Telemetry data processing task allocation method and device and storage medium

Abstract

The invention discloses a telemetry data processing task allocation method and device and a storage medium, belongs to the technical field of telemetry data processing, and realizes intelligent scheduling through a reinforcement learning model. The model can dynamically select the most suitable device for task allocation according to the resource demand of the task and the real-time state of the device, so that the task scheduling efficiency and flexibility are greatly improved. Through continuous training, the model can gradually optimize an allocation strategy, so that the task can be efficiently and safely completed in a complex and high-dynamic working environment, and meanwhile, the waste of equipment resources and the fault risk are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of telemetry data processing, and in particular to a telemetry data processing task allocation method, device and storage medium. Background Art

[0002] In the aerospace field, telemetry data generated by spacecraft such as rockets and satellites is crucial. This data records the status of the spacecraft in real time, including information such as position, attitude, temperature, pressure, and power system. Due to the complex on-orbit environment of spacecraft, real-time monitoring and analysis of this data is the key to ensuring mission success and safety. However, as the complexity of missions and the amount of data increase, the ability of a single device to process massive amounts of data at the ground station is gradually insufficient. Distributed multi-device collaborative processing solutions can improve processing efficiency and ensure the integrity and real-time nature of data when failures occur.

[0003] Existing aerospace technology solutions usually use static task allocation or simple rule-based scheduling methods, and introduce redundancy mechanisms to ensure the progress of tasks. Although these methods can complete task allocation to a certain extent, they lack intelligent dynamic scheduling capabilities and cannot effectively cope with highly dynamic telemetry data streams, which can easily lead to uneven equipment load and waste of resources. In addition, when equipment or communication links fail, tasks are prone to failure.

[0004] In view of this, it is necessary to provide a new technical solution to solve the above problems. Summary of the invention

[0005] To solve the above technical problems, the present application provides a telemetry data processing task allocation method, device and storage medium, which can dynamically adjust the task allocation strategy, realize efficient resource scheduling and load balancing, ensure the optimal utilization of equipment resources, and improve the overall performance and reliability of the system.

[0006] A telemetry data processing task allocation method, comprising: Monitor the key attributes of device nodes in real time and obtain the resource usage status data of the device; The obtained resource usage status data is stored in the database; Build a task scheduling model based on the reinforcement learning model; Train the constructed task scheduling model; According to the real-time resource usage status of the equipment and the resource requirements of the telemetry data processing tasks, the telemetry data processing task allocation strategy is dynamically determined using the trained task scheduling model.

[0007] Preferably, the resource usage status data of the device includes CPU usage, memory usage, network bandwidth occupancy, network latency, disk I / O load and health status of the device.

[0008] Preferably, the acquired resource usage status data is stored in a database, and the data structure stored in the database includes: measurement value, label, field and timestamp; Wherein, the measurement value is used to store the acquired resource usage status data; The tag serves as a unique identifier for the device, making it easy to query the resource usage status data of each device; Fields include CPU usage, memory usage, network bytes sent, network bytes received, network latency, disk read bytes, disk write bytes, and device health status; The timestamp is used to record the collection time of resource usage status data.

[0009] Preferably, storing the acquired resource usage status data in a database includes: Use the client library influxdb-client provided by the InfluxDB database to periodically write the resource usage status data of the device into the database to store the resource usage status data; Each time after the resource usage status data of the device is obtained, the data is inserted into the database as a new record, and the corresponding timestamp of the resource usage status data of the device is recorded.

[0010] 5 Preferably, the task scheduling model is constructed based on the reinforcement learning model, including: The usage status of the device resources at each time step and the resource requirements of the task are used as the status of the task scheduling model; Define the size of the action space and the representation of the action space; Designed as a reward signal: Build a DQN network model.

[0011] Preferably, the usage status of the device resources at each time step and the resource requirements of the task are used as the status of the task scheduling model, including: Define the usage status of device resources in each time step; Define the resource requirements of the task at each time step; Combining the usage status information of device resources and the resource requirement information of tasks, a joint state space containing high-dimensional vectors is formed.

[0012] Preferably, when designing a reward signal, the reward signal is designed based on whether the task is successfully executed, the equipment load and resource matching, the time efficiency of the task execution, and the health status of the equipment; Among them, the reward function expression is: ; In the formula, Rewards for successful or unsuccessful execution of tasks; The reward weight for whether the task is successfully executed or not; Rewards for equipment load and resource matching; Reward weights for matching device load and resources; Reward for time efficiency in task execution; Reward weights for time efficiency of task execution; Reward for equipment health; Reward weight for device health status.

[0013] Preferably, the training of the constructed task scheduling model includes: S41, using the state vector of the device and the task, the task assignment probability of each device as output, and the total number of devices as action, initialize the Q network with random network parameters; S42, initializing the experience replay pool; S43, selecting an action using an epsilon-greedy strategy according to the resource usage status of the current device and the resource requirements of the telemetry data processing task; S44, execute the selected action, obtain the corresponding reward value and environment state change value, and put the next state after the selected action into the experience replay pool; S45, selecting a number of data from the experience replay pool, and calculating each selected data using the target network; S46, calculating the minimized target loss, and updating the current Q network with the minimized target loss; Steps S43 to S46 are repeated until a preset time step is reached or the model converges.

[0014] According to another aspect of the present application, a computing device is also provided, including: a processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the telemetry data processing task allocation method is executed.

[0015] According to another aspect of the present application, a computer-readable storage medium is provided, on which computer instructions are stored. When the instructions are executed on a computer, the computer executes the telemetry data processing task allocation method.

[0016] Compared with the prior art, this application has at least the following beneficial effects: 1. The present invention can dynamically adjust the task allocation strategy, achieve efficient resource scheduling and load balancing, ensure the optimal utilization of equipment resources, and improve the overall performance and reliability of the system.

[0017] 2. The present invention uses a deep learning model to monitor the device status in real time and adaptively adjusts task allocation according to task priority and resource requirements. This scheduling mechanism ensures efficient task allocation and maximizes device resource utilization.

[0018] 3. Through intelligent monitoring and scheduling, the present invention can quickly adjust task allocation when equipment or communication links fail, avoiding data loss and task interruption. The system automatically reallocates tasks according to equipment status and failure conditions to ensure task continuity and system stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings: Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 This is the network structure diagram of the DQN network. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0021] like Figure 1 As shown, a telemetry data processing task allocation method includes the following steps: S1. Monitor the key attributes of device nodes in real time and obtain the resource usage status data of the device.

[0022] Specifically, in order to implement a task scheduling load balancing mechanism, it is necessary to obtain the resource usage status data of the device by monitoring the key attributes of the device nodes in real time.

[0023] Among them, the device's resource usage status data includes information such as CPU usage, memory usage, network bandwidth occupancy, network latency, disk I / O load, and device health status.

[0024] CPU usage (CPU_usage): The CPU load of the device, reflecting the device's ability to process tasks.

[0025] Memory usage (Memory_usage): The memory resource usage of the device, which measures the remaining memory availability.

[0026] Network bandwidth usage (Bandwidth_usage): The network traffic usage of the device reflects the bandwidth usage of the device during data transmission.

[0027] Network latency: The delay in network data transmission affects the communication efficiency between devices.

[0028] Disk I / O Load (Disk_IO): The disk input / output load of the device, indicating the burden of disk read and write operations.

[0029] Device health status (Device_health_status): The operating status of the device. If the device is normal, it is 1; if an abnormality occurs, it is 0.

[0030] By collecting this data, the current load status of the equipment, resource availability and task requirements can be evaluated, thus providing a basis for task allocation.

[0031] When obtaining the resource usage status data of the device, the Python language is used to implement the device monitoring script, and the monitoring interface provided by the operating system (such as psutil, ping3) is used to regularly collect the resource usage of the device. The specific implementation steps are as follows: CPU usage (CPU_usage): Get the device's CPU usage through psutil.cpu_percent().

[0032] Memory usage (Memory_usage): Get the device's memory usage through psutil.virtual_memory().

[0033] Network bandwidth usage (Bandwidth_usage): Get the device's network traffic information through psutil.net_io_counters() and calculate the number of bytes sent and received respectively.

[0034] Network latency: Use the ping3.ping() command to measure the network latency between the device and the target host (such as 8.8.8.8).

[0035] Disk I / O load (Disk_IO): Get the number of read and written bytes of the disk through psutil.disk_io_counters().

[0036] Device health status (Device_health_status): Based on the various resource indicators of the device, evaluate whether the device is in an abnormal state (such as high CPU usage, excessive memory usage, etc.), and determine whether the device is in a healthy state based on the threshold.

[0037] S2. Store the acquired resource usage status data in a database.

[0038] Specifically, in order to store the monitoring data of the device in the database, the InfluxDB database is used as the time series database for data storage and management. InfluxDB is designed for efficient storage and query of time series data, and can provide powerful time series query and data analysis capabilities, which is very suitable for storing device monitoring data.

[0039] When storing data, in the InfluxDB database, monitoring data will be stored according to timestamps, and measurement will be used to store device status information. The monitoring data of each device will be inserted into the database as a point.

[0040] Specifically, it includes: using the client library influxdb-client provided by the InfluxDB database to periodically write the resource usage status data of the device into the database to realize the storage of the resource usage status data; Each time the resource usage status data of the device is obtained, the data is inserted into the database as a new record, and the corresponding timestamp of the resource usage status data of the device is recorded, and the timestamp is used to ensure the order and queryability of the data.

[0041] When storing data, the data structure is designed as follows: Measurement:device_status: Used to store device status data.

[0042] Tags: device_id: The unique identifier of the device, which facilitates querying device-related monitoring data.

[0043] Fields: cpu_usage: CPU usage; memory_usage: memory usage; bandwidth_sent: the number of network bytes sent; bandwidth_recv: the number of network bytes received; latency: network delay; disk_read_bytes: number of bytes read from disk; disk_write_bytes: number of bytes written to disk; health_status: device health status (1: normal, 0: abnormal); Timestamp: The timestamp of the record. The time when the device monitoring data was collected is used as the timestamp.

[0044] S3. Build a task scheduling model based on the reinforcement learning model.

[0045] The present invention constructs an intelligent scheduling system by using a reinforcement learning model to automatically adjust task allocation in response to changes in task volume and equipment resource overload.

[0046] (1) Solution selection In this invention, reinforcement learning (RL) is selected as the model for task scheduling, and the deep Q network (DQN) algorithm is selected to handle large-scale, continuous state space task scheduling problems. Reinforcement learning continuously optimizes decision-making strategies through the interaction between the agent and the environment. The core of the model is to guide the agent to make the optimal task allocation decision based on the resource usage and task requirements of the device through the reward mechanism.

[0047] (2) Environment modeling and state representation The training of the model needs to consider both the state information of the device and the resource requirements of the task. The present invention uses the resource requirements of the task as the task state and combines the current state of the device to make scheduling decisions.

[0048] State: The device resource state and task requirements at each time step, including all monitored data (such as CPU usage, memory usage, bandwidth usage, latency, disk I / O load, etc.). Task requirements can be obtained from historical data. The task resource requirements and device resource usage are considered as a "joint state", and then the device is selected based on this state. The details are as follows: Device status information: The status of each device includes information such as its resource usage and health status, which reflects the resource availability and performance of the device. The device status includes: CPU usage (cpu_usage): The current CPU usage ratio of the device, ranging from [0, 1].

[0049] Memory usage (memory_usage): The current usage ratio of the device's memory resources, ranging from [0, 1].

[0050] Bandwidth usage (bandwidth_usage): The network bandwidth usage of the device, which can be calculated by the sum of the number of bytes sent and received, and the range is [0, 1].

[0051] Network latency: The network latency of a device, which indicates the response time from the device to other devices or servers, is usually measured in milliseconds and ranges from [0, 1].

[0052] Disk I / O load (disk_io): The disk input / output load of the device, including the number of read and written bytes, in the range [0, 1].

[0053] Device health status (health_status): Whether the device is operating normally, usually 0 (abnormal) or 1 (normal).

[0054] This information together constitutes the device status, indicating the current resource usage of the device.

[0055] Task requirements information: The resource requirements of a task describe the amount of resources required to execute the task. Task status includes: CPU demand (task_cpu_demand): The CPU resources required for task execution. The range is [0, 1], indicating the proportion of CPU occupied by the task.

[0056] Memory demand (task_memory_demand): The memory resources required for task execution. The range is [0, 1], indicating the proportion of memory occupied by the task.

[0057] Bandwidth demand (task_bandwidth_demand): The network bandwidth required for task execution. The range is [0,1], which indicates the proportion of network bandwidth occupied by the task.

[0058] Latency requirement (task_latency_demand): The network latency requirement of a task, ranging from [0, 1], indicating the maximum latency that a task can tolerate. This information represents the resource requirements of a task and is a key factor in task scheduling decisions.

[0059] Joint state space S: Combining the device state information and task requirement information, the joint state space S forms a high-dimensional vector that contains all relevant information about the device and the task.

[0060] Joint state vector: S = [{cpu_usage}, {memory_usage}, {bandwidth_usage}, {latency},{disk_io}, {health_status}, {task_cpu_demand}, {task_memory_demand}, {task_bandwidth_demand}, {task_latency_demand}].

[0061] Specifically, this state space consists of the following: Device resource status: device CPU usage, memory usage, bandwidth usage, network latency, disk I / O load, and device health status.

[0062] Resource requirements of the task: CPU requirements, memory requirements, bandwidth requirements, and latency requirements of the task.

[0063] Therefore, the dimension of the state space is 10, corresponding to the 6 state attributes of the equipment and the 4 requirement attributes of the task.

[0064] (3) Action space: including the definition of the size of the action space and the representation of the action space.

[0065] Action space definition: Number of devices N: Assuming there are N devices in the system, the size of the action space is N, and each action corresponds to selecting a device to perform the current task.

[0066] Action representation: When selecting action a, the model selects the a∈[0,N−1]th action, where N is the total number of devices.

[0067] (4) Reward Model The goal of the reward model is to encourage the model to select reasonable devices to perform tasks in order to improve the overall system efficiency, avoid device overload or resource waste, and ensure that tasks can be successfully executed within a reasonable time.

[0068] The design of reward signals is mainly based on the following aspects: The success or failure of task execution, equipment load and resource matching, time efficiency of task execution and equipment health status.

[0069] Whether the task is executed successfully or not: Whether the task can be successfully executed on the device is the core of the reward mechanism.

[0070] Successfully executed task: If the task is successfully executed on the selected device, a positive reward is given .

[0071] Task execution failure: If the task fails to execute on the selected device due to insufficient resources or device failure, a negative reward is given. .

[0072] Equipment load and resource matching degree: The matching degree between device load and resource usage will affect the reward of task scheduling. If the device resources (such as CPU usage, memory usage, network bandwidth, etc.) match the task requirements, a positive reward is given. A simple way to calculate the resource matching degree is to calculate the difference between the task requirements and the current resources of the device and obtain the matching degree through a weighted formula.

[0073] Matching Reward Formula: ; in, is the weight, It is the difference between task resource requirements and equipment resources.

[0074] In actual aerospace applications, different tasks may have different requirements for resources and the importance of equipment capabilities. For example, for the task of analyzing complex data in telemetry data, CPU may be more important than bandwidth, while disk performance is more critical for the storage of high-frequency telemetry data. Therefore, for each task, a weight can be assigned to each resource, and the difference metric is calculated according to the weight. The weighted difference metric is calculated as follows: ; in, is the weight of the i-th resource dimension, and are the resource requirements and status of tasks and devices in the i-th dimension respectively.

[0075] Task execution time performance: Task execution time is an important factor in task scheduling, especially for real-time tasks. When executing a task, timeout will cause system performance to degrade, so execution time will affect the reward: Complete the task on time: If the task is completed within a reasonable time, a positive reward is given .

[0076] Task timeout: If the task is not completed on time, a negative reward is given .

[0077] Device health status: The health status of the device also affects the reward. If the device is in an unhealthy state (such as hardware failure or abnormality), assigning a task to the device may cause the task to fail. Therefore, the health status factor should also be included in the reward.

[0078] Healthy Equipment: If the equipment health status is 1 (normal), a positive reward is given .

[0079] Unhealthy device: If the device health status is 0 (abnormal), a negative reward is given .

[0080] Based on these factors, the present invention designs the reward function as a weighted sum, where each factor has a corresponding weight. The specific reward function is as follows: ; In the formula, Rewards for successful or unsuccessful execution of tasks; The reward weight for whether the task is successfully executed or not; Rewards for equipment load and resource matching; Reward weights for matching device load and resources; Reward for time efficiency in task execution; Reward weights for time efficiency of task execution; Reward for equipment health; Reward weight for device health status.

[0081] (5) Building a DQN network model The present invention selects the deep Q network (DQN) model. DQN is composed of a deep neural network, the input is the current state (the combination of the device state and the task state), and the output is the Q value of each device. The Q value represents the expected return of selecting each device to perform the task under the current state. In this embodiment, the network structure of the deep Q network model is as follows Figure 2 shown.

[0082] The DQN network model includes an input layer, a hidden layer, and an output layer.

[0083] Input layer: The size is the dimension of the state space, that is, the number of features of the device state and task state. In this solution, the device state has 6 features, the task state has 4 features, and the total feature dimension is 10.

[0084] Hidden layer: In this solution, the network contains 3 hidden layers, each of which consists of multiple neurons. Each layer uses the ReLU activation function to increase nonlinearity, so that the model can better fit complex tasks and device relationships.

[0085] Output layer: The dimension of the output layer is equal to the size of the action space. In the device selection task, each output neuron represents the Q value of selecting a specific device, that is, the expected reward of the action. The output layer uses a linear activation function because the Q value is a continuous value and is suitable for regression tasks.

[0086] S4. Train the constructed task scheduling model.

[0087] S41. With the state vector of the device and the task, the task assignment probability of each device is output, and the total number of devices is taken as the action, the Q network is initialized with random network parameters.

[0088] Specifically, the network is initialized with random network parameters , the input is the state vector of the device and task, the output is the task allocation probability of each device, and the action is the total number of devices N. The purpose of the Q network is to learn how to output the optimal task allocation strategy through the input device state. Copy the same parameters To initialize the target network .

[0089] S42. Initialize the experience replay pool.

[0090] In order to avoid correlation between samples, the present invention uses an experience replay mechanism to initialize the experience replay pool R.

[0091] S43. According to the resource usage status of the current device and the resource requirements of the telemetry data processing task, an action is selected using an epsilon-greedy strategy.

[0092] Specifically, according to the status (Current device status and task requirements) Select actions using an epsilon-greedy strategy (Task allocation method).

[0093] S44. Execute the selected action, obtain the corresponding reward value and environment state change value, and put the next state after the selected action into the experience replay pool.

[0094] Specifically, perform actions , get rewards , the environment state becomes ,Will(,) , that is, the task allocation for a certain requirement under a certain equipment state, and the next state after the action is selected and put into the experience replay pool R.

[0095] S45. Select a number of data from the experience replay pool, and calculate each selected data using the target network.

[0096] S46, calculate the minimized target loss, and update the current Q network with the minimized target loss.

[0097] If there are enough data in R, use N data from R , for each data, use the target network to calculate .

[0098] Among them, the minimized target loss is: ; Update the current network with this and .

[0099] Steps S43 to S46 are repeated until a preset time step is reached or the model converges.

[0100] The model is evaluated by testing the scheduling performance of the agent under different equipment load conditions and task requirements. The following indicators can be used for evaluation: The efficiency of task allocation: that is, whether the assigned tasks are evenly distributed to devices with lower loads.

[0101] Latency of task execution: Whether the task allocation results in too high a delay.

[0102] Device load balancing: whether any device is unable to work properly due to overload.

[0103] S5. According to the real-time resource usage status of the equipment and the resource requirements of the telemetry data processing task, the telemetry data processing task allocation strategy is dynamically determined using the trained task scheduling model.

[0104] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0105] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A telemetry data processing task allocation method, characterized in that: include: Monitor the key attributes of device nodes in real time and obtain the resource usage status data of the device; The obtained resource usage status data is stored in the database; Build a task scheduling model based on the reinforcement learning model; Train the constructed task scheduling model; According to the real-time resource usage status of the equipment and the resource requirements of the telemetry data processing tasks, the telemetry data processing task allocation strategy is dynamically determined using the trained task scheduling model.

2. The telemetry data processing task allocation method according to claim 1, characterized in that: The resource usage status data of the device includes CPU usage, memory usage, network bandwidth occupancy, network latency, disk I / O load and device health status.

3. The telemetry data processing task allocation method according to claim 2, characterized in that: The acquired resource usage status data is stored in a database, wherein the data structure stored in the database includes: measurement value, label, field and timestamp; Wherein, the measurement value is used to store the acquired resource usage status data; The tag serves as a unique identifier for the device, making it easy to query the resource usage status data of each device; Fields include CPU usage, memory usage, network bytes sent, network bytes received, network latency, disk read bytes, disk write bytes, and device health status; The timestamp is used to record the collection time of resource usage status data.

4. The telemetry data processing task allocation method according to claim 3, characterized in that: The storing of the acquired resource usage status data into a database includes: Use the client library influxdb-client provided by the InfluxDB database to periodically write the resource usage status data of the device into the database to store the resource usage status data; Each time after the resource usage status data of the device is obtained, the data is inserted into the database as a new record, and the corresponding timestamp of the resource usage status data of the device is recorded.

5. The telemetry data processing task allocation method according to claim 4, characterized in that: The task scheduling model is constructed based on the reinforcement learning model, including: The usage status of the device resources at each time step and the resource requirements of the task are used as the status of the task scheduling model; Define the size of the action space and the representation of the action space; Designed as a reward signal: Build a DQN network model.

6. The telemetry data processing task allocation method according to claim 5, characterized in that: The usage status of the device resources at each time step and the resource requirements of the task are used as the status of the task scheduling model, including: Define the usage status of device resources in each time step; Define the resource requirements of the task at each time step; Combining the usage status information of device resources and the resource requirement information of tasks, a joint state space containing high-dimensional vectors is formed.

7. The telemetry data processing task allocation method according to claim 6, characterized in that: When designing a reward signal, the reward signal is designed based on the success or failure of task execution, device load and resource matching, time efficiency of task execution, and device health status; Among them, the reward function expression is: ； In the formula, Rewards for successful or unsuccessful execution of tasks; The reward weight for whether the task is successfully executed or not; Rewards for matching equipment load and resources; Reward weights for matching device load and resources; Reward for time efficiency in task execution; Reward weights for time efficiency of task execution; Reward for equipment health; Reward weight for device health status.

8. The telemetry data processing task allocation method according to claim 7, characterized in that: The training of the constructed task scheduling model includes: S41, using the state vector of the device and the task, the task assignment probability of each device as output, and the total number of devices as action, initialize the Q network with random network parameters; S42, initializing the experience replay pool; S43, selecting an action using an epsilon-greedy strategy according to the resource usage status of the current device and the resource requirements of the telemetry data processing task; S44, execute the selected action, obtain the corresponding reward value and environment state change value, and put the next state after the selected action into the experience replay pool; S45, selecting a number of data from the experience replay pool, and calculating each selected data using the target network; S46, calculating the minimized target loss, and updating the current Q network with the minimized target loss; Steps S43 to S46 are repeated until a preset time step is reached or the model converges.

9. A computing device, characterized in that include: A processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the telemetry data processing task allocation method according to any one of claims 1 to 8 is executed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the instructions are executed on a computer, the computer executes the telemetry data processing task allocation method according to any one of claims 1 to 8.

Images