Intelligent management method and system for offshore low-broadband data transmission

Abstract

The invention discloses an intelligent management method and system for offshore low-broadband data transmission, and the method comprises the following steps: S1, dividing an offshore communication network into a plurality of bandwidth sensitive intervals, collecting dynamic environment data, and marking a network operation mode state; s2, associating and matching the dynamic environment data with a network operation mode state; s3, generating a data transmission strategy group based on the network state feature set; s4, simulating a path selection behavior and generating a path load characteristic spectrum; s5, performing clustering processing on the path load characteristic spectrum to form a path strategy optimization library; s6, determining a data transmission strategy and path combination by using a double-library dynamic mapping mechanism; and S7, monitoring the network state in real time, and dynamically updating the path strategy optimization library and the network state feature set. According to the invention, the maritime communication bandwidth utilization efficiency and the transmission stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission, and in particular to an intelligent management method and system for low-bandwidth data transmission at sea. Background Art

[0002] With the continued development of the marine economy and maritime activities, the reliability and effectiveness of maritime communications are receiving increasing attention. However, due to the complexity and severity of the marine environment, particularly the long-standing limitations of communication coverage blind spots, low bandwidth, and poor link stability, existing low-bandwidth maritime communication technologies cannot fully meet the diverse needs of data transmission. In particular, they have many shortcomings in efficiently transmitting and managing data under the requirements of multi-tasking concurrency and high real-time performance.

[0003] Traditional maritime communication networks generally use fixed transmission protocols and simple bandwidth allocation strategies. The technical principle of this type of method is to allocate and manage communication resources based on fixed rules and static planning, and lacks the ability to perceive and dynamically adjust the network environment status in real time. For example, by pre-setting bandwidth allocation rules, link switching mechanisms and static routing tables to complete the allocation and scheduling of communication resources, this type of method is relatively simple to implement and relatively low in cost, and is suitable for scenarios where the network environment does not change frequently. However, when faced with a complex marine communication environment with drastic bandwidth fluctuations, unstable links or drastic changes in network load, the above-mentioned fixed strategies are often difficult to respond quickly to environmental changes, and are prone to reduced communication efficiency, increased data transmission delays and even severe packet loss. It is difficult to meet the needs of complex scenarios such as real-time ocean monitoring, maritime operation coordination, and emergency rescue communications.

[0004] In recent years, some advanced dynamic bandwidth management technologies have begun to be applied to maritime communications. For example, adaptive bandwidth adjustment methods based on conventional reinforcement learning or heuristic path optimization algorithms use partial perception of network status to optimize transmission strategies and paths to a certain extent. While these solutions have a certain degree of adaptability, they still lack real-time and accurate perception of environmental conditions, flexible path selection, and refined network load management, failing to balance the comprehensive performance of communication quality and resource utilization.

[0005] Therefore, how to provide an intelligent management method and system for low-bandwidth data transmission at sea is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One purpose of the present invention is to propose an intelligent management method for low-bandwidth data transmission at sea. The present invention has real-time network status perception and intelligent path optimization capabilities, which significantly improves data transmission stability and bandwidth utilization efficiency.

[0007] According to an embodiment of the present invention, a method for intelligent management of low-bandwidth data transmission at sea includes the following steps: S1. Divide the maritime communication network into multiple bandwidth-sensitive intervals. Use multi-source sensors to collect dynamic environmental data on bandwidth, link latency, and packet loss rate within each bandwidth-sensitive interval, and simultaneously mark the network operation mode status of each interval. S2. Correlate and match the dynamic environment data with the network operation mode status to obtain a network status feature set; S3. Based on the network status feature set, deep reinforcement learning is used to automatically generate multiple data transmission strategy groups corresponding to different bandwidth-sensitive intervals, and evaluation indicators are set based on data transmission delay and bandwidth utilization as parameters; S4. Based on the ant colony algorithm, simulate the path selection behavior of different data transmission strategy groups in each bandwidth-sensitive interval, record the actual load distribution data on each path, and generate the corresponding path load characteristic map; S5. Perform hierarchical clustering on the path load characteristic map to select the optimal load distribution path groups under different bandwidth-sensitive intervals and corresponding policy groups, forming a path policy optimization library. S6. During real-time data transmission, based on the specific data task type and priority, a dual-library dynamic mapping mechanism combining a path strategy optimization library and a network status feature set is used to determine and execute the optimal data transmission strategy and path combination; S7. Continuously monitor changes in network status after the data transmission strategy and path combination are executed, dynamically update the path strategy optimization library and network status feature set, and form a closed-loop optimization feedback mechanism.

[0008] Optionally, the S1 specifically includes: S11. Select all physical links in the maritime communication network as analysis objects. For each physical link, monitor and record in real time the interaction duration, number of signal round trips, data packet transmission success rate, and interaction event interval of each signal interaction event between the link and adjacent links, and establish a multi-dimensional link interaction behavior original dataset. S12. Continuously processing the original data set of multi-dimensional link interaction behavior using a preset weighted sliding time window to obtain a real-time characteristic indicator sequence of link interaction stability for each physical link in different time periods; S13. Constructing a comprehensive link sensitivity evaluation index based on the real-time characteristic indicator sequence of link interaction stability, wherein the comprehensive link sensitivity evaluation index includes a stability index under the link load level, a signal crosstalk interference coefficient, a link data packet success transmission rate, a number of data transmissions per unit time, and a link bandwidth occupancy rate; S14. Based on the comprehensive link sensitivity evaluation index, link sensitivity feature vectors of all physical links are calculated through multi-dimensional feature vector space mapping, and clustering is performed using a preset multi-dimensional feature space sensitivity level threshold matrix to obtain several link sensitivity level subsets with clear feature differentiation. S15. Analyze and calibrate each obtained link sensitivity level subset one by one, adopt similarity cross-coupling mapping, use the Euclidean distance of the link sensitivity feature vectors in each subset as the similarity measurement benchmark, divide multiple bandwidth sensitive intervals with consistent characteristics, and standardize and number and code the divided bandwidth sensitive intervals one by one; S16. Deploy multi-source sensors in the bandwidth-sensitive intervals, collect dynamic environmental data of bandwidth, link delay, and packet loss rate in each bandwidth-sensitive interval through the multi-source sensors, and synchronously mark the network operation mode status of each bandwidth-sensitive interval.

[0009] Optionally, the S2 specifically includes: S21, using non-uniformly spaced adaptive sampling, taking the instantaneous mutation point of the dynamic environment data change rate as the basis for timestamp marking, and generating a non-uniform time interval feature sequence; S22. Using the change rate and direction mutation point of the network operation mode state as the basis for division, adaptive window division with a variable step size is used to dynamically determine the start and end positions of each analysis window, thereby obtaining a divided network operation mode state sequence and forming a corresponding window index set; S23. Based on the non-uniform time interval feature sequence and the window index set, high-order difference processing is performed on the dynamic environment data and the network operation mode state in each analysis window to extract the trend change amplitude, change rate, and direction of the sequence in the window, thereby forming a corresponding window feature vector group; S24. Based on the window feature vector group, a nonlinear correlation calculation model between the dynamic environment data and the network operation mode state is constructed through nonlinear multi-scale kernel function mapping, and the correlation matching degree of the sequence within the window is calculated; S25, setting a multi-level dynamic threshold based on the correlation matching degree, selecting a window feature vector group whose correlation matching degree exceeds the multi-level dynamic threshold, and forming a high-correlation feature data subset with a level identifier; S26, performing incremental feature fusion on the highly correlated feature data subsets in a progressive manner, performing correlation redundancy testing and real-time feature updating after each level of fusion, and generating a fusion feature matrix of network state features; S27. Dynamically sort the fused feature matrices based on the correlation stability and redundancy constraints, and eliminate the fused feature matrices whose correlation stability is lower than the dynamic stability threshold to obtain a network state feature set.

[0010] Optionally, the S3 specifically includes: S31. Based on the network status feature set, for each bandwidth sensitive interval, a plurality of network status feature clusters with spatial similarity are formed through dynamic hierarchical clustering; S32. For the network status feature clusters, construct a multidimensional cross-mapping matrix between the network status feature clusters and the data transmission tasks based on the type, data size, real-time requirements, and delay sensitivity of the data transmission tasks. The matrix elements are task status sensitivity levels, which are divided according to the combination of task sensitivity to bandwidth utilization and transmission delay. S33. Based on a multi-dimensional cross-mapping matrix, deep reinforcement learning is used to establish a state-action decision mapping relationship between network state feature clusters and data transmission task sensitivity levels; S34. Using a comprehensive evaluation method for data transmission delay and bandwidth utilization, set a performance reward function in the deep reinforcement learning model to evaluate and quantify the performance of the data transmission strategy during the deep reinforcement learning model training process; S35. Through iterative training of the deep reinforcement learning model, with the performance reward value as the optimization target, the action combination is subjected to rolling real-time simulation training, and the bandwidth utilization and data transmission delay performance corresponding to each action combination are dynamically recorded to obtain an action combination performance trend data set; S36. Set a multi-dimensional dynamic threshold based on the fluctuation period, fluctuation amplitude, and convergence stability rate in the action combination performance trend data set, perform deep reinforcement learning evaluation on the performance stability of the action combination, and obtain an action combination with excellent performance stability and convergence speed; S37. Based on the action combination with excellent performance stability and convergence speed, a data transmission strategy group for each bandwidth sensitive interval is formed.

[0011] Optionally, the S4 specifically includes: S41. For the data transmission strategy group, the path selection behavior is decomposed into multiple micro-decision stages. The initial pheromone distribution coefficient is independently set for each micro-decision stage, and differentiated values ​​are assigned based on the historical usage count and historical load balancing degree of the path in the current stage. S42, for each path node link in the micro decision stage, calculate the link load perception factor in real time; S43, using the link load perception factor to capture the instantaneous state of the path link, obtaining instantaneous fluctuation data of the link load perception factor through high-frequency time scanning at each micro-decision stage, and generating a link load instantaneous fluctuation data sequence; S44. Build a real-time dynamic adjustment mechanism for path selection behavior based on the link load instantaneous fluctuation data sequence, and determine the real-time dynamic path selection probability by jointly evaluating the similarity of the instantaneous fluctuation data sequence and the deviation degree of the path selection historical behavior; S45. Based on the real-time dynamic path selection probability, iteratively update the pheromone concentration on each link stage by stage, record the actual load distribution data of each path link in real time, and form a dynamic data matrix of path load distribution; S46. Based on the dynamic data matrix, with the time dimension and space dimension of the link load fluctuation as coordinate axes, a path load feature map is generated through the path link time-space dual-dimensional feature association mapping.

[0012] Optionally, the S5 specifically includes: S51. Based on the path load characteristic map, for each bandwidth-sensitive interval and the corresponding policy group, a spatiotemporal fusion feature field of the path link load is constructed. The feature field uses the link load fluctuation amplitude and load duration as spatial dimensions, and the link load fluctuation period as temporal dimension. S52: using the spatiotemporal fusion feature field of the path link load and taking the change trend of the link load fluctuation characteristics as a basis, dynamically generating an evolution path diagram between each path link, marking the transition frequency of the path link between different states, and constructing a path link state transition sequence; S53. For the path link state transition sequence, a double correlation matching method is used to construct a cross-correlation mapping between the link load feature space and the path link evolution trend space to obtain the path link evolution feature of the load feature and the evolution trend; S54. According to the path link evolution characteristics, a multi-level dynamic similarity threshold is set, and hierarchical clustering of the path load characteristics and the path link evolution trends is performed step by step to generate a set of hierarchical evolution patterns of the path links; S55. Based on the set of hierarchical evolution patterns of path links, a path link evolution stability evaluation function is constructed with the path link state migration stability and the path load fluctuation trend convergence rate as dual indicators, and a path link evolution stability ranking sequence is determined; S56. Filter and sort the path links above a preset threshold one by one based on the path link evolution stability ranking sequence to form an optimal path group for load distribution; S57: Based on the load distribution optimal path group, establish a path strategy optimization library corresponding to the corresponding bandwidth sensitive interval and data transmission strategy group.

[0013] Optionally, the S6 specifically includes: S61, dynamically coupling the link load fluctuation trend of the path in the path strategy optimization library with the network operation status characteristics of the network status characteristic set in real time to form an instantaneous correlation mapping matrix between the path and the network status; S62. For the data task to be transmitted, a three-dimensional dynamic adaptation space is constructed between the task characteristics and the instantaneous association mapping matrix using the task type, scale, real-time requirements, and priority level. The position in the adaptation space is determined by the path load trend level, the network status characteristic level, and the data task priority level. S63: Based on the three-dimensional dynamic adaptation space, a path-state-task feature ternary collaborative matching process is implemented, and a real-time sliding window scanning mechanism is used in the adaptation space to determine the peak matching area between the path and the network state under the current task requirements; S64: Using the instantaneous matching characteristics within the matching degree peak area, construct a dynamic determination threshold for path link selection and data transmission strategy, and dynamically update the determination threshold according to the real-time matching degree; S65: Filtering the optimal matching path link and data transmission strategy combination in real time based on the dynamic determination threshold, identifying the path link load trend and network status feature level of the combination, and generating a corresponding data transmission strategy execution sequence in real time; S66. According to the data transmission strategy execution sequence, dynamic control of transmission rate, dynamic path selection and dynamic differential configuration of data packet priority are gradually implemented, and data transmission is performed in real time.

[0014] Optionally, an intelligent management system for low-bandwidth data transmission at sea includes the following modules: The real-time network status monitoring module is used to collect dynamic environmental data such as bandwidth, link delay, and packet loss rate within bandwidth-sensitive intervals, and mark the network operation mode status; A network state feature construction module is used to generate a network state feature set based on dynamic environment data and network operation mode status, and to generate a non-uniform time interval feature sequence and a network operation mode status sequence; The data transmission strategy generation module is used to automatically generate a data transmission strategy group based on the network status feature set through deep reinforcement learning methods, and construct a performance reward function based on data transmission delay and bandwidth utilization; The path link load characteristic analysis module is used to simulate path selection behavior based on the ant colony algorithm, record and analyze the load distribution data of path links in real time, and generate path load characteristic maps; The path strategy optimization library construction module is used to perform hierarchical clustering on the path load characteristic map, determine the optimal path group for load distribution, and form a path strategy optimization library; The dual-library dynamic mapping scheduling module is used to dynamically couple the path strategy optimization library with the network status feature set in real time, determine and execute data transmission strategies and path combinations in real time; The real-time feedback update module is used to record the actual link load, transmission delay and bandwidth utilization in real time after the data transmission strategy is executed, and dynamically update the network status feature set and path strategy optimization library based on the real-time recording results.

[0015] The beneficial effects of the present invention are: (1) The present invention divides the maritime communication network into bandwidth-sensitive intervals and collects dynamic environmental data of bandwidth, delay and packet loss rate in real time. It then constructs a network status feature set based on the network operation mode status, thereby achieving a fine perception of the communication environment, effectively improving the pertinence and response speed of the data transmission strategy, and enhancing the real-time scheduling capability of the system in the low-bandwidth environment at sea.

[0016] (2) The present invention constructs a multidimensional cross-mapping matrix based on the network status feature set and data task features, and uses a deep reinforcement learning method to generate a data transmission strategy group. Combined with an improved ant colony algorithm to simulate path selection behavior, the present invention significantly improves the balance performance between bandwidth utilization and data transmission delay, and exhibits better dynamic adaptability in multi-task concurrent scenarios.

[0017] (3) The present invention achieves a detailed analysis of the path load evolution trend by hierarchically clustering the path load feature map and constructing a spatiotemporal fusion feature field of the path link. It effectively solves the technical problems of inaccurate path selection and frequent load fluctuations in the existing technology, breaks through the bottleneck of traditional static routing, and realizes efficient load balancing of maritime communication links.

[0018] (4) The present invention couples the path strategy optimization library with the network status feature set in real time through a dual-library dynamic mapping mechanism, and continuously feeds back and updates during the data transmission process, thereby achieving closed-loop self-optimization of transmission strategy and path selection, thereby effectively improving the data transmission stability and overall communication quality in low-bandwidth environments at sea. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of an intelligent management method for low-bandwidth data transmission at sea proposed by the present invention. DETAILED DESCRIPTION

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0021] refer to Figure 1, a method for intelligent management of data transmission for low-bandwidth at sea, comprising the following steps: S1. Divide the maritime communication network into multiple bandwidth-sensitive intervals. Use multi-source sensors to collect dynamic environmental data on bandwidth, link latency, and packet loss rate within each bandwidth-sensitive interval, and simultaneously mark the network operation mode status of each interval. S2. Correlate and match the dynamic environment data with the network operation mode status to obtain a network status feature set; S3. Based on the network status feature set, deep reinforcement learning is used to automatically generate multiple data transmission strategy groups corresponding to different bandwidth-sensitive intervals, and evaluation indicators are set based on data transmission delay and bandwidth utilization as parameters; S4. Based on the ant colony algorithm, simulate the path selection behavior of different data transmission strategy groups in each bandwidth-sensitive interval, record the actual load distribution data on each path, and generate the corresponding path load characteristic map; S5. Perform hierarchical clustering on the path load characteristic map to select the optimal load distribution path groups under different bandwidth-sensitive intervals and corresponding policy groups, forming a path policy optimization library. S6. During real-time data transmission, based on the specific data task type and priority, a dual-library dynamic mapping mechanism combining a path strategy optimization library and a network status feature set is used to determine and execute the optimal data transmission strategy and path combination; S7. Continuously monitor changes in network status after the data transmission strategy and path combination are executed, dynamically update the path strategy optimization library and network status feature set, and form a closed-loop optimization feedback mechanism.

[0022] By dividing the maritime communication network into bandwidth-sensitive intervals and collecting dynamic environmental data such as bandwidth, delay and packet loss rate in real time, a network status feature set is constructed in combination with the network operation mode status. Based on this feature set, deep reinforcement learning is used to generate transmission strategies, and the ant colony algorithm is combined to perform path load feature analysis and hierarchical clustering, effectively forming a path strategy optimization library. The closed-loop optimization of real-time strategy and path combination is then achieved through a dual-library dynamic mapping mechanism, which adaptively responds to changes in network status, thereby improving bandwidth utilization, reducing transmission delay and enhancing the system's real-time adjustment capability in the low-bandwidth environment at sea.

[0023] In this embodiment, S1 specifically includes: S11. Select all physical links in the maritime communication network as analysis objects. For each physical link, monitor and record in real time the interaction duration, number of signal round trips, data packet transmission success rate, and interaction event interval of each signal interaction event between the link and adjacent links, and establish a multi-dimensional link interaction behavior original dataset. S12. Continuously processing the original data set of multi-dimensional link interaction behavior using a preset weighted sliding time window to obtain a real-time characteristic indicator sequence of link interaction stability for each physical link in different time periods; S13. Constructing a comprehensive link sensitivity evaluation index based on the real-time characteristic indicator sequence of link interaction stability, wherein the comprehensive link sensitivity evaluation index includes a stability index under the link load level, a signal crosstalk interference coefficient, a link data packet success transmission rate, a number of data transmissions per unit time, and a link bandwidth occupancy rate; S14. Based on the comprehensive link sensitivity evaluation index, link sensitivity feature vectors of all physical links are calculated through multi-dimensional feature vector space mapping, and clustering is performed using a preset multi-dimensional feature space sensitivity level threshold matrix to obtain several link sensitivity level subsets with clear feature differentiation. S15. Analyze and calibrate each obtained link sensitivity level subset one by one, adopt similarity cross-coupling mapping, use the Euclidean distance of the link sensitivity feature vectors in each subset as the similarity measurement benchmark, divide multiple bandwidth sensitive intervals with consistent characteristics, and standardize and number and code the divided bandwidth sensitive intervals one by one; S16. Deploy multi-source sensors in the bandwidth-sensitive intervals, collect dynamic environmental data of bandwidth, link delay, and packet loss rate in each bandwidth-sensitive interval through the multi-source sensors, and synchronously mark the network operation mode status of each bandwidth-sensitive interval.

[0024] By collecting multi-dimensional interactive behavior data of all physical links in the maritime communication network and continuously processing them using a weighted sliding time window, the link sensitivity is graded and finely divided by combining the link interaction stability feature sequence and comprehensive evaluation indicators. Multi-source sensors are deployed in the bandwidth-sensitive interval to synchronously mark the network operation mode status, achieving fine-grained perception and dynamic partition management of different areas of the network, thereby providing high-precision basic data support for subsequent data transmission strategies and path optimization.

[0025] In this embodiment, S2 specifically includes: S21, using non-uniformly spaced adaptive sampling, taking the instantaneous mutation point of the dynamic environment data change rate as the basis for timestamp marking, and generating a non-uniform time interval feature sequence; S22. Using the change rate and direction mutation point of the network operation mode state as the basis for division, adaptive window division with a variable step size is used to dynamically determine the start and end positions of each analysis window, thereby obtaining a divided network operation mode state sequence and forming a corresponding window index set; S23. Based on the non-uniform time interval feature sequence and the window index set, high-order difference processing is performed on the dynamic environment data and the network operation mode state in each analysis window to extract the trend change amplitude, change rate, and direction of the sequence in the window, thereby forming a corresponding window feature vector group; S24. Based on the window feature vector group, a nonlinear correlation calculation model between the dynamic environment data and the network operation mode state is constructed through nonlinear multi-scale kernel function mapping, and the correlation matching degree of the sequence within the window is calculated; S25, setting a multi-level dynamic threshold based on the correlation matching degree, selecting a window feature vector group whose correlation matching degree exceeds the multi-level dynamic threshold, and forming a high-correlation feature data subset with a level identifier; S26, performing incremental feature fusion on the highly correlated feature data subsets in a progressive manner, performing correlation redundancy testing and real-time feature updating after each level of fusion, and generating a fusion feature matrix of network state features; S27. Dynamically sort the fused feature matrices based on the correlation stability and redundancy constraints, and eliminate the fused feature matrices whose correlation stability is lower than the dynamic stability threshold to obtain a network state feature set.

[0026] By accurately dividing the network operation mode state sequence based on non-uniformly spaced adaptive sampling and variable step-size adaptive windows, and using high-order differences and nonlinear multi-scale kernel function mapping to construct a correlation calculation model between dynamic environmental data and network status, we achieve a deep coupling analysis of network status change trends and dynamic environmental data, thereby obtaining a more accurate and stable network status feature set.

[0027] In this embodiment, S3 specifically includes: S31. Based on the network status feature set, for each bandwidth sensitive interval, a plurality of network status feature clusters with spatial similarity are formed through dynamic hierarchical clustering; S32. For the network status feature clusters, construct a multidimensional cross-mapping matrix between the network status feature clusters and the data transmission tasks based on the type, data size, real-time requirements, and delay sensitivity of the data transmission tasks. The matrix elements are task status sensitivity levels, which are divided according to the combination of task sensitivity to bandwidth utilization and transmission delay. S33. Based on a multi-dimensional cross-mapping matrix, deep reinforcement learning is used to establish a state-action decision mapping relationship between network state feature clusters and data transmission task sensitivity levels; S34. Use a comprehensive evaluation method for data transmission delay and bandwidth utilization to set a performance reward function in the deep reinforcement learning model: used to evaluate and quantify the performance of the data transmission strategy during the deep reinforcement learning model training process.

[0028] ; in, Indicates the The performance reward value of the action combination, For the The action combination is in The actual bandwidth value used during the simulation period, For the The action combination is in The total bandwidth value available during the simulation period, Indicates the The action combination is in The data transmission delay within the simulation period is is the total number of simulation test periods, and is the weight coefficient of bandwidth utilization and data transmission delay, and satisfies ; For comprehensive quantification The bandwidth utilization and data transmission delay performance of the action combination in the rolling simulation process are first analyzed. Action combinations in each simulation period Sum the actual bandwidth usage during the simulation period and divide it by the total number of simulation periods , to reflect the bandwidth utilization efficiency; then the data transmission delay in each period Sum and divide , to characterize the delay burden; finally, the coefficient and The weighted subtraction of the two is used to obtain the comprehensive score , thereby guiding the strategy selection to optimize towards high bandwidth utilization and low latency in reinforcement learning training.

[0029] S35. Through iterative training of the deep reinforcement learning model, with the performance reward value as the optimization target, the action combination is subjected to rolling real-time simulation training, and the bandwidth utilization and data transmission delay performance corresponding to each action combination are dynamically recorded to obtain an action combination performance trend data set; S36. Set a multi-dimensional dynamic threshold based on the fluctuation period, fluctuation amplitude, and convergence stability rate in the action combination performance trend data set, perform deep reinforcement learning evaluation on the performance stability of the action combination, and obtain an action combination with excellent performance stability and convergence speed; S37. Based on the action combination with excellent performance stability and convergence speed, a data transmission strategy group for each bandwidth sensitive interval is formed.

[0030] By implementing dynamic hierarchical clustering on the network state feature set and constructing a multidimensional cross-mapping matrix, combining deep reinforcement learning to establish a state-action decision mapping relationship, and building a performance reward function based on bandwidth utilization and data transmission delay, the performance trend of the strategy-path combination is accurately learned in iterative simulation training, achieving fine optimization of the data transmission strategy in the low-bandwidth network environment at sea, significantly improving the accuracy of strategy selection and the stability of transmission performance.

[0031] In this embodiment, the S4 specifically includes: S41. For the data transmission strategy group, the path selection behavior is decomposed into multiple micro-decision stages. The initial pheromone distribution coefficient is independently set for each micro-decision stage, and differentiated values ​​are assigned based on the historical usage count and historical load balancing degree of the path in the current stage. S42. Calculate the link load perception factor in real time for each path node link in each micro-decision stage: ; in, For the moment Path Node To Node The link load perception factor reflects the actual load status of the link in real time. For the moment Path Node To Node The communication capacity occupied by the link, specifically the ratio of the actual amount of data transmitted to the maximum amount of data carried by the link, For the moment Path Node To Node The average delay of the link actually transmitting data packets, For the moment Path Node To Node The current available bandwidth capacity of the link, is the weight factor between link load occupied capacity and link delay; The formula is used to calculate the link load perception factor of each path node in each micro-strategy stage , to evaluate the current load conditions of the path links. Represents a path node To Node At the moment The link load status reflects the actual load status of the link; Path node To Node At the moment The data transmission delay represents the delay burden of the link. For the moment The available bandwidth of the link reflects the bandwidth resource status of the link. The formula is weighted summation, where is the weight coefficient between bandwidth and delay, satisfying , taking into account the load and delay factors of the link, the load perception factor of the path link is obtained .

[0032] S43, using the link load perception factor to capture the instantaneous state of the path link, obtaining instantaneous fluctuation data of the link load perception factor through high-frequency time scanning at each micro-decision stage, and generating a link load instantaneous fluctuation data sequence; S44. A real-time dynamic adjustment mechanism for path selection behavior is established based on the link load instantaneous fluctuation data sequence. The real-time dynamic path selection probability is determined by jointly evaluating the similarity of the instantaneous fluctuation data sequence and the deviation degree of the path selection historical behavior: ; in, For the moment At the path node Select Node As the path selection probability of the next transmission node, Indicates time Path Node To Node The pheromone concentration on 、 Represent the weight coefficients of pheromone concentration and link load perception factor, For the moment Path Node To Node Sensitivity adjustment factor between the historical path selection frequency and the current path selection frequency: ; in, For the moment Previous historical cumulative path nodes To Node Number of path selections, For the moment Nodes in the current decision stage To Node The cumulative number of path selections; The formula is used to determine the dynamic probability of path selection by jointly evaluating the similarity of instantaneous fluctuation data series and the historical deviation of path selection. . Indicates time At the path node Select Node The probability of path selection as the next transmission node. This probability is obtained by comprehensively considering the pheromone concentration of the path node. and link load perception factor The influence of weight coefficient and The weights of pheromone concentration and load perception factor are controlled, combined with the historical path selection frequency Further adjustments.

[0033] Specifically, the pheromone concentration Represents a path node To Node At the moment The historical selection intensity reflects the attractiveness of the path; is the link load perception factor, which indicates the link load condition; This is a factor influencing the historical frequency of path selection, indicating path selection preference. By comprehensively considering various factors, the probability of path selection is dynamically adjusted to optimize path selection and load balancing, ensuring network transmission efficiency and stability.

[0034] S45. Based on the real-time dynamic path selection probability, iteratively update the pheromone concentration on each link stage by stage, record the actual load distribution data of each path link in real time, and form a dynamic data matrix of path load distribution; S46. Based on the dynamic data matrix, with the time dimension and space dimension of the link load fluctuation as coordinate axes, a path load feature map is generated through the path link time-space dual-dimensional feature association mapping.

[0035] By decomposing the path selection behavior into micro-decision-making stages and setting the pheromone distribution coefficient, the path links are captured in an instantaneous state and analyzed for high-frequency fluctuations according to the link load perception factor. Combined with the historical selection sensitivity, the path selection probability is adjusted in real time. This achieves a refined characterization and dynamic adaptation of the link load distribution, effectively improving the accuracy of path selection and the network load balancing capability.

[0036] In this embodiment, the S5 specifically includes: S51. Based on the path load characteristic map, for each bandwidth-sensitive interval and the corresponding policy group, a spatiotemporal fusion feature field of the path link load is constructed. The feature field uses the link load fluctuation amplitude and load duration as spatial dimensions, and the link load fluctuation period as temporal dimension. S52: using the spatiotemporal fusion feature field of the path link load and taking the change trend of the link load fluctuation characteristics as a basis, dynamically generating an evolution path diagram between each path link, marking the transition frequency of the path link between different states, and constructing a path link state transition sequence; S53. For the path link state transition sequence, a double correlation matching method is used to construct a cross-correlation mapping between the link load feature space and the path link evolution trend space to obtain the path link evolution feature of the load feature and the evolution trend; S54. According to the path link evolution characteristics, a multi-level dynamic similarity threshold is set, and hierarchical clustering of the path load characteristics and the path link evolution trends is performed step by step to generate a set of hierarchical evolution patterns of the path links; S55. Based on the set of hierarchical evolution patterns of path links, a path link evolution stability evaluation function is constructed with the path link state migration stability and the path load fluctuation trend convergence rate as dual indicators, and a path link evolution stability ranking sequence is determined; S56. Filter and sort the path links above a preset threshold one by one based on the path link evolution stability ranking sequence to form an optimal path group for load distribution; S57: Based on the load distribution optimal path group, establish a path strategy optimization library corresponding to the corresponding bandwidth sensitive interval and data transmission strategy group.

[0037] By constructing a spatiotemporal fusion feature field of path link loads and dynamically generating an evolutionary path diagram, combined with dual correlation matching across space and evolutionary trends and a multi-level clustering hierarchical evolutionary model, we can achieve accurate analysis of path load fluctuation trends and state migration characteristics, effectively screen out the optimal load distribution path group, and form an optimization library corresponding to bandwidth-sensitive intervals and strategy groups, thereby improving the accuracy and stability of path selection in low-bandwidth maritime environments.

[0038] In this embodiment, S6 specifically includes: S61, dynamically coupling the link load fluctuation trend of the path in the path strategy optimization library with the network operation status characteristics of the network status characteristic set in real time to form an instantaneous correlation mapping matrix between the path and the network status; S62. For the data task to be transmitted, a three-dimensional dynamic adaptation space is constructed between the task characteristics and the instantaneous association mapping matrix using the task type, scale, real-time requirements, and priority level. The position in the adaptation space is determined by the path load trend level, the network status characteristic level, and the data task priority level. S63: Based on the three-dimensional dynamic adaptation space, a path-state-task feature ternary collaborative matching process is implemented, and a real-time sliding window scanning mechanism is used in the adaptation space to determine the peak matching area between the path and the network state under the current task requirements; S64: Using the instantaneous matching characteristics within the matching degree peak area, construct a dynamic determination threshold for path link selection and data transmission strategy, and dynamically update the determination threshold according to the real-time matching degree; S65: Filtering the optimal matching path link and data transmission strategy combination in real time based on the dynamic determination threshold, identifying the path link load trend and network status feature level of the combination, and generating a corresponding data transmission strategy execution sequence in real time; S66. According to the data transmission strategy execution sequence, dynamic control of transmission rate, dynamic path selection and dynamic differential configuration of data packet priority are gradually implemented, and data transmission is performed in real time.

[0039] By real-time coupling of path load fluctuation trends and network operation status characteristics, and implementing path-state-task collaborative matching and sliding window scanning in a three-dimensional dynamic adaptation space, accurate positioning of the peak area of ​​path and state matching under task requirements is achieved; combined with dynamic judgment thresholds, the optimal path and transmission strategy combination is screened in real time, and rate control and priority differential configuration are continuously implemented during execution, thereby improving the accuracy and stability of data transmission in complex low-bandwidth maritime environments.

[0040] An intelligent management system for low-bandwidth data transmission at sea, comprising the following modules: The real-time network status monitoring module is used to collect dynamic environmental data such as bandwidth, link delay, and packet loss rate within bandwidth-sensitive intervals, and mark the network operation mode status; A network state feature construction module is used to generate a network state feature set based on dynamic environment data and network operation mode status, and to generate a non-uniform time interval feature sequence and a network operation mode status sequence; The data transmission strategy generation module is used to automatically generate a data transmission strategy group based on the network status feature set through deep reinforcement learning methods, and construct a performance reward function based on data transmission delay and bandwidth utilization; The path link load characteristic analysis module is used to simulate path selection behavior based on the ant colony algorithm, record and analyze the load distribution data of path links in real time, and generate path load characteristic maps; The path strategy optimization library construction module is used to perform hierarchical clustering on the path load characteristic map, determine the optimal path group for load distribution, and form a path strategy optimization library; The dual-library dynamic mapping scheduling module is used to dynamically couple the path strategy optimization library with the network status feature set in real time, determine and execute data transmission strategies and path combinations in real time; The real-time feedback update module is used to record the actual link load, transmission delay and bandwidth utilization in real time after the data transmission strategy is executed, and dynamically update the network status feature set and path strategy optimization library based on the real-time recording results.

[0041] By organically integrating modules such as real-time network status monitoring, feature construction, strategy generation, path load analysis, strategy optimization library construction, dual-library dynamic mapping scheduling and real-time feedback update, collaborative closed-loop management of network status and transmission requirements in low-bandwidth maritime environments is achieved, thereby improving the accuracy of data transmission strategy selection and continuously optimizing path load balancing, enhancing the system's real-time adaptability and overall communication performance stability.

[0042] Example 1: In order to verify the feasibility of the present invention in implementation, this embodiment applies the present invention to the real-time data transmission optimization task in a low-bandwidth data transmission intelligent management system at sea. This task mainly involves improving the data transmission rate and reducing the transmission delay by optimizing the transmission path in a low-bandwidth maritime communication environment, while ensuring the stability and efficiency of data transmission. In this application scenario, the problems faced by traditional methods are mainly reflected in the allocation and optimization of bandwidth resources, the real-time performance of path selection, and the dynamic adaptability of data transmission strategies. Existing methods rely on manually setting transmission strategies or selecting paths based on static network status. This approach often responds slowly to dynamic changes in the maritime environment, cannot efficiently utilize bandwidth resources, results in low data transmission efficiency, and is easily affected by fluctuations in network status.

[0043] In order to solve this problem, the present invention combines deep reinforcement learning with ant colony algorithm to dynamically adjust the data transmission strategy in real time according to the changes in the low-bandwidth communication environment at sea, ensuring that path selection and bandwidth allocation can make optimal decisions based on the real-time network load and bandwidth-sensitive intervals. During the implementation process, the system first collects network status data in real time through multi-source sensors, including information such as bandwidth, link delay, packet loss rate, etc., and combines it with the deep reinforcement learning model for real-time training to generate the optimal data transmission strategy. Subsequently, based on these strategies and the path load characteristic map, the path selection behavior is simulated through the ant colony algorithm to optimize the path selection and bandwidth resource allocation in real time. The system dynamically adjusts the data transmission strategy to ensure the path stability and transmission efficiency during the data transmission process, and ultimately significantly improves the transmission rate and stability of the low-bandwidth communication system at sea.

[0044] In practical applications, researchers first set up a test scenario in a low-bandwidth offshore environment to simulate bandwidth resource fluctuations and conducted experiments on typical data tasks. Using the intelligent management method presented in this paper, the system automatically generated transmission strategies based on real-time network status characteristics and determined the optimal path combination through a dual-library dynamic mapping mechanism. During the experiment, the system automatically updated the path strategy optimization library and network status feature set based on a dynamic feedback mechanism, achieving real-time optimization and adaptive adjustment of data transmission.

[0045] Experimental results show that after adopting the intelligent management method of the present invention, the data transmission rate increased by 23.4%, the transmission delay decreased by 15.8%, and the system showed strong adaptability and stability in a frequently fluctuating bandwidth environment. The specific data is shown in Table 1: Table 1: Performance comparison of low-bandwidth data transmission at sea before and after optimization

[0046] Table 1 shows a comparison of the original and optimized performance, including data transmission rate, transmission delay, bandwidth utilization, and system stability, under different test cases. As can be seen, in all test cases, the optimized system achieved a significant increase in data transmission rate, with an average increase of 24.2%. Transmission delay was also significantly reduced, with an average reduction of 17.0%. System stability was significantly improved, with an average increase of 22.1%. These results demonstrate that the proposed method can significantly improve the efficiency and reliability of low-bandwidth data transmission at sea, demonstrating strong dynamic adaptability, particularly in unstable bandwidth environments.

[0047] Further experimental data also demonstrated the system's high adaptability in diverse maritime environments. Even in environments with significant bandwidth fluctuations, the system was able to rapidly respond to network changes, automatically adjusting data transmission strategies while maintaining a low error rate and high transmission efficiency. For example, within a single test cycle, the system successfully identified and automatically corrected six transmission bottlenecks caused by sudden network load fluctuations, ultimately achieving a 94% data transmission success rate after optimization.

[0048] Comprehensive analysis of the tabular data reveals that this invention, through its combination of deep reinforcement learning and an ant colony algorithm, can dynamically adapt to bandwidth resource fluctuations in low-bandwidth maritime environments and precisely optimize path selection and bandwidth allocation, demonstrating significant advantages in increasing data transmission rates, reducing transmission latency, and improving bandwidth utilization. In this low-bandwidth maritime environment, the invention effectively addresses the inability of traditional methods to cope with bandwidth resource fluctuations and optimize transmission strategies, providing reliable support for the intelligent management of maritime communication systems.

[0049] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for intelligent management of low-bandwidth data transmission at sea, characterized in that: The steps include: S1. Divide the maritime communication network into multiple bandwidth-sensitive intervals. Use multi-source sensors to collect dynamic environmental data on bandwidth, link latency, and packet loss rate within each bandwidth-sensitive interval, and simultaneously mark the network operation mode status of each interval. S2. Correlate and match the dynamic environment data with the network operation mode status to obtain a network status feature set; S3. Based on the network status feature set, deep reinforcement learning is used to automatically generate multiple data transmission strategy groups corresponding to different bandwidth-sensitive intervals, and evaluation indicators are set based on data transmission delay and bandwidth utilization as parameters; S4. Based on the ant colony algorithm, simulate the path selection behavior of different data transmission strategy groups in each bandwidth-sensitive interval, record the actual load distribution data on each path, and generate the corresponding path load characteristic map; S5. Perform hierarchical clustering on the path load characteristic map to select the optimal load distribution path groups under different bandwidth-sensitive intervals and corresponding policy groups, forming a path policy optimization library. S6. During real-time data transmission, based on the specific data task type and priority, a dual-library dynamic mapping mechanism combining a path strategy optimization library and a network status feature set is used to determine and execute the optimal data transmission strategy and path combination; S7. Continuously monitor changes in network status after the data transmission strategy and path combination are executed, dynamically update the path strategy optimization library and network status feature set, and form a closed-loop optimization feedback mechanism.

2. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: Said S1 specifically includes: S11. Select all physical links in the maritime communication network as analysis objects. For each physical link, monitor and record in real time the interaction duration, number of signal round trips, data packet transmission success rate, and interaction event interval of each signal interaction event between the link and adjacent links, and establish a multi-dimensional link interaction behavior original dataset. S12. Continuously processing the original data set of multi-dimensional link interaction behavior using a preset weighted sliding time window to obtain a real-time characteristic indicator sequence of link interaction stability for each physical link in different time periods; S13. Constructing a comprehensive link sensitivity evaluation index based on the real-time characteristic indicator sequence of link interaction stability, wherein the comprehensive link sensitivity evaluation index includes a stability index under the link load level, a signal crosstalk interference coefficient, a link data packet success transmission rate, a number of data transmissions per unit time, and a link bandwidth occupancy rate; S14. Based on the comprehensive link sensitivity evaluation index, link sensitivity feature vectors of all physical links are calculated through multi-dimensional feature vector space mapping, and clustering is performed using a preset multi-dimensional feature space sensitivity level threshold matrix to obtain several link sensitivity level subsets with clear feature differentiation. S15. Analyze and calibrate each obtained link sensitivity level subset one by one, adopt similarity cross-coupling mapping, use the Euclidean distance of the link sensitivity feature vectors in each subset as the similarity measurement benchmark, divide multiple bandwidth sensitive intervals with consistent characteristics, and standardize and number and code the divided bandwidth sensitive intervals one by one; S16. Deploy multi-source sensors in the bandwidth-sensitive intervals, collect dynamic environmental data of bandwidth, link delay, and packet loss rate in each bandwidth-sensitive interval through the multi-source sensors, and synchronously mark the network operation mode status of each bandwidth-sensitive interval.

3. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: The S2 specifically includes: S21, using non-uniformly spaced adaptive sampling, taking the instantaneous mutation point of the dynamic environment data change rate as the basis for timestamp marking, and generating a non-uniform time interval feature sequence; S22. Using the change rate and direction mutation point of the network operation mode state as the basis for division, adaptive window division with a variable step size is used to dynamically determine the start and end positions of each analysis window, thereby obtaining a divided network operation mode state sequence and forming a corresponding window index set; S23. Based on the non-uniform time interval feature sequence and the window index set, high-order difference processing is performed on the dynamic environment data and the network operation mode state in each analysis window to extract the trend change amplitude, change rate, and direction of the sequence in the window, thereby forming a corresponding window feature vector group; S24. Based on the window feature vector group, a nonlinear correlation calculation model between the dynamic environment data and the network operation mode state is constructed through nonlinear multi-scale kernel function mapping, and the correlation matching degree of the sequence within the window is calculated; S25, setting a multi-level dynamic threshold based on the correlation matching degree, selecting a window feature vector group whose correlation matching degree exceeds the multi-level dynamic threshold, and forming a high-correlation feature data subset with a level identifier; S26, performing incremental feature fusion on the highly correlated feature data subsets in a progressive manner, performing correlation redundancy testing and real-time feature updating after each level of fusion, and generating a fusion feature matrix of network state features; S27. Dynamically sort the fused feature matrices based on the correlation stability and redundancy constraints, and eliminate the fused feature matrices whose correlation stability is lower than the dynamic stability threshold to obtain a network state feature set.

4. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: The S3 specifically includes: S31. Based on the network status feature set, for each bandwidth sensitive interval, a plurality of network status feature clusters with spatial similarity are formed through dynamic hierarchical clustering; S32. For the network status feature clusters, construct a multidimensional cross-mapping matrix between the network status feature clusters and the data transmission tasks based on the type, data size, real-time requirements, and delay sensitivity of the data transmission tasks. The matrix elements are task status sensitivity levels, which are divided according to the combination of task sensitivity to bandwidth utilization and transmission delay. S33. Based on a multi-dimensional cross-mapping matrix, deep reinforcement learning is used to establish a state-action decision mapping relationship between network state feature clusters and data transmission task sensitivity levels; S34. Using a comprehensive evaluation method for data transmission delay and bandwidth utilization, set a performance reward function in the deep reinforcement learning model to evaluate and quantify the performance of the data transmission strategy during the deep reinforcement learning model training process; S35. Through iterative training of the deep reinforcement learning model, with the performance reward value as the optimization target, the action combination is subjected to rolling real-time simulation training, and the bandwidth utilization and data transmission delay performance corresponding to each action combination are dynamically recorded to obtain an action combination performance trend data set; S36. Set a multi-dimensional dynamic threshold based on the fluctuation period, fluctuation amplitude, and convergence stability rate in the action combination performance trend data set, perform deep reinforcement learning evaluation on the performance stability of the action combination, and obtain an action combination with excellent performance stability and convergence speed; S37. Based on the action combination with excellent performance stability and convergence speed, a data transmission strategy group for each bandwidth sensitive interval is formed.

5. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: The S4 specifically includes: S41. For the data transmission strategy group, the path selection behavior is decomposed into multiple micro-decision stages. The initial pheromone distribution coefficient is independently set for each micro-decision stage, and differentiated values ​​are assigned based on the historical usage count and historical load balancing degree of the path in the current stage. S42, for each path node link in the micro decision stage, calculate the link load perception factor in real time; S43, using the link load perception factor to capture the instantaneous state of the path link, obtaining instantaneous fluctuation data of the link load perception factor through high-frequency time scanning at each micro-decision stage, and generating a link load instantaneous fluctuation data sequence; S44. Build a real-time dynamic adjustment mechanism for path selection behavior based on the link load instantaneous fluctuation data sequence, and determine the real-time dynamic path selection probability by jointly evaluating the similarity of the instantaneous fluctuation data sequence and the deviation degree of the path selection historical behavior; S45. Based on the real-time dynamic path selection probability, iteratively update the pheromone concentration on each link stage by stage, record the actual load distribution data of each path link in real time, and form a dynamic data matrix of path load distribution; S46. Based on the dynamic data matrix, with the time dimension and space dimension of the link load fluctuation as coordinate axes, a path load feature map is generated through the path link time-space dual-dimensional feature association mapping.

6. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: The S5 specifically includes: S51. Based on the path load characteristic map, for each bandwidth-sensitive interval and the corresponding policy group, a spatiotemporal fusion feature field of the path link load is constructed. The feature field uses the link load fluctuation amplitude and load duration as spatial dimensions, and the link load fluctuation period as temporal dimension. S52: using the spatiotemporal fusion feature field of the path link load and taking the change trend of the link load fluctuation characteristics as a basis, dynamically generating an evolution path diagram between each path link, marking the transition frequency of the path link between different states, and constructing a path link state transition sequence; S53. For the path link state transition sequence, a double correlation matching method is used to construct a cross-correlation mapping between the link load feature space and the path link evolution trend space to obtain the path link evolution feature of the load feature and the evolution trend; S54. According to the path link evolution characteristics, a multi-level dynamic similarity threshold is set, and hierarchical clustering of the path load characteristics and the path link evolution trends is performed step by step to generate a set of hierarchical evolution patterns of the path links; S55. Based on the set of hierarchical evolution patterns of path links, a path link evolution stability evaluation function is constructed with the path link state migration stability and the path load fluctuation trend convergence rate as dual indicators, and a path link evolution stability ranking sequence is determined; S56. Filter and sort the path links above a preset threshold one by one based on the path link evolution stability ranking sequence to form an optimal path group for load distribution; S57: Based on the load distribution optimal path group, establish a path strategy optimization library corresponding to the corresponding bandwidth sensitive interval and data transmission strategy group.

7. The method for intelligent management of low-bandwidth data transmission at sea according to claim 1, characterized in that: The S6 specifically includes: S61, dynamically coupling the link load fluctuation trend of the path in the path strategy optimization library with the network operation status characteristics of the network status characteristic set in real time to form an instantaneous correlation mapping matrix between the path and the network status; S62. For the data task to be transmitted, a three-dimensional dynamic adaptation space is constructed between the task characteristics and the instantaneous association mapping matrix using the task type, scale, real-time requirements, and priority level. The position in the adaptation space is determined by the path load trend level, the network status characteristic level, and the data task priority level. S63: Based on the three-dimensional dynamic adaptation space, a path-state-task feature ternary collaborative matching process is implemented, and a real-time sliding window scanning mechanism is used in the adaptation space to determine the peak matching area between the path and the network state under the current task requirements; S64: Using the instantaneous matching characteristics within the matching degree peak area, construct a dynamic determination threshold for path link selection and data transmission strategy, and dynamically update the determination threshold according to the real-time matching degree; S65: Filtering the optimal matching path link and data transmission strategy combination in real time based on the dynamic determination threshold, identifying the path link load trend and network status feature level of the combination, and generating a corresponding data transmission strategy execution sequence in real time; S66. According to the data transmission strategy execution sequence, dynamic control of transmission rate, dynamic path selection and dynamic differential configuration of data packet priority are gradually implemented, and data transmission is performed in real time.

8. An intelligent management system for low-bandwidth data transmission at sea, applied to the intelligent management method for low-bandwidth data transmission at sea according to any one of claims 1 to 7, characterized in that: Includes the following modules: The real-time network status monitoring module is used to collect dynamic environmental data such as bandwidth, link delay, and packet loss rate within bandwidth-sensitive intervals, and mark the network operation mode status; A network state feature construction module is used to generate a network state feature set based on dynamic environment data and network operation mode status, and to generate a non-uniform time interval feature sequence and a network operation mode status sequence; The data transmission strategy generation module is used to automatically generate a data transmission strategy group based on the network status feature set through deep reinforcement learning methods, and construct a performance reward function based on data transmission delay and bandwidth utilization; The path link load characteristic analysis module is used to simulate path selection behavior based on the ant colony algorithm, record and analyze the load distribution data of path links in real time, and generate path load characteristic maps; The path strategy optimization library construction module is used to perform hierarchical clustering on the path load characteristic map, determine the optimal path group for load distribution, and form a path strategy optimization library; The dual-library dynamic mapping scheduling module is used to dynamically couple the path strategy optimization library with the network status feature set in real time, determine and execute data transmission strategies and path combinations in real time; The real-time feedback update module is used to record the actual link load, transmission delay and bandwidth utilization in real time after the data transmission strategy is executed, and dynamically update the network status feature set and path strategy optimization library based on the real-time recording results.

Images