Ice condition change considering frozen harbor water unmanned aerial vehicle multi-frequency monitoring planning method

Abstract

The application discloses a kind of frozen port water area unmanned plane multi-frequency monitoring planning method considering ice condition change, the method includes constructing unmanned plane multi-frequency ice condition monitoring path planning model;And the unmanned plane multi-frequency ice condition monitoring path planning model includes objective function and corresponding constraint condition;Based on the unmanned plane multi-frequency ice condition monitoring path planning model, construct unmanned plane path and ice condition information coupling's double space-time network model;By designing path reduction type two-stage double space-time network optimization algorithm, double space-time network model is solved, and the collaborative optimization of unmanned plane multi-frequency monitoring time, monitoring point, flight path and ice condition information update process is carried out, to realize considering ice condition change frozen port water area unmanned plane multi-frequency monitoring planning process.The present application solves the problem that existing method is difficult to realize monitoring time, monitoring point, flight path and ice condition information update process collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of port winter waterway navigation assurance and UAV intelligent monitoring technology, and in particular to a UAV multi-frequency monitoring planning method for frozen port waterways that takes into account ice condition changes. Background Technology

[0002] Current ice condition monitoring in frozen port waters mainly relies on fixed monitoring equipment, shore-based manual observation, vessel reports, on-site manual inspections, and remote sensing imagery to obtain ice condition information. Fixed monitoring equipment is typically deployed on shore or in specific water areas to continuously observe local ice condition changes; manual inspections and vessel reports depend on feedback from on-site personnel or navigating vessels regarding channel ice conditions; remote sensing imagery can provide information on ice distribution over a large area, but its effectiveness is affected by factors such as imaging cycle, weather conditions, and spatial resolution. In recent years, drones have begun to be used for port ice condition inspections due to their high mobility, flexible deployment, and ease of acquiring close-range image information; however, this is still in its early stages, primarily involving temporary patrols, fixed-route inspections, or localized auxiliary observations.

[0003] However, existing technologies typically treat ice monitoring points as static inspection targets, rarely considering the differences in vessel navigation needs at different times and navigating different nodes. They also struggle to characterize the impact of non-homogeneous degradation of ice information over time, post-monitoring information reset, and the cumulative effect of information failure on monitoring decisions. For frozen port waters, the backfilling, accumulation, and compression of ice fragments after tugboat icebreaking exhibit significant spatiotemporal variations, and the rate of information failure and monitoring importance vary across different nodes. If only fixed-cycle, fixed-path, or distance-cost-based UAV inspection schemes are adopted, problems such as untimely monitoring of high-demand nodes, repeated monitoring of low-risk nodes, and significant cumulative ice information failures can easily arise, making it difficult to achieve coordinated optimization of monitoring time, monitoring locations, flight paths, and the ice information update process. Summary of the Invention

[0004] This invention provides a method for planning multi-frequency monitoring of frozen harbor waters by unmanned aerial vehicles (UAVs) that takes into account changes in ice conditions, in order to overcome the above-mentioned technical problems.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: A method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes, comprising the following steps: S1: Construct a path planning model for multi-frequency UAV ice condition monitoring; and the UAV multi-frequency ice condition monitoring path planning model includes an objective function and corresponding constraints; The objective function is used to minimize the loss of ice condition information at monitoring nodes in frozen port waters considering differences in navigation demand during the planning period. S2: Based on the aforementioned UAV multi-frequency ice condition monitoring path planning model, construct a dual spatiotemporal network model that couples UAV path and ice condition information; S3: By designing a path reduction-type two-stage dual spatiotemporal network optimization algorithm, the dual spatiotemporal network model is solved, and the multi-frequency monitoring time, monitoring points, flight paths and ice condition information update process of UAVs are coordinated and optimized to realize the planning process of multi-frequency UAV monitoring in frozen port waters considering ice condition changes.

[0006] Furthermore, the objective function in S1 corresponds to the UAV multi-frequency ice condition monitoring path planning model. As shown in formula (1): (1) In the formula: The node number represents the monitoring point location. Represents the set of monitoring nodes and , This represents the discrete time within the planning period. Denotes the set of discrete time steps within the planning period and , This represents a weighting parameter used to reflect the importance of information at different nodes and time periods. Represents decision variables, Represents a node At time step The ice condition information is invalid, otherwise it is 0; Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the additional penalty coefficient after the cumulative failure threshold of ice conditions is exceeded.

[0007] Furthermore, the constraints of the UAV multi-frequency ice condition monitoring path planning model include the cumulative state constraint of ice condition information under the failure of cross-time period differentiated information, the frequency and path planning constraints of UAV monitoring, the selection of candidate start time and the continuation constraint of monitoring node arrival time for each frequency under UAV multi-frequency monitoring, the return time, flight duration, power supply continuation and cost upper limit constraints for each frequency under UAV multi-frequency monitoring. Considering the cumulative state constraints of ice condition information under the failure of cross-time period differentiated information, as shown in formulas (2)-(11): (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) In the formula: As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As decision variables, Represents a node In flights Time step within If monitoring has begun, then it is 0; otherwise, it is 0. Indicates the set of drone monitoring sorties and , Indicates at node Monitoring time, Indicates each monitoring point The maximum number of times monitored during the planning period Represents a node At time step The effective duration of ice condition information, Represents a node At time step Single-step failure increment, Indicates the node at the start of the planning period The initial cumulative failure process of ice conditions, Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the node at the beginning of the initial state, i.e., at the end of time step t=0. The cumulative failure process of ice conditions, Represents a node The cumulative ice condition failure process logic constraints use sufficiently large constants.

[0008] Furthermore, the method for constructing the dual spatiotemporal network model described in S2 specifically includes: S21: Acquire the spatiotemporal network of UAV multi-frequency monitoring and ;in, Indicates the first One sortie of drones monitored the spatiotemporal network. Represents a spatiotemporal network node and , Indicates a virtual source point. Indicates a virtual sink; This is a state node, representing the drone at the moment of the event. Located at node ; Represents the spatiotemporal network arc and , The starting arc represents the time when a candidate starting point is selected from the virtual source point. , Indicates the start time of the candidate monitoring task. This represents the access arc, which is the process by which a drone moves from a base or monitoring point to a node. And complete the node The monitoring process, Indicates the arc cost of the access arc; This indicates the energy consumption of the access arc. The return arc represents the process by which the drone returns from the monitoring point to the base and ends its sortie. , These represent the arc cost and energy consumption of the return arc, respectively. S22: Obtaining Ice Condition Information Spatiotemporal Network and ;in, Multiple monitoring points The spatiotemporal network of ice condition information states constituted by the changes in ice condition information states; Represents a spatiotemporal network node. Represents a node At any moment Ice condition information status, Represents the spatiotemporal network arc of ice condition information status. Represents a node From the last monitoring time Until the next monitoring time The information invalidation interval between them; Arc cost represents the cumulative loss of information failure within the information failure interval, and includes the degree of continuous cumulative failure formed during non-homogeneous degradation and the additional penalty coefficient after exceeding the cumulative failure threshold of ice conditions; Indicates the time since the last monitoring Until the next monitoring time The information invalidation interval between them; S23: Spatiotemporal network of UAV multi-frequency monitoring obtained through S21 This describes the monitoring process of UAVs during different sorties, including takeoff, flight, monitoring, and return to base; and the spatiotemporal network of ice condition information acquired through S22. This describes the information failure accumulation process of each monitoring point between two adjacent information updates during the monitoring process, in order to construct a dual spatiotemporal network model; And the objective function of the dual spatiotemporal network model To minimize the cumulative information failure in the spatiotemporal network of ice condition information, as shown in formula (12): (12) In the formula: Represents the spatiotemporal network arc of ice condition information status. Represents decision variables, Represents a node Information failure network selection interval arc Conversely, it is 0; The constraints corresponding to the dual spatiotemporal network model include: The spatiotemporal network path constraints for multi-frequency monitoring by UAVs are shown in formulas (13)-(17): (13) (14) (15) (16) (17) In the formula: As decision variables, Indicates the first Each sortie selects the drone path spatiotemporal arc Conversely, it is 0; Indicates from virtual source The set of starting arcs and the virtual source point The set of arcs; Indicates entering the node The set of arcs and their branches from the nodes The set of starting arcs, Indicates the number of flights All access monitoring points The set of access arcs; The constraints on the number of UAV flights and their order in the spatiotemporal network for multi-frequency monitoring are shown in formulas (18)-(21): (18) (19) (20) (twenty one) In the formula: Indicates the number of flights In time step The set of starting and ending arcs, As decision variables, Indicates the number of flights Selecting the UAV path spacetime arc Conversely, it is 0; This indicates the fixed refueling and turnaround time required between adjacent flights; This represents the operating cost and energy consumption of each arc within the complete path; The path integrity constraints within the spatiotemporal network of ice condition information and the coupling constraints of the dual spatiotemporal network are shown in formulas (22)-(26): (twenty two) (twenty three) (twenty four) (25) (26) In the formula: As a decision variable, when When all values ​​are 1, they represent nodes respectively. Information failure network selection interval arc Conversely, it is 0; Indicates the number of flights All corresponding nodes In time step Completed monitoring of the set of access arcs, As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As a decision variable, when When the value is 1, it represents the node. Information failure network selection interval arc Conversely, it is 0.

[0009] Furthermore, the path reduction two-stage dual spatiotemporal network optimization algorithm designed in S3 includes: S31: The first stage of constructing a path-reduction type two-stage dual-temporal network optimization algorithm: defining the solution of the dual-temporal network model as... and , ;in, Indicates the first Candidate initiation arcs of a monitoring task in the spatiotemporal network of the UAV path; Indicates the first Candidate access arcs and prohibited access arcs within each monitoring task; and Represents a node In the Within a monitoring task, nodes are considered prohibited from access; otherwise, they are considered accessible. Represents the set of candidate starting arcs and the set of visiting arcs for all tasks; Solution based on dual spatiotemporal network model Obtain the set of spatiotemporal network arcs for reducing the drone path; Furthermore, the method for obtaining the set of spatiotemporal network arcs for reducing the drone path is as follows: S311: Solution to the dual-spatiotemporal network model In the middle, the first The starting arc of each monitoring task The start time inside If the reduction is performed, a new starting arc will be formed. for: (27) in, For the initial arc The corresponding start time; S312: Solution to the dual-temporal network model In the middle, the first Candidate access arcs and forbidden access arcs within each monitoring task If the reduction is performed, the new access arc for: (28) in, Indicates access arc The corresponding visited node, Indicates closing the task All access nodes Access arcs are defined, otherwise they are reserved; S313: Based on S311 to S312, obtain the set of spatiotemporal network arcs for reducing the UAV path. : (29) in, Indicates the number of flights The set of terminated arcs; S32: The second stage of the two-stage dual-temporal network optimization algorithm for path reduction: Based on the reduced UAV path temporal network arc set combined with the constraints corresponding to the dual-temporal network model, solve the dual-temporal coupled network model to obtain the UAV multi-frequency monitoring planning scheme for each monitoring task. Furthermore, the UAV multi-frequency monitoring planning scheme includes, but is not limited to, the selection of monitoring tasks, candidate times, monitoring nodes, access order, and ice condition information failure intervals; The UAV multi-frequency monitoring planning scheme is defined as a candidate solution; the candidate solutions are screened based on the adaptive large neighborhood search algorithm combined with the simulated annealing algorithm, and the optimal solution is output, which is the optimal scheme for UAV multi-frequency monitoring planning in frozen port waters considering ice condition changes.

[0010] Furthermore, the screening of candidate solutions in S32 based on the adaptive large neighborhood search algorithm combined with the simulated annealing algorithm specifically includes the following steps: S321: Based on the adaptive large neighborhood search algorithm, define composite perturbation operators based on information failure feedback perturbation objects and composite perturbation operators based on candidate space perturbation objects; Furthermore, the composite perturbation operator based on the information failure feedback perturbation object: The objective function value of the dual spatiotemporal network model is used as the total information failure loss index; and based on the total information failure loss index, the node with the largest information failure, the node with the largest unit information loss, the associated task corresponding to the largest information failure, and the task pair with the largest contribution to information failure are selected in sequence. Maximum node of information failure : ; The node with the greatest information loss per unit : ; The corresponding task for the maximum information failure : ; The task pair with the largest contribution from information failure :

[0011] By using the total information failure index within the statistical information failure network, we can identify and obtain nodes or task pairs at different levels that cause high information failure, and set the perturbation objects at the node level and the task level: The node level includes the node with the largest information failure and the node with the largest unit information loss; the task level includes the associated task corresponding to the largest information failure and the task pair that contributes the most to the information failure. The composite perturbation operator based on candidate space perturbation objects includes a node access perturbation operator, a node time perturbation operator, a node spatiotemporal perturbation operator, an associated task time perturbation operator, a task candidate node perturbation operator, and a task pair spatiotemporal perturbation operator. The node access perturbation operator grants access permissions to the node with the largest unit information loss. ; Represents a node In the Each monitoring task contains accessible nodes; The node-time perturbation operator: determines the preceding and following monitoring tasks corresponding to the information failure interval of the node with the largest information failure; for the preceding and following monitoring tasks, based on empirical values, replaces the starting arc corresponding to the task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs. This enables the second-stage reduced dual-temporal network to autonomously select the actual takeoff time from the candidate set of starting arcs, under the conditions of satisfying path continuity, flight time, energy consumption, and refueling turnaround constraints. The node spatiotemporal perturbation operator: for the associated task corresponding to the maximum information failure, it simultaneously performs node access perturbation and node time perturbation; The associated task time-perturbation operator: For the high-cost information failure arc corresponding to the node with the largest information failure, determine the associated monitoring tasks corresponding to the previous and next update times of the high-cost information failure arc, and replace the corresponding starting arc of the associated monitoring task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs based on empirical values. This allows the second phase to reselect the actual takeoff time from the initial arc candidate set; The task candidate node perturbation operator: For the associated task corresponding to the maximum information failure, grant access permissions to the corresponding high-failure node. ; The task employs a spatiotemporal perturbation operator that simultaneously performs perturbations of the associated task time and perturbations of the task candidate nodes. S322: Take the objective function value of the dual spatiotemporal network model corresponding to each candidate solution as the fitness value of the current candidate solution; based on the fitness value of the current candidate solution and according to the current operator weight, select the combination of each perturbation operator in the composite perturbation operator based on the information failure feedback perturbation object and the composite perturbation operator based on the candidate space perturbation object, and perturb the candidate solution according to the combined perturbation operator to obtain the latest candidate solution; S323: Obtain the fitness value of the latest candidate solution; confirm whether the latest candidate solution is accepted based on the simulated annealing acceptance criterion of the simulated annealing algorithm; if not, repeat step S322; if yes, confirm whether the preset number of iterations has been reached; if the preset number of iterations has been reached, output the optimal solution; if the preset number of iterations has not been reached, repeat step S322; thus obtaining the optimal scheme for planning the multi-frequency monitoring of frozen port waters by UAVs considering ice condition changes.

[0012] Furthermore, in S322, based on the fitness value of the current candidate solution and the current operator weights, a combination of perturbation operators is selected from the composite perturbation operator based on the information failure feedback perturbation object and the composite perturbation operator based on the candidate space perturbation object, specifically as follows: Let the set of composite perturbation operators be... , No. The selection weights, cumulative scores, and number of uses of the perturbation operator are respectively: Then the first The probability that the perturbation operator is selected in the current iteration. for: (30) In the formula: Indicates the first The current selection weights of the composite perturbation operator; A reward value is assigned to the selected perturbation operator based on the quality of the candidate solution, i.e., the fitness value of the current candidate solution. and update its cumulative rating. With the number of times it is used After each iteration update segment, obtain the first... The average reward of the perturbation operator is Then, the selection weights are updated: (31) (32) In the formula: This indicates that the reaction coefficient is updated by selecting weights. This represents the lower limit of the average reward.

[0013] This invention provides a method for planning multi-frequency monitoring of frozen harbor waters using unmanned aerial vehicles (UAVs) considering ice condition changes, with the following advantages: (1) Based on AIS traffic flow, waterway layout and historical ice condition data, this invention identifies key monitoring nodes and constructs a UAV multi-frequency ice condition monitoring path planning model. Its objective function is used to minimize the combined loss of ice condition information degradation and failure during the planning period, which is weighted by traffic demand. At the same time, it considers the degradation loss before information failure and the failure state loss after exceeding the failure threshold, so that the monitoring scheme is transformed from traditional fixed inspection or simple coverage to optimization oriented towards traffic demand and the effectiveness of ice condition information.

[0014] (2) This invention constructs a dual spatiotemporal network model that couples UAV path and ice condition information. It aligns the UAV monitoring completion event and node information update event through network coupling constraints, so that the UAV path selection, monitoring time decision and information failure accumulation process can be described collaboratively in a unified dual spatiotemporal coupled network. It can take into account the dynamic change process of frozen port ice condition information during the planning period, avoids simply regarding ice condition information as a fixed validity period or static coverage state, and enables the UAV monitoring scheme to adjust the monitoring time and monitoring path in a timely manner according to the information failure changes of different nodes and different time periods, thereby improving the timeliness of ice condition information updates.

[0015] (3) By designing a path reduction type two-stage dual spatiotemporal network optimization algorithm, this invention can improve the solution efficiency and solution quality of the multi-frequency UAV ice condition monitoring path planning problem. In the face of the problem of highly coupled path selection and information failure process under multiple tasks, multiple nodes, and multiple time periods, this invention can reduce the size of the second-stage candidate path while preserving the coupling relationship between path continuity, flight time, energy consumption, energy replenishment turnover and information failure, and obtain better monitoring time, monitoring point and monitoring path scheme through targeted search. Attached Figure Description

[0016] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This invention provides a route map for the planning method of multi-frequency monitoring of frozen harbor waters by unmanned aerial vehicles, taking into account changes in ice conditions. Figure 2 This is a flowchart illustrating the framework of the path reduction two-stage dual-temporal network optimization algorithm in this embodiment. Figure 3 This is a graph showing the iteration curves of the path reduction two-stage dual spatiotemporal network optimization algorithm, the adaptive large neighborhood search algorithm, and the large neighborhood search algorithm in this embodiment. Figure 4 This diagram illustrates the use of internal operators in the path-reduction two-stage dual-temporal network optimization algorithm in this embodiment. Figure 5 This is a graph showing the number of paths reduced by the path reduction two-stage dual-temporal network optimization algorithm in this embodiment; Figure 6 This is a graph showing the results of multiple frequency monitoring task paths in the solution scheme of this embodiment; Figure 7 This is a graph showing the results of the ice condition information changes at the monitoring points in this embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This embodiment provides a method for planning multi-frequency UAV monitoring of frozen harbor waters that takes into account ice condition changes, such as... Figure 1 As shown, the specific steps include: S1: Acquire AIS trajectory, channel layout, historical ice conditions, and monitoring data from the previous planning period for the nearshore navigable waters of the frozen port to identify key monitoring nodes; simultaneously, determine the navigation demand weight of nodes based on statistical navigation flow based on AIS data, and determine the time-segmented ice condition information failure rate of each node by combining historical and previous monitoring data, so as to construct a UAV multi-frequency ice condition monitoring path planning model; and the UAV multi-frequency ice condition monitoring path planning model includes an objective function and corresponding constraints; Specifically, the objective function corresponding to the UAV multi-frequency ice condition monitoring path planning model. As shown in formula (1): (1) In the formula: The node number represents the monitoring point location. Represents the set of monitoring nodes and , This represents the discrete time within the planning period. Denotes the set of discrete time steps within the planning period and , The weighting parameter represents the importance of information at different nodes and time periods. In this embodiment, it is reflected by mining the historical traffic and navigation flow of the monitoring points. Represents decision variables, Represents a node At time step The ice condition information is invalid, otherwise it is 0; Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the additional penalty coefficient after the cumulative ice condition failure threshold is exceeded. In this embodiment, the objective function is used to minimize the ice condition information failure loss of monitoring nodes in frozen port waters considering differences in navigation demand during the planning period. This objective function also characterizes the cumulative failure degree formed by ice condition information during non-homogeneous degradation, as well as the additional failure state penalty generated after the cumulative failure degree exceeds the acceptable threshold, so that the monitoring plan can minimize the degradation loss before the information fails and reduce the additional failure loss after the information enters the failure state.

[0020] Specifically, the constraints for the UAV multi-frequency ice condition monitoring path planning model include: Considering the cumulative state constraints of ice condition information under the failure of cross-time period differentiated information, as shown in formulas (2)-(11): (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) In the formula: As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As decision variables, Represents a node In flights Time step within If monitoring has begun, then it is 0; otherwise, it is 0. Indicates the set of drone monitoring sorties and , Indicates at node Monitoring time, Indicates each monitoring point The maximum number of times monitored during the planning period Represents a node At time step The effective duration of ice condition information varies between nodes and time periods. Represents a node At time step The single-step failure increment, while the failure changes during the planning period are cross-time periods and non-homogeneous; Indicates the node at the start of the planning period The initial cumulative failure process of ice conditions, Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the node at the end of the initial state (time step t=0). The cumulative failure process of ice conditions, Represents a node The cumulative ice condition failure process logic constraints use sufficiently large constants; As decision variables, Represents a node At time step The ice condition information is in an invalid state. In this embodiment, formulas (2) to (3) represent the time delay relationship between the start monitoring event and the information update event and the upper limit of the number of monitoring times. Formulas (4) to (5) represent the definition of the non-homogeneous failure rate of ice condition at different time periods obtained from historical data and previous periodic monitoring. Formulas (5) to (9) represent the cumulative failure process of the ice condition information status of the monitoring point. Formula (5) represents the initial state of the ice condition information. Formulas (6) to (9) represent the cumulative ice condition failure information when the monitoring point is not monitored, and the reset of the ice condition information status after being monitored. If in time step Unmonitored Nodes The cumulative failure process increases. If the time step node Once monitoring is complete, the cumulative failure process is reset to 0 at time step; Formulas (10) to (11) are constraints for identifying ice condition information status failures. When the cumulative failure process exceeds the threshold of 1, the node information is judged as an additional failure. The constraints ensure the information status decision variables. The value of .

[0021] The frequency of UAV monitoring and path planning constraints are shown in formulas (12)-(20): (12) (13) (14) (15) (16) (17) (18) (19) (20) In the formula: As decision variables, Represents a node In flights If it is monitored, it is 0; otherwise, it is 0. As decision variables, Indicates the number of flights Chinese drones from nodes Fly directly to the node Conversely, it is 0; They represent the number of flights. In the monitoring path, from the shore-based station to the first monitoring point and the last monitoring point Decision variables for shore-based operations. Denotes the set of nodes containing a base and , Indicates a self-loop within the path. Represents a node In flights Decision variables for the order of access in the monitoring path. The formula (12) indicates the number of inspection points; in this embodiment, formula (13) indicates that the number of times a monitoring point is monitored does not exceed the upper limit under multiple monitoring by UAVs during the planning period; formulas (14) to (15) indicate that under each round of UAV monitoring, the monitoring path includes departure, node entry and exit, and return, reflecting the path flow balance; formulas (16) to (17) indicate the elimination of sub-loops in each frequency monitoring path of the UAV to ensure the rationality of the monitoring path; formula (20) indicates the correlation of monitoring variables. Under the multi-frequency monitoring by UAV, the selection of candidate start time and the continuation constraint of the arrival time of monitoring node for each frequency are shown in formulas (21)-(28): (twenty one) (twenty two) (twenty three) (twenty four) (25) (26) (27) (28) In the formula: As decision variables, Indicates the number of flights Select candidate takeoff time Conversely, it is 0; Indicates the first A collection of candidate takeoff times for each flight. Indicates the number of flights The decision variable for the actual takeoff time, Indicates the number of flights The One candidate takeoff time Representing nodes respectively In flights The decision variable for the actual arrival time in the data. These represent the distance from the shore base to the node. With slave node To the node Flight time, The end time of the planning period is indicated; in this embodiment, formulas (21) to (22) indicate that each flight must select only one time from its candidate takeoff time set, and the definition of the actual takeoff time is given. Formulas (23) to (28) indicate the association between the arrival time of nodes in the monitoring mission path and the node selection variable.

[0022] The return time, flight duration, power supply continuity, and cost upper limit constraints for each frequency of UAV multi-frequency monitoring are shown in formulas (29)-(36): (29) (30) (31) (32) (33) (34) (35) (36) In the formula: Indicates the number of flights Upon returning to base, This indicates the drone's flight time from the last monitoring node to the base. Indicates the number of flights Continuous working time, This indicates the maximum permissible continuous operation time for a single sortie. Indicates the set of flightable arcs of drones and , Indicates along the arc Energy consumption during flight Indicates at node The energy consumption of the service generated during one monitoring cycle. This indicates the maximum available energy per flight. This indicates the fixed refueling and turnaround time required between adjacent flights. Indicates execution of the first The fixed start-up cost per flight Indicates along the arc The operating costs incurred by flight Represents a node The cost of one monitoring session This indicates the upper limit of the operating cost of a single monitoring mission by the UAV. In this embodiment, formulas (29) to (30) indicate the time when the UAV returns to the base under the current monitoring path, formulas (31) to (33) indicate the definition of flight time and the limitation of the duration of a single mission within a sortie, while ensuring that the monitoring mission does not exceed the time boundary of the planning period, formula (34) ensures that the sum of the flight energy consumption and service energy consumption of each sortie does not exceed the maximum available energy under full charge, formula (35) ensures that there is a fixed recharge turnaround time between adjacent sorties, and formula (36) indicates that the cost of a single monitoring mission must not exceed the limit.

[0023] S2: Based on the aforementioned UAV multi-frequency ice condition monitoring path planning model, a dual spatiotemporal network model coupling UAV path and ice condition information is constructed. In this embodiment, the information failure calculation in the original model, i.e., the UAV multi-frequency ice condition monitoring path planning model, is essentially a cross-time period state recursion process. For a monitoring point, its information failure level at a given time is closely related to the previous update time, whether it is currently being monitored, and the failure accumulation process between adjacent time periods. Therefore, if this process is directly described in the original mixed integer programming model, a large number of time state variables, update indicator variables, and adjacent time period recursion constraints need to be introduced. When the planning period is discretized into many time steps, the number of constraints will increase rapidly with the number of monitoring points and the number of time steps, forming a strong cross-time period coupling relationship, thereby increasing the difficulty of solving the model. Therefore, in this embodiment, the original model is converted into a dual spatiotemporal network model, i.e., a UAV runtime spatiotemporal network and an information failure spatiotemporal network are constructed respectively: the UAV runtime spatiotemporal network is used to describe the take-off, flight, monitoring, and return processes of the UAV under different sorties, and the information failure spatiotemporal network is used to describe the information failure accumulation process of each monitoring point between two adjacent information updates. By pre-generating network nodes and arcs, and transforming UAV operating time, energy consumption, cost, and information failure loss into arc attributes, the path continuity and information failure recursive relationship in the original model can be characterized by network flow constraints and arc selection constraints, respectively. Specifically, each arc in the information failure network represents the cumulative failure value of a monitoring point from the last update time to the next update time, and its arc cost can be pre-calculated. Thus, the complex time-by-time recursive relationship is transformed into an arc selection problem between adjacent update times. Therefore, dual-spatiotemporal network modeling not only more clearly expresses the coupling relationship between the UAV operation process and the information failure evolution process, but also reduces the impact of numerous recursive constraints in the original model on solution efficiency.

[0024] The process of constructing the dual spatiotemporal network model in this embodiment specifically includes the following steps: S21: Acquire the spatiotemporal network of UAV multi-frequency monitoring and ;in, Indicates the first One sortie of drones monitored the spatiotemporal network. Represents a spatiotemporal network node and , Indicates a virtual source point. Indicates a virtual sink; This is a state node, representing the drone at the moment of the event. Located at node ; Represents the spatiotemporal network arc and , The starting arc represents the time when a candidate starting point is selected from the virtual source point. , Indicates the start time of the candidate monitoring task. This represents the access arc, which is the process by which a drone moves from a base or monitoring point to a node. And complete the node The monitoring process, Indicates the arc cost of the access arc; This indicates the energy consumption of the access arc. The return arc represents the process by which the drone returns from the monitoring point to the base and ends its sortie. , These represent the arc cost and energy consumption of the return arc, respectively. In this embodiment, the UAV multi-frequency monitoring spatiotemporal network represents the UAV monitoring path using time-monitoring point, which can indirectly obtain the monitoring time, points, and point order within each frequency monitoring task. S22: Obtaining Ice Condition Information Spatiotemporal Network and ;in, Multiple monitoring points The spatiotemporal network of ice condition information states constituted by the changes in ice condition information states; Represents a spatiotemporal network node. Represents a node At any moment Ice condition information status, Represents the spatiotemporal network arc of ice condition information status. Represents a node From the last monitoring time Until the next monitoring time The information invalidation interval between them; Arc cost represents the cumulative loss of information failure within the information failure interval, and includes the degree of continuous cumulative failure formed during the non-homogeneous degradation process and the additional penalty coefficient after exceeding the cumulative failure threshold of ice conditions. This arc cost can be pre-calculated based on the navigation demand weight of each time period within the interval, the increment of non-homogeneous ice condition degradation, and the failure threshold. Indicates the time since the last monitoring Until the next monitoring time The information invalidation interval between them; S23: Spatiotemporal network of UAV multi-frequency monitoring obtained through S21 This describes the monitoring process of UAVs during different sorties, including takeoff, flight, monitoring, and return to base; and the spatiotemporal network of ice condition information acquired through S22. This describes the information failure accumulation process of each monitoring point between two adjacent information updates during the monitoring process, in order to construct a dual spatiotemporal network model; And the objective function of the dual spatiotemporal network model To minimize the cumulative information failure in the spatiotemporal network of ice condition information, as shown in formula (37): (37) In the formula: Represents the spatiotemporal network arc of ice condition information status. Represents decision variables, Represents a node Information failure network selection interval arc Conversely, it is 0; the objective function of the dual-spatiotemporal network model is to minimize the cumulative information failure of the ice condition information state spatiotemporal network. Indicates each monitoring point. Represents the set of monitoring nodes. From the last monitoring time Until the next monitoring time The information invalidation interval between them; The constraints corresponding to the dual spatiotemporal network model include: The spatiotemporal network path constraints for multi-frequency monitoring by UAVs are shown in formulas (38)-(42): (38) (39) (40) (41) (42) In the formula: As decision variables, Indicates the first Each sortie selects the drone path spatiotemporal arc Conversely, it is 0; Indicates from virtual source The set of starting arcs and the virtual source point The set of arcs; Indicates entering the node The set of arcs and their branches from the nodes The set of starting arcs, Indicates the number of flights All access monitoring points The set of access arcs; in this embodiment, formulas (38) to (40) ensure that the monitoring path of each UAV flight completely contains a spatiotemporal path from start to end, the first The next inspection task must select an arc starting from the starting point, the first... The inspection path must ultimately return to or terminate at the mission endpoint, and flow conservation constraints apply. Formula (41) for each sortie At least one access arc must be selected to indirectly ensure that at least one monitoring point is completed for each flight. Formula (42) indicates that in the same flight Within, at the same monitoring point It can be accessed at most once.

[0025] The constraints on the number of UAV flights and their order in the spatiotemporal network for multi-frequency monitoring are shown in formulas (43)-(46): (43) (44) (45) (46) In the formula: Indicates the number of flights In time step The set of starting and ending arcs, As decision variables, Indicates the number of flights Selecting the UAV path spacetime arc Conversely, it is 0; This indicates the fixed refueling and turnaround time required between adjacent flights; This represents the operating cost and energy consumption of each arc within the complete path; in this embodiment, formula (43) represents the cost and energy consumption of each flight. The flight time limit, formula (44) means to ensure a fixed refueling turnaround time between adjacent flights, and formulas (45)~(46) mean that each flight Operating costs and maximum available energy limits.

[0026] The path integrity constraints within the spatiotemporal network of ice condition information and the coupling constraints of the dual spatiotemporal network are shown in formulas (47)-(51): (47) (48) (49) (50) (51) In the formula: As a decision variable, when When all values ​​are 1, they represent nodes respectively. Information failure network selection interval arc Conversely, it is 0; Indicates the number of flights All corresponding nodes In time step Completed monitoring of the set of access arcs, As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As a decision variable, when When the value is 1, it represents the node. Information failure network selection interval arc Conversely, it is 0. In this embodiment, formula (47) represents the node. The information status, starting from the initial moment of the planning cycle, must be linked to a subsequent update time. Formula (48) represents the expression for each monitoring point Information failure networks must eventually reach a virtual endpoint. Once, that is, from the last information update time. From the beginning until the end of the planning period. Formulas (49) to (51) are the core coupling constraints of the dual spatiotemporal network. Formula (49) represents the conversion of the path within the spatiotemporal network monitored by the UAV in multiple frequencies into the variables monitored by the nodes. Formula (50) represents the condition if the node... At any moment If an information update occurs, then the information becomes invalid. There must be one and only one arc in the network from time [time]. Enter. Formula (51) indicates that if the node At any moment If an information update occurs, then the information becomes invalid. There must be one and only one arc in the network from time [time]. Leaving essentially ensures that the drone's path and information failure are synchronized in time and space.

[0027] S3: By designing a path-reduction two-stage dual-spatiotemporal network optimization algorithm, the dual-spatiotemporal network model is solved, and the multi-frequency monitoring time, monitoring points, flight paths, and ice condition information update process of UAVs are collaboratively optimized to realize the planning process of multi-frequency UAV monitoring in frozen port waters considering ice condition changes. Specifically, in this embodiment, the selection of the inner arc of the dual-spatiotemporal coupled network includes the decision of the candidate start time of the monitoring task, the monitoring node and the monitoring order, and the spatiotemporal alignment with the information failure arc. Given the large number of spatiotemporal network arcs and the complexity of decision factors, this embodiment designs a path-reduction two-stage dual-spatiotemporal network optimization algorithm. Using an improved adaptive large neighborhood search as the core iterative framework, the complete solution process is divided into two stages: upper-level candidate space reduction and lower-level reduction dual-spatiotemporal network optimization. In this embodiment, the path-reduction two-stage dual-spatiotemporal network optimization algorithm first adopts a node priority-based approach. The greedy insertion strategy generates the first-stage upper-level solution, which includes candidate start times for monitoring tasks and monitoring nodes. Specifically, firstly, under the premise of satisfying constraints such as adjacent task energy replenishment and turnaround (refer to constraints (31)-(36)), candidate start times are constructed for each monitoring task; then, according to the nodes... The cumulative failure rate is calculated based on information loss due to long-term lack of monitoring, navigation demand weighting, and ice degradation rate. , ,in, It is used to reflect the importance of information at different nodes and time periods, and is reflected by mining the historical traffic and navigation flow of monitoring points. Represents a node At time step The ice condition information is invalid if it is valid, otherwise it is 0. Indicates time step End node The cumulative failure process of ice conditions, Additional penalty for exceeding the threshold; Represents a node At time step The single-step failure increment is calculated, and nodes are inserted into different monitoring tasks sequentially according to the failure severity level. During the node insertion process, the candidate insertion scheme is defined as placing nodes... Insert the first The first monitoring task There are 1 position, denoted as _ . First, determine whether the candidate insertion scheme meets the feasibility conditions such as mission duration, flight energy consumption, mission cost, and time continuity of adjacent missions (refer to constraints (31)-(36)).

[0028] , in, Representing nodes respectively Insert the first The first task After this location, the corresponding mission duration, flight energy consumption, and mission cost are... These represent the maximum duration, maximum energy consumption, and maximum cost of the task, respectively. Indicates the first insertion. The mission's return time, Indicates the first The moment the task begins. This represents the fixed refueling and turnaround time required between adjacent flights. For candidate insertion schemes that meet the above conditions, their comprehensive insertion evaluation value is further calculated, i.e., the reduction in marginal information failure loss: ,in Indicates completion After the insertion of the next node, the ice condition information corresponding to the current partial solution becomes invalid, resulting in cumulative losses. Indicates the first During the next insertion, the node Insert the first The first monitoring task The cumulative loss of ice condition information obtained after each location is considered. Finally, the scheme with the highest comprehensive insertion evaluation value is selected as the insertion result for the current node. , This represents the final insertion scheme. After insertion, a task-node accessibility matrix is ​​generated, and a small number of candidate nodes are retained to maintain the local optimization space in the second stage, ultimately forming the initial solution, which is the upper-level solution of the first stage. In the first stage, the corresponding starting arc is retained according to the candidate start time of each monitoring task, and the corresponding access arc is retained or closed according to the relationship between accessible and prohibited nodes within the task, thus forming a reduced set of candidate arcs in the spatiotemporal network. In the second stage, based on the reduced set of candidate arcs and the spatiotemporal network of ice condition information failure, the actual starting arc, access arc, return arc, and information failure interval arc are selected through a dual spatiotemporal network model, thereby determining the actual take-off time of the UAV, the access nodes, the access order, the return time, and the selection result of the ice condition information failure interval.

[0029] Specifically, the designed path-reduction two-stage dual-temporal network optimization algorithm includes: S31: The first stage of constructing a path-reduction type two-stage dual-temporal network optimization algorithm: defining the solution of the dual-temporal network model as... and , ;in, Indicates the first Candidate initiation arcs of a monitoring task in the spatiotemporal network of the UAV path; Indicates the first Candidate access arcs and prohibited access arcs within each monitoring task; and Represents a node In the Within a monitoring task, nodes are considered prohibited from access; otherwise, they are considered accessible. The set of candidate start arcs and access arcs for all tasks determines the start time of each task, the candidate access nodes within the task, and the forbidden access nodes. The upper-level solution is used to limit the candidate start arcs and candidate access arcs in the second-stage dual-spatiotemporal network. This embodiment also includes, during solution construction, a feasibility analysis of adjacent candidate start times to avoid significant temporal conflicts between adjacent monitoring tasks. Specifically, this involves analyzing the feasibility of adjacent tasks' start times. candidate start time Is the difference greater than or equal to Even if the candidate start times of adjacent monitoring tasks are at least separated by the UAV refueling and turnaround time. (Based on constraint 35), if the candidate start time does not meet the above conditions, then the time adjacent to the candidate time is selected sequentially for traversal, and the candidate time with the smallest difference that meets the conditions is selected. , The candidate times are adjacent to each other. The selected new candidate timestamps meet the criteria. This detection only guarantees the basic continuity of the time sequence of the upper-level candidate tasks; the actual flight duration within a task, the return time, and the complete time continuity of adjacent tasks are still determined by the second-stage dual-spatiotemporal network model.

[0030] Solution based on dual spatiotemporal network model Obtain the set of spatiotemporal network arcs for reducing UAV paths; and the method for obtaining the set of spatiotemporal network arcs for reducing UAV paths is as follows: S311: Solution based on the first stage Clearly define the task start time and the nodes that can and cannot be monitored, reduce the spatiotemporal arc paths in the dual networks that exceed the start time and are prohibited from access, based on the first The starting arc of each monitoring task The start time inside If the reduction is performed, a new starting arc will be formed. for: (52) in, For the initial arc The corresponding start time; S312: Solution to the dual-temporal network model In the middle, the first Candidate access arcs and forbidden access arcs within each monitoring task If the reduction is performed, the new access arc for: (53) in, Indicates access arc The corresponding visited node, Indicates closing the task All access nodes Access arcs are defined, otherwise they are reserved; S313: Based on S311 to S312, obtain the set of spatiotemporal network arcs for reducing the UAV path. : (54) in, Indicates the number of flights The set of arcs that have ended; in this embodiment, the arc set reduction process is equivalent to fixing the path selection variables corresponding to the prohibited starting arcs and access arcs to 0, that is... ,in, Indicates the first Whether the monitoring task selects the UAV path spatiotemporal arc For the retained candidate arcs, the second stage still needs to decide whether to actually select them based on the relevant constraints; S32: The second stage of the two-stage dual-temporal network optimization algorithm for path reduction: Based on the reduced UAV path temporal network arc set combined with the constraints corresponding to the dual-temporal network model, solve the dual-temporal coupled network model to obtain the UAV multi-frequency monitoring planning scheme for each monitoring task; and the UAV multi-frequency monitoring planning scheme includes, but is not limited to, the selection of monitoring tasks, candidate times, monitoring nodes, access order, and ice condition information failure intervals. Specifically, in this embodiment, the second stage of the path-reduction two-stage dual-spatiotemporal network optimization algorithm takes the reduced UAV path spatiotemporal network arc set obtained in the first stage as input, and solves the dual-spatiotemporal coupled network model on the reduced UAV path spatiotemporal network and the ice condition information failure spatiotemporal network. The second stage does not change the path integrity, flow conservation, flight time continuity, energy consumption, cost, refueling turnaround, and information failure coupling constraints in the original dual-spatiotemporal network model. Instead, it fixes the path selection variables corresponding to the candidate start arcs and candidate access arcs closed in the first stage to 0, and only performs path selection on the retained reduced arc set. In the second stage, the candidate start times and task-node accessibility relationships determined in the first stage serve as input conditions for the reduced dual-spatiotemporal network. Under these input conditions, the model further optimizes the selection of actual access arcs and information failure interval arcs, thereby obtaining the actual monitoring points, monitoring order, and task sequence for each monitoring task, as well as the information failure accumulation and reset process matching the actual monitoring completion time. The second stage also undertakes the fitness function calculation function, that is, demapping the first stage into a reduced dual-spatiotemporal network and solving its objective function value, which is used to evaluate the scheme quality and guide subsequent neighborhood search.

[0031] The aforementioned UAV multi-frequency monitoring planning scheme is defined as a candidate solution. Based on the adaptive large neighborhood search algorithm combined with simulated annealing, the candidate solutions are screened, and the optimal solution is output, thus obtaining the optimal scheme for UAV multi-frequency monitoring planning in frozen harbor waters considering ice condition changes. Specifically, in this embodiment, the screening of candidate solutions based on the adaptive large neighborhood search algorithm combined with simulated annealing includes the following steps: S321: In this embodiment, the perturbation object identification, candidate space perturbation, and lightweight rule repair are combined into a composite perturbation operator; and in each iteration, based on the failure feedback index of the statistical information of the second-stage solution results, and according to the selected information failure arc and node time period failure variables in the spatiotemporal network of ice condition information failure, the total information failure loss index is calculated. The algorithm sequentially selects the node with the largest information failure, the node with the largest unit information loss, the associated task corresponding to the largest information failure, and the task with the largest contribution to information failure. Specifically, it defines a composite perturbation operator based on the information failure feedback perturbation object and a composite perturbation operator based on the candidate space perturbation object based on the adaptive large neighborhood search algorithm. Furthermore, the composite perturbation operator based on the information failure feedback perturbation object: The objective function value of the dual spatiotemporal network model is used as the total information failure loss index; and based on the total information failure loss index, the node with the largest information failure, the node with the largest unit information loss, the associated task corresponding to the largest information failure, and the task pair with the largest contribution to information failure are selected in sequence. Maximum node of information failure : ; The node with the greatest information loss per unit : ; The corresponding task for the maximum information failure : ; The task pair with the largest contribution from information failure :

[0032] By using the total information failure index within the statistical information failure network, we can identify and obtain nodes or task pairs at different levels that cause high information failure, and set the perturbation objects at the node level and the task level: The node level includes the node with the largest information failure and the node with the largest unit information loss; the task level includes the associated task corresponding to the largest information failure and the task pair that contributes the most to the information failure. In this embodiment, multiple perturbation methods are designed based on the perturbation object, which are combined with the perturbation object to form a composite perturbation operator. The composite perturbation operator based on the candidate space perturbation object includes a node access perturbation operator, a node time perturbation operator, a node spatiotemporal perturbation operator, an associated task time perturbation operator, a task candidate node perturbation operator, and a task-to-spatiotemporal perturbation operator. The node access perturbation operator grants access permissions to the node with the largest unit information loss. ; Represents a node In the Within each monitoring task, nodes are accessible; that is, access is granted to several tasks whose periods of high loss are close to these nodes. This increases its chances of being selected for the second-stage path selection. The node-time perturbation operator: For the node with the largest information failure, obtain the preceding and following monitoring tasks associated with its corresponding information failure interval; for the preceding and following monitoring tasks, in the first-stage solution, release the candidate set of starting arcs corresponding to the task, that is, based on empirical values, replace the starting arc corresponding to the task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs. This allows the second-stage reduction of the dual-temporal network to autonomously select its actual takeoff time while satisfying constraints on path continuity, flight time, energy consumption, and refueling turnaround. The node spatiotemporal perturbation operator: For the associated task corresponding to the maximum information failure, by combining the node access perturbation operator and the node time perturbation operator, on the one hand, the target node access permission is opened in the monitoring tasks adjacent to or overlapping with its high information failure interval, and on the other hand, the candidate start time of the associated task is released, so that the accessibility relationship of the target node and the monitoring time enter the second stage of the path reduction type two-stage dual spatiotemporal network optimization algorithm together, thereby increasing the possibility that the node will be selected by the path and complete the information update; The associated task time perturbation operator: For the high-cost information failure arc corresponding to the node with the largest information failure, locate the monitoring tasks associated with its update times before and after, and use the associated monitoring tasks as candidate time release objects. That is, based on empirical values, replace the corresponding starting arc of the associated monitoring task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs. The second stage of the path reduction two-stage dual spatiotemporal network optimization algorithm reselects the actual takeoff time from the candidate start arc set of the associated monitoring tasks, so that the update time is closer to the period when information failure is rapidly increasing or high costs are accumulating. The task candidate node perturbation operator: For the associated task corresponding to the maximum information failure, grant access permissions to the corresponding high-failure node. This increases the chance of selecting high-failure nodes during the second phase. The task-to-spatiotemporal perturbation operator: For the task pair contributing the most to information failure or the associated task corresponding to the largest information failure, it combines the spatiotemporal perturbation of the task to the candidate node with the temporal perturbation of the associated task, and jointly opens access permissions for high-failure nodes in the related tasks. The algorithm releases the candidate start times for related tasks. Then, in the second stage of the path reduction two-stage dual spatiotemporal network optimization algorithm, the actual monitoring node, access order, and take-off time are jointly selected from the node access arc after opening and the candidate start arc after release, so as to improve the impact of task timing and node access relationship on information failure accumulation. In this embodiment, four types of information failure feedback disturbance objects (the node with the largest information failure, the node with the largest unit information loss, the associated task corresponding to the largest information failure, and the task pair with the largest information failure contribution) and their respective three types of candidate space disturbances (at the node level: node access disturbance, node time disturbance, node spatiotemporal disturbance; at the task level: associated task time disturbance, task candidate node disturbance, and task pair spatiotemporal disturbance) are combined to form twelve composite disturbance operators. These operators are used to adaptively select specific operators to execute during the iteration process based on operator scores and selection weights. Specifically, in S322, based on the current candidate solution fitness value and the current operator weights, a combination of the composite disturbance operators based on the information failure feedback disturbance object and the composite disturbance operators based on the candidate space disturbance object is selected. Specifically, let the set of composite disturbance operators be... , No. The selection weights, cumulative scores, and number of uses of the perturbation operator are as follows: Then the first The probability that the perturbation operator is selected in the current iteration. for: (55) In the formula: Indicates the first The current selection weights of the composite perturbation operator; A reward value is assigned to the selected perturbation operator based on the quality of the candidate solution, i.e., the fitness value of the current candidate solution. and update its cumulative rating. With the number of times it is used After each iteration update segment, obtain the first... The average reward of the perturbation operator is Then, the selection weights are updated: (56) (57) In the formula: This indicates that the reaction coefficient is updated by selecting weights. This indicates the lower limit of the average reward; If a certain type of operator achieves a high average reward in the current iteration segment, its subsequent selection probability increases; if its search performance is poor, its selection weight decreases, but through... Maintaining the minimum update level, after the weight update is complete, calculate the cumulative score within this iteration segment. Number of times used The process is cleared and proceeds to the next iteration. In this embodiment, only the perturbation of node access relationships is directly formed as a candidate solution through lightweight rules. The perturbation of the candidate start time is repaired and fitness evaluated by locally releasing the starting arc and calling the second-stage reduced bi-temporal network once. S322: Take the objective function value of the dual spatiotemporal network model corresponding to each candidate solution as the fitness value of the current candidate solution; based on the fitness value of the current candidate solution and according to the current operator weight, select the combination of each perturbation operator in the composite perturbation operator based on the information failure feedback perturbation object and the composite perturbation operator based on the candidate space perturbation object, and perturb the candidate solution according to the combined perturbation operator to obtain the latest candidate solution; S323: Obtain the fitness value of the latest candidate solution; confirm whether the latest candidate solution is accepted based on the simulated annealing acceptance criterion of the simulated annealing algorithm; if not, repeat step S322; if yes, confirm whether the preset number of iterations has been reached; if the preset number of iterations has been reached, output the optimal solution; if the preset number of iterations has not been reached, repeat step S322; thus obtaining the optimal scheme for planning the multi-frequency monitoring of frozen port waters by UAVs considering ice condition changes.

[0033] In this embodiment, during the iteration of perturbation operators, weight updates, and candidate solution selection, the algorithm maintains the selection weights and cumulative scores of each composite perturbation operator. In each iteration, the algorithm first extracts information failure feedback indicators based on the previous round's second-stage solution results. Then, based on the objective function value and the current operator weights, it selects a composite perturbation operator to perturb the candidate start times and task-node accessibility relationships in the first-stage solution, forming candidate solutions. After candidate solutions are formed, the algorithm only calls the second-stage reduced dual-spatiotemporal network model once to calculate fitness. Subsequently, simulated annealing acceptance criteria are used to select candidate solutions: if the candidate solution's fitness is better than the current solution, it is directly accepted; if the candidate solution's fitness is worse than the current solution, it is accepted with a certain probability based on the current temperature parameter and the degree of degradation of the target value. If a candidate solution is better than the historical best solution, better than the current solution, or accepted by the acceptance criteria, different rewards are given to the corresponding operators. After each set iteration segment, the selection weights of the operators are updated based on their cumulative scores, so that operators performing better in subsequent iterations receive a higher selection probability. As iterations proceed, the temperature parameter gradually decreases. The algorithm maintains strong search diversity in the early stages, but gradually reduces its acceptance of inferior candidate solutions in the later stages. The algorithm terminates when it reaches the maximum number of iterations, fails to improve for several consecutive rounds, or reaches the computation time limit, outputting the historical best solution. The algorithm flow is as follows: Figure 2 As shown.

[0034] This embodiment also includes the following application examples: Taking the nearshore navigation waters of a northern port in China as an example, this study considers the impact of winter sea ice on port navigation safety, the maintenance of tugboat icebreaking effectiveness, and the organization of vessel entry and exit. Monitoring points are mainly deployed at the outer edge of the main channel, the middle section of the main channel, the intersection of the main channel and the east-west channel, the outer edge of the east and west port entrances, the near edge of the east and west channels, and the sensitive waters south of the entrances where ice fragments are easily backfilled. The study focuses on key navigational points where ice fragments are easily backfilled, accumulated, and compressed after tugboat icebreaking, thus providing a basis for subsequent icebreaking scheduling, vessel entry and exit arrangements, and port winter navigation assurance.

[0035] The example uses 5-8 monitoring points and 1 shore-based node, with a planning period of 80 discrete time steps. A maximum of 5 UAV monitoring missions are scheduled, with candidate time intervals of 0-12, 16-28, 32-44, 48-60, and 64-76. The traffic demand weight for each node is determined by traffic flow time periods, with a weight value of 1-3. The UAV flight time is 1-5 minutes, the unit flight cost is 3.0, and the unit flight energy consumption is 2.0. Each monitoring node has a service time of 1 minute, service energy consumption of 1.5 minutes, and service cost of 1.0. The fixed cost per mission is 5.0, the maximum duration of a single mission is 15 minutes, the maximum energy consumption is 60.0, and the recharge turnaround time between adjacent missions is 5 minutes. Ice condition information degradation parameters are determined based on the previous planning period and historical statistical data, considering the non-homogeneous effective time for different nodes and time periods, ranging from 15-30 minutes, and the degradation rate is generated by its reciprocal. The initial information failure level is 0, and the state penalty coefficient after exceeding the failure threshold is 3.0.

[0036] The path-reduction two-stage dual-spatiotemporal network optimization algorithm uses 12 types of composite perturbation operators for iterative search. The perturbation objects include the node with the largest information failure, the node with the highest information failure per unit time, the task with the largest failure association, and task pairs contributing high information failure. Perturbation methods include access perturbation, time perturbation, and spatiotemporal joint perturbation. Operator weights employ an adaptive update mechanism. The initial selection weights of all 12 types of composite perturbation operators are set to 1.0, the initial cumulative score is 0, and the initial number of uses is 0. Each iteration uses a roulette wheel selection based on the current weights. If a candidate solution is better than the historical best solution, the operator reward value is 8.0; if a candidate solution is better than the current solution, the reward value is 4.0; if a candidate solution does not improve but is accepted by the simulated annealing criterion, the reward value is 1.0; if a candidate solution is not accepted, the reward value is 0.2. Every 10 generations constitutes an update segment. The selection weights of operators within this segment are updated based on their average reward. The weight update response coefficient is set to 0.25, meaning the new weights are determined by 75% of the original weights and 25% of the average reward. Simultaneously, to prevent the weights of certain operator types from degenerating to zero, the lower limit of the average reward is set to 0.05. After the update, the cumulative score and usage count for this segment are cleared, and the next iteration begins. The maximum number of iterations is set to 100, and the initial simulated annealing temperature is set to... ,in The objective function value is the current solution; if a candidate solution is better than the current solution, it is accepted directly; if a candidate solution is worse than the current solution, it is accepted according to... The probability is acceptable. During regular iterations, the temperature is gradually reduced with a cooling coefficient of 0.96; when there is no improvement after 5 consecutive generations, an enhanced search is initiated, maintaining the temperature at no less than 0.10. The algorithm aims to improve its ability to escape local optima. Termination conditions include reaching the maximum number of iterations, failing to achieve effective improvement for several consecutive iterations, or reaching a set computation time limit.

[0037] Table 1-3 shows the comparison results of the two-stage dual spatiotemporal network optimization algorithm with the Gurobi solver, ALNS, and LNS algorithms. The number of monitoring nodes in different examples is 5-8. In the comparison with the solver, the Gurobi solver can find the optimal solution within the specified time (3600 seconds), while the method described in this embodiment can obtain a high-quality feasible solution within 60 seconds, with a difference of less than 2% from the optimal solution, reflecting that the method described in this embodiment has extremely high solution efficiency. At the same time, there are ice condition failure nodes in different examples, indicating that the planned UAV monitoring task cannot completely and timely cover the ice condition change information of all points, but it can grasp the ice condition of all high-demand points according to different weights and control the ice condition coverage of other points, achieving basic coverage of ice condition change information.

[0038] Table 1. Comparison of the two-stage dual-temporal network optimization algorithm and the solver.

[0039] In both ALNS (Adaptive Large Neighborhood Search) and LNS (Large Neighborhood Search), the operator settings are random nodes, task pair visits, and candidate time perturbations. The former uses weight-based dynamic operator selection, while the latter uses fixed operator selection. By comparing with the method described in this embodiment, it can be found that the solution time is slightly faster than that of the method described in this embodiment, but the quality of the obtained planning scheme is lower. The maximum difference between the scheme obtained by the ALNS algorithm and that of the method described in this embodiment is 14.53%, and the maximum difference of LNS is 19.51%, indicating that their operator optimization capabilities are limited and both are trapped in local optima. At the same time, the number of ice condition failure points gradually increases under different calculation examples, indicating that the UAV monitoring planning scheme obtained by the comparative algorithms cannot effectively cover the ice condition change information of all nodes, resulting in significant information loss. The method described in this embodiment has made significant improvements over traditional algorithms in terms of customized perturbation operators and algorithm structure.

[0040] Table 2. Comparison results between the two-stage dual-temporal network optimization algorithm and ALNS

[0041] Table 3. Comparison results between the two-stage dual-temporal network optimization algorithm and LNS.

[0042] Specifically, taking the monitoring scheme of the 8 monitoring nodes in Example 4 during the planning period as an example, the feasibility of the solution obtained by the method described in this embodiment is verified. First, at the algorithm level, the method described in this embodiment is compared with the ALNS and LNS algorithms. The algorithm iteration curves are as follows: Figure 3As shown, since the operators can only change the quality of the current solution to a limited extent, the algorithm's iterative process converges slowly, and both ALNS and LNS get stuck in local optima. In contrast, the customized operator combination in the method described in this embodiment makes targeted improvements to nodes and task pairs with high information failure and high failure contribution, and combines a two-stage spatiotemporal network framework for search, so that the algorithm continuously seeks optimization and converges within 100 generations.

[0043] Figure 4 The cumulative usage of operators during the iteration process is shown. Among them, operators M4, M6, M8, and M12 have higher usage rates, namely the node with the highest information failure per unit time + access perturbation, the node with the highest information failure per unit time + spatiotemporal joint perturbation, the task with the largest failure association + time perturbation, and the task pair with high information failure contribution + spatiotemporal joint perturbation. This indicates that the perturbation related to candidate time under the unit time information failure of the task pair has a good effect, which verifies the effectiveness of customized operators.

[0044] Figure 5 The figure illustrates the number of path reductions in each iteration of the two-stage dual-spatiotemporal network optimization algorithm. As shown in the figure, the number of path reductions is significant in the early iterations, indicating that the algorithm can significantly compress the candidate arc size in the second-stage UAV path spatiotemporal network by utilizing the candidate start times of the first stage and the task-node accessibility relationship in the early search phase. As the iterations progress, the number of path reductions gradually stabilizes, indicating that the solution structure of the first stage gradually converges, and the candidate path space transferred to the second stage also tends to stabilize. This verifies that the candidate solutions obtained in the first stage can be effectively transformed into spatiotemporal network arc path reduction information and transferred to the second-stage dual-spatiotemporal network model for solving. This reduces the size of the second-stage model and improves the algorithm's solution efficiency while preserving path feasibility and information failure coupling.

[0045] At the solution level, the monitoring task planning obtained by the two-stage dual spatiotemporal network optimization algorithm is shown in Table 4. The monitoring start times of the five tasks are 0, 16, 32, 48 and 64 respectively. The number of monitoring nodes in each monitoring task is 6-7 per round. Only in task 3 is there a node ice condition information failure. The ice condition information of the other points is covered by the multi-frequency monitoring tasks. Figure 6The diagram illustrates the drone monitoring path within a single task. Due to factors such as the rate of change of node information across time periods, differences in the weight of node air traffic demand, and flight and energy consumption limitations, the drone monitoring paths for multiple tasks differ. Since multi-frequency monitoring is a continuous and dynamically updated process, the actual monitoring results of the previous task will change the information failure status of each node in subsequent time periods, thus affecting the selection and access order of monitoring nodes in the next task. Therefore, different monitoring tasks do not simply repeat fixed inspection paths, but rather coordinate adjustments based on the cumulative process of ice condition information failure within the planning period, the importance of node air traffic demand, and the feasibility of the drone path. As shown in the figure, the algorithm can form differentiated monitoring path arrangements in different tasks, ensuring that key monitoring nodes are updated in multiple time periods, while also taking into account constraints on drone energy consumption, flight distance, and task duration. This verifies the applicability of the method described in this embodiment to multi-frequency ice condition monitoring path planning in frozen port waters.

[0046] Table 4. Optimization results of the two-stage dual-temporal network optimization algorithm

[0047] Figure 7 The dynamic changes in ice condition information at eight nodes are illustrated. After a drone completes its monitoring task at a node, the node's information status is updated, and its information failure level is reset to 0. Subsequently, over time, the node's ice condition information gradually accumulates and fails at a non-homogeneous degradation rate. As shown in the figure, the slopes of the information failure curves differ for different nodes, indicating that each node has different information change rates at different times. Simultaneously, the reset times for different nodes also differ, reflecting that the algorithm differentiates the monitoring times and sequences based on the node's traffic demand weight, information failure status, and drone path constraints. Node 5 only briefly exceeded the failure threshold between the second and third monitoring tasks, and then quickly reset after the next monitoring, demonstrating that the algorithm can promptly identify and update high-failure nodes, preventing nodes from remaining in a failure state for extended periods. Overall, the method described in this embodiment can perform multiple information updates for key nodes during the planning period, keeping the information failure level of most nodes near or below the threshold, thus verifying the model's ability to characterize the non-homogeneous degradation of ice condition information across time periods and the information reset process after monitoring.

[0048] The method described in this embodiment is applicable to main channels, port entrances, channel intersections, and waterways sensitive to ice backfilling in ports affected by sea ice in winter. By dynamically monitoring the ice conditions at key navigation points, it provides a basis for decision-making regarding icebreaking operations, vessel entry and exit organization, and port navigation assurance. Compared with existing technologies, the beneficial effects are as follows: (1) To address the varying degrees of impact from different locations on navigation demand and ice condition changes in nearshore navigable waters of frozen ports, the method described in this embodiment identifies key monitoring nodes based on AIS navigation flow, channel layout, and historical ice condition data, and constructs a joint loss objective function for ice condition information degradation and failure weighted by navigation demand. This objective function considers both degradation loss before information failure and failure state loss after exceeding the failure threshold, transforming the monitoring scheme from traditional fixed inspections or simple coverage to optimization oriented towards navigation demand and the effectiveness of ice condition information. The method described in this embodiment can improve the pertinence and timeliness of ice condition monitoring in key navigable waters of frozen ports, reduce the accumulation of information failures caused by untimely monitoring at nodes with high navigation demand, and enable UAV monitoring resources to be rationally allocated according to navigation support needs and ice condition change characteristics, prioritizing the dynamic monitoring needs of key areas and critical periods. This enhances the continuity, accuracy, and support of ice condition information acquisition in frozen port waters, and improves its service capabilities for ship entry and exit organization, navigation safety control, and icebreaking support decision-making.

[0049] (2) Existing technologies mostly involve static monitoring at fixed locations and times. The method described in this embodiment determines the node information failure rate by time period based on historical ice condition data and monitoring results from the previous planning period, and constructs a mechanism for cross-time period non-homogeneous degradation and post-monitoring information reset. Furthermore, it constructs a spatiotemporal network for UAV paths and a spatiotemporal network for ice condition information failure, aligning UAV monitoring completion events and node information update events through network coupling constraints. This allows the UAV path selection, monitoring time decision-making, and information failure accumulation process to be collaboratively described in a unified dual spatiotemporal coupling network. The method described in this embodiment can consider the dynamic changes in frozen port ice condition information during the planning period, avoiding the simple view of ice condition information as a fixed validity period or static coverage state. This enables the UAV monitoring scheme to adjust the monitoring time and monitoring path in a timely manner according to the information failure changes of different nodes and time periods, thereby improving the timeliness of ice condition information updates.

[0050] (3) Existing solution techniques are inefficient. The method described in this embodiment designs a path-reduction two-stage dual-spatiotemporal network optimization algorithm: In the first stage, candidate starting arcs and task-node accessibility relationships are determined using upper-level candidate solutions, and arc reduction is performed on the UAV path spatiotemporal network; in the second stage, the actual monitoring path and information failure interval are solved in the reduced dual-spatiotemporal coupled network. Simultaneously, based on the information failure contribution and failure degree feedback obtained in the second stage, this invention identifies high-failure nodes, high-failure nodes per unit time, the most failed associated task, and high-contribution task pairs, and combines them with access perturbation, time perturbation, and spatiotemporal joint perturbation to form a customized composite perturbation operator. Combined with adaptive weight update and simulated annealing mechanisms, the search efficiency and solution quality are improved. The method described in this embodiment can improve the solution efficiency and solution quality of the multi-frequency UAV ice condition monitoring path planning problem. In response to the problem of highly coupled path selection and information failure processes under multiple tasks, multiple nodes, and multiple time periods, the method described in this embodiment can reduce the scale of candidate paths in the second stage while preserving the coupling relationship between path continuity, flight time, energy consumption, refueling turnaround, and information failure. It can also obtain better monitoring time, monitoring points, and monitoring path schemes through targeted searches.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for planning multi-frequency monitoring of frozen harbor waters using unmanned aerial vehicles (UAVs) considering ice condition changes, characterized in that: Specifically, the following steps are included: S1: Construct a path planning model for multi-frequency UAV ice condition monitoring; and the UAV multi-frequency ice condition monitoring path planning model includes an objective function and corresponding constraints; The objective function is used to minimize the loss of ice condition information at monitoring nodes in frozen port waters considering differences in navigation demand during the planning period. S2: Based on the aforementioned UAV multi-frequency ice condition monitoring path planning model, construct a dual spatiotemporal network model that couples UAV path and ice condition information; S3: By designing a path reduction-type two-stage dual spatiotemporal network optimization algorithm, the dual spatiotemporal network model is solved, and the multi-frequency monitoring time, monitoring points, flight paths and ice condition information update process of UAVs are coordinated and optimized to realize the planning process of multi-frequency UAV monitoring in frozen port waters considering ice condition changes.

2. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 1, characterized in that, The objective function in S1 corresponding to the UAV multi-frequency ice condition monitoring path planning model As shown in formula (1): （1） In the formula: The node number represents the monitoring point location. Represents the set of monitoring nodes and , This represents the discrete time within the planning period. Denotes the set of discrete time steps within the planning period and , This represents a weighting parameter used to reflect the importance of information at different nodes and time periods. Represents decision variables, Represents a node At time step The ice condition information is invalid, otherwise it is 0; Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the additional penalty coefficient after the cumulative failure threshold of ice conditions is exceeded.

3. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 2, characterized in that, The constraints of the UAV multi-frequency ice condition monitoring path planning model include the state accumulation constraint of ice condition information under the failure of cross-time period differentiated information, the frequency and path planning constraint of UAV monitoring, the candidate start time selection and monitoring node arrival time continuity constraint for each frequency under UAV multi-frequency monitoring, the return time, flight duration, power continuity and cost upper limit constraint for each frequency under UAV multi-frequency monitoring. Considering the cumulative state constraints of ice condition information under the failure of cross-time period differentiated information, as shown in formulas (2)-(11): （2） （3） （4） （5） （6） （7） （8） （9） （10） （11） In the formula: As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As decision variables, Represents a node In flights Time step within If monitoring has begun, then it is 0; otherwise, it is 0. Indicates the set of drone monitoring sorties and , Indicates at node Monitoring time, Indicates each monitoring point The maximum number of times monitored during the planning period Represents a node At time step The effective duration of ice condition information, Represents a node At time step Single-step failure increment, Indicates the node at the start of the planning period The initial cumulative failure process of ice conditions, Indicates time step End node The decision variables for the cumulative failure process of ice conditions. This represents the node at the beginning of the initial state, i.e., at the end of time step t=0. The cumulative failure process of ice conditions, Represents a node The cumulative ice condition failure process logic constraints use sufficiently large constants.

4. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 3, characterized in that, The method for constructing the dual spatiotemporal network model described in S2 specifically includes: S21: Acquire the spatiotemporal network of UAV multi-frequency monitoring and ;in, Indicates the first One sortie of drones monitored the spatiotemporal network. Represents a spatiotemporal network node and , Indicates a virtual source point. Indicates a virtual sink; This is a state node, representing the drone at the moment of the event. Located at node ; Represents the spatiotemporal network arc and , The starting arc represents the time when a candidate starting point is selected from the virtual source point. , Indicates the start time of the candidate monitoring task. This represents the access arc, which is the process by which a drone moves from a base or monitoring point to a node. And complete the node The monitoring process, Indicates the arc cost of the access arc; This represents the energy consumption of the access arc. The return arc represents the process by which the drone returns from the monitoring point to the base and ends its sortie. , These represent the arc cost and energy consumption of the return arc, respectively. S22: Obtaining Ice Condition Information Spatiotemporal Network and ;in, Multiple monitoring points The spatiotemporal network of ice condition information states constituted by the changes in ice condition information states; Represents a spatiotemporal network node. Represents a node At any moment Ice condition information status, Represents the spatiotemporal network arc of ice condition information status. Represents a node From the last monitoring time Until the next monitoring time The information invalidation interval between them; Arc cost represents the cumulative loss of information failure within the information failure interval, and includes the degree of continuous cumulative failure formed during non-homogeneous degradation and the additional penalty coefficient after exceeding the cumulative failure threshold of ice conditions; Indicates the time since the last monitoring Until the next monitoring time The information invalidation interval between them; S23: Spatiotemporal network of UAV multi-frequency monitoring obtained through S21 This describes the monitoring process of UAVs during different sorties, including takeoff, flight, monitoring, and return to base; and the spatiotemporal network of ice condition information acquired through S22. This describes the information failure accumulation process of each monitoring point between two adjacent information updates during the monitoring process, in order to construct a dual spatiotemporal network model; And the objective function of the dual spatiotemporal network model To minimize the cumulative information failure in the spatiotemporal network of ice condition information, as shown in formula (12): （12） In the formula: Represents the spatiotemporal network arc of ice condition information status. Represents decision variables, Represents a node Information failure network selection interval arc Conversely, it is 0; The constraints corresponding to the dual spatiotemporal network model include: The spatiotemporal network path constraints for multi-frequency monitoring by UAVs are shown in formulas (13)-(17): （13） （14） （15） （16） （17） In the formula: As decision variables, Indicates the first Each sortie selects the drone path spatiotemporal arc Conversely, it is 0; Indicates from virtual source The set of starting arcs and the virtual source point The set of arcs; Indicates entering the node The set of arcs and their branches from the nodes The set of starting arcs, Indicates the number of flights All access monitoring points The set of access arcs; The constraints on the number of UAV flights and their order in the spatiotemporal network for multi-frequency monitoring are shown in formulas (18)-(21): （18） （19） （20） （21） In the formula: Indicates the number of flights In time step The set of starting and ending arcs, As decision variables, Indicates the number of flights Selecting the UAV path spacetime arc Conversely, it is 0; This indicates the fixed refueling and turnaround time required between adjacent flights; This represents the operating cost and energy consumption of each arc within the complete path; The path integrity constraints within the spatiotemporal network of ice condition information and the coupling constraints of the dual spatiotemporal network are shown in formulas (22)-(26): （22） （23） （24） （25） （26） In the formula: As a decision variable, when When both values ​​are 1, they represent nodes respectively. Information failure network selection interval arc Conversely, it is 0; Indicates the number of flights All corresponding nodes In time step Completed monitoring of the set of access arcs, As decision variables, Represents a node At time step Monitoring is completed, and the ice condition information status is updated at that moment; otherwise, it is 0. As a decision variable, when When the value is 1, it represents the node. Information failure network selection interval arc Conversely, it is 0.

5. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 4, characterized in that, The path reduction two-stage dual spatiotemporal network optimization algorithm designed in S3 specifically includes: S31: The first stage of constructing a path-reduction type two-stage dual-temporal network optimization algorithm: defining the solution of the dual-temporal network model as... and , ;in, Indicates the first Candidate initiation arcs of a monitoring task in the spatiotemporal network of the UAV path; Indicates the first Candidate access arcs and prohibited access arcs within each monitoring task; and Represents a node In the Within a monitoring task, nodes are considered prohibited from access; otherwise, they are considered accessible. Represents the set of candidate starting arcs and the set of visiting arcs for all tasks; Solution based on dual spatiotemporal network model Obtain the set of spatiotemporal network arcs for reducing the drone path; Furthermore, the method for obtaining the set of spatiotemporal network arcs for reducing the drone path is as follows: S311: Solution to the dual-spatiotemporal network model In the middle, the first The starting arc of each monitoring task The start time inside If the reduction is performed, a new starting arc will be formed. for: （27） in, For the initial arc The corresponding start time; S312: Solution to the dual-temporal network model In the middle, the first Candidate access arcs and forbidden access arcs within each monitoring task If the reduction is performed, the new access arc for: （28） in, Indicates access arc The corresponding visited node, Indicates closing the task All access nodes Access arcs are defined, otherwise they are reserved; S313: Based on S311 to S312, obtain the set of spatiotemporal network arcs for reducing the UAV path. : （29） in, Indicates the number of flights The set of terminated arcs; S32: The second stage of the two-stage dual-temporal network optimization algorithm for path reduction: Based on the reduced UAV path temporal network arc set combined with the constraints corresponding to the dual-temporal network model, solve the dual-temporal coupled network model to obtain the UAV multi-frequency monitoring planning scheme for each monitoring task. Furthermore, the UAV multi-frequency monitoring planning scheme includes, but is not limited to, the selection of monitoring tasks, candidate times, monitoring nodes, access order, and ice condition information failure intervals; The UAV multi-frequency monitoring planning scheme is defined as a candidate solution; the candidate solutions are screened based on the adaptive large neighborhood search algorithm combined with the simulated annealing algorithm, and the optimal solution is output, which is the optimal scheme for UAV multi-frequency monitoring planning in frozen port waters considering ice condition changes.

6. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 5, characterized in that, The screening of candidate solutions in S32, based on the adaptive large neighborhood search algorithm combined with the simulated annealing algorithm, specifically includes the following steps: S321: Based on the adaptive large neighborhood search algorithm, define composite perturbation operators based on information failure feedback perturbation objects and composite perturbation operators based on candidate space perturbation objects; Furthermore, the composite perturbation operator based on the information failure feedback perturbation object: The objective function value of the dual spatiotemporal network model is used as the total information failure loss index; and based on the total information failure loss index, the node with the largest information failure, the node with the largest unit information loss, the associated task corresponding to the largest information failure, and the task pair with the largest contribution to information failure are selected in sequence. Maximum node of information failure : ; The node with the greatest information loss per unit : ; The most important information failure corresponds to the associated task. : ; The task pair with the largest contribution from information failure : By using the total information failure index within the statistical information failure network, we can identify and obtain nodes or task pairs at different levels that cause high information failure, and set the perturbation objects at the node level and the task level: The node level includes the node with the largest information failure and the node with the largest unit information loss; the task level includes the associated task corresponding to the largest information failure and the task pair that contributes the most to the information failure. The composite perturbation operator based on candidate space perturbation objects includes a node access perturbation operator, a node time perturbation operator, a node spatiotemporal perturbation operator, an associated task time perturbation operator, a task candidate node perturbation operator, and a task pair spatiotemporal perturbation operator. The node access perturbation operator grants access permissions to the node with the largest unit information loss. ; Represents a node In the Each monitoring task contains accessible nodes; The node-time perturbation operator: determines the preceding and following monitoring tasks corresponding to the information failure interval of the node with the largest information failure; for the preceding and following monitoring tasks, based on empirical values, replaces the starting arc corresponding to the task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs. This enables the second-stage reduced dual-temporal network to autonomously select the actual takeoff time from the candidate set of starting arcs, under the conditions of satisfying path continuity, flight time, energy consumption, and refueling turnaround constraints. The node spatiotemporal perturbation operator: for the associated task corresponding to the maximum information failure, it simultaneously performs node access perturbation and node time perturbation; The associated task time-perturbation operator: For the high-cost information failure arc corresponding to the node with the largest information failure, determine the associated monitoring tasks corresponding to the previous and next update times of the high-cost information failure arc, and replace the corresponding starting arc of the associated monitoring task in the first-stage solution with a candidate set of starting arcs containing multiple candidate starting arcs based on empirical values. This allows the second phase to reselect the actual takeoff time from the initial arc candidate set; The task candidate node perturbation operator: For the associated task corresponding to the maximum information failure, grant access permissions to the corresponding high-failure node. ; The task employs a spatiotemporal perturbation operator that simultaneously performs perturbations of the associated task time and perturbations of the task candidate nodes. S322: Take the objective function value of the dual spatiotemporal network model corresponding to each candidate solution as the fitness value of the current candidate solution; based on the fitness value of the current candidate solution and according to the current operator weight, select the combination of each perturbation operator in the composite perturbation operator based on the information failure feedback perturbation object and the composite perturbation operator based on the candidate space perturbation object, and perturb the candidate solution according to the combined perturbation operator to obtain the latest candidate solution; S323: Obtain the fitness value of the latest candidate solution corresponding to the latest candidate solution; confirm whether the latest candidate solution is accepted based on the simulated annealing acceptance criterion of the simulated annealing algorithm; if not, repeat step S322; if yes, confirm whether the preset number of iterations has been reached. If the preset number of iterations is reached, the optimal solution is output. If the preset number of iterations is not reached, step S322 is repeated; thus, the optimal solution for planning the multi-frequency monitoring of frozen port waters by UAVs considering ice condition changes is obtained.

7. The method for planning multi-frequency UAV monitoring of frozen harbor waters considering ice condition changes according to claim 6, characterized in that, In S322, based on the fitness value of the current candidate solution and the current operator weights, a combination of perturbation operators is selected from the composite perturbation operator based on the information failure feedback perturbation object and the composite perturbation operator based on the candidate space perturbation object. Specifically: Let the set of composite perturbation operators be... , No. The selection weights, cumulative scores, and number of uses of the perturbation operator are respectively: Then the first The probability that a perturbation operator is selected in the current iteration. for: （30） In the formula: Indicates the first The current selection weights of the composite perturbation operator; A reward value is assigned to the selected perturbation operator based on the quality of the candidate solution, i.e., the fitness value of the current candidate solution. and update its cumulative rating. With the number of times used After each iteration update segment, obtain the first... The average reward of the perturbation operator is Then, the selection weights are updated: （31） （32） In the formula: This indicates that the reaction coefficient is updated by selecting weights. This represents the lower limit of the average reward.

Images