Backscatter communication node multivariate algorithm fusion task scheduling and scheduling method

Abstract

The invention provides a backscatter communication node multivariate algorithm fusion task scheduling and scheduling method, and belongs to the technical field of passive communication. According to the method, through command interaction between a gateway and a sensing node, LSTM model parameters are optimized by using a genetic algorithm, and a sensing chip scheduling strategy is dynamically adjusted in combination with Q-learning. And the gateway predicts the data uploading time point of the sensing node according to the calculated data packet uploading time interval, determines the available time period of an uplink, and realizes efficient task scheduling. According to the method, the communication resource occupation condition can be accurately grasped, the task execution efficiency of the sensing node is remarkably improved, and the communication collision problem under high concurrency control is effectively solved.

Description

Technical Field

[0001] The present invention belongs to the field of passive communication, and specifically relates to a multi-algorithm fusion task scheduling and scheduling method for backscatter communication nodes, aiming to improve the application performance of a fully passive communication system in the Internet of Things environment through refined communication resource management and intelligent scheduling strategies. Background Art

[0002] In passive communication, the backscattering of signals relies on the carrier assistance provided by the gateway, resulting in a half-duplex working mode of communication. To ensure the successful reception of sensor node data, the gateway needs to continuously provide a carrier signal before reception. Otherwise, it will affect downlink communication and it is difficult to achieve efficient concurrent access to multiple sensor nodes. At the same time, the dependence of uplink communication on the gateway carrier makes it difficult for the gateway to perform concurrent control on multiple sensor nodes. Specifically, the uplink is achieved by controlling the reflection of the gateway carrier signal, which leads to a large amount of redundant carrier transmission duration, affecting downlink communication, and causing the gateway to control each sensor node sequentially and unable to execute multi-sensor node tasks concurrently. Object of the Invention

[0003] The present invention proposes a multi-algorithm fusion task scheduling and scheduling method for backscatter communication nodes. By combining technologies such as genetic algorithms, Q-learning, and LSTM, the task scheduling strategy of sensor nodes is optimized. This method enables the Internet of Things gateway to accurately predict future communication resource occupancy, significantly enhancing the scheduling ability for multiple sensor nodes. In a fully passive communication environment, it effectively improves the task execution efficiency of sensor nodes and successfully solves the problem of communication collisions under high-concurrency control, thus providing strong support for the efficient operation of the Internet of Things system.

[0004] The technical solution adopted by the present invention is as follows:

[0005] An intelligent scheduling and accurate prediction method for sensor node tasks based on multi-algorithms in a fully passive communication system, the method comprising:

[0006] 1) Command interaction:

[0007] The gateway sends a control command to the sensor node. After receiving it, the sensor node selects the corresponding scheduling method for the sensor chip and returns the relevant parameters to the gateway.

[0008] 2) Data processing and strategy generation:

[0009] A. The gateway uses the starfish optimization algorithm to optimize the parameters of the LSTM model. First, randomly initialize the positions of starfish individuals to represent the parameters of the LSTM model, and determine algorithm parameters such as population size and maximum number of iterations. Then, calculate the fitness value of each starfish individual (based on the performance metrics of the LSTM model on the validation dataset). Next, update the positions of starfish individuals according to the rules of the starfish optimization algorithm and adjust the parameters of the LSTM model. Continuously repeat the steps of fitness evaluation and position update until the maximum number of iterations is reached or the preset convergence condition is satisfied.

[0010] B. Use the optimized LSTM model to process historical task data (including task type, data volume, communication quality, scheduling parameters, task execution time) and current task real-time data, and extract task features (such as the change trend of task data volume over time, communication quality fluctuation).

[0011] C. Q-learning uses the extracted task features as state inputs to learn the value of state-action pairs. Task features are key information extracted from historical task data and current task real-time data, and are used to describe the state of tasks. These features include: Task type: such as data collection, control command execution, etc. Data volume: the size of data involved in the task. Communication quality: such as signal strength, signal-to-noise ratio, etc. Scheduling parameters: such as task priority, scheduling frequency, etc. Task execution time: the estimated execution time of the task. The task features are transformed into the Q-learning state space, and the Q-learning state space includes:

[0012] Feature selection and normalization: Select features that have an obvious impact on task scheduling, such as task type, data volume, communication quality, etc. Perform normalization processing on the features so that they are in the range of (0, 1] to avoid certain features affecting model training due to different dimensions. For example: where x is the original feature value, x min and x max are the minimum and maximum values of this feature respectively.

[0013] Construction of the state space: Combine the normalized features into a state vector as the state input of Q-learning. For example: S = [TaskType norm , DataSize norm , CommunicationQuality norm , Priority normal where each feature is a normalized value

[0014] ​State space dimension: The dimension of the state space is determined by the number of selected features. For example, if 5 features are selected, the state space is 5-dimensional.

[0015] State-action mapping in Q-learning: Q-learning selects an action A (such as selecting a certain scheduling strategy) according to the current state S, and receives an immediate reward r t . Its immediate reward r t Through the formula: r t = w1·Task Efficiency + w2·Communication Quality - w3·Collision Rate where w1, w2, and w3 are weight coefficients, Task Efficiency is the speed and success rate of task completion, Communication Quality is the reliability and latency of data transmission, and Collision Rate is the frequency of uplink and downlink communication collisions.

[0016] Q-learning update rule: Q-learning updates the Q value through the following formula: where α is the learning rate and γ is the discount factor

[0017] 3) Time calculation and prediction:

[0018] Based on the parameters returned by the sensing nodes and the generated scheduling strategy, the gateway calculates the time interval between adjacent data packet uploads on the sensing nodes and the time interval between the last data packet and the previous data packet, and then predicts the time point of sensing node data upload.

[0019] A. Adaptive scheduling mode: In the adaptive scheduling mode: In the adaptive scheduling mode, the calculation model of the time interval between adjacent data packet transmissions of the sensing node during task execution can be expressed as: For the jth data packet, its time interval from the previous data packet is jointly determined by the chip instruction processing cycle, data conversion duration, and scheduling weight. The specific expression is: where τ i represents the instruction stream length of the ith sensing chip, δ i corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, the total time required for the chip to complete the instruction sequence is θ i , the dynamic scheduling weight coefficient is ω i , T represents the fixed time consumption for single data encapsulation; this model passes through the dynamic weight coefficient ωi Reflect the chip scheduling priority, and parse the instruction time as τ i / f, the data conversion time is δ i / f and the instruction execution time is θ i Perform weighted integration, and finally superimpose the data encapsulation time nT of a fixed period to form a complete time interval calculation system; In the adaptive scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data packet upload mechanism. Assuming that the node has accumulated K data packets transmitted within the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ i Represents the instruction stream length of the i-th sensing chip, δ i Corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , the dynamic scheduling weight coefficient is ω i , T represents the fixed time for single data encapsulation;

[0020] B. Cyclic scheduling mode: In the cyclic scheduling mode: In the cyclic scheduling mode, the calculation model of the transmission interval between adjacent data packets of the sensing node during task execution can be expressed as: For the j-th data packet, its time interval from the previous data packet is determined by the chip instruction processing cycle, data conversion duration, and scheduling weight. The specific expression is: Among them, τ i Represents the instruction stream length of the i-th sensing chip, δ i Corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time for single data encapsulation; the instruction parsing time is τ i / f, the data conversion time is δ i / f and the instruction execution time is θ i Perform weighted integration, and finally superimpose the data encapsulation time nT of a fixed period to form a complete time interval calculation system; In the cyclic scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data packet upload mechanism. Assuming that the node has accumulated K data packets transmitted within the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ iThe instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation;

[0021] C. Random scheduling mode: In random scheduling mode: In the random scheduling mode, the calculation model of the transmission interval of adjacent data packets of sensor nodes during task execution can be expressed as follows: for the jth data packet, the time interval between it and the previous data packet is determined by the chip instruction processing cycle, data conversion time and scheduling weight. The specific expression is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i is the number of times the i-th chip is scheduled in the corresponding scheduling queue; the model takes the instruction parsing time as τ i / f, data conversion time is δ i / f and the execution time of the instruction is θ i Perform weighted integration, and finally superimpose the fixed period data encapsulation time nT to form a complete time interval calculation system; In the random scheduling mode, when there are data fragments that do not reach the preset threshold when the task is terminated, the sensor node will still trigger the final data packet upload mechanism. Assuming that the node has cumulatively transmitted K data packets during the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i is the number of times the i-th chip is scheduled in the corresponding scheduling queue;

[0022] D. Priority scheduling mode: In priority scheduling mode: In the priority scheduling mode, the calculation model of the transmission interval between adjacent data packets of the sensing node during task execution can be expressed as follows: for the j-th data packet, its time interval from the previous data packet is jointly determined by the chip instruction processing cycle, the data conversion duration, and the scheduling weight. The specific expression is: Among them, τ i represents the instruction stream length of the i-th sensing chip, δ i corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single data encapsulation, r i is related to the chip priority p i and the task execution weight w i ; this model weights and integrates the instruction parsing time τ i / f, the data conversion time δ i / f, and the instruction execution time θ i , and finally superimposes the fixed-cycle data encapsulation time consumption nT to form a complete time interval calculation system; In the priority scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data packet upload mechanism. Assuming that the node has cumulatively transmitted K data packets during the task cycle, the time interval between the last data packet and the K-th data packet can be modeled as: Among them, τ i represents the instruction stream length of the i-th sensing chip, δ i corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single data encapsulation, r i is related to the chip priority p i and the task execution weight w i ;

[0023] 4) Bandwidth allocation and data transmission:

[0024] The gateway determines the available time period of the future uplink bandwidth according to the predicted data upload time point, sends a carrier signal at the corresponding time point, the sensing node packs and uploads the data output by the sensing chip according to the carrier signal, and the gateway uses carrier reflection to receive the data.

[0025] 5) Feedback and optimization:

[0026] A. The sensing node feeds back the task execution result to the Q-learning model for further optimizing the parameters of the Q-learning model.

[0027] B. Regularly recollect historical task data and current task real-time data, reuse the starfish optimization algorithm to optimize the LSTM model parameters, and update the LSTM model and Q-learning model to adapt to new task types and communication environment changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a drawing of the mechanism abstract of the multi-algorithm fusion task scheduling and scheduling method for backscatter communication nodes in an embodiment of the present invention.

[0029] Figure 2 It is a simple flowchart of the multi-algorithm fusion task scheduling and scheduling method mechanism for backscatter communication nodes in an embodiment of the present invention.

[0030] Figure 3 It is a detailed flowchart of the multi-algorithm fusion task scheduling and scheduling method mechanism for backscatter communication nodes in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0032] To achieve large-scale deployment of the Internet of Things system, the present invention provides a multi-algorithm fusion task scheduling and scheduling method for backscatter communication nodes. This method is implemented through the following steps:

[0033] 1) Command transmission and reception:

[0034] The gateway generates a control command containing scheduling instructions, task types, and other information according to the system task requirements. Through a specific communication protocol, the control command is sent to the sensing node in the form of a wireless signal. The sensing node is equipped with a corresponding signal receiving module to accurately receive the control command sent by the gateway and parse the command. According to the parsing result, the sensing node selects a sensing chip scheduling method that conforms to the current task scenario and its own resource status from the four preset scheduling methods: cyclic scheduling, random scheduling, priority scheduling, and adaptive scheduling.

[0035] 2) Parameter return:

[0036] A. Cyclic scheduling mode: The sensing node determines the number of cycles α of cyclic scheduling, and counts the length c of the chip control flow i , the output data length d i , the total time t required to execute each command i , and information such as the data collection time T, etc., and packs these parameters and returns them to the gateway.

[0037] B. Random scheduling mode: Record the number of scheduling times r of each chip in the scheduling queuei Collect the chip control flow length c i and the output data length d i as well as the total time t required to execute each command i and the data collection time T and other data, and send them to the gateway

[0038] C. Priority scheduling mode: Calculate the priority p of each chip i and the execution task weight w i and then determine the number of scheduling times r i , together with the chip control flow length c i and the output data length d i as well as the total time t required to execute each command i and the data collection time T and other parameters, and feedback them to the gateway

[0039] D. Adaptive scheduling mode: The sensing node collects historical task data, covering aspects such as task type, data volume, communication quality, scheduling parameters, and task execution time, and performs data cleaning, feature extraction, and normalization processing. Use the Q-Learning algorithm to train the model and adjust the model parameters in real time according to the current task execution situation. The adjusted scheduling parameters, such as task priority, scheduling order, scheduling frequency, and scheduling time, are combined with the chip control flow length c i and the output data length d i as well as the total time t required to execute each command i and the data collection time T and return them to the gateway together

[0040] 3) Time interval calculation and prediction:

[0041] A. Cyclic scheduling mode: Under the cyclic scheduling mode: Under the cyclic scheduling mode, the calculation model of the adjacent packet transmission interval of the sensing node during task execution can be expressed as: For the jth packet, its time interval from the previous packet is jointly determined by the chip instruction processing cycle, data conversion duration, and scheduling weight, and the specific expression is: where τ i represents the instruction stream length of the ith sensing chip, δ i corresponds to the generated data volume of the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single data encapsulation; the instruction parsing time is τ i / f, the data conversion time is δ i / f, and the instruction execution time is θ iPerform weighted integration, and finally superimpose the data encapsulation time consumption nT of a fixed period to form a complete time interval calculation system; In the cyclic scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data packet upload mechanism. Assuming that the node has accumulated and transmitted K data packets within the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ i The instruction stream length of the i-th sensing chip, δ i The generated data volume corresponding to the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single data encapsulation;

[0042] B. Random scheduling mode: In the random scheduling mode: In the random scheduling mode, the calculation model of the transmission interval between adjacent data packets of the sensing node during task execution can be expressed as: For the j-th data packet, its time interval from the previous data packet is jointly determined by the chip instruction processing cycle, data conversion duration, and scheduling weight. The specific expression is: Among them, τ i represents the instruction stream length of the i-th sensing chip, δ i The generated data volume corresponding to the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single data encapsulation, r i is the scheduling times of the i-th chip in the corresponding scheduling queue; this model takes the instruction parsing time as τ i / f, the data conversion time as δ i / f and the instruction execution time as θ i Perform weighted integration, and finally superimpose the data encapsulation time consumption nT of a fixed period to form a complete time interval calculation system; In the random scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data packet upload mechanism. Assuming that the node has accumulated and transmitted K data packets within the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ i represents the instruction stream length of the i-th sensing chip, δ i The generated data volume corresponding to the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i, T represents the fixed time consumption for single - data encapsulation, r i is the scheduling times of the i - th chip in the corresponding scheduling queue;

[0043] C. Priority scheduling mode: Under the priority scheduling mode: Under the priority scheduling mode, the calculation model of the transmission interval between adjacent data packets of the sensing node during task execution can be expressed as: for the j - th data packet, its time interval from the previous data packet is jointly determined by the chip instruction processing cycle, data conversion duration, and scheduling weight. The specific expression is: Among them, τ i represents the instruction stream length of the i - th sensing chip, δ i the generated data volume corresponding to the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single - data encapsulation, r i is related to the chip priority p i and the execution task weight w i ; this model weights and integrates the instruction parsing time τ i / f, the data conversion time δ i / f and the instruction execution time θ i , and finally superimposes the fixed - cycle data encapsulation time consumption nT to form a complete time - interval calculation system; Under the priority scheduling mode, when there are data segments that do not reach the preset threshold when the task terminates, the sensing node will still trigger the final data - packet uploading mechanism. Assume that the node has cumulatively transmitted K data packets during the task cycle, then the time interval between the last data packet and the K - th data packet can be modeled as: Among them, τ i represents the instruction stream length of the i - th sensing chip, δ i the generated data volume corresponding to the chip, f is the local clock reference frequency of the node, and the total time required for the chip to complete the instruction sequence is θ i , T represents the fixed time consumption for single - data encapsulation, r i is related to the chip priority p i and the execution task weight w i ;

[0044] D. Adaptive scheduling mode: Under the adaptive scheduling mode: In the adaptive scheduling mode, the calculation model of the transmission interval of adjacent data packets of sensor nodes during task execution can be expressed as follows: for the jth data packet, the time interval between it and the previous data packet is determined by the chip instruction processing cycle, data conversion time and scheduling weight. The specific expression is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , the dynamic scheduling weight coefficient is w, T represents the fixed time consumption of a single data package; the model uses the dynamic weight coefficient ω i Reflects the chip scheduling priority and sets the instruction parsing time to τ i / f, data conversion time is δ i / f and the execution time of the instruction is θ i Perform weighted integration, and finally superimpose the fixed period data encapsulation time nT to form a complete time interval calculation system; In the adaptive scheduling mode, when there are data segments that do not reach the preset threshold when the task is terminated, the sensor node will still trigger the final data packet upload mechanism. Assuming that the node has cumulatively transmitted K data packets during the task cycle, the time interval between the last data packet and the Kth data packet can be modeled as: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , the dynamic scheduling weight coefficient is ω i , T represents the fixed time consumption of single data encapsulation;

[0045] 4) Carrier allocation and data reception:

[0046] The gateway rationally plans the available time period of future uplink bandwidth based on the predicted sensor node data upload time point. At the determined time point, the gateway sends a carrier signal through the transmitting module. After the sensor node receives the carrier signal, it triggers the data packaging operation and packages the sensor chip output data in a specific format. Using the carrier reflection principle, the sensor node uploads the packaged data to the gateway. The gateway's receiving module receives the reflected carrier signal and parses the data uploaded by the sensor node from it, completing the data reception and collection work.

[0047] 5) Feedback and Optimization:

[0048] A. Q - learning model optimization: The sensing nodes feedback the task execution results, such as the task completion speed, success rate, data transmission reliability, latency, and communication collision situation, etc., to the Q - learning model. The model adjusts its own parameters according to this feedback data, optimizes the value evaluation of the state - action pairs, and improves the model's decision - making ability for task scheduling strategies.

[0049] B. Data collection and model update: Regularly initiate the re - collection of historical task data and current task real - time data. Use the starfish optimization algorithm to optimize the LSTM model parameters. Represent the LSTM model parameters by randomly initializing the positions of starfish individuals, determine algorithm parameters such as population size and maximum number of iterations, calculate the fitness value and update the positions of starfish individuals until the preset conditions are met. Update the LSTM model and the Q - learning model so that the models can better adapt to new task types and changes in the communication environment, and improve the accuracy of task scheduling and prediction.

[0050] After testing, the task status prediction mechanism of the fully passive communication control provided by the embodiments of the present invention can increase the device capacity in the network by 3.42 times respectively in practical applications.

[0051] In summary, the technical solution provided by the present invention brings at least the following beneficial effects:

[0052] A. Fine - grained carrier allocation: The gateway can calculate the future occupancy requirements of each sensing node for the communication link according to parameters such as the operating mode and clock frequency of the sensing nodes, so as to allocate carrier supply in a fine - grained manner, leaving more communication resources for planning concurrent tasks on multiple sensing nodes.

[0053] B. Improve task execution efficiency: Through the adaptive scheduling mode, the sensing nodes can dynamically adjust the scheduling strategy according to the current task execution situation and historical data, optimizing the task execution efficiency and communication performance.

[0054] C. Avoid communication collisions: Effectively avoid the problem that it is difficult to improve task efficiency due to uplink - downlink communication collisions under high - concurrency control, ensuring that the gateway can correctly receive the data information sent by the sensing nodes.

Claims

1. A multi-algorithm fusion task scheduling and scheduling method for backscatter communication nodes, characterized in that: The following steps are involved: Command interaction: The gateway sends control commands to the sensor nodes. After receiving the commands, the sensor nodes select the corresponding sensor chip scheduling mode and return the relevant parameters to the gateway. Data processing and strategy generation: The Starfish optimization algorithm is used to optimize the LSTM model parameters, and the optimized LSTM model is used to extract task features; Based on the Q-learning algorithm, the sensor chip scheduling strategy is dynamically generated with task characteristics as state input; Time calculation and prediction: The gateway calculates the time interval between the uploads of adjacent data packets on the sensor node and the time interval between the last data packet and the previous data packet based on the parameters returned by the sensor node and the generated scheduling strategy, and then predicts the time point for data upload of the sensor node; Bandwidth allocation and data transmission: The gateway determines the available time period of future uplink bandwidth based on the predicted data upload time point, and sends a carrier signal at the corresponding time point. The sensor node packages and uploads the sensor chip output data based on the carrier signal, and the gateway uses carrier reflection to receive data.

2. The method according to claim 1, characterized in that The specific steps of optimizing the LSTM model parameters by the starfish optimization algorithm include: Initialization: Randomly generate a number of starfish individuals to represent the parameters of the LSTM model, set the population size and the maximum number of iterations; Fitness evaluation: The fitness value of each starfish individual is calculated based on the accuracy and mean square error of the LSTM model on the validation set; Position update: adjust the individual position according to the starfish optimization rule to optimize the LSTM model parameters; in is the updated position, is the current position, X best is the current optimal individual position, X rand is the random individual position, α and β are control parameters; Iterative optimization: Repeat the fitness evaluation and position update steps until the preset maximum number of iterations is reached or the convergence condition is met.

3. The method according to claim 1, characterized in that The specific steps of optimizing the LSTM model parameters by the starfish optimization algorithm include: Feature selection and normalization: select task type, data volume, and communication quality as key features and perform normalization; State space construction: The normalized features are combined into a state vector as the input of the Q-learning algorithm. The state vector is specifically expressed as: S=[TaskType norm ,DataDize norm ,CommunicationQuality norm ,Priority normal ] Among them, each feature is a normalized value. This state vector can fully reflect the state information of the current task and provide a basis for the Q-learning algorithm to accurately select appropriate actions; Reward function design: The immediate reward is calculated by comprehensively considering factors such as task efficiency, communication quality, and collision rate. The reward formula is: r t =w1·Task Efficiency+w2·Communication Quality-w3·Collision Rate Among them, w1, w2, and w3 are weight coefficients, Task Efficiency is the speed and success rate of task completion, Communication Quality is the reliability and delay of data transmission, and Collision Rate is the frequency of collisions in uplink and downlink communications. Reasonable setting of these weight coefficients can allow the Q-learning algorithm to pay more attention to certain key factors according to different application scenarios and requirements, thereby generating a more appropriate scheduling strategy; Q value update: Dynamically adjust the Q value of state-action based on the Bellman equation to continuously optimize the state-action mapping relationship; the specific update formula is: Among them, α is the learning rate, which is used to control the step size of each update. If the learning rate is too large, the algorithm will be unstable, and if it is too small, the convergence speed will be slowed down. γ is the discount factor, which is used to measure the importance of future rewards. It reflects the degree of attention the algorithm pays to long-term interests.

4. The method according to claim 1, characterized in that The scheduling modes include adaptive scheduling, cyclic scheduling, random scheduling and priority scheduling. The calculation formulas for the time interval of data packet upload in different modes are: In adaptive scheduling mode: During task execution, the time interval between the jth data packet and the previous data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , the dynamic scheduling weight coefficient is ω i , T represents the fixed time consumption of single data encapsulation; When the task terminates, if there are data segments that do not reach the preset threshold, the time interval between the last data packet and the Kth data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , the dynamic scheduling weight coefficient is ω i , T represents the fixed time consumption of single data encapsulation; In round-robin scheduling mode: During task execution, the time interval between the jth data packet and the previous data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation; When the task terminates, if there are data segments that do not reach the preset threshold, the time interval between the last data packet and the Kth data packet is: Among them, τ i The instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation; In random scheduling mode: During task execution, the time interval between the jth data packet and the previous data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i is the number of times the i-th chip is scheduled in the corresponding scheduling queue; When the task terminates, if there are data segments that do not reach the preset threshold, the time interval between the last data packet and the Kth data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i is the number of times the i-th chip is scheduled in the corresponding scheduling queue; In priority scheduling mode: During task execution, the time interval between the jth data packet and the previous data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i With chip priority p i and the execution task weight w i Related; When the task terminates, if there are data segments that do not reach the preset threshold, the time interval between the last data packet and the Kth data packet is: Among them, τ i represents the instruction stream length of the i-th sensor chip, δ i The amount of data generated by the corresponding chip, the local clock reference frequency of the f node, and the total time required for the chip to complete the instruction sequence are θ i , T represents the fixed time consumption of single data encapsulation, r i With chip priority p i and the execution task weight w i Related.

5. The method according to claim 1, characterized in that The method also includes feedback and optimization steps for sensor node task scheduling: The sensor node feeds back the task execution results to the Q-learning model to further optimize the parameters of the Q-learning model; Regularly re-collect historical task data and current task real-time data, reuse the Starfish optimization algorithm to optimize the LSTM model parameters, and update the LSTM model and Q-learning model to adapt to new task types and changes in the communication environment.

Images