An unmanned ship real-time scheduling control system

Abstract

The application belongs to the technical field of unmanned ship scheduling, and discloses an unmanned ship real-time scheduling control system; the system comprises: a region construction module, which pre-constructs the observation regions corresponding to N unmanned ships based on the environmental perception data and the state parameters of the N unmanned ships in a target observation sea area within a unit time; a first processing module, which extracts features from the environmental perception data in the target observation sea area within a unit time based on the observation regions corresponding to the N unmanned ships, and obtains the environmental perception feature data corresponding to the N unmanned ships; a second processing module, which extracts features from the state parameters of the N unmanned ships within a unit time, and obtains the state feature data corresponding to the N unmanned ships; and a scheduling decision module, which is used for inputting the environmental perception feature data and the state feature data corresponding to the N unmanned ships into a scheduling decision level setting model respectively, and obtaining the corresponding scheduling decision level; and the intelligent scheduling and collaborative control of multiple unmanned ships are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned ship scheduling, more particularly, to an unmanned ship real-time scheduling control system. BACKGROUND

[0002] With the deepening of tasks such as ocean resource survey, environmental monitoring and disaster warning, multi-point cooperative observation based on unmanned ship cluster has become an important development direction of ocean observation technology. The existing unmanned ship scheduling control system mainly uses task pre-distribution, path planning and static scheduling strategy to perform simple division based on the initial conditions of the task.

[0003] However, in actual application, the ocean environment has high dynamicity and uncertainty, and external environmental parameters such as wind speed, sea current, and solar radiation intensity fluctuate frequently. In addition, the unmanned ship will also deteriorate in the running process due to long-time sailing, resulting in problems such as yaw deviation, rapid decline in power, and unstable attitude. The existing scheduling control technology generally lacks dynamic perception and response capability to the above environmental changes and the running state of the unmanned ship, and cannot realize adaptive adjustment and fine distribution of the task scheduling strategy, resulting in some unmanned ships quitting the observation midway due to abnormal state during task execution, and further causing the observation data chain to be broken, the system load to be unbalanced, and the overall task continuity and observation stability of the cluster system to be affected.

[0004] Therefore, it is urgent to provide an intelligent scheduling control system that integrates multi-dimensional environmental perception and running state evaluation to improve the resource allocation accuracy, adaptive scheduling capability and task execution robustness of the unmanned ship cluster in complex and dynamic ocean environment. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical scheme: an unmanned ship real-time scheduling control system, comprising:

[0006] A region construction module, based on the environmental perception data in the target observation sea area within a unit time and the state parameters of N unmanned ships, pre-constructs the observation regions corresponding to the N unmanned ships;

[0007] A first processing module, based on the observation regions corresponding to the N unmanned ships, extracts features from the environmental perception data in the target observation sea area within a unit time, to obtain environmental perception feature data corresponding to the N unmanned ships;

[0008] A second processing module, based on the state parameters of the N unmanned ships within a unit time, extracts features to obtain state feature data corresponding to the N unmanned ships;

[0009] The scheduling decision module is configured to input the environment perception feature data and the state feature data corresponding to the N unmanned ships into a scheduling decision level setting model to obtain corresponding scheduling decision levels.

[0010] The intelligent scheduling module is configured to intelligently schedule the N unmanned ships based on the environment perception feature data, the state feature data, and the scheduling decision levels corresponding to the N unmanned ships.

[0011] Further, the scheduling decision levels include a first scheduling level, a second scheduling level, and a third scheduling level; and the method of intelligently scheduling the N unmanned ships includes:

[0012] The unmanned ships corresponding to the first scheduling level are constructed into a first scheduling unmanned ship set;

[0013] The unmanned ships corresponding to the second scheduling level are constructed into a second scheduling unmanned ship set, and the unmanned ships in the second scheduling unmanned ship set are intelligently scheduled;

[0014] The unmanned ships corresponding to the third scheduling level are constructed into a third scheduling unmanned ship set;

[0015] The environment perception feature data and the state feature data corresponding to the first scheduling unmanned ship set and the third scheduling unmanned ship set are analyzed and processed to obtain a replacement candidate triple set;

[0016] The unmanned ships in the first scheduling unmanned ship set and the second scheduling unmanned ship set are intelligently scheduled based on the replacement candidate triple set.

[0017] Further, the method of intelligently scheduling the unmanned ships in the second scheduling unmanned ship set includes:

[0018] The environment perception feature data and the state feature data corresponding to each unmanned ship in the second scheduling unmanned ship set are input into an unmanned ship parameter setting model to obtain corresponding unmanned ship target adjustment parameters, including propulsion power, rudder angle adjustment frequency, and attitude control accuracy; and the current state parameters of the unmanned ship are adjusted based on the unmanned ship target adjustment parameters to promote the unmanned ship from the second scheduling level to the third scheduling level; specifically as follows:

[0019] S700: The number of unmanned ships in the second scheduling unmanned ship set is denoted as EJ; the initial value of ej is 1, and the value range of ej is 1 to EJ;

[0020] S701: Obtain the e jth unmanned ship from the secondary dispatch unmanned ship set, input the environment perception feature data and state feature data of the e jth unmanned ship into the unmanned ship parameter setting model, obtain the unmanned ship adjustment parameter corresponding to the e jth unmanned ship, denoted as the unmanned ship target adjustment parameter; the unmanned ship adjustment parameter includes the propulsion power, the rudder angle adjustment frequency and the attitude control precision;

[0021] S702: Adjust the current state parameter of the e jth unmanned ship to the unmanned ship target adjustment parameter; make the unmanned ship rise from the secondary dispatch level to the tertiary dispatch level;

[0022] S703: Let e j = e j + 1, if e j is less than or equal to EJ, continue to execute S701 to S702; if e j is greater than EJ, end the current process.

[0023] Further, the obtaining method of the replacement candidate triple set comprises:

[0024] The environment perception feature data and state feature data corresponding to the primary dispatch unmanned ship and the tertiary dispatch unmanned ship are constructed into replacement adaptation degree data, and the replacement adaptation degree between each combination of the primary and tertiary unmanned ships is calculated according to the data;

[0025] According to the replacement adaptation degree result, a triple containing the serial number of the unmanned ship to be replaced, the serial number of the candidate replacement unmanned ship and the corresponding adaptation degree is constructed, and a replacement candidate triple set is formed; specifically as follows:

[0026] S500: Record the number of unmanned ships in the primary dispatch unmanned ship set as YJ, and the number of unmanned ships in the tertiary dispatch unmanned ship set as SJ; let the initial value of yj be 1, and the value range of yj be 1 to YJ; let the initial value of sj be 1, and the value range of sj be 1 to SJ;

[0027] S501: Construct the environment perception feature data and state feature data of the y jth unmanned ship in the primary dispatch unmanned ship set into first replacement adaptation degree data; construct the environment perception feature data and state feature data of the sjth unmanned ship in the tertiary dispatch unmanned ship set into second replacement adaptation degree data;

[0028] S502: Calculate the replacement adaptation degree of the y jth unmanned ship and the sjth unmanned ship based on the first replacement adaptation degree data and the second replacement adaptation degree data; construct the replacement candidate triple with yj, sj and the replacement adaptation degree, and add the replacement candidate triple to the replacement candidate triple set;

[0029] S503: Let sj=sj+1, if sj is less than or equal to SJ, continue to execute S501 to S502; if sj is greater than SJ, let yj=yj+1, if yj is less than or equal to YJ, let sj=1, and continue to execute S501 to S502, if yj is greater than YJ, get the replacement candidate triplet set, and end the current process.

[0030] Further, the method for intelligently scheduling unmanned ships in the primary scheduling unmanned ship set and the secondary scheduling unmanned ship set based on the replacement candidate triplet set comprises:

[0031] Obtaining the replacement candidate triplet with the highest replacement fitness, and recalculating the updated observation area of the two based on the observation area of the to-be-replaced unmanned ship and the candidate replacement unmanned ship in the replacement candidate triplet according to a preset observation area replacement ratio, as a scheduling optimization result;

[0032] Removing all candidates involved in the used replacement candidate triplet, and repeating the above operation on the remaining replacement candidate triplets until the replacement candidate triplet set is empty; specifically as follows:

[0033] S600: Obtaining the replacement candidate triplet with the highest replacement fitness from the replacement candidate triplet set, denoted as the current replacement candidate triplet; the first position of the replacement candidate triplet is the serial number of the to-be-replaced unmanned ship; the second position of the replacement candidate triplet is the serial number of the candidate replacement unmanned ship; and the third position of the replacement candidate triplet is the replacement fitness of the second position corresponding unmanned ship replacing the first position corresponding unmanned ship;

[0034] S601: Obtaining the observation area of the unmanned ship corresponding to the first position of the replacement candidate triplet, denoted as the first observation area; and obtaining the observation area of the unmanned ship corresponding to the second position of the replacement candidate triplet, denoted as the second observation area;

[0035] S602: Multiplying the first observation area by a preset observation area replacement ratio to obtain a shared observation area; subtracting the shared observation area from the first observation area to obtain a first updated observation area, which is the updated observation area of the unmanned ship corresponding to the first position of the replacement candidate triplet; and summing the second observation area and the shared observation area to obtain a second updated observation area, which is the updated observation area of the unmanned ship corresponding to the second position of the replacement candidate triplet;

[0036] S603: Removing all replacement candidate triplets containing the first position or the second position corresponding to the current replacement candidate triplet from the replacement candidate triplet set;

[0037] S604: Repeat S600-S603, stop when the replacement candidate triple set is empty.

[0038] Further, the method for pre-building the observation area corresponding to the N unmanned ships comprises:

[0039] Obtain environmental perception data in the target observation sea area, which includes sea current velocity field, sea current direction map, wind speed, wind direction, sea wave level, and solar radiation intensity;

[0040] Obtain the state parameters of the N unmanned ships, including real-time position coordinates, battery capacity, sailing speed, and heading;

[0041] Input the state parameters and environmental perception data corresponding to the N unmanned ships into the comprehensive feasibility evaluation model respectively to obtain the comprehensive feasibility scores corresponding to the N unmanned ships;

[0042] Normalize the comprehensive feasibility scores corresponding to the N unmanned ships to obtain the regional allocation factors corresponding to the N unmanned ships, and combine the total area of the target observation sea area to calculate the observation area that the N unmanned ships should undertake. The observation area that the N unmanned ships should undertake is the observation area corresponding to the N unmanned ships.

[0043] Further, the method for obtaining the environmental perception feature data corresponding to the N unmanned ships comprises:

[0044] Obtain the observation area environmental perception data corresponding to each unmanned ship, calculate the sea current velocity mean and the sea current velocity stability index from the sea current velocity field, and calculate the sea current direction mean and the sea current direction stability index from the sea current direction map;

[0045] Calculate the solar energy obtainable index from the solar radiation intensity, obtain the wind speed disturbance degree and wind direction jump frequency index from the wind speed and wind direction data, and obtain the wind energy disturbance level based on the pre-set wind energy disturbance level mapping table;

[0046] Construct the above calculated indexes into the environmental perception feature data of each unmanned ship; specifically as follows:

[0047] S100: Let the initial value of n be 1, and the value range of n be 1 to N; divide the unit time into G time points;

[0048] S101: Obtain the observation area environmental perception data corresponding to the nth unmanned ship; based on the sea current velocity field in the environmental perception data, calculate the corresponding sea current velocity mean and sea current velocity stability index; based on the sea current direction map in the environmental perception data, calculate the corresponding sea current direction mean and sea current direction stability index; based on the solar radiation intensity in the environmental perception data, calculate the corresponding solar energy obtainable index;

[0049] The wind speed disturbance degree corresponding to the wind speed is calculated based on the wind speed in the environmental perception data; the wind direction jump frequency index corresponding to the wind direction is calculated based on the wind direction in the environmental perception data; the wind energy disturbance level index is calculated based on the wind speed disturbance degree and the wind direction jump frequency index; and the wind energy disturbance level index is matched with the pre-constructed wind energy disturbance level mapping table to obtain the corresponding wind energy disturbance level.

[0050] S102: The mean sea current speed, the sea current speed stability index, the mean sea current direction, the sea current direction stability index, the solar energy obtainable index, and the wind energy disturbance level are constructed into the environmental perception feature data of the nth unmanned ship.

[0051] S103: Let n = n + 1, if n is less than or equal to N, continue to execute S101 to S102; if n is greater than N, the environmental perception feature data corresponding to N unmanned ships is obtained, and the current process is ended.

[0052] Further, the state feature data acquisition method of the N unmanned ships comprises:

[0053] The real-time state parameters of each unmanned ship are obtained, the unmanned ship deviation identifier is identified by analyzing the position coordinate change trend, the battery power change trend type is obtained by analyzing the battery power change trend, and the navigation stability index is calculated according to the sailing speed and the heading stability;

[0054] The state feature data corresponding to each unmanned ship is constructed by the unmanned ship deviation identifier, the battery power change trend type, and the navigation stability index; specifically as follows:

[0055] S200: Let the initial value of n be 1, and the value range of n be 1 to N; divide the unit time into G time points;

[0056] S201: Obtain the state parameters of the nth unmanned ship; analyze and process the position coordinate change trend based on the real-time position coordinates in the state parameters to obtain the unmanned ship deviation identifier of the nth unmanned ship; analyze and process the battery power change trend based on the battery power in the state parameters to obtain the battery power change trend type of the nth unmanned ship; and calculate the navigation stability index of the nth unmanned ship based on the sailing speed and the heading in the state parameters;

[0057] S202: The state feature data of the nth unmanned ship is constructed by the unmanned ship deviation identifier, the battery power change trend type, and the navigation stability index;

[0058] S203: Let n = n + 1, if n is less than or equal to N, continue to execute S201 to S202; if n is greater than N, the state feature data corresponding to N unmanned ships is obtained, and the current process is ended.

[0059] Further, the method for obtaining the unmanned ship deviation identifier of the nth unmanned ship comprises:

[0060] The nearest distance between the unmanned ship historical trajectory point and the observation area boundary is calculated, and the proportion of the out-of-bound trajectory point is counted. If the proportion exceeds the preset out-of-bound proportion threshold, the unmanned ship deviation identifier is set to yes, otherwise, it is set to no. Specifically, the following is performed:

[0061] S300: Obtain the observation area corresponding to the nth unmanned ship, extract the boundary coordinates corresponding to the observation area, set the initial value of g to 1, and the value range of g is 1 to G. Set the initial value of the out-of-bound counting variable to 0.

[0062] S301: Obtain the real-time position coordinates corresponding to the nth unmanned ship at the gth time point, calculate the Euclidean distance between the real-time position coordinates and all boundary coordinates of the observation area respectively, and select the smallest Euclidean distance as the unmanned ship boundary distance corresponding to the gth time point. The unmanned ship boundary distance refers to the nearest distance between the unmanned ship and the observation area boundary.

[0063] S302: If the unmanned ship boundary distance corresponding to the gth time point is greater than or equal to the preset boundary safety distance threshold, the out-of-bound counting variable is incremented by one.

[0064] S303: Set g = g + 1, if g is less than or equal to G, continue to perform S301 to S302, if g is greater than G, perform S304.

[0065] S304: Obtain the out-of-bound proportion by dividing the out-of-bound counting variable by G, if the out-of-bound proportion is greater than or equal to the preset out-of-bound proportion threshold, set the unmanned ship deviation identifier to yes, if the out-of-bound proportion is less than the preset out-of-bound proportion threshold, set the unmanned ship deviation identifier to no.

[0066] Further, the method for obtaining the power change trend type of the nth unmanned ship comprises:

[0067] According to the battery power at consecutive time points, the battery power change rate is calculated, and the preset battery power change rate threshold one and battery power change rate threshold two are combined to count the change times of stable power change, slow power decline and rapid power decline, respectively.

[0068] The change type with the most change times is taken as the power change trend type of the nth unmanned ship. Specifically, the following is performed:

[0069] S400: Let the initial value of t be 1, the value range of t is 1 to G-1; preset battery power change rate threshold one and battery power change rate threshold two, battery power change rate threshold one is less than battery power change rate threshold two; let the initial value of power stable change number, power slow decline change number and power rapid decline change number be 0;

[0070] S401: Obtain the battery power of the nth unmanned ship at the tth time point; obtain the battery power of the nth unmanned ship at the t+1th time point; calculate the battery power change rate between the tth and t+1th time points based on the battery power at the tth and t+1th time points;

[0071] S402: If the battery power change rate is less than the battery power change rate threshold one, then the power stable change number is increased by one; if the battery power change rate is greater than or equal to the battery power change rate threshold one and less than the battery power change rate threshold two, then the power slow decline change number is increased by one; if the battery power change rate is greater than or equal to the battery power change rate threshold two, then the power rapid decline change number is increased by one;

[0072] S403: Let t=t+1, if t is less than or equal to G-1, then continue to execute S401 to S402; if t is greater than G-1, then execute S404;

[0073] S404: If the power stable change number is the maximum, then the power change trend type is the power stable change type; if the power slow decline change number is the maximum, then the power change trend type is the power slow decline change type; if the power rapid decline change number is the maximum, then the power change trend type is the power rapid decline change type.

[0074] Compared with the prior art, the technical effects and advantages of the unmanned ship real-time scheduling control system are as follows:

[0075] The unmanned ship real-time scheduling control system provided by the application constructs an intelligent scheduling mechanism that integrates environmental perception features and state features, and can realize intelligent task allocation and collaborative control of multiple unmanned ships in a dynamic marine environment.

[0076] By deploying edge computing nodes, the system performs distributed collection and real-time analysis on environmental perception data such as current velocity field, current direction map, wind speed, wind direction, wave level and solar radiation intensity of the target observation sea area, and combines the real-time position, battery power, speed, heading and other operating states of the unmanned ship to generate environmental perception feature data and state feature data of each unmanned ship. The system further inputs the feature data into a comprehensive feasibility evaluation model constructed by a multilayer perceptron neural network to obtain a comprehensive feasibility score of each unmanned ship, thereby realizing dynamic observation area division based on capability difference.

[0077] On this basis, the unmanned ship running state is intelligently divided into multiple levels by a dispatch decision level setting model, and a differentiated dispatch response strategy is supported: for the first-level dispatch level unmanned ship, a replacement candidate triple set is constructed and a replacement fitness is calculated to realize task automatic transfer and relay dispatch; for the second-level dispatch level unmanned ship, a pre-trained machine learning model is used to generate a targeted control adjustment parameter, which is sent to a propulsion power, rudder angle control, attitude stabilization control and other execution units to make the unmanned ship running state tend to be stable and have the ability to be converted to a third-level dispatch level. Through the above steps, the application realizes a closed-loop intelligent dispatch mechanism of unmanned ship state perception, dispatch level determination, task load adjustment and state optimization feedback, which significantly improves the resource allocation accuracy, adaptive dispatching capability and task execution robustness of the system in a complex and unstructured marine environment. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 It is a real-time dispatch control system schematic diagram of an unmanned ship of embodiment 1 of the application;

[0079] Figure 2 It is a real-time dispatch control method flowchart of an unmanned ship of embodiment 3 of the application;

[0080] Figure 3 It is a real-time dispatch control system schematic diagram of an unmanned ship of embodiment 2 of the application;

[0081] Figure 4 It is a method flowchart for intelligently dispatching N unmanned ships;

[0082] Figure 5 It is a method flowchart for intelligently dispatching unmanned ships in a first-level dispatch unmanned ship set and a second-level dispatch unmanned ship set based on a replacement candidate triple set. DETAILED DESCRIPTION

[0083] The technical solutions in the embodiments of the application will be described in detail below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only used to better illustrate and describe the technical solutions of the application, and are intended to enable those skilled in the art to better understand and implement the application, and should not be understood as limiting the scope of protection of the application. Those skilled in the art can modify, adjust or equivalently replace them according to the content disclosed in the application without departing from the spirit and essence of the application, and these should be regarded as the protection scope of the application.

[0084] Embodiment 1

[0085] Please refer to Figure 1As shown, the embodiment discloses a real-time scheduling control system for unmanned ships, which comprises a region construction module, a first processing module, a second processing module, a scheduling decision module and an intelligent scheduling module, each module is connected through wired and / or wireless connection to realize data transmission.

[0086] The region construction module pre-constructs the observation region corresponding to the N unmanned ships based on the environmental perception data in the target observation sea area and the state parameters of the N unmanned ships within a unit time.

[0087] The method for pre-construction of the observation region corresponding to the N unmanned ships comprises:

[0088] The environmental perception data in the target observation sea area is obtained, which includes sea current velocity field, sea current direction diagram, wind speed, wind direction, sea wave level and solar radiation intensity.

[0089] It should be noted that the sea current velocity field and the sea current direction diagram can be perceived in real time by deploying buoys and ADCP (acoustic Doppler current profiler) devices in the target observation sea area. The sea current velocity field is a vector field data describing the flow velocity of seawater at different geographic coordinates in the target sea area, and the sea current direction diagram is an angle distribution diagram describing the flow direction of sea current in each region. The wind speed, wind direction and sea wave level are obtained through the marine weather station. The solar radiation intensity is measured in real time by the light sensor integrated on the top of the unmanned ship. The wind direction is expressed in angle, and the numerical range of the wind direction is 0° to 360°, which is used to indicate the direction angle of the wind direction, wherein 0° represents the north direction, 90° represents the east direction, 180° represents the south direction, and 270° represents the west direction, and the rest of the angles are used in the same way.

[0090] The state parameters of the N unmanned ships are obtained, which include real-time position coordinates, battery capacity, sailing speed and heading;

[0091] It should be noted that the real-time position coordinates are obtained through the global positioning system or Beidou navigation, the battery capacity is obtained through the battery management system of the unmanned ship, and the sailing speed and heading are obtained through the sailing control system of the unmanned ship. The heading is expressed in angle, and the numerical range of the heading is 0° to 360°, which is used to indicate the direction angle of the heading, wherein 0° represents the north direction, 90° represents the east direction, 180° represents the south direction, and 270° represents the west direction, and the rest of the angles are used in the same way.

[0092] The state parameters and environmental perception data corresponding to the N unmanned ships are input into the comprehensive feasibility evaluation model respectively to obtain the comprehensive feasibility score corresponding to the N unmanned ships; the higher the comprehensive feasibility score, the stronger the observation ability of the unmanned ship in the current environment, which can undertake more tasks and divide larger observation area.

[0093] The comprehensive feasibility scores corresponding to the N unmanned ships are normalized to obtain regional allocation factors corresponding to the N unmanned ships, and the observation area that the N unmanned ships should undertake is calculated in combination with the total area of the target observation sea area; the observation area that the N unmanned ships should undertake is the observation area corresponding to the N unmanned ships.

[0094] The method for obtaining the regional allocation factor comprises:

[0095]

[0096] FPYZ n is the regional allocation factor of the nth unmanned ship, ZHPF n is the comprehensive feasibility score of the nth unmanned ship, is the total sum of the comprehensive feasibility scores of the N unmanned ships, i is an index variable of the summation formula, ZHPF i is the comprehensive feasibility score of the ith unmanned ship.

[0097] The method for obtaining the observation area comprises:

[0098] GCMJ n = FPYZ n * ZMJ;

[0099] GCMJ n is the observation area of the nth unmanned ship, and ZMJ is the total area of the target observation sea area.

[0100] It should be noted that in an embodiment of the present application, in order to achieve efficient coverage of the target observation sea area and reasonable allocation of the task load of the unmanned ship, the system proposes an observation area division method based on the comprehensive feasibility score. Specifically, the system first calculates the comprehensive feasibility score of each of the N unmanned ships to be dispatched, which reflects the task execution ability and stability of each unmanned ship under the current environmental conditions and its own running state. The higher the comprehensive feasibility score, the better the environmental adaptability, the more stable the navigation state, and the more sustainable the energy support ability of the unmanned ship, which is suitable for undertaking larger observation task load.

[0101] In order to realize the dynamic division of the task area based on the capability difference, the system normalizes the comprehensive feasibility scores of all the unmanned ships to construct the regional allocation factor of the nth unmanned ship, which represents the score proportion of the unmanned ship in the cluster, i.e. the proportion weight it should undertake in the entire observation task. The larger the regional allocation factor, the stronger the capability of the unmanned ship, which is suitable for undertaking a larger observation area. Therefore, based on the regional allocation factor, the system further multiplies it with the total area of the target observation sea area to obtain the observation area corresponding to the nth unmanned ship.

[0102] Through the above derivation process, the feasibility score obtained by fusing multi-source features is converted into a basis for spatial division of the observation task area, so as to dynamically allocate the observation task area according to the comprehensive capability of each unmanned ship, and to make the task distribution and resource utilization of the system more reasonable, flexible and stable. This method not only avoids the waste or overload risk that may be caused by static average allocation, but also enhances the real-time scheduling capability and task completion rate of the system in a complex and dynamic environment.

[0103] The training method of the comprehensive feasibility evaluation model comprises:

[0104] A pre-constructed comprehensive feasibility evaluation dataset is provided, which comprises Y sets of comprehensive feasibility evaluation data and comprehensive feasibility scores corresponding to the Y sets of comprehensive feasibility evaluation data, Y being a positive integer greater than 0, the comprehensive feasibility evaluation data comprising environmental perception data and state parameters of the unmanned ship; the comprehensive feasibility evaluation dataset is divided into a comprehensive feasibility evaluation data training set and a comprehensive feasibility evaluation data verification set, wherein the comprehensive feasibility evaluation data training set is used for parameter learning of the comprehensive feasibility evaluation model, and the comprehensive feasibility evaluation data verification set is used for real-time evaluation of the generalization capability of the comprehensive feasibility evaluation model;

[0105] In the training process of the comprehensive feasibility evaluation model, a deep neural network structure based on a multilayer perceptron is used to convert the comprehensive feasibility evaluation data into a feature vector as input, extract nonlinear features in the data through a hidden layer, and finally generate a probability distribution of the comprehensive feasibility score in the output layer using a softmax activation function, and output the comprehensive feasibility score corresponding to the maximum probability as the final prediction result; the training process aims to minimize the cross-entropy loss function, and an early stopping strategy is introduced to monitor the performance of the comprehensive feasibility evaluation data verification set; when the prediction accuracy on the comprehensive feasibility evaluation data verification set reaches a preset accuracy, it is considered that the comprehensive feasibility evaluation model has converged, and the training is stopped.

[0106] It should be noted that when training the comprehensive feasibility evaluation model, the environment perception data includes sea current velocity field, sea current direction map, wind speed, wind direction, sea wave level and solar radiation intensity, and the state parameters of the unmanned ship include real-time position coordinates, battery power, sailing speed and heading; the environment perception data and the state parameters of the unmanned ship are normalized and quantized into parameters that can be used for machine learning model training. Specifically, to realize unified modeling and quantifiable processing of multi-dimensional heterogeneous data, the system normalizes various raw data into a set of standardized parameter features with consistent dimensions and dimensionless, and the quantized data has a value range of [0, 1]. When the comprehensive feasibility evaluation data set is pre-constructed, the comprehensive feasibility score can be evaluated by the person skilled in the art according to the normalized data, and the comprehensive feasibility score corresponding to each set of comprehensive feasibility evaluation data is obtained.

[0107] The range of the comprehensive feasibility score is also [0, 1], and if the score value approaches 1, it means that the current state of the unmanned ship is stable, the environmental adaptability is strong, and it has the priority to perform the main observation task; if the score value is in the middle interval (such as 0.4-0.7), it means that there is slight fluctuation in the running state or resource limitation, and the task load should be appropriately reduced; if the score value is less than 0.4, it means that the running ability is significantly weakened or the failure risk is high, and it is not suitable to continue to undertake the original task, and the relay or evacuation strategy should be triggered.

[0108] It should be noted that the environment perception data can comprehensively reflect the dynamic environmental characteristics of the target observation sea area, in which the sea current velocity and direction affect the sailing stability and energy consumption of the unmanned ship, the wind speed and direction affect the navigation deviation and propulsion efficiency, the sea wave level reflects the anti-disturbance ability requirement of sea surface operation, and the solar radiation intensity affects the effectiveness of solar power supply; the state parameters of the N unmanned ships can reflect the current operation ability, energy state and task scheduling response ability of each unmanned ship.

[0109] The environment perception data and the state parameters of the unmanned ship are combined with the comprehensive feasibility evaluation model to quantitatively score the feasibility of each unmanned ship in the current conditions to perform observation tasks, and the obtained comprehensive feasibility score can reflect the comprehensive ability level of the unmanned ship in the target sea area to carry out the task. Based on the comprehensive feasibility score, the observation area matching the ability of each unmanned ship is divided, and the dynamic balance of task load and optimization of resource utilization efficiency are realized.

[0110] The division of the observation area not only fully considers the environmental change characteristics of the target sea area and the dynamic performance difference of the unmanned ships, but also forms an exclusive operation range for each unmanned ship, which helps the subsequent path planning process to generate an optimal path in a limited area, reduces path conflicts and redundant intersections, and improves the accuracy of the path control and the convergence efficiency of the scheduling algorithm. In addition, based on the divided observation area, individual path planning can further combine key parameters such as the remaining power, sensor state, heading attitude, etc. of each unmanned ship to realize an adaptive control strategy for individual state, so that the task execution is more in line with the actual ability boundary, and the success rate and energy utilization efficiency of task completion are improved, and the multi-level collaborative optimization effect from "regional load division" to "individual path optimization" is realized as a whole.

[0111] The first processing module is configured to extract features from the environmental perception data in the target observation sea area per unit time based on the observation areas corresponding to the N unmanned ships, to obtain environmental perception feature data corresponding to the N unmanned ships;

[0112] The method for obtaining the environmental perception feature data corresponding to the N unmanned ships comprises:

[0113] S100: set the initial value of n to 1, and the value range of n is 1 to N; divide the unit time into G time points;

[0114] S101: obtain the environmental perception data corresponding to the observation area of the nth unmanned ship; calculate the corresponding mean current velocity and current velocity stability index based on the current velocity field in the environmental perception data; calculate the corresponding mean current direction and current direction stability index based on the current direction graph in the environmental perception data; calculate the corresponding solar energy availability index based on the solar radiation intensity in the environmental perception data;

[0115] Calculate the corresponding wind speed disturbance degree based on the wind speed in the environmental perception data; calculate the corresponding wind direction jump frequency index based on the wind direction in the environmental perception data; calculate the corresponding wind energy disturbance level index based on the wind speed disturbance degree and the wind direction jump frequency index; match the wind energy disturbance level index with the pre-constructed wind energy disturbance level mapping table to obtain the corresponding wind energy disturbance level;

[0116] S102: construct the mean current velocity, the current velocity stability index, the mean current direction, the current direction stability index, the solar energy availability index and the wind energy disturbance level into the environmental perception feature data of the nth unmanned ship;

[0117] S103: set n = n + 1, if n is less than or equal to N, continue to execute S101 to S102; if n is greater than N, obtain the environmental perception feature data corresponding to the N unmanned ships, and end the current process.

[0118] The method for calculating the mean value of the current velocity comprises:

[0119]

[0120] wherein, is the mean value of the current velocity, g is an index variable of the summation formula, the value range of g is 1 to G, and HLSD g is the current velocity corresponding to the gth time point.

[0121] The method for calculating the stability index of the current velocity comprises:

[0122]

[0123] wherein, HLWD is the stability index of the current velocity.

[0124] The method for calculating the mean value of the current direction comprises:

[0125]

[0126] wherein, is the mean value of the current direction, and HLFX g is the current direction corresponding to the gth time point.

[0127] The method for calculating the stability index of the current direction comprises:

[0128]

[0129] wherein, FXWD is the stability index of the current direction.

[0130] The method for calculating the disturbance degree of the wind speed comprises:

[0131]

[0132] wherein, FSRD is the disturbance degree of the wind speed, FS g is the wind speed corresponding to the gth time point, represents the mean value of the wind speed.

[0133] The method for calculating the frequency index of the wind direction jump comprises:

[0134]

[0135] wherein, TBPL is the frequency index of the wind direction jump, θ thresh is the jump threshold value (for example, the jump threshold value can be set to 15°), FX g is the wind direction corresponding to the gth time point, FX g-1 is the wind direction corresponding to the (g-1)th time point, and δ() is a logical judgment function.

[0136] If |FX g -FX g-1 |>θ thresh , then the corresponding value of δ(|FX g -FX g-1 |>θ thresh ) is 1, and if |FX g -FX g-1 |>θ thresh , then the corresponding value of δ(|FX g -FX g-1 |>θ thresh ) is 0.

[0137] The calculation method of the wind energy disturbance level index comprises:

[0138] WDI = a x FSRD + b x TBPL;

[0139] Wherein, WDI is the wind energy disturbance level index, a is the weighted coefficient of wind speed disturbance degree, b is the weighted coefficient of wind direction jump frequency index, a and b reflect the importance of wind speed disturbance degree and wind direction jump frequency index to the wind energy disturbance level index, for example, a can be set to 0.7, and b can be set to 0.3.

[0140] It should be noted that, in order to realize effective evaluation of the wind field disturbance degree in the target observation area, and improve the path planning and energy management capability of the unmanned ship scheduling system under complex weather conditions, the wind energy disturbance level index is designed based on wind speed change and wind direction instability. The wind energy disturbance level index is used to quantify the wind field disturbance intensity in unit time, and provides a key basis for unmanned ship task allocation, path selection and energy budget.

[0141] In the present application, in order to accurately depict the hydrodynamic and wind energy disturbance characteristics of the target observation sea area in unit time, a plurality of mathematical indexes for representing environmental state stability and change trend are constructed, specifically including sea current velocity mean, sea current velocity stability index, sea current direction mean, sea current direction stability index, wind speed disturbance degree and wind direction jump frequency index. The above indexes all take time series as input, and extract the environmental dynamic characteristics which have practical significance for unmanned ship observation and scheduling task through statistical or change rate calculation method.

[0142] The mean value of the current speed is used to measure the average current speed per unit time in the observation area where the unmanned ship is located, and reflects the influence of the current intensity on the ship body power compensation and propulsion strategy. The current speed stability index is constructed based on the standard deviation, and is used to describe the fluctuation degree of the current speed in the time dimension. The higher the index, the more intense the current disturbance, and the greater the stability of the ship and the energy consumption adjustment pressure. Similarly, the mean value of the current direction represents the trend of the main flow direction of the current, which helps to assist in judging the ship heading control and predicting the risk of drift. The current direction stability index reflects the direction jump degree, which can be used to identify vortex, cross flow or other complex flow direction interference environment, and then evaluate the risk level of the unmanned ship attitude control.

[0143] In addition, the wind speed disturbance degree is used to describe the disturbance intensity of the wind speed in the task period by calculating the standard deviation of the wind speed, which is an important quantitative index for evaluating the intensity of aerodynamic disturbance. The wind direction jump frequency index is constructed based on the adjacent difference value of the wind direction time series to construct a jump judgment function, and the ratio of the number of jumps to the unit time is used to quantify the suddenness of the wind field change, which can effectively identify whether the current is in a complex wind direction fluctuation section.

[0144] The above-mentioned multiple indexes are important components of the environmental perception feature data in the present application, which on the one hand provides a high-reliability input basis for the scheduling decision level setting model, and on the other hand can be used as a disturbance constraint condition input in the subsequent unmanned ship parameter adjustment model, which helps to improve the accuracy of the scheduling response and the overall anti-disturbance ability of the system.

[0145] An example of the wind energy disturbance level mapping table is shown in Table 1:

[0146] Table 1: Wind energy disturbance level mapping table

[0147] Wind Disturbance Index (WDI) Wind Disturbance Index 0 < WDI < 0.5 1 (Very Weak) 0.5 < WDI < 1.0 2 (Weak) 1.0 < WDI < 1.5 3 (Moderate) 1.5 < WDI < 2.0 4 (Strong) WDI > 2.0 5 (Strong Disturbance)

[0148] The method for obtaining the solar energy obtainable index comprises:

[0149]

[0150] Wherein, TYHQ is the solar energy obtainable index, FSQD is the solar radiation intensity, NJD curr is the current sea visibility, NJD ideal is the ideal sea visibility, SD curr is the current sea humidity, YL curr is the current sea cloud cover, ω1, ω2 and ω3 are corresponding weight coefficients, satisfying ω1+ω2+ω3=1, for example, ω1 can be set to 0.4, ω2 can be set to 0.3, and ω3 can be set to 0.3.

[0151] The visibility correction term represents the shielding strength of the visible environment. The lower the visibility, the stronger the shielding ability of suspended particles, water vapor and other light shielding in the air, and the lower the sunlight penetration rate. Multiply the visibility correction term by the weighting coefficient ω1 to reflect the contribution of ω1 to the overall shielding. The humidity correction term is the humidity correction term. As the humidity increases, the condensation, fogging and scattering effects of water vapor are enhanced, which inhibits the irradiance flux. Therefore, multiply SD by the weighting coefficient ω2 to obtain the second attenuation factor; ω3 x YL curr The humidity correction term is the humidity correction term. As the humidity increases, the condensation, fogging and scattering effects of water vapor are enhanced, which inhibits the irradiance flux. Therefore, multiply SD by the weighting coefficient ω2 to obtain the second attenuation factor; ω3 x YL curr The cloudiness correction term is the cloudiness correction term. Clouds are the most significant factor directly shielding sunlight. The greater the cloudiness, the more serious the direct irradiation is blocked.

[0152] The visibility correction term, humidity correction term and cloudiness correction term are combined into a total shielding influence value. The ratio factor of effective solar energy is represented by 1-total shielding influence value. Finally, multiply the ratio factor of effective solar energy by FSQD to obtain the solar energy obtainable index. Not only can it comprehensively reflect the multi-dimensional interference of various environmental factors on photovoltaic power supply, but also has good flexibility and real-time performance, which can provide reliable illumination environment input support for path selection, power budget, return judgment and the like of unmanned ships in the dispatching period.

[0153] The second processing module performs feature extraction based on the state parameters of the N unmanned ships in a unit time to obtain state feature data corresponding to the N unmanned ships; the state feature data is used to assist in judging whether task re-allocation or path scheduling optimization of the unmanned ships is needed.

[0154] The method for obtaining the state feature data corresponding to the N unmanned ships comprises:

[0155] S200: Let the initial value of n be 1, and the value range of n be 1 to N; divide the unit time into G time points;

[0156] S201: Obtain the state parameters of the nth unmanned ship; analyze and process the position coordinate change trend based on the real-time position coordinates in the state parameters to obtain the unmanned ship deviation identifier of the nth unmanned ship; analyze and process the battery power change trend based on the battery power in the state parameters to obtain the power change trend type of the nth unmanned ship; calculate the navigation stability index of the nth unmanned ship based on the sailing speed and the heading in the state parameters;

[0157] S202: Construct the state feature data of the nth unmanned ship from the unmanned ship deviation identifier, the power change trend type and the navigation stability index;

[0158] S203: Let n=n+1, if n is less than or equal to N, continue to execute S201 to S202; if n is greater than N, obtain the state feature data corresponding to the N unmanned ships, and end the current process.

[0159] The method for obtaining the unmanned ship deviation identifier of the nth unmanned ship comprises the following steps:

[0160] S300: An observation area corresponding to the nth unmanned ship is obtained, and boundary coordinates corresponding to the observation area are extracted; an initial value of g is 1, and a value range of g is 1 to G; an initial value of a boundary-exceeding counting variable is 0;

[0161] S301: Real-time position coordinates corresponding to the nth unmanned ship at a gth time point are obtained, and the real-time position coordinates and all boundary coordinates of the observation area are respectively subjected to Euclidean distance calculation, and the smallest Euclidean distance is selected as an unmanned ship boundary distance corresponding to the gth time point, wherein the unmanned ship boundary distance refers to a distance between the unmanned ship and the boundary of the observation area closest to the unmanned ship;

[0162] S302: If the unmanned ship boundary distance corresponding to the gth time point is greater than or equal to a preset boundary safety distance threshold, the boundary-exceeding counting variable is increased by 1;

[0163] S303: g is set to g+1, and if g is less than or equal to G, the steps S301 to S302 are continuously executed; if g is greater than G, the step S304 is executed;

[0164] S304: A boundary-exceeding proportion is obtained by performing a division operation on the boundary-exceeding counting variable and G, and if the boundary-exceeding proportion is greater than or equal to a preset boundary-exceeding proportion threshold, the unmanned ship deviation identifier is set to yes; if the boundary-exceeding proportion is less than the preset boundary-exceeding proportion threshold, the unmanned ship deviation identifier is set to no.

[0165] It should be noted that the boundary safety distance threshold is used to determine whether the unmanned ship is within a safe distance of the observation area, that is, the distance between the unmanned ship and the boundary of the observation area needs to be less than the boundary safety distance threshold, and the boundary safety distance threshold can be set by a person skilled in the art according to the area of the observation area and the volume of the unmanned ship, for example, the boundary safety distance threshold is set to 20 meters; the boundary-exceeding proportion threshold is set by a person skilled in the art, for example, the boundary-exceeding proportion threshold can be set to 0.5. By constructing the unmanned ship deviation identifier, that is, whether the unmanned ship is close to or exceeds the boundary of the observation area for a long time within a unit time, the rapid identification of whether the unmanned ship has a boundary deviation trend in the observation area is realized, and then risk intervention is performed in advance.

[0166] The method for obtaining the power change trend type of the nth unmanned ship comprises the following steps:

[0167] S400: An initial value of t is 1, and a value range of t is 1 to G-1; a preset battery power change rate threshold one and a battery power change rate threshold two are set, and the battery power change rate threshold one is less than the battery power change rate threshold two; initial values of a power stable change number, a power slow decline change number and a power rapid decline change number are all 0;

[0168] S401: Obtain the battery power of the nth unmanned ship at the tth time point; obtain the battery power of the nth unmanned ship at the t+1th time point; calculate the battery power change rate between the tth and t+1th time points based on the battery powers at the tth and t+1th time points;

[0169] S402: If the battery power change rate is less than the first battery power change rate threshold, increase the number of stable power changes by one; if the battery power change rate is greater than or equal to the first battery power change rate threshold and less than the second battery power change rate threshold, increase the number of slow power decrease changes by one; if the battery power change rate is greater than or equal to the second battery power change rate threshold, increase the number of rapid power decrease changes by one.

[0170] It should be noted that the first battery power change rate threshold is used to determine whether the power change is slow enough, that is, if the power decrease rate in a certain time period is lower than the first battery power change rate threshold, it is considered that the current energy consumption is basically stable; the second battery power change rate threshold is used to define the critical boundary between "slow decrease" and "rapid decrease", if the power decrease rate in a certain time period is higher than the second battery power change rate threshold, it is considered that the current energy consumption is too fast, which may be caused by high load or system abnormality. The first battery power change rate threshold and the second battery power change rate threshold are set by a person skilled in the art according to the upper limit of the safe discharge rate supported by the battery type and environmental factors.

[0171] For example, the average battery decrease rate of a certain model of unmanned ship under standard speed and medium task load conditions is 0.3% / min; under strong interference, high wind and high load tasks, the average battery decrease rate is 1.2% / min. The first battery power change rate threshold can be set to 0.5% / min to identify the stable power state, and the second battery power change rate threshold can be set to 1.0% / min to determine whether to enter the rapid power decrease state.

[0172] S403: Let t = t + 1, if t is less than or equal to G-1, continue to execute S401 to S402; if t is greater than G-1, execute S404;

[0173] S404: Compare the sizes of the number of stable power changes, the number of slow power decrease changes and the number of rapid power decrease changes, if the number of stable power changes is the largest, the power change trend type is the stable power change type; if the number of slow power decrease changes is the largest, the power change trend type is the slow power decrease change type; if the number of rapid power decrease changes is the largest, the power change trend type is the rapid power decrease change type.

[0174] The method for calculating the battery power change rate comprises:

[0175]

[0176] wherein, DDL t，t+1 is the battery power change rate between the tth and (t+1)th time points, DCDL t+1 is the battery power at the (t+1)th time point, DCDL t is the battery power at the tth time point, and Δt is the time interval between the tth and (t+1)th time points.

[0177] It should be noted that by continuously analyzing the battery power change rates of the N unmanned ships in a unit of time over multiple time periods, and in combination with the set change rate classification threshold, a power change trend type judgment mechanism is constructed, which can effectively identify the power consumption trend of the current unmanned ship during task execution. The complex continuous energy consumption behavior is abstracted into three types of "stable", "slowly decreasing" or "rapidly decreasing", thereby providing an efficient, clear and predictive input basis for subsequent decision-making of task scheduling, return judgment, energy compensation path optimization and the like of the unmanned ship.

[0178] The calculation method of the navigation stability index of the nth unmanned ship includes:

[0179] DHWD n = λ1×SBZC n + λ2×XBZC n ;

[0180]

[0181] wherein, DHWD n is the navigation stability index of the nth unmanned ship, SBZC n is the variance of the sailing speed of the nth unmanned ship, XBZC n is the variance of the heading of the nth unmanned ship, λ1 is the weighted coefficient of the variance of the sailing speed, and λ2 is the weighted coefficient of the variance of the heading, HXSD g is the sailing speed at the gth time point, HX g is the heading at the gth time point, and λ1 and λ2 reflect the importance of the variance of the sailing speed and the variance of the heading to the navigation stability index. For example, λ1 can be set to 0.6, and λ2 can be set to 0.4. The navigation stability index is used to evaluate whether the heading of the unmanned ship is stable and whether there is a significant running deviation caused by sea conditions or wind interference.

[0182] The scheduling decision module is configured to input the environment perception feature data and the state feature data corresponding to the N unmanned ships into a scheduling decision level setting model to obtain corresponding scheduling decision levels. The scheduling decision levels include a first scheduling level, a second scheduling level, and a third scheduling level. The first scheduling level indicates that the unmanned ship is currently in an unsustainable observation state, and there are problems such as serious energy consumption risk, regional deviation, or stability anomaly, and a scheduling response operation needs to be performed immediately. The second scheduling level indicates that the unmanned ship is currently in a runnable but unstable state, and there are certain risk trends or local abnormalities, and slight scheduling is needed. The third scheduling level indicates that the unmanned ship is currently in a stable running state, and the environment and state conditions are good, and it is suitable to continue to perform the current observation task.

[0183] The training method of the scheduling decision level setting model includes:

[0184] A scheduling decision level setting data set is collected in advance. The scheduling decision level setting data set includes DJ group scheduling decision level setting data and scheduling decision levels corresponding to the DJ group scheduling decision level setting data. DJ is a positive integer greater than 0. The scheduling decision level setting data includes environment perception feature data and state feature data. The scheduling decision level setting data set is divided into a training set and a validation set. The training set is used to train the scheduling decision level setting model, and the validation set is used to evaluate the generalization performance of the scheduling decision level setting model.

[0185] During the training of the scheduling decision level setting model, the cross-entropy loss function is minimized as the optimization target, the performance of the validation set is monitored using the early stopping strategy, the model performance is optimized by continuously adjusting the network parameters, and when the prediction accuracy on the validation set reaches the expected accuracy, it is considered that the scheduling decision level setting model has converged, and the training is stopped. The scheduling decision level setting model is trained using a deep neural network based on a multilayer perceptron.

[0186] The scheduling decision level setting data is converted into a feature vector. The input layer of the scheduling decision level setting model receives the feature vector, extracts the nonlinear relationship in the data through the hidden layer, and finally the output layer of the scheduling decision level setting model calculates the probability distribution of the scheduling decision level through the softmax activation function, and outputs the scheduling decision level corresponding to the maximum probability as the final prediction result.

[0187] It should be noted that, in an embodiment of the present application, in order to realize intelligent judgment of scheduling levels of multiple unmanned ships under complex environmental conditions, the system constructs a scheduling decision level setting model based on environmental perception feature data and state feature data. The model is trained by supervised learning, the input end includes a multi-dimensional feature vector, and the output end is a corresponding scheduling decision level label, which is used to classify the current scheduling control level of the unmanned ship. When training the scheduling decision level setting model, the environmental perception feature data includes mean sea current speed, sea current speed stability index, mean sea current direction, sea current direction stability index, solar energy availability index, and wind energy disturbance level; the state feature data includes unmanned ship deviation identifier, power change trend type, and navigation stability index.

[0188] The mean sea current speed reflects the resistance level in the propulsion process; the sea current speed stability index is used to measure the fluctuation amplitude of the flow rate; the mean sea current direction is used to evaluate the need for course correction; the sea current direction stability index is used to reflect the direction disturbance intensity; the solar energy availability index is used to reflect the future endurance of the unmanned ship; the wind energy disturbance level is used to evaluate the interference degree of the wind field on the attitude and track; the unmanned ship deviation identifier is used to indicate whether there is a trend of crossing the border or approaching the border; the power change trend type is used to indicate the change rate of the battery power in the continuous time period (such as stable, slow decline, rapid decline); the navigation stability index is used to measure the stability of the heading, and the higher the value, the more unstable the attitude.

[0189] The environmental perception feature data and the state feature data are both normalized to the numerical interval [0, 1] or mapped into a classification vector to construct a standard input feature vector. The system inputs the feature vectors of multiple samples and the corresponding labels (scheduling levels: first-level scheduling, second-level scheduling, and third-level scheduling) into the scheduling decision level setting model for training. After the training of the scheduling decision level setting model is completed, only the real-time extraction of the current environmental and state features of the unmanned ship is required in the running stage, and the corresponding scheduling decision level at the current time can be output by inputting the model. The scheduling decision level is used to guide whether to perform regional adjustment, parameter adjustment, or task relay on the current unmanned ship, ensuring that the scheduling control has state responsiveness, environmental adaptability, and classification operability.

[0190] It should be further noted that the scheduling decision level setting model is based on a preset multi-dimensional evaluation criterion and hierarchical judgment logic, and jointly analyzes the key features such as environmental adaptability, energy state, navigation stability, and spatial position deviation of each unmanned ship, and outputs the corresponding scheduling decision level of each unmanned ship. The scheduling decision level is used to comprehensively evaluate the current task adaptability and running stability of each unmanned ship.

[0191] In the present application, the scheduling decision level setting model as the core intelligent judgment component, based on the environmental perception feature data and the state feature data of the unmanned ship in the observation area, dynamically judges whether each unmanned ship is currently suitable to continue to perform the observation task, and outputs the corresponding scheduling level (first-level scheduling, second-level scheduling or third-level scheduling) accordingly. The introduction of this model breaks through the traditional scheduling mode based on static rules or fixed threshold setting, and can realize the adaptive identification and intelligent grading of the running state of the unmanned ship in the complex dynamic marine environment.

[0192] Specifically, through quantitative analysis of the environmental perception feature data and the state feature data, the scheduling decision level setting model can comprehensively evaluate whether a single ship has running risks such as rapid energy consumption decline, navigation instability or observation deviation, so as to accurately identify the scheduling level of immediate replacement (first-level scheduling), mild adjustment (second-level scheduling) or suitable operation (third-level scheduling).

[0193] The deployment of the scheduling decision level setting model not only improves the timeliness and accuracy of the system in identifying abnormal ships, but also provides clear and structured input basis for subsequent replacement scheduling or parameter adjustment strategies, realizes the change of unmanned ship cluster scheduling from "static preset" to "dynamic closed loop", and significantly enhances the stable operation ability and resource coordination efficiency of the system in complex and volatile marine environments.

[0194] The intelligent scheduling module intelligently schedules the N unmanned ships based on the corresponding environmental perception feature data, state feature data and scheduling decision level of the N unmanned ships.

[0195] As shown in Figure 4 The method for intelligently scheduling the N unmanned ships includes:

[0196] The unmanned ships corresponding to the first-level scheduling level are constructed into a first-level scheduling unmanned ship set;

[0197] The unmanned ships corresponding to the second-level scheduling level are constructed into a second-level scheduling unmanned ship set, and the unmanned ships in the second-level scheduling unmanned ship set are intelligently scheduled;

[0198] The unmanned ships corresponding to the third-level scheduling level are constructed into a third-level scheduling unmanned ship set;

[0199] The corresponding environmental perception feature data and state feature data are analyzed and processed based on the first-level scheduling unmanned ship set and the third-level scheduling unmanned ship set, and a replacement candidate triple set is obtained;

[0200] The unmanned ships in the first-level scheduling unmanned ship set and the second-level scheduling unmanned ship set are intelligently scheduled based on the replacement candidate triple set.

[0201] The obtaining method of the succession candidate triplet set comprises:

[0202] S500: the number of unmanned ships in the first-level scheduling unmanned ship set is denoted as YJ, and the number of unmanned ships in the third-level scheduling unmanned ship set is denoted as SJ; the initial value of yj is 1, and the value range of yj is 1 to YJ; the initial value of sj is 1, and the value range of sj is 1 to SJ;

[0203] S501: the environment perception feature data and the state feature data of the yjth unmanned ship in the first-level scheduling unmanned ship set are constructed into first succession adaptation degree data; and the environment perception feature data and the state feature data of the sth unmanned ship in the third-level scheduling unmanned ship set are constructed into second succession adaptation degree data;

[0204] S502: the succession adaptation degree of the yjth unmanned ship and the sth unmanned ship is calculated based on the first succession adaptation degree data and the second succession adaptation degree data; yj, sj and the succession adaptation degree are constructed into a succession candidate triplet, and the succession candidate triplet is added to the succession candidate triplet set;

[0205] S503: sj is set to sj+1, if sj is less than or equal to SJ, S501 to S502 are continuously executed; if sj is greater than SJ, yj is set to yj+1, if yj is less than or equal to YJ, sj is set to 1, and S501 to S502 are continuously executed, if yj is greater than YJ, the succession candidate triplet set is obtained, and the current process is ended.

[0206] The succession adaptation degree of the yjth unmanned ship and the sth unmanned ship is calculated based on the first succession adaptation degree data and the second succession adaptation degree data; yj, sj and the succession adaptation degree are constructed into a succession candidate triplet, and the succession candidate triplet is added to the succession candidate triplet set;

[0207]

[0208] SPD yj，sj is the succession adaptation degree of the yjth unmanned ship and the sth unmanned ship, ZXJL(yj, sj) represents the distance between the yjth unmanned ship and the sth unmanned ship, TYHQ sj represents the solar energy obtainable index of the sth unmanned ship, WDI sj represents the wind energy disturbance level index of the sth unmanned ship, DHWD sj represents the navigation stability index of the sth unmanned ship. μ1, μ2, μ3 and μ4 are weighting coefficients, and μ1+μ2+μ3+μ4=1. For example, μ1 can be set to 0.25, μ2 can be set to 0.25, μ3 can be set to 0.25, and μ4 can be set to 0.25.

[0209] As a distance factor (positive correlation), the closer the distance between the yth unmanned ship and the sth unmanned ship, the shorter the task migration path and the lower the scheduling response time. μ2×TYHQ sj As a solar energy factor (positive correlation), TYHQ sj Reflects the energy supply capability of the sth unmanned ship during the relay task, the better the light condition and the stronger the power supply capability, the longer the sth unmanned ship can support the task, so the larger the solar energy factor, the higher the fitness. μ3×WDI sj As a wind field interference factor (negative correlation), the larger the wind field interference factor, the stronger the wind disturbance, the lower the path tracking and heading stability, and the worse the fitness, so the wind field interference factor is a negative coefficient. μ4×DHWD sj As a navigation stability factor (negative correlation): DHWD sj The larger, the more unstable the navigation, the higher the task risk, and the lower the fitness, so the navigation stability factor is a negative coefficient. By combining the above distance factor, solar energy factor, wind field interference factor and navigation stability factor, a multi-dimensional evaluation of the candidate unmanned ship is formed, which makes the task migration strategy more suitable for the comprehensive operation capability of the unmanned ship and the current environmental conditions.

[0210] As Figure 5 shown, the method for intelligently scheduling the unmanned ships in the primary scheduling unmanned ship set and the secondary scheduling unmanned ship set based on the replacement candidate triple set comprises:

[0211] S600: Obtain the replacement candidate triple with the highest replacement fitness from the replacement candidate triple set, and denote it as the current replacement candidate triple; the first position of the replacement candidate triple is the serial number of the unmanned ship to be replaced; the second position of the replacement candidate triple is the serial number of the candidate replacement unmanned ship; and the third position of the replacement candidate triple is the replacement fitness when the unmanned ship corresponding to the second position replaces the unmanned ship corresponding to the first position;

[0212] S601: Obtain the observation area of the unmanned ship corresponding to the first position of the replacement candidate triple, and denote it as the first observation area; and obtain the observation area of the unmanned ship corresponding to the second position of the replacement candidate triple, and denote it as the second observation area;

[0213] S602: multiply the first observation area by a preset observation area replacement ratio (for example, the observation area replacement ratio can be set to one fifth) to obtain an allocated observation area; subtract the allocated observation area from the first observation area to obtain a first updated observation area, which is the updated observation area of the first unmanned ship corresponding to the first position of the replacement candidate triple; sum the second observation area and the allocated observation area to obtain a second updated observation area, which is the updated observation area of the second unmanned ship corresponding to the second position of the replacement candidate triple;

[0214] Optionally, the system can dynamically adjust the replacement ratio according to the state characteristics such as the power and stability of the relay unmanned ship, so as to improve the sustainability and load balancing degree of task relay.

[0215] S603: remove all replacement candidate triples containing the first position or the second position corresponding to the current replacement candidate triple from the replacement candidate triple set, to prevent a single unmanned ship from being repeatedly involved in dispatch relay;

[0216] S604: repeat S600 to S603, and stop when the replacement candidate triple set is empty.

[0217] The method for intelligently scheduling unmanned ships in the secondary scheduling unmanned ship set comprises:

[0218] S700: record the number of unmanned ships in the secondary scheduling unmanned ship set as EJ; set the initial value of ej as 1, and the value range of ej as 1 to EJ;

[0219] S701: obtain the ejth unmanned ship from the secondary scheduling unmanned ship set, input the environmental perception feature data and state feature data of the ejth unmanned ship into the unmanned ship parameter setting model to obtain the unmanned ship adjustment parameter corresponding to the ejth unmanned ship, which is recorded as the unmanned ship target adjustment parameter; the unmanned ship adjustment parameter includes propulsion power, rudder angle adjustment frequency and attitude control accuracy;

[0220] S702: adjust the current state parameter of the ejth unmanned ship to the unmanned ship target adjustment parameter; thereby promoting the unmanned ship to rise from the secondary scheduling level to the tertiary scheduling level, i.e. from the runnable but unstable state to the stable state;

[0221] S703: set ej = ej + 1, if ej is less than or equal to EJ, continue to execute S701 to S702; if ej is greater than EJ, end the current process.

[0222] The training method of the unmanned ship parameter setting model comprises:

[0223] A pre-constructed unmanned ship parameter setting data set is provided, the unmanned ship parameter setting data set including CS group unmanned ship parameter setting data and unmanned ship adjustment parameters corresponding to the CS group unmanned ship parameter setting data, CS being a positive integer greater than 0, the unmanned ship parameter setting data including environmental perception feature data and state feature data; the unmanned ship parameter setting data set is divided into an unmanned ship parameter setting data training set and an unmanned ship parameter setting data verification set, wherein the unmanned ship parameter setting data training set is used for parameter learning of an unmanned ship parameter setting model, and the unmanned ship parameter setting data verification set is used for real-time evaluation of the generalization ability of the unmanned ship parameter setting model.

[0224] In the training process of the unmanned ship parameter setting model, a deep neural network structure based on a multilayer perceptron is adopted, the unmanned ship parameter setting data is converted into a feature vector as input, nonlinear features in the data are extracted through a hidden layer, and finally a probability distribution of the unmanned ship adjustment parameter is generated by using a softmax activation function in an output layer, and the unmanned ship adjustment parameter corresponding to the maximum probability is output as the final prediction result; the training process aims to minimize the cross-entropy loss function, and an early stopping strategy is introduced to monitor the performance of the unmanned ship parameter setting data verification set; when the prediction accuracy on the unmanned ship parameter setting data verification set reaches a preset accuracy, it is considered that the unmanned ship parameter setting model has converged, and the training is stopped.

[0225] It should be noted that the propulsion power controls the speed of the unmanned ship and the energy consumption per unit time, and high speed and high frequency start will cause the battery power to decrease rapidly, causing the unmanned ship to be in an energy risk state. The rudder angle adjustment frequency directly affects the navigation stability index, i.e., the direction fluctuation during navigation, and frequent and large amplitude changes in the rudder angle will cause the ship body to swing and the route to oscillate, especially in the wind and wave area, which will be judged as "unstable navigation". In the wind and wave interference area, the ship body of the unmanned ship will tilt slightly or sway, affecting the instrument stability and navigation stability of the unmanned ship, and if the sampling period corresponding to the attitude control accuracy is too long or the correction frequency is too low, it will be judged as "unstable running".

[0226] Embodiment 2

[0227] Please refer to Figure 3 The embodiment provides an unmanned ship real-time scheduling control system, and further comprises:

[0228] The proportion setting module inputs the environmental perception feature data and the state feature data of the unmanned ship to be replaced into the replacement proportion setting model to obtain a dynamically set observation area replacement proportion.

[0229] The training method of the replacement proportion setting model comprises:

[0230] Pre-construct a replacement ratio setting data set, the replacement ratio setting data set includes X sets of replacement ratio setting data and the corresponding observation area replacement ratio of X sets of replacement ratio setting data, X is a positive integer greater than 0, the replacement ratio setting data includes environmental perception feature data and state feature data; the replacement ratio setting data set is divided into a replacement ratio setting data training set and a replacement ratio setting data validation set, wherein the replacement ratio setting data training set is used for parameter learning of the replacement ratio setting model, and the replacement ratio setting data validation set is used for real-time evaluation of the generalization ability of the replacement ratio setting model;

[0231] In the replacement ratio setting model training process, a deep neural network structure based on a multilayer perceptron is used to convert the replacement ratio setting data into a feature vector as input, extract nonlinear features in the data through a hidden layer, and finally generate a probability distribution of the observation area replacement ratio in the output layer using a softmax activation function. The output of the maximum probability is used as the final prediction result of the observation area replacement ratio; the training process aims to minimize the cross-entropy loss function, and an early stopping strategy is introduced to monitor the performance of the replacement ratio setting data validation set. When the prediction accuracy on the replacement ratio setting data validation set reaches the preset accuracy, it is considered that the replacement ratio setting model has converged, and the training is stopped.

[0232] Embodiment 3

[0233] Please refer to Figure 2 As shown in the figure, the embodiment provides a real-time scheduling control method for unmanned ships, which includes:

[0234] Based on the environmental perception data in the target observation sea area and the state parameters of N unmanned ships in a unit time, the observation areas corresponding to the N unmanned ships are pre-constructed;

[0235] Based on the observation areas corresponding to the N unmanned ships, the environmental perception data in the target observation sea area in a unit time is feature extracted to obtain the environmental perception feature data corresponding to the N unmanned ships;

[0236] Based on the state parameters of the N unmanned ships in a unit time, feature extraction is performed to obtain the state feature data corresponding to the N unmanned ships;

[0237] The environmental perception feature data and the state feature data corresponding to the N unmanned ships are input into a scheduling decision level setting model respectively to obtain the corresponding scheduling decision level;

[0238] Based on the environmental perception feature data, the state feature data and the scheduling decision level corresponding to the N unmanned ships, the N unmanned ships are intelligently scheduled.

[0239] The above merely describes specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any modification or substitution within the technical range disclosed by the present application can be easily thought of by those skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0240] Finally, the above merely describes preferred embodiments of the present application, and is not used to limit the present application, and any modification, equivalent substitution, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A real-time scheduling and control system for unmanned vessels, characterized in that, include: The region construction module pre-constructs the observation area corresponding to N unmanned vessels based on the environmental perception data within the target observation sea area and the state parameters of N unmanned vessels within a unit of time. The first processing module extracts features from the environmental perception data of the target sea area within a unit of time based on the observation areas corresponding to N unmanned vessels, and obtains the environmental perception feature data corresponding to N unmanned vessels. The second processing module extracts features based on the state parameters of N unmanned vessels per unit time to obtain the state feature data corresponding to the N unmanned vessels. The scheduling decision module is used to input the environmental perception feature data and state feature data of N unmanned vessels into the scheduling decision level setting model to obtain the corresponding scheduling decision level. The scheduling decision levels include Level 1 scheduling, Level 2 scheduling, and Level 3 scheduling. The intelligent scheduling module performs intelligent scheduling of N unmanned vessels based on their environmental perception feature data, status feature data, and scheduling decision level. Methods for intelligent scheduling of N unmanned vessels include: The unmanned vessels corresponding to the first-level scheduling decision level are constructed into a set of first-level scheduling unmanned vessels; The unmanned vessels corresponding to the second-level scheduling decision level are constructed into a second-level scheduling unmanned vessel set, and intelligent scheduling is performed on the unmanned vessels in the second-level scheduling unmanned vessel set; The unmanned vessels corresponding to the scheduling decision level of Level 3 are constructed into a set of Level 3 scheduling unmanned vessels; Based on the first-level and third-level scheduling unmanned vessel sets, the corresponding environmental perception feature data and state feature data are analyzed and processed to obtain the successor candidate triplet set. Intelligent scheduling of unmanned vessels in the first-level and second-level scheduling unmanned vessel sets is performed based on the successor candidate triple set.

2. The unmanned vessel real-time scheduling and control system according to claim 1, characterized in that, Methods for intelligent scheduling of unmanned vessels in a secondary scheduling unmanned vessel ensemble include: For each unmanned vessel in the set of unmanned vessels under secondary scheduling, the corresponding environmental perception feature data and state feature data are input into the unmanned vessel parameter setting model to obtain the corresponding unmanned vessel target adjustment parameters, which include propulsion power, rudder angle adjustment frequency and attitude control accuracy. Based on the unmanned vessel target adjustment parameters, the current state parameters of the unmanned vessel are adjusted so that the unmanned vessel is upgraded from the secondary scheduling level to the tertiary scheduling level.

3. The unmanned vessel real-time scheduling and control system according to claim 1, characterized in that, The method for obtaining the set of successor candidate triples includes: The environmental perception feature data and state feature data corresponding to the first-level and third-level unmanned ships respectively are used to construct the succession adaptation data, and the succession adaptation between each group of first-level and third-level unmanned ships is calculated based on the data. Based on the succession suitability results, a triplet is constructed containing the serial number of the unmanned vessel to be replaced, the serial number of the candidate unmanned vessel to be replaced, and the corresponding suitability, thus forming a set of candidate succession triplets.

4. The unmanned vessel real-time scheduling and control system according to claim 1, characterized in that, Methods for intelligent scheduling of unmanned surface vessels (USVs) in the first-level and second-level scheduling USV sets based on the successor candidate triplet set include: Obtain the replacement candidate triplet with the highest replacement suitability. Based on the observation area of ​​the unmanned vessel to be replaced and the candidate unmanned vessel to be replaced contained in the replacement candidate triplet, recalculate the updated observation area of ​​the two according to the preset observation area replacement ratio, and use them as the scheduling optimization results respectively. Remove all candidates involved in the used replacement candidate triplet, and repeat the above operation for the remaining replacement candidate triplets until the set of replacement candidate triplets is empty.

5. The unmanned vessel real-time scheduling and control system according to claim 1, characterized in that, Methods for pre-constructing the observation areas corresponding to N unmanned surface vessels include: Acquire environmental perception data within the target observation sea area, including ocean current velocity field, ocean current direction map, wind speed, wind direction, wave level, and solar radiation intensity; Obtain the status parameters of N unmanned vessels, including real-time position coordinates, battery level, speed, and heading; The state parameters and environmental perception data of N unmanned vessels are input into the comprehensive feasibility assessment model to obtain the comprehensive feasibility score of N unmanned vessels. The comprehensive feasibility scores corresponding to N unmanned vessels are normalized to obtain the regional allocation factors corresponding to N unmanned vessels. Combined with the total area of ​​the target observation sea area, the observation area to be undertaken by N unmanned vessels is calculated. The observation area to be undertaken by N unmanned vessels is the observation area corresponding to N unmanned vessels.

6. The unmanned vessel real-time scheduling and control system according to claim 1, characterized in that, Methods for obtaining environmental perception feature data for N unmanned surface vessels include: The system acquires environmental perception data of the observation area corresponding to each unmanned vessel, calculates the mean ocean current velocity and ocean current velocity stability index by analyzing the ocean current velocity field, and calculates the mean ocean current direction and ocean current direction stability index by analyzing the ocean current direction map. The solar energy availability index is calculated based on solar radiation intensity, the wind speed disturbance degree and wind direction jump frequency index are obtained based on wind speed and wind direction data, and the wind energy disturbance level is obtained based on the preset wind energy disturbance level mapping table. The calculated indicators are used to construct environmental perception feature data for each unmanned vessel.

7. The real-time scheduling and control system for unmanned vessels according to claim 1, characterized in that, Methods for obtaining the state feature data corresponding to N unmanned vessels include: The system acquires real-time status parameters of each unmanned vessel, identifies deviation indicators of the unmanned vessel by analyzing the trend of position coordinate changes, determines the type of power change trend by analyzing the trend of battery power changes, and calculates navigation stability indicators based on sailing speed and heading stability. The deviation markers, battery level change trends, and navigation stability indicators of each unmanned vessel are used to construct the state characteristic data corresponding to each unmanned vessel.

8. The unmanned vessel real-time scheduling and control system according to claim 7, characterized in that, The methods for obtaining the deviation marker of the nth unmanned surface vessel include: The closest distance between the historical trajectory point of the unmanned vessel and the boundary of its observation area is calculated, and the proportion of trajectory points that cross the boundary is counted. If the proportion exceeds the preset threshold for the proportion of the unmanned vessel that crosses the boundary, the deviation flag of the unmanned vessel is set to yes; otherwise, it is set to no.

9. The unmanned vessel real-time scheduling and control system according to claim 8, characterized in that, Methods for obtaining the battery power change trend type of the nth unmanned vessel include: Based on the battery charge at continuous time points, calculate the battery charge change rate. Combined with the preset battery charge change rate threshold one and battery charge change rate threshold two, count the number of changes in charge that are stable, slow, and rapid. The type of change with the most frequent changes is taken as the type of power change trend for the nth unmanned vessel.

Images