Whole-process collaborative scheduling management system for unmanned ship autonomous refueling operation

Abstract

This invention discloses a full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels, belonging to the field of operational process management technology. It includes: a state encapsulation module, used to convert the real-time state of physical entities into timestamped state units and encapsulate operational requirements into task cards, defining at least one capability procurement item; and a network construction module, used to construct a multi-dimensional, time-varying virtual capability requirement relationship network based on the state units, where nodes include capability providers and capability procurement items as demand initiators. This addresses the problems of traditional unmanned vessel refueling scheduling relying on single-point technologies and lacking global coordination. Through state encapsulation and capability requirement network construction, it achieves spatiotemporal matching of multiple tasks and devices, avoids resource conflicts by combining a price screening mechanism, prevents task delays caused by localized conservative operations, improves the overall spatiotemporal resource utilization rate, reduces operational environmental risks, and adapts to the normalized refueling needs of multiple nodes in deep-sea areas.

Description

Technical Field

[0001] This invention relates to the field of operational process management technology, and more specifically, to a collaborative scheduling and management system for the entire process of autonomous refueling operations of unmanned vessels. Background Technology

[0002] For a long time, ship refueling has relied on traditional refueling ships or fixed refueling points. The operation and scheduling of these ships are highly dependent on human experience and on-site command. With the widespread application of autonomous equipment such as unmanned surface vessels, the demand for routine refueling in deep sea and at multiple nodes is increasing. The limitations of the traditional model in terms of efficiency, safety and cost are becoming more and more prominent. To address this, unmanned refueling ship technology with autonomous navigation and robotic arm docking capabilities has emerged. Existing technologies mainly focus on the realization of single-point technologies such as high-precision positioning, stable berthing and automatic docking. For example, by integrating GNSS, vision and robotic arm control to achieve precise oil line connection between ships.

[0003] Existing technological solutions mostly focus on improving the automation and reliability of individual operational links, which essentially still represent a step-by-step replacement of traditional manual operations. When faced with a complex scenario of continuously resupplying multiple dynamically distributed, differently positioned, and task-priority receiving vessels, the overall operational efficiency faces severe challenges. Each functional module, such as navigation control, robotic arm operation, and status monitoring, often operates independently based on its own optimal strategy, lacking a unified intelligent scheduling center for global resource integration and process coordination. For example, to ensure absolute safety during a single docking, the vessel may adopt an overly conservative speed or wait for an excessively long operation window. While this ensures the reliability of local operations, it may lead to a chain reaction of delays in subsequent tasks, thereby reducing the overall efficiency of the time and space resource utilization of the task chain and potentially exposing the system to greater environmental risks in the future. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels. This system solves the problems of traditional unmanned vessel refueling scheduling relying on single-point technology and lacking global coordination. By encapsulating state and constructing a capability requirement network, it achieves spatiotemporal matching of multiple tasks and devices. Combined with a quotation screening mechanism, it avoids resource conflicts, avoids task delays caused by local conservative operations, improves the utilization rate of spatiotemporal resources throughout the entire process, reduces operational environmental risks, and adapts to the normalized refueling needs of multiple nodes in the deep sea.

[0005] To solve the above problems, the present invention adopts the following technical solution:

[0006] The end-to-end collaborative scheduling and management system for autonomous refueling operations of unmanned vessels includes:

[0007] The state encapsulation module is used to convert the real-time state of a physical entity into a timestamped state unit and encapsulate the job requirements into a task card, which defines at least one capability procurement item.

[0008] The network construction module is used to build a multi-dimensional, time-varying virtual capability demand relationship network based on state units. The nodes include capability providers and capability procurement items as demand initiators, and the weight of the edges between nodes represents the estimated cost of completing the corresponding capability procurement item.

[0009] The quotation receiving module is used to broadcast task cards to capability providers and receive quotations submitted independently by each capability provider based on its own status and perceived cost. The quotation is the number of resource vouchers required to complete the corresponding capability procurement item.

[0010] The bid filtering module is used to collect all bids. Based on the global spatiotemporal mutual exclusion constraint, the multi-objective adjudication engine filters out a set of winning bids that can be chained together to form a complete task chain from all bid combinations, and generates the corresponding adjudication execution sequence and resource voucher budget allocation.

[0011] The voucher issuance module is used to generate and issue execution vouchers containing the approved quotes and action time windows according to the ruling execution sequence. The corresponding execution vouchers must be verified and cancelled before physical actions are executed.

[0012] The budget reallocation module re-executes the process of receiving, filtering, and issuing vouchers based on the updated real-time status, and performs rolling reallocation of the allocated resource voucher budget.

[0013] The parameter adjustment module is used to compare the actual completion status with the corresponding quote after the task is completed, and adjust the parameters used by the capability provider to generate the quote based on the comparison results.

[0014] Furthermore, the state encapsulation module includes:

[0015] The heterogeneous real-time state parameters of physical entities are transformed into state equivalents to obtain a sequence of dimensionless state intensity values ​​corresponding to each physical entity with synchronized timestamps.

[0016] Based on the state intensity value sequence, and according to the predefined job atomic action dictionary, the job requirements are decomposed into discrete capability primitives. By matching the difference between the task target entity and its corresponding state intensity value, the combination of capability primitives is dynamically activated and defined as capability procurement items.

[0017] Furthermore, the state encapsulation module also includes:

[0018] A dynamic time reference anchor is introduced. The dynamic time reference anchor takes the earliest critical physical event predicted to occur within the current scheduling cycle as the reference zero point. The timestamps of all state intensity value sequences and the expected time constraints of capability procurement items are uniformly converted to a relative time coordinate system with the dynamic time reference anchor as the origin. The converted state intensity values ​​with relative timestamps are encapsulated as state units, and the set of capability procurement items with relative time constraints is encapsulated as task cards.

[0019] Furthermore, the network building module includes:

[0020] Based on state units and task cards, the spatiotemporal envelope of each capability provider and capability procurement item is calculated. The spatiotemporal envelope is defined by the spatial boundary and the time interval, and only nodes with non-empty intersections of the spatiotemporal envelope are retained.

[0021] Based on the spatiotemporal envelope, the nodes of the capability procurement items that can be executed are enumerated starting from each capability provider node. The edge weights between each pair of nodes are obtained through virtual pre-occupancy simulation. The edge weights are determined by evaluating the degree of impact of the pre-occupancy behavior on the execution of capability procurement items.

[0022] Furthermore, the quotation receiving module includes:

[0023] A nonlinear decay mapping is performed on the real-time internal state of each capability provider. A pricing factor is generated for each key state dimension through a decay function with saturation characteristics. The pricing factors of all dimensions are then combined into a comprehensive state decay coefficient.

[0024] Based on the comprehensive state decay coefficient and edge weights, a competitive feedback adjustment is used to generate a bid. The competitive feedback adjustment includes forming a dynamic bid baseline based on the historical bid intentions of similar capability procurement items, and then adding additional items related to edge weights to the dynamic bid baseline after modulating it with the comprehensive state decay coefficient to form the final bid.

[0025] Furthermore, the quote filtering module includes:

[0026] The entire scheduling cycle is divided into discrete time slots. The spatiotemporal envelope of each capability provider is mapped to a Boolean variable indicating whether it is occupied in each time slot. The time window of each capability procurement item is mapped to a Boolean sequence of the corresponding time slot. A set of Boolean logic expressions is generated as a hard constraint skeleton. The Boolean logic expressions ensure that any capability provider is occupied by at most one capability procurement item in the same time slot and any capability procurement item is undertaken by at most one capability provider.

[0027] Each quote is associated with a supplier, purchase item, and time slot sequence as a label. Hypothesis propagation is performed through constraint propagation and pruning processes to eliminate conflicting quote combinations and obtain a candidate set of feasible quote combinations.

[0028] Furthermore, the quote filtering module also includes:

[0029] Using total resource voucher consumption and task chain completion time as optimization axes, the gradient-based Pareto front construction method searches along the gradient descent direction of total resource voucher consumption and simultaneously detects incremental changes in task chain completion time. It records the inflection point where resource voucher consumption decreases but the increase in completion time does not exceed a preset threshold to form a Pareto front. Based on preset weight preferences, it selects an equilibrium point from the Pareto front, decodes the candidate scheme corresponding to the equilibrium point into a decision execution sequence, and determines the resource voucher budget allocation.

[0030] Furthermore, the credential issuance module includes:

[0031] Based on the ruling execution sequence and resource voucher budget allocation, a credential chain is constructed. The credential chain consists of credential nodes encapsulated in sequence. Each credential node encodes the action instruction, the approved quote, and the action time window, and implants a time lock and a dependency tag of the preceding node.

[0032] The time lock stipulates that the credential node can only be activated when the global time reaches the start of its time window and the deviation between the current state and the predicted state of the associated capability provider is less than a preset threshold. The preceding node dependency flag stipulates that the activation of the current credential node must be after the execution completion signal of its specified preceding node has been verified and received.

[0033] Based on the credential chain, condition-triggered action unlocking and resource voucher revocation are performed. When the time lock condition of the credential node is met and the dependency of the preceding node is met, its state is changed to unlocked and an action execution token carrying the credential node is sent. When the action is started, a portion of resource vouchers is pre-deducted from the associated resource voucher budget. After the action is completed and confirmed, the resource vouchers are revoked according to the actual resource consumption.

[0034] Furthermore, the budget reconfiguration module includes:

[0035] The deviation between the updated state unit and the predicted state recorded by each voucher node in the voucher chain is compared. When the deviation exceeds the specific dynamic tolerance threshold of the node, the node and its subsequent dependent nodes are marked as the affected domain. Based on the current state, the actual consumption of the executed nodes and the original budget of the unexecuted nodes are calculated in reverse to generate the budget balance map corresponding to the affected domain.

[0036] Using the budget margin graph as the constraint boundary, the unexecuted credential nodes within the affected domain are treated as contracts to be renegotiated. A renegotiation request is broadcast to the associated capability providers. New quotations generated by each capability provider within the allowable range of the budget margin graph and the original quotations are received. The local adjudication process defined by the quotation filtering module is run within the solution space defined by the budget margin graph to generate a redistribution patch to update the credential chain and resource voucher budget allocation.

[0037] Furthermore, the parameter adjustment module includes:

[0038] After the task is completed, based on the resource voucher redemption record and disturbance detection record, a structured training tuple is constructed by merging the voucher nodes. The structured training tuple contains the predicted state, the actual state and the resource consumption deviation of the node.

[0039] A historical bias spectrum is formed based on multiple structured training tuples belonging to the same capability provider, and the key parameters of the nonlinear attenuation mapping function in the bid receiving module are adjusted compensatorily according to the historical bias spectrum.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] (1) This solution solves the problem that traditional unmanned vessel refueling scheduling relies on single-point technology and lacks global coordination. Through state encapsulation and capability requirement network construction, it realizes spatiotemporal matching of multiple tasks and multiple devices. Combined with the quotation screening mechanism, it avoids resource conflicts, avoids task delays caused by local conservative operations, improves the utilization rate of spatiotemporal resources throughout the process, reduces the risk of the operating environment, and adapts to the normalized refueling needs of multiple nodes in the deep sea.

[0042] (2) This scheme uses resource vouchers to construct a pricing and reimbursement system, combined with nonlinear decay mapping and competitive feedback adjustment, so that the pricing can accurately match the equipment status and operating costs. Then, Pareto frontier screening is used to achieve a balance between total resource consumption and task completion time, avoiding resource waste or cost overruns, ensuring reasonable budget allocation, and improving the accuracy of resource utilization.

[0043] (3) This solution has the ability to dynamically reconfigure the budget, detect state deviations in real time and define the disturbance impact domain. It can quickly adjust the price and budget through local renegotiation and adjudication without global rescheduling, reduce task interruption caused by disturbances, ensure that the operation can still proceed in an orderly manner under complex sea conditions and equipment status fluctuations, and improve the system's anti-interference and stability.

[0044] (4) This solution forms a closed-loop optimization by adjusting parameters after the task, adjusts the parameters for generating the quotation based on the historical deviation spectrum, corrects the deviation between the equipment quotation and the actual consumption, makes the subsequent quotation more in line with the actual operating capacity of the equipment, continuously optimizes the scheduling matching accuracy, and improves the overall scheduling efficiency and budget control level in the long term, adapting to the dynamic needs of different operating scenarios. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the various modules within the collaborative scheduling and management system for autonomous refueling operations of unmanned vessels, as described in this invention. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 The whole-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels includes:

[0049] The state encapsulation module is used to convert the real-time state of a physical entity into a timestamped state unit and encapsulate the job requirements into a task card, which defines at least one capability procurement item.

[0050] The state encapsulation module also performs the following operations:

[0051] S11 performs state equivalence transformation on the heterogeneous real-time state parameters of physical entities to obtain a time-stamped sequence of dimensionless state strength values ​​for each physical entity. The specific operation is as follows:

[0052] First, state equivalence transformation is performed on the heterogeneous real-time state parameters of various physical entities within the system. These physical entities include unmanned refueling vessels, receiving vessels, and key monitoring nodes in the operational environment. The heterogeneous state parameters include real-time data of different dimensions and units, such as ship speed, attitude angles, remaining fuel, position coordinates, ambient wind speed and wave height, and robotic arm joint angles. Since these heterogeneous parameters are incomparable due to dimensional differences when directly used for subsequent scheduling and analysis, the state encapsulation module uses a preset equivalence transformation algorithm to map the heterogeneous parameters of each physical entity to dimensionless state intensity values ​​within a unified numerical range. For example, the remaining fuel level of 0 to 100% is mapped to a state intensity value of 0 to 1. The environmental wave height of 0-5m is mapped to a state intensity value of 0.2-1.0 according to the safety level, ensuring that different types of parameters have a unified characterization standard. At the same time, in order to eliminate the analysis bias caused by the time difference of state acquisition of different physical entities, the state encapsulation module calibrates the acquisition time of all state parameters through a global clock synchronization protocol, so that the converted state intensity values ​​of each physical entity form a time-stamped sequence data. That is, the same timestamp corresponds to the state intensity value of each entity at that moment, ensuring that the subsequent demand decomposition and scheduling decisions based on the state sequence have a precise time consistency basis, avoiding state misjudgment caused by time asynchrony, and providing a standardized and synchronized state input basis for global collaborative scheduling.

[0053] S12, based on the state strength value sequence and according to the predefined job atomic action dictionary, decomposes the job requirements into discrete capability primitives. Then, by matching the difference between the task target entity and its corresponding state strength value, it dynamically activates the combination of capability primitives and defines them as capability procurement items. The specific operation is as follows:

[0054] After obtaining the dimensionless state strength value sequence synchronized with the timestamp, the state encapsulation module, based on this sequence and a predefined dictionary of job atomic actions, completes the reverse decomposition of job requirements and the definition of capability procurement items. The dictionary of job atomic actions pre-stores the basic action units required for the entire process of autonomous refueling of unmanned vessels, including core atomic actions such as navigation adjustment, high-precision positioning, robotic arm docking, oil circuit connection, status monitoring, and emergency disengagement. Each atomic action is associated with corresponding capability requirements and state matching conditions. The state encapsulation module first decomposes the overall job requirements, such as the continuous refueling mission of multiple receiving vessels, into discrete capability primitives according to the job logic. Each capability primitive corresponds to one or more atomic actions in the dictionary of job atomic actions. For example, completing a single refueling docking requirement can be decomposed into navigation. The system first identifies capability primitives such as target area, high-precision positioning calibration, robotic arm attitude adjustment, and oil line interface docking. Then, the state encapsulation module dynamically activates the appropriate combination of capability primitives by calculating the difference in state intensity values ​​between the target entity and the execution entity. In this case, the execution entity is the unmanned refueling vessel, and the target entity is the receiving vessel. For example, when the difference in position state intensity values ​​between the receiving vessel and the unmanned refueling vessel is large, navigation adjustment and high-precision positioning-related capability primitives are activated first. When the difference in position is small and the attitude state intensity values ​​are not well matched, attitude fine-tuning and robotic arm calibration-related capability primitives are activated. Finally, the state encapsulation module defines the activated combination of capability primitives as at least one capability procurement item, specifying the action execution requirements, state constraints, and target output corresponding to each procurement item.

[0055] S13 introduces a dynamic time reference anchor. This anchor uses the earliest predicted critical physical event within the current scheduling cycle as its reference zero. It uniformly converts the timestamps of all state intensity value sequences and the expected time constraints of capability procurement items to a relative time coordinate system with the dynamic time reference anchor as its origin. The converted state intensity values ​​with relative timestamps are then encapsulated as state units, and the set of capability procurement items with relative time constraints is encapsulated as task cards. The specific operations are as follows:

[0056] To resolve scheduling conflicts caused by inconsistent time bases among different entities, the state encapsulation module introduces a dynamic time base anchoring mechanism. This mechanism uses the earliest predicted critical physical event within the current scheduling cycle as the reference zero point. Critical physical events include emergency resupply request triggers, initial departure of unmanned refueling vessels, and completion of preceding tasks. The state encapsulation module analyzes the state intensity value sequence and task priority in real time to accurately predict the occurrence time of the critical event and set it as the origin of the relative time coordinate system, i.e., time 0. Subsequently, the state encapsulation module unifies the original absolute timestamps of the state intensity value sequences of all physical entities, as well as the expected time constraints of each capability procurement item, such as the planned execution window and the latest completion time. Converted to this relative time coordinate system, for example, if the original absolute timestamp is T1 and the dynamic time reference anchor zero point is T0, then the converted relative timestamp is T1-T0, so that all time information forms a unified representation based on the same dynamic reference. After the time conversion is completed, the state encapsulation module encapsulates the state intensity value sequence with relative timestamps into state units. Each state unit contains an entity identifier, a relative timestamp sequence, and the corresponding state intensity value, ensuring the time traceability and consistency of state information. At the same time, the set of capability procurement items with relative time constraints is encapsulated into task cards, which clearly defines the relative execution time window, time priority, and associated state requirements of each capability procurement item, so that the task cards have standardized time dimension information.

[0057] In a preferred embodiment of the present invention, a network construction module is further included, which is used to construct a multi-dimensional, time-varying virtual capability demand relationship network based on state units, wherein the nodes include capability providers and capability procurement items as demand initiators, and the weight of the edge between nodes represents the estimated cost of completing the corresponding capability procurement item.

[0058] The network construction module also performs the following operations:

[0059] S21, based on state units and task cards, calculates the corresponding spatiotemporal envelope for each capability provider and capability procurement item. The spatiotemporal envelope is defined by the spatial boundary and the time interval, and only nodes with non-empty intersections in the spatiotemporal envelope are retained. The specific operations are as follows:

[0060] The network construction module, based on the state units and task cards output by the state encapsulation module, calculates the corresponding spatiotemporal envelope for each capability provider and each capability procurement item within the system. The spatiotemporal envelope, as the core representation of a node in a multi-dimensional time-varying capability demand relationship network, is composed of spatial boundaries and time intervals, enabling precise definition of the spatiotemporal attributes of nodes. The calculation of spatial boundaries is based on the positional state intensity values ​​of physical entities within the state units, combined with the entity's operational characteristics and environmental constraints. For example, the spatial boundary of a capability provider, such as an unmanned refueling vessel, needs to be comprehensively delineated based on its current position coordinates, navigation range, robotic arm operating radius, and obstacle avoidance boundaries of the operating area, forming a closed boundary range in three-dimensional or two-dimensional space. The spatial boundary of a capability procurement item, such as the refueling docking requirement of an oil receiving vessel, is determined based on the real-time position fluctuation range of the oil receiving vessel, the necessary spatial gap for refueling operations, and the estimated positional offset under environmental disturbances. To ensure that the spatial boundaries cover the spatial range required for actual operations, the calculation of time intervals is based on the relative time constraints of the capability procurement items in the task card and the available time window of the capability provider in the state unit. Combined with the estimated time consumption of the atomic actions of the operation, the effective time interval corresponding to each node is determined. For example, the time interval of the capability procurement item is the relative start time to the latest completion time allowed for its execution, and the time interval of the capability provider is the time range of the operation it can undertake in its current state. After completing the spatiotemporal envelope calculation of all nodes, the network construction module uses the spatiotemporal intersection judgment algorithm to filter out the node combinations with non-empty spatiotemporal envelopes. That is, the spatiotemporal envelope of the capability provider overlaps with the spatiotemporal envelope of the capability procurement item in space and intersects in time, which means that the capability provider has the potential to undertake the corresponding capability procurement item in the spatiotemporal dimension. Nodes with no spatiotemporal envelope intersection are eliminated to simplify the network node scale.

[0061] S22, based on the spatiotemporal envelope, enumerates the possible capability procurement item nodes that each capability provider node can execute, starting from each capability provider node. The edge weights between each pair of nodes are obtained through virtual pre-occupancy simulation. The edge weights are determined by evaluating the impact of this pre-occupancy behavior on the execution of capability procurement items. The specific operations are as follows:

[0062] After completing node screening, the network construction module enumerates all capability procurement item nodes with non-empty intersections in their spatiotemporal envelopes, starting with each capability provider node, forming a potential capability supply and demand correspondence. This provides a basic framework for subsequent edge weight calculation. To define the edge weight between each pair of supply and demand nodes, i.e., to estimate the cost of the corresponding capability procurement item, the network construction module introduces a virtual pre-occupancy simulation mechanism. By simulating the resource occupancy status after a capability provider undertakes a target capability procurement item, the module assesses the impact of this pre-occupancy behavior on the execution of all other capability procurement items in the system, and then quantifies the impact as edge weight values. During the virtual pre-occupancy simulation, the network construction module first assumes that the capability provider undertakes the target procurement item and virtually occupies the resources corresponding to its spatiotemporal envelope, including navigation resources, operational resources, and time resources, and marks the time period and spatial range of resource occupation. Subsequently, it analyzes the impact of this pre-occupancy behavior on all other capability procurement items, including whether some procurement items lose available capability providers and whether the execution time window of the procurement items needs to be adjusted to avoid resource conflicts. The assessment of the impact of factors such as the need to activate backup capability providers and the resulting additional cost increases employs multi-dimensional quantitative indicators. These indicators include the decrease in the spatiotemporal matching degree of other procurement items after pre-allocation, the reduction in resource redundancy, and the increase in the overall delay risk of the task chain. The module converts these multi-dimensional impact indicators into a unified cost quantification value through preset weight allocation rules. This quantification value serves as the edge weight between the corresponding supply and demand nodes. For example, if a pre-allocation behavior causes a high-priority procurement item to be delayed, and the delay time exceeds a preset threshold, the corresponding edge weight will be significantly increased to reflect the higher indirect cost brought about by the pre-allocation behavior. If the pre-allocation behavior has a small impact on other procurement items, causing only a slight decrease in resource redundancy, the edge weight will remain at a low level. The edge weights determined in this way not only cover the direct cost of capability providers executing procurement items but also incorporate the indirect impact cost from a global perspective. This ensures that the edge weights can comprehensively reflect the overall cost of supply and demand matching and also guarantees that the capability demand relationship network can dynamically adapt to complex scheduling scenarios involving multiple tasks and multiple nodes, thereby improving the network's global collaborative optimization capabilities.

[0063] In a preferred embodiment of the present invention, a quotation receiving module is further included, which is used to broadcast task cards to capability providers and receive quotations submitted independently by each capability provider based on its own status and perceived cost edge, wherein the quotation is the number of resource vouchers required to complete the corresponding capability procurement item.

[0064] The quotation receiving module also performs the following steps:

[0065] S31, perform nonlinear decay mapping on the real-time internal state of each capability provider, generate a pricing factor for each key state dimension through a decay function with saturation characteristics, and synthesize the pricing factors of all dimensions into a comprehensive state decay coefficient. The specific operation is as follows:

[0066] The quotation receiving module first performs nonlinear decay mapping processing on the real-time internal state of each capability provider. The real-time internal state covers the capability provider's core operational state dimensions, including key dimensions such as remaining power of the power system, wear and tear of the robotic arm joints, pressure stability of the hydraulic system, historical job success rate, and real-time fault warning level. The state parameters for each dimension are extracted from the corresponding state unit and standardized to ensure compatibility with subsequent mapping calculations. To accurately quantify the impact of each state dimension on the quotation, the quotation receiving module configures a decay function with saturation characteristics for each key state dimension. The characteristics of this type of function are: when the state dimension is in the superior range, the decay is gradual, and the generated quotation factor approaches 1; when the state dimension drops to the critical range, the decay significantly increases, and the quotation factor decreases rapidly; when the state dimension is below the threshold range, the function enters the saturation stage, and the quotation factor tends to a stable minimum value, avoiding the influence of factors beyond the threshold. Extreme deterioration in the status leads to an unlimited decrease in the pricing factor. For example, when the remaining power of the power system is 80%, which is the excellent range, the corresponding pricing factor is 0.95; when the remaining power drops to 30%, which is the critical range, the pricing factor drops to 0.4; when the remaining power is below 10%, which is the threshold range, the pricing factor saturates at 0.2. After generating pricing factors for each dimension, the pricing receiving module synthesizes the pricing factors of all dimensions into a comprehensive status decay coefficient through a preset weighted synthesis algorithm. The weight allocation is determined based on the priority of the impact of each status dimension on the operation execution. For example, if the weight of the power system status is higher than the weight of the historical operation success rate, the value of the synthesized comprehensive status decay coefficient ranges from 0 to 1. The closer the coefficient is to 1, the better the current internal status of the capability provider, and the smaller the cost premium space when pricing. The closer the coefficient is to 0, the worse the internal status, and the higher the price is needed to cover potential operational risks and resource consumption.

[0067] S32, based on the comprehensive state decay coefficient and edge weights, generates a bid through competitive feedback adjustment. Competitive feedback adjustment includes forming a dynamic bid baseline based on historical bid intentions for similar capability procurement items, and then modulating the dynamic bid baseline with the comprehensive state decay coefficient and adding additional items related to the edge weights to form the final bid. The specific operation is as follows:

[0068] After obtaining the comprehensive state decay coefficient, the bid receiving module generates the final bid based on this coefficient and the edge weights output by the network construction module through a competitive feedback adjustment mechanism. The bid is characterized by the number of resource vouchers required to complete the corresponding capability procurement item. The first step of the competitive feedback adjustment is to construct a dynamic bid baseline. The bid receiving module retrieves historical operation data from the system to extract historical bid intention records for similar capability procurement items, including historical winning bids, distribution of unsuccessful bids, and the correlation between bids and success rates. Combined with the supply and demand ratio of similar capability procurement items in the current scheduling cycle, specifically the ratio of the number of capability providers to the number of procurement items, the bid baseline is then dynamically generated. For example, if the historical average winning bid for similar procurement items is 100 resource vouchers, and the current supply and demand ratio is 1:3 (less providers than procurement items), the dynamic bid baseline is raised to 120 resource vouchers; if the supply and demand ratio is 3:1 (more providers than procurement items), the baseline is lowered to 90 resource vouchers. Subsequently, the bid receiving module modulates the dynamic bid baseline with the comprehensive state decay coefficient. The calculation involves multiplying the baseline value by a comprehensive state attenuation coefficient to quantitatively adjust the price based on the state dimension. For example, with a baseline of 120 resource vouchers and a comprehensive state attenuation coefficient of 0.8, the modulated base price is 96 resource vouchers. Finally, the price receiving module adds an additional item related to the edge weight. This additional item is obtained by multiplying the edge weight value by a preset proportional coefficient, which is dynamically adjusted according to the operational complexity. For example, with an edge weight of 50, a medium estimated cost, and a proportional coefficient of 0.2, the additional item is 10 resource vouchers, and the final price after addition is 96 + 10 = 106 resource vouchers. Throughout the competitive feedback adjustment process, the price receiving module synchronizes the real-time bidding intentions of similar capability procurement items in real time, dynamically adjusting the price baseline and proportional coefficient to ensure that the generated final price can cover the capability provider's operating costs and risks. Specifically, it is based on the comprehensive attenuation coefficient and edge weight, thus avoiding the loss of bidding opportunities due to excessively high prices or resource losses due to excessively low prices. At the same time, it enables the price to accurately reflect the capability provider's own state and estimated operating costs.

[0069] In a preferred embodiment of the present invention, a bid screening module is further included, which is used to collect all bids, and based on global spatiotemporal mutual exclusion constraints, select a set of winning bids that can be chained together to form a complete task chain from all bid combinations through a multi-objective adjudication engine, and generate the corresponding adjudication execution sequence and resource voucher budget allocation.

[0070] The quote filtering module also performs the following steps:

[0071] S41, the entire scheduling cycle is divided into discrete time slots. The spatiotemporal envelope of each capability provider is mapped to a Boolean variable indicating whether it is occupied in each time slot. The time window of each capability procurement item is mapped to a Boolean sequence of the corresponding time slot. A set of Boolean logic expressions is generated as a hard constraint skeleton. The Boolean logic expressions ensure that any capability provider is occupied by at most one capability procurement item in the same time slot and that any capability procurement item is undertaken by at most one capability provider. The specific operations are as follows:

[0072] The quotation filtering module first divides the entire scheduling cycle into discrete time slots according to a preset duration. The duration of each time slot is determined based on a combination of the minimum estimated time of the job's atomic actions and the scheduling decision accuracy. For example, a single time slot duration is set to 5 minutes to ensure accurate matching of the job execution time granularity while avoiding computational redundancy caused by overly fine time slots. Next, the quotation filtering module maps the spatiotemporal envelope of each capability provider to Boolean variables corresponding to each time slot. A Boolean variable value of 1 indicates that the capability provider's spatiotemporal envelope is occupied within the corresponding time slot, meaning it has accepted a job or cannot accept new jobs. A value of 0 indicates that it is idle. During the mapping process, the correspondence between the time interval of the capability provider's spatiotemporal envelope and the discrete time slots must be accurately matched to ensure the accuracy of the time representation of the occupied state. Simultaneously, the quotation filtering module maps the time window of each capability procurement item, i.e., the allowed execution time interval, to a Boolean sequence of the corresponding time slot. Each element in the sequence... Each element corresponds to a time slot. A value of 1 indicates that the time slot is within the execution time window of the procurement item, while a value of 0 indicates that it is outside the window. This achieves a discretized representation of the execution time of the procurement item. Based on the above Boolean variables and Boolean sequences, the quotation screening module generates a set of Boolean logic expressions as a hard constraint skeleton. The core logic expressions include two types: one type is used to constrain the time occupancy of the same capability provider, ensuring that in the Boolean variables of all time slots, at most only one procurement item-related variable has a value of 1 in any time slot, that is, the provider is occupied by at most one procurement item in the same time slot; the other type is used to constrain the acceptance relationship of the same capability procurement item, ensuring that in the time slot with a value of 1 in its Boolean sequence, at most only one capability provider's corresponding Boolean variable has a value of 1, that is, the procurement item is accepted by at most one provider. This hard constraint skeleton defines the feasible boundary of quotation combinations from the perspective of spatiotemporal mutual exclusion, completely avoiding problems such as repeated allocation of the same resource and operation timing conflicts.

[0073] S42 uses the supplier, purchase item, and time slot sequence associated with each quotation as a label. Through constraint propagation and pruning, hypothesis propagation is performed to eliminate conflicting quotation combinations, resulting in a candidate set of feasible quotation combinations. The specific operations are as follows:

[0074] After constructing the hard constraint framework, the bid filtering module labels each received bid. Each bid's label includes its associated capability provider identifier, capability procurement item identifier, and corresponding time slot sequence—the set of time slots for which the bid promises to perform the task. The Boolean variables for each time slot are all set to 1, ensuring a precise association between the bid and the supply and demand entities, as well as spatiotemporal information. This provides a clear information carrier for constraint verification. Subsequently, the bid filtering module initiates constraint propagation and pruning, performing hypothesis propagation. This involves assuming each individual bid will win, and then deriving the impact of this hypothesis on other bids based on the hard constraint framework, thereby eliminating conflicting bid combinations. Specifically, when a bid is assumed to win, the bid filtering module substitutes the provider, procurement item, and time slot sequence information from its label into a Boolean logic expression to lock the provider's occupancy status within the corresponding time slot and simultaneously lock the procurement item's acceptance relationship. The module then filters out bids that conflict with the hypothesis, including other bids from the same supplier within the same time slot, other bids for the same procurement item, and other supply-demand combinations that overlap with the bid time slot sequence. These conflicting bids are pruned from the candidate pool. The bid filtering module iteratively repeats this hypothesis propagation process, starting with the bids corresponding to high-priority capability procurement items and gradually expanding to all bids. After each iteration, the feasible bids in the candidate pool are updated until no new conflicting bids can be pruned. Through this process, the bid filtering module ultimately eliminates all bid combinations that violate hard constraints and selects bid combinations that meet the requirements of spatiotemporal mutual exclusion and can form an effective supply-demand match. This constitutes the feasible bid combination candidate set, which significantly simplifies the calculation scope of subsequent optimization decisions, improves filtering efficiency, and ensures that each bid combination in the candidate set has actual spatiotemporal feasibility for execution.

[0075] S43 uses total resource voucher consumption and task chain completion time as optimization axes. Through a gradient-based Pareto front construction method, it searches along the gradient descent direction of total resource voucher consumption and simultaneously detects incremental changes in task chain completion time. It records inflection points where resource voucher consumption decreases but the increase in completion time does not exceed a preset threshold to form a Pareto front. Based on preset weight preferences, it selects an equilibrium point from the Pareto front. The candidate solutions corresponding to this equilibrium point are decoded into a decision execution sequence, and the resource voucher budget allocation is determined. The specific operations are as follows:

[0076] For the feasible bid combination candidate set, the bid selection module activates a multi-objective adjudication engine. Using total resource voucher consumption and task chain completion time as dual optimization axes, it conducts global optimization screening through a gradient Pareto front construction method to achieve a dynamic balance between resource cost and operational efficiency. Total resource voucher consumption is the sum of the resource voucher quantities corresponding to all winning bids in the combination, and task chain completion time is the total time from the start of the first procurement item to the completion of the last procurement item when all capable procurement items in the combination are executed sequentially. Both are indicators for evaluating the quality of bid combinations and there is a certain trade-off relationship. Generally, a decrease in total resource voucher consumption may be accompanied by an increase in task chain completion time, and vice versa. During the gradient Pareto front construction process, the bid selection module... The block first uses total resource voucher consumption as the optimization dimension, and searches for bid combinations in the candidate set one by one along the consumption gradient descent direction, that is, starting from the combination with the highest consumption and gradually searching for combinations with lower consumption. At the same time, it simultaneously detects the incremental change in the task chain completion time of each combination compared to the previous group, calculates the difference between the completion time increment and the preset threshold, and marks the combination as an inflection point when a combination is found that satisfies the condition that the total resource voucher consumption has decreased compared to the previous group and the task chain completion time increment has not exceeded the preset threshold. After continuing to search for all candidate combinations, all inflection points constitute a Pareto front, that is, no combination on the front can improve the other metric without worsening one of the optimization metrics, thus realizing the non-dominated solution set of the bi-objective optimization. Subsequently, the module performs a comprehensive evaluation of combinations on the Pareto front based on preset weight preferences, such as setting the total resource voucher consumption weight to 0.6 for resource conservation priority and the task chain completion time weight to 0.6 for efficiency priority. It then selects the balance point combination with the highest score. Finally, the quotation screening module decodes the balance point combination into a decision execution sequence, sorts the execution order of each procurement item, the corresponding capacity provider, and the execution time slot according to the temporal logic of the task chain, and determines the total resource voucher budget allocation scheme based on the number of resource vouchers in each quotation in the combination, clarifying the budget amount and payment recipient for each procurement item.

[0077] In a preferred embodiment of the present invention, a voucher issuance module is further included, which is used to generate and issue an execution voucher containing the approved quote and action time window according to the adjudication execution sequence. The corresponding execution voucher needs to be verified and cancelled before the physical action is executed.

[0078] The voucher issuance module also performs the following steps:

[0079] S51, construct a certificate chain based on the ruling execution sequence and resource voucher budget allocation. The certificate chain consists of certificate nodes encapsulated in sequence. Each certificate node encodes the action instruction, the approved quote, and the action time window, and implants a time lock and a dependency tag on the preceding node. The specific operations are as follows:

[0080] The voucher issuance module, based on the adjudication execution sequence and resource voucher budget allocation scheme output by the quotation screening module, constructs a structured voucher chain. This chain achieves integrated encapsulation and orderly association of job execution instructions, resource budgets, and spatiotemporal constraints. The construction of the voucher chain strictly follows the temporal logic of the adjudication execution sequence, encapsulating each job step in the sequence as a voucher node. These nodes are linked together in the order of execution to form a chain structure, ensuring the temporal continuity and traceability of job execution. Each voucher node serves as an independent execution instruction unit, internally encoding three types of information: first, action instructions, precisely corresponding to the specific action combinations in the job's atomic action dictionary, clearly defining the operational details that the capability provider needs to perform, such as... The parameters include: 1) the oil tanker's navigation path parameters, robotic arm docking angle, and oil circuit connection sequence; 2) the approved bid, i.e., the number of resource vouchers awarded for the corresponding capability procurement item, which serves as the basis for resource verification; 3) the action time window, which clarifies the start and end times of execution corresponding to the node based on the relative time coordinate system, consistent with the time interval in the previous spatiotemporal envelope. At the same time, the voucher issuance module implants a time lock and a predecessor node dependency mark into each voucher node. The time lock is used to limit the activation time conditions of the node, while the predecessor node dependency mark is used to associate the identification information of its directly preceding operation node. Through the two types of marks, a dual constraint on node activation is constructed to ensure that the execution of the voucher chain strictly follows the preset sequence and constraint requirements.

[0081] S52, the time lock stipulates that the credential node can only be activated when the global time reaches the start of its time window and the deviation between the current state and the predicted state of the associated capability provider is less than a preset threshold. The preceding node dependency flag stipulates that the activation of the current credential node requires the completion signal of its specified preceding node to be verified and received. The specific operation is as follows:

[0082] The time lock embedded in the voucher node, together with the dependency flag of the preceding node, constitutes a rigid constraint for node activation, clarifying the conditions for the voucher node to transition from a pending activation state to an executable state. The constraint logic of the time lock includes a dual verification dimension: the first is a time base verification, where the voucher issuance module synchronizes global relative time in real time. When the global time reaches the start of the voucher node's action time window, the time base condition is met. The second is a state deviation verification, where the voucher issuance module collects the current state strength value sequence of the associated capability provider in real time and compares it dimension-by-dimensionally with the preset predicted state strength value sequence in the adjudication execution sequence, calculating the deviation value. When the deviation values ​​of all key state dimensions are less than a preset threshold (e.g., the state strength value deviation threshold is set to 0), the node is considered activated. 1. To ensure state stability, the time lock can only be released when both checks pass. The constraint logic of the preceding node dependency mark focuses on the continuity of the operation sequence. The mark clearly records the identifier of the preceding operation node corresponding to the current credential node. The module continuously monitors the execution status of the preceding node. The preceding node dependency constraint can only be satisfied when the preceding node completes the operation and sends an execution completion signal, and the signal passes the verification, that is, confirming that the operation execution meets the requirements and there is no abnormal feedback. The two types of constraints are independent of each other and indispensable. This not only avoids the capability provider starting the operation at an unspecified time or when the state is unstable, but also prevents the sequence disorder caused by skipping steps in the operation process, thus ensuring the orderliness and safety of the entire autonomous refueling operation process.

[0083] S53, based on the credential chain, performs condition-triggered action unlocking and resource voucher reimbursement. When the time lock condition of a credential node is met and the dependency of its predecessor node is satisfied, its state is changed to unlocked and an action execution token carrying the credential node is sent. When the action is started, a portion of resource vouchers is pre-deducted from the associated resource voucher budget. After the action is completed and confirmed, resource vouchers are reimbursed based on the actual resource consumption. The specific operations are as follows:

[0084] The voucher issuance module uses a voucher chain to perform condition-triggered action unlocking and resource voucher verification, achieving coordinated operation and resource management. During the action unlocking process, the voucher issuance module monitors the constraint satisfaction of each voucher node in real time. When a node simultaneously meets the time lock release condition and the dependency constraint of the preceding node, it immediately changes the node's state from pending activation to unlocked and sends an action execution token containing the voucher node information to the associated capability provider. The token contains complete action instructions, time windows, and quotation information, serving as the capability provider's sole legal basis for initiating the operation. Resource voucher verification is divided into two stages: pre-deduction and final verification. When the operation starts, the voucher issuance module pre-deducts a portion of the resource vouchers from the corresponding resource voucher budget based on the winning bid in the voucher node. The pre-deduction ratio is set according to the operation complexity, such as the pre-deduction ratio for a regular refueling operation. A 50% deduction is applied, with a 70% deduction for complex integration tasks. This pre-deduction ensures the dedicated use of resource budgets and prevents them from being misappropriated by other operational stages. After the task is completed, the module receives actual resource consumption data from the capability provider. Combined with the task execution acceptance results (e.g., whether the task was completed on time or if there were any abnormal consumption), the pre-deducted resource vouchers are finally reconciled and adjusted. If the actual consumption matches the winning bid price, the remaining resource vouchers are deducted. If the actual consumption is lower than the winning bid price, the difference is returned to the total budget. If the actual consumption is higher than the winning bid price, the difference is made up from the remaining total budget. If insufficient, a budget reallocation warning is triggered. The entire process achieves precise linkage between tasks and resources through a conditional trigger mechanism, ensuring the compliance of task execution and realizing dynamic control of resource voucher budgets, thereby improving the accuracy and security of resource utilization.

[0085] In a preferred embodiment of the present invention, a budget reallocation module is also included, which re-executes the process of receiving quotes, filtering quotes, and issuing vouchers based on the updated real-time status, and performs rolling reallocation of the allocated resource voucher budget.

[0086] The budget reconfiguration module also performs the following steps:

[0087] S61, compare the deviation between the updated state unit and the predicted state recorded by each voucher node in the voucher chain. When the deviation exceeds the specific dynamic tolerance threshold of the node, mark the node and its subsequent dependent nodes as the affected domain. Based on the current state, reverse calculate the actual consumption of executed nodes and the original budget of unexecuted nodes to generate a budget margin map corresponding to the affected domain. The specific operations are as follows:

[0088] The budget reallocation module takes the status units updated in real time by the system as input, conducts deviation comparison with the predicted status recorded in each voucher node in the voucher chain, realizes the accurate identification of status disturbances and the definition of the influence scope during the operation execution process. The dynamic tolerance threshold is set exclusively for each voucher node and is configured differently according to the importance, status sensitivity and operation fault tolerance ability of the operation link corresponding to the node. For example, the dynamic tolerance threshold for high-precision operation links such as robotic arm docking is set to 0.08, and the threshold for relatively rough links such as navigation adjustment is set to 0.15, ensuring that the threshold can adapt to the constraint requirements of different operation scenarios. During the deviation comparison process, the budget reallocation module calculates the difference between the updated status intensity value and the predicted status intensity value dimension by dimension. When the deviation value of any key dimension exceeds the dynamic tolerance threshold of the node, it is determined that the node is affected by the status disturbance. At this time, the node and all its subsequent dependent nodes are marked as the affected domain. Since the subsequent dependent nodes logically depend on the status output of the previous node, the status deviation of the previous node will form a chain effect, so they need to be included in the affected domain for unified management. Subsequently, the budget reallocation module reversely calculates the actual resource consumption of the executed nodes based on the current real-time status, and combines the resource voucher withholding records, cancellation vouchers and operation acceptance data to accurately calculate the actual expenditure of the resource vouchers for the executed links. At the same time, the original budget amount of the unexecuted nodes is extracted, that is, the budget amount of the resource vouchers allocated in the previous period. By subtracting the actual consumption of the executed nodes and the original budget of the unexecuted nodes from the total budget amount, the total budget surplus corresponding to the affected domain is obtained, and further decomposed into the budget adjustment range of each unexecuted node, and finally a budget surplus map is generated. This map clearly marks the budget distribution, surplus scale and budget constraints of each node within the affected domain, providing a clear resource boundary basis for subsequent renegotiation and reallocation, and avoiding the budget reallocation exceeding the actual resource carrying capacity.

[0089] S62, taking the budget surplus map as the constraint boundary, regarding the unexecuted voucher nodes within the affected domain as contracts to be renegotiated, broadcasting a renegotiation request to the associated capability providers, receiving the new quotes generated by each capability provider within the range allowed by the budget surplus map and the original quotes, and running the local adjudication process defined by the quote screening module within the solution space defined by the budget surplus map to generate a reallocation patch to update the voucher chain and the resource voucher budget allocation. The specific operations are as follows:

[0090] The budget reconfiguration module first treats all unexecuted credential nodes within the affected domain as contracts to be renegotiated. It broadcasts renegotiation requests to the capability providers associated with these nodes. These requests include constraints from the budget margin map, adjustment requirements for the pending jobs, and the bid fluctuation range. Bid fluctuations must simultaneously meet the upper limit of the corresponding node's budget in the budget margin map (not exceeding the maximum adjustable range) and the lower limit (not falling below the minimum resource requirements to guarantee job execution). Furthermore, the fluctuation must be within ±20% of the original winning bid, balancing bid stability and state adaptability. Each capability provider independently generates a new bid based on its real-time status, current job load, and renegotiation constraints, and feeds it back to the module. The new bid must be synchronously associated with the adjusted job time window and state commitment to adapt to the actual situation after the disturbance. The budget reconfiguration module then collects all new bids. Within the solution space defined by the budget margin map, the local adjudication process set by the bid screening module is run. This process follows the core logic of the global adjudication, with the optimization objectives of minimizing total resource voucher consumption and shortening the task chain completion time. Combined with the hard constraint of spatiotemporal mutual exclusion, the optimal bid combination is selected. Finally, the budget reallocation module decodes the optimal combination into a budget reallocation patch. The patch contains the adjusted bids of each unexecuted node, resource voucher budget amount, action time window, and node dependencies. By updating the corresponding content of the voucher chain and the resource voucher budget allocation scheme through the patch, the rolling reallocation of the budget is realized. The entire process focuses on local optimization in the affected domain to avoid efficiency loss caused by global reallocation. At the same time, the feasibility and optimality of the reallocation scheme are ensured through budget constraints and bid screening, ensuring that autonomous refueling operations can still proceed in an orderly manner under state disturbances and reducing the risk of task interruption.

[0091] In a preferred embodiment of the present invention, a parameter adjustment module is further included, which is used to compare the actual completion status with the corresponding quotation after the task is completed, and adjust the parameters used by the capability provider to generate the quotation based on the comparison result.

[0092] The parameter adjustment module also performs the following steps:

[0093] S71, after the task is completed, based on the resource voucher redemption records and disturbance detection records, structured training tuples are constructed by merging voucher nodes. The structured training tuples contain the predicted state, actual state, and resource consumption deviation of the nodes. The specific operations are as follows:

[0094] After the autonomous refueling operation is completed, the parameter adjustment module initiates a data collection and structured processing flow. Using resource voucher redemption records and disturbance detection records as data sources, it merges data according to voucher nodes, constructing standardized structured training tuples to provide accurate data support for subsequent parameter adjustments. The resource voucher redemption records cover information such as the winning bid price, the number of resource vouchers pre-deducted, the actual resource consumption amount, the redemption adjustment difference, and the operation acceptance results for each voucher node, comprehensively reflecting the actual resource consumption situation at each operational stage. The disturbance detection records contain state deviation data during the execution of each node, the time of disturbance occurrence, the degree of disturbance impact, and budget reallocation correlation information, clearly recording the potential impact of state changes on resource consumption. During data merging, the parameter adjustment module uses the voucher node as a unique identifier to correlate resource consumption data with disturbance-related data for the same node. The process involves integration to ensure that all data from each node forms a complete dataset. Then, structured training tuples are constructed based on the merged dataset. Each tuple contains three types of information: first, the node's predicted state, which is the sequence of preset state intensity values ​​recorded in the voucher node along with its corresponding timestamps; second, the node's actual state, which is the sequence of actual state intensity values ​​collected after the job is completed, maintaining consistency with the predicted state in the time dimension for easy deviation comparison; and third, the resource consumption deviation, calculated by the difference between the node's actual resource consumption and the winning bid price. If the actual consumption is higher than the bid price, the deviation is positive; if it is lower, the deviation is negative. The absolute value of the deviation reflects the degree of deviation between the bid price and the actual consumption. The construction of structured training tuples achieves a deep correlation between the predicted state, the actual state, and the resource consumption deviation, laying a data foundation for subsequent analysis of the rationality of the provider's bidding parameters and for uncovering the causes of deviations.

[0095] S72, based on multiple structured training tuples belonging to the same capability provider, forms a historical deviation spectrum, and then compensates for the key parameters of the nonlinear attenuation mapping function in the bid receiving module according to the historical deviation spectrum. The specific operation is as follows:

[0096] The parameter adjustment module aggregates and organizes all structured training tuples belonging to the same capability provider to form a historical deviation spectrum specific to that provider. Through pattern analysis of the historical deviation data, it compensates for key parameters of the nonlinear decay mapping function in the bid receiving module, improving the accuracy of subsequent bids. The construction of the historical deviation spectrum is based on a time series model, integrating training tuple data corresponding to all operation nodes undertaken by the capability provider. It focuses on analyzing the deviation patterns between predicted and actual states, the distribution characteristics of resource consumption deviations, and the correlation between the two types of deviations. For example, it statistically analyzes the correlation coefficient between a certain state dimension, such as the deviation of the remaining power of the power system, and the resource consumption deviation, identifying key state factors that cause deviations between bids and actual consumption. Simultaneously, the historical deviation spectrum is also categorized and statistically analyzed according to operation type (e.g., routine refueling, emergency supply) and environmental conditions (e.g., calm seas, complex sea conditions), clarifying the bid deviation characteristics of the provider under different scenarios. Based on the analysis results of the historical deviation spectrum, the parameter adjustment module compensates for key parameters of the nonlinear decay mapping function. These key parameters include various state... The parameters include the saturation threshold, attenuation coefficient, and weight allocation ratio of the attenuation function in the state dimension. For example, if the historical deviation spectrum shows that a certain capacity provider has a consistently positive resource consumption deviation when the remaining power of the power system is in the range of 30%-50%, and the actual consumption is higher than the quoted price, it indicates that the attenuation factor set for this range is too low, resulting in a low quoted price. In this case, the attenuation coefficient for this range needs to be increased to increase the quoted price factor generated in this state, thereby improving the rationality of subsequent quotes. If the deviation of a certain state dimension has a very low correlation with the resource consumption deviation, the weight of this dimension in the synthesis of the comprehensive state attenuation coefficient is reduced to reduce its interference with the quote. During the parameter adjustment process, the parameter adjustment module follows the principle of small-amplitude, gradual compensation. Each adjustment does not exceed 15% of the original parameter value, and the parameter values ​​before and after the adjustment and the corresponding deviation expectations are recorded to ensure that the adjusted nonlinear attenuation mapping function can more accurately match the actual operating state and resource consumption characteristics of the capacity provider, thereby reducing the deviation between subsequent quotes and actual consumption, improving the reliability and rationality of quotes in global scheduling, and optimizing the allocation efficiency of resource voucher budgets.

[0097] The above description is merely a preferred embodiment of the present invention; however, 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 its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A collaborative scheduling and management system for the entire process of autonomous refueling operations of unmanned vessels, characterized in that: include: The state encapsulation module is used to convert the real-time state of a physical entity into a timestamped state unit and encapsulate the job requirements into a task card, which defines at least one capability procurement item. The network construction module is used to build a multi-dimensional, time-varying virtual capability demand relationship network based on state units. The nodes include capability providers and capability procurement items as demand initiators, and the weight of the edges between nodes represents the estimated cost of completing the corresponding capability procurement item. The quotation receiving module is used to broadcast task cards to capability providers and receive quotations submitted independently by each capability provider based on its own status and perceived cost. The quotation is the number of resource vouchers required to complete the corresponding capability procurement item. The bid filtering module is used to collect all bids. Based on the global spatiotemporal mutual exclusion constraint, the multi-objective adjudication engine filters out a set of winning bids that can be chained together to form a complete task chain from all bid combinations, and generates the corresponding adjudication execution sequence and resource voucher budget allocation. The voucher issuance module is used to generate and issue execution vouchers containing the approved quotes and action time windows according to the ruling execution sequence. The corresponding execution vouchers must be verified and cancelled before physical actions are executed. The budget reallocation module re-executes the process of receiving, filtering, and issuing vouchers based on the updated real-time status, and performs rolling reallocation of the allocated resource voucher budget. The parameter adjustment module is used to compare the actual completion status with the corresponding quote after the task is completed, and adjust the parameters used by the capability provider to generate the quote based on the comparison results.

2. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 1, characterized in that, The state encapsulation module includes: The heterogeneous real-time state parameters of physical entities are transformed into state equivalents to obtain a sequence of dimensionless state intensity values ​​corresponding to each physical entity with synchronized timestamps. Based on the state intensity value sequence, and according to the predefined job atomic action dictionary, the job requirements are decomposed into discrete capability primitives. By matching the difference between the task target entity and its corresponding state intensity value, the combination of capability primitives is dynamically activated and defined as capability procurement items.

3. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 2, characterized in that, The state encapsulation module further includes: A dynamic time reference anchor is introduced. The dynamic time reference anchor takes the earliest critical physical event predicted to occur within the current scheduling cycle as the reference zero point. The timestamps of all state intensity value sequences and the expected time constraints of capability procurement items are uniformly converted to a relative time coordinate system with the dynamic time reference anchor as the origin. The converted state intensity values ​​with relative timestamps are encapsulated as state units, and the set of capability procurement items with relative time constraints is encapsulated as task cards.

4. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 3, characterized in that, The network construction module includes: Based on state units and task cards, the spatiotemporal envelope of each capability provider and capability procurement item is calculated. The spatiotemporal envelope is defined by the spatial boundary and the time interval, and only nodes with non-empty intersections of the spatiotemporal envelope are retained. Based on the spatiotemporal envelope, the nodes of the capability procurement items that can be executed are enumerated starting from each capability provider node. The edge weights between each pair of nodes are obtained through virtual pre-occupancy simulation. The edge weights are determined by evaluating the degree of impact of the pre-occupancy behavior on the execution of capability procurement items.

5. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 4, characterized in that, The quotation receiving module includes: A nonlinear decay mapping is performed on the real-time internal state of each capability provider. A pricing factor is generated for each key state dimension through a decay function with saturation characteristics. The pricing factors of all dimensions are then combined into a comprehensive state decay coefficient. Based on the comprehensive state decay coefficient and edge weights, a competitive feedback adjustment is used to generate a bid. The competitive feedback adjustment includes forming a dynamic bid baseline based on the historical bid intentions of similar capability procurement items, and then adding additional items related to edge weights to the dynamic bid baseline after modulating it with the comprehensive state decay coefficient to form the final bid.

6. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 5, characterized in that, The quote filtering module includes: The entire scheduling cycle is divided into discrete time slots. The spatiotemporal envelope of each capability provider is mapped to a Boolean variable indicating whether it is occupied in each time slot. The time window of each capability procurement item is mapped to a Boolean sequence of the corresponding time slot. A set of Boolean logic expressions is generated as a hard constraint skeleton. The Boolean logic expressions ensure that any capability provider is occupied by at most one capability procurement item in the same time slot and any capability procurement item is undertaken by at most one capability provider. Each quote is associated with a supplier, purchase item, and time slot sequence as a label. Hypothesis propagation is performed through constraint propagation and pruning processes to eliminate conflicting quote combinations and obtain a candidate set of feasible quote combinations.

7. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 6, characterized in that, The quote filtering module also includes: Using total resource voucher consumption and task chain completion time as optimization axes, the gradient-based Pareto front construction method searches along the gradient descent direction of total resource voucher consumption and simultaneously detects incremental changes in task chain completion time. It records the inflection point where resource voucher consumption decreases but the increase in completion time does not exceed a preset threshold to form a Pareto front. Based on preset weight preferences, it selects an equilibrium point from the Pareto front, decodes the candidate scheme corresponding to the equilibrium point into a decision execution sequence, and determines the resource voucher budget allocation.

8. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 7, characterized in that, The certificate issuance module includes: Based on the ruling execution sequence and resource voucher budget allocation, a credential chain is constructed. The credential chain consists of credential nodes encapsulated in sequence. Each credential node encodes the action instruction, the approved quote, and the action time window, and implants a time lock and a dependency tag of the preceding node. The time lock stipulates that the credential node can only be activated when the global time reaches the start of its time window and the deviation between the current state and the predicted state of the associated capability provider is less than a preset threshold. The preceding node dependency flag stipulates that the activation of the current credential node must be after the execution completion signal of its specified preceding node has been verified and received. Based on the credential chain, condition-triggered action unlocking and resource voucher revocation are performed. When the time lock condition of the credential node is met and the dependency of the preceding node is met, its state is changed to unlocked and an action execution token carrying the credential node is sent. When the action is started, a portion of resource vouchers is pre-deducted from the associated resource voucher budget. After the action is completed and confirmed, the resource vouchers are revoked according to the actual resource consumption.

9. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 8, characterized in that, The budget reconfiguration module includes: The deviation between the updated state unit and the predicted state recorded by each voucher node in the voucher chain is compared. When the deviation exceeds the specific dynamic tolerance threshold of the node, the node and its subsequent dependent nodes are marked as the affected domain. Based on the current state, the actual consumption of the executed nodes and the original budget of the unexecuted nodes are calculated in reverse to generate the budget balance map corresponding to the affected domain. Using the budget margin graph as the constraint boundary, the unexecuted credential nodes within the affected domain are treated as contracts to be renegotiated. A renegotiation request is broadcast to the associated capability providers. New quotations generated by each capability provider within the allowable range of the budget margin graph and the original quotations are received. The local adjudication process defined by the quotation filtering module is run within the solution space defined by the budget margin graph to generate a redistribution patch to update the credential chain and resource voucher budget allocation.

10. The full-process collaborative scheduling and management system for autonomous refueling operations of unmanned vessels according to claim 9, characterized in that, The parameter adjustment module includes: After the task is completed, based on the resource voucher redemption records and disturbance detection records, structured training tuples are constructed by merging voucher nodes. The structured training tuples contain the predicted state, actual state, and resource consumption deviation of the nodes. A historical bias spectrum is formed based on multiple structured training tuples belonging to the same capability provider, and the key parameters of the nonlinear attenuation mapping function in the bid receiving module are adjusted compensatorily according to the historical bias spectrum.

Images