Intelligent decision support system for unmanned plane autonomous guarantee platform on unmanned ship

CN118960736APending Publication Date: 2024-11-15HARBIN ENG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411013269.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing autonomous drone support technology has a low degree of automation in a ship-based environment, making it difficult to achieve high-density unmanned aerial vehicle loading and autonomous guarantee, affecting the endurance and service life of the drone.

Method used

Design an intelligent decision support system for unmanned boats, including an intelligent decision-making system for drones, an autonomous guarantee platform intelligent decision-making control system and a cloud platform. Through neural network algorithms and heuristic search algorithms, the automated takeoff, landing and Transfer, optimize the outbound takeoff path planning, select the most suitable drone and perform automation support.

Benefits of technology

It realizes high-density loading and autonomous guarantee of drones in specific spaces of unmanned boats, improves the degree of automation and unmanned drones, extends the cruise time of drones, reduces losses, and enhances the system Survival ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118960736A_ABST
    Figure CN118960736A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of cooperative operation of unmanned ships and unmanned aerial vehicles, and particularly relates to an intelligent decision support system for an unmanned aerial vehicle autonomous guarantee platform on an unmanned ship. According to the invention, for the unmanned aerial vehicle with a specific model and size, high-density loading and autonomous guarantee are realized in a space with a specific size of the unmanned surface vehicle, and the unmanned aerial vehicle is automatically transferred and lifted to the autonomous guarantee platform in an unmanned manner after landing. According to the invention, a heuristic search and reinforcement learning fusion algorithm for unmanned aerial vehicle autonomous guarantee task planning is provided, and the optimal path planning of takeoff and landing is realized; and designing an aircraft which can automatically select the most suitable aircraft with the highest score through related influence factors and alterable weight proportions thereof, and adjusting and perfecting the neural network system in combination with later specific actual flight data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of cooperative operation of unmanned boats and unmanned aerial vehicles, and in particular relates to an intelligent decision support system for an autonomous support platform of an unmanned aerial vehicle on an unmanned boat. Background Art

[0002] In recent years, with the rapid development of drone-related technologies, drones have been applied to various fields. Nowadays, drones are playing an increasingly important role in various scenarios such as reconnaissance, surveillance, strike, logistics and transportation. Drones and unmanned boats have become research hotspots due to their unique advantages in ocean exploration. As a carrier for drones to perform maritime exploration missions, unmanned boats provide drones with various autonomous support services (such as refueling, charging, parking, maintenance, etc.). This type of autonomous support service system is an important part of the unmanned boat system and is of great significance for realizing unmanned operations such as maritime exploration and inspection.

[0003] With the development of science and technology towards informatization and intelligence, the role of UAV systems in reconnaissance and detection has become increasingly prominent. However, traditional single-machine detection execution is difficult to meet today's needs, and multi-machine coordination has become the focus of research. By loading UAVs with high density on unmanned boats, the requirements of multiple UAVs coordinating and cooperating to perform tasks can be achieved. Through the mutual cooperation between the UAVs and the various functional modules of the hangar of the unmanned boat, a large number of UAVs can be deployed in a short time to perform tasks, thereby improving execution efficiency.

[0004] For drone systems to be able to perform tasks continuously and at a high density, shipboard support technology is also crucial. By refueling, charging, preheating, and storing drones through autonomous support platforms, the cruising time of drones can be greatly extended, the loss of drones can be reduced, and the survivability of the system can be enhanced. At present, research on autonomous support for drones is relatively weak. Therefore, research on autonomous support technology for drones is of great significance for improving the endurance of drones and the service life of drones.

[0005] With the development of technology, the research on UAV loading systems is gradually oriented towards integration, automation and intelligence. From the traditional separation of take-off and recovery platforms, manual control and adjustment, and manual health detection, refueling, and support of UAVs, to the gradual integration of take-off, loading, recovery, and support, coupled with unmanned, automated, and intelligent control of the entire system, reducing manual participation has greatly improved efficiency and has good application prospects. The traditional separation of take-off and landing, recovery systems, and support detection systems, each part is manually connected. In the process of preparing for UAV launch and recovery and support detection, a large number of collaborative operators are required, and the operation steps are cumbersome. Its work efficiency and safety depend to a large extent on human control.

[0006] At present, countries around the world are actively promoting the development of drones. European and American countries such as the United States, Germany, Israel, Canada, and Japan have developed automated drone airports, which are widely used in communication facility inspections, logistics distribution, fire monitoring, emergency search and rescue, and other purposes.

[0007] For example, Canada's SkyX has developed SkyOne, a drone for surveying, and a drone airport. The SkyOne vertical take-off and landing drone is equipped with a beautiful clamshell hangar and an automatic airport. Figure 1 The main purpose is to inspect oil pipelines and natural gas pipelines. The automatic airport is arranged along the oil pipeline for SkyOne drones to land and charge. After the airport guides the aircraft to land, it will automatically close the hangar protection cover to keep warm and prevent wind, and charge the drone wirelessly.

[0008] German transport company DHL has developed a drone airport, Parcelcopter Skyport, with a take-off and landing platform on top of the central airport, covered by a dome. Figure 2 When the drone approaches, the dome opens and the drone lands on the platform. The Parcelcopter drone can complete the tasks of automatic unloading and automatic reloading, and then the drone will fly to the next destination. The whole process is fully automated and does not require human intervention.

[0009] In addition, some domestic commercial companies have also begun to conduct extensive research on small drone loading systems and the development of related platforms, such as the Fuya UAV automatic airport intelligent battery replacement, charging, and recovery storage drone designed by Shanghai Weiman Industrial Product Design Co., Ltd. Figure 3 As shown; the UAV carrier platform designed by Zhejiang Hede Technology Co., Ltd. (such as Figure 4 ), each component of the shelter has a sequence of work, including small drone doors, small drone heavy-duty slide rail trays, large drone lifting platforms, roof sliding cover systems, control integration parts, etc.; the vertical take-off and landing fixed-wing drone automatic airport system developed and designed by Xi'an Lingkong Electronic Technology Co., Ltd. (such as Figure 5 ), an automated and intelligent drone inspection and supervision system. It can be used as a mobile vehicle chassis as a carrier to provide an installation and application platform for the vertical take-off and landing fixed-wing honeycomb box. The drone can operate flexibly with the vehicle-mounted platform, automatically take off and land, automatically charge, operate autonomously, and automatically recover throughout the process, which can meet the needs of long-distance, mobile, and high-intensity continuous operations.

[0010] In addition, in terms of patents, the patents for drone hangars are mainly aimed at single drones, vehicle-mounted, and various hangar design structures. For example, Zhang Zhi et al. proposed a mobile cabin for a vehicle-mounted drone hangar, which solved the problems of fixed deployment limitations of drone automatic airports and the lack of power supply and poor signal. However, this is only for single drones used on land, and the degree of automation is poor; Chen Fangping et al. proposed patents for the scheduling method, device and service end of drones and drone hangars, which can effectively improve the degree of automation when scheduling drones and drone hangars, and provide a communication method between drones and drone hangars, receiving pending job requests to the drones. The target drone hangar corresponding to the human hangar identifier sends a takeoff preparation instruction to enable the target drone hangar to perform takeoff preparation actions, but it is mainly for a single drone and only gives a rough framework, without corresponding implementation for land or shipborne situations; Li Wei and others proposed a patent for a shipborne drone hangar, which is a hangar applied to the hull for carrying drones, which can improve the stability of the horizontal landing platform and prevent it from shaking with the hull, thereby reducing the difficulty of landing the drone and avoiding collisions of drones during landing. It is only for a single drone and is not for improving automation and better autonomous protection. Summary of the invention

[0011] The purpose of the present invention is to provide an intelligent decision support system for an autonomous support platform for unmanned aerial vehicles on unmanned boats. It can realize high-density loading and autonomous support for unmanned aerial vehicles of specific models and sizes in a space of a specific size in the unmanned boat, and complete the automatic and unmanned transfer and lifting of the unmanned aerial vehicle to the autonomous support platform after landing.

[0012] The intelligent decision support system for the autonomous support platform of unmanned aerial vehicles on unmanned boats includes the intelligent decision system of unmanned aerial vehicles, the intelligent decision control system of the autonomous support platform of unmanned aerial vehicles, the communication interface and the cloud platform; the intelligent decision system of unmanned aerial vehicles and the intelligent decision control system of the autonomous support platform of unmanned aerial vehicles are connected to the cloud platform through the communication interface to realize data intercommunication;

[0013] The cloud platform includes a cloud database and a cloud server, and the cloud server includes a task planning server, a control command server, a neural network algorithm server and an autonomous support platform operation algorithm server; the task planning server is used to generate task requirements; the neural network algorithm server is used to select the most suitable drone for dispatch according to the task requirements; the autonomous support platform operation algorithm server is used to select the optimal outbound takeoff path planning and movement steps according to the information of the drone to be dispatched; the control command server is used to generate an execution command set according to the outbound takeoff path planning and movement steps of the drone to be dispatched, and transmit it to the drone to be dispatched for execution.

[0014] Furthermore, the UAV intelligent decision-making system includes a UAV real-time control system and a UAV information management system; the UAV autonomous support platform intelligent decision-making control system includes a hangar lifting platform module, a hangar UAV transfer device, a hangar scheduling and operation module, a hangar intelligent support platform and apron expansion module;

[0015] The UAV real-time control system is used to control the modules of the intelligent decision-making control system of the UAV autonomous support platform, including the lifting control of the hangar, the expansion control of the apron, and the transfer control of the UAV in the hangar; the UAV information management system is used for the UAV to receive and execute relevant execution commands, including the relevant task scheduling of the UAV, and on the other hand, to obtain relevant information of the UAV through various sensors, including the fuel and power of the UAV, the number of the UAV, and the equipment of the load, and transmit and display it in real time on the user interaction interface;

[0016] The hangar lifting platform module is used to achieve the attitude adjustment, precise fixation and lifting and transportation of the UAV; the hangar UAV transfer device is used to provide the fixation, transfer and handling of the UAV; the hangar scheduling and operation module receives the control command signal from the cloud platform through the communication interface, and controls the hangar lifting platform module and the hangar UAV transfer device to work together according to the command to complete the lifting and transfer functions, thereby realizing the take-off and landing of the UAV; the hangar intelligent support platform is used to provide the UAV power-on, circulation during high-density storage, and pre-preparation functions for go-around; the apron extension device is used to provide a larger support space for the UAV when the UAV lands in the marine environment.

[0017] Furthermore, the communication interface includes a satellite communication module, a 5G communication module, an optical fiber communication module and a Wifi communication module;

[0018] Both the drone and the autonomous support platform are equipped with satellite communication modules and 5G communication modules. When the autonomous support platform and the drone are close to the land or within the coverage area, the existing 5G mobile communication network is used to achieve high-speed data transmission with the cloud platform. When there is no signal on the ocean, the satellite communication module transmits the data to the satellite, and then transmits it to the cloud platform through the ground station for communication interaction.

[0019] When the UAV is within the communication range of the UAV autonomous support platform, data communication is carried out through the satellite communication module; when the UAV autonomously lands and docks in the UAV autonomous support platform on the unmanned boat, data is transmitted through the optical fiber communication module to provide extremely high speed and security;

[0020] Managers use registered account identity information to interact with the cloud platform through the Internet on PC web pages or mobile apps to monitor and manage real-time data and status.

[0021] Furthermore, the specific method by which the neural network algorithm server selects the most suitable drone for dispatch according to the mission requirements is:

[0022] (1) Collect mission data, drone data, and environment data to construct a training set. Each set of training data in the training set includes an input vector X and a score y.

[0023] X=[T,D,E]

[0024] Where T = [T t ,T p ,T lt ,T lg ,T d ], T t Indicates the task type code, T p Indicates the task priority, T lt With T lg Indicates the latitude and longitude coordinates of the mission location, T d represents the task time window interval; D = {D e ,D p ,D s ,D lt ,D lg ,D b ], D e Indicates the battery life, D p Indicates load capacity, D s Indicates speed, D lt With D lg Indicates the latitude and longitude coordinates of the current location, D b Indicates the current power percentage; E=[E e ,E w ,E p ,E t ,E o ], E e Indicates temperature, E w Indicates wind speed, E p Indicates precipitation, E t Indicates the terrain type, E o Indicates the distribution of obstacles;

[0025] (2) Use the training set to train the neural network model. The input of the neural network model is vector X, and the output is

[0026] Initially, 3 hidden layers are selected, and the optimal number of layers is determined through subsequent experiments. The hidden layer consists of multiple neurons, each of which performs linear transformation and nonlinear activation processing on the input data; the corresponding initial number of neurons in each layer is 2dx, dx, In a layer-by-layer decreasing manner, dx is the input dimension, and the three-layer neural network model formula is expressed as:

[0027] Z1=σ1(W1X+b1)

[0028] Z2=σ1(W2Z1+b2)

[0029] Z3=σ1(W3Z2+b3)

[0030]

[0031] Among them, W1, W2, W3, W out is the weight matrix, b1, b2, b3, b out is the bias vector, Z1, Z2, and Z3 are the outputs of the corresponding hidden layers; σ1 and σ2 are activation functions;

[0032] σ1(x)=max(0,x)

[0033]

[0034] Use mean square error as the loss function to measure the difference between the model prediction value and the actual value:

[0035]

[0036] Where m is the number of samples; y i and Represents the true score of the i-th group of training samples and the score output by the neural network model;

[0037] Calculate the gradient of the loss function with respect to the weights and biases of each layer using the chain rule;

[0038]

[0039] in, Represents the loss function L relative to the j-th layer weight matrix W j The gradient of Z j is the activation output of the jth layer, Y is the desired output, is the transpose of the activation output of the j-1th layer;

[0040] Perform weight updates and use the gradient descent algorithm to update weights and biases to minimize the loss function:

[0041]

[0042] Among them, α is the learning rate;

[0043] After the training is completed, for a given task and various related environmental data identified, the data of each drone is automatically uploaded, the score of each drone is calculated, and the drone with the highest score is selected.

[0044] Furthermore, the method for the autonomous support platform operation algorithm server to select the optimal take-off path planning and movement steps according to the information of the drone to be dispatched is specifically as follows:

[0045] (1) Define the environment map information used for search. Since the drone transportation in the hangar only involves up, down, left, and right movements, the three-dimensional operation is simplified to a two-dimensional movement problem.

[0046] (2) Determine the starting point S and the end point G of the drone. When the drone takes off, the lifting platform of the hangar numbered 0 needs to be raised to form a space for transfer and movement. The position of the space is zeroPos;

[0047] (3) Create lists Openlist and Closelist and initialize them to empty. The Openlist is used to store nodes whose costs have been calculated and their cost values, and the Closelist stores the node with the minimum cost for each expansion and the parent node information of the node.

[0048] (4) Put the starting node S into the Openlist, set the cost function of the initial node S to g(S) = 0, set the parent node of S to itself, and calculate the heuristic function h(S) of S;

[0049] (5) Calculate the f(n) values ​​of the nodes in the four neighborhoods around the starting node S, and put the calculation results and the nodes into the Openlist;

[0050] f(n)=g(n)+h(n)

[0051] Where f(n) represents the actual cost of node n; g(n) represents the current cost of node n, that is, the cost required to travel from the starting point to node n; h(n) represents the estimated cost required to move from node n to the target node;

[0052] h(n)=abs(nx-goal.x)+abs(ny-goal.y)

[0053] Among them, nx represents the horizontal coordinate of the current node n; ny represents the vertical coordinate of the current node n; goal.x represents the horizontal coordinate of the target node; goal.x represents the vertical coordinate of the target node;

[0054] (6) Remove the starting node from the Openlist and add it to the Closelist;

[0055] (7) Select the hangar node n with the smallest f(n) value from Openlist, determine whether the position of the current hangar node n is consistent with the position of the target hangar node G to be reached by the starting point S, and whether the space position zeroPos is in the hangar numbered 0; if so, the path search is successful, the cyclic path planning ends, and the shortest movement path and the minimum number of moves are output; otherwise, execute steps (8) to (10);

[0056] (8) If hangar node n is not the same as target hangar node G, move hangar node n from the Openlist table to the Closelist table;

[0057] (9) Traverse the nodes m in the four neighborhoods around the hangar node n, that is, the child nodes of node n, determine the movement amount in each direction, and whether it is a legal movement. If it is a legal movement, perform the movement operation, obtain the new state and make the following judgments:

[0058] 1) Determine whether m is already in the Closelist. If so, skip it directly;

[0059] 2) Determine whether m is already in the Openlist. If the child node m is already in the list Openlist, recalculate f(m) and compare whether the current estimated minimum cost of the node is the same as the previous estimated value. If the current cost of node m is smaller, change the cost of node m in the Openlist to a smaller new cost, and the parent node of node m becomes n at the same time;

[0060] 3) If m is not in either the Openlist or the Closelist, put node m into the Openlist, confirm its estimated cost f(m) and set its parent node to n;

[0061] (10) Return to step (7) and continue calculation.

[0062] The beneficial effects of the present invention are:

[0063] The present invention is aimed at drones of specific models and sizes, and realizes high-density loading and autonomous support in a space of a specific size of an unmanned boat, and completes the automatic and unmanned transfer and lifting of the drone to the autonomous support platform after landing. The present invention provides a heuristic search and reinforcement learning fusion algorithm for autonomous support mission planning of drones, realizes the optimal path planning for takeoff and landing, and combines neural networks with the requirements of the mission to design an aircraft that can automatically select the most suitable and highest-scoring aircraft for dispatch through relevant influencing factors and their changeable weight ratios, and adjusts and improves the neural network system in combination with specific actual flight data in the later stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A schematic diagram of the SkyOne drone library.

[0065] Figure 2 A diagram of the Parcelcopter Skyport drone hangar.

[0066] Figure 3 This is a schematic diagram of Fuya UAV automatic airport.

[0067] Figure 4 Schematic diagram of the drone carrier platform.

[0068] Figure 5 This is a schematic diagram of the Xi'an Lingkong UAV automatic airport system.

[0069] Figure 6 It is the overall system architecture diagram of the present invention.

[0070] Figure 7 It is a system composition structure diagram of the present invention.

[0071] Figure 8 This is a system deployment architecture design diagram of the present invention.

[0072] Fig. 9 It is a communication flow chart of the whole system of the present invention.

[0073] Fig.10 It is a control flow chart of the present invention.

[0074] Fig.11 This is a flowchart of the neural network in the present invention.

[0075] Fig.12 The present invention is a flowchart of the heuristic search algorithm for selecting the optimal take-off path planning and movement steps. DETAILED DESCRIPTION

[0076] The present invention is further described below in conjunction with the accompanying drawings.

[0077] The present invention designs an autonomous support system for high-density loading of unmanned aerial vehicles. For unmanned aerial vehicles of specific models and sizes, high-density loading and autonomous support are achieved in a space of a specific size of an unmanned boat, and the unmanned aerial vehicle is automatically and unmannedly transferred and lifted to the autonomous support platform after landing. In order to solve the problem of how to operate and schedule the autonomous support platform during the take-off and landing of the unmanned aerial vehicle in the autonomous support platform, the present invention proposes a heuristic search and reinforcement learning fusion algorithm for autonomous support task planning of unmanned aerial vehicles to achieve optimal path planning for take-off and landing, and combines a neural network with relevant influencing factors and their changeable weight ratios according to the requirements of the dispatching task, and designs an aircraft that can automatically select the most suitable and highest-scoring dispatching aircraft through the relevant influencing factors and their changeable weight ratios, and adjusts and improves the neural network system in combination with specific actual flight data in the later stage; at the same time, the system architecture and task planning of the entire autonomous support system are carried out, and the entire control display interface is designed by separating the front end and the back end; finally, physical experiments are carried out for verification.

[0078] like Figure 6 and Figure 7 As shown in the figure, the whole system architecture is mainly composed of the UAV intelligent decision-making system, the UAV autonomous support platform intelligent decision-making control system, the communication interface and the cloud platform. The UAV intelligent decision-making system and the UAV autonomous support platform intelligent decision-making control system are connected to the cloud platform through the communication interface to achieve data intercommunication. The cloud platform uses the algorithm server to obtain the optimal solution and planning steps for the processed data, and at the same time passes this result to the control command server to generate control commands for each module and pass it to each actuator, and each actuator executes the relevant commands.

[0079] like Figure 8 As shown in the figure, the whole system function is divided into three parts: data collection and transmission, cloud server and Web interactive interface: the data collection and transmission part mainly includes drones and drone autonomous support platform. First, it will obtain data from the relevant sensors of drones and drone autonomous support platform, and transmit the data to the cloud platform through the communication interface module. At the same time, the cloud platform will also transmit back relevant data and control commands, such as how the modules of the autonomous support platform should coordinate and operate to make the aircraft stop at the designated location of the autonomous support platform. The cloud platform is mainly composed of cloud database, task planning server, neural network algorithm server, control command server, and autonomous support platform operation algorithm server, which processes, analyzes and stores data, and transmits the processed data to the corresponding server for processing, and transmits the obtained data results to the control command server to generate the relevant execution command set, and then transmits it to the execution agency for execution. The interactive interface is based on B / S architecture, allowing users to operate on the Web interface to control the background logic, data transmission and storage.

[0080] like Fig. 9As shown in the figure, the UAV and the UAV autonomous support platform are equipped with satellite communication modules and 5G communication modules. When the ship and the UAV are close to the land or within the coverage area, the existing 5G mobile communication network can be used to achieve high-speed data transmission with the cloud platform; when there is no signal on the ocean, the data can be transmitted to the satellite through the satellite communication module, and then transmitted to the cloud platform for communication interaction through the ground station. When the UAV communicates data through the satellite communication module within the communication range of the ship, and the UAV autonomously lands and docks in the autonomous support platform, the data can be transmitted through optical fiber or wired connection to provide extremely high speed and security. Managers can log in to the user interaction interface on the PC web page or mobile APP through the registered account identity information to monitor and manage real-time data and status.

[0081] like Fig.10 As shown, a neural network algorithm is trained in the cloud platform neural network algorithm server to select the most suitable UAV according to the relevant task requirements, and then upload the information such as the number of the selected UAV to the autonomous support platform operation algorithm server, and then obtain the specific control instructions of the autonomous support platform after processing. The autonomous support platform operation algorithm server selects the optimal outbound takeoff path planning and movement steps through the "heuristic search and reinforcement learning fusion algorithm for UAV autonomous support task planning", and then the autonomous support platform executes the relevant commands and displays the execution results on the page to feedback the success / failure of the execution. If successful, it enters the standby state.

[0082] Embodiment 1:

[0083] The intelligent decision support system for the UAV autonomous support platform includes the UAV intelligent decision system, the UAV autonomous support platform intelligent decision control system, the communication interface and the cloud platform; the UAV intelligent decision system and the UAV autonomous support platform intelligent decision control system are connected to the cloud platform through the communication interface to achieve data intercommunication;

[0084] The cloud platform includes a cloud database and a cloud server. The cloud server includes a task planning server, a control command server, a neural network algorithm server and an autonomous support platform operation algorithm server. The task planning server is used to generate task requirements. The neural network algorithm server is used to select the most suitable UAV for dispatch according to the task requirements. The autonomous support platform operation algorithm server is used to select the optimal take-off path planning and movement steps based on the information of the UAV to be dispatched. The control command server is used to generate an execution command set based on the take-off path planning and movement steps of the UAV to be dispatched, and transmit it to the UAV to be dispatched for execution.

[0085] The UAV intelligent decision-making system includes the UAV real-time control system and the UAV information management system; the UAV autonomous support platform intelligent decision-making control system includes the hangar lifting platform module, the hangar UAV transfer device, the hangar scheduling and operation module, the hangar intelligent support platform and the apron expansion module;

[0086] UAV real-time control system: Users can switch the control system to manual mode through the human-machine exchange interface, so as to control the movement of various modules and devices of the intelligent decision-making control system of the UAV autonomous support platform in real time, such as the lifting control of the hangar, the expansion control of the apron, and the transfer control of the UAV in the hangar.

[0087] UAV information management system: It mainly consists of two aspects. On the one hand, the UAV receives and executes relevant execution commands, such as the relevant task scheduling of the UAV; on the other hand, it obtains relevant information of the UAV through various sensors, such as the fuel level of the UAV, the number of the UAV and the equipment loaded, and transmits and displays it in real time on the user interaction interface.

[0088] The hangar lifting platform device is used to achieve the attitude adjustment, precise fixation and lifting and transportation of the UAV; the hangar UAV transfer device is composed of a multi-link mechanism. When the pre-fixation function tethers the UAV to the take-off and landing plane of the cabin, it provides the fixation, transfer, and transportation of the UAV; the hangar scheduling and operation module receives the control command signal from the cloud platform through the communication interface, and controls the hangar lifting platform module and the hangar UAV transfer device to work together according to the command to complete the lifting and transfer functions, thereby realizing the take-off and landing of the UAV; the hangar intelligent support platform provides the UAV power-on, circulation during high-density storage, and pre-preparation for go-around functions; the apron extension device is used to provide a larger support space for the UAV when the UAV lands in the marine environment;

[0089] The communication interface includes satellite communication module, 5G communication module, optical fiber communication module and Wifi communication module;

[0090] Both the drone and the autonomous support platform are equipped with satellite communication modules and 5G communication modules. When the autonomous support platform and the drone are close to the land or within the coverage area, the existing 5G mobile communication network is used to achieve high-speed data transmission with the cloud platform. When there is no signal on the ocean, the satellite communication module transmits the data to the satellite, and then transmits it to the cloud platform through the ground station for communication interaction.

[0091] When the drone is within the communication range of the drone autonomous support platform, data communication is carried out through the satellite communication module; when the drone autonomously lands and docks in the drone autonomous support platform, data is transmitted through the optical fiber communication module to provide extremely high speed and security;

[0092] Managers use registered account identity information to interact with the cloud platform through the Internet on PC web pages or mobile apps to monitor and manage real-time data and status.

[0093] The present invention designs a heuristic search and reinforcement learning fusion algorithm for autonomous support mission planning of unmanned aerial vehicles to achieve optimal path planning for take-off and landing, and combines neural networks with the requirements of the mission to design an automatic selection of the most suitable and highest-scoring aircraft for dispatch through relevant influencing factors and their modifiable weight ratios, and adjusts and improves the neural network system in combination with specific actual flight data in the later stage.

[0094] The UAV task scheduling problem for the UAV autonomous support system can be regarded as a complex optimization problem. When the UAV is located on the autonomous support platform, the most suitable UAV is selected to take off and perform a specific task based on factors such as mission requirements, UAV performance and environmental conditions through a neural network model.

[0095] For data collection and preprocessing, there are mainly the following types of data:

[0096] ① Mission data: including mission type, such as reconnaissance, transportation, patrol, etc., and digitally coding the mission type, with reconnaissance mission coded as 1, transportation mission coded as 2, and patrol mission coded as 3; mission priority: the urgency or importance of the mission; mission location: the geographical location of the mission; mission time window: the start and end time of the mission;

[0097] ② Drone data: mainly obtain the performance and current status of each drone. Mainly includes endurance time: the time the drone can continue to fly with the current amount of fuel; load capacity: the maximum load that the drone can carry; speed: the flight speed of the drone; current position: the position of the drone in the hangar; current power: the battery power percentage of the drone;

[0098] ③ Environmental data: mainly obtains relevant information about the environment at the time of task execution. This includes weather conditions: including temperature, wind speed, precipitation, etc.; geographic information: including terrain type, obstacle distribution, etc.;

[0099] like Fig.11 As shown in the figure, the neural network selects the most suitable drone through the following analysis data process:

[0100] (1) The above-mentioned mission data, UAV data and environmental data are cleaned, standardized and normalized, and the data is divided into a training set and a test set. Then, feature extraction is performed on the training set and provided as input data to the neural network.

[0101] (2) Complete the design of the neural network model, use the neural network to train the samples, complete the simulation of the neural network, and backtrack;

[0102] (3) Check whether the neural network has completed training. If so, save the trained network structure. Otherwise, adjust the initialization parameters.

[0103] (4) Use the test set to evaluate the final performance of the model and optimize the model parameters;

[0104] (5) Extract relevant data features based on real-time mission requirements and use neural networks to select the most suitable UAV for takeoff and execution;

[0105] The neural network model is mainly composed of three layers. The first layer is the input layer, the second layer is the hidden layer, and the third layer is the output layer. Each layer is composed of a certain number of neurons, and the layers are connected to each other through neurons. For the input layer, the input data is first processed. According to the data types given above, three vector representations can be obtained, namely, the mission data combination vector T, the drone data combination vector D, and the environment data combination vector E. The relevant vectors are defined as follows: T = [T t ,T p ,T lt ,T lg ,T d ], where T t T is the mission type code, with values ​​of 1 (reconnaissance), 2 (transportation), and 3 (patrol); p is the task priority; T lt With T lg It represents the latitude and longitude coordinates of the mission location; the mission time window time interval is T d . D=[D e ,D p ,D s ,D lt ,D lg ,D b ], where D e D is endurance, p is the load capacity (PayloadCapacity), D s Speed ​​(Speed) Current Location (Current Location) is the longitude and latitude coordinates (D lt , D lg ), the current battery (CurrentBattery) is D b (percentage). E=[E e ,E w ,E p ,E t ,Eo ], where E e is temperature, E w is the wind speed, E p is the precipitation, E t Represents the terrain type, E o Indicates the obstacle distribution. The input vector is: X = [T, D, E]. For the output, there is only one final comprehensive score of the drone.

[0106] Initially, 3 hidden layers are selected, and the optimal number of layers is determined through subsequent experiments. The hidden layer consists of multiple neurons, and each neuron performs linear transformation and nonlinear activation processing on the input data. The corresponding initial selection of the number of neurons in each layer is 2dx, dx, The method of decreasing layer by layer, where dx is the dimension of the input. The formula of the neural network model with a three-layer structure is expressed as:

[0107] Z1=σ1(W1X+b1) (4-1)

[0108] Z2=σ1(W2Z1+b2) (4-2)

[0109] Z3=σ1(W3Z2+b3) (4-3)

[0110]

[0111] Among them, W1, W2, W3, W out is the weight matrix, b1, b2, b3, b out is the bias vector, Z1, Z2, and Z3 are the outputs of the corresponding hidden layers, σ1 and σ2 are activation functions, and the activation function establishes a functional relationship between the data at both ends of the neuron. The choice of the activation function has a significant impact on the effect of the neural network. The activation functions σ1 and σ2 used in the present invention are the ReLU function and the sigmoid function formulas are as follows:

[0112] σ1(x)=max(0,x) (4-5)

[0113]

[0114] Finally, the final comprehensive score of the predicted drone is obtained through formula (4-4)

[0115] Then use the mean square error (MSE) as the loss function to measure the difference between the model prediction value and the actual value:

[0116]

[0117] Where m is the number of samples, yi It is a real rating. is the predicted score. Then the neural network model is trained and the gradient of the loss function relative to the weights and biases of each layer is calculated by the chain rule:

[0118]

[0119] in, Represents the loss function L relative to the j-th layer weight matrix W j The gradient of Z, m is the number of training samples, which is used to calculate the average gradient. j is the activation output of the jth layer, Y is the desired output, is the transpose of the activation output of the j-1th layer. Then the weights are updated, and the gradient descent algorithm is used to update the weights and biases to minimize the loss function:

[0120]

[0121]

[0122] Here, α is the learning rate.

[0123] After the training is completed, for a given task and various related environmental data identified, the data of each drone is automatically uploaded, the score of each drone is calculated, and the drone with the highest score is selected. The drone number is passed as a parameter to the following A* algorithm to realize the task scheduling of the drone, select the most suitable and highest-scoring drone, select the optimal out-of-stock movement steps, and perform the specified task.

[0124] The heuristic search and reinforcement learning fusion algorithm for autonomous support mission planning of UAVs is as follows:

[0125] The A* algorithm is a heuristic search algorithm used to find the shortest path in a graph. The cost of each node in the A* algorithm is calculated by the corresponding function, so the determination of the estimated cost function is also crucial, which will have an important impact on the entire search process of the A* algorithm. The current estimated cost function of the A* algorithm is shown in Formula 5-1:

[0126] f(n)=g(n)+h(n) (5-1)

[0127] Where n represents the node that needs to be expanded and the cost calculated during the execution of the A* algorithm. f(n) represents the actual cost of node n, and g(n) represents the current cost of node n, that is, the cost required for the robot to travel from the starting point to node n. h(n) represents the estimated value of the cost required to move from node n to the target node, which is not the actual value and needs to be calculated through the corresponding heuristic function.

[0128] The heuristic function type used by the A* algorithm in the present invention is the Manhattan distance, which represents the absolute value of the difference between two nodes in the horizontal direction plus the absolute value of the difference in the vertical direction. Its expression is shown in Formula 5-2:

[0129] h(n)=abs(nx-goal.x)+abs(ny-goal.y) (5-2)

[0130] In the formula, nx represents the horizontal coordinate of the current node n, ny represents the vertical coordinate of the current node n, goal.x represents the horizontal coordinate of the target node, and goal.x represents the vertical coordinate of the target node. It is suitable for searching in only four directions, up, down, left, and right. Since the drone transportation in the hangar only involves the movement of up, down, left, and right, the Manhattan distance is very suitable as the heuristic function of the A* algorithm in the present invention.

[0131] like Fig.12 As shown, the main process of the A* algorithm in the present invention when finding the shortest moving path and the minimum number of moves is as follows:

[0132] (1) Define the environment map information used for search. Since the drone transportation in the hangar only involves up, down, left, and right movements, the three-dimensional operation is simplified to a two-dimensional movement problem.

[0133] (2) Determine the starting point S and the end point G. When the UAV takes off, the lifting platform of the hangar numbered 0 needs to be raised to form a space for transfer and movement. The position of the space is zeroPos;

[0134] (3) Create the lists Openlist and Closelist and initialize them to empty. The Openlist table is mainly used to store the nodes whose costs have been calculated and the cost values ​​of the nodes. The Closelist table is used to store the node with the minimum cost for each expansion and the parent node information of the node.

[0135] (4) Put the starting node S into the Openlist, set the cost function of the initial node S to g(S) = 0, set the parent node of S to itself, and calculate the heuristic function h(S) of S;

[0136] (5) Calculate the f(n) values ​​of the nodes in the four neighborhoods around the starting node S, and put the calculation results and the nodes into the Openlist;

[0137] (6) Remove the starting node from the Openlist and add it to the Closelist;

[0138] (7) Select the hangar node n with the smallest f(n) value from Openlist, determine whether the position of the current hangar node n is consistent with the position of the target hangar node G to be reached by the starting point S of the hangar starting target drone, and whether the space position zeroPos is consistent with the initial state. In the hangar numbered 0, if so, the path search is successful, the cyclic path planning is ended, and the shortest movement path and the minimum number of moves are output.

[0139] (8) If the hangar number node n is not the same as the target hangar number node G, move the hangar number node n from the Openlist table to the Closelist table;

[0140] (9) Traverse the nodes m in the four neighborhoods around the hangar number node n, that is, the child nodes of node n, determine the movement amount in each direction, and whether it is a legal movement. If it is a legal movement, perform the move operation, obtain the new state and make the following judgments:

[0141] 1) Determine whether m is already in the Closelist. If so, skip it directly;

[0142] 2) Determine whether m is already in the Openlist. If the child node m is already in the list Openlist, recalculate f(m) and compare whether the current estimated minimum cost of the node is the same as the previous estimated value. If the current cost of node m is smaller, change the cost of node m in the Openlist to a smaller new cost, and the parent node of node m becomes n at the same time;

[0143] 3) If m is not in either the Openlist or the Closelist, put node m into the Openlist, confirm its estimated cost f(m) and set its parent node to n;

[0144] (10) Go to (7) and continue execution.

[0145] The neural network mentioned above is combined with the A* algorithm. The neural network algorithm server predicts the most suitable drone for deployment in real time according to different mission requirements. The number of this drone is passed into the A* algorithm as a parameter. The A* algorithm calculates the optimal movement path and number of movements based on the positional relationship between the target and the destination, and issues relevant control instructions to realize the take-off and landing of the drone.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Intelligent decision support system for autonomous support platform of unmanned aerial vehicle on unmanned boat, Features: It includes the UAV intelligent decision-making system, the UAV autonomous support platform intelligent decision-making control system, the communication interface and the cloud platform; the UAV intelligent decision-making system and the UAV autonomous support platform intelligent decision-making control system are connected to the cloud platform through the communication interface to achieve data intercommunication; The cloud platform includes a cloud database and a cloud server, and the cloud server includes a task planning server, a control command server, a neural network algorithm server and an autonomous guarantee platform operation algorithm server; The task planning server is used to generate task requirements; The neural network algorithm server is used to select the most suitable drone for dispatch according to the mission requirements; the autonomous support platform operation algorithm server is used to select the optimal outbound takeoff path planning and movement steps based on the information of the drone to be dispatched; the control command server is used to generate an execution command set based on the outbound takeoff path planning and movement steps of the drone to be dispatched, and transmit it to the drone to be dispatched for execution.

2. According to claim 1, the intelligent decision support system for the autonomous support platform of unmanned aerial vehicles on unmanned boats, Features: The UAV intelligent decision-making system includes a UAV real-time control system and a UAV information management system; the UAV autonomous support platform intelligent decision-making control system includes a hangar lifting platform module, a hangar UAV transfer device, a hangar scheduling and operation module, a hangar intelligent support platform and apron expansion module; The UAV real-time control system is used to control the modules of the intelligent decision-making control system of the UAV autonomous support platform, including the lifting control of the hangar, the expansion control of the apron, and the transfer control of the UAV in the hangar; the UAV information management system is used for the UAV to receive and execute relevant execution commands, including the relevant task scheduling of the UAV, and on the other hand, to obtain relevant information of the UAV through various sensors, including the fuel and power of the UAV, the number of the UAV, and the equipment of the load, and transmit and display it in real time on the user interaction interface; The hangar lifting platform module is used to achieve attitude adjustment, precise fixation and lifting and transportation of the UAV; The hangar drone transfer device is used to provide drone fixation, transfer and transportation services; the hangar scheduling and operation module receives control command signals from the cloud platform through the communication interface, and controls the hangar lifting platform module and the hangar drone transfer device to work together according to the commands to complete the lifting and transfer functions, thereby realizing the take-off and landing of the drone; the hangar intelligent support platform is used to provide drone power-on, circulation during high-density storage, and pre-preparation functions for go-around; the apron extension device is used to provide a larger support space for the drone when the drone lands in a marine environment.

3. According to claim 1, the intelligent decision support system for the autonomous support platform of unmanned aerial vehicles on unmanned boats, Features: The communication interface includes a satellite communication module, a 5G communication module, an optical fiber communication module and a Wifi communication module; Both the drone and the autonomous support platform are equipped with satellite communication modules and 5G communication modules. When the autonomous support platform and the drone are close to the land or within the coverage area, the existing 5G mobile communication network is used to achieve high-speed data transmission with the cloud platform. When there is no signal on the ocean, the satellite communication module transmits the data to the satellite, and then transmits it to the cloud platform through the ground station for communication interaction. When the UAV is within the communication range of the UAV autonomous support platform, data communication is carried out through the satellite communication module; when the UAV autonomously lands and docks in the UAV autonomous support platform on the unmanned boat, data is transmitted through the optical fiber communication module to provide extremely high speed and security; Managers use registered account identity information to interact with the cloud platform through the Internet on PC web pages or mobile apps to monitor and manage real-time data and status.

4. According to claim 1, the intelligent decision support system for the autonomous support platform of unmanned aerial vehicles on unmanned boats, Features: The specific method by which the neural network algorithm server selects the most suitable UAV for dispatch according to the mission requirements is: (1) Collect mission data, drone data, and environment data to construct a training set. Each set of training data in the training set includes an input vector X and a score y. X = [T, D, E] Where T = [T t , T p , T lt , T lg , T d ], T t Indicates the task type code, T p Indicates the task priority, T lt With T lg Indicates the latitude and longitude coordinates of the mission location, T d Indicates the task time window interval; D = [D e , D p , D s , D lt , D lg , D b ], D e Indicates the battery life, D p Indicates load capacity, D s Indicates speed, D lt With D lg Indicates the latitude and longitude coordinates of the current location, D b Indicates the current power percentage; E=[E e , E w , E p , E t , E o ], E e Indicates temperature, E w Indicates wind speed, E p Indicates precipitation, E t Indicates the terrain type, E o Indicates the distribution of obstacles; (2) Use the training set to train the neural network model. The input of the neural network model is vector X, and the output is Initially, 3 hidden layers are selected, and the optimal number of layers is determined through subsequent experiments. The hidden layer consists of multiple neurons, each of which performs linear transformation and nonlinear activation processing on the input data; the corresponding initial number of neurons in each layer is 2dx, dx, In a layer-by-layer decreasing manner, dx is the input dimension, and the three-layer neural network model formula is expressed as: Z 1 =σ 1 (W 1 X+b 1 ) WITH 2 =σ 1 (IN 2 WITH 1 +b 2 ) WITH 3 =σ 1 (IN 3 WITH 2 +b 3 ) Among them, W 1 , W 2 , W 3 , W out is the weight matrix, b 1 、b 2 、b 3 、b out is the bias vector, Z 1 , Z 2 , Z 3 are the outputs of the corresponding hidden layers; σ 1 , σ 2 is the activation function; σ 1 (x)=max(0,x) Use mean square error as the loss function to measure the difference between the model prediction value and the actual value: Where m is the number of samples; yi and Represents the true score of the i-th group of training samples and the score output by the neural network model; Calculate the gradient of the loss function with respect to the weights and biases of each layer using the chain rule; in, Represents the loss function L relative to the j-th layer weight matrix W j The gradient of Z j is the activation output of the jth layer, Y is the desired output, is the transpose of the activation output of the j-1th layer; Perform weight updates and use the gradient descent algorithm to update weights and biases to minimize the loss function: Among them, α is the learning rate; After the training is completed, for a given task and various related environmental data identified, the data of each drone is automatically uploaded, the score of each drone is calculated, and the drone with the highest score is selected.

5. The intelligent decision support system for the autonomous support platform of unmanned aerial vehicles on unmanned boats according to claim 1, Features: The method in which the autonomous support platform operation algorithm server selects the optimal take-off path planning and movement steps according to the information of the unmanned aerial vehicle to be dispatched is as follows: (1) Define the environment map information used for search. Since the drone transportation in the hangar only involves up, down, left, and right movements, the three-dimensional operation is simplified to a two-dimensional movement problem. (2) Determine the starting point S and the end point G of the drone. When the drone takes off, the lifting platform of the hangar numbered 0 needs to be raised to form a space for transfer and movement. The position of the space is zeroPos; (3) Create lists Openlist and Closelist and initialize them to empty. The Openlist is used to store nodes whose costs have been calculated and their cost values, and the Closelist stores the node with the minimum cost for each expansion and the parent node information of the node. (4) Put the starting node S into the Openlist, set the cost function of the initial node S to g(S) = 0, set the parent node of S to itself, and calculate the heuristic function h(S) of S; (5) Calculate the f(n) values ​​of the nodes in the four neighborhoods around the starting node S, and put the calculation results and the nodes into the Openlist; f(n)=g(n)+h(n) Where f(n) represents the actual cost of node n; g(n) represents the current cost of node n, that is, the cost required to travel from the starting point to node n; h(n) represents the estimated cost required to move from node n to the target node; h(n)=abs(nx-goal.x)+abs(ny-goal.y) Among them, nx represents the horizontal coordinate of the current node n; ny represents the vertical coordinate of the current node n; goal.x represents the horizontal coordinate of the target node; goal.x represents the vertical coordinate of the target node; (6) Remove the starting node from the Openlist and add it to the Closelist; (7) Select the hangar node n with the smallest f(n) value from Openlist, determine whether the position of the current hangar node n is consistent with the position of the target hangar node G to be reached by the starting point S, and whether the space position zeroPos is in the hangar numbered 0; if so, the path search is successful, the cyclic path planning ends, and the shortest movement path and the minimum number of moves are output; otherwise, execute steps (8) to (10); (8) If hangar node n is not the same as target hangar node G, move hangar node n from the Openlist table to the Closelist table; (9) Traverse the nodes m in the four neighborhoods around the hangar node n, that is, the child nodes of node n, determine the movement amount in each direction, and whether it is a legal movement. If it is a legal movement, perform the movement operation, obtain the new state and make the following judgments: 1) Determine whether m is already in the Closelist. If so, skip it directly; 2) Determine whether m is already in the Openlist. If the child node m is already in the list Openlist, recalculate f(m) and compare whether the current estimated minimum cost of the node is the same as the previous estimated value. If the current cost of node m is smaller, change the cost of node m in the Openlist to a smaller new cost, and the parent node of node m becomes n at the same time; 3) If m is not in either the Openlist or the Closelist, put node m into the Openlist, confirm its estimated cost f(m) and set its parent node to n; (10) Return to step (7) and continue calculation.