load trend characteristics of the control nodes, combined with the load change amplitude sequence over a continuous period, intraday periodic fluctuation information, mutation frequency records and load peak records, the future load status is inferred, the possible load offset direction and degree of change of each control node in the future are identified, and a control node load prediction difference sequence is established. S313: Based on the load prediction difference sequence of the control nodes, extract the real-time load value and prediction deviation of each control node at the current moment, determine the risk of abnormal fluctuation in the current load status, combine the control node processing capability score and historical scheduling response record, adjust the task allocation ratio of the control nodes in advance, update the task assignment mapping relationship, and obtain the load fluctuation prediction and pre-adjustment scheme. The specific steps for obtaining the adaptive load adjustment strategy are as follows: S411: Based on the load fluctuation prediction and pre-adjustment scheme and the control node load monitoring data, construct a simulation input data structure according to the combination of task load index and control node capability index, simulate the load distribution state after the task migrates between different control nodes in turn, compare the output of the task distribution under each combination, identify and output the load distribution response trend characteristics under the current state. S412: Based on the load distribution response trend characteristics, extract the average resource utilization rate of each control node under each round of task simulation configuration, compare it with the set overload threshold and resource waste threshold, determine the overload and resource waste control nodes, perform anomaly screening on the current task configuration status of relevant control nodes, and statistically analyze the proportion and distribution of abnormal control nodes in the overall control node set to obtain the resource anomaly scheduling control node set. S413: Based on the set of resource anomaly scheduling control nodes, extract the task processing records and parameter information of unassigned tasks from the anomaly control nodes, determine whether the current task configuration causes abnormal deviation in resource utilization efficiency, evaluate the matching rationality of the current task on the control node, eliminate and reconstruct configurations with low matching between task execution and control node performance, update the task scheduling mapping, and generate an adaptive load adjustment strategy.

2. The fully intelligent analog load distributed cooperative control method according to claim 1, characterized in that, The control node load monitoring data includes load health assessment labels, control node resource consumption levels, and system response status records. The load allocation strategy includes resource scheduling priority levels, task transfer weights, and control node adaptation coefficients. The load fluctuation prediction and pre-adjustment scheme includes future load peak positions and trend change confidence intervals. The adaptive load adjustment strategy includes task migration threshold setting results, dynamic rebalancing mechanisms, and resource idle detection rules.

3. The fully intelligent analog load distributed cooperative control method according to claim 1, characterized in that, The specific steps for obtaining the control node load monitoring data are as follows: S111: Acquire and monitor the raw load data of each control node, including CPU load, memory usage, and ready queue length; record the load sampling value sequence of each control node within a unit of time; complete the preliminary grouping and alignment of data according to the control node number and timestamp order to obtain the raw dataset of control node load. S112: Based on the original dataset of control node load, outlier removal operations are performed on CPU load, memory usage and ready queue length along the time axis. Consistency judgment is performed on the range of fluctuation amplitude between adjacent time points of control nodes. Data that deviates from the range of control node group is removed to obtain control node load cleaned data set. S113: Based on the control node load cleaning data set, normalize it according to a unified time axis, rely on time series consistency verification, identify the information distribution characteristics and fluctuation status in the load sequence, and obtain control node load monitoring data.

4. The fully intelligent analog load distributed cooperative control method according to claim 1, characterized in that, The specific steps for obtaining the load distribution strategy are as follows: S211: Based on the load monitoring data of the control nodes, obtain the processing capability score corresponding to each control node, match it according to the control node number, construct the corresponding mapping between the control node number and the processing capability score, synchronously collect the number of tasks processed by the control node and the task execution completion time within a unit time, calculate the task processing throughput and form a two-dimensional mapping table with the processing capability score, and establish a set of control node performance parameters. S212: Based on the set of performance parameters of the control nodes, extract the processing capability score and task processing throughput of each control node, analyze the load balancing status between control nodes according to the load status of each control node at the current moment and the load difference between control nodes, calculate and obtain the load balancing index between control nodes, judge the combination between each control node, and establish the load balancing measurement result. S213: Based on the load balancing measurement results, using the current load value and corresponding processing capacity score of each control node as the judgment benchmark, identify the control node group whose current load deviation exceeds the load deviation threshold, and synchronously redistribute tasks to obtain a load allocation strategy.

5. The fully intelligent analog load distributed cooperative control method according to claim 1, characterized in that, The specific steps for obtaining the fully intelligent simulated load distributed collaborative control scheme are as follows: S511: Based on the load monitoring data of the control nodes and the adaptive load adjustment strategy, construct a set of scheduling information structures for each control node, and extract load status information and resource utilization based on the real-time performance of the control nodes during the scheduling process. Analyze the distribution of the scheduling capabilities of each control node in the current system, define the scheduling resource status mapping range, and establish a scheduling resource status mapping matrix. S512: Based on the resource utilization of each control node in the scheduling resource status mapping matrix, adjust the scheduling coordination relationship between control nodes, analyze the resource utilization level and the priority distribution pattern of scheduling tasks, identify the load transfer relationship between control nodes whose resource utilization exceeds the scheduling threshold and control nodes in the scheduling buffer, allocate the task scheduling amount of each control node, and obtain the control node task allocation adjustment result. S513: Based on the task allocation adjustment results of the control nodes, determine the collaborative relationship between each control node, compare and analyze the resource utilization level, task distribution status and scheduling tendency among the control nodes, identify control node pairs with stable collaborative relationships, screen control node groups with collaborative control foundation, construct scheduling link structure, and obtain a fully intelligent simulated load distributed collaborative control scheme.

Images