A dynamic scheduling method for offshore rocket self-lifting platform based on collaborative intelligence

Abstract

The application discloses a kind of based on collaborative intelligence's offshore rocket self-lifting platform dynamic scheduling method, comprising the following steps: multiple platforms are constructed as intelligent unit that can be autonomously responded to task, and the adaptive matching between task and platform is realized by dynamic grouping through collaborative mechanism.In the method, multiple indicators are introduced as scheduling targets, and the weight parameters are dynamically adjusted according to real-time task, resource and environmental information to improve scheduling flexibility.Through learning optimization mechanism to generate scheduling strategy, control platform task allocation, path planning and execution timing.In the execution process, scheduling strategy and platform grouping structure can be updated in time according to environmental disturbance and task state change, to ensure smooth execution of task and stable cooperation of platform.The application is suitable for multi-task, high-frequency launch scene.

Description

Technical Field

[0001] This invention relates to the field of aerospace scheduling and control technology, and in particular to a dynamic scheduling method for a marine rocket self-elevation platform based on collaborative intelligence. Background Technology

[0002] With the rapid development of aerospace technology and commercial space missions, sea-based rocket launches have become an important supplement to high-frequency launch modes. Sea-based rocket self-elevating platforms possess flexible mobility and long-range deployment capabilities, and are increasingly being used in multi-batch, intensive launch scenarios. However, against the backdrop of continuously increasing mission density and drastic changes in sea conditions, the platform's intelligent scheduling and adaptive capabilities face severe challenges.

[0003] In existing technologies, scheduling methods for marine rocket platforms mainly rely on rule-based or pre-defined path scheduling systems. These methods typically employ centralized control and static optimization mechanisms, making them ill-suited for complex missions and dynamic disturbance scenarios. Specifically, existing scheduling technologies have significant shortcomings in the following aspects:

[0004] 1. Lack of multi-platform collaboration capability: Existing methods are difficult to achieve autonomous response and group collaboration of tasks across multiple platforms, resulting in rigid scheduling response and low resource allocation efficiency.

[0005] 2. Rigid and fixed objective function: Traditional scheduling models use a static weighting method, which cannot dynamically adjust the scheduling strategy according to the urgency of the task, platform load or environmental disturbances, resulting in poor adaptability.

[0006] 3. Disconnect between scheduling and learning mechanisms: Scheduling optimization is often separated from execution feedback, lacking a data-driven policy training path, resulting in weak policy generalization ability and high response latency.

[0007] 4. Lack of disturbance recovery mechanism: Under abnormal conditions such as sudden changes in sea state or mission interruption, there is a lack of a scheduling strategy update mechanism based on real-time feedback, resulting in insufficient overall robustness of the platform system.

[0008] Therefore, how to provide a dynamic scheduling method for marine rocket self-elevation platforms based on collaborative intelligence is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] One objective of this invention is to propose a dynamic scheduling method for marine rocket self-elevation platforms based on collaborative intelligence. This invention employs collaborative game modeling, information entropy-driven multi-objective scheduling optimization, multi-agent reinforcement learning strategy training, and a disturbance-aware scheduling self-recovery mechanism to systematically construct an intelligent closed-loop scheduling process for platform task matching, behavior sequence generation, path planning, and strategy updating. It has the advantages of strong adaptability, high collaborative efficiency, and strong disturbance recovery capability.

[0010] A dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence, according to an embodiment of the present invention, includes the following steps:

[0011] S1. Model multiple sea-based rocket self-elevating platforms as intelligent agents, collect mission requirement data, platform operation status data, and environmental observation data, construct a unified state representation, and set an initial scheduling objective function;

[0012] S2. Based on the constructed unified state representation, establish an asymmetric cooperative game model, calculate the task response benefits between agents, generate a task cooperation graph, and construct a dynamic cooperative grouping structure between platforms based on the task cooperation graph.

[0013] S3. Using the collected task requirement data, platform operation status data, and environmental observation data, calculate the corresponding task priority indicators, platform resource load indicators, and environmental disturbance level indicators, and construct an information entropy function to dynamically adjust the weight parameters of each sub-objective in the scheduling objective function, forming a dynamic objective function with adaptive capabilities.

[0014] S4. Input the task collaboration graph and dynamic objective function into the multi-agent reinforcement learning algorithm for training, and output the scheduling strategy.

[0015] S5. Generate the matching relationship between tasks and the platform and the platform behavior sequence according to the scheduling strategy, complete task allocation, path planning and time window coordination, and form an executable scheduling scheme.

[0016] S6. Collect task execution status and disturbance feedback data in the executable scheduling scheme, dynamically update the scheduling strategy parameters based on the collected data, and when a sudden disturbance or task abnormality is detected, regenerate the task collaboration graph and dynamic collaborative grouping structure based on the updated data, and generate a new scheduling scheme.

[0017] Optionally, S1 specifically includes:

[0018] S11. Model each sea-based rocket self-elevating platform as an intelligent agent with a unique state mapping relationship based on its structural parameters, execution capabilities and communication topology. Define the perception space, action space and state transition function, and establish the state transition matrix.

[0019] S12. Obtain the task requirement data of the task to be scheduled, including task type, resource requirement vector, time window boundary and priority label, and establish a task parameter set for joint encoding of the state set.

[0020] S13. Collect platform operation status data, including platform location, remaining energy, current task load and task acceptance capacity scalar, and simultaneously collect environmental observation data, including current sea state disturbance index, wind and wave intensity level and historical disturbance trend vector, to form an environmental status group.

[0021] S14. Merge the task parameter set, platform operation status data, and environment status group to generate a dynamic status representation vector. The vector evolves over time and serves as the input state stream for the scheduling system;

[0022] S15. Constructing the initial scheduling objective function based on the state representation vector. This function contains at least three types of sub-objectives: task timeliness response sub-objective. Platform resource load balancing sub-objective Adaptability to environmental disturbances sub-objectives Its weighted form is:

[0023] ;

[0024] Among them, weight parameters ,satisfy The initial configuration is automatically generated from the task structure distribution in the task parameter set;

[0025] S16. Using the aforementioned scheduling objective function as the direction of scheduling optimization, bind it to the agent to complete the integrated closed-loop modeling between state representation, agent behavior constraints, and task objectives.

[0026] Optionally, S2 specifically includes:

[0027] S21. Based on the constructed unified state representation vector and state transition matrix, a task response payoff function is constructed in the asymmetric cooperative game model for each sea-based rocket self-elevating platform. The task response payoff function comprehensively considers the task matching degree, platform resource load status, communication cost, and state transition probability.

[0028] ;

[0029] in, This refers to the intelligent agent of the sea-based rocket self-elevating platform. Intelligent agents for sea-based rocket self-elevating platforms Task response reward value, , , These are the task matching function, resource load function, and communication cost function, respectively, with weight parameters. , , ,satisfy ;

[0030] S22. The task response benefit function calculation results are constructed into a task collaboration graph in an asymmetric collaborative game model. The task collaboration graph is a directed weighted graph structure. In the directed weighted graph, the nodes correspond to the intelligent agents of the sea rocket self-elevating platform, and the edge weights represent the one-way task response benefit values.

[0031] S23. Based on the edge weight strength and the task load state of the corresponding node in the task collaboration graph, perform sparsification processing, and apply the heterogeneous density clustering algorithm to divide the task collaboration graph into subgraphs and extract the local subgraph structure in the asymmetric collaborative game model.

[0032] S24. Based on the node benefit ranking result of the subgraph in the task collaboration graph and the urgency of the task response window, construct a dynamic collaborative grouping structure between platforms. The dynamic collaborative grouping structure between platforms includes a main response marine rocket self-elevating platform intelligent agent and at least one auxiliary response marine rocket self-elevating platform intelligent agent, and generate a collaborative task mapping graph under the task drive of the main response marine rocket self-elevating platform intelligent agent.

[0033] Optionally, S3 specifically includes:

[0034] S31. Based on the collected task requirement data, a task priority index is constructed. The task priority index is calculated and generated according to the urgency of the task time window, the task resource consumption ratio, and the task collaboration dependency, and serves as the input basis for task hierarchical sorting in the scheduling system.

[0035] S32. Based on the collected operational status data of the marine rocket self-elevating platform, a platform resource load index is constructed. The platform resource load index is calculated and generated according to the current available resource ratio and task occupancy load ratio of each marine rocket self-elevating platform, reflecting the current execution capacity pressure status of the platform.

[0036] S33. Based on the collected environmental observation data, an environmental disturbance level index is constructed. The environmental disturbance level index is calculated and generated according to the rate of change of sea state disturbance amplitude, wind and wave intensity frequency and disturbance trend slope, reflecting the level of impact of external environmental interference on scheduling behavior.

[0037] S34. Normalize the task priority index, platform resource load index, and environmental disturbance level index respectively, and construct the probability distribution of the three indexes. ,in, These correspond to task priority metrics, platform resource load metrics, and environmental disturbance level metrics, respectively. Indicates the first The first category of indicators The proportion of the normalized indicator value, This represents the number of indicator items contained in this type of indicator vector;

[0038] S35. Construct information entropy functions based on the probability distributions under the three indexes, wherein the information entropy function is defined as:

[0039] ;

[0040] in, Indicates the first The information entropy function value corresponding to a category of indicators measures the degree of uncertainty of that indicator set in the current task scheduling scenario. Indicates the first The first category of indicators The proportion of the normalized indicator value, for The corresponding logarithmic function term;

[0041] S36. Based on the real-time calculation results of each type of information entropy function, calculate the dynamic weight parameters of each sub-objective in the scheduling objective function. The dynamic weight parameters are calculated based on the proportion of the information entropy function and satisfy the following normalization relationship:

[0042] ;

[0043] in, Indicates the first Sub-targets in time Dynamic weight parameters at time step This represents the current information entropy value of the corresponding sub-target. This represents the weighted sum of the entropy values ​​of all three types of information, ensuring that the normalized weight parameters satisfy the following conditions: ;

[0044] S37. Based on the above dynamic weight parameters , , The initial scheduling objective function is updated to form a dynamic objective function with adaptive capabilities:

[0045] ;

[0046] in, For at any time The global expression of the dynamic objective function. , , These are the task response time sub-objective, the platform resource balance sub-objective, and the environmental disturbance adaptability sub-objective.

[0047] Optionally, S4 specifically includes:

[0048] S41. The task collaboration graph and the dynamic objective function are used as joint inputs to the training of the multi-agent reinforcement learning algorithm. The task collaboration graph represents the topological structure of the collaboration relationship between the agents of each marine rocket self-elevation platform. The dynamic objective function is used as the objective function for scheduling strategy optimization.

[0049] S42. Based on a unified state representation, define the local state space and action space of each marine rocket self-elevating platform agent, and construct a state-action-reward-state transition quadruple.

[0050] S43. Construct a scheduling policy network and a policy value network based on the state-action-reward-state transition quadruple, which are used to generate policy functions respectively. With the strategy value function ;

[0051] S44. With the goal of maximizing the long-term expected value of the dynamic objective function, the network parameters of the policy function and the policy value function are jointly updated using a centralized training method.

[0052] S45. Define the instantaneous reward function of the scheduling policy network. This represents the increment of the dynamic objective function between two consecutive time steps.

[0053] S46. After the scheduling strategy network converges, output the scheduling strategy for the collaborative scheduling of intelligent agents of the marine rocket self-elevating platform.

[0054] Optionally, S5 specifically includes:

[0055] S51. Based on the scheduling strategy, and combining the task requirement data contained in the unified state representation with the operating status data of the marine rocket self-elevating platform, a matching relationship is generated between the task and the marine rocket self-elevating platform agent. The matching relationship is used to define the task allocation structure in the scheduling system.

[0056] S52. Based on the matching relationship between the mission and the intelligent agent of the sea-based rocket self-elevating platform, determine the mission execution order of each intelligent agent of the sea-based rocket self-elevating platform and generate a platform behavior sequence, which includes mission acceptance behavior, take-off behavior, path execution behavior, platform positioning behavior and launch preparation behavior.

[0057] S53. Based on the platform behavior sequence and the current position of the intelligent agent of the sea rocket self-elevation platform, combined with the environmental disturbance level index, the path planning process is executed to generate the platform execution path trajectory, which serves as the navigation input in the platform execution unit.

[0058] S54. Based on the platform execution path trajectory, and according to the time window constraints of each task, the platform reachability time, and the path consumption time parameters, calculate the task scheduling time coordination function:

[0059] ;

[0060] in, To coordinate time discrepancies, This indicates the estimated arrival time of the intelligent agent on the sea-based rocket self-elevating platform. This indicates the lower bound of the earliest executable time window for the task;

[0061] S55. Based on the platform behavior sequence, platform execution path trajectory and task scheduling time coordination function calculation results of the marine rocket self-elevating platform intelligent agent, an executable scheduling scheme is generated. The executable scheduling scheme includes task number, platform behavior timing and scheduling instruction set, which serve as the control input for the task execution of the marine rocket self-elevating platform intelligent agent.

[0062] Optionally, S6 specifically includes:

[0063] S61. During the execution of the executable scheduling scheme, task execution status data and environmental disturbance feedback data are collected in real time. The task execution status data includes task completion progress, task anomaly identifier and platform response delay. The environmental disturbance feedback data includes wind speed, wave height change rate and sudden meteorological anomaly indicators.

[0064] S62. Perform comprehensive analysis of task execution status data and environmental disturbance feedback data to identify whether there are sudden disturbance events and task abnormal events. If the set disturbance judgment conditions are met, trigger the scheduling strategy parameter update process.

[0065] S63. Reconstruct a unified state representation based on the collected task execution state data and environmental disturbance feedback data, and use the reconstructed state as input to the scheduling policy network for policy function parameters. With the parameters of the policy value function Dynamic updates;

[0066] S64. The updated strategy function and strategy value function are output to regenerate the task matching relationship, and the task collaboration graph is recalculated based on the reconstructed unified state representation and the current task requirement data to build a new inter-platform dynamic collaborative grouping structure.

[0067] S65. Based on the new task collaboration graph and the dynamic collaboration grouping structure between platforms, and combined with the current scheduling objective function, a new executable scheduling scheme is generated. The scheduling scheme includes the updated task number, platform behavior sequence, and scheduling control instructions.

[0068] The beneficial effects of this invention are:

[0069] (1) By introducing an asymmetric cooperative game mechanism and a multi-agent modeling method, and combining the task response benefit function with the construction of a dynamic cooperation graph, this invention realizes the temporary cooperative grouping between the self-elevating platforms of the sea rocket, breaks through the static allocation bottleneck of the traditional scheduling method in multi-platform task cooperation, and significantly improves the resource allocation flexibility and platform response efficiency in high-frequency launch scenarios.

[0070] (2) This invention constructs a multi-objective weight adjustment mechanism based on the information entropy function, dynamically integrates three types of indicators—task urgency, platform load, and environmental disturbance—into the input of the scheduling objective function, and realizes online updating and adaptive evolution of weight parameters, overcoming the problem that the static objective function of the existing scheduling model cannot adapt to task changes and environmental disturbances.

[0071] (3) This invention constructs a scheduling strategy network and a value network by integrating a task collaboration graph and a dynamic objective function, and combines reinforcement learning algorithms to realize centralized training and distributed execution of scheduling strategies, thereby improving the generalization ability of scheduling strategies and the optimal task response ability in complex environments.

[0072] (4) This invention establishes an integrated modeling method for task matching, platform behavior sequence, path planning and time coordination through an executable scheduling scheme generation mechanism, realizes the structured output of platform scheduling instructions, and improves the system consistency and precision control capability of task execution.

[0073] (5) This invention uses a disturbance event perception and feedback update mechanism to reconstruct the state input and collaborative graph structure based on task execution status and environmental disturbance data. Under the condition of sudden events, it realizes the rapid update of scheduling strategy and the reconstruction of platform collaborative structure, which significantly enhances the recovery capability and system stability of the scheduling system. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1 This is a flowchart of a dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence, as proposed in this invention. Detailed Implementation

[0076] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0077] refer to Figure 1A dynamic scheduling method for marine rocket self-elevation platforms based on collaborative intelligence includes the following steps:

[0078] S1. Model multiple sea-based rocket self-elevating platforms as intelligent agents, collect mission requirement data, platform operation status data, and environmental observation data, construct a unified state representation, and set an initial scheduling objective function;

[0079] S2. Based on the constructed unified state representation, establish an asymmetric cooperative game model, calculate the task response benefits between agents, generate a task cooperation graph, and construct a dynamic cooperative grouping structure between platforms based on the task cooperation graph.

[0080] S3. Using the collected task requirement data, platform operation status data, and environmental observation data, calculate the corresponding task priority indicators, platform resource load indicators, and environmental disturbance level indicators, and construct an information entropy function to dynamically adjust the weight parameters of each sub-objective in the scheduling objective function, forming a dynamic objective function with adaptive capabilities.

[0081] S4. Input the task collaboration graph and dynamic objective function into the multi-agent reinforcement learning algorithm for training, and output the scheduling strategy.

[0082] S5. Generate the matching relationship between tasks and the platform and the platform behavior sequence according to the scheduling strategy, complete task allocation, path planning and time window coordination, and form an executable scheduling scheme.

[0083] S6. Collect task execution status and disturbance feedback data in the executable scheduling scheme, dynamically update the scheduling strategy parameters based on the collected data, and when a sudden disturbance or task abnormality is detected, regenerate the task collaboration graph and dynamic collaborative grouping structure based on the updated data, and generate a new scheduling scheme.

[0084] In this embodiment, S1 specifically includes:

[0085] S11. Model each sea-based rocket self-elevating platform as an intelligent agent with a unique state mapping relationship based on its structural parameters, execution capabilities and communication topology. Define the perception space, action space and state transition function, and establish the state transition matrix.

[0086] S12. Obtain the task requirement data of the task to be scheduled, including task type, resource requirement vector, time window boundary and priority label, and establish a task parameter set for joint encoding of the state set.

[0087] S13. Collect platform operation status data, including platform location, remaining energy, current task load and task acceptance capacity scalar, and simultaneously collect environmental observation data, including current sea state disturbance index, wind and wave intensity level and historical disturbance trend vector, to form an environmental status group.

[0088] S14. Merge the task parameter set, platform operation status data, and environment status group to generate a dynamic status representation vector. The vector evolves over time and serves as the input state stream for the scheduling system;

[0089] S15. Constructing the initial scheduling objective function based on the state representation vector. This function contains at least three types of sub-objectives: task timeliness response sub-objective. Platform resource load balancing sub-objective Adaptability to environmental disturbances sub-objectives Its weighted form is:

[0090] ;

[0091] Among them, weight parameters ,satisfy The initial configuration is automatically generated from the task structure distribution in the task parameter set;

[0092] The principle of this formula is to transform each sub-objective into a quantified objective function value, and to control the importance of different objectives by allocating weight factors. This enables the scheduling process to have controllability and multi-dimensional optimization capabilities in the initial stage, providing a structural foundation and parameter interface for subsequent dynamic adjustment of information entropy. The constructed initial scheduling objective function, by introducing a weighted summation of multiple sub-objectives, achieves a comprehensive consideration of task response time, platform resource load balancing, and environmental disturbance adaptability.

[0093] S16. Using the aforementioned scheduling objective function as the direction of scheduling optimization, bind it to the agent to complete the integrated closed-loop modeling between state representation, agent behavior constraints, and task objectives.

[0094] This implementation model each sea-based rocket self-elevating platform as an intelligent agent with a unique state mapping relationship based on its structural parameters, execution capabilities, and communication topology. By combining collected mission requirements, platform operating status, and environmental disturbance data, a dynamic state representation vector is constructed. Based on this state vector, an initial scheduling objective function is set, which includes three sub-objectives: mission timeliness, resource load, and environmental adaptability. This achieves a unified expression of the three data elements of mission, platform, and environment, and drives the scheduling objective. Furthermore, the scheduling objective function is bound to the agent's behavioral strategy, constructing a closed-loop relationship between state representation, behavioral space, and scheduling intent. This enhances the agent's mission perception capability and objective consistency, realizes intelligent coordination of multi-platform scheduling infrastructure, and systematically expresses parameter initialization, laying a modeling foundation for subsequent strategy training and dynamic game theory.

[0095] In this embodiment, S2 specifically includes:

[0096] S21. Based on the constructed unified state representation vector and state transition matrix, a task response payoff function is constructed in the asymmetric cooperative game model for each sea-based rocket self-elevating platform. The task response payoff function comprehensively considers the task matching degree, platform resource load status, communication cost, and state transition probability.

[0097] ;

[0098] in, This refers to the intelligent agent of the sea-based rocket self-elevating platform. Intelligent agents for sea-based rocket self-elevating platforms Task response reward value, , , These are the task matching function, resource load function, and communication cost function, respectively, with weight parameters. , , ,satisfy ;

[0099] The design principle of this formula lies in comprehensively considering the capability adaptability, resource carrying capacity, and communication overhead of each platform in completing a specific task, thereby quantifying the differences in collaborative benefits between agents. By adjusting the weight ratio of each factor, a dynamic expression of collaborative preferences under different scheduling requirements can be achieved, providing an optimization basis for subsequent generation of task collaboration graphs and dynamic collaborative grouping structures between platforms. The collaborative game payoff function constructs an asymmetric scoring mechanism for measuring the collaborative value between any two marine rocket self-elevating platform agents by weighted summation of task matching degree, platform resource load status, and communication cost.

[0100] S22. The task response benefit function calculation results are constructed into a task collaboration graph in an asymmetric collaborative game model. The task collaboration graph is a directed weighted graph structure. In the directed weighted graph, the nodes correspond to the intelligent agents of the sea rocket self-elevating platform, and the edge weights represent the one-way task response benefit values.

[0101] S23. Based on the edge weight strength and the task load state of the corresponding node in the task collaboration graph, perform sparsification processing, and apply the heterogeneous density clustering algorithm to divide the task collaboration graph into subgraphs and extract the local subgraph structure in the asymmetric collaborative game model.

[0102] S24. Based on the node benefit ranking result of the subgraph in the task collaboration graph and the urgency of the task response window, construct a dynamic collaborative grouping structure between platforms. The dynamic collaborative grouping structure between platforms includes a main response marine rocket self-elevating platform intelligent agent and at least one auxiliary response marine rocket self-elevating platform intelligent agent, and generate a collaborative task mapping graph under the task drive of the main response marine rocket self-elevating platform intelligent agent.

[0103] This implementation constructs a unified state representation vector and state transition matrix, combines task matching degree, platform load state, and communication cost to define a task response reward function, establishes an asymmetric collaborative game model, generates a task collaboration graph based on the reward function output, and uses a heterogeneous density clustering algorithm to subdivide the collaboration graph, extracting substructures with high local collaboration relationships. Then, based on the node reward ranking and task time urgency, it constructs a dynamic collaborative grouping structure including main response and auxiliary response roles, generates a collaborative task mapping graph, and realizes temporary collaborative grouping and efficient task allocation among multiple platforms during task execution, effectively improving the system's resource collaboration efficiency and response flexibility in complex task scenarios.

[0104] In this embodiment, S3 specifically includes:

[0105] S31. Based on the collected task requirement data, a task priority index is constructed. The task priority index is calculated and generated according to the urgency of the task time window, the task resource consumption ratio, and the task collaboration dependency, and serves as the input basis for task hierarchical sorting in the scheduling system.

[0106] S32. Based on the collected operational status data of the marine rocket self-elevating platform, a platform resource load index is constructed. The platform resource load index is calculated and generated according to the current available resource ratio and task occupancy load ratio of each marine rocket self-elevating platform, reflecting the current execution capacity pressure status of the platform.

[0107] S33. Based on the collected environmental observation data, an environmental disturbance level index is constructed. The environmental disturbance level index is calculated and generated according to the rate of change of sea state disturbance amplitude, wind and wave intensity frequency and disturbance trend slope, reflecting the level of impact of external environmental interference on scheduling behavior.

[0108] S34. Normalize the task priority index, platform resource load index, and environmental disturbance level index respectively, and construct the probability distribution of the three indexes. ,in, These correspond to task priority metrics, platform resource load metrics, and environmental disturbance level metrics, respectively. Indicates the first The first category of indicators The proportion of the normalized indicator value, This represents the number of indicator items contained in this type of indicator vector;

[0109] S35. Construct information entropy functions based on the probability distributions under the three indexes, wherein the information entropy function is defined as:

[0110] ;

[0111] in, Indicates the first The information entropy function value corresponding to a category of indicators measures the degree of uncertainty of that indicator set in the current task scheduling scenario. Indicates the first The first category of indicators The proportion of the normalized indicator value, for The corresponding logarithmic function term;

[0112] This formula is used to construct the information entropy function, which quantifies the uncertainty of three types of indicators—task priority, platform resource load, and environmental disturbances—under the current scheduling state. The principle of the information entropy function is based on the calculation of information content in a probability distribution; that is, the more dispersed the distribution of a certain type of indicator and the higher its uncertainty, the greater its information entropy value. Through this function, the volatility and dominance of each scheduling factor under the current environment can be dynamically measured, providing a quantitative basis for subsequent adjustments to weight parameters. This enables the scheduling objective function to respond in real time to changes in external conditions and achieve adaptive optimization.

[0113] S36. Based on the real-time calculation results of each type of information entropy function, calculate the dynamic weight parameters of each sub-objective in the scheduling objective function. The dynamic weight parameters are calculated based on the proportion of the information entropy function and satisfy the following normalization relationship:

[0114] ;

[0115] in, Indicates the first Sub-targets in time Dynamic weight parameters at time step This represents the current information entropy value of the corresponding sub-target. This represents the weighted sum of the entropy values ​​of all three types of information, ensuring that the normalized weight parameters satisfy the following conditions: ;

[0116] This formula dynamically adjusts the weight parameters for each type of scheduling objective. Its core idea is to reflect the degree of uncertainty of each objective in its current state by calculating the information entropy values ​​of three indicators: task priority, platform resource load, and environmental disturbances. A higher information entropy value indicates greater uncertainty for that indicator, and its importance in the overall scheduling should be increased accordingly. Therefore, its entropy value is normalized and used as the basis for dynamic weight parameter allocation. This formula transforms the target weights from static setting to real-time adjustment, enabling the scheduling strategy to more accurately respond to changes in task urgency, resource pressure, and environmental fluctuations, thereby improving the adaptability and multi-objective balancing capability of the scheduling system.

[0117] S37. Based on the above dynamic weight parameters , , The initial scheduling objective function is updated to form a dynamic objective function with adaptive capabilities:

[0118] ;

[0119] in, For at any time The global expression of the dynamic objective function. , , These are the task response time sub-objective, the platform resource balance sub-objective, and the environmental disturbance adaptability sub-objective.

[0120] This formula is used to construct a dynamic objective function. By introducing dynamic weight parameters for three sub-objectives, it weights and combines each sub-objective to achieve real-time adjustment of the scheduling objective and balance of multiple factors. The principle behind this formula is that it uses the information entropy results of task priority, platform load, and environmental disturbances calculated in the previous step, and sets the weight of each sub-objective according to their normalized proportion. This allows the scheduling objective function to automatically adjust its optimization direction based on the current task complexity, resource scarcity, and environmental changes, enabling dynamic shifts in the focus of the objective under different scheduling scenarios, thus enhancing the flexibility and environmental adaptability of the scheduling strategy.

[0121] This implementation constructs task priority indicators, platform resource load indicators, and environmental disturbance level indicators from collected task requirement data, platform operation status data, and environmental observation data, respectively. After normalizing these three types of indicators to form probability distributions, corresponding information entropy functions are constructed to measure the uncertainty level of each indicator category in the current scheduling environment. Based on the proportion of each type of information entropy function, the weight parameters of the three sub-objectives—task timeliness, platform load balancing, and environmental adaptability—in the scheduling objective function are dynamically adjusted. Finally, an adaptive scheduling objective function is constructed, enabling real-time reconstruction and weight updates of the scheduling objective. This effectively enhances the sensitivity and adaptability of the scheduling strategy to changes in task structure and environmental disturbances, improving the rationality of resource allocation and the stability of the scheduling process.

[0122] In this embodiment, S4 specifically includes:

[0123] S41. The task collaboration graph and the dynamic objective function are used as joint inputs to the training of the multi-agent reinforcement learning algorithm. The task collaboration graph represents the topological structure of the collaboration relationship between the agents of each marine rocket self-elevation platform. The dynamic objective function is used as the objective function for scheduling strategy optimization.

[0124] S42. Based on a unified state representation, define the local state space and action space of each marine rocket self-elevating platform agent, and construct a state-action-reward-state transition quadruple.

[0125] S43. Construct a scheduling policy network and a policy value network based on the state-action-reward-state transition quadruple, which are used to generate policy functions respectively. With the strategy value function ;

[0126] S44. With the goal of maximizing the long-term expected value of the dynamic objective function, the network parameters of the policy function and the policy value function are jointly updated using a centralized training method.

[0127] S45. Define the instantaneous reward function of the scheduling policy network. This represents the increment of the dynamic objective function between two consecutive time steps.

[0128] S46. After the scheduling strategy network converges, output the scheduling strategy for the collaborative scheduling of intelligent agents of the marine rocket self-elevating platform.

[0129] This implementation uses a task coordination graph and a dynamic objective function as joint inputs to construct the local state space and action space of the agent of the self-elevating platform for marine rockets. It generates a state-action-reward-state transition quadruple, and on this basis, constructs a scheduling policy network and a policy value network. The policy function and value function are jointly optimized through centralized training, with the long-term expectation of the dynamic objective function as the optimization objective. During training, the scheduling policy network calculates the increment of the objective function as an immediate reward signal, guiding the agent to continuously adjust its scheduling behavior strategy. The final output scheduling policy possesses adaptability and optimality in multi-task, high-disturbance environments, improving task allocation efficiency and inter-platform collaborative response capabilities.

[0130] In this embodiment, S5 specifically includes:

[0131] S51. Based on the scheduling strategy, and combining the task requirement data contained in the unified state representation with the operating status data of the marine rocket self-elevating platform, a matching relationship is generated between the task and the marine rocket self-elevating platform agent. The matching relationship is used to define the task allocation structure in the scheduling system.

[0132] S52. Based on the matching relationship between the mission and the intelligent agent of the sea-based rocket self-elevating platform, determine the mission execution order of each intelligent agent of the sea-based rocket self-elevating platform and generate a platform behavior sequence, which includes mission acceptance behavior, take-off behavior, path execution behavior, platform positioning behavior and launch preparation behavior.

[0133] S53. Based on the platform behavior sequence and the current position of the intelligent agent of the sea rocket self-elevation platform, combined with the environmental disturbance level index, the path planning process is executed to generate the platform execution path trajectory, which serves as the navigation input in the platform execution unit.

[0134] S54. Based on the platform execution path trajectory, and according to the time window constraints of each task, the platform reachability time, and the path consumption time parameters, calculate the task scheduling time coordination function:

[0135] ;

[0136] in, To coordinate time discrepancies, This indicates the estimated arrival time of the intelligent agent on the sea-based rocket self-elevating platform. This indicates the lower bound of the earliest executable time window for the task;

[0137] The core principle of this formula is based on the relationship between path time, the platform's current position, and the mission start time limit to determine whether the platform can complete the scheduling response within the mission time window. This provides a time constraint basis for the subsequent generated scheduling scheme, ensuring the executability and rationality of the scheduling scheme in terms of timing. The time coordination function is used to calculate the time deviation between the expected arrival time of the marine rocket self-elevating platform agent and the earliest executable time window of the mission. This function achieves a quantitative assessment of the degree of time matching by comparing the expected arrival time of the platform during its journey with the lower limit of the time window set by the mission.

[0138] S55. Based on the platform behavior sequence, platform execution path trajectory and task scheduling time coordination function calculation results of the marine rocket self-elevating platform intelligent agent, an executable scheduling scheme is generated. The executable scheduling scheme includes task number, platform behavior timing and scheduling instruction set, which serve as the control input for the task execution of the marine rocket self-elevating platform intelligent agent.

[0139] This implementation method, based on a scheduling strategy, integrates task requirement data and platform operational status data from a unified state representation to dynamically generate a matching relationship between the task and the intelligent agent of the sea-based rocket self-elevating platform. Based on this, it determines the platform's task execution sequence and constructs a complete platform behavior sequence including task acceptance, path execution, and launch preparation. Path planning is performed by combining the platform's current position and environmental disturbance level indicators to generate a platform navigation path. Simultaneously, time coordination deviations are calculated based on the task time window, platform reachability time, and path latency to ensure accurate task execution within a controllable timeframe. Finally, the behavior sequence, path trajectory, and time coordination parameters are integrated to generate a structured, executable scheduling scheme to guide the platform's actual launch operations, significantly improving scheduling accuracy, platform coordination, and task response efficiency.

[0140] In this embodiment, S6 specifically includes:

[0141] S61. During the execution of the executable scheduling scheme, task execution status data and environmental disturbance feedback data are collected in real time. The task execution status data includes task completion progress, task anomaly identifier and platform response delay. The environmental disturbance feedback data includes wind speed, wave height change rate and sudden meteorological anomaly indicators.

[0142] S62. Perform comprehensive analysis of task execution status data and environmental disturbance feedback data to identify whether there are sudden disturbance events and task abnormal events. If the set disturbance judgment conditions are met, trigger the scheduling strategy parameter update process.

[0143] S63. Reconstruct a unified state representation based on the collected task execution state data and environmental disturbance feedback data, and use the reconstructed state as input to the scheduling policy network for policy function parameters. With the parameters of the policy value function Dynamic updates;

[0144] S64. The updated strategy function and strategy value function are output to regenerate the task matching relationship, and the task collaboration graph is recalculated based on the reconstructed unified state representation and the current task requirement data to build a new inter-platform dynamic collaborative grouping structure.

[0145] S65. Based on the new task collaboration graph and the dynamic collaboration grouping structure between platforms, and combined with the current scheduling objective function, a new executable scheduling scheme is generated. The scheduling scheme includes the updated task number, platform behavior sequence, and scheduling control instructions.

[0146] This implementation method dynamically identifies sudden disturbance events and task anomaly events by collecting task execution status data and environmental disturbance feedback data in real time during scheduling execution. Based on this, a unified state representation is reconstructed, driving the online update of the scheduling policy function and policy value function, thereby achieving continuous optimization of policy decision parameters. Subsequently, based on the updated policy output, task matching relationships, task collaboration graphs, and dynamic inter-platform collaborative grouping structures are regenerated, ultimately forming a new executable scheduling scheme. This process constructs a closed-loop linkage mechanism between scheduling execution and disturbance feedback, effectively improving the scheduling system's adaptability, policy robustness, and platform reorganization efficiency in complex environments.

[0147] Example 1:

[0148] To verify the feasibility of this invention in practice, it was applied to three sea-based rocket launch missions in a certain area of ​​the South China Sea as a typical scenario for application testing. This area has typical complex sea state characteristics, and the missions involved three sea-based rocket self-elevating platforms configured by a commercial aerospace company. The execution cycle covered key scheduling stages such as pre-launch mission preparation, platform transfer, sea state disturbance, and coordinated mission execution.

[0149] In this scenario, the system first models three sea-based rocket self-elevating platforms, forming a multi-agent system, and integrates mission requirements, platform operating status, and environmental disturbance data into a unified scheduling system. Mission requirements include launch windows for different types of satellites, launch vehicle models, and propellant usage requirements; platform status encompasses remaining resources, movement trajectories, and real-time mission load; environmental disturbances such as wind speed and direction, wave height, and sudden weather events are also integrated into the scheduling model in real time. Through an information entropy mechanism, the system dynamically adjusts the priority configuration of the scheduling objective function, enabling it to adaptively adjust the scheduling direction based on mission urgency, platform status pressure, and sea state changes.

[0150] In actual execution, the system constructs a task coordination graph through asymmetric cooperative game theory and generates a dynamic collaborative grouping structure between platforms based on it. Combining a unified state representation and a dynamic weight objective function, the scheduling strategy is trained using a multi-agent reinforcement learning algorithm. The output of the scheduling strategy is used to generate the behavior sequence, path trajectory, and task control instructions for each platform. During execution, the system continuously monitors the platform execution status and environmental feedback. In the third round of launch missions, it successfully identified a sudden low-pressure disturbance and path deviation event, triggering a dynamic strategy update mechanism to regenerate the task coordination graph and complete the task grouping reconstruction.

[0151] Regarding task execution performance, the scheduling system of this invention was compared with a control group using a traditional centralized scheduling mechanism throughout the entire scheduling cycle. The scheduling experiment results are as follows:

[0152] Table 1 Comparison data between the dynamic scheduling system of this invention and the traditional scheduling system (based on the average value of 3 rounds of tasks)

[0153] index Traditional centralized dispatching system This invention is a collaborative intelligent scheduling system. Average task response time (minutes) 27.5 13.2 Average number of scheduling policy updates (times / round) 1.1 3.8 Platform task reconstruction success rate 68.3% 93.6% Platform resource utilization 72.4% 89.1% Average path deviation control accuracy (percentage within the error range) 76.2% 95.4% Number of mission interruptions due to sea state disturbances (cumulative) 4 times 1 time

[0154] As can be seen from the data in the table, the scheduling system of this invention performs superiorly in terms of multi-platform dynamic collaboration, task response efficiency, and disturbance adaptability. Especially in the policy update and platform reconstruction stages, the system can achieve online state reconstruction and policy parameter optimization based on task execution status and disturbance feedback, significantly reducing task interruption rate and improving the success rate of platform collaborative scheduling. Simultaneously, because the platform behavior sequence and path trajectory have a unified structural output format, the task execution process is clearer, platform execution control latency is reduced by approximately 52%, and task completion stability is significantly enhanced.

[0155] For example, during the third round of launches, the second platform encountered continuous wind and wave disturbances for 7 hours during the transfer process. In the traditional scheduling scheme, the mission execution deviation caused launch delays. However, the scheduling system of this invention updates the status representation in real time and reconstructs the scheduling scheme, so that the platform is reorganized into an auxiliary response node, and the mission is switched to the main response platform, ultimately ensuring that the mission is executed on time and achieving platform load balancing.

[0156] In summary, this embodiment fully demonstrates the system response capability, scheduling flexibility, and execution robustness of the present invention in the context of multi-mission, high-frequency launch scheduling. By integrating mechanisms such as collaborative intelligence, learning-driven mechanisms, and disturbance self-recovery, a closed-loop evolution of platform scheduling from "planning-execution-feedback-correction" is achieved, providing an efficient and adaptive intelligent scheduling solution for sea-based rocket launch scenarios.

[0157] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence, characterized in that, Includes the following steps: S1. Model multiple sea-based rocket self-elevating platforms as intelligent agents, collect mission requirement data, platform operation status data, and environmental observation data, construct a unified state representation, and set an initial scheduling objective function; S2. Based on the constructed unified state representation, establish an asymmetric cooperative game model, calculate the task response benefits between agents, generate a task cooperation graph, and construct a dynamic cooperative grouping structure between platforms based on the task cooperation graph. S3. Utilizing the collected task requirement data, platform operation status data, and environmental observation data, calculate the corresponding task priority indicators, platform resource load indicators, and environmental disturbance level indicators. Construct an information entropy function to dynamically adjust the weight parameters of each sub-objective in the scheduling objective function, forming a dynamic objective function with adaptive capabilities, including: S31. Based on the collected task requirement data, construct a task priority index, which is calculated and generated according to the urgency of the task time window, the task resource consumption ratio, and the task collaboration dependency. S32. Based on the operational status data of the marine rocket self-elevating platform collected in step S1, a platform resource load index is constructed. The platform resource load index is calculated and generated based on the current available resource ratio and the task occupancy load ratio of each marine rocket self-elevating platform. S33. Based on the collected environmental observation data, an environmental disturbance level index is constructed. The environmental disturbance level index is calculated and generated according to the rate of change of sea state disturbance amplitude, wind and wave intensity frequency and disturbance trend slope. S34. Normalize the task priority index, platform resource load index, and environmental disturbance level index respectively, and construct the probability distribution of the three indexes. ,in, These correspond to task priority metrics, platform resource load metrics, and environmental disturbance level metrics, respectively. Indicates the first The first category of indicators The proportion of the normalized indicator value, This refers to the number of indicator items included in this type of indicator; S35. Construct information entropy functions based on the probability distributions under the three indexes, wherein the information entropy function is defined as: ； in, Indicates the first The information entropy function value corresponding to the class index, Indicates the first The first category of indicators The proportion of the normalized indicator value, for The corresponding logarithmic function term; S36. Based on the real-time calculation results of each type of information entropy function, calculate the dynamic weight parameters of each sub-objective in the scheduling objective function. The dynamic weight parameters are calculated based on the proportion of the information entropy function and satisfy the following normalization relationship: ； in, Indicates the first Sub-targets in time Dynamic weight parameters at time step This represents the current information entropy value of the corresponding sub-target. This represents the weighted sum of the entropy values ​​of all three types of information, and the normalized weight parameters satisfy the following: ; S37. Based on the above dynamic weight parameters , , The initial scheduling objective function is updated to form a dynamic objective function with adaptive capabilities; S4. Input the task collaboration graph and dynamic objective function into the multi-agent reinforcement learning algorithm for training, and output the scheduling strategy. S5. Generate the matching relationship between tasks and the platform and the platform behavior sequence according to the scheduling strategy, complete task allocation, path planning and time window coordination, and form an executable scheduling scheme. S6. Collect task execution status and disturbance feedback data in the executable scheduling scheme, dynamically update the scheduling strategy parameters based on the collected data, and when a sudden disturbance or task abnormality is detected, regenerate the task collaboration graph and dynamic collaborative grouping structure based on the updated data, and generate a new scheduling scheme.

2. The dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence as described in claim 1, characterized in that, S1 specifically includes: S11. Model each sea-based rocket self-elevating platform as an intelligent agent with a unique state mapping relationship based on its structural parameters, execution capabilities and communication topology. Define the perception space, action space and state transition function, and establish the state transition matrix. S12. Obtain the task requirement data of the task to be scheduled, including task type, resource requirement vector, time window boundary and priority label, and establish a task parameter set for joint encoding of the state set. S13. Collect platform operation status data, including platform location, remaining energy, current task load and task acceptance capacity scalar, and simultaneously collect environmental observation data, including current sea state disturbance index, wind and wave intensity level and historical disturbance trend vector, to form an environmental status group. S14. Merge the task parameter set, platform operation status data and environment status group to generate a dynamic status representation vector; S15. Constructing the initial scheduling objective function based on the state representation vector. This function contains at least three types of sub-objectives: task timeliness response sub-objective. Platform resource load balancing sub-objective Adaptability to environmental disturbances (sub-objective); S16. Using the scheduling objective function as the scheduling optimization direction, bind it to the agent to complete the integrated closed-loop modeling between state expression, agent behavior constraints and task objectives.

3. The dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence as described in claim 1, characterized in that, S2 specifically includes: S21. Based on the constructed unified state representation vector and state transition matrix, a task response payoff function is constructed in the asymmetric cooperative game model for each sea-based rocket self-elevating platform. The task response payoff function comprehensively considers the task matching degree, platform resource load status, communication cost, and state transition probability. ； in, This refers to the intelligent agent of the sea-based rocket self-elevating platform. Intelligent agents for sea-based rocket self-elevating platforms Task response reward value, , , These are the task matching function, resource load function, and communication cost function, respectively, with weight parameters. , , ,satisfy ; S22. The task response benefit function calculation results are constructed into a task collaboration graph in an asymmetric collaborative game model. The task collaboration graph is a directed weighted graph structure. In the directed weighted graph, the nodes correspond to the intelligent agents of the sea rocket self-elevating platform, and the edge weights represent the one-way task response benefit values. S23. Based on the edge weight strength and the corresponding task load state of the task collaboration graph, perform sparsification processing, and apply the heterogeneous density clustering algorithm to divide the task collaboration graph into subgraphs, and extract the local subgraph structure in the asymmetric collaborative game model. S24. Based on the node benefit ranking result of the subgraph in the task collaboration graph and the urgency of the task response window, construct a dynamic collaborative grouping structure between platforms. The dynamic collaborative grouping structure between platforms includes a main response marine rocket self-elevating platform intelligent agent and at least one auxiliary response marine rocket self-elevating platform intelligent agent, and generate a collaborative task mapping graph under the task drive of the main response marine rocket self-elevating platform intelligent agent.

4. The dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence as described in claim 1, characterized in that, S4 specifically includes: S41. Input the task collaboration graph and the dynamic objective function as joint inputs into the training of the multi-agent reinforcement learning algorithm. S42. Based on a unified state representation, define the local state space and action space of each marine rocket self-elevating platform agent, and construct a state-action-reward-state transition quadruple. S43. Construct a scheduling policy network and a policy value network based on the state-action-reward-state transition quadruple, which are used to generate policy functions respectively. With the strategy value function ; S44. With the goal of maximizing the long-term expected value of the dynamic objective function, the network parameters of the policy function and the policy value function are jointly updated using a centralized training method. S45. Define the instantaneous reward function of the scheduling policy network. This represents the increment of the dynamic objective function between two consecutive time steps. S46. After the scheduling strategy network converges, output the scheduling strategy for the collaborative scheduling of intelligent agents of the marine rocket self-elevating platform.

5. The dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence as described in claim 1, characterized in that, S5 specifically includes: S51. Based on the scheduling strategy, and combining the task requirement data contained in the unified state representation with the operating status data of the marine rocket self-elevating platform, a matching relationship is generated between the task and the marine rocket self-elevating platform agent. The matching relationship is used to define the task allocation structure in the scheduling system. S52. Based on the matching relationship between the mission and the intelligent agent of the sea-based rocket self-elevating platform, determine the mission execution order of each intelligent agent of the sea-based rocket self-elevating platform and generate the platform behavior sequence. S53. Based on the platform behavior sequence and the current position of the intelligent agent of the sea rocket self-elevation platform, combined with the environmental disturbance level index, the path planning process is executed to generate the platform execution path trajectory. S54. Based on the platform execution path trajectory, calculate the task scheduling time coordination function according to the time window constraints of each task, the platform reachability time and path consumption parameters; S55. Based on the platform behavior sequence, platform execution path trajectory and task scheduling time coordination function of the marine rocket self-elevating platform intelligent agent, generate an executable scheduling scheme.

6. The dynamic scheduling method for a marine rocket self-elevating platform based on collaborative intelligence as described in claim 1, characterized in that, S6 specifically includes: S61. During the execution of the executable scheduling scheme, real-time data on task execution status and environmental disturbance feedback are collected. S62. Perform comprehensive analysis of task execution status data and environmental disturbance feedback data to identify whether there are sudden disturbance events and task abnormal events. If the set disturbance judgment conditions are met, trigger the scheduling strategy parameter update process. S63. Reconstruct a unified state representation based on the collected task execution state data and environmental disturbance feedback data, and use the reconstructed state as input to the scheduling policy network for policy function parameters. With the parameters of the policy value function Dynamic updates; S64. The updated strategy function and strategy value function are output to regenerate the task matching relationship, and the task collaboration graph is recalculated based on the reconstructed unified state representation and the current task requirement data to build a new inter-platform dynamic collaborative grouping structure. S65. Based on the new task collaboration graph and the dynamic collaborative grouping structure between platforms, and combined with the current scheduling objective function, a new executable scheduling scheme is generated.

Images