Emergency command fusion communication system resource dynamic scheduling method based on artificial intelligence

Abstract

The application provides an emergency command fusion communication system resource dynamic scheduling method based on artificial intelligence, which comprises the following steps: S1, acquiring a plurality of sensors and log data of an emergency command system; constructing a dynamic graph neural network to optimize a communication topology, and selecting an optimal communication path; combining resource prediction and the optimized communication topology to formulate a globally optimal resource scheduling strategy, and generating an optimal resource allocation scheme for each task; the resource pool comprises available computing resources and communication bandwidth; dynamically adjusting resource allocation and communication paths during task execution, generating task execution feedback corresponding to the tasks; monitoring the system state in real time, and optimizing resource prediction, topology adjustment and scheduling strategy based on the execution error feedback of the task execution feedback. The application realizes efficient and intelligent dynamic resource scheduling in the emergency command fusion communication system through self-supervised learning resource prediction, adaptive topology optimization of a variable topology graph neural network and global scheduling driven by reinforcement learning.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of emergency command, and particularly relates to a resource dynamic scheduling method for an emergency command fusion communication system based on artificial intelligence. BACKGROUND

[0002] In a modern emergency command system, a fusion communication system plays a crucial role, which can integrate various communication means (satellite communication, shortwave radio, private network, public network, etc.), realize cross-regional and cross-departmental information sharing and unified scheduling. However, in a sudden event (such as earthquake, flood, terrorist attack, military exercise, etc.), the traditional fusion communication system still has many technical bottlenecks in resource scheduling, mainly in the following aspects:

[0003] Firstly, resource demand is difficult to predict in advance, leading to scheduling lag or resource waste. The occurrence of emergency events is often accompanied by dramatic changes in communication demand, and the traditional static configuration or rule-driven resource scheduling method is difficult to adapt to complex emergencies. For example, in a sudden natural disaster, the communication demand in the affected area increases dramatically, and the scheduling center often has difficulty in timely sensing these demands, leading to lag in communication resource allocation, thereby affecting the efficiency of rescue. In addition, the current mainstream resource scheduling method mostly adopts an experience rule-based method, such as presetting a fixed communication topology or allocating resources according to a predefined scheduling priority. Although this method can play a certain role in normal circumstances, when the scale, type and influence range of the event are unpredictable, the preset rules often cannot adapt to new demands, easily leading to resource scheduling imbalance, and some areas have communication resource shortage, while other areas have idle resources.

[0004] Secondly, the traditional communication topology structure is rigid and cannot adapt to dynamic changes in emergencies. The physical and logical topology structure of the fusion communication system may change dramatically in emergencies, for example, some communication nodes may fail due to damage or overload, and some communication links may be unavailable due to interference, while the traditional network usually adopts a fixed topology or relies only on static backup links, and it is difficult to flexibly adjust the network structure in emergencies. For example, in a battlefield or disaster site, some communication base stations may be damaged, and the existing system is difficult to quickly reconfigure the optimal communication path, leading to the loss of contact of some rescue units or command nodes, seriously affecting the effectiveness of scheduling and command. Even though some systems support a certain degree of link switching and load balancing, most still rely on static preset rules, lack intelligent optimization capability, and the communication link adjustment process is slow, affecting real-time performance.

[0005] Furthermore, the resource scheduling lacks global optimization capability, and it is difficult to balance communication stability and resource efficient utilization. The traditional integrated communication system usually allocates resources according to a single dimension (such as bandwidth, computing resources), and fails to comprehensively consider various factors such as the priority, bandwidth demand, and computing load of different tasks, resulting in poor global optimality of the scheduling decision. For example, in the same emergency task, some units may need high bandwidth for video transmission, while other units mainly perform low-bandwidth voice scheduling, and the traditional scheduling system is difficult to intelligently adjust resource allocation, resulting in that high-priority tasks cannot be fully supported, while low-priority tasks occupy too many resources. In addition, with the wide application of edge computing and cloud computing, the complexity of resource scheduling is further increased, and the existing scheduling system is difficult to realize intelligent collaboration of edge computing nodes and cloud resources, resulting in that computing resources cannot be efficiently utilized in emergency situations.

[0006] In summary, the current integrated communication system faces core problems such as insufficient resource demand prediction capability, difficulty in self-adaptive adjustment of communication topology rigidity, and lack of global optimization of resource scheduling in emergency command scheduling, which seriously affects the response speed and stability of the emergency command system. Therefore, an intelligent resource demand prediction, self-adaptive communication topology adjustment, and globally optimized dynamic scheduling method is urgently needed to improve the reliability and scheduling efficiency of the emergency command system. SUMMARY

[0007] The purpose of the present application is to propose an artificial intelligence-based emergency command integrated communication system resource dynamic scheduling method, which realizes efficient, stable, and intelligent dynamic resource scheduling in the emergency command integrated communication system through self-supervised learning resource prediction, adaptive topology optimization of variable topology graph neural network, and reinforcement learning-driven global scheduling.

[0008] In order to achieve the above purpose, the present application provides an artificial intelligence-based emergency command integrated communication system resource dynamic scheduling method, which comprises the following steps:

[0009] S1, obtaining a plurality of sensors and log data of the emergency command system, adopting task similarity-based contrast learning to construct a label-free learning task, combining time series Transformer to automatically learn the resource consumption pattern in the historical data and generate resource prediction for the future time period;

[0010] S2, constructing a dynamic graph neural network to optimize the communication topology, dynamically and adaptively adjusting the communication topology according to the resource prediction combined with reinforcement learning, and selecting the optimal communication path;

[0011] S3, obtaining a resource pool, combining the resource prediction and the optimized communication topology to formulate a globally optimal resource scheduling strategy, and generating an optimal resource allocation scheme for each task; the resource pool includes available computing resources and communication bandwidth;

[0012] S4, according to the node state information of the emergency command system for each task Optimal resource allocation scheme and optimized communication topology Intelligent matching calculation task and computing node, and dynamically adjusting resource allocation and communication path in the task execution process, generating task execution feedback corresponding to the task; The node state information includes: maximum computing capacity, current load, maximum bandwidth and current bandwidth occupation;

[0013] S5, real-time monitoring system state, and based on the execution error feedback of task execution feedback, optimizing resource prediction, topology adjustment and scheduling strategy.

[0014] Preferably, the data sources of the plurality of sensors and log data include: network monitoring equipment, computing node management system and equipment health monitoring system;

[0015] S1, further comprising: using a sliding window method, the data of continuous Time step is divided into pieces to form a history sequence; Normalize the data of different dimensions to learn on the same scale.

[0016] Preferably, the task similarity-based contrast learning constructs a no-label learning task, including:

[0017] Using a transformer model to calculate the historical resource usage pattern of the task, generating a task feature vector to represent the characteristics of the task, and extracting the key resource usage pattern of the task

[0018] Based on normalized cross-entropy analysis, the feature vectors of similar tasks are closer, and the feature vectors of different tasks are far away;

[0019] Wherein, the construction of no-label learning task combines time series Transformer, automatically learns the resource consumption pattern in the historical data and generates resource prediction in the future time period, including:

[0020] Based on historical data to predict future resource demand, using time series Transformer for modeling, taking the task feature vector as input, and outputting the bandwidth demand prediction value, computing resource and equipment health status in the future time step; The time series Transformer uses weighted mean square error to calculate the prediction error, so that the prediction error of the key resource is more important.

[0021] Preferably, the S2 specifically includes:

[0022] Building a dynamic communication graph , initialize node state, wherein the dynamic communication graph Including: node set Representing communication equipment; edge set Represents the communication link between nodes; weight matrix Record link quality;

[0023] Based on dynamic communication graph The system employs a dynamic graph neural network to model the communication topology and iteratively propagates link information, enabling each node to perceive the global state. Through multiple rounds of iterative calculations, each node dynamically adjusts its own connections based on the state of its neighboring nodes, calculates the retention probability of each link, decides whether to retain or delete the link, and establishes new links between nodes with high demand, generating an optimized communication topology.

[0024] On the optimized communication topology, the path with the lowest latency is selected based on the bandwidth requirements of the task, and the links in the path are ensured to have sufficient bandwidth margin, thereby generating the path cost of the communication topology.

[0025] Preferably, the calculation of the retention probability of each link to determine whether to retain or delete the link specifically involves:

[0026] ;

[0027] when Delete the link when necessary;

[0028] when Add links between nodes with high demand to form a new network topology. ;

[0029] in, The probability of retention or rejection for each link. For the edge The weight, For the edge The weight, For the edge and , To set a threshold;

[0030] The edge weight The calculation is as follows:

[0031] ;

[0032] in, For the physical distance of the link, For link reliability, For link latency; Based on predicted bandwidth requirements, the topology is guided to optimize towards nodes with high demand; This is a hyperparameter used to balance various factors.

[0033] Preferably, the resource scheduling strategy meets the task priority; the task priority is obtained by weighted linear analysis based on the task urgency, the bandwidth demand proportion of the task, the influence of the task on the scheduling priority in the case of resource shortage, and the adjacent link quality of the communication topology where the task is located.

[0034] Preferably, the resource scheduling strategy is optimized in combination with the task priority, with the constraint condition of available computing resources and communication bandwidth not exceeding the maximum capacity and the target of maximizing resource utilization; the scheduling loss function of the resource scheduling strategy is:

[0035] ;

[0036] wherein, and represent the matching degree of the allocated resources and the demand; The bandwidth imbalance of adjacent tasks is constrained to ensure that the link will not be overloaded due to a single task. is a regularization coefficient for controlling the trade-off between resource balance and priority.

[0037] Preferably, the S4 comprises:

[0038] Based on the required computing resources and bandwidth of the task, the maximum computing resources and bandwidth capacity of the computing node, the current computing resources and bandwidth occupation of the node, and the link score associated with the task, the adaptation degree of the task executing on the computing node is calculated; wherein the link score associated with the task is the weight of the corresponding edge.

[0039] The computing node with the highest adaptation degree is selected for task execution.

[0040] Meanwhile, during the task execution process, the load of each computing node changes over time, and if an overload condition occurs, the task allocation strategy needs to be adjusted; if the communication link on which the task depends is congested, the communication path of the task needs to be dynamically adjusted.

[0041] After each task is executed, the actual consumption value of the computing resources and the communication bandwidth is recorded and compared with the predicted value to calculate the error; if the error exceeds the set threshold, the resource prediction value of step S1 is returned to improve the accuracy of the next round of task execution; if the error continues to exceed the standard, the resource scheduling strategy of step S3 is adjusted to improve the adaptation degree calculation of task execution and optimize the matching rule of computing task and computing node.

[0042] Preferably, the path selection and switching rule for adjusting the task allocation strategy if an overload condition occurs is:

[0043] The maximum load threshold of the computing node is set If a computing node If the current computing load of the node exceeds the threshold, the task allocation needs to be adjusted:

[0044] ;

[0045] wherein, represents the task set currently allocated to the node ; is the computing resource required by the task ;

[0046] If the load exceeds the threshold, part of the task is preferentially migrated to the node with rich computing resources , or the computing priority of part of the task is reduced to make it wait for execution.

[0047] If the communication link on which the task depends is congested, the path selection and switching rules of the communication path of the task need to be dynamically adjusted as follows:

[0048] If the bandwidth of the new path is sufficient, the switching is directly performed.

[0049] If the bandwidth is insufficient, the bandwidth occupation of the low-priority task is reduced to make the high-priority task be preferentially executed.

[0050] Preferably, the execution error is a bandwidth allocation error and a computing resource allocation error, and if the bandwidth allocation error or the computing resource allocation error exceeds a set threshold, it indicates that the resource prediction is inaccurate, and the prediction parameters of the resource prediction need to be optimized.

[0051] The S5 further comprises:

[0052] If the bandwidth utilization of the link on which the task depends exceeds a preset threshold, the weight of the link is reduced, a new optimal path is tried to find, and the communication topology is adjusted.

[0053] The computing task is adapted to the new path . If , the task is migrated.

[0054] If the adaptation degree is lower than a threshold, the bandwidth occupation of the task is adjusted to reduce the influence on the network resource

[0055] The present application has at least the following beneficial technical effects:

[0056] First, to address the problem of difficult early prediction of resource demand, the present application uses self-supervised learning (Self-Supervised Learning, SSL), combined with time series Transformer and contrastive learning (Contrastive Learning), to automatically learn the resource demand patterns of different tasks from historical emergency event data, without the need for large amounts of manually labeled data, to accurately predict future bandwidth, computing resources and device usage demand. Through this innovation, the system can perceive the impending resource bottleneck in advance and provide data support for subsequent scheduling, thereby avoiding the problem of resource waste or communication interruption caused by scheduling lag.

[0057] Second, to address the problem of rigid traditional communication topology that cannot adapt to sudden changes, the present application uses a dynamic topology graph neural network (Dynamic Topology Graph Neural Network, DT-GNN) to construct a dynamic graph model, analyze the communication network structure in real time, and combine with graph reinforcement learning (Graph Reinforcement Learning, Graph RL) to autonomously learn the optimal link adjustment strategy. When some communication nodes are blocked due to damage or overload, DT-GNN can adaptively adjust the communication topology according to the resource demand prediction results, automatically select the optimal communication path, and ensure uninterrupted command and dispatch. In addition, by combining the collaborative optimization of edge computing and cloud computing, the present application can quickly perform topology adjustment under low latency conditions, improving the real-time response capability of the system.

[0058] Finally, to address the problem of lack of global optimization in resource scheduling, the present application uses reinforcement learning combined with multi-objective optimization method on the basis of resource prediction and topology optimization, considering factors such as task priority, resource consumption, computing and communication load, to intelligently adjust the computing resource and bandwidth allocation strategy. Through this mechanism, the present application not only improves the utilization rate of resources, but also ensures that the resource demand of different tasks is reasonably met, avoiding the situation of resource shortage for critical tasks and resource waste for low-priority tasks. BRIEF DESCRIPTION OF DRAWINGS

[0059] The present application is further illustrated by the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present application. For those skilled in the art, other drawings can be obtained without creative labor on the basis of the following drawings.

[0060] Figure 1 The present application discloses a resource dynamic scheduling method flowchart of an emergency command integrated communication system based on artificial intelligence. DETAILED DESCRIPTION

[0061] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0062] like Figure 1 As shown in the figure, the method for dynamic resource scheduling of an emergency command converged communication system based on artificial intelligence provided in this embodiment of the invention includes the following steps:

[0063] S1. Acquire multiple sensor and log data from the emergency command system, construct an unlabeled learning task using comparative learning based on task similarity, and combine it with a time series Transformer to automatically learn resource consumption patterns in historical data and generate resource predictions for future time periods.

[0064] Specifically, the data sources are: multiple sensors and log data from the emergency command system, including:

[0065] Network monitoring equipment: collects bandwidth usage information. , indicating task Network resource usage at different times.

[0066] Compute Node Management System: Records computing resource usage. Reflecting the task The computational cost at runtime.

[0067] Equipment health monitoring system: provides equipment status data Such as CPU load, storage usage, and channel interference.

[0068] Further, data preprocessing: A sliding window method is used to process continuous data. Data at each time step is segmented to form a historical sequence. This ensures the model can learn time dependencies. Normalizing data of different dimensions allows it to learn on the same scale, improving model convergence speed.

[0069] Furthermore, the self-supervised learning objective is constructed by employing contrastive learning based on task similarity to construct an unlabeled learning task.

[0070] Computational tasks Historical resource usage patterns This represents the characteristics of the task and extracts the key resource usage patterns of the task:

[0071] ;

[0072] in, is a parametric transformer model that can automatically learn the resource consumption patterns of tasks.

[0073] The feature vectors of similar tasks are made closer and the feature vectors of different tasks are made further apart:

[0074] ;

[0075] where, represents the cosine similarity, and is the temperature coefficient.

[0076] Further, a resource demand prediction model is constructed: based on historical data to predict future resource demand, using time series Transformer for modeling.

[0077] Input: task feature vector , extract trends from the historical resource usage patterns of tasks.

[0078] Output: future bandwidth demand prediction value ;

[0079] Computational resources and device health status ;

[0080] The weighted mean square error is used to calculate the prediction error, ensuring that the prediction error of key resources has a greater impact:

[0081]

[0082] where, are the loss weights of bandwidth, computational resources and device health status, respectively.

[0083] Further, the optimization goal: comprehensive consideration of task similarity (self-supervised learning) and prediction error (supervised learning) for joint training, so that the model can learn from unlabeled data and accurately predict future resource demand.

[0084] Final output: resource demand prediction for future time period .

[0085] S2, construct a dynamic graph neural network to optimize the communication topology, dynamically and adaptively adjust the communication topology according to the resource prediction combined with reinforcement learning, and select the optimal communication path.

[0086] Specifically, in the emergency command fusion communication system, sudden events may cause partial communication node failure, link congestion, and affect the stability of command and dispatch. To ensure that the communication network can dynamically adapt to changes in resource demand, this step is based on the resource prediction results of step 1 , optimizes the communication topology to adapt to future bandwidth, computing resource demand, and device health status. This step uses a dynamic graph neural network (DT-GNN) to optimize the communication topology, adjusts the link weight according to the task demand, and dynamically reconstructs the network structure to ensure the stability and efficiency of the communication link. Finally, the optimized communication topology is output for subsequent resource scheduling.

[0087] Further, the communication network topology is constructed and the node state is initialized:

[0088] Input data: use the resource prediction of step 1 , combined with the current communication network structure for dynamic adjustment.

[0089] Topology definition: construct a dynamic communication graph , where:

[0090] Node set represents command centers, base stations, terminals, and other communication devices.

[0091] Edge set represents the communication link between nodes.

[0092] Weight matrix records the link quality, including bandwidth, latency, packet loss rate, etc.

[0093] Initialize node state: each node has an initial state , including bandwidth demand , computing resources , and device health status , to ensure that topology adjustment can adapt to changes in resource demand. Set the weight of the edge , the calculation method is as follows:

[0094] ;

[0095] where, is the physical distance of the link, is the link reliability, is the link latency; Combine the predicted bandwidth demand of step 1 to guide the topology to optimize towards high demand nodes; is a hyperparameter that balances various factors.

[0096] Further, information propagation and update: adopt dynamic graph neural network (DT-GNN) to model communication topology, iteratively propagate link information, and make each node perceive the global state.

[0097] Topology adjustment calculation: through multiple rounds of iteration calculation, each node dynamically adjusts its connection according to the state of neighbor nodes, ensures that high-reliability links are preferentially retained, and low-quality links are dynamically removed.

[0098] Calculate the keep or delete probability of each link to decide whether to retain or delete the link, and establish new links between high-demand nodes:

[0099] ;

[0100] When (set threshold) is deleted;

[0101] When , add links between high-demand nodes to form a new network topology .

[0102] Further, calculate the optimal communication path: on the optimized topology , select the lowest latency path based on the bandwidth demand of the task , and ensure that the links in the path have sufficient bandwidth margin.

[0103] Path cost calculation: define the optimization goal, minimize the path delay, and increase the bandwidth margin, the formula is as follows:

[0104] ;

[0105] Where, is the maximum bandwidth capacity of the link, as an adjustment factor, to ensure that high-demand tasks have low-latency paths.

[0106] Output the optimized topology information: output the new communication topology for resource scheduling in step 3, to ensure efficient matching of computing resources and communication resources; output the score of each communication link for subsequent scheduling process to preferentially select high-quality communication links, and improve the reliability of the emergency command system.

[0107] S3, obtain the resource pool, combine resource prediction and optimized communication topology to develop a globally optimal resource scheduling strategy, and generate an optimal resource allocation scheme for each task; the resource pool includes available computing resources and communication bandwidth.

[0108] Specifically, in an emergency command converged communication system, the goal of resource scheduling is to ensure the rational allocation of computing resources and communication bandwidth to meet the dynamic needs of tasks under emergencies. Since emergencies may cause some communication links to become unstable, computing nodes to experience excessive load, or resource imbalances, relying solely on traditional static scheduling schemes cannot guarantee the system's efficiency and stability. This step is based on the communication topology optimized in step 2. and link weight Combined with the resource demand forecast in step 1 A globally optimal resource scheduling strategy should be formulated. This strategy needs to consider the dynamic topology changes of communication links, task priorities, and the global optimality of resource allocation to ensure that the command system can still operate efficiently in case of emergencies.

[0109] Construct a resource scheduling optimization model:

[0110] Input data: The input for this step includes:

[0111] Prediction results of step 1 It is used to guide demand-side decisions in resource allocation.

[0112] Step 2 Optimization Topology and its link weight It is used to assess the availability of computing resources and the stability of the link.

[0113] Task resource status representation: Define the task demand vector ,in: For the task Required bandwidth resources; For computing resource requirements; It represents the health status of the equipment and determines whether a task should be prioritized for scheduling.

[0114] Resource pool definition: Defines available computing resources and communication bandwidth This is to ensure that the allocation scheme does not exceed the system's capacity.

[0115] Furthermore, task priorities are calculated: since task urgency, device health status, and bandwidth requirements have different impacts on scheduling decisions, a task priority score is defined. As a key basis for resource allocation:

[0116] ;

[0117] in, The urgency level of the mission is provided by the command system. This reflects the bandwidth requirements of each task, ensuring that high-demand tasks are allocated resources first. Impact the scheduling priority of tasks in resource-constrained situations. Additional items Compute tasks The quality of adjacent links in the communication topology, ensuring that tasks prefer high-stability paths. Weight factor, control the impact of different factors.

[0118] Further, the goal is to maximize the utilization efficiency of bandwidth and computing resources while ensuring task priority order, avoiding resource waste or overload.

[0119] Optimization strategy: based on task priority , calculate the optimal resource allocation scheme :

[0120] Set constraints to ensure that the total amount of resource allocation does not exceed system capacity:

[0121] ;

[0122] Adopt a dynamic adjustment mechanism based on link weight If the reliability of the communication link on which the task mainly depends decreases, reduce the bandwidth allocation of this task and preferentially allocate it to other path stable tasks.

[0123] Calculate the scheduling loss function: maximize resource utilization rate, and optimize in combination with task priority:

[0124] ;

[0125] where, and represent the matching degree of allocated resources and demand. Restrict the bandwidth imbalance of adjacent tasks to ensure that the link will not be overloaded due to a single task. Regularization coefficient, control the trade-off between resource balance and priority.

[0126] Final output: generate the optimal resource allocation scheme for each task and pass the results to the next step to ensure that tasks can be efficiently executed based on allocated resources.

[0127] At the same time, provide resource usage state feedback to ensure that system load balancing can be adjusted in subsequent steps to optimize resource utilization efficiency.

[0128] Adjust the scheduling strategy of computing resources. In the case of computing node overload, automatically allocate computing tasks to low-load computing units to ensure the running stability of the entire system. ​

[0129] S4. Based on the node status information of the emergency command system, intelligently match the computing task with the computing node according to the optimal resource allocation scheme and optimized communication topology for each task, and dynamically adjust the resource allocation and communication path during the task execution process to generate corresponding task execution feedback; the node status information includes: maximum computing power, current load, maximum bandwidth and current bandwidth usage.

[0130] Specifically, in an emergency command converged communication system, task execution not only relies on reasonable resource scheduling (the output of step 3), but also requires ensuring a high degree of matching between the computational tasks and the communication network to guarantee stable and efficient system operation during execution. The goal of this step is to optimize the resource scheduling scheme based on the results of step 3. Combined with the optimized communication topology from step 2 It intelligently matches computing tasks with computing nodes and dynamically adjusts resource allocation and communication paths during task execution to cope with possible computing load overload or network status fluctuations.

[0131] Get input data:

[0132] Resource scheduling results in step 3 That is, the task The optimal bandwidth and computing resource allocation scheme.

[0133] Step 2 Optimization Topology This is used to select the execution node for computing tasks and ensure the stability of the communication path.

[0134] Compute node status information: Compute node set Includes each computing node computing power Current load Maximum bandwidth and current bandwidth usage .

[0135] Furthermore, the task execution node selection:

[0136] Computational tasks At the computing node Adaptability of execution Considering overall computing resources, bandwidth, and communication link quality:

[0137] ;

[0138] in, and Tasks Required computing resources and bandwidth. and They are computing nodes Maximum computing resources and bandwidth capacity. and Representing nodes respectively Based on the current computing resources and bandwidth usage, ensure that computing nodes with lower loads are selected for task scheduling. For the task Link scoring is used to ensure the selection of a stable communication path. The adjustment coefficient balances computing power, bandwidth adaptability, and communication stability.

[0139] Task Select the computing node with the highest compatibility. To execute, that is:

[0140] ;

[0141] If the fit of all computing nodes is below the threshold This indicates that the current system's computing resources are insufficient, so it is necessary to adjust the task execution priority and delay the execution of low-priority tasks to ensure that high-priority tasks can be scheduled.

[0142] Furthermore, compute node load monitoring: During task execution, each compute node... The load varies over time, and if overload occurs, the task allocation strategy needs to be adjusted.

[0143] Task load adjustment rules: Set the maximum load threshold for compute nodes. If a certain computing node If the current computational load exceeds this threshold, task allocation needs to be adjusted:

[0144] ;

[0145] in Indicates the current allocation to the node. A set of tasks.

[0146] If the load exceeds the limit, prioritize migrating some tasks to nodes with abundant computing resources. Alternatively, the computational priority of some tasks can be reduced, causing them to wait for execution.

[0147] Furthermore, communication path dynamic optimization: during task execution, if the task... The dependent communication link is congested (bandwidth utilization exceeds the limit). If so, the communication path of the task needs to be dynamically adjusted.

[0148] Path selection and switching rules: In optimizing the topology Internally, searching for new low-latency paths , and re-allocate bandwidth to ensure uninterrupted task execution.

[0149] Bandwidth adjustment strategy: if the link bandwidth on the new path is sufficient, switch directly. If the bandwidth is insufficient, reduce the bandwidth occupation of low-priority tasks to give priority to high-priority tasks.

[0150] Further, task completion state recording: after each task execution, record the actual consumption values of computing resources and communication bandwidth , and compare them with the predicted values to calculate the error:

[0151] ;

[0152] If the error exceeds the set threshold , adjust the resource prediction model in step 1 to improve the accuracy of the next round of task execution.

[0153] Error feedback mechanism: if the error continues to exceed the standard, adjust the resource scheduling strategy in step 3 to improve the adaptation calculation of task execution and optimize the matching rules of computing tasks and computing nodes.

[0154] This step combines computing resource matching, dynamic load balancing, communication link switching, and error feedback mechanism during task execution to realize an adaptive task execution optimization scheme. Through intelligent task scheduling and resource adaptation, this step ensures that the emergency command integrated communication system can operate stably and efficiently in complex environments, improving the reliability of overall task execution.

[0155] S5, real-time monitoring of system state, and based on task execution feedback, optimizing resource prediction, topology adjustment, and scheduling strategy.

[0156] Specifically, in the emergency command integrated communication system, during task execution, changes in the environment, fluctuations in computing resource load, or congestion in communication links may cause resource scheduling imbalance or task execution failure. Therefore, it is necessary to real-time monitor the system state and optimize resource prediction, topology adjustment, and scheduling strategy based on execution error feedback. This step compares the actual state of task execution in step 4 with the resource allocation scheme in step 3 , calculates the error, and optimizes the resource prediction model through online adjustment to improve the dynamic adaptability of the system.

[0157] Further, obtain input data:

[0158] Task execution feedback in step 4 , i.e., the actual consumption of bandwidth and computing resources by tasks .​

[0159] Resource scheduling result of step 3 , as the allocation benchmark before task execution.

[0160] Optimized topology of step 2 , for evaluating the communication link state.

[0161] Error calculation: calculate bandwidth and computing resource allocation error:

[0162] ;

[0163] If or exceeds the set threshold , indicating that the resource prediction is inaccurate, and the resource prediction model needs to be optimized.

[0164] Link quality change evaluation: analyze the task dependent link bandwidth utilization , if it exceeds the set threshold , the communication topology or resource scheduling strategy needs to be adjusted.

[0165] Further, the optimization goal: improve the resource demand prediction model of step 1, adapt to the real-time changing system load, and reduce the resource allocation error.

[0166] Update prediction model parameters: define error correction terms , used to adjust the resource prediction model parameters to reduce the prediction deviation of future tasks:

[0167] ;

[0168] Where, is the learning rate, controlling the update step size. represents the gradient of the resource prediction model to the input data .

[0169] Dynamic adjustment of prediction weights: for high-error tasks, increase the weight of historical data, so that it has a greater impact on model adjustment, and improve prediction accuracy.

[0170] Further, link bandwidth utilization monitoring: if the task dependent link bandwidth utilization exceeds , reduce the weight of the link, and try to find a new optimal path.

[0171] Further, topology adjustment strategy: calculate the task in the new path adaptability on the task , if , migrate the task. If the adaptability is lower than the threshold, adjust the bandwidth occupation of the task , reduce the impact on network resources.

[0172] Further, dynamic adjustment of task priority: based on historical execution error, adjust the task scheduling priority, ensure that high error tasks have priority to obtain resources to reduce future scheduling deviation.

[0173] Load balancing optimization: if the load of a certain computing node exceeds , adjust the task to a node with more abundant computing resources. If the overall computing resources of the system are tight, reduce the allocation of computing resources to low-priority tasks, and make critical tasks a priority.

[0174] This step realizes resource prediction, communication topology adjustment and scheduling strategy optimization through real-time monitoring and error feedback optimization, ensures that the emergency command system can dynamically adapt to environmental changes, and improves the stability and resource utilization efficiency of task execution.

[0175] In summary, the present application realizes efficient, stable and intelligent dynamic resource scheduling in the emergency command integrated communication system through self-supervised learning resource prediction, adaptive topology optimization of variable topology graph neural network and global scheduling driven by reinforcement learning. Compared with traditional methods, the present application reduces scheduling lag, improves communication stability, optimizes resource allocation efficiency, and can significantly improve the reliability of the emergency command system and the ability to respond to emergencies.

[0176] The above describes certain embodiments of the specification, and other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily have to be implemented in the specific order shown or in a continuous sequence to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or can be advantageous.

[0177] The system, device, module or unit illustrated in the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions. A typical implementation device is a computer. Specifically, the computer may, for example, be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0178] For the sake of presentation, the detailed description will often refer to a numerical example and to an operational mathematics. However, the examples set forth in the detailed description are not intended to limit the application to a single preferred embodiment or single preferred method. Rather, as explained above, various

[0179] Those skilled in the art will appreciate that the embodiments of the present description can be further provided as methods, systems, or computer program products. Accordingly, embodiments of the present description can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, and the like) embodying computer readable program code.

[0180] The present description is described with reference to flowcharts and / or block diagrams illustrating the architecture, functionality, and operation of embodiments of methods, apparatuses (systems), and computer program products according to the present description. It will be understood that each block of the flowchart and / or block diagrams and combinations of blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0181] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0182] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0183] In one typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0184] Memory can include non-persistent memory, Random Access Memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory, in computer readable media. Memory is an example of computer readable media.

[0185] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0186] It is also important to note that the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the named element.

[0187] The specification can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, that perform particular tasks or implement particular abstract data types. The specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices.

[0188] The various embodiments in the specification are described in a progressive manner, and the same or similar parts between the various embodiments can be mutually referred to, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0189] Finally, it should be noted that the lithium battery core chip equalization control platform disclosed in the embodiments of the present application is only the preferred embodiment of the present application, and is used to illustrate the technical solutions of the present application, but not to limit it. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions described in the foregoing embodiments can be modified or some technical features can be replaced by equivalents. The modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A resource dynamic scheduling method for an emergency command fusion communication system based on artificial intelligence, characterized in that, The method comprises the following steps: S1, acquiring a plurality of sensors and log data of an emergency command system, adopting task similarity-based contrast learning to construct a no-label learning task, combining a time series Transformer to automatically learn resource consumption patterns in historical data and generate resource predictions for future time periods; S2, constructing a dynamic graph neural network to optimize the communication topology, dynamically and adaptively adjusting the communication topology according to the resource predictions combined with reinforcement learning, and selecting the optimal communication path; S3, acquiring a resource pool, combining the resource predictions and the optimized communication topology to formulate a globally optimal resource scheduling strategy, and generating an optimal resource allocation scheme for each task; the resource pool includes available computing resources and communication bandwidth; S4, intelligently matching the computing tasks and computing nodes for the optimal resource allocation scheme of each task and the optimized communication topology according to the node state information of the emergency command system, and dynamically adjusting the resource allocation and the communication path during task execution to generate task execution feedback corresponding to the tasks; the node state information includes maximum computing capacity, current load, maximum bandwidth, and current bandwidth occupancy; S5, real-time monitoring of system state, and feedback optimization of resource prediction, topology adjustment, and scheduling strategy based on the execution error feedback of task execution feedback.

2. The method of claim 1, wherein the method further comprises: The data sources of the plurality of sensors and log data include network monitoring devices, computing node management systems, and device health monitoring systems; The S1 further comprises: adopting a sliding window method to fragment data of continuous time steps to form a history sequence; and normalizing data of different dimensions to enable learning on the same scale.

3. The method of claim 1, wherein the method further comprises: The task similarity-based contrast learning to construct a no-label learning task comprises: Using a transformer model to calculate the historical resource usage patterns of tasks, generating task feature vectors to represent the characteristics of tasks, and extracting the key resource usage patterns of tasks; Based on normalized cross-entropy analysis, the feature vectors of similar tasks are closer, and the feature vectors of different tasks are farther apart; The construction of the no-label learning task combined with the time series Transformer to automatically learn the resource consumption patterns in the historical data and generate resource predictions for future time periods comprises: Based on historical data, predict future resource demand, use time series Transformer for modeling, use task feature vectors as input, and output bandwidth demand prediction value, computing resource and device health status at future time steps; the time series Transformer uses weighted mean square error to calculate prediction error to ensure that the prediction error of key resources has a greater impact.

4. The method of claim 1, wherein the method further comprises: The S2 specifically comprises: Constructing dynamic communication graphs , initializing node states, wherein the dynamic communication graph comprises: a set of nodes representing communication devices; a set of edges representing communication links between the nodes; and a weight matrix recording link qualities; Dynamic communication graph based The dynamic graph neural network is used for modeling the communication topology, and the link information is iteratively propagated to enable each node to perceive the global state. Through multiple rounds of iterative calculation, each node dynamically adjusts its connection according to the state of the neighbor nodes, and calculates the keep or delete probability of each link to determine whether to keep or delete the link, and a new link is established between the nodes with high demand to generate an optimized communication topology. On the optimized communication topology, the lowest latency path is selected based on the bandwidth demand of the task, and it is ensured that the links in the path have sufficient bandwidth margin to generate the path cost of the communication topology.

5. The method of claim 4, wherein the method further comprises: The calculation of the keep-or-delete probability of each link determines whether to keep or delete the link, specifically: ； When the link is deleted; When Adding links between high demand nodes, forming a new network topology ; wherein, is the stay probability for each link, is the weight of edge , is the weight of edge , are the first and nodes, respectively, is a set threshold; The edge of the weight is calculated as: ； wherein, is a link physical distance, is a link reliability, is a link latency; combined with the predicted bandwidth demand, guide the topology to optimize towards high demand nodes; is a hyper parameter, balancing the factors; is a bandwidth demand prediction value, is a bandwidth demand prediction value.

6. The method of claim 1, wherein the method further comprises: The resource scheduling strategy satisfies the task priority; the task priority is obtained by weighted linear analysis based on the task urgency, the bandwidth demand proportion of the task, the scheduling priority of the task in the resource shortage situation, and the adjacent link quality of the current task in the communication topology.

7. The method of claim 5, wherein the method further comprises: The S4 comprises: The adaptation degree of the task to be executed on the computing node is calculated based on the required computing resources and bandwidth of the task, the maximum computing resource and bandwidth capacity of the computing node, the current computing resource and bandwidth occupation of the node, and the link score associated with the task, wherein the link score associated with the task is the weight of the corresponding edge; The computing node with the highest adaptation degree of task selection is executed; Meanwhile, during the task execution process, the load of each computing node changes over time, and if an overload situation occurs, the task allocation strategy needs to be adjusted; if the communication link on which the task depends is congested, the communication path of the task needs to be dynamically adjusted; After each task is executed, the actual consumption value of the computing resource and the communication bandwidth is recorded, compared with the predicted value, the error is calculated, and if the error exceeds the set threshold, the resource prediction value of the adjustment step S1 is returned to improve the accuracy of the next round of task execution; if the error continues to exceed the standard, the resource scheduling strategy of the step S3 is adjusted to improve the adaptation degree calculation of the task execution and optimize the matching rule of the computing task and the computing node.

8. The method of claim 7, wherein the method further comprises: The path selection and switching rule for the case where the overload situation occurs is that: Setting a maximum load threshold for a computing node If the current computing load of a computing node exceeds the threshold, task allocation needs to be adjusted; If the load exceeds the standard, part of the task is migrated to the node with rich computing resources in priority Or reduce the computing priority of part of the task to wait for execution The path selection and switching rule for the case where the communication link on which the task depends is congested is that: If the link bandwidth on the new path is sufficient, the path is directly switched; If the bandwidth is insufficient, the bandwidth occupation of the low-priority task is reduced to make the high-priority task execute preferentially.

9. The method of claim 1, wherein the method further comprises: The execution error is the bandwidth allocation error and the computing resource allocation error, and if the bandwidth allocation error or the computing resource allocation error exceeds the set threshold, it indicates that the resource prediction is inaccurate, and the prediction parameters of the resource prediction need to be optimized; The S5 further comprises: If the bandwidth utilization rate of the link on which the task depends exceeds the preset threshold, the weight of the link is reduced, a new optimal path is tried to find, and the communication topology is adjusted: Fitness of the computing task T on the new path , if , then migrate the task; if the fitness is below a threshold, then adjust the bandwidth occupancy of the task T to reduce the impact on network resources.

Images