Marine satellite Internet of Things multi-modal data narrowband transmission scheduling method and device

Abstract

The invention provides an ocean satellite Internet of Things multi-modal data narrow-band transmission scheduling method and device, belongs to the technical field of ocean satellite Internet of Things, and innovatively provides a multi-modal data narrow-band transmission scheduling strategy for challenges of ocean observation equipment in an energy-limited and dynamic satellite channel environment. The core of the method comprises the following steps: constructing a multi-modal data dynamic transmission model, and carrying out data value evaluation based on space-time intelligence; designing a congestion perception and hierarchical congestion control strategy to guarantee data timeliness; the transmission fairness is optimized in combination with the data value weight factor, the PAoI and the Jain fairness index; an energy dynamic evolution model is constructed, and long-term energy revenue and expenditure balance of equipment is guaranteed; based on a deep Q network and imitation learning, in combination with a security reinforcement learning framework, information timeliness, energy consumption and congestion are jointly optimized. According to the method, the deep and far sea observation and detection data transmission time delay, energy consumption and the congestion occurrence rate are remarkably reduced, and the information timeliness and the multi-modal data transmission scheduling fairness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of marine satellite Internet of Things, and particularly to a narrowband transmission scheduling method and device for multi-modal data of marine satellite Internet of Things. Background Art

[0002] The integration of marine observation and detection equipment and low-earth orbit satellite Internet has become a key infrastructure for far-reaching sea environment observation and target detection. To meet the multi-dimensional perception requirements of complex marine environments and targets, devices such as drifting buoys, wave energy gliders, Argo buoys, and real-time moored buoys are equipped with meteorological sensors, hydrological sensors, and optoelectronic cameras to achieve the collection of multi-modal data such as marine meteorology, hydrology, and targets. Based on the data backhaul technology of low-earth orbit satellite Internet, the coverage limitation of traditional shore-based communication has been broken through, providing wide-area interconnection capabilities for far-reaching sea areas. However, the contradiction between the characteristics of satellite narrowband communication and the differentiated transmission requirements of multi-modal data has become increasingly prominent, and it is difficult to guarantee the transmission scheduling efficiency and service quality of multi-modal data. How to achieve efficient narrowband transmission scheduling of multi-modal data under multi-modal data, dynamic network status, and strict energy constraints has become a core problem to be solved urgently.

[0003] In the far-reaching sea observation and detection scenario, the transmission requirements of multi-modal data show significant spatio-temporal heterogeneity. Periodic meteorological and hydrological data need to be continuously and stably backhauled to support long-term monitoring of the marine environment; while event-triggered data such as target discovery, hydrological anomalies, and meteorological warnings require low latency and high-reliability transmission. In addition, affected by satellite communication windows, the volatility of device solar power supply, and the complex marine environment, data transmission needs to maximize the transmission value within a limited communication window. Existing transmission scheduling methods have deficiencies in data value quantification, multi-objective collaborative optimization, and adaptive congestion control, and cannot meet the differentiated transmission requirements of multi-modal data in complex scenarios.

[0004] Currently, the narrowband transmission of multi-modal data in marine satellite Internet of Things faces multiple challenges: First, insufficient mining of spatio-temporal characteristics leads to low transmission efficiency. Existing mechanisms lack in-depth analysis of the spatio-temporal correlation characteristics of multi-modal data, and the division of data transmission priorities lacks a scientific basis; Second, the congestion control problem under satellite link resource constraints. Traditional congestion perception usually relies on a single index and lacks the multi-dimensional perception ability of bandwidth utilization, delay jitter, and queue length change rate, making it difficult to adapt to dynamic scenarios such as bad weather and data peaks; Third, the energy supply of far-reaching sea observation and detection equipment is unstable. Traditional energy acquisition models do not consider the day-night difference of solar power supply, and data transmission does not consider the long-term energy balance of devices. Therefore, facing the transmission requirements of multi-modal data in the complex far-reaching sea environment, there is an urgent need to design a narrowband transmission scheduling method that integrates spatio-temporal intelligent analysis, adaptive congestion control, and energy dynamic optimization.

[0005] Existing transmission scheduling strategies are mostly based on a single optimization goal, ignoring the resource competition and coupling relationship among information timeliness, energy consumption, and congestion. Traditional methods may reduce energy consumption by sacrificing information timeliness, or blindly discard data during congestion, resulting in the loss of critical information and unable to achieve multi-objective dynamic trade-off. The multi-objective joint optimization model proposed in the present invention incorporates the Peak Age of Information (PAoI) with data value weights, device energy consumption, and network congestion into a unified objective function, transforming the multi-modal data transmission scheduling problem into a Pareto optimal problem of minimizing the weighted sum. This model can adaptively adjust the weight coefficients of each optimization term, balance information freshness, energy efficiency, and network load while ensuring the transmission priority of critical data, and significantly improve the overall performance of the system.

[0006] In solving high-dimensional dynamic optimization problems, existing methods such as Lyapunov optimization-based greedy strategies, heuristic algorithms, and linear programming have difficulty handling the non-linear, time-varying constraints, and large-scale state space of multi-modal data scheduling in marine satellite Internet of Things. Although traditional reinforcement learning has the ability to handle complex decisions, it is prone to unstable training due to low exploration efficiency and high risk of constraint violation in a dynamic marine environment. The present invention innovatively integrates the Deep Q-Network (DQN) with imitation learning to construct a safe reinforcement learning framework: accelerating the convergence of DQN through expert policy pre-training and guiding the exploration direction using the action probability distribution provided by imitation learning; at the same time, introducing a safety constraint layer to transform bandwidth limitations and energy constraints into the feasible domain boundary of the state space, ensuring that the actions generated by the policy naturally satisfy physical constraints. This hybrid solution method not only reduces the training cost but also can stably generate optimal scheduling strategies that meet multi-objective trade-offs in a complex dynamic marine environment, providing theoretical feasibility and technical advancement guarantees for multi-modal data transmission in narrowband satellite Internet of Things. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a method and device for narrowband transmission scheduling of multi-modal data in marine satellite Internet of Things to eliminate or improve one or more defects existing in the prior art and solve the problems of insufficient analysis of spatio-temporal characteristics of existing multi-modal data, limited satellite communication link resources, unstable device energy supply, and the fact that existing transmission scheduling methods are not fully applicable to the complex dynamic environment of marine satellite Internet of Things.

[0008] On the one hand, the present invention provides a method for narrowband transmission scheduling of multi-modal data in marine satellite Internet of Things, characterized in that the method includes the following steps: Construct a multi-modal data dynamic transmission model for deep and far-sea observation and detection equipment: This model receives multi-modal data collected from meteorological sensors, hydrological sensors, and optoelectronic cameras, and establishes independent data queues respectively according to the nature of the data (periodic meteorological and hydrological data, event-triggered target discovery, hydrological anomalies, and meteorological warning data), and dynamically updates the queue length.

[0009] In view of the bandwidth limitation of satellite narrowband communication, construct a bandwidth resource constraint model to ensure efficient data transmission within limited bandwidth resources.

[0010] Conduct spatio-temporal intelligent-driven data value evaluation. Use the Long Short-Term Memory (LSTM) network to analyze the fluctuation characteristics in the time dimension, and combine the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to perform spatial clustering on hot regions in the deep and far sea, quantify data value factors, and achieve differential scheduling of data.

[0011] Design a congestion awareness and hierarchical congestion control strategy. Calculate the network congestion index based on bandwidth utilization, delay jitter, and the change rate of queue length, and dynamically adjust the index weights according to different scenarios (severe weather, data transmission peak period, and sudden data influx) to achieve scenario-adaptive network congestion assessment. Design a hierarchical congestion control strategy based on thresholds to adjust the data transmission strategy according to the congestion degree.

[0012] Construct an information timeliness model. Evaluate the timeliness of data packets through PAoI, and combine the data value weight factor to construct a weighted PAoI index to evaluate the information timeliness of data with different values. Construct a multi-sensor PAoI fairness function that integrates data value weights based on the Jain fairness index to ensure the fairness of multi-sensor data transmission and avoid high-value data monopolizing resources for a long time.

[0013] Establish an energy dynamic evolution model for deep and far-sea observation and detection equipment. Combine the characteristics of solar power supply and the diurnal cycle change to establish a dynamic energy acquisition model, and design an energy consumption model according to the transmission and sleep energy consumption characteristics of sensors to ensure that the equipment meets the long-term energy balance.

[0014] Construct a multi-objective optimization function to minimize the weighted sum of weighted PAoI, energy consumption, and congestion. Adopt a reinforcement learning strategy that combines DQN and imitation learning. Model the state space and action space through the Markov decision process, introduce a safety constraint layer to limit bandwidth and energy overbound behavior, and generate a dynamic optimal transmission scheduling strategy.

[0015] In some embodiments of the present invention, the method further includes: The hierarchical congestion control strategy dynamically adjusts the weight allocation for different scenarios. For example, it increases the weight of delay jitter in bad weather and increases the weight of queue change rate when burst data surges, enhancing the adaptability to scenarios.

[0016] The fairness transmission scheduling balances the priority transmission of high-value data and the basic service requirements of low-value data by fusing the data value weight and the Jain fairness index, avoiding the phenomenon of data starvation.

[0017] The energy harvesting model designs a mechanism for adaptive adjustment in the day-night cycle, maximizing energy storage during the day and optimizing transmission tasks at night to reduce energy consumption and extend the device life cycle.

[0018] The optimization problem solving method based on DQN and imitation learning pre-trains the DQN network using expert historical data, accelerates convergence by combining imitation learning, and dynamically adjusts the exploration rate to adapt to changes in the ocean environment; a constraint layer is introduced through safety reinforcement learning to limit the action space to avoid violating bandwidth and energy constraints.

[0019] On the other hand, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps of the narrowband transmission scheduling method for multi-modal data of the marine satellite Internet of Things mentioned in any one of the above.

[0020] The beneficial effects of the present invention are at least: The present invention provides a narrowband transmission scheduling method and device for multi-modal data of the marine satellite Internet of Things, including: proposing a multi-objective joint optimization method based on weighted PAoI, energy consumption, and congestion, incorporating PAoI fused with data value weight, device energy consumption, and network congestion into a unified objective function, adaptively adjusting the weight coefficients of each optimization item, and balancing information freshness, energy efficiency, and network load while ensuring the transmission priority of key data, significantly improving the overall performance of the system; proposing an optimization problem solving method based on DQN and imitation learning, integrating a safety reinforcement learning framework, which not only reduces the training cost but also can stably generate an optimal scheduling strategy that meets multi-objective trade-offs in a complex and dynamic ocean environment, providing a theoretical feasibility and technical advancement guarantee for the multi-modal data transmission of narrowband satellite Internet of Things.

[0021] The additional advantages, objectives, and features of the present invention will be partially elaborated in the following description, and will become partially obvious to those of ordinary skill in the art after studying the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the specification and the drawings.

[0022] Those skilled in the art will understand that the purposes and advantages achievable by the present invention are not limited to those specifically described above, and the above and other purposes achievable by the present invention will be more clearly understood according to the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and do not limit the present invention. In the drawings: Figure 1 It is a schematic diagram of the steps of the narrowband transmission scheduling method for multi-modal data of the ocean satellite Internet of Things in an embodiment of the present invention.

[0024] Figure 2 It is a scenario diagram of the narrowband transmission of data of ocean observation and detection equipment based on the low-earth orbit satellite Internet in an embodiment of the present invention.

[0025] Figure 3 It is a technical roadmap of the narrowband transmission scheduling of multi-modal data of the ocean satellite Internet of Things in an embodiment of the present invention.

[0026] Figure 4 It is a schematic diagram of the dynamic transmission scheduling process of the ocean satellite Internet of Things based on DQN and imitation learning in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] To make the purposes, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. Here, the illustrative embodiments and descriptions thereof of the present invention are used to explain the present invention, but do not limit the present invention.

[0028] Here, it should also be noted that in order to avoid obscuring the present invention with unnecessary details, only the structures and / or processing steps closely related to the solution of the present invention are shown in the drawings, and other details less related to the present invention are omitted.

[0029] It should be emphasized that the term "including / comprising" when used herein refers to the presence of features, elements, steps, or components, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0030] Here, it should also be noted that if not otherwise specified, the term "connection" in this article can not only refer to a direct connection, but also represent an indirect connection with an intermediate.

[0031] In the following, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0032] To solve the problems of insufficient analysis of the spatio-temporal characteristics of existing multi-modal data, limited satellite communication link resources, unstable device energy supply, and the fact that existing transmission scheduling methods are not fully applicable to the complex dynamic environment of the marine satellite Internet of Things, the present invention provides a narrowband transmission scheduling method for multi-modal data in the marine satellite Internet of Things, as Figure 1 shown, the method includes the following steps S101~S106: Step S101: Construct a dynamic transmission model for multi-modal data of deep-sea and far-sea observation and detection devices. The model input includes multi-modal data collected by meteorological sensors, hydrological sensors, and target detection sensors in the marine satellite Internet of Things. Independent data queues are established for periodic meteorological and hydrological data and event-triggered target discovery, hydrological anomalies, and meteorological warning data, and the queue lengths are dynamically updated; in view of the bandwidth limitation of satellite narrowband communication, a bandwidth resource constraint model is constructed; spatio-temporal intelligent-driven data value evaluation is carried out, the LSTM is used to analyze the fluctuation characteristics in the time dimension, and the DBSCAN is combined to perform spatial clustering on deep-sea and far-sea hot spots, quantify the data value factors, and realize differential data scheduling.

[0033] Step S102: Design a congestion awareness and hierarchical congestion control strategy. Based on the bandwidth utilization rate, delay jitter, and queue length change rate, the network congestion index is comprehensively calculated, and the index weights are dynamically adjusted for scenarios such as bad weather, data transmission peak periods, and sudden data influxes to realize scenario-adaptive network congestion assessment; design a hierarchical congestion control strategy based on thresholds. In the low congestion mode, data is transmitted according to the data value. In the medium congestion mode, the single transmission volume is limited and high-value data is preferentially transmitted. In the high congestion mode, the transmission of low-value data is suspended and expired data is discarded.

[0034] Step S103: Information timeliness model and fairness transmission scheduling. The Age of Information (AoI) is the time difference from the packet generation time to the successful reception time, and the PAoI is the maximum AoI value within the packet transmission period. The data value weight factor is combined with the PAoI to construct a weighted PAoI index to realize the information timeliness evaluation of different value data; a multi-sensor PAoI fairness function integrating the data value weight is constructed based on the Jain fairness index to ensure the fairness of multi-sensor data transmission.

[0035] Step S104: Establish a dynamic energy evolution model for deep-sea and far-sea observation and detection devices. By combining the solar power supply characteristics and the day-night cycle changes, a dynamic energy acquisition model is established, and according to the transmission and sleep energy consumption characteristics of the sensors, an energy consumption model is designed, the battery energy constraint is introduced, the remaining energy is dynamically updated, and the energy fluctuation range is limited to ensure that the device meets the long-term energy budget balance.

[0036] Step S105: Jointly optimize information timeliness, energy consumption, and congestion, construct a multi-objective optimization function, and minimize the weighted sum of weighted PAoI, energy consumption, and congestion.

[0037] Step S106: Adopt a strategy that combines DQN and imitation learning, introduce safety reinforcement learning, model the state space, action space, safety constraints, and reward function through a Markov decision process, generate optimal transmission decisions including sensor transmission decision variables and the amount of transmitted data for each time slot, and achieve dynamic transmission scheduling.

[0038] In step S101, based on the generation rules of sensor data from ocean observation equipment, establish independent data queues for each sensor; dynamically update the data queue length according to the arrival and transmission processes of data packets; construct a bandwidth resource constraint model according to the bandwidth limitations of satellite narrowband communication modules; conduct data value evaluation based on spatio-temporal intelligence to provide a basis for differential data scheduling.

[0039] In the multi-sensor data queue model, based on the generation rules of sensor data from ocean observation equipment, establish independent data queues that match the number of sensors; periodic meteorological and hydrological observation data generate data packets at fixed time intervals, and the arrival volume follows a uniform distribution; event-triggered target discovery, hydrological anomalies, and meteorological warning data generate data packets according to a Poisson process, and the arrival volume .

[0040] In the dynamic update model of each sensor queue, according to the arrival and transmission processes of data packets, the formula for dynamically updating the data queue length is , where is the queue length of sensor at time , is the transmission decision variable, is the amount of data transmitted in a single transmission, is the amount of newly arrived data, is the maximum queue length to avoid storage overflow.

[0041] In the bandwidth resource constraint model, as Figure 2 shown, satellite narrowband communication modules are carried on devices such as drifting buoys, wave energy gliders, Argo buoys, and real-time moored buoys. The upper limit of the total bandwidth resource within a single time slot is , and the bandwidth requirement for a single transmission of sensor is . The narrowband communication bandwidth constraint formula is constructed as .

[0042] In the above-mentioned spatio-temporal intelligence-based data value evaluation model, the data value weight factors of each sensor are calculated by weighting according to the sensor data type, time characteristics, and spatial characteristics; spatio-temporal intelligence-based data value analysis is adopted. In the time dimension, based on LSTM, the fluctuation amplitude of marine environment data in the time dimension is perceived, and abnormal data changes are identified to obtain the marine environment fluctuation rating. , and the corresponding time weighting coefficient is ; in the spatial dimension, DBSCAN is used to cluster the high-precision navigation and positioning data in the spatial hot spot areas, and the regional division and pattern recognition of the maritime space are realized to achieve the spatial hot spot rating. , and the corresponding spatial weighting coefficient is ; through the dynamic data value weight factor , the differential service quality guarantee is realized. Exemplarily, if LSTM detects that the water temperature rises by 2°C within 1 hour, then the environmental fluctuation rating in the time dimension is , and the corresponding time weighting coefficient is ; if 10 ships are detected to gather in a certain area and are marked as hot spots by DBSCAN, the hot spot area rating in the spatial dimension is , and the corresponding time weighting coefficient is ; calculate the data value weight factor .

[0043] In step S102, the network congestion perception model is constructed by real-time monitoring of the bandwidth utilization rate, delay jitter, and queue length change rate, the network congestion index is calculated, the index weights are dynamically adjusted to adapt to the marine satellite Internet of Things scenario, and the transmission strategy is controlled by grading according to the threshold.

[0044] In the above-mentioned network congestion perception model, the congestion degree is dynamically quantified by real-time monitoring of the bandwidth utilization rate, delay jitter, and queue length change rate.

[0045] The satellite link bandwidth utilization rate is , which reflects the current channel load. Exemplarily, when the bandwidth utilization rate reaches 80%, it indicates that the channel is approaching saturation.

[0046] The difference between the current Round-Trip Time (RTT) and the minimum RTT characterizes the network fluctuation intensity , effectively identifying the unstable state of the link. Exemplarily, in bad weather it may increase from 10ms to 200ms.

[0047] The queue length change rate is , and the weighted average of the queue length change rates of each sensor can be obtained as , which characterizes the queue change trend.

[0048] Calculate the congestion index based on three metrics: bandwidth utilization, delay jitter, and the change rate of queue length , , exemplarily, the bandwidth utilization weight , the delay jitter weight , and the queue change rate weight .

[0049] In some embodiments, the weights are dynamically adjusted according to the characteristics of the marine environment: in bad weather, the atmospheric attenuation causes the signal quality to deteriorate, and the delay jitter is more sensitive. It is necessary to respond to the unstable link preferentially. Therefore, increase the weight; during the peak data transmission period, the channel load is close to saturation, and it is necessary to strictly control the bandwidth allocation to avoid congestion collapse. Therefore, increase the weight; when burst data floods in, the queue grows rapidly, and it is necessary to relieve the queue pressure preferentially to prevent overflow. Therefore, increase the weight.

[0050] In the threshold-based hierarchical congestion control strategy, according to the real-time congestion degree , exemplarily, design two congestion thresholds and ; when , in the low congestion mode, determine the data transmission priority according to the data value factor , and allow the maximum transmission volume ; when , in the medium congestion model, start traffic shaping, limit the upper limit of the data volume transmitted each time, and give priority to transmitting high-value data ; when , in the high congestion mode, suspend the transmission of low-value data . If the queue length , the queue is close to overflow, and discard the oldest data packet according to the timeliness.

[0051] In step S103, combine the data value weight and PAoI to construct a weighted PAoI to quantify the timeliness of multi-modal data, constrain the multi-sensor resource allocation based on the Jain fairness index, and design a fair transmission scheduling mechanism for multi-sensor data to balance the preferential transmission of high-value data and the basic service requirements of low-value data.

[0052] In the information timeliness modeling, quantify the timeliness difference of data with different values, and preferentially ensure the low-latency transmission of high-value data; the age of information AoI is an index to measure the information timeliness. The peak age of information PAoI of the th data packet of the sensor is denoted as , where is the data generation time; the average PAoI of the statistical system is calculated to reflect the overall timeliness. The data value weight of the sensor is combined with the PAoI, and the system average weighted PAoI is defined as .

[0053] In the multi-sensor data fairness transmission scheduling mechanism, the data collection periods, data volumes, and data values of multiple sensors are different. To quantify the timeliness fairness, combined with the data value weight factor, the Jain fairness index is improved, and a multi-sensor information timeliness fairness function is constructed. The formula is , to avoid high-value data monopolizing resources for a long time and ensure the basic transmission requirements of low-value data; let represent the timeliness fairness threshold. When , it indicates that the timeliness of sensor updates is in an unfair state. Otherwise, the timeliness of sensor updates is in a fair state; Exemplarily, the system fairness threshold is set as . If , the multi-sensor transmission does not meet the fairness constraint, and resources are forcibly allocated to low-priority sensors.

[0054] In step S104, since the far-sea observation and detection equipment can only operate relying on renewable green energy, and the complex marine environment cannot always guarantee sufficient and stable energy, the equipment has strict requirements for energy consumption. The deep-sea observation and detection equipment energy dynamic evolution model quantifies the direct impact of data transmission scheduling on the equipment's survival ability by coupling the randomness of energy acquisition, multi-mode energy consumption, and the hard constraint of battery energy storage, providing an energy dimension decision boundary for jointly optimizing weighted PAoI, energy consumption, and congestion; Based on solar power supply and day-night cycle, an energy acquisition model is established, combined with the transmission / sleep energy consumption model, and under the battery energy constraint, a device dynamic energy management mechanism is designed to ensure long-term energy balance.

[0055] In the energy acquisition model, the dynamic energy acquisition process of solar power supply is modeled to adapt to the day-night cycle and seasonal changes. Based on the characteristics of solar power supply, the energy acquisition process is modeled with an exponential distribution with as a parameter ; According to the day-night cycle , the energy acquisition model is adjusted, is the daylight duration, which is dynamically adjusted according to the season ; Exemplarily, the energy acquisition rate during daylight hours follows an exponential distribution of . The daylight duration in summer is 14 hours, and the daylight duration in winter is 10 hours. The specific situation is dynamically calibrated according to the geographical location. The season correction coefficient: summer ​, the total charging energy is increased by 20% in winter , the total charging energy is reduced by 15%.

[0056] In the energy consumption model, the energy consumption of sensor transmission and dormancy is quantified, that is, the sensor The energy consumption of a single transmission and the energy consumption of dormancy Meet , the long-term average energy consumption of the system is .

[0057] In the battery energy constraint, the upper and lower limits of the battery capacity are defined and , the remaining energy is dynamically updated , preventing overcharging / overdischarging of the battery and extending the device life.

[0058] In the dynamic energy management mechanism of the device, considering the battery life and usage stability of the deep-sea and far-sea observation equipment, in order to achieve long-term stable operation under energy fluctuations, the device needs to meet the long-term energy balance constraint. High-energy-consuming transmission tasks are preferentially executed during the day, and low-energy-consuming operations are restricted at night; for example, if , force it to enter the low-power mode and suspend data transmission.

[0059] As Figure 3 shown, steps S102~S104 are the specific steps for constructing the model of the narrowband transmission scheduling system for multi-modal data of the ocean satellite Internet of Things, and steps S105 and S106 are the specific steps for constructing and solving the multi-objective joint optimization problem.

[0060] In step S105, a multi-objective optimization problem that jointly optimizes weighted PAoI, energy consumption, and congestion is constructed.

[0061] In the construction of the joint optimization objective function, in the scenario where the deep-sea and far-sea observation equipment performs data backhaul based on the low-earth orbit satellite Internet, under the constraints of narrowband communication bandwidth, dynamic energy acquisition, and battery capacity, seek an optimal transmission scheduling plan for timely backhaul of high-value data, sustainable operation of device energy, and adaptive network congestion control, balance weighted PAoI, energy consumption, and congestion, construct a multi-objective joint optimization problem, and the objective function is to minimize the weighted sum of the long-term time-average system weighted PAoI, energy consumption, and congestion. The formula is , where is the energy consumption penalty coefficient, is the congestion penalty coefficient, which is used to adjust the balance of weighted PAoI, energy consumption, and congestion; the scheduling strategy is used to determine when to transmit the update information of the sensor , that is The design of needs to balance the immediate timeliness benefit, long-term system energy efficiency, and network congestion control under multiple constraints.

[0062] In step S106, DQN and imitation learning are adopted to generate the optimal dynamic transmission decision under safety constraints.

[0063] In the optimization problem solving method based on DQN and imitation learning, the data transmission scheduling problem is modeled as a Markov decision process, and the agent perceives the current state and executes actions by interacting with the environment; the state space includes the sensor queue length, the current congestion index, the current device energy state, the data value weight factor, and the weighted PAoI; the action space includes the transmission decision variable and the amount of data transmitted per single transmission ; combined with safety reinforcement learning, a safety constraint layer is introduced, and the restricted actions include bandwidth constraint and battery energy constraint; the reward function is the negative value of the joint optimization objective to minimize the weighted PAoI, energy consumption, and congestion penalty; In the combination strategy of DQN and imitation learning, the training efficiency and policy safety are improved by integrating expert experience and self-exploration. As Figure 4 shown, it includes the following steps: collecting historical expert transmission scheduling data to construct an expert data set containing state-action pairs; using supervised learning to pre-train the DQN network and initialize the Q-value function parameters to enable the model to quickly master the basic decision-making mode; in the online training stage, mixing the expert data and real-time interaction data in proportion and inputting them into the experience replay pool, combining the action probability distribution of imitation learning and the Q-value update of DQN to dynamically balance exploration and exploitation; adopting a decaying exploration rate mechanism. Exemplarily, the initial exploration rate is 0.5, and it decays by 0.1 every 1000 steps, gradually reducing the random exploration ratio, enabling the model to transition from relying on expert experience to self-optimization, and finally generating an efficient and safe transmission strategy in a complex dynamic marine environment.

[0064] Safety reinforcement learning ensures that the transmission decision complies with physical limitations and system stability requirements by introducing a constraint layer and a penalty mechanism. In the action selection stage, a two-level filtering mechanism is designed: the first level excludes the over-limit action combinations based on the bandwidth constraint; the second level restricts the risky operations based on the energy constraint. At the same time, a violation penalty term is embedded in the reward function, and the reward value is additionally deducted when the constraint is violated, guiding the agent to learn safe actions through negative feedback. Exemplarily, when the remaining energy of the device is lower than a certain value, the safety layer only allows low-energy-consuming operations to avoid device downtime. This mechanism takes effect throughout the training and inference stages to ensure the reliable operation of the system in a narrowband communication and energy-constrained environment.

[0065] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the narrowband transmission scheduling method for multi-modal data of the marine satellite Internet of Things are implemented.

[0066] Correspondingly to the above method, the present invention further provides a device, which includes a computer device. The computer device includes a processor and a memory. Computer instructions are stored in the memory, and the processor is configured to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0067] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the foregoing edge computing server deployment method. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0068] In summary, the present invention provides a narrowband transmission scheduling method and device for multimodal data in a marine satellite Internet of Things, including: proposing a multi-objective joint optimization method based on weighted PAoI, energy consumption, and congestion, incorporating PAoI integrating data value weights, device energy consumption, and network congestion into a unified objective function, adaptively adjusting the weight coefficients of each optimization term, and balancing information freshness, energy efficiency, and network load while ensuring the transmission priority of critical data, significantly improving the overall performance of the system; proposing an optimization problem solving method based on DQN and imitation learning, integrating a security reinforcement learning framework. This method not only reduces the training cost but also can stably generate an optimal scheduling strategy that meets multi-objective trade-offs in a complex and dynamic marine environment, providing a theoretical feasibility and technical advancement guarantee for the multimodal data transmission of narrowband satellite Internet of Things.

[0069] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Specifically, whether to implement in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, and so on. When implemented in software, the elements of the present invention are programs or code segments used to execute the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave on a transmission medium or a communication link.

[0070] It should be clear that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present invention is not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present invention.

[0071] In the present invention, features described and / or illustrated for one embodiment can be used in the same way or in a similar way in one or more other embodiments, and / or combined with the features of other embodiments or replace the features of other embodiments.

[0072] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the embodiments of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for scheduling narrowband transmission of multimodal data of ocean satellite Internet of Things, characterized in that: The method comprises the following steps: Construct a dynamic transmission model of multimodal data for deep-sea observation and detection equipment, whose input includes multimodal data collected by meteorological, hydrological and target monitoring sensors, establish independent queues for periodic data and event-triggered data, and dynamically update each queue; construct a bandwidth resource constraint model for satellite narrowband communication; adopt a data value evaluation mechanism that combines long short-term memory (LSTM) network time series data feature analysis with density clustering algorithm DBSCAN spatial clustering, quantify data value factors and realize differentiated scheduling; Design congestion awareness and hierarchical congestion control strategies, build scenario-adaptive network congestion index based on bandwidth utilization, delay jitter, and queue length change rate, and establish a three-level congestion control strategy: low congestion is transmitted according to data value, medium congestion is limited and high-value data is transmitted first, and high congestion is suspended. Low-value data transmission and expired data are discarded; Information timeliness model and fair transmission scheduling, construct the weighted peak information age PAoI indicator that integrates data value weights, and evaluate the information timeliness of data of different values; combine the Jain fairness index to construct a multi-sensor PAoI fairness function that integrates data value weights to ensure transmission fairness; Establish a dynamic energy evolution model for deep-sea observation and detection equipment, establish a dynamic energy acquisition model based on the seasonal and diurnal variation characteristics of solar power supply, design an energy consumption model based on transmission and dormancy energy consumption, and establish a long-term energy balance mechanism under battery energy constraints; Construct a multi-objective optimization problem that jointly optimizes weighted PAoI, energy consumption, and congestion; An optimization problem solving method based on DQN and imitation learning is adopted, and secure reinforcement learning is introduced to solve the optimal transmission strategy that meets the constraints through the Markov decision process.

2. The method for scheduling multimodal data narrowband transmission of ocean satellite Internet of Things according to claim 1 is characterized in that: The multi-modal data dynamic transmission model of deep-sea observation and detection equipment includes the following steps: In multi-queue modeling and queue dynamic update, based on the data generation rules of ocean observation and detection equipment sensors, an independent data queue matching the number of sensors is established; periodic meteorological and hydrological observation data are generated at fixed time intervals to generate data packets. Obey uniform distribution; event-triggered target discovery, hydrological anomaly and meteorological warning data generate data packets according to the Poisson process, and the arrival volume; build a multi-queue structure, and dynamically update the data queue length formula according to the data packet arrival and transmission process: ,in, For sensor At the moment The queue length, is the transmission decision variable, is the amount of data transmitted in a single time, is the amount of newly arrived data, is the maximum queue length to avoid storage overflow; In the satellite narrowband communication bandwidth constraint modeling, according to the bandwidth limitation of the satellite narrowband communication module carried by the device, the upper limit of the total bandwidth resource in a single time slot is ,sensor The bandwidth requirement for a single transmission is , construct the narrowband communication bandwidth constraint formula as ; In constructing the dynamic data value weight model, the value weight factor of each sensor data is calculated based on the sensor data type, time characteristics and spatial characteristics; the data value analysis based on spatiotemporal intelligence is adopted. In the time dimension, based on the LSTM network, the fluctuation amplitude of the marine environment data in the time dimension is perceived, abnormal data changes are identified, and the marine environment fluctuation rating is obtained. , the corresponding time weighting coefficient is In the spatial dimension, DBSCAN is used to cluster the spatial hotspots of high-precision navigation and positioning data, and regional division and pattern recognition of maritime space are performed to achieve spatial hotspot rating. , the corresponding spatial weighting coefficient is ; Through dynamic data value weighting factors , to achieve differentiated service quality assurance.

3. The method for scheduling narrowband transmission of multimodal data of ocean satellite Internet of Things according to claim 1 is characterized in that: The congestion perception and hierarchical congestion control strategy includes the following steps: Real-time monitoring of satellite link bandwidth utilization in the congestion-aware model , reflects the current channel load; calculates the difference between the current round-trip time RTT and the minimum RTT, characterizing the intensity of network fluctuations , effectively identify the unstable state of the link; queue length change rate , the weighted average of the change rate of each sensor queue length can be obtained , characterizes the queue change trend; calculates the congestion index based on three indicators: bandwidth utilization, delay jitter, and queue length change rate , , dynamically adjust the weight according to the characteristics of the ocean environment, and increase Weight, data transmission peak period increases Weight, increased when burst data flows in Weight; According to the real-time congestion level , design a hierarchical congestion control strategy; when When low congestion mode is used, the data value factor Determines data transmission priority and allows maximum transmission volume ;when , medium congestion model, start traffic shaping, limit the upper limit of single transmission data volume, and give priority to high-value data; when In high congestion mode, low-value data transmission is suspended. If the queue is close to overflow, the oldest data packet is discarded according to timeliness.

4. The method for scheduling multimodal data narrowband transmission of ocean satellite Internet of Things according to claim 1, characterized in that: The information timeliness model and fair transmission scheduling include the following steps: In information timeliness modeling, information age PAoI is an indicator to measure information timeliness. No. The peak information age PAoI of a data packet is denoted as ,in, is the transmission completion time, is the data generation time; the data value weight factor is introduced into the system average PAoI calculation to form the weighted PAoI indicator formula: ; In the PAoI-driven multi-sensor data fairness transmission scheduling optimization method, multiple sensors have different data collection cycles, data volumes, and data values. In order to quantify timeliness fairness, the Jain fairness index is introduced and combined with the data value weight factor to construct a multi-sensor information timeliness fairness function. The formula is: .

5. The method for scheduling multimodal data narrowband transmission of ocean satellite Internet of Things according to claim 1, characterized in that: The energy dynamic evolution model of the deep-sea observation and detection equipment includes the following steps: In the energy acquisition model, based on the characteristics of solar power supply, Modeling the energy harvesting process for an exponential distribution of parameters , according to the diurnal cycle Adjust the energy acquisition model, The length of daylight hours is adjusted dynamically with seasonal changes; In the energy consumption model, the sensor Single transmission consumption Sleep energy consumption satisfy , the long-term average energy consumption of the system is ; In battery energy constraints, define the upper and lower limits of battery capacity and , dynamically update the remaining energy ; Taking into account the battery life and usage stability of deep-sea observation and detection equipment, the energy fluctuation range is limited, and the equipment meets the long-term energy balance constraints.

6. The method for scheduling multimodal data narrowband transmission of ocean satellite Internet of Things according to claim 1, characterized in that: The multi-objective optimization problem of jointly optimizing weighted PAoI, energy consumption and congestion includes the following steps: In constructing the joint optimization objective function, in order to achieve the optimal trade-off between system information timeliness, energy consumption and congestion, a multi-objective joint optimization problem is constructed. The objective function is to minimize the weighted sum of the long-term average system weighted PAoI, energy consumption and congestion. The formula is: ,in, is the energy consumption penalty coefficient, is the congestion penalty coefficient, which is used to adjust the balance among weighted PAoI, energy consumption and congestion. DQN and imitation learning are used to solve the optimization problem.

7. The method for scheduling multimodal data narrowband transmission of ocean satellite Internet of Things according to claim 1, characterized in that: The optimization problem solving method based on DQN and imitation learning comprises the following steps: In the modeling of the interaction between the agent and the environment, the data transmission scheduling problem is modeled as a Markov decision process. The agent perceives the current state and performs actions by interacting with the environment. The state space includes the sensor queue length, the current congestion index, the current device energy state, the data value weight factor and the weighted PAoI. The action space includes the transmission decision variables And the amount of data transferred in a single time ; Combined with safety reinforcement learning, a safety constraint layer is introduced to restrict actions including bandwidth constraints and battery energy constraints; the reward function is the negative value of the joint optimization objective to minimize weighted PAoI, energy consumption and congestion penalty; In the combination strategy of DQN and imitation learning, demonstration data of expert transmission scheduling decisions, including historical state sequences and action sequences, are collected to construct an expert data set; the expert data set is used to pre-train the DQN network through supervised learning to initialize the Q-value function parameters; during the DQN training process, the expert action probability distribution obtained by imitation learning is combined with the samples in the experience replay mechanism to generate mixed training data; based on the current state and model convergence, the exploration rate parameters are dynamically adjusted to reduce the cost of invalid exploration; through the training model that combines DQN with imitation learning, DQN can quickly learn the action distribution that meets the constraints and adapt to the complex dynamic ocean environment.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images