Heterogeneous interconnected multi-core distributed simulation experiment system and collaborative control method

Abstract

The application relates to the field of simulation technology, and particularly discloses a heterogeneous interconnection multi-core distributed simulation experiment system and a cooperative control method. The system comprises a semi-physical simulation system and a virtual system. The semi-physical simulation system comprises a plurality of semi-physical simulation nodes, each of which can simulate a semi-physical simulation entity; a plurality of multi-core parallel simulation nodes, each of which can simulate a plurality of parallel semi-physical simulation entities. The virtual system comprises a digital simulation system, which can simulate a plurality of parallel digital simulation entities; an experiment environment model, which is used for generating simulation experiment environment modeling data; a virtual-real combined simulation experiment platform, which is used for deducing and rendering a working scene; and a dynamic deducing visual simulation engine, which is used for rendering environment information by using the simulation experiment environment modeling data. The system has the capability of multi-core simulation, can further improve the upper limit of semi-physical simulation model access to a semi-physical system, and enables the semi-physical system to perform cluster simulation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of simulation technology, in particular to a heterogeneous interconnected multi-core distributed simulation experiment system and a collaborative control method. BACKGROUND

[0002] The idea of cluster adaptive communication originated in the late 1990s. The U.S. Air Force drew inspiration from simulated biological clusters and proposed the theory of "swarm warfare". In 2016, the U.S. Air Force demonstrated the technical framework of "swarm warfare" and strengthened the interconnection and intercommunication technology of clusters. In the early research, the communication networking was mainly focused on fixed topological forms, such as star topology, tree topology, mesh topology, and forest topology. It is difficult to objectively evaluate the rationality of simulation task decomposition when conducting distributed simulation. Jane Bachman proposed an intelligent grouping task decomposition method. Subsequently, a task decomposition strategy for load balancing of computing resources was proposed, including equivalent load balancing strategy and non-equivalent load balancing strategy. Traditional gateways achieve "one-to-one" parameter conversion through hard coding, and the processing logic is simple and the conversion relationship is clear, but lack of flexibility and scalability. Domestic scholar Gao Yuan proposed a "message-to-message" conversion method based on configuration files to achieve scalable adaptive message configuration, but the processing capacity is limited. To solve the problem of low processing capacity, a distributed information interaction mechanism based on shadow model is proposed. The shadow model realizes the synchronization of the behavior and state of the entity model, but does not focus on the processing logic of the model. Therefore, the shadow model adopts the "facade" mode for the internal simulation node, and follows the modeling of the simulation system to realize the model framework and interface. The shadow model adopts the "proxy" mode for the internal simulation node, and subscribes to messages through the distributed communication message middleware. Currently, there is still little research on how to achieve rapid and effective communication topology structure adaptation in the high dynamic, multi-task and complex environment of clusters, while reducing network traffic and improving the robustness of clusters. And the understanding of the basic principles of evolutionary communication self-organization of swarm intelligence is still insufficient, and there is still a big gap between the multi-node distributed simulation process and the actual situation. How to accurately reason about the situation of cluster warfare and make effective and reliable communication decisions under actual constraints still faces severe challenges, and new types of swarm intelligence communication organization architecture and methods need to be explored.

[0003] The prior art Chinese patent application CN202210718734.8 discloses a simulation test platform and method for cooperative confrontation of unmanned vehicle formation, but it only has the ability of single-core semi-physical simulation, and does not have the ability of multi-core parallel simulation and digital simulation multi-core parallel simulation calculation. SUMMARY

[0004] The application aims to provide a heterogeneous interconnected multi-core distributed simulation experiment system and a collaborative control method, partially solve or alleviate the above-mentioned deficiencies in the prior art, have the ability of multi-core simulation, and further improve the upper limit of the semi-physical simulation model accessing the semi-physical system, so that the semi-physical system can perform cluster simulation.

[0005] In order to solve the above-mentioned technical problems, the application specifically adopts the following technical solutions: the first aspect of the application is to provide a heterogeneous interconnected multi-core distributed simulation experiment system, which comprises a semi-physical simulation system and a virtual system; the semi-physical simulation system comprises: a plurality of semi-physical simulation nodes, each of which can simulate a semi-physical simulation entity; a plurality of parallel simulation nodes, each of which can simulate a plurality of parallel semi-physical simulation entities; the virtual system comprises: a digital simulation system, which can simulate a plurality of parallel digital simulation entities; an experiment environment model, which is used for generating simulation experiment environment modeling data; a virtual-real combined simulation experiment platform, which is used for deducing and rendering a working scene; and a dynamic deduction visual simulation engine, which is used for rendering environment information by using the simulation experiment environment modeling data. Further, the semi-physical simulation node comprises: a real-time simulation machine, which is used for constructing a semi-physical simulation entity model; a flight control computer, which is used for performing motion control on the semi-physical simulation entity model constructed by the real-time simulation machine; an external task computer, which comprises an online task planning module used for simulation entity cluster task planning and distribution to collaboratively control the cluster and an adaptive communication module used for realizing communication between simulation entities; and a self-organizing network data link, which is a physical communication link between semi-physical simulation entities. The multi-core parallel simulation node comprises: a real-time simulation machine, which is used for constructing a parallel semi-physical simulation entity model; a multi-core parallel simulation framework deployed in an RTX real-time system, which is used for parallel semi-physical simulation entity test state configuration, experiment process control, and storage and processing of test data; an online task planning simulation system, which is a digital simulation of the online task planning module, and is used for simulation entity cluster task planning and distribution to collaboratively control the cluster; and a self-organizing network data link simulation system, which is a digital simulation of the adaptive communication module and the self-organizing network data link, and is used for realizing communication between simulation entities. The digital simulation system comprises: a multi-core parallel simulation framework deployed in a Windows system and a digital simulation model based on the multi-core parallel simulation framework, which are used for parallel digital simulation entity test state configuration, experiment process control, and storage and processing of test data; an online task planning simulation system, which is a digital simulation of the online task planning module, and is used for simulation entity cluster task planning and distribution to collaboratively control the cluster; and a self-organizing network data link simulation system, which is a digital simulation of the adaptive communication module and the self-organizing network data link, and is used for realizing communication between simulation entities.

[0006] This invention also provides a collaborative control method for a heterogeneous interconnected multi-core distributed simulation experimental system, deployed in the online task planning module of an online task planning simulation system and / or an external task computer for collaborative control of simulation entities, including: task allocation, assigning tasks to different simulation entity clusters; planning actions for the host of each simulation entity cluster and calculating the follow-up action instructions of the slave machines within the simulation entity cluster; cluster control, controlling the actions of each simulation entity within the simulation entity cluster and updating the individual machine action information of each simulation entity, and feeding back the individual machine action information; and calculating the overall action control quantity of the simulation entity cluster based on the fed-back individual machine action information. The task allocation steps include: mathematically modeling the tasks and simulation entity clusters, defining the relevant attributes of the tasks and simulation entity clusters; constructing a task cost function based on the characteristics of the tasks and the relevant attributes of the tasks and simulation entity clusters; filtering out legal tasks and simulation entity clusters, generating a set of legal tasks and a set of legal simulation entity clusters that can be allocated; comparing the number of tasks with the number of resources, if the number of tasks is greater than the number of resources, creating a full-rank sparse matrix with the number of tasks as the order, and if the number of tasks is less than or equal to the number of resources, creating a full-rank sparse matrix with the number of resources as the order; establishing a random particle mathematical model by randomly shuffling the column order of the full-rank sparse matrix; indexing each column using the task cost function, with legal tasks as columns and legal simulation entity clusters as rows. The task cost matrix is ​​calculated for each task pair under the row index and the result is filled into a zero matrix with the same order as the particle to generate a set of backup task pairs. A loop iterates through the new particles, performing random genetic operations and evaluating the value and variance of each particle using the task cost matrix. Locally optimal particles and the current globally optimal particle are recorded at different iteration counts until the iteration is complete. The globally optimal particle is decoded to obtain a set of urgent task pairs. For simulation entity clusters without assigned tasks, the allocation scheme in the backup task pair set is used to supplement them. Finally, the urgent task pair set is combined with the unassigned portions of the backup task pair set to obtain a task allocation scheme.

[0007] Furthermore, the task cost function is: in, For the value of the task, For the task Execution priority For cluster Not capable of executing the first The ability to perform such tasks It is a task Is there a completion indicator? For the task The remaining valid time, For cluster Execute the task Cost time spent, To let the cluster Perform the task The allocation task pair; task The remaining effective time of the formula: Calculate; wherein, The remaining effective time of the task , The start execution time of the task , The end time of the task ; The task execution time threshold at The moment, The cluster Perform the task The cost time spent is calculated by the formula: ; wherein, The cost time spent by the cluster Perform the task , The center position of the task , The number of simulation entities in the cluster , The average speed of the cluster , The center position of the cluster . Further, at time t, the formula: ; screening legal simulation entity cluster; wherein, The legal simulation entity cluster set , The operation of traversing from the first task to the Task to determine whether the cluster Has the ability to perform tasks, The operation of traversing from the first task to the Task to determine whether the cluster Can reach a certain task location within a specified time, The time criterion, U is the simulation entity cluster set, U i The ith simulation entity cluster , The number of simulation entities in the cluster Calculate the time criterion by the formula: ; wherein, The time criterion , The start execution time of the task , The center position of the task , The central location, For cluster Average velocity, where t is at time t; at time t, using the formula: Filtering for legitimate tasks; among them, For a set of legitimate tasks, For the task The end time, M is the task set, M j For the j-th task, For the first cluster to the second The cluster is traversed to determine the task. Whether the operation can be executed by a certain cluster. Further, using the formula: The computation task is a set; where, For a set of task pairs, For a set of legal tasks The j-th task in A collection of legitimate simulated entity clusters The i-th cluster in For task value evaluation function, To enable cluster Execute the task The task allocation pair To retrieve the cluster from the set of legitimate tasks The task with the highest value is selected and output. Further, using the formula: Calculate the globally optimal particle; where, The globally optimal particle. It is a set of positive integers. The number of iterations for particle optimization. This is the sequence number of the loop. For the first The local optimum particle in the second cycle of optimization. For the first Local optimal particle in the second cycle The random permutation matrix, For the matrix and Perform matrix multiplication. This indicates that the standard deviation is calculated only for all non-zero elements in the matrix; using the formula: ; Calculate the local optimal particle; where, For the first The local optimum particle in the second cycle of optimization. It is a set of positive integers. For random particles The randomly inserted portion comes from the globally optimal particle. Gene or the local optimum particle from the previous cycle Genetic particles Genetic particles The permutation matrix, The task cost matrix, Let h be the number of random particle iterations, and h be the random particle iteration number. For genetic operators, This indicates that the standard deviation is calculated only for all non-zero elements in the matrix. Further, using the formula: , Calculate emergency task pairs; among them, For the urgent task pairing, To determine the number of legitimate simulated entity clusters, For the number of legal tasks, Assignment matrix for emergency rescue plans, Emergency Rescue Allocation Matrix The Line 1 The task pair of the column , and These are the globally optimal particle matrix and its permutation matrix, respectively. It is a task-to-relationship mapping matrix.

[0008] Furthermore, when the simulated entity is a digital simulated UAV, the swarm control steps include: solving for the height difference between each simulated entity and the reference point; calculating the layered height difference between each simulated entity; determining whether layered control is needed based on the heading angle deviation and formation control deviation; if layered control is needed, controlling the horizontal position of the simulated entity according to the set formation and controlling the height position of the simulated entity according to the layered height difference; if layered control is not needed, controlling the horizontal and height positions of the simulated entity according to the set formation; calculating the formation control command for the aircraft; when the main unit turns, the heading angle control command is calculated based on the heading angle commanded by the main unit; when the main unit does not turn, the heading angle control command is calculated based on the actual heading angle of the main unit.

[0009] Beneficial effects: The heterogeneous interconnected multi-core distributed simulation experiment system provided by the application effectively combines a dynamic deduction visual simulation engine, an environment physical model, a virtual-real combined simulation experiment platform, a semi-physical simulation system and an RTX / WIN multi-core parallel simulation framework, forms a general unmanned aerial vehicle online cluster simulation experiment platform, can perform unmanned aerial vehicle cluster semi-physical simulation, digital semi-physical simulation and virtual-real combined cluster simulation experiment, adaptive communication algorithm, cluster planning algorithm and task-driven large-scale cluster scene simulation verification, can realize unmanned aerial vehicle task-level scene playback experiment and virtual-real combined cluster simulation experiment. The heterogeneous interconnected multi-core distributed simulation experiment system has the capability of multi-core parallel simulation, can break through the hardware limit under the condition of ensuring data accuracy, realizes the multi-core parallel simulation capability of the semi-physical simulation model, and makes the scale of the cluster simulation multiple. The experiment platform has an environment model, and can realize high-fidelity simulation of environment data in the environment construction scene in combination with the physical simulation engine. The standardized data interface can realize cluster semi-physical simulation experiment of different adaptive communication algorithms, intelligent task planning algorithms and path planning algorithms, reduces the unmanned aerial vehicle cluster intelligent planning and scheduling research and development cost, improves the equipment key parameter verification capability and sample experiment playback technical difficulty. The cluster simulation experiment platform is used for constructing a virtual-real combined integrated experiment simulation environment, reduces the equipment experiment research and development cost, constructs an environment close to the real physical simulation environment by means of the virtual engine and the physical engine, constructs a semi-physical simulation model and introduces physical parameter related interference, and builds a virtual-real integrated simulation experiment platform. The cluster simulation experiment platform improves the capability of the computer simulation formal environment, reduces the experiment research and development cycle of the cluster method verification. The cluster simulation experiment platform adopts a standard data interface, can integrate different types of adaptive communication task planning algorithms and path planning algorithms, and can verify different types of unmanned aerial vehicle control algorithms by means of the experiment platform. The cluster simulation experiment platform can analyze the key samples of the unmanned aerial vehicle cluster simulation experiment, improves the simulation experiment reliability and fidelity. At the same time, the task-level large-scale cluster simulation experiment task-level verification can be performed. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description.

[0011] Figure 1 It is a structure principle diagram of the multi-core parallel heterogeneous interconnected simulation experiment platform; Figure 2 It is a structure principle diagram of the multi-core simulation management system; Figure 3 It is a communication architecture schematic diagram of the multi-core parallel framework; Figure 4 It is a principle diagram of the ad hoc network data link simulation system; Figure 5 It is a principle diagram of the heterogeneous multi-node adaptive communication mechanism; Figure 6Schematic diagram for whether to meet the Beta angle detection algorithm condition (meet the Beta angle detection algorithm); Figure 7 Schematic diagram for whether to meet the Beta angle detection algorithm condition (not meet the Beta angle detection algorithm); Figure 8 Flow chart of the cluster cooperative control method; Figure 9 Flow chart of the cluster task allocation algorithm; Figure 10 Flow chart of the cluster formation control algorithm; Figure 11 The plane coordinate system of the north east ground. DETAILED DESCRIPTION

[0012] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Those of ordinary skill in the art can understand the specific meanings of the above-mentioned terms in the present application according to the specific circumstances. In this article, “multiple” means two or more, that is, it includes two, three, four, five, etc.

[0013] Embodiment one: the present application is based on the design of the RTX / Window multi-core parallel simulation parallel framework, and combines a semi-physical framework adaptive communication mechanism and an unmanned aerial vehicle cooperative control algorithm to construct a multi-core parallel heterogeneous interconnected multi-core distributed simulation experiment system, such as Figure 1As shown, it specifically includes a semi-physical simulation system and a virtual system. Specifically, the semi-physical simulation system includes: a plurality of semi-physical simulation nodes, each of which can simulate a semi-physical simulation entity; a plurality of multi-core parallel simulation nodes, each of which can simulate a plurality of parallel semi-physical simulation entities; the virtual system includes: a digital simulation system, which can simulate a plurality of parallel digital simulation entities; an experimental environment model for simulating the generation of experimental environment modeling data; a virtual-real combined simulation experiment platform for deducing and rendering a working scene; and a dynamic deduction visual simulation engine for rendering environmental information using simulated experimental environment modeling data. The above-mentioned semi-physical simulation entities, parallel semi-physical simulation entities, and parallel digital simulation entities are collectively referred to as simulation entities hereinafter. In addition, for ease of description, the simulation entities in this embodiment are taken as simulation aircrafts (drones) as an example. The semi-physical simulation nodes are usually connected to some real hardware devices, which can be some key components on the drone, such as flight control system hardware, sensor hardware, etc. Taking the flight control system of the drone as an example, a real flight control computer can be directly used to participate in the simulation by using its actual computing power and control logic, which makes the semi-physical simulation entity closer to the physical characteristics and hardware response of the real drone. Since it contains real hardware, the semi-physical simulation entity can accurately simulate the actual working characteristics of the drone hardware, such as the measurement error of the sensor and the response delay of the flight control system, which has high precision for verifying algorithms and systems related to the hardware layer. The multi-core parallel simulation node mainly relies on the computing power of the multi-core processor. The multi-core parallel simulation node uses multiple processor cores to process computing tasks in parallel, fully utilizes the parallel computing performance of the hardware, and improves the efficiency of the simulation. For example, in large-scale drone cluster simulation, each core can be responsible for a part of the computing task of a drone, such as dynamics calculation, sensor simulation, etc. Through accurate mathematical models and efficient algorithms, the parallel semi-physical simulation entity can accurately simulate the dynamics, kinematics, and other characteristics of the drone. In processing large-scale drone cluster simulation, the computing accuracy and real-time performance of each drone model can be ensured through reasonable task allocation and parallel computing. The digital simulation system completely simulates various behaviors and characteristics of the drone based on mathematical models and software codes. By writing program codes to describe the dynamics equation, sensor model, communication model, etc. of the drone, numerical calculations are performed using the computing resources of the computer, so as to present the running state of the drone in the virtual environment. The parallel digital simulation entity can flexibly adjust the simulation accuracy according to specific requirements and the complexity of the model. Through the establishment of detailed mathematical models and the use of advanced numerical calculation methods, various behaviors of the drone can be simulated in depth, but the authenticity of the simulation is lower than that of the semi-physical simulation entity and the parallel semi-physical simulation entity.

[0014] The heterogeneous interconnected multi-core distributed simulation experiment system combines a dynamic deduction visual simulation engine and an environment physical model, a virtual-real combined simulation experiment platform, a semi-physical simulation system and an RTX / WIN multi-core parallel simulation framework, forms a general unmanned aerial vehicle online cluster simulation experiment platform, can perform unmanned aerial vehicle cluster semi-physical simulation, digital semi-physical simulation and virtual-real combined cluster simulation experiment, adaptive communication algorithm, cluster planning algorithm and task-driven large-scale cluster scene simulation verification, can realize unmanned aerial vehicle task level scene playback experiment and cluster virtual-real combined joint simulation experiment. The heterogeneous interconnected multi-core distributed simulation experiment system has the capability of multi-core parallel simulation, can break through the hardware limitation under the condition of ensuring data precision, realizes the multi-core parallel simulation capability of the semi-physical simulation model, and makes the scale of the cluster simulation multiple. The above parts are described in detail below.

[0015] One, the half physical simulation system. In this embodiment, the heterogeneous half physical simulation system is composed of multiple half physical simulation nodes and multiple multi-core parallel simulation nodes. 1. The half physical simulation node is mainly composed of a real-time simulation machine, a flight control computer, an external task computer, a communication module and a half physical simulation framework and the like, which is used to build a lightweight simulation node. The additional torque load simulator, intelligent IO device, full-quantization turntable, GPS simulator and inertial navigation simulator and the like constitute a full-quantization simulation node. (1) The real-time simulation machine (not shown in the figure) is used to build a half physical simulation entity model. (2) The flight control computer is used to control the motion of the half physical simulation entity built by the real-time simulation machine, and is specifically used for flight control of a six-degree-of-freedom aircraft. The flight control computer is composed of a sensor, an ARM processor and a high-performance DSP processor, and mainly completes functions such as attitude information solving, position information node, attitude control, position control and task control. The high-performance DSP processor is designed to have high control precision, high wind resistance, high operation speed and the like, the ARM processor has the ability of data access through a variety of hardware peripheral interfaces, and is combined with a high-performance inertial navigation sensor (IMU) and a high-precision GPS through external interfaces, so that the DSP processor is quickly solved, so that the half physical simulation node has good control effect. (3) The external task computer includes an online task planning module for simulating cluster task planning and distribution to cooperatively control the cluster and an adaptive communication module for realizing communication between simulation entities. The external task computer is based on an NVIDIA series core processor, loads algorithms such as cluster task distribution, flight path planning and formation keeping and switching into the internal core processor, realizes combat modes such as autonomous decision of the unmanned aerial vehicle cluster, online task distribution and formation switching, and the like. The data communication module of the intelligent planner is mainly responsible for data interaction with the flight control terminal and the self-organizing network data link and the like, effectively guarantees information transmission between the unmanned aerial vehicles, and provides an effective data channel for online dynamic planning and formation switching of the unmanned aerial vehicle cluster. (4) The self-organizing network data link is a physical communication link between the half physical simulation entities, has a high-performance wireless MESH data transmission module, and has multiple anti-interference channels, so as to realize high-speed and large-bandwidth data transmission in long-distance communication. Through the self-organizing network data link, data interaction between the unmanned aerial vehicles can be efficiently realized, the real-time data communication process in the unmanned aerial vehicle cluster is completed, and effective online simulation verification of the virtual-real combined half physical system of the unmanned aerial vehicle cluster is ensured. It is worth noting that the multi-core parallel node and the digital simulation system are limited by hardware, and the self-organizing network data link is simulated by the self-organizing network data link simulation system through optical fiber reflection memory.2. The multi-core parallel simulation node is mainly composed of a real-time simulation machine, a multi-core parallel simulation framework, a self-organizing network data link simulation system and an online task planning simulation system, and is used for constructing a functional level and a functional level lightweight hardware-in-the-loop agent simulation node. The node has the ability of multi-core simulation, which can further improve the upper limit of the hardware-in-the-loop model accessing the hardware-in-the-loop system. The hardware-in-the-loop system has the ability of cluster simulation. (1) The real-time simulation machine (not shown in the figure) is used for constructing a parallel hardware-in-the-loop entity model. The six-degree-of-freedom model of the hardware-in-the-loop simulation fixed-wing unmanned aerial vehicle has the functions of flight control, attitude control and trajectory tracking. Through the model, multiple virtual aircraft devices can be created to effectively simulate the functions of the real aircraft control device. (2) The multi-core parallel simulation framework deployed in the RTX real-time system is used for parallel hardware-in-the-loop simulation entity test state configuration, experimental process control, and test data storage and processing. The RTX multi-core parallel simulation framework is a multi-core parallel extension based on the hardware-in-the-loop integration framework, which has the ability of multi-core parallel simulation of hardware-in-the-loop models. In order to break through the hardware limitation of the hardware-in-the-loop simulation system and ensure the model precision, the framework is mainly used for functional level and functional level lightweight hardware-in-the-loop model simulation, and provides real-time simulation solution results of key simulation parameters for the hardware-in-the-loop system. (3) The online task planning simulation system is a digital simulation of the online task planning module, which is used for simulating entity cluster task planning and distribution to cooperatively control the cluster. The online task planning simulation system constructs and loads algorithm modules such as cluster task distribution, flight path planning, formation control and formation change, which has the same effect as the online task planning module in the external task computer. (4) The self-organizing network data link simulation system is a digital simulation of the adaptive communication module and the self-organizing network data link, which is used for realizing the communication between simulation entities.

[0016] II. Virtual system. The virtual system is composed of an experimental environment model, a virtual-real combined simulation experiment platform, a digital simulation system, and a dynamic deduction visual simulation engine. The virtual simulation system adopts a digital simulation mode, can realize the same cluster simulation function as the real system, and can freely expand the number of simulated unmanned aerial vehicles, thereby realizing large-scale unmanned aerial vehicle cluster virtual-real combined online simulation verification on the basis of saving experimental costs. 1. The digital simulation system includes a self-organizing network data link simulation system, an online task planning simulation system, and a multi-core parallel simulation framework. (1) The multi-core parallel simulation framework deployed in the Windows system and the digital simulation model based on the multi-core parallel simulation framework are used for parallel digital simulation entity test state configuration, experimental process control, and test data storage and processing. The digital simulation model in this embodiment is a six-degree-of-freedom model of a fixed-wing unmanned aerial vehicle, which has functions such as flight control, attitude control, and trajectory tracking. Through the model, multiple virtual aircraft devices can be created to effectively simulate the functions of real flight control devices. The Window multi-core parallel simulation framework is a model filling and integration framework for connecting digital models in a digital system to a digital simulation system, mainly used for unmanned aerial vehicle digital model simulation, and has the ability of multi-thread running, which can realize single model and multi-model simulation calculation. It provides key simulation parameter calculation results for the digital system. In the laboratory environment, both the semi-physical simulation system and the digital simulation system can effectively simulate the unmanned aerial vehicle cluster flight process. (2) The online task planning simulation system is a digital simulation of the online task planning module, which is used for simulation entity cluster task planning and distribution to cooperatively control the cluster. The online task planning simulation system plays the same role as the online task planning module in the external task computer by constructing and loading algorithm modules such as cluster task distribution, flight path planning, formation control, and formation change. (3) The self-organizing network data link simulation system is a digital simulation of the adaptive communication module and the self-organizing network data link, which is used to realize communication between simulation entities. It is a data interaction channel between each simulation unmanned aerial vehicle in the virtual simulation system. 2. The experimental environment model is used to simulate the generation of experimental environment modeling data. Specifically, it is used to simulate the generation of experimental environment modeling data such as wind, rain, fog, and snow, and to input the environmental data into the dynamic deduction visual simulation engine to render and construct multi-dimensional scene information. 3. The virtual-real combined simulation experiment platform is used for the deduction and rendering of the working scene. Specifically, it is jointly developed based on UE5 and C++ programming under the GUI framework platform and mainly includes combat scene deduction and rendering functions such as cluster control instruction issuance, task distribution, flight path planning, scene construction, and data communication, effectively completing the 3D rendering display of the corresponding combat tasks. Through simulation flight path and data online analysis, the online real-time task distribution efficiency evaluation and online evaluation of multi-task combat capability of the cluster combat are completed, and the verification and evaluation of the online simulation algorithm of the unmanned aerial vehicle cluster are efficiently completed.4. Dynamic deduction visual simulation engine, used for rendering environment information by using simulation experiment environment modeling data. The main function is to build a cluster combat simulation scene, access experiment environment simulation model environment data, and render high-fidelity environment information. With the help of fiber reflection memory network, six-degree-of-freedom data of each simulation model in the semi-physical simulation system and the digital simulation system is obtained. The cluster simulation control command sent by the virtual-real combined simulation experiment platform drives the unmanned aerial vehicle cluster to perform three-dimensional path deduction display in real time online. Finally, the three-dimensional experiment scene rendered and constructed is returned to the virtual-real combined simulation experiment platform for deduction display.

[0017] III. Multi-core parallel simulation system. In this embodiment, two sets of multi-core parallel simulation frameworks are included, i.e., a multi-core parallel simulation framework deployed in an RTX real-time system (belonging to a multi-core parallel simulation node) and a multi-core parallel simulation framework deployed in a Windows system (belonging to a digital simulation system). The two sets of multi-core parallel simulation frameworks are respectively used for management of parallel semi-physical simulation entities and parallel digital simulation entities. The RTX version adopts a classic real-time operating system architecture, and through hardware interrupts and priority scheduling, strong real-time performance is ensured, and the response period is stably controlled within 1 ms. The Windows 11 version is based on a standard Windows system, and a response service is implemented by using a Windows program. Although the strong real-time performance of the RTX cannot be achieved, through optimization of a scheduling algorithm and system resource management, a relatively low response delay is achieved, and the fluctuation is controlled within 12 ms. The two sets of multi-core parallel simulation frameworks belong to the multi-core parallel simulation system. For example, Figure 2As shown, the multi-core parallel simulation system adopts a hierarchical host-computer structure. The host computer is the control host of the multi-core parallel simulation system, runs the Windows 11 system, and is mainly responsible for test state configuration, experimental process control, and test data storage and processing. The lower computer module is divided into two groups: the lower computer runs the Windows 7 system and carries the RTX high-version real-time operating subsystem supporting multi-core parallel calculation, and the other group of lower computers runs the Windows 11 system to adapt to the dependence of some models on high-version systems. The host computer deploys a virtual-real combined simulation experiment platform, in which the multi-core simulation system is integrated inside the platform. The main functions of the platform are to configure the test state, control the experimental process, and store and process the test data. The host computer and the lower computer communicate through a reflective memory. The lower computer mainly runs a simulation scheduling management program and a real-time simulation, responds to the instructions of the host computer in real time, and schedules the simulation tasks in real time according to the clock signal to complete the simulation test. The two sets of multi-core parallel simulation frameworks on the lower computer both adopt the implementation scheme of "management task scheduling and operation part periodic execution", combined with the Reactor+thread pool mode based on NIO. The Reactor mode is an event-driven asynchronous mode that handles I / O events (multi-node simulation events) through one or more Reactors, suitable for high-concurrency scenarios. It separates the listening and processing of multi-node simulation events, improving the response speed of the system. The thread pool can effectively manage thread resources, avoiding the overhead of thread creation and destruction, and improving the concurrent processing capability of the system. To ensure reasonable use of thread resources, the Reactor listening scheduling thread is strongly bound to CPU0 to ensure that multi-node simulation events are processed in real time and efficiently. The multi-node simulation events are distributed to the corresponding event processing threads for real-time simulation, minimizing the coupling degree of software structure and control flow. In multi-core simulation task scheduling, the RTX version realizes task scheduling and data exchange through the RTX response service once every 1 ms. The control end issues a task scheduling instruction every 1 ms, and the real-time simulation operation end polls the task scheduling instruction every 1 ms and responds. The communication between simulation cores strictly follows the order of judging the interrupt state first and then accessing the memory, ensuring efficient coordination of multi-task processing. The Windows 11 version adopts a similar mechanism, but relies on the kernel optimization and user-state scheduling of the Windows system to achieve near-real-time task response. The real-time parallel simulation Reactor listening thread is responsible for monitoring the simulation test process, scheduling tasks for each core, coordinating inter-core communication, allocating shared resources, handling simulation test faults, and collecting and processing data in both frameworks. In the RTX version, strong real-time performance makes it more suitable for high-precision simulation tasks sensitive to time delay; while the Windows 11 version provides better system compatibility and model adaptation capability, suitable for application scenarios that require low real-time performance but high-version system support.

[0018] IV. Multi-core parallel simulation framework communication architecture. For example... Figure 3 As shown, the framework management mainly includes the division of real-time parallel simulation task modules, the allocation of simulation tasks on each core, and the real-time scheduling of simulation tasks. Real-time task scheduling is a crucial aspect of parallel simulation system design. The main purpose of task scheduling is to rationally schedule tasks to execute on various processors, minimizing program execution time while ensuring the accuracy of simulation results. The designed real-time parallel simulation task scheduling adopts an RTX clock signal-driven scheduling mechanism. The RTX clock signal is generated by the RTX software system, which has dynamic priority allocation capabilities, task switching time of less than 5.2µs, and interrupt response time of less than 5µs, making it a real-time simulation operating system. The interrupt service routine under the RTX system notifies the occurrence of the simulation event in the shortest possible time, while other high-frequency calculations and data interactions are performed within the triggered simulation task itself through the communication mechanism between the interrupt and the task. This avoids the various limitations of writing interrupt service routines and further reduces interrupt latency. On the Windows side, the real-time parallel simulation framework adopts a similar task scheduling mechanism, but since the Windows system itself is not a real-time operating system, the response time and the determinism of task scheduling are somewhat lower than RTX. Task scheduling on the Windows side primarily relies on kernel optimization and user-mode scheduling of the operating system. Through high-priority threads and fine-grained task management, near real-time task response is achieved. The response time of the Windows system is typically controlled within 12ms. While it cannot achieve the strong real-time performance of RTX, it still meets the needs of simulation scenarios with lower real-time requirements, while providing better system compatibility and model adaptation capabilities. After a simulation task is activated, the simulation results from the previous cycle are first updated to the shared memory cache. When the framework management detects updates to simulation event data on other CPU cores in the cache, it transmits the data to the host computer via reflected memory and then activates the real-time step-by-step solution tasks on other cores. Driven by external event instructions, the parallel simulation framework management periodically activates model tasks, updates simulation results, and sends them to the host computer until the simulation experiment ends. The inter-task communication models supported by parallel simulation mainly include: shared memory, semaphores, and reflected memory. The advantages of using shared memory for inter-task communication are the fastest access speed, the ability to specify the size of the data area, and the ability to freely configure the structure of the data area as needed. Using reflective memory for interaction with the host computer and external devices, the Fiber-Optic Reflective Memory Network adopts high-speed fiber optic hardware data sharing technology. Compared with Ethernet communication networks, it not only has strict transmission determinism and predictability, but also features simple data communication protocols, fast data transmission, strong real-time performance, and good platform versatility.

[0019] Five, the node communication architecture of the heterogeneous interconnected multi-core distributed simulation experiment system. The traditional adaptive communication only solves the problem of cooperative communication between simulation nodes of the same type, and the data communication problem between virtual and real nodes has been a major technical difficulty in engineering. The adaptive communication and distributed simulation provided in this embodiment solve the multi-point communication problem between semi-physical simulation nodes and between semi-physical simulation nodes and digital simulation nodes in joint simulation. Combined with the optical fiber reflection memory network, the cooperative communication efficiency between large-scale heterogeneous cluster semi-physical simulation nodes is further improved. As shown in Figure 1 , the node communication architecture of the heterogeneous interconnected multi-core distributed simulation experiment system includes a self-organizing network data link, an adaptive communication module (deployed in an external task computer) and a self-organizing network data link simulation system provided in the multi-core parallel simulation node. The self-organizing network data link simulation system is actually a digital simulation of the self-organizing network data link and the adaptive communication module. The semi-physical simulation entities simulated by the semi-physical simulation node, the parallel simulation entities simulated by the multi-core parallel simulation node, and the semi-physical simulation entities and the parallel simulation entities communicate with each other using the self-organizing network data link or / and the self-organizing network data link simulation system. To meet the distributed simulation needs of large-scale simulation entities, an optimal communication mechanism and a semi-physical model distributed simulation strategy for large-scale simulation entity scheduling management are constructed to further improve the virtual and real interconnection and cluster control semi-physical simulation capability of the simulation experiment platform. The principle of the self-organizing network data link simulation system is shown in Figure 4 . 1. Adaptive communication method. To improve the communication efficiency of large-scale simulation entities of the simulation experiment platform, in view of the problems of high communication delay, limited bandwidth, frequent communication failures and the like of large-scale entities, a distributed communication interaction strategy design is adopted, the optimal interaction time selection and the least interaction content extraction are studied, and a distributed adaptive communication mechanism is constructed in a multi-task scenario to form the adaptive communication simulation capability of the simulation experiment platform. Specifically, as shown in Figure 5As shown, the adaptive communication method between simulation entities comprises: S1, detecting the state of the simulation entity and judging the event triggering occasion; the simulation entity comprises a semi-physical simulation entity and a parallel simulation entity; S2, when the event communication is triggered, determining the simulation entities for mutual communication; S3, transmitting data between the determined simulation entities for mutual communication, and encoding and decoding the data. Each simulation node in the figure adopts the above adaptive communication mechanism. The semi-physical simulation nodes communicate with each other by using Ethernet; the multi-core parallel simulation nodes communicate with each other by using optical fiber reflection memory; the semi-physical simulation nodes and the multi-core parallel simulation nodes communicate with each other by using Ethernet; the parallel simulation entities in the multi-core parallel simulation nodes communicate with each other by using shared memory. As can be seen, due to hardware limitations, the semi-physical simulation nodes will externally access various hardware devices, the semi-physical nodes adopt the adaptive communication mechanism to build a traditional Ethernet mode, the multi-core parallel simulation nodes adopt a multi-core simulation form by means of reflection memory to break through the hardware limitations, and the ad hoc network communication part adopts a data link simulator to complete the ad hoc network communication link simulation. The adaptive communication mechanism between the multi-core parallel simulation nodes is built by using optical fiber reflection simulation. The adaptive communication algorithm and the cluster task planning algorithm of the semi-physical simulation node run on an external task computer, the adaptive communication algorithm of the multi-core parallel simulation node is realized by an ad hoc network data link simulation system, mainly used for building the data link between various system simulation subsystems, the cluster task planning algorithm is completed by an online task planning simulation system, mainly used for task planning and flight path planning of each node in the cluster simulation node and cluster formation formation maintenance and other functions. In the adaptive discrete event triggered communication mechanism, whether the sampling data is transmitted depends on whether the error between the current time and the transmission value at the last time is greater than a fixed threshold. Although the introduction of the event triggering mechanism saves certain network resources on the premise of ensuring system performance, the fixed threshold value may cause the following problems if the value is not properly selected: (1) when the system is in a stable state, a large number of data packets are still transmitted, and in actual engineering, when the system is stable and has good performance, the transmission of data packets needs to be reduced; (2) when the system is not triggered for a long time, the time delay will increase, and thus the system stability will be poor. In order to solve this problem, the embodiment dynamically adjusts the triggering threshold according to the running behavior of the system, trying to achieve a better balance between system performance and network resource saving. More specifically, the step of judging the event triggering occasion in step S1 comprises:

[0020] S11, calculating the state offset according to the current state estimation value of the simulation entity and the system state estimation value satisfying the event triggering condition at the last time; and calculating the dynamic change scaling factor by using the state offset.

[0021] The method for calculating the dynamic change scaling factor comprises using the formula: , a dynamic changing scaling factor is calculated; wherein λ is the dynamic changing scaling factor, Δ is the normalized state change rate, ε is the offset correction, α and is the fuzzy rule strength coefficient, is the current system state estimation value satisfying the event trigger condition, is the last system state estimation value satisfying the event trigger condition. If , , is a given constant, then is the time trigger mechanism of the dynamic threshold. The arctangent function combines the parameters and to automatically adjust the threshold parameter , . If , then , , a smaller is needed to speed up the network data transmission; otherwise, a larger is used to reduce the transmission frequency. The dynamic change of the trigger parameter dynamically adjusts the data transmission, and ultimately promotes the optimal compromise balance between system performance and communication resources. In particular, if or , then it is the event trigger mechanism of the fixed threshold. The upper limit of the automatically adjusted threshold parameter is set to limit the transmission delay when the system is in a steady state, i.e., to prevent long-term non-transmission of data when the system is in a steady state. A small positive number is introduced as an offset correction to adjust the output of appropriately. S12 adjusts the automatically adjusted threshold parameter satisfying the event trigger condition at the last time using the dynamic changing scaling factor and the upper and lower bounds of the automatically adjusted threshold parameter, thereby obtaining the automatically adjusted threshold parameter at the current time. Specifically, the automatically adjusted threshold parameter at the current time is calculated using the formula: ; wherein is the automatically adjusted threshold parameter at the current time, is the automatically adjusted threshold parameter at the last time, λ is the dynamic changing scaling factor, is the lower limit of the automatically adjusted threshold parameter, is the upper limit of the automatically adjusted threshold parameter. S13 performs threshold adjustment using the automatically adjusted threshold parameter at the current time, and takes the time satisfying the adjusted threshold requirement as the time satisfying the event trigger condition at the next time. Specifically, the time satisfying the event trigger condition at the next time is calculated using the formula:

[0022] ; wherein t k+1 h is the time satisfying the event trigger condition at the next time, t kh is the current time satisfying the event trigger condition, h is the sampling period, Z is the number of sampling periods, N is a natural number; is the state estimation error, is the system state estimation value at the current time, is the system state estimation value at the current time satisfying the event trigger condition; T is the transpose symbol, is the automatic adjustment threshold parameter at the current time, is a positive definite weight matrix. is an integer and .

[0023] The starting time is in a discrete state at each node, and when the communication event is triggered, monotonically does not increase with time, that is: By mathematical induction, it can be proved that: When t = h, it satisfies ; When t = h + Z, it can be obtained that:

[0024] , where: represents the number of monitoring periods with as the base over time is the exponential function of the variable, and c is a constant, which is an empirical value, and represent the weight, and . Since , there is When t = h + Z, it satisfies ; When t = h + 2Z, assuming , from the above formula, it can be deduced that:

[0025] ;

[0026] According to the assumption, it is deduced that:

[0027] , and then . That is, the dynamic threshold function is a monotonically non-increasing function. The threshold change law can be consistent with the choice of communication trigger time. Moreover different values of can be used to represent different communication mechanisms. When changes over time, it represents a dynamic event trigger mechanism; when The constant value represents a static event-triggered communication mechanism, i.e., a traditional periodic communication mechanism. S14 determines whether the next time the event trigger condition is met belongs to Zeno behavior. If not, the simulation entity can communicate at the next time the event trigger condition is met. When an event is triggered, it's necessary to determine if Zeno behavior exists. Zeno behavior refers to the behavior of generating an infinite number of operations within a finite amount of time, meaning the event trigger interval tends to be infinitesimally small. This leads to frequent communication, consuming excessive communication resources and failing to achieve resource conservation. Therefore, Zeno behavior should be avoided when designing trigger-based methods; otherwise, communication cannot be optimized and may even increase the communication burden. This step identifies Zeno behavior by determining whether the next event time is too dense; it allows reasonable events to trigger communication, avoiding unreasonable events that cause simulation crashes; and it ensures that the communication behavior of the simulation entity is both timely and does not disrupt time continuity, which is a key guarantee for the stable operation of complex event-triggered simulation systems. If it is not Zeno behavior, it means that the simulation entity can establish a communication connection at this moment. As the number of nodes within the communication radius increases, the number of links that each node needs to maintain under the communication mechanism increases significantly, and the corresponding computational and communication complexity also increases significantly. Furthermore, algorithms that do not consider the upper bound of vertex degree often suffer from the problem of maximum vertex degree, so it is also necessary to determine the simulation entities that communicate with each other. Step S2, determining the simulation entities that communicate with each other, includes: S21 constructing the communication mapping matrix and communication topology graph for each simulation entity. The Beta angle detection algorithm is a distributed communication topology control algorithm that enables each node to maintain necessary neighbor node links and remove redundant communication links. Due to the limitation of communication resources and the absence of centralized base stations or other facilities for allocating communication resources, the number of communication links that each node can maintain is limited. Therefore, the Beta angle detection algorithm must ensure that the network graph it generates has a finite upper bound of vertex degree, that is, the maximum number of edges of any node must be less than a certain constant value. To meet the distributed requirements, each node should be able to obtain information from its neighbor nodes within one hop range. The principle of the Beta angle detection algorithm is as follows:

[0028] Diagram The vertex set is ,vertex With vertex The edge between It exists if and only if for any vertex , interior angle satisfy ,in . Figure 6 The edges are shown in two dimensions. Does it meet the requirements? Conditions for corner detection algorithm. Nodes in the diagram. Located in the region outer, i.e. corresponding to are all less than satisfy , indicating that the edge satisfy the angle detection algorithm condition. And Figure 7 in are all located in inner, i.e. , i.e. the edge does not satisfy the angle detection algorithm condition. For ease of description, the above is based on the two-dimensional case, and the algorithm is still effective in the three-dimensional case.

[0029] The Beta angle detection algorithm has the following advantages: essentially, the Beta angle detection mechanism follows the principles of distribution and symmetry. Therefore, as the number of nodes increases, the generated network structure has certain scalability and robustness. In addition, by modeling the communication link between nodes as a bidirectional link, the symmetry of the Beta angle detection rule helps to avoid unnecessary trouble caused by hidden terminals and the like due to unidirectional links. As the number of nodes within the communication radius increases, the number of links each node needs to maintain under the full connectivity mechanism increases greatly, and the corresponding calculation and communication complexity also increases greatly. In addition, algorithms that do not consider the upper bound of vertex degree usually have the problem of . Under the Beta angle detection algorithm mechanism, the vertex degree has a constant vertex degree upper bound, which is a function of and is independent of the number of nodes . The maximum vertex degree upper bound is shown in the following formula.

[0030] In the two-dimensional case: ; in the three-dimensional case: .

[0031] In summary, the Beta angle detection algorithm has the following characteristics: ① the vertex degree upper bound is controllable, ② adaptive implementation, ③ symmetry is met. In this embodiment, the specific steps of constructing the communication mapping matrix of each simulation entity and the communication topology graph using the Beta angle detection algorithm include: S211 selecting a neighbor node from the neighbor nodes of a simulation entity, and calculating the included angle formed by the simulation entity, the neighbor node and other neighbor nodes. For example, the neighbor nodes of the simulation entity A are B, C and D, and the preset Beta is 30°; the neighbor node B is selected, and the included angle between the connection A-B and the connection A-C is calculated as ∠CAB=25°, and the included angle between the connection A-B and the connection A-D is calculated as ∠DAB=28°. S212 comparing the included angle with the preset Beta value, and recording the number of times that the included angle is less than the Beta value. In the above example, both of the included angles are <30°, and the number of times is 2. S213 establishing a communication link between the simulation entity and the neighbor node in the case that the number of times that the included angle is less than the Beta value is equal to the number of neighbor nodes. In the above example, when the number of times is equal to the number of neighbors, the A-B link is established, i.e. a=1. If one of the included angles is ≥30°, i.e. the number of times is 1≠the number of neighbors, the A-B link is not established, i.e. a=0. S214 traversing all the neighbor nodes of the simulation entity and repeating steps S211-S213, so as to construct the communication mapping matrix and the communication topology graph of the simulation entity.

[0032] S22 calculates the final linking point using the communication mapping matrix and the communication topology graph to determine the simulation entities that communicate with each other. The communication topology formed by the Beta angle detection algorithm is usually modeled as an undirected graph, whose vertices and edges represent nodes and communication links between them, respectively. In addition, the second smallest eigenvalue of the Laplacian matrix corresponding to the communication topology graph, i.e., the algebraic connectivity, is used to measure the pros and cons of network connectivity. To better achieve data interaction, global dynamic data of nodes is obtained as much as possible within a non-period. The Beta angle detection algorithm can obtain the upper and lower bounds of the algebraic connectivity, and the node with the minimum vertex degree determines the upper bound of the algebraic connectivity of the network graph. Therefore, the existence of this node has a greater impact on the connectivity of the network, and therefore the number of low vertex degree nodes needs to be reduced to improve the distributed connectivity of the network. In this embodiment, the step of calculating the final linking point using the communication mapping matrix and the communication topology graph includes: S221 calculates the second smallest eigenvalue of the Laplacian matrix using the communication mapping matrix and the identity matrix. The second smallest eigenvalue of the Laplacian matrix is a core indicator of network connectivity, and the larger the second smallest eigenvalue, the better the connectivity; when the second smallest eigenvalue is 0, the network is disconnected. This step converts the topological relationship into a numerical feature and quantifies the global connectivity. S222 forms a feature pair by power iteration between the number of nodes in the communication topology graph and the second smallest eigenvalue of the Laplacian matrix. In this step, based on the number of vertices N determined by the communication topology graph, the feature pair, i.e., the eigenvalue and the corresponding eigenvector, is generated by the power iteration method. S223 calculates the estimate of the connectivity of the communication topology graph using the estimate of the eigenvector corresponding to the second smallest eigenvalue in the feature pair; the estimate of the connectivity of the communication topology graph is calculated using the formula:

[0033] ; wherein, is the estimate of the connectivity, L ij is the Laplacian matrix formed by node i and node j, is the element value corresponding to the Fiedler vector of the jth node in the kth iteration, is the element estimate of the Fiedler vector of the ith node, N i is the neighbor set of node i; define the coefficient matrix W such that the principal eigenvalue corresponds to the all-1 vector of the Laplacian matrix: ; wherein, is the identity matrix, denotes the minimum value, and satisfies , then the eigenvector of the algebraic connectivity is the principal eigenvector of the matrix. When the initial value satisfies , the iteration vector will converge to the second eigenvector of the Laplacian matrix the algebraic connectivity vector. Thus each node in the multi-node can update the corresponding element in the algebraic connectivity vector by a distributed power iteration method as follows: using the formula to calculate the element estimate of the Fiedler vector after iteration; where is the element value of the Fiedler vector corresponding to the ith node in the kth iteration, is the element value of the Fiedler vector corresponding to each node in the k-1th iteration, W i is the coefficient matrix of the ith node. Using the formula to calculate the element value of the Fiedler vector after iteration; where is the element value of the Fiedler vector corresponding to each node in the kth iteration. However, the power iteration has the following two problems: the existence of rounding errors in the iteration process, and only using the initial vector to obtain may gradually approach . . the calculation of requires global information, which will lead to frequent global information exchange, increase resource overhead, and cause the distributed characteristics to fail. To solve the above problems, we consider obtaining information globally to update at a specified number of iterations, while introducing a correction term , the iteration coefficient matrix is as follows: ; where is the coefficient matrix in the kth iteration, I is the identity matrix, L is the Laplacian matrix, J is the all-one matrix, is a very small positive value, k w is a given positive integer, n is the number of nodes, k is the iteration number, and "mod" represents the modulo operation. Thus each node only exchanges a large amount of data at , thereby reducing the global exchange of information. Under this mechanism, each node can obtain information and connected nodes in a distributed manner.

[0034] S224 increases the number of connected nodes if the estimated value of the connectivity of the communication topology graph is greater than the number of nodes, thereby forming the final link points. In graph theory and distributed systems, if the estimated value of the connectivity is much greater than n, it indicates that the theoretical connectivity potential of the current network is not fully released, i.e., the actual number of connected nodes is insufficient; on the contrary, if it is close to or less than n, the network connectivity is basically adapted to the scale. If the estimated value of the connectivity is greater than the number of vertices n, it indicates that the actual number of connected nodes of the current network is insufficient, and there are nodes that can be connected but are not connected, and the number of connected nodes needs to be increased. If the estimated value of the connectivity is less than or equal to the number of vertices n, it indicates that the number of connected nodes of the current network has adapted to the scale, and there is no need for additional expansion, and the existing link points are maintained.

[0035] In step S3, Huffman algorithm is used for encoding and decoding of data. In order to reduce the transmission amount of communication data and reduce the time delay, Huffman coding is used, which is an entropy coding (weight coding) algorithm for lossless data compression, which is a kind of variable length coding (VLC). This method constructs the code word with the shortest average length of different prefixes according to the character occurrence probability, sometimes called optimal coding. Huffman coding constructs an optimal binary tree with the help of the tree structure in the data structure under the support of Huffman algorithm. In computers, data storage and processing are based on bytes as the basic unit, a Western character is expressed by a byte, and a Chinese character is expressed by two bytes. We call this coding form that each character is expressed by the same number of bytes as fixed length coding. Huffman coding is a variable length coding, which constructs the coding with the shortest average length according to the probability of character occurrence. In other words, if a character appears frequently in a document, its coding is short, and if a character appears rarely in a document, its coding is long. When the length of each code word in the coding is strictly arranged in reverse order according to the probability of the corresponding symbol, the average length of the coding is the smallest. In an extended binary tree, the path from the root to the external node can be used for coding, which is to move one step to the left sub-tree by using 0 and to move one step to the right sub-tree by using 1. The path code from the root to the external node is 010. The path codes from the root to the node are (00, 010, 011, 100, 101, 11) respectively. Because each path is exactly the first half of another path, none of the path codes is a prefix of another path code. Therefore, these codes can be used to encode the characters respectively. Let S be a string composed of these characters, and be the frequency of the character , where belongs to the set If S is encoded using these codes, the length of the encoded bit string is:

[0036] For an extended binary tree with external nodes, and the external nodes are labeled as , the length of the corresponding encoding bit string is: .

[0037] where the path length from the root to the external node (i.e., the number of edges of the path); WEP is the weighted external path length of the binary tree. In order to shorten the length of the encoding string, the binary tree code must be used, the external nodes of the binary tree correspond to the characters of the string to be encoded, and the WEP is the minimum. A binary tree is called a Huffman tree if its WEP is the minimum for a given set of frequencies. The process of constructing a Huffman tree is to first establish a set of binary trees, each containing only one external node, each external node representing a symbol of the string, and its weight is equal to the frequency of the symbol. Then, constantly select two binary trees with the smallest weight from the set, merge them into a new binary tree, and the merging method is to add a root node and take the two binary trees as left and right child trees. The weight of the new binary tree is the sum of the weights of the two child trees. This process continues until only one tree is left.

[0038] For a cluster numbered , , it has the following properties: 1, the number of surviving UAVs in the cluster is , is an integer greater than or equal to zero; 2, at time , the geometric center coordinates of the cluster are , is the matrix transpose symbol, , and are the north direction coordinates, east direction coordinates and height coordinates of the center coordinates of the cluster at time ; 3, at time , the average flight speed of the cluster is , , and are the north direction speed component, east direction speed component and vertical to the horizontal plane upward speed component of the average flight speed of the cluster ; 4, the set of task type execution capabilities possessed by the cluster is , where is a symbol number equal to 0 or 1, representing a cluster not having the ability to perform the first kind of task, representing a cluster having the ability to perform the first kind of task. For the combat task numbered , , , it has the following attributes: 1, the type of task , , is a set of positive integers; 2, the task completion flag is , a symbol number equal to 0 or 1, indicating that the task has not been completed, indicating that the task has been completed; 3, the task execution priority is , is a set of positive integers, the greater the absolute value, the higher the execution priority of the combat task; 4, at the moment , the geometric center coordinates of the task are , , and are respectively the north coordinate, east coordinate and height coordinate perpendicular to the horizontal plane of the center coordinate of the task at the moment , , , , , , .

[0039] S1012 According to the characteristics of the task and the task and the simulation entity cluster related attributes to build the task cost function. Let the cluster perform the task allocation task pair is represented as , the value of the rescue task can be evaluated by designing the function . Given is the execution priority of the task , is the cluster not having the ability to perform the first kind of task, is the flag whether the task has been completed, then the function is defined as: wherein, is the remaining valid time of task , is the assigned task pair for cluster to execute task , and the start execution time and end time of task are known ; , the task execution time threshold at time , then the remaining valid time of task is defined as: is the cost time for cluster to execute task , and the number of UAVs, average flight speed and center position of cluster are known , and respectively, and the center position of combat task is , is the operator for calculating the Euclidean distance between two three-dimensional coordinate points, then is defined as: .

[0040] S1013 screens out the legal task and simulation entity cluster, and generates the legal task set and legal simulation entity cluster set that can be assigned. At time , the legal and valid cluster set is defined as: wherein, is a positive integer and ; is a positive integer and ; is the number of UAVs of cluster ; is an operation for traversing the first task to the nth task to determine whether cluster has the task execution capability, if there is at least one task that is not completed and the type is , so that cluster has the execution capability for the nth task type , then , otherwise ; ; is a time criterion for determining whether cluster can arrive at the designated place in time to execute the task , The definition of is: In the formula It is the present moment. and Clusters Average flight speed and center position, It is a cluster Execute the task The cost and time spent It is a task The start time of execution; This represents a sequence of tasks from the first task to the second task. Each task is used to iterate and determine the cluster. The calculation determines whether a task can be completed within a specified time; if at least one task exists... And the task was not completed. This enables the cluster Execute the task Time criteria ,but ,otherwise .exist At any given moment, a legal and valid set of tasks Defined as: ,in, It is the present moment; It is a positive integer and ; It is a positive integer and ; It is a task The end execution time; Indicates task The executable task has not yet ended; Represents a kind of cluster from the first cluster to the second cluster. The cluster is traversed to determine the task. Whether the operation can be executed by a cluster, if the task It was not completed And at least one cluster exists. The cluster Having the ability to execute Type of task The ability is , but ,otherwise ; It is a time criterion used in clusters. Whether they can arrive at the designated location in time to carry out the mission ; This represents a sequence of tasks from the first task to the second task. The task involves iterating through and judging each cluster. Whether the task can be successfully executed by a cluster within its lifecycle, if the task It was not completed And at least one cluster exists. Make the cluster Execute the task Time criteria ,but ,otherwise .

[0041] S1014 compares the number of tasks and the number of resources. If the number of tasks is greater than the number of resources, a full-rank sparse matrix is ​​created with the number of tasks as the order; if the number of tasks is less than or equal to the number of resources, a full-rank sparse matrix is ​​created with the number of resources as the order. A random particle mathematical model is established by randomly shuffling the column order of the full-rank sparse matrix. Random particle matrix. . OK column identity matrix The original particle before random transformation is defined as: ,in, The order of the matrix is ​​the larger of the total number of elements in the set of valid clusters and the total number of elements in the set of valid tasks. The formula is: , It is the total number of legitimate clusters. It is the total number of legitimate tasks. It is the operator for calculating the number of elements in a set. This is a maximum value comparison output operator. (Full-rank sparse matrix) It is a random permutation matrix, which is defined by random permutations. Diagonal matrix of order 1 The matrix obtained by taking the column vectors in their positional order is defined as follows: ,in, The order of the matrix is ​​the larger of the total number of elements in the set of valid clusters and the total number of elements in the set of valid tasks. It is a diagonal matrix and its expression is: ,in, It is used for randomly swapping diagonal matrices. Random operators for column vectors The matrix represents A square matrix of order.

[0042] S1015 uses legal tasks as columns and legal simulated entity clusters as rows. It calculates the task cost function for each task pair under each column and row index, filling the results into a zero matrix with the same order as the particles to obtain the task cost matrix. It also records the maximum value task that each legal simulated entity cluster can execute to generate a set of alternative task pairs. (Task Cost Matrix) Defined as:

[0043] wherein, and are the number of legal clusters and the number of legal tasks, respectively; is the number of legal clusters and the number of legal tasks ; is a zero column vector of 1 column; is a valid task cost matrix of rows columns, which is defined as:

[0044] wherein the indices and are integers, is a task value evaluation function, is an assigned task pair that makes cluster perform task ; denotes that the matrix is a matrix of rows columns. The set of spare task pairs records the maximum value task for each legal cluster at the present time, which is defined as: wherein, is an integer, denotes a certain task in the legal task set, denotes a certain cluster in the legal cluster set, is a task value evaluation function, is an assigned task pair that makes cluster perform task ; denotes retrieving the task with the maximum value for cluster from the legal task set and outputting.

[0045] Further, in the step S6, the number of cycles of the particle optimization of the present algorithm is set to a positive integer , and the number of random particles generated each cycle is a positive integer . The global optimal particle of the present algorithm operation is defined as:

[0046] wherein, is a set of positive integers, is the number of cycles of the particle optimization of the present algorithm, is the cycle number, is the local optimal particle of the th cycle optimization,​ is the th local optimal particle of the th iteration represents the matrix multiplication operation on matrix and represents the matrix multiplication operation on matrix and represents the calculation of the sum of all elements in the matrix in the parentheses represents the traversal of all th iteration of the algorithm and obtaining the particle matrix with the maximum value represents the standard deviation operation only for all non-zero elements in the matrix represents the particle matrix with the minimum variance in the th iteration represents the standard deviation operation only for all non-zero elements in the matrix

[0047] S1016 performs a loop iteration, performs random genetic operations on new particles, and performs screening and evaluation on the value and variance of each particle through the task cost matrix. The local optimal particle at different iteration times and the current global optimal particle are recorded until the iteration is completed.

[0048] the th local optimal particle of the th iteration is defined as:

[0049] ,

[0050] ;

[0051] wherein, is a set of positive integers; is the random particle iteration number of the current round of iteration; is a genetic operation that randomly copies part of the column data of matrix to the corresponding column data position of matrix , and the operation result can ensure that each column of the matrix has only one non-zero element with a value of 1; is the task cost matrix; is the th particle in the random particles generated in the th iteration; matrix is the random particle randomly inserted part of the genes from the global optimal particle or the local optimal particle of the last iteration ​The genetic particles of a gene; It is the first Random particles A random permutation matrix; a matrix Genetic particles The permutation matrix, which is used in relation to the genetic particles The same random gene insertion order is replaced by the permutation matrix of the locally optimal particles in the previous cycle. Permutation matrix with the globally optimal particle Perform matrix column data exchange; Represents a matrix and Perform matrix multiplication. Indicates to and Perform matrix dot product calculation; This indicates all generated in this loop. The system iterates through a random number of particles and obtains the particle matrix with the highest value. This indicates that the standard deviation is calculated only for all non-zero elements in the matrix. Specifically, the number of iterations... ,when From time to time and Established.

[0052] S1017 decodes the globally optimal particle to obtain the set of emergency task pairs; for the cluster of simulation entities without assigned tasks, it supplements the allocation scheme in the set of backup task pairs.

[0053] For the optimal particle Decoding the matrix, the task is to map the relation matrix. Defined as:

[0054] ,in, It is by OK The effective task pair mapping matrix, consisting of columns, is defined as follows:

[0055] In the formula, the index and All are integers. Is to make cluster Execute the task The task allocation pairs. For the globally optimal particle. Emergency rescue mission after matrix decoding to set Defined as: , ,in, It is an emergency relief allocation scheme matrix. Emergency rescue allocation scheme matrix the first row and the first column of the task pair , is the allocation task pair for the cluster to execute the task , and and are the global optimal particle matrix and its permutation matrix respectively, is the task pair relationship mapping matrix, represents the inverse matrix operation of the matrix , represents the matrix multiplication operation of the matrix and the matrix , represents the matrix point multiplication operation of the matrix and the matrix .

[0056] S1018 combines the emergency task pair set with the part of the standby task pair set that has not been allocated to obtain a task allocation scheme. At the moment, the final task allocation scheme of the secondary algorithm is obtained, which is defined as the target task pair set , and is specifically represented as: , wherein is the emergency rescue task pair set; is the standby task pair set; is the set addition merge symbol; is a conditional set subtraction operator, represents that the elements in the standby task pair set are excluded, and a new set is obtained by subtracting the task pair elements in the emergency rescue task pair set that have been allocated to the task drone cluster.

[0057] S102 cluster control, control the actions of each simulation entity in the simulation entity cluster and update the single-machine action information of each simulation entity, and feedback the single-machine action information; calculate the overall action control quantity of the simulation entity cluster according to the feedback single-machine action information. All followers in the cluster take the host as the benchmark, and fix the center of the relative coordinate system on the host. All wingmen in the formation take their coordinates in the relative coordinate system as the control benchmark. When each wingman stabilizes near the required position, the required formation shape of the formation is formed. All wingmen only need to interact with the host to accept the control of the host, and do not transfer information among wingmen or even measure the relative position. Compared with the pure centralized control mode, this formation mode belongs to the single-point centralized control mode. For example Figure 10As shown, the steps of the cluster control include: S1021 solving the height difference between each simulated entity and the reference point; S1022 calculating the layered height difference between each simulated entity; S1023 determining whether layered control is needed based on the heading angle deviation and formation control deviation; S1024 controlling the horizontal position of the simulated entity according to the set formation control and the height position of the simulated entity according to the layered height difference when layered control is needed; controlling the horizontal and height positions of the simulated entity according to the set formation control when layered control is not needed; S1025 calculating the formation control command of the aircraft through the PD controller.

[0058] like Figure 11 As shown, the lead aircraft (main unit) uses It is said that the wingman used express, Let be the heading angle and velocity of the lead aircraft and wingman, respectively. Assume the lead aircraft's position in the northeast coordinate system is . The speed is The wingman's position in the northeast coordinate system is: The speed is Relative formation in reference coordinate system The heading angle can be obtained from the speed of the lead aircraft: The relative formation of the lead and wingmen will be transferred from the lead aircraft's heading reference coordinate system to the inertial coordinate system: Then the expected position of the wingman in the inertial frame is: The difference between the wingman's actual position and his desired position is: Switch to the (lead aircraft heading coordinate system): The relative speeds of the lead and wingman aircraft in the northeast coordinate system are: Switching to the lead aircraft's heading coordinate system yields: In the wingman's heading coordinate system, the x-axis component corresponds to the speed command, the y-axis component corresponds to the heading angle command, and the z-axis component corresponds to the altitude command. If the lead aircraft's status is... Based on the PD control strategy, the control commands for the wingman are obtained: .

[0059] When the main engine is turning, the heading angle control command is calculated based on the heading angle commanded by the main engine; when the main engine is not turning, the heading angle control command is calculated based on the actual heading angle of the main engine.

[0060] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for cooperative control of a heterogeneous interconnected multi-core distributed simulation experiment system, deployed in an online task planning simulation system of the heterogeneous interconnected multi-core distributed simulation experiment system or / and an online task planning module of an external task computer, for cooperative control of simulation entities, characterized in that The heterogeneous interconnected multi-core distributed simulation experiment system comprises a semi-physical simulation system and a virtual system; The semi-physical simulation system comprises: a plurality of semi-physical simulation nodes, each of which can simulate a semi-physical simulation entity; a plurality of multi-core parallel simulation nodes, each of which can simulate a plurality of parallel semi-physical simulation entities; The virtual system comprises: a digital simulation system, which can simulate a plurality of parallel digital simulation entities; an experiment environment model, which is used for generating simulation experiment environment modeling data; a virtual-real combined simulation experiment platform, which is used for deducing and rendering a working scene; a dynamic deduction visual simulation engine, which is used for rendering environment information by using the simulation experiment environment modeling data; The semi-physical simulation node comprises: a real-time simulation machine, which is used for constructing a semi-physical simulation entity model; a flight control computer, which is used for controlling the motion of the semi-physical simulation entity model constructed by the real-time simulation machine; an external task computer, which comprises an online task planning module used for simulating the task planning and distribution of a simulation entity cluster to cooperatively control the cluster and an adaptive communication module used for realizing the communication between simulation entities; a self-organizing network data link, which is a physical communication link between semi-physical simulation entities; The multi-core parallel simulation node comprises: a real-time simulation machine, which is used for constructing a parallel semi-physical simulation entity model; a multi-core parallel simulation framework deployed in an RTX real-time system, which is used for the test state configuration, experiment process control, and storage and processing of test data of the parallel semi-physical simulation entities; an online task planning simulation system, which is a digital simulation of the online task planning module, and is used for simulating the task planning and distribution of a simulation entity cluster to cooperatively control the cluster; a self-organizing network data link simulation system, which is a digital simulation of the adaptive communication module and the self-organizing network data link, and is used for realizing the communication between simulation entities; The digital simulation system comprises: a multi-core parallel simulation framework deployed in a Windows system and a digital simulation model based on the multi-core parallel simulation framework, which are used for the test state configuration, experiment process control, and storage and processing of test data of parallel digital simulation entities; an online task planning simulation system, which is a digital simulation of the online task planning module, and is used for simulating the task planning and distribution of a simulation entity cluster to cooperatively control the cluster; a self-organizing network data link simulation system, which is a digital simulation of the adaptive communication module and the self-organizing network data link, and is used for realizing the communication between simulation entities; The steps of the cooperative control method comprise: task distribution, which distributes tasks to different simulation entity clusters, plans the actions of the master of each simulation entity cluster, and calculates the following action instructions of the slaves in the simulation entity cluster; cluster control, which controls the actions of each simulation entity in the simulation entity cluster, updates the single-machine action information of each simulation entity, feeds back the single-machine action information, and calculates the overall action control amount of the simulation entity cluster according to the feedback single-machine action information; The steps of the task distribution comprise: mathematical modeling of the tasks and simulation entity clusters, and definition of the related attributes of the tasks and simulation entity clusters; According to the characteristics of the task and the task and simulation entity cluster related attributes to build a task cost function; Screen out legal tasks and simulation entity clusters, generate a set of legal tasks and a set of legal simulation entity clusters that can be assigned; Compare the number of tasks and resources, if the number of tasks is greater than the number of resources, create a full rank sparse matrix with the number of tasks as the order, if the number of tasks is less than or equal to the number of resources, create a full rank sparse matrix with the number of resources as the order; Build a random particle mathematical model by randomly shuffling the column order of the full rank sparse matrix; Use the task cost function to calculate the task pair under each column index and row index, fill the result into a zero matrix with the same order as the particle to obtain a task cost matrix, and record the maximum value task that each legal simulation entity cluster can execute to generate a backup task pair set; Perform loop iteration, randomly perform genetic operation on new particles, filter and evaluate the value and variance of each particle through the task cost matrix, record the local optimal particle and the current global optimal particle under different iteration numbers until the iteration is completed; Decode the global optimal particle to obtain an emergency task pair set; for the simulation entity cluster that has not been assigned a task, supplement according to the assignment scheme in the backup task pair set; Combine the emergency task pair set with the part of the backup task pair set that has not been assigned to obtain a task assignment scheme; The task cost function is: in, For the value of the task, For the task Execution priority For cluster Not capable of executing the first The ability to perform such tasks It is a task Is there a completion indicator? For the task The remaining valid time, For cluster Execute the task The cost and time spent To enable cluster Execute the task The task allocation pairs; Task The remaining valid time of the task is calculated using the formula: performing the calculation; wherein the remaining effective time for the task , the start execution time for the task , the end time for the task ; the task time checkpoint at time instant ; Cluster Performing tasks Cost time utilization formula: ； performing the calculation; wherein is the cluster performing the task cost time spent, is the task center position, is the cluster number of simulated entities in the cluster, is the cluster average speed, is the cluster center position.

2. The collaborative control method of the heterogeneous interconnected multi-core distributed simulation experiment system according to claim 1, characterized in that: At time t, the formula is used: ； Screening legal simulation entity cluster; wherein, For legal simulation entity cluster set, From the first task to the first Task to determine whether the cluster Has the ability to perform the operation, From the first task to the first Task to determine whether the cluster Can reach a certain task place within a specified time operation, For time criterion, U is a simulation entity cluster set, U i For the ith simulation entity cluster, For the number of simulation entities in the cluster . The formula is used: ； a computation time criterion; wherein is a time criterion, is a start execution time of the task is a center position of the task is a center position of the cluster is a center position of the cluster is a center position of the cluster is an average speed of the cluster is an average speed of the cluster is an average speed of the cluster At time t, the formula is used: ； Screening legal tasks; wherein, for a legal task set, for a task an end time, M is a task set, M j for the jth task, for the operation of traversing from the 1st cluster to the cluster to determine whether the task can be executed by a certain cluster.

3. The cooperative control method for a heterogeneous interconnected multi-core distributed simulation experimental system according to claim 1, characterized in that... The formula is used: ； A set of task pairs; wherein, A set of legal tasks, A set of legal task clusters The jth task in the set, A set of legal task clusters The ith cluster in the set, A task value evaluation function, A task pair that causes the cluster To perform the task With the highest value, To retrieve the task in the set of legal tasks that has the highest value for the cluster And output.

4. The method of claim 1, wherein the method further comprises: The formula is used: ； calculating global optimal particles; wherein, for global optimal particles, is a positive integer set, is the number of cycles of particle optimization, is the cycle number, is the local optimal particle of the th cycle optimization, is the random permutation matrix of the th cycle local optimal particle , and is the matrix multiplication operation on the matrix and , and denotes the standard deviation operation only for all non-zero elements in the matrix. The formula is used: ； ； calculating local optimal particles; wherein, is the first local optimal particle of the first loop optimization, is a positive integer set, is a random particle part of the random insertion is from the global optimal particle gene or the last loop local optimal particle genetic particle of the gene, is a genetic particle is a permutation matrix of the genetic particle, is a task cost matrix, is the number of random particle iterations, h is the random particle iteration number, is a genetic operator, represents the standard deviation operation only for all non-zero elements in the matrix.

5. The method of claim 1, wherein the method further comprises: The formula is used: ， ； Compute the emergency task pairs; wherein, is the set of emergency task pairs, is the number of legal simulation entity clusters, is the number of legal tasks, is the emergency rescue allocation scheme matrix, is the emergency rescue allocation scheme matrix is the task pair in the th row and the th column of the , and are the global optimal particle matrix and its permutation matrix, respectively, is the task pair relationship mapping matrix.

6. The method of claim 1, wherein the method further comprises: In the case of simulation entities being digital simulation unmanned aerial vehicles, the steps of the cluster control include: Solving the height difference of each simulation entity relative to the reference point; Calculating the layered height difference of each simulation entity; Judging whether layered control is needed through the heading angle deviation and formation control deviation; In the case of needing layered control, controlling the horizontal position of the simulation entity according to the set formation control, and controlling the height position of the simulation entity according to the layered height difference; in the case of not needing layered control, controlling the horizontal position and height position of the simulation entity according to the set formation control; Calculating the formation control command of the aircraft; In the case of the host turning, the heading angle control instruction is calculated based on the host instruction heading angle; in the case of the host not turning, the heading angle control instruction is calculated based on the actual heading angle of the host.

Images