Adaptive adjustment method and system for a deep learning-based TT&C system

Abstract

The application discloses a method and system for self-adaptive adjustment of a measurement and control system based on deep learning, and relates to the technical field of intelligent regulation and control. Real-time operation data of the measurement and control system is collected, and the real-time operation data is constructed into a system state vector. Based on the system state vector, feature extraction is performed on tasks in the measurement and control system, and the tasks are classified by using a support vector machine. The resource demand of different types of tasks is predicted, the task priority is dynamically adjusted based on the resource demand, task priority information is generated, global resource allocation optimization is performed by using a genetic algorithm through a macroscopic scheduling layer, real-time scheduling decisions are made by using a reinforcement learning model through a microscopic scheduling layer, and specific resource allocation instructions are generated based on the resource allocation scheme output after optimization of the macroscopic scheduling layer. The system resource utilization rate is improved, and the task response delay is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent regulation and control, in particular to a deep learning-based self-adaptive adjustment method and system for a measurement and control system. BACKGROUND

[0002] Modern measurement and control systems are facing increasingly complex task environments and variable workloads, and traditional static resource allocation methods have been unable to meet the needs of efficient operation. Measurement and control systems have limited resources, while the number and types of tasks continue to increase, leading to widespread resource contention. In particular, in some key areas, the system often faces challenges such as sudden surge of tasks and fluctuation of resource demand, and traditional fixed priority scheduling strategies and pre-allocated resource methods are difficult to cope with these dynamic changes, which can easily cause high-priority task response delays and low resource utilization.

[0003] To this end, the present application proposes a deep learning-based self-adaptive adjustment method and system for a measurement and control system. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application proposes a deep learning-based self-adaptive adjustment method and system for a measurement and control system, which improves system resource utilization and reduces task response delay.

[0005] To achieve the above-mentioned purpose, a deep learning-based self-adaptive adjustment system for a measurement and control system is proposed, which includes a resource state perception module, a priority adjustment module, a double-layer scheduling decision module, and a resource scheduling execution module; wherein each module is connected through electrical means;

[0006] The resource state perception module collects real-time operation data of the measurement and control system, pre-processes the real-time operation data, and constructs the pre-processed real-time operation data into a system state vector; and sends the system state vector to the priority adjustment module and the double-layer scheduling decision module as an input for resource scheduling decision;

[0007] The priority adjustment module receives the system state vector, extracts features of tasks in the measurement and control system, classifies the tasks using a support vector machine, predicts resource demand for different types of tasks, dynamically adjusts the priority of the tasks based on resource demand, and transmits the adjusted task priority information to the double-layer scheduling decision module;

[0008] The double-layer scheduling decision module is composed of a macro-scheduling layer and a micro-scheduling layer; the macro-scheduling layer receives the system state vector and the adjusted task priority information, and performs global resource allocation optimization by using a genetic algorithm; the micro-scheduling layer performs real-time scheduling decision by using a reinforcement learning model based on the optimization result of the macro-scheduling layer, and takes the resource allocation scheme of the macro-scheduling layer as a constraint condition of the micro-scheduling layer in a top-down guidance feedback manner; the double-layer scheduling decision module sends the optimized resource allocation scheme and the result of the real-time scheduling decision to the resource scheduling execution module.

[0009] The resource scheduling execution module receives the resource allocation scheme output by the double-layer scheduling decision module, and generates a specific resource allocation instruction.

[0010] The resource state perception module collects real-time running data of the measurement and control system, including the following steps:

[0011] Step 11: Collecting real-time running data; the real-time running data includes CPU usage, memory occupancy, network bandwidth usage, storage space usage, and task queue state information;

[0012] Step 12: Constructing the real-time running data into a system state vector through feature engineering;

[0013] The priority adjustment module includes the following steps for feature extraction of tasks in the measurement and control system:

[0014] Step 21: Extracting task features of each task from the task queue state information in the system state vector;

[0015] Step 22: Classifying the extracted task features by using a support vector machine, and dividing the tasks into three types of calculation-intensive, IO-intensive, and mixed;

[0016] Step 23: Predicting the resource demand of different types of tasks classified by using a linear regression model.

[0017] The priority adjustment module establishes a dynamic task priority adjustment mechanism in the following manner:

[0018] Extracting the waiting time in the task queue state information in the system state vector, and adjusting the task priority adaptively according to the waiting time and the system state vector;

[0019] The adaptive adjustment of the task priority includes:

[0020] The priority adjustment module is composed of a resource competition analysis unit, a priority calculation unit, and a priority update unit.

[0021] The resource competition analysis unit calculates the current system resource competition degree based on the system resource state sub-vector and the network resource state sub-vector in the system state vector. The system resource competition degree is described by a resource competition index, which is calculated by weighting factors such as CPU usage, memory occupancy, network bandwidth usage, and storage space usage to quantify the degree of system resource tension.

[0022] The priority calculation unit calculates a dynamic priority adjustment factor for each task based on the waiting time, the resource competition index, and the resource demand of the task.

[0023] The dynamic priority adjustment factor calculation method implemented by the priority calculation unit includes the following sub-steps:

[0024] Step 221: Construct a waiting time compensation function, which increases non-linearly with the increase of task waiting time, to prevent long-term starvation of tasks;

[0025] Step 222: Construct a resource demand penalty function to appropriately lower the priority of tasks with large resource demand, but set an upper limit to avoid high resource demand tasks from never being executed;

[0026] Step 223: Construct a system load adaptive function to dynamically adjust the weights of waiting time compensation and resource demand penalty according to the current system load condition;

[0027] Step 224: Weighted fusion of task priority, waiting time compensation value, resource demand penalty value, and system load adaptive factor in the queue state information to calculate the final dynamic priority adjustment factor;

[0028] The priority updating unit receives the dynamic priority adjustment factor and updates the priority information of the task.

[0029] The macro scheduling layer in the double-layer scheduling decision module adopts a genetic algorithm including the following steps:

[0030] Step 31: Receive the system state vector and the adjusted task priority information, and design a chromosome coding method for global resource allocation optimization of the macro scheduling layer to represent the resource allocation scheme;

[0031] The design of the chromosome coding method includes the following sub-steps:

[0032] Step 311: Design an integer coding-based chromosome structure to represent the mapping relationship between the predicted task resource demand and the system resource;

[0033] Step 312: Based on the chromosome structure, design a segmentation strategy for the chromosome, which respectively encodes the allocation scheme of different types of resources;

[0034] Step 313: Based on the segmentation strategy, design a constraint condition coding mode composed of a hard constraint coding unit, a soft constraint coding unit and a constraint priority unit, so that the constraint resource allocation satisfies the system limit;

[0035] Step 314: Based on the task priority information, design a priority coding mode composed of a priority mapping unit, an execution order coding unit and an importance coding unit, reflecting the execution order and importance of the task;

[0036] Step 32: For the global resource allocation optimization of the macro-scheduling layer, design a fitness function composed of a target function construction unit, a weight allocation unit and a normalization processing unit, which comprehensively considers system throughput and load balancing degree;

[0037] For the genetic algorithm, the corresponding fitness function includes the following sub-steps:

[0038] Step 321: Based on the chromosome coding mode, construct a system throughput evaluation function composed of a task execution time estimation unit, a parallelism analysis unit and a throughput calculation unit, to calculate the number of tasks completed per unit time;

[0039] Step 322: Based on the chromosome coding mode, construct a load balancing degree evaluation function composed of a resource utilization rate calculation unit, a variance calculation unit and an equalization degree scoring unit, to calculate the variance of each resource utilization rate;

[0040] Step 323: Based on the chromosome coding mode, construct an energy consumption evaluation function composed of a resource energy consumption model unit, a task energy consumption calculation unit and a system energy consumption optimization unit, to calculate the energy consumption of the resource allocation scheme;

[0041] Step 324: Based on the chromosome coding mode, construct a time delay evaluation function composed of a task scheduling sequence generation unit, a critical path analysis unit and a response time calculation unit, to calculate the average task completion time.

[0042] Step 33: For the global resource allocation optimization of the macro-scheduling layer, introduce an adaptive crossover and mutation operator composed of a population diversity evaluation unit, a parameter self-adaptive adjustment unit and an operation execution unit, to dynamically adjust the parameters according to the population diversity;

[0043] For the global resource allocation optimization of the macro-scheduling layer, the introduction of the adaptive crossover and the mutation operator includes the following sub-steps:

[0044] Step 331: Based on the fitness function evaluation results, the fitness difference analysis unit, the crossover probability calculation unit, and the individual selection unit are constructed to dynamically adjust the crossover probability according to the fitness difference of the population individuals;

[0045] Step 332: Based on the results of the dynamically adjusted crossover probability, the evolution state monitoring unit, the convergence speed calculation unit, and the mutation probability adjustment unit are constructed to dynamically adjust the mutation probability according to the evolution generation number and the convergence speed;

[0046] Step 333: Based on the chromosome encoding mode, a special crossover operation for resource allocation problems is designed, which is composed of the crossover point selection unit, the intra-segment crossover unit, and the effectiveness repair unit;

[0047] Step 334: Based on the special crossover operation, a local search mutation operation for scheduling optimization is designed, which is composed of the mutation position selection unit, the neighborhood generation unit, and the optimal replacement unit;

[0048] Step 34: An elite preservation strategy is designed for the global resource allocation optimization of the macro-scheduling layer, which is composed of the elite selection unit, the elite preservation unit, and the elite re-introduction unit;

[0049] The micro-scheduling layer in the double-layer scheduling decision module uses a reinforcement learning model for real-time scheduling decision, including the following steps:

[0050] Step 35: Receive the system state vector and the updated task queue state information, and define the state space of the micro-scheduling layer;

[0051] Step 36: Based on the resource allocation scheme of the macro-scheduling layer, define the action space of the micro-scheduling layer as a specific resource allocation decision;

[0052] Step 37: Based on the energy consumption evaluation function and the time delay evaluation function, design the reward function of the micro-scheduling layer as the weighted sum of task completion time and energy consumption;

[0053] Step 38: Based on the defined state space, action space, and reward function, use the Q-learning algorithm to optimize the policy of the reward function;

[0054] The top-down guidance feedback includes the following steps:

[0055] Step 41: Receive the resource allocation scheme optimized by the macro-scheduling layer, and construct the guidance information receiving unit, the constraint condition conversion unit, and the constraint condition application unit to use the resource allocation scheme as the constraint condition of the micro-scheduling layer;

[0056] Step 42: the macro-scheduling layer periodically transmits the evaluated global optimization target and resource usage boundary to the micro-scheduling layer;

[0057] A deep learning-based TT&C system adaptive adjustment method is proposed, comprising the following steps:

[0058] Step 1: collect real-time running data of the TT&C system, and construct the pre-processed real-time running data into a system state vector;

[0059] Step 2: based on the system state vector, feature extraction is performed on the tasks in the TT&C system, support vector machines are used to classify the tasks, resource demand of different types of tasks is predicted, task priority is dynamically adjusted based on resource demand, and task priority information is generated;

[0060] Step 3: the system state vector and the adjusted task priority information are received by a pre-constructed macro-scheduling layer, a genetic algorithm is used for global resource allocation optimization, a reinforcement learning model is used for real-time scheduling decision based on the optimization result of the macro-scheduling layer by a pre-constructed micro-scheduling layer, and the macro-scheduling layer and the micro-scheduling layer are optimized by top-down guidance feedback and bottom-up correction feedback;

[0061] Step 4: based on the resource allocation scheme output by the macro-scheduling layer after optimization, specific resource allocation instructions are generated.

[0062] Compared with the prior art, the beneficial effects of the present application are:

[0063] The present application firstly collects real-time running data of the system and pre-processes it into a system state vector to provide a basis for subsequent decision-making. Then, support vector machines are used to classify different types of tasks, predict resource demand, and dynamically adjust task priority, so that the system can intelligently allocate resources according to the current environment. At the scheduling level, the macro-scheduling layer uses an improved genetic algorithm for global resource allocation optimization to provide the optimal resource allocation strategy for the overall system; the micro-scheduling layer realizes real-time scheduling decision based on a lightweight reinforcement learning model, quickly responds to local changes, and solves the technical problems of inflexible resource allocation, untimely priority adjustment, and difficulty in balancing global optimization and local response in traditional TT&C systems, significantly improving system resource utilization, reducing task response delay, and enhancing the system's emergency handling capability. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The flowchart of the deep learning-based TT&C system adaptive adjustment method in embodiment 1 of the present application;

[0065] Figure 2A module connection relationship diagram of the adaptive adjustment system of the deep learning-based measurement and control system in Embodiment 2 of the present application. DETAILED DESCRIPTION

[0066] The technical solutions of the present application will be described clearly and completely in combination with embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application. Embodiment 1

[0067] As shown in Figure 1 , the adaptive adjustment system of the deep learning-based measurement and control system comprises:

[0068] The adaptive adjustment system of the deep learning-based measurement and control system comprises a resource state perception module, a priority adjustment module, a double-layer scheduling decision module and a resource scheduling execution module. The modules are connected through electrical connection.

[0069] The resource state perception module collects real-time running data of the measurement and control system, pre-processes the real-time running data, and constructs the pre-processed real-time running data into a system state vector. The system state vector is sent to the priority adjustment module and the double-layer scheduling decision module as an input of resource scheduling decision.

[0070] The priority adjustment module receives the system state vector, extracts features of tasks in the measurement and control system, classifies the tasks by using a support vector machine, predicts resource demand of different types of tasks, dynamically adjusts the priority of the tasks based on the resource demand, and transmits the adjusted task priority information to the double-layer scheduling decision module.

[0071] The double-layer scheduling decision module comprises a macro-scheduling layer and a micro-scheduling layer. The macro-scheduling layer receives the system state vector and the adjusted task priority information, and performs global resource allocation optimization by using a genetic algorithm. The micro-scheduling layer performs real-time scheduling decision by using a reinforcement learning model based on the optimization result of the macro-scheduling layer, and takes the resource allocation scheme of the macro-scheduling layer as a constraint condition of the micro-scheduling layer by using a top-down guidance feedback mode. The double-layer scheduling decision module sends the optimized resource allocation scheme and the result of real-time scheduling decision to the resource scheduling execution module.

[0072] The resource scheduling execution module receives the resource allocation scheme output by the double-layer scheduling decision module, and generates a specific resource allocation instruction.

[0073] The resource state perception module collects real-time running data of the TT&C system, including the following steps:

[0074] Step 11: Collecting real-time running data; the real-time running data includes CPU usage, memory occupancy, network bandwidth usage, storage space usage, and task queue state information;

[0075] Specifically, in this embodiment, the real-time running data is collected through a distributed resource monitoring network composed of a system resource monitor, a network resource monitor, and a task queue monitor.

[0076] The system resource monitor is deployed on each computing node of the TT&C system and monitors system resource indicators such as CPU usage, memory occupancy, and storage space usage in real time; the computing node refers to a combination of hardware devices and software environments that perform computing tasks in the TT&C system;

[0077] The network resource monitor is deployed on the network key node of the TT&C system and monitors network performance indicators such as network bandwidth usage, network delay, and data packet loss rate; the network key node can be a core switch or core router in the network where the TT&C system is located;

[0078] The task queue monitor monitors the queue state information of various tasks in the TT&C system, including task queue length, task waiting time, task priority distribution, task type, computing complexity, data volume, and time constraints, task resource demand characteristics, and task execution status, and other task scheduling related parameters. The task queue monitor analyzes the task descriptor in real time through a task descriptor parser to extract the basic attributes and resource demand characteristics of the task and form structured task queue state information.

[0079] The system resource indicators, network performance indicators, and task scheduling related parameters collected each time form real-time running data;

[0080] Step 12: Constructing the real-time running data into a system state vector through feature engineering;

[0081] Specifically, in the feature engineering process, first, the system resource indicators, network performance indicators, and task scheduling related parameters are aggregated in a time window to calculate statistical features such as average value, maximum value, minimum value, and change rate within several time windows; then, the statistical features are analyzed for correlation to identify and retain the most critical feature combinations representing the system state; finally, these feature combinations are organized into a fixed-dimensional system state vector according to a predefined order and weight.

[0082] For task queue state information, the feature engineering process includes: task clustering analysis, clustering similar task features; queue state feature extraction, calculating queue depth, average waiting time, priority distribution entropy and other statistical features; task flow feature extraction, analyzing task arrival rate, completion rate and blocking rate and other dynamic features; resource matching degree analysis, evaluating the matching degree of the current task set and available resources. These features together constitute the task queue state sub-vector, which comprehensively represents the task scheduling state of the measurement and control system.

[0083] The constructed system state vector contains three main parts: system resource state sub-vector, representing the use of CPU, memory, storage and other system resources; network resource state sub-vector, representing network bandwidth, delay and other network performance indicators; task queue state sub-vector, representing task queue length, waiting time, task type distribution, priority distribution, resource demand characteristics and other task scheduling parameters. The three sub-vectors together constitute a complete system state vector, which comprehensively reflects the real-time running state of the measurement and control system.

[0084] Further, the priority adjustment module extracts the following steps for the features of the tasks in the measurement and control system:

[0085] Step 21: Extract the task features of each task from the task queue state information in the system state vector; the task features include but are not limited to task type, computational complexity, data volume, priority and time constraint, etc.

[0086] Step 22: Classify the extracted task features using a support vector machine to divide tasks into three categories: compute-intensive, IO-intensive and hybrid;

[0087] In the specific implementation process of the present application, the main features of compute-intensive tasks are high computational complexity and relatively low IO operation frequency, typical examples being scientific computing, image processing and data analysis tasks; the main features of IO-intensive tasks are frequent input and output operations and relatively low computational complexity, typical examples being data transmission, file reading and writing and network communication tasks; hybrid tasks have high demand in both computing and IO operations, such as database transaction processing and real-time data stream processing tasks.

[0088] Specifically, the support vector machine is a supervised learning classifier based on kernel functions, used to map task feature vectors to predefined task categories. The support vector machine is composed of a model training unit, a feature mapping unit and a classification decision unit.

[0089] The model training unit uses historical task data to train a support vector machine model. The historical task data contains a number of labeled task samples, each of which contains a vector of task features and a corresponding task category label (determined to be compute-intensive, IO-intensive, and mixed). By minimizing the structural risk, the support vector machine learns the optimal classification hyperplane in the feature space, achieving accurate classification of the task category of the new task.

[0090] The feature mapping unit uses a radial basis function as a kernel function to map the task feature vector to a high-dimensional feature space, enhancing the separability of the features. In the high-dimensional feature space, different types of tasks exhibit more obvious clustering characteristics, facilitating accurate classification by the support vector machine.

[0091] The classification decision unit receives the mapped high-dimensional feature vector, calculates the distance of the high-dimensional feature vector to each classification hyperplane, and determines the task category based on the distance relationship. The decision-making process adopts a one-versus-all strategy, constructing a number of binary classifiers to determine whether the task belongs to compute-intensive, IO-intensive, or mixed, and finally synthesizing the results of each classifier to determine the final category of the task.

[0092] Step 23: For different types of tasks classified, a linear regression model is used to predict the resource demand;

[0093] Specifically, the linear regression model is a statistical learning model based on task feature vectors to predict resource demand. The linear regression model consists of a feature selection unit, a model construction unit, and a prediction execution unit.

[0094] Among them, the feature selection unit selects the most relevant feature subset according to the task type. For example, for compute-intensive tasks, select features such as computational complexity, algorithm type, and data size; for IO-intensive tasks, select features such as IO operation frequency, data transfer volume, and storage access mode; for mixed tasks, consider both compute and IO-related features. The feature selection process uses a correlation analysis-based method to retain highly correlated features and eliminate redundant and irrelevant features.

[0095] The model construction unit constructs an independent linear regression model for each resource type, including a CPU demand prediction model, a memory demand prediction model, a network bandwidth demand prediction model, and a storage space demand prediction model. Each model uses a weighted least squares method to estimate model parameters, with weights adjusted dynamically according to historical prediction accuracy to improve prediction accuracy.

[0096] The prediction execution unit receives a task feature vector and task type information, selects a corresponding linear regression model to perform resource demand prediction.

[0097] The priority adjustment module establishes a task priority dynamic adjustment mechanism in the following manner:

[0098] The waiting time in the task queue state information in the system state vector is extracted, and the task priority is adaptively adjusted according to the waiting time and the system state vector.

[0099] Specifically, the adaptive adjustment of the task priority comprises:

[0100] The priority adjustment module comprises a resource competition analysis unit, a priority calculation unit, and a priority updating unit.

[0101] The resource competition analysis unit calculates the current system resource competition degree based on the system resource state sub-vector and the network resource state sub-vector in the system state vector. The system resource competition degree is described by a resource competition index, which is calculated by weighting factors such as CPU usage, memory occupancy, network bandwidth usage, and storage space usage, and has quantified the system resource tension degree. Obviously, the higher the resource competition index, the more intense the system resource competition, and more accurate task priority adjustment is needed.

[0102] The priority calculation unit calculates a dynamic priority adjustment factor for each task based on the waiting time, the resource competition index, and the resource demand of the task. The dynamic priority adjustment factor is calculated by an adaptive weighting method, for example, the waiting time weight, the resource demand weight, and the original priority weight are dynamically adjusted according to the system load condition, to ensure the rationality of the priority adjustment under different system states.

[0103] In the specific implementation process of the present application, the dynamic priority adjustment factor calculation method implemented by the priority calculation unit comprises the following sub-steps:

[0104] Step 221: Construct a waiting time compensation function, which increases non-linearly with the increase of the task waiting time, to prevent long-term starvation of the task;

[0105] Step 222: Construct a resource demand penalty function, which appropriately reduces the priority of the task with large resource demand, but sets an upper limit to avoid the task with high resource demand from being executed forever;

[0106] Step 223: Construct a system load adaptive function to dynamically adjust the weights of the latency compensation and resource demand penalty according to the current system load condition;

[0107] Step 224: Weighted fusion of task priority, latency compensation value, resource demand penalty value and system load adaptive factor in queue state information to calculate the final dynamic priority adjustment factor;

[0108] The priority updating unit receives the dynamic priority adjustment factor and updates the priority information of the task.

[0109] Further, the macro-scheduling layer in the double-layer scheduling decision module adopts a genetic algorithm, including the following steps:

[0110] Step 31: Receive the system state vector and the adjusted task priority information, and design a chromosome coding mode for global resource allocation optimization of the macro-scheduling layer to represent the resource allocation scheme;

[0111] Specifically, the design of the chromosome coding mode includes the following sub-steps:

[0112] Step 311: Design an integer coding-based chromosome structure to represent the mapping relationship between the predicted task resource demand and the system resources;

[0113] Specifically, the integer coding-based chromosome structure is composed of a chromosome coding unit and a resource mapping unit.

[0114] The chromosome coding unit uses a multi-dimensional integer array to represent the resource allocation scheme, and each chromosome is composed of several gene bits, each of which corresponds to an allocation relationship between a task and a resource. In the chromosome structure, the arrangement order of the gene bits is consistent with the order of the tasks in the task queue, ensuring the consistency and interpretability of the coding structure. The value of each gene bit represents the resource identifier allocated to the task, which is directly mapped to specific computing nodes, memory blocks, network channels or storage units through integer coding.

[0115] The resource mapping unit converts the chromosome coding into actual resource allocation instructions. The resource mapping unit maintains a mapping table between resource identifiers and physical resources to ensure that the integer coding in the chromosome can accurately correspond to the actual resources of the measurement and control system. The resource mapping process considers the hierarchical structure and dependency relationship of resources, such as allocating computing nodes first, then allocating memory and storage resources within the nodes, and finally allocating network channels to form a complete resource allocation scheme.

[0116] Step 312: Based on the chromosome structure, design a chromosome segmentation strategy to encode the allocation schemes of different types of resources respectively;

[0117] The different types of resources can include, but are not limited to, computing resources, memory resources, network resources, and storage resources.

[0118] Specifically, the chromosome segmentation strategy divides the chromosome into several segments, each corresponding to an allocation scheme of a resource type. The chromosome segmentation strategy consists of a resource type division unit, an intra-segment encoding unit, and an inter-segment association unit.

[0119] The resource type division unit divides the chromosome into computing resource segments, memory resource segments, network resource segments, and storage resource segments according to the resource types of the measurement and control system. The length of each resource segment is related to the number of resources of the corresponding type and the number of tasks in the system, ensuring that the allocation scheme of this type of resource can be completely represented.

[0120] The intra-segment encoding unit designs a dedicated encoding method for each resource type. For the computing resource segment, a combination of processor identifier and core identifier is used for encoding; for the memory resource segment, a combination of memory block identifier and allocation size is used for encoding; for the network resource segment, a combination of channel identifier and bandwidth allocation ratio is used for encoding; for the storage resource segment, a combination of storage unit identifier and access permission is used for encoding.

[0121] The inter-segment association unit maintains the association between different resource segments to ensure consistency in resource allocation. For example, when a task is allocated to a specific computing node, the allocation of its memory and storage resources should give priority to resources that are physically close to the computing node to reduce data transmission overhead. The inter-segment association unit quantifies the degree of association between different resources by establishing a resource affinity matrix to guide the coordinated allocation of multiple types of resources.

[0122] Step 313: Based on the segmentation strategy, design a constraint condition encoding method consisting of a hard constraint encoding unit, a soft constraint encoding unit, and a constraint priority unit to ensure that the resource allocation satisfies the system limitations;

[0123] Specifically, the hard constraint encoding unit encodes the system limitation conditions that must be met, including resource capacity constraints, task dependency constraints, and resource mutual exclusion constraints. Resource capacity constraints ensure that the total amount of resources allocated to all tasks does not exceed the available resources of the system; task dependency constraints ensure that tasks with dependency relationships are allocated resources in the correct order; resource mutual exclusion constraints ensure that exclusive resources are not occupied by multiple tasks at the same time. Hard constraint conditions are encoded by a feasibility check function, which verifies each chromosome. Chromosomes that do not meet the hard constraints will be marked as invalid solutions.

[0124] The soft constraint coding unit codes constraints that affect system performance but can be appropriately relaxed, such as load balancing constraints, energy consumption control constraints, and response time constraints. Soft constraints are coded by a penalty function, which quantifies the degree of violation of soft constraints and incorporates a penalty value into the calculation of the fitness function, guiding the genetic algorithm to evolve in the direction of satisfying soft constraints.

[0125] The constraint priority unit sets the priority of different constraints and weighs when multiple constraints conflict. Constraint priority is coded by a weight coefficient, and violation of a high-priority constraint will result in a larger penalty value. The constraint priority can be dynamically adjusted according to the current state of the measurement and control system, for example, when the system load is low, the priority of the energy consumption control constraint can be increased; when the task is urgent, the priority of the response time constraint can be increased.

[0126] Step 314: Based on the task priority information, design a priority coding method composed of a priority mapping unit, an execution order coding unit, and an importance coding unit, reflecting the execution order and importance of the task;

[0127] Specifically, the priority mapping unit maps the task priority information output by the priority adjustment module into the chromosome coding. The mapping process uses a priority conversion function to convert the task priority value into adjustment factors for selection probability, crossover probability, and mutation probability in genetic algorithms, so that high-priority tasks receive more attention in resource allocation.

[0128] The execution order coding unit encodes the execution order information of the task in the chromosome. The execution order is coded by the relative position of the gene site and the special marker gene in the chromosome, ensuring that tasks with dependencies are executed in the correct order. For tasks with predecessor tasks, the gene site contains a reference to the completion flag of the predecessor task, and the task can only be scheduled for execution after the completion of the predecessor task.

[0129] The importance coding unit encodes the importance information of the task in the chromosome. Importance is coded by adding a weight field in the gene site, affecting the priority order and resource allocation proportion of resource allocation. The gene site of a high-importance task obtains a higher weight value and has priority in resource competition and may obtain more resource allocation. Importance coding also affects the calculation of the fitness function, so that the completion of important tasks contributes more to the fitness, guiding the genetic algorithm to preferentially meet the resource needs of important tasks.

[0130] Step 32: For the global resource allocation optimization of the macro-scheduling layer, design a fitness function composed of a target function construction unit, a weight allocation unit, and a normalization processing unit, considering system throughput and load balancing degree;

[0131] Specifically, the objective function construction unit constructs a mathematical model for evaluating the quality of the resource allocation scheme. Based on multiple evaluation indexes such as system throughput, load balancing degree, energy consumption efficiency, and task response time, the unit constructs a multi-objective optimization function. The system throughput objective function calculates the number of tasks completed per unit time; the load balancing degree objective function calculates the resource utilization rate variance of each computing node, memory unit, network channel, and storage unit, and the smaller the variance, the more balanced the load; the energy consumption efficiency objective function calculates the energy consumption required to complete all tasks; and the task response time objective function calculates the average time from task submission to completion.

[0132] The weight allocation unit allocates weight coefficients to each objective in the multi-objective optimization function. Weight allocation is based on the current state and operating requirements of the measurement and control system, dynamically adjusting the importance of each objective. For example, in a high-load situation, the weight of system throughput can be increased; in an energy-limited situation, the weight of energy consumption efficiency can be increased; and in a situation requiring high real-time performance, the weight of task response time can be increased. Weight allocation uses an adaptive mechanism to automatically adjust according to the indicators in the system state vector, ensuring that the fitness function can reflect the optimization requirements of the current system.

[0133] The normalization processing unit normalizes the calculation results of each objective function, allowing indicators of different dimensions to be weighted and summed. Normalization uses the maximum and minimum value normalization method to map the values of each objective function to the [0, 1] interval. After normalization, the objective function values are multiplied by the corresponding weight coefficients, and then summed to obtain the final fitness value. The higher the fitness value, the better the resource allocation scheme, and the greater the chance of being retained and reproduced in the selection operation of the genetic algorithm.

[0134] The corresponding fitness function for the genetic algorithm is designed to include the following sub-steps:

[0135] Step 321: Based on the chromosome coding mode, construct a system throughput evaluation function composed of a task execution time estimation unit, a parallelism analysis unit, and a throughput calculation unit to calculate the number of tasks completed per unit time;

[0136] Specifically, the task execution time estimation unit estimates the execution time of each task based on the resource demand characteristics of the task and the allocated resources. The estimation process takes into account the computational complexity of the task, data volume, resource utilization efficiency, and resource performance parameters, and calculates the expected execution time of the task under the given resource configuration through a performance model.

[0137] The parallelism analysis unit analyzes the task parallel execution in the resource allocation scheme. The analysis process considers the dependency between tasks, the independence of resource allocation, and the parallel processing capability of the system, and determines the maximum number of tasks that can be executed simultaneously by the system under the given resource allocation scheme.

[0138] The throughput calculation unit calculates the number of tasks that can be completed by the system per unit time based on task execution time and system parallelism. The calculation process considers the arrival mode of tasks, execution time distribution, and system scheduling strategy, and obtains the expected throughput of the system through queuing theory model or discrete event simulation method. The throughput calculation result is an important part of the fitness function, guiding the genetic algorithm to optimize the processing capability of the system.

[0139] Step 322: Based on the chromosome coding mode, construct a load balancing degree evaluation function composed of resource utilization rate calculation unit, variance calculation unit and balance degree scoring unit, calculate the variance of each resource utilization rate;

[0140] Specifically, the resource utilization rate calculation unit calculates the expected utilization rate of each resource unit in the system based on the chromosome coding resource allocation scheme. The calculation process considers the number of tasks allocated to each resource, the resource demand intensity of tasks, and the processing capability of resources, and obtains the utilization rate prediction value of each resource unit.

[0141] The variance calculation unit calculates the statistical variance of the utilization rate of the same type of resource. The calculation process groups the utilization rates of computing resources, memory resources, network resources and storage resources respectively, calculates the variance of the resource utilization rate in each group, and the smaller the variance value, the more balanced the load distribution of the type of resource.

[0142] The balance degree scoring unit evaluates the load balancing degree of each type of resource as a balance degree score. The scoring process considers the importance and sensitivity of different types of resources, and performs weighted average on the variance values of each type of resource to obtain the overall load balancing degree score of the system. The balance degree score is a part of the fitness function, guiding the genetic algorithm to optimize the balanced use of resources and avoid resource hotspots and resource starvation.

[0143] Step 323: Based on the chromosome coding mode, construct an energy consumption evaluation function composed of resource energy consumption model unit, task energy consumption calculation unit and system energy consumption optimization unit, calculate the energy consumption of the resource allocation scheme;

[0144] Specifically, the resource energy consumption model unit establishes energy consumption models of various resources in the monitoring and control system. The model considers the relationship between the static power consumption, dynamic power consumption and resource utilization rate of the resources, and describes the energy consumption performance of the resources under different loads through an energy consumption characteristic curve. The energy consumption model covers the energy consumption characteristics of computing resources, memory resources, network resources and storage resources, and provides basic data for energy consumption evaluation.

[0145] The task energy consumption calculation unit calculates the energy consumption of each task execution process based on the resource energy consumption model and the resource usage of the task. The calculation process considers the execution time, resource usage intensity and energy consumption characteristics of the resources, and obtains the total energy consumption prediction value of the task through energy consumption integration method.

[0146] The system energy consumption optimization unit considers the energy consumption of all tasks and the energy consumption management strategy of the system to calculate the energy efficiency of the entire resource allocation scheme. The optimization process considers dynamic power management of resources, energy-aware scheduling of tasks and heat dissipation constraints of the system, and evaluates the energy efficiency of the resource allocation scheme through the energy efficiency ratio index. The energy consumption evaluation result is used as a component of the fitness function to guide the genetic algorithm to optimize energy use while ensuring performance.

[0147] Step 324: Based on the chromosome coding mode, a time delay evaluation function composed of a task scheduling sequence generation unit, a critical path analysis unit and a response time calculation unit is constructed to calculate the average completion time of the tasks;

[0148] Specifically, the task scheduling sequence generation unit generates the scheduling and execution sequence of the tasks based on the chromosome-coded resource allocation scheme and the task dependency relationship. The generation process considers the priority, dependency constraints and resource availability time of the tasks, and constructs the execution Gantt chart of the tasks through list scheduling algorithm or critical path method to clearly define the start time and completion time of each task.

[0149] The critical path analysis unit identifies the critical path in the task execution sequence. The analysis process considers the dependency relationship and execution time between tasks to traverse the longest execution path from the first task to the last task. The execution time of the tasks on the critical path determines the shortest completion time of the entire task set and is the focus of optimization scheduling.

[0150] The response time calculation unit calculates the average response time of the tasks from submission to completion. The calculation process considers the waiting time, execution time and resource competition of the tasks, and obtains the average response time index of the system through weighted average method. The response time evaluation result is used as a component of the fitness function to guide the genetic algorithm to optimize the timely processing of the tasks and improve the real-time performance of the system.

[0151] Step 33: Introducing an adaptive crossover and mutation operator composed of a population diversity evaluation unit, a parameter self-adaptive adjustment unit, and an operation execution unit into the global resource allocation optimization of the macro-scheduling layer, and dynamically adjusting the parameters according to the population diversity;

[0152] Specifically, the population diversity evaluation unit evaluates the diversity level of the current population. The evaluation process adopts a method combining genetic diversity indicators and phenotype diversity indicators to calculate the average difference between individuals in the population. Genetic diversity measures the difference in chromosome coding through Hamming distance or edit distance; phenotype diversity measures the difference in individual performance through the distribution range and variance of fitness values.

[0153] The parameter self-adaptive adjustment unit dynamically adjusts the crossover probability and mutation probability according to the population diversity evaluation results. When the population diversity is low, increase the mutation probability and reduce the crossover probability to promote the population to explore new solution space; when the population diversity is high, increase the crossover probability and reduce the mutation probability to accelerate the population convergence to the optimal solution. Parameter adjustment uses fuzzy control rules or adaptive functions to achieve smooth mapping between parameters and diversity.

[0154] The operation execution unit executes crossover and mutation operations. Crossover operation adopts resource allocation problem-specific crossover methods such as partial mapping crossover or sequential crossover to ensure that the generated offspring chromosomes still represent valid resource allocation schemes; mutation operation adopts local search strategies such as neighborhood search or simulated annealing to explore better resource allocation schemes while maintaining the effectiveness of chromosome structure.

[0155] Introducing the adaptive crossover and mutation operator for the global resource allocation optimization of the macro-scheduling layer includes the following sub-steps:

[0156] Step 331: Based on the fitness function evaluation results, dynamically adjust the crossover probability according to the fitness difference of population individuals by constructing a fitness difference analysis unit, a crossover probability calculation unit, and an individual selection unit;

[0157] The fitness difference analysis unit calculates the statistical characteristics of individual fitness in the population, including maximum fitness, minimum fitness, average fitness, and fitness standard deviation. The analysis results reflect the evolution state and convergence trend of the population, providing a basis for dynamic adjustment of the crossover probability.

[0158] The crossover probability calculation unit calculates the crossover probability of each pair of individuals to be crossed based on the fitness difference analysis result. The calculation process considers the difference between the fitness value of an individual and the average fitness of the population. For individuals with a fitness value higher than the average, a lower crossover probability is given to protect their superior genes. For individuals with a fitness value lower than the average, a higher crossover probability is given to promote their gene recombination. The calculation of the crossover probability uses a nonlinear mapping function, and different adjustment strategies are used at the beginning and end of population convergence.

[0159] The individual selection unit selects pairs of individuals to participate in the crossover operation according to the calculated crossover probability. The selection process uses the roulette method or the tournament method, combined with the crossover probability for random selection, to ensure that the selection process has both randomness and a tendency to select individuals with higher fitness. The selected pairs of individuals will undergo crossover operations to generate new offspring individuals, enriching the genetic diversity of the population.

[0160] Step 332: Based on the results of the dynamic adjustment of the crossover probability, the evolution state monitoring unit, the convergence speed calculation unit, and the mutation probability adjustment unit are constructed to dynamically adjust the mutation probability according to the evolution generation number and the convergence speed;

[0161] The evolution state monitoring unit tracks and records the evolution generation number and the population state changes of the genetic algorithm. The monitoring process records the optimal fitness, average fitness, and fitness variance of each generation, constructing a historical trajectory of population evolution to provide time series data for the dynamic adjustment of the mutation probability.

[0162] The convergence speed calculation unit calculates the convergence speed of the population based on the evolution state monitoring data. The calculation process considers the optimal fitness change rate and population diversity change trend of consecutive generations, and obtains the convergence speed index of the current population through the sliding window method or the exponential smoothing method.

[0163] The mutation probability adjustment unit dynamically adjusts the mutation probability according to the evolution generation number and the convergence speed. The adjustment strategy follows the principle of "low mutation in early stage, high mutation in middle stage, and fine mutation in later stage". In the early stage of the algorithm, a lower mutation probability is maintained to avoid destroying superior genes; in the middle stage, the mutation probability is increased to enhance the exploration ability of the population; in the later stage, the mutation probability is reduced for fine search to accelerate convergence to the optimal solution. The specific adjustment of the mutation probability uses a piecewise function or an adaptive function to achieve dynamic matching of the mutation probability and the evolution state.

[0164] Step 333: Based on the chromosome encoding method, a special crossover operation for resource allocation problems is designed, which is composed of a crossover point selection unit, an intra-segment crossover unit, and an effectiveness repair unit;

[0165] Specifically, the crossover point selection unit selects appropriate crossover points in the parent chromosome. The selection process considers the segment structure of the chromosome and the integrity of resource allocation, preferring to select crossover points at the boundaries of resource segments to ensure that the crossover operation does not disrupt the internal consistency of resource allocation. For multi-point crossover, the selection of crossover points also considers the dependency between resources, avoiding splitting strongly related resource allocation schemes into different crossover segments.

[0166] The intra-segment crossover unit performs specialized crossover operations for different resource segments. For computing resource segments, a partial mapping crossover method is used to maintain one-to-one mapping of tasks to processors; for memory resource segments, a uniform crossover method is used to independently exchange the memory allocation of each task; for network resource segments, a sequential crossover method is used to maintain the relative order of network channel allocation; for storage resource segments, a single-point crossover method is used to exchange large blocks of storage allocation schemes.

[0167] The validity repair unit checks whether the child chromosome generated after crossover meets the constraint conditions of resource allocation. The checking process verifies hard constraint conditions such as resource capacity constraints, task dependency constraints, and resource mutual exclusion constraints. For chromosomes that violate constraints, local adjustment or greedy repair methods are used for correction to ensure that the child chromosome represents an effective resource allocation scheme.

[0168] Step 334: Based on the specialized crossover operation, design a local search mutation operation for scheduling optimization composed of a mutation position selection unit, a neighborhood generation unit, and an optimal replacement unit;

[0169] Specifically, the mutation position selection unit selects the positions in the chromosome that need to be mutated. The selection process considers the unevenness of resource utilization and the criticality of tasks, preferring to select highly loaded resource units or tasks on the critical path for mutation to improve the relevance and effectiveness of the mutation operation.

[0170] The neighborhood generation unit generates multiple candidate mutation schemes based on the selected mutation position. The generation process uses problem-specific neighborhood definitions such as resource reallocation neighborhood, task exchange neighborhood, or priority adjustment neighborhood to construct the local search space of the current solution. Neighborhood generation considers the constraint conditions of resource allocation and only generates effective mutation schemes that meet the constraints.

[0171] The optimal replacement unit evaluates the fitness of all candidate mutation schemes and selects the optimal scheme to replace the original solution. The evaluation process uses an incremental calculation method to only calculate the impact of the mutated part on fitness, improving evaluation efficiency. The replacement strategy can be greedy selection (always selecting the best scheme) or probabilistic acceptance (accepting suboptimal schemes with a certain probability), balancing the convergence speed of the algorithm and the ability to escape local optima.

[0172] Step 34: An elite preservation strategy composed of an elite selection unit, an elite preservation unit, and an elite re-introduction unit is designed for the global resource allocation optimization of the macro-scheduling layer to ensure that the optimal solution of the evaluation is not destroyed;

[0173] Specifically, the elite selection unit selects several individuals with the highest fitness from the current population as elite individuals according to the evaluation results of the fitness function during the evolution process of each generation. The number of elite individuals is dynamically determined according to the population size, and is usually maintained at a certain proportion of the total population. The elite selection process adopts a ranking-based selection method, which first sorts all individuals in descending order of fitness values, and then selects the top-ranked individuals as elite individuals.

[0174] The elite preservation unit directly copies the selected elite individuals to the next generation population without crossing and mutation operations, ensuring that these high-quality solutions are not destroyed during the evolution process. The elite preservation unit maintains an elite pool to store the optimal solutions found in historical iterations, and regularly updates the individuals in the elite pool to ensure that the elite pool always contains high-quality solutions representing different optimization directions.

[0175] The elite re-introduction unit monitors population diversity during the evolution process, and when it detects that the population diversity has fallen below a preset threshold, it selects elite individuals that are significantly different from the current population from the elite pool and reintroduces them to the current population to increase population diversity and avoid the algorithm falling into a local optimal solution. The re-introduction process adopts a distance metric-based diversity evaluation method to ensure that the reintroduced elite individuals can maximize the increase in population diversity.

[0176] The micro-scheduling layer in the double-layer scheduling decision module uses a reinforcement learning model for real-time scheduling decisions, including the following steps:

[0177] Step 35: Receive the system state vector and the updated task queue state information, and define the state space of the micro-scheduling layer;

[0178] Specifically, the state space of the micro-scheduling layer is composed of a state representation unit, a state compression unit, and a state update unit.

[0179] The state representation unit combines the system state vector and the updated task queue state information into a complete state description. The state description includes three parts: the current available resource state, the to-be-scheduled task state, and the system environment state. The available resource state describes the current available amount and utilization rate of various resources; the to-be-scheduled task state describes the type, priority, resource demand, and waiting time of each task in the task queue; and the system environment state describes environmental factors such as system load trend, resource competition degree, and task arrival mode.

[0180] The state compression unit reduces the dimension of the complete state description, reduces the dimension of the state space, and improves the convergence speed of the reinforcement learning algorithm. The state compression adopts the principal component analysis method, retains the state characteristics most critical to decision-making, and eliminates redundant and noise characteristics. The compressed state representation retains the main features of the original state information while reducing the dimension of the state space.

[0181] The state update unit updates the system state at each decision-making time point. The state update process includes: obtaining the latest system state vector from the resource state perception module; obtaining the updated task queue state information from the priority adjustment module; fusing the newly obtained information with the historical state information to form a time sequence state representation; and compressing the fused state through the state compression unit to generate the state representation at the current decision-making time.

[0182] Step 36: Based on the resource allocation scheme of the macro scheduling layer, define the action space of the micro scheduling layer as a specific resource allocation decision;

[0183] Specifically, the action space of the micro scheduling layer is composed of an action definition unit, an action constraint unit and an action execution unit.

[0184] The action definition unit represents the resource allocation decision as a multi-dimensional vector, each dimension corresponding to an allocation decision of a resource type. For each task to be scheduled, the action vector describes the specific amount of each type of resource allocated to the task. The design of the action space considers the divisibility of resources and the parallelism of tasks, supporting fine-grained resource allocation and concurrent task execution.

[0185] The action constraint unit sets constraint conditions for the action space of the micro scheduling layer based on the resource allocation scheme of the macro scheduling layer. The constraint conditions include resource total amount constraint, task priority constraint and system policy constraint. The resource total amount constraint ensures that the total amount of resources allocated by the micro scheduling layer does not exceed the total amount of resources allocated by the macro scheduling layer; the task priority constraint ensures that high-priority tasks are given priority in resource allocation; and the system policy constraint sets specific resource allocation rules according to the system management policy, such as resource reservation, resource isolation and resource reservation.

[0186] The action execution unit converts the selected action into specific resource allocation instructions and implements resource allocation through the resource scheduling execution module. The action execution process includes: parsing the selected action vector to determine the resource allocation scheme for each task; checking whether the resource allocation scheme meets all the constraint conditions; generating a sequence of resource allocation instructions; and passing the instruction sequence to the resource scheduling execution module for execution.

[0187] Step 37: Based on the energy consumption evaluation function and the delay evaluation function, design the reward function of the micro scheduling layer as the weighted sum of task completion time and energy consumption;

[0188] Specifically, the reward function of the micro-scheduling layer is composed of a reward calculation unit, a reward normalization unit and a reward adjustment unit.

[0189] The reward calculation unit considers two key indicators, task completion time and energy consumption, to construct a weighted reward function. The task completion time is calculated by a delay evaluation function, reflecting the influence of resource allocation decisions on system response performance; the energy consumption is calculated by an energy consumption evaluation function, reflecting the influence of resource allocation decisions on system energy efficiency. The reward calculation uses negative values, the shorter the completion time and the lower the energy consumption, the higher the reward value (the smaller the absolute value of the negative value).

[0190] The reward normalization unit normalizes the original reward value calculated to make the reward values generated by tasks of different scales and types comparable. The normalization process considers the complexity and resource requirements of the task, and adjusts the scale of the reward value for different tasks to ensure that the reward signal accurately reflects the decision quality and is not affected by the scale and type of the task.

[0191] The reward adjustment unit dynamically adjusts the weights of completion time and energy consumption according to the system state and optimization target. When the system load is light, increase the weight of completion time to prioritize improving system response performance; when the system load is heavy, increase the weight of energy consumption to prioritize improving system energy efficiency. The weight adjustment process uses an adaptive method to dynamically calculate the optimal weight combination according to system load conditions, task queue length and resource utilization, etc.

[0192] Step 38: Based on the defined state space, action space and reward function, a Q-learning algorithm is used to optimize the reward function;

[0193] Specifically, the Q-learning algorithm is composed of a Q value update unit, an exploration strategy unit and a policy extraction unit.

[0194] The Q value update unit maintains a Q table or Q network to store the value estimates of state-action pairs. After each decision, the Q value is updated using the observed reward and the next state using the time difference learning method. The Q value update uses an experience replay mechanism to store historical decision experiences in a replay buffer, and randomly samples a number of experiences from the buffer for batch update each time, improving learning efficiency and stability.

[0195] The exploration strategy unit balances exploration and utilization by using an ε-greedy strategy. During decision-making, an action is randomly selected with a probability of ε for exploration, and an action with the highest current Q value is selected with a probability of 1-ε for utilization. The value of ε gradually decreases with the learning process, and exploration is mainly performed in the early stage, and utilization is mainly performed in the later stage, ensuring that the algorithm can fully explore the state-action space and gradually converge to the optimal strategy.

[0196] The policy extraction unit extracts the decision-making policy according to the learned Q value. At each decision-making moment, the Q table or Q network is queried according to the current state, and the action with the highest Q value is selected as the decision-making result. The policy extraction process considers the action constraint conditions to ensure that the extracted policy meets all constraint requirements. For complex state spaces, the policy extraction unit uses a function approximation method to approximate the Q function through a neural network model, effectively handling continuous state spaces.

[0197] Further, the top-down guidance feedback includes the following steps:

[0198] Step 41: Receive the resource allocation scheme optimized by the macro scheduling layer, and construct a guidance information receiving unit, a constraint condition conversion unit, and a constraint condition application unit to use the resource allocation scheme as a constraint condition for the micro scheduling layer;

[0199] The guidance information receiving unit receives the resource allocation scheme output by the macro scheduling layer. The resource allocation scheme contains global optimization information such as system resource total allocation, task priority sorting, and resource usage boundary. The guidance information receiving unit establishes a connection with the macro scheduling layer through a high-speed data channel to ensure that the resource allocation scheme can be transmitted to the micro scheduling layer in real time, providing global guidance for real-time scheduling decisions.

[0200] The constraint condition conversion unit converts the resource allocation scheme into a constraint condition format that can be directly used by the micro scheduling layer. The conversion process includes resource boundary extraction, priority rule conversion, and time window division. Resource boundary extraction extracts the usage upper limit of each type of resource from the resource allocation scheme; priority rule conversion maps the macro task priority to the priority constraint of micro scheduling; and time window division divides the global resource allocation scheme into several scheduling periods according to the time dimension, providing time-periodic resource constraints for the micro scheduling layer.

[0201] The constraint application unit applies the converted constraint conditions to a decision-making process of the micro-scheduling layer. The application modes include action space constraint, reward function adjustment, and state space reconstruction. The action space constraint limits the micro-scheduling layer to make decisions only in an action subspace that meets the constraint conditions; the reward function adjustment adjusts the reward signal of the micro-scheduling layer according to the optimization target of the macro-scheduling layer; and the state space reconstruction adds a constraint condition related feature to the state representation of the micro-scheduling layer, so that the micro-scheduling layer can perceive the global constraint.

[0202] Step 42: The macro-scheduling layer periodically transmits the evaluated global optimization target and resource usage boundary to the micro-scheduling layer;

[0203] Specifically, the macro-scheduling layer is composed of a target transmission unit, a boundary update unit, and a transmission period control unit.

[0204] The target transmission unit is responsible for transmitting the global optimization target of the macro-scheduling layer to the micro-scheduling layer. The global optimization target includes a system throughput target, a load balancing degree target, an energy efficiency target, and a task response time target. The target transmission unit not only transmits target values, but also transmits weight coefficients of each target, guiding the micro-scheduling layer to weigh among multiple targets. Target transmission adopts a structured target description format, including target type, target value, weight coefficient, and timeliness label information.

[0205] The boundary update unit is responsible for updating and transmitting the resource usage boundary. The resource usage boundary defines the usage limit of the micro-scheduling layer on various resources, including a computing resource boundary, a memory resource boundary, a network resource boundary, and a storage resource boundary. The boundary update process considers the dynamic changes of system resource state, dynamically adjusts the resource usage boundary according to the resource utilization in the system state vector, and ensures that the boundary setting can meet the global optimization target and adapt to the real-time fluctuations of system resources.

[0206] The transmission period control unit controls the timing and frequency of target and boundary transmission. Transmission period control adopts an adaptive mechanism to dynamically adjust the transmission period according to system load conditions and task change frequency. When the system load is stable, a longer transmission period is used to reduce communication overhead; when the system load fluctuates greatly, the transmission period is shortened to ensure that the micro-scheduling layer can obtain the latest global guidance information in time. Transmission period control also includes a trigger mechanism that triggers a target and boundary transmission immediately when the system state changes significantly, to ensure that the micro-scheduling layer can quickly respond to system changes.

[0207] The resource scheduling execution module generates specific resource allocation instructions in the following manner:

[0208] According to the total system resource allocation in the resource allocation scheme, task priority sorting and resource use boundary, a corresponding resource total allocation instruction, task priority sorting instruction and resource use convenient instruction are generated for the TT&C system to constitute the resource allocation instruction. Embodiment 2

[0209] As shown in the figure, the deep learning-based TT&C system adaptive adjustment method comprises the following steps: Figure 2

[0210] Step one: collect real-time running data of the TT&C system, and construct the pre-processed real-time running data into a system state vector;

[0211] Step two: based on the system state vector, feature extraction is performed on the tasks in the TT&C system, support vector machine is used for classification of the tasks, resource demand of different types of tasks is predicted, task priority is dynamically adjusted based on the resource demand, and task priority information is generated;

[0212] Step three: the system state vector and the adjusted task priority information are received through the pre-constructed macro-scheduling layer, genetic algorithm is used for global resource allocation optimization; based on the optimization result of the macro-scheduling layer, real-time scheduling decision is made through the pre-constructed micro-scheduling layer using reinforcement learning model; and the macro-scheduling layer and the micro-scheduling layer are optimized using top-down guidance feedback and bottom-up correction feedback;

[0213] Step four: based on the resource allocation scheme output by the macro-scheduling layer after optimization, specific resource allocation instructions are generated.

[0214] The above embodiments are only used to illustrate the technical method of the present application and are not limited. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.​

Claims

1. A deep learning-based adaptive adjustment system for measurement and control systems, characterized in that, It includes a resource status awareness module, a priority adjustment module, a two-layer scheduling decision module, and a resource scheduling execution module; the modules are connected electrically. The resource status perception module collects real-time operating data from the measurement and control system, preprocesses the real-time operating data, and constructs a system status vector from the preprocessed real-time operating data; and sends the system status vector to the priority adjustment module and the two-layer scheduling decision module. The priority adjustment module receives the system state vector, extracts features from the tasks in the measurement and control system, classifies the tasks using a support vector machine, predicts the resource requirements for different types of tasks, dynamically adjusts the task priorities based on the resource requirements, and transmits the adjusted task priority information to the two-layer scheduling decision module. The two-layer scheduling decision module consists of a macro-scheduling layer and a micro-scheduling layer. The macro-scheduling layer receives the system state vector and the adjusted task priority information, and uses a genetic algorithm to optimize global resource allocation. The micro-scheduling layer, based on the optimization results of the macro-scheduling layer, uses a reinforcement learning model to make real-time scheduling decisions. Furthermore, it uses a top-down guidance feedback mechanism to apply the resource allocation scheme from the macro-scheduling layer as a constraint on the micro-scheduling layer. The two-layer scheduling decision module then sends the optimized resource allocation scheme and the results of the real-time scheduling decisions to the resource scheduling execution module. The resource scheduling execution module receives the resource allocation scheme output by the two-layer scheduling decision module and generates specific resource allocation instructions.

2. The adaptive adjustment system for measurement and control system based on deep learning according to claim 1, characterized in that, The resource status perception module collects real-time operational data from the monitoring and control system through the following steps: Step 11: Collect real-time running data; the real-time running data includes CPU utilization, memory usage, network bandwidth utilization, storage space utilization, and task queue status information; Step 12: Construct the real-time running data into a system state vector through feature engineering; for task queue state information, the feature engineering process includes: task clustering analysis, clustering similar task features; queue state feature extraction, calculating queue depth, average waiting time, and priority distribution entropy statistical features; task flow feature extraction, analyzing dynamic features including task arrival rate, completion rate, and blocking rate; and resource matching degree analysis, evaluating the matching degree between the current task set and available resources.

3. The adaptive adjustment system for measurement and control system based on deep learning according to claim 2, characterized in that, The priority adjustment module performs feature extraction on tasks in the measurement and control system, including the following steps: Step 21: Extract the task features of each task from the task queue state information in the system state vector; Step 22: Use a support vector machine to classify the extracted task features, dividing the tasks into three categories: computationally intensive, I / O intensive, and hybrid. Step 23: For different types of tasks in the above classification, use a linear regression model to predict their resource requirements.

4. The adaptive adjustment system for measurement and control system based on deep learning according to claim 3, characterized in that, The priority adjustment module establishes a dynamic task priority adjustment mechanism in the following way: Extract the waiting time from the task queue status information in the system state vector, and adaptively adjust the task priority based on the waiting time and the system state vector; The adaptive adjustment of task priority includes: The priority adjustment module consists of a resource competition analysis unit, a priority calculation unit, and a priority update unit. The resource competition analysis unit calculates the current system resource competition level based on the system resource state sub-vector and network resource state sub-vector in the system state vector. The system resource competition level is described by the resource competition index, which is calculated by weighting factors such as CPU utilization, memory usage, network bandwidth utilization, and storage space utilization to quantify the system resource tension. The priority calculation unit calculates a dynamic priority adjustment factor for each task based on the task's waiting time, resource contention index, and resource requirement; the dynamic priority adjustment factor is calculated using an adaptive weighting method. The dynamic priority adjustment factor calculation method implemented by the priority calculation unit includes the following sub-steps: Step 221: Construct a waiting time compensation function. As the task waiting time increases, the compensation value increases non-linearly to prevent the task from starving for a long time. Step 222: Construct a resource requirement penalty function to reduce the priority of tasks with high resource requirements, but set an upper limit to prevent high resource requirement tasks from being not executed; Step 223: Construct a system load adaptive function to dynamically adjust the weights of waiting time compensation and resource demand penalty based on the current system load status; Step 224: Weight and fuse the task priority, waiting time compensation value, resource requirement penalty value and system load adaptive factor in the queue status information, and calculate the final dynamic priority adjustment factor.

5. The adaptive adjustment system for measurement and control system based on deep learning according to claim 4, characterized in that, The macro-scheduling layer in the two-layer scheduling decision module employs a genetic algorithm, including the following steps: Step 31: Receive the system state vector and the adjusted task priority information, and design a chromosome encoding method for global resource allocation optimization of the macro scheduling layer to represent the resource allocation scheme; Step 32: Design a fitness function for the global resource allocation optimization of the macro scheduling layer, which is composed of an objective function construction unit, a weight allocation unit, and a normalization processing unit, taking into account both system throughput and load balancing. Designing the corresponding fitness function for the genetic algorithm includes the following sub-steps: Step 321: Based on the chromosome encoding method, construct a system throughput evaluation function composed of a task execution time estimation unit, a parallelism analysis unit, and a throughput calculation unit to calculate the number of tasks completed per unit time. Step 322: Based on the chromosome encoding method, construct a load balance evaluation function consisting of a resource utilization calculation unit, a variance calculation unit, and a load balance scoring unit, and calculate the variance of each resource utilization rate; Step 323: Based on the chromosome encoding method, construct an energy consumption evaluation function composed of a resource energy consumption model unit, a task energy consumption calculation unit, and a system energy consumption optimization unit to calculate the energy consumption of the resource allocation scheme; Step 324: Based on the chromosome encoding method, construct a latency evaluation function composed of a task scheduling sequence generation unit, a critical path analysis unit, and a response time calculation unit, and calculate the average task completion time; Step 33: Introduce an adaptive crossover and mutation operator, consisting of a population diversity assessment unit, a parameter adaptive adjustment unit, and an operation execution unit, to optimize the global resource allocation of the macro scheduling layer, and dynamically adjust the parameters according to the population diversity. Step 34: Design an elite retention strategy for the global resource allocation optimization of the macro scheduling layer, which consists of an elite selection unit, an elite preservation unit, and an elite re-introduction unit.

6. The adaptive adjustment system for measurement and control system based on deep learning according to claim 5, characterized in that, The micro-scheduling layer in the two-layer scheduling decision module uses a reinforcement learning model to make real-time scheduling decisions, including the following steps: Step 35: Receive the system state vector and the updated task queue state information, and define the state space of the micro-scheduling layer; Step 36: Based on the resource allocation scheme of the macro scheduling layer, define the action space of the micro scheduling layer as specific resource allocation decisions; Step 37: Based on the energy consumption evaluation function and the time delay evaluation function, design the reward function of the micro-scheduling layer as a weighted sum of task completion time and energy consumption; Step 38: Based on the defined state space, action space, and reward function, the Q-learning algorithm is used to optimize the reward function.

7. The adaptive adjustment system for measurement and control system based on deep learning according to claim 6, characterized in that, The design of the chromosome encoding method includes the following sub-steps: Step 311: Design a chromosome structure based on integer encoding to represent the mapping relationship between the predicted task resource requirements and system resources; Step 312: Based on the chromosome structure, design a chromosome segmentation strategy to encode allocation schemes for different types of resources; Step 313: Based on the segmentation strategy, design a constraint condition coding method consisting of hard constraint coding units, soft constraint coding units, and constraint priority units, so that the constraint resource allocation meets the system constraints; Step 314: Based on the task priority information, design a priority encoding method consisting of a priority mapping unit, an execution order encoding unit, and an importance encoding unit to reflect the execution order and importance of the task.

8. The adaptive adjustment system for measurement and control system based on deep learning according to claim 7, characterized in that, Introducing the adaptive crossover and mutation operators for global resource allocation optimization at the macro-scheduling layer includes the following sub-steps: Step 331: Based on the fitness function evaluation results, and according to the fitness differences of individuals in the population, construct a fitness difference analysis unit, a crossover probability calculation unit, and an individual selection unit to dynamically adjust the crossover probability together; Step 332: Based on the results of the dynamic adjustment of the crossover probability, according to the number of generations and the convergence speed, construct an evolutionary state monitoring unit, a convergence speed calculation unit, and a mutation probability adjustment unit to dynamically adjust the mutation probability together; Step 333: Based on the chromosome encoding method, design a crossover operation for the resource allocation problem, consisting of a crossover point selection unit, an intra-segment crossover unit, and a validity repair unit; Step 334: Based on the aforementioned crossover operation, design a local search mutation operation for scheduling optimization, consisting of a mutation location selection unit, a neighborhood generation unit, and an optimal replacement unit.

9. The adaptive adjustment system for measurement and control system based on deep learning according to claim 8, characterized in that, The top-down guidance and feedback includes the following steps: Step 41: Receive the resource allocation scheme optimized by the macro scheduling layer, and construct a guidance information receiving unit, a constraint condition conversion unit, and a constraint condition application unit to jointly use the resource allocation scheme as the constraint condition of the micro scheduling layer; Step 42: The macro scheduling layer periodically transmits the evaluated global optimization target and resource usage boundary to the micro scheduling layer.

10. A deep learning-based adaptive adjustment method for a measurement and control system, implemented based on any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Collect real-time operating data from the measurement and control system, and construct a system state vector from the preprocessed real-time operating data; Step 2: Based on the system state vector, extract features from the tasks in the measurement and control system, classify the tasks using a support vector machine, predict the resource requirements for different types of tasks, dynamically adjust the task priorities based on the resource requirements, and generate task priority information. Step 3: Receive the system state vector and the adjusted task priority information through a pre-constructed macro-scheduling layer, and optimize global resource allocation using a genetic algorithm; based on the optimization results of the macro-scheduling layer, use a reinforcement learning model to make real-time scheduling decisions through a pre-constructed micro-scheduling layer; and optimize the macro-scheduling layer and micro-scheduling layer using top-down guidance feedback and bottom-up correction feedback. Step 4: Based on the resource allocation scheme optimized by the macro scheduling layer, generate specific resource allocation instructions.

Images