Unmanned cluster distributed credible optimization method and system in uncertain scene

Abstract

The invention discloses an unmanned cluster distributed credible optimization method and system in an uncertain scene. The method comprises the steps that each unit of an unmanned cluster is regarded as a node, and a distributed online optimization model adaptive to dynamic change of the number of nodes is constructed with the purpose of minimizing the total cost aiming at node sudden failure and random access uncertainty scenes; converting the consistency constraint in the model into a penalty function, and constructing a penalty optimization model; and aiming at the problem that the communication topology is continuously changed due to the change of the number of the nodes, redesigning the communication weight, proposing a distributed credible optimization algorithm, solving the penalty optimization model, and obtaining the optimization state variable of the active node at each moment. According to the invention, the problem of low decision credibility caused by dynamic change of the number of nodes in an open network is solved, the preset limitation on the change of the number of the nodes is relaxed, and the credible decision of the unmanned cluster system in an uncertain environment is realized.

Description

Technical Field

[0001] This invention relates to a distributed trusted optimization method and system for unmanned clusters in uncertain scenarios, belonging to the field of distributed optimization technology for multi-agent systems. Background Technology

[0002] An unmanned swarm system is a system composed of multiple unmanned devices with autonomous perception, decision-making, and execution capabilities, where each unmanned device is defined as a "node." Through information exchange and collaboration mechanisms among the individuals, unmanned swarm systems can accomplish complex tasks that are difficult for a single unmanned device to complete, and have been widely applied in civilian and military fields. For example, unmanned vehicle swarms complete logistics delivery tasks through collaborative path planning, and unmanned aerial vehicle swarms conduct reconnaissance of designated areas through collaborative combat.

[0003] However, in uncertain environments, unmanned swarm systems typically face complex challenges, including communication link interruptions caused by external environmental interference, functional failures due to node malfunctions, and node deployment changes caused by dynamic task adjustments. These challenges lead to frequent "exit-entry" dynamics among nodes in the swarm system, resulting in performance degradation in unmanned swarm systems and significantly reducing the reliability of collaborative decision-making. Therefore, how to achieve distributed reliable optimization decision-making in uncertain environments has become a research hotspot in fields such as distributed optimization and intelligent systems.

[0004] Distributed trusted optimization methods employ a distributed execution architecture, running independently on each node. By designing specific optimization decision-making mechanisms, they leverage information exchange between nodes and their neighbors to achieve collaborative optimization, ensuring the robustness of decisions and thus guaranteeing the effectiveness and reliability of system decisions in complex dynamic environments. For example, the distributed trusted optimization algorithm [N. Hayashi, Distributed subgradient method in open multiagent systems[J]. IEEE Transactions on Automatic Control, 2023, 68(10).] updates the local estimates of each active node through gradient descent and exchanges estimation information with neighboring active nodes, solving the constrained optimization problem in scenarios with dynamically changing nodes. However, this algorithm has limitations on the "cumulative number of arriving nodes, cumulative number of departing nodes, and cumulative number of active nodes," which deviates from the randomness and unpredictability of dynamic node changes in actual systems. This makes it difficult to meet the requirements in most application scenarios, thus limiting the algorithm's practical application scope to some extent. By storing historical data in multiple sets, the distributed dual average algorithm [Y. Liu, M. Ye, L. Ding and Q. -L. Han, Distributed strategy design for free-in and free-out aggregative games[J]. IEEE Transactions on Automatic Control, 2025, 70(8).] solves the aggregation game problem under dynamic node changes. While this design can meet the needs of dynamic scenarios, it imposes a significant storage burden on each node, causing the algorithm to lose its lightweight characteristics and resulting in poor adaptability in resource-constrained scenarios such as edge computing and low-power devices. Summary of the Invention

[0005] Purpose of the invention: To address the shortcomings of existing technologies, this invention proposes a distributed trusted optimization method and system for unmanned clusters in uncertain scenarios. By designing an optimization algorithm that adapts to scenarios with dynamically changing node numbers, the system can maintain stable computational trust in an open environment where nodes can freely enter and leave, and the limitations on the cumulative number of nodes entering and leaving are reduced.

[0006] Technical solution: To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] A distributed trust optimization method for unmanned clusters in uncertain scenarios includes the following steps:

[0008] S1. Map each unit of the unmanned cluster to a node. For uncertain scenarios such as sudden node failure and random access, introduce a state awareness mechanism to capture the dynamic changes in the node's state in real time. With the goal of minimizing the total cost, construct a distributed online optimization model based on time-varying topology.

[0009] S2. By adopting a penalty factor adjustment strategy, the consistency constraints in the distributed online optimization model are transformed into dynamic penalty functions. Through the real-time linkage between the penalty factor and the node state deviation, a penalty optimization model is constructed. This penalty optimization model is an equivalent solvable form of the distributed online optimization model.

[0010] S3. To address the communication topology change problem caused by the change in the number of nodes, a dynamic communication weight allocation mechanism based on topological correlation is designed. Combining historical interaction data and real-time connection quality optimization weight parameters, a distributed trusted optimization algorithm is proposed to solve the penalty optimization model and obtain the optimized state variables of active nodes at each time. The distributed trusted optimization algorithm first initializes the active node state, initial edge weights, step size, and penalty factor parameters; then it iteratively updates the active node set and communication weights at each time, calculates the state value according to the node type, and outputs the optimized state variables of active nodes at each time after iterating to the maximum number of times.

[0011] Furthermore, the process of constructing the distributed online optimization model includes:

[0012] Define time interval The nodes that participate in the iterative calculation are active nodes. At any given time, the communication graph between active nodes is represented as an undirected connected graph. ,in Let be the set of active nodes at time t. Let be the set of edges connecting active nodes at time t;

[0013] like Then the edge weight This is used to characterize the intensity of information interaction between nodes i and j; otherwise... ;

[0014] Define the set of nodes to leave at time t as The set of nodes that can be reached is The set of retained nodes is Set the upper bound of the number of active nodes at any given time as . ,satisfy , , as well as ;

[0015] remember Let i be the time when node i enters the system. Let i be the time node i leaves the system, satisfying Each node that is entered is considered a new node;

[0016] At each time point t, node It has a private time-varying convex cost function Where R is the set of real numbers, and the state variables of node i satisfy the closed convex set constraint. ;

[0017] The expression for the distributed online optimization model is: The local cost function subgradient Bounded, that is, for , There exists a positive constant M such that .

[0018] Furthermore, the process of constructing the penalty optimization model includes:

[0019] The distributed online optimization model can be equivalently represented as: The optimal solution to this optimization problem at time t is denoted as . ;

[0020] Define penalty function ,in This is a column vector consisting of the state values ​​of the active nodes at time t;

[0021] The expression for the penalty optimization model is: ,in This is a penalty factor.

[0022] Furthermore, the design rules for dynamic communication weights include:

[0023] If node i exits the system and becomes inactive, then the edges associated with node i... Disappearance, corresponding weight ;

[0024] If node i is a newly added active node and is connected to node j Then the weight ;

[0025] Communication weights satisfy any row sum , where N is a positive constant.

[0026] Furthermore, the iterative process of the distributed trustworthy optimization algorithm includes:

[0027] S31: Initialization: Set the total number of active nodes at the initial time. Number of nodes reached Number of departing nodes Number of remaining nodes Within the constraint set, the initial state vectors of all active nodes are randomly set as follows: The weights of the edges connecting active nodes Set as Step size Punishment factor ,in It is a positive number. It represents the minimum non-zero value of the edge weight between active nodes;

[0028] S32: Iterative Update: Loop ,in Represents the maximum time iteration step, updating the set of left nodes. Arrive at the set of nodes and the remaining set of nodes Determine the set of active nodes at the current moment. And update the communication weights between active nodes according to the communication weight design rules. ;

[0029] For each The state value of node i is calculated as follows:

[0030] like ,but and with node Disconnect all related edges and set all edge weights to 0;

[0031] like ,but Randomly set as constraint set Any value in;

[0032] like ,remember for In the set The projection, then The update method is ;

[0033] After the current loop ends, the loop variable t is incremented by 1, and the loop continues until the maximum number of iterations is reached, returning the state variables of the active nodes at each time step. .

[0034] Furthermore, the method also includes: introducing dynamic regret and static regret as core evaluation indicators to quantitatively evaluate the effectiveness of the algorithm in uncertain environments;

[0035] Among them, nodes The static regret is represented as: , It is a node The fixed optimal strategy during its active period;

[0036] Node The dynamic regret is represented as: , It is a node At the current time The dynamic optimal strategy within.

[0037] A distributed, trusted optimization system for unmanned clusters in uncertain scenarios includes:

[0038] The problem construction module is used to map each unit of the unmanned cluster to a node. For uncertain scenarios such as sudden node failure and random access, a state awareness mechanism is introduced to capture the dynamic changes in the node state in real time. With the goal of minimizing the total cost, a distributed online optimization model based on time-varying topology is constructed.

[0039] The penalty model transformation module is used to transform the consistency constraints in the distributed online optimization model into dynamic penalty functions by adopting a penalty factor adjustment strategy. Through the real-time linkage between the penalty factor and the node state deviation, a penalty optimization model is constructed, which is an equivalent solvable form of the distributed online optimization model.

[0040] The algorithm design and solution module addresses the communication topology change problem caused by variations in the number of nodes. It designs a dynamic communication weight allocation mechanism based on topological correlation, combines historical interaction data with real-time connection quality optimization weight parameters, and proposes a distributed trusted optimization algorithm to solve the penalty optimization model, obtaining the optimized state variables of active nodes at each time step. The distributed trusted optimization algorithm first initializes the active node states, initial edge weights, step sizes, and penalty factor parameters; then it iteratively updates the active node set and communication weights at each time step, calculates state values ​​according to node type, and outputs the optimized state variables of active nodes at each time step after iterating to the maximum number of iterations.

[0041] The present invention also provides a computer device, characterized in that it includes: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs, when executed by the processors, implement the steps of the unmanned cluster distributed trusted optimization method under uncertain scenarios as described above.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the unmanned cluster distributed trusted optimization method under uncertain scenarios as described above.

[0043] The present invention also provides a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the unmanned cluster distributed trusted optimization method under uncertain scenarios as described above.

[0044] Beneficial Effects: This invention treats each unit of an unmanned swarm as a node. Addressing uncertain scenarios such as sudden node failures and random access, it constructs a distributed online optimization model adapted to dynamically changing node numbers, aiming to minimize total cost. Subsequently, the consistency constraints in the model are transformed into penalty functions, constructing a penalty optimization model. To address the issue of constantly changing communication topology due to varying node numbers, communication weights are redesigned, and a distributed trusted optimization algorithm is proposed to solve the penalty optimization model, obtaining the optimized state variables of active nodes at each time step. This invention not only solves the technical challenge of low decision-making reliability caused by dynamically changing node numbers in open networks but also relaxes the pre-defined restrictions on node number changes, providing a new and broader solution paradigm for distributed trusted optimization under uncertain environments, and realizing trusted decision-making in unmanned swarm systems under uncertain conditions. Attached Figure Description

[0045] Figure 1 The flowchart below illustrates a distributed trusted optimization method and system for unmanned clusters in uncertain scenarios according to the present invention.

[0046] Figure 2 This is the initial communication topology diagram in this embodiment of the invention;

[0047] Figure 3 This is a graph showing the number of active nodes at different times in an embodiment of the present invention;

[0048] Figure 4 This represents the ratio of the static regret to the number of iterations for nodes 1, 2, and 3 under the algorithm proposed in this embodiment of the invention.

[0049] Figure 5 This represents the ratio of the dynamic regret to the number of iterations for nodes 1, 2, and 3 under the algorithm proposed in this embodiment of the invention.

[0050] Figure 6 This is a state trajectory diagram comparing nodes 1, 2, and 3 with the optimal solution in an embodiment of the present invention. Detailed Implementation

[0051] To provide a clearer understanding of the features and advantages of the technical solution of the present invention, the composition and implementation of the specific solution are described below in conjunction with the accompanying drawings.

[0052] like Figure 1 As shown, this embodiment provides a distributed trust optimization method for unmanned clusters in uncertain scenarios, including the following specific steps:

[0053] S1 treats each unit of the unmanned cluster as a node. For uncertain scenarios such as sudden node failures and random access, a distributed online optimization model is constructed to adapt to the dynamic changes in the number of nodes with the goal of minimizing the total cost.

[0054] This invention uses a drone swarm performing a collaborative reconnaissance mission as a typical example. This swarm possesses dynamic and variable node characteristics; that is, drone nodes within the swarm can passively exit due to uncontrollable factors such as sudden failures, or new nodes can be randomly added according to mission requirements. The swarm will be divided into groups within a time interval... Nodes that participate in the iterative computation are defined as active nodes. At any given time, the communication graph between active nodes is represented as an undirected connected graph. ,in and Represent The set of active nodes and their connected edges. If , remember the edge rights Used to characterize nodes and The intensity of information exchange between them, otherwise .definition The sets of nodes that leave, arrive, and remain at each moment are respectively , , Let the upper bound of the number of active nodes at any given time be given. We can know the cardinality of the node set. , , as well as All of these hold true. Furthermore, to characterize the dynamic properties of individual nodes, let's denote... For nodes The time of entering the system, For the time spent leaving the system, satisfy Meanwhile, this invention treats each newly entered node as a new node.

[0055] When a drone swarm performs a coordinated reconnaissance mission, on the one hand, to achieve covert operations, the swarm needs to maintain a consistent flight altitude to reduce the risk of exposure. On the other hand, each drone aims to minimize reconnaissance risk, and the reconnaissance risk at a given flight altitude can be quantified as a private, time-varying convex function. ,in Represents the set of real numbers. For drone nodes At any moment The state variable is flight altitude. This reconnaissance risk quantification function is generally referred to as the cost function. This is a mathematical description of the quantification of task execution costs, and its form can be flexibly adjusted according to task type, node characteristics, and optimization objectives. For example, in an unmanned vehicle cluster logistics delivery scenario, the cost function can be defined as a comprehensive assessment of the path length and collision risk of the unmanned vehicle node. In this embodiment of the invention, due to the limitations of the UAV's power system performance, its flight altitude has a minimum safety threshold and a maximum ceiling; therefore, the state variables must satisfy the closed convex set constraint. At every moment Each active node selects decision variables that satisfy the constraints based on the currently acquired local information and the information sharing mechanism. The process will be repeated at the next moment, or the system will be switched to an inactive state.

[0056] In the aforementioned uncertain combat scenario where the number of nodes is dynamically variable, the goal of the UAV swarm is to minimize the sum of reconnaissance risks of all currently online nodes while ensuring that all UAVs maintain a consistent flight altitude. To this end, a distributed online optimization model is constructed as follows:

[0057]

[0058] Define local cost function subgradient Bounded, that is, for , There are positive numbers Make .

[0059] S2 transforms the consistency constraints in the model into penalty functions to construct a penalty optimization model.

[0060] The consistency constraints in the above distributed online optimization model are explicitly represented in the model formulation, meaning the original problem can be equivalently expressed as:

[0061]

[0062] The optimization problem is in The optimal solution at time t is denoted as It is easy to see that the optimal solution to this problem is the same as that to the original problem.

[0063] To ensure the stealth of collaborative reconnaissance missions, all UAV nodes must maintain the same flight altitude, which can be modeled as a consistency constraint. The penalty function is defined as follows:

[0064]

[0065] in represent A column vector consisting of the flight altitudes of all active UAV nodes at any given time. If they are connected, then If and only if This is true. Therefore, the penalty optimization model corresponding to the original problem is:

[0066]

[0067] in It is a punishment factor. When When large enough, minimize It will force This means that the states of active nodes tend to be consistent. Therefore, minimizing the penalty function... The goal is to minimize the original objective function while satisfying the consistency constraint; therefore, the optimal solution set of the penalized optimization problem is... ,in The original problem is... The optimal solution at time t. It is a column vector whose elements are all 1s.

[0068] S3 addresses the problem of constantly changing communication topology caused by variations in the number of nodes by redesigning communication weights and proposing a distributed trustworthy optimization algorithm.

[0069] Regarding the current moment Connecting edges Its weight The node state adjustment rules are as follows: If the node... If an edge exits the system and becomes inactive due to malfunction, performance degradation, or other reasons, then the connection... It then disappears, and the corresponding weight is reset to If node For newly added active nodes in the system, and related to nodes Establish edges Then the weight is set to , As an upper bound on the number of nodes, it can be seen that the weight matrix between nodes under this design does not satisfy the constraint that the row sum is 1 under the conventional condition. In fact, the edge weight can be designed as an arbitrary constant according to communication requirements or topological characteristics, only needing to satisfy any row sum. That's all, among which Let be any positive integer.

[0070] Subsequently, to minimize the objective function of the penalty optimization problem, an iterative solution based on gradient descent is employed. Specifically, the local cost function... The corresponding gradient is represented as penalty function The corresponding gradient is calculated as follows: .

[0071] In summary, assuming that the sum of the number of reached nodes increases at a sublinear rate with time, that is... The designed distributed trustworthy optimization algorithm based on gradient descent is shown below:

[0072] S31: Initialization: Set the total number of active nodes at the initial time. Number of nodes reached Number of departing nodes Number of remaining nodes Within the constraint set, the initial state vectors of all active nodes are randomly set as follows: The weights of the edges connecting active nodes Set as Step size Punishment factor ,in It is a positive number. It is the smallest non-zero value of the edge weight between active nodes.

[0073] S32: Update active nodes iteratively Decision variables as follows:

[0074] cycle ,in Represents the maximum number of time iterations;

[0075] Update the set of leaving nodes Arrive at the set of nodes and the remaining set of nodes Determine the set of active nodes at the current moment. And update the communication weights between active nodes according to the above weight adjustment rules. ;

[0076] cycle ;

[0077] Calculate the nodes as follows: State values:

[0078] if ,but and with node Disconnect all related edges and set all edge weights to 0;

[0079] if ,but Randomly set as constraint set Any value in;

[0080] if ,but The update method is

[0081] ,

[0082] in for In the set The projection.

[0083] Loop ends

[0084] set up Continue the loop until the maximum number of iterations is reached;

[0085] Loop ends

[0086] Final result returned: .

[0087] The returned result is the state variables of the active nodes at each time step.

[0088] Compared to existing technologies that require both the sum of arriving nodes to grow at a sublinear rate over time and the sum of departing nodes to grow at a sublinear rate over time, the algorithm proposed in this invention has a wider range of applicability. It only needs to satisfy the former, while allowing the number of departing nodes to vary randomly without additional restrictions. Therefore, the preset conditions required by this invention are more lenient, thus significantly expanding its applicability.

[0089] S4 introduces dynamic regret and static regret as core evaluation indicators to quantitatively evaluate the effectiveness of the algorithm in uncertain environments.

[0090] The core function of static regret is to quantify the difference between the cumulative cost of the actual decision made by the online algorithm at each step based on currently available information and the cumulative cost of the long-term optimal response strategy. Specifically, for the original problem, the node... The static regret is represented as:

[0091]

[0092] in, It is a node The fixed optimal strategy adopted during its active time. Furthermore, dynamic regret quantifies the node. At every moment The loss between the actual cost and the optimal response cost that dynamically adjusts over time. Due to the dynamic nature of the number of system nodes, the dynamic regret value is more effective than the static regret value in evaluating the dynamic performance of the proposed strategy, i.e.:

[0093]

[0094] in It is a node At the current time The best strategy adopted internally is dynamically adjusted over time.

[0095] By calculating static and dynamic regret, it can be observed that, driven by the proposed algorithm, as time approaches infinity, the difference between the total system cost and the total cost corresponding to the optimal value will approach 0, meaning that the node... The state value continuously tracks the optimal solution, ultimately achieving a reliable solution in uncertain scenarios where the number of nodes changes dynamically.

[0096] This invention treats each unit of an unmanned swarm as a node. Addressing uncertain scenarios such as sudden node failures and random access, it constructs a distributed online optimization model adapted to dynamically changing node numbers, aiming to minimize total cost. Subsequently, the consistency constraints in the model are transformed into penalty functions, constructing a penalty optimization model. To address the issue of constantly changing communication topology due to varying node numbers, communication weights are redesigned, and a distributed reliable optimization algorithm is proposed. Finally, dynamic regret and static regret are introduced as core evaluation indicators to quantitatively assess the algorithm's effectiveness in uncertain environments. This invention not only overcomes the technical bottleneck of low decision-making reliability caused by dynamically changing node numbers in open networks but also relaxes the pre-set restrictions on node number changes in existing technologies, providing a new and broader solution paradigm for distributed reliable optimization in uncertain environments, and realizing reliable decision-making in unmanned swarm systems under uncertain conditions.

[0097] To verify the performance of the proposed method, MATLAB 2023b was used as the simulation software in this embodiment. A typical application scenario was a UAV swarm performing a collaborative reconnaissance mission, with an initial swarm configuration of 5 UAVs. Since excessively high flight altitude reduces the clarity of target reconnaissance, while excessively low altitude increases the risk of self-exposure, the reconnaissance risk of each UAV can be modeled as the following quadratic function:

[0098]

[0099] in Representing drones The flight altitude. Considering the real-time impact of dynamic factors such as environmental interference and changes in the number of active nodes on reconnaissance risks during the reconnaissance process, the coefficients of this function... These are set as time-varying coefficients. The initial values ​​of each coefficient are shown in Table 1, and their values ​​during task execution are... and The fluctuation range is , The fluctuation range is Furthermore, due to limitations imposed by the drone's propulsion system, a minimum safety threshold and a maximum service ceiling are set for each drone, meaning the flight altitude must meet local ensemble constraints. The communication topology of the drone swarm is set to a ring shape.

[0100] Table 1. Parameters of the local cost function of each active node at the initial time.

[0101]

[0102] The simulation uses an unmanned cluster distributed trustworthy optimization method and system with parameters set as follows:

[0103] The initial communication weight is set to 1 / 3, and the total time interval is... Step length Punishment factor The upper bound of the number of nodes Furthermore, to more clearly verify the adaptability of the proposed reliable optimization algorithm in dynamic environments, nodes 1, 2, and 3 are set to always remain in the system, while the remaining seven nodes simulate random entry and exit behavior. This not only fully demonstrates the algorithm's ability to track the optimal solution but also intuitively presents the sublinear regret bound characteristics of static and dynamic regret during long-term operation.

[0104] Under the condition that "the total number of arriving nodes increases at a sublinear rate with time", the simulation results of the proposed reliable optimization algorithm are as follows: Figures 3 to 6 As shown. Figure 3 The dynamic changes in the number of nodes in the unmanned cluster are presented. Due to the random arrival and departure behavior of nodes, the total number of nodes exhibits a random fluctuation trend; in Example 1, Figure 4 The ratio curves of static regret to iteration count for nodes 1, 2, and 3 under the proposed algorithm are presented. Figure 5 This is the curve showing the ratio of dynamic regret to the number of iterations for nodes 1, 2, and 3; Figure 6 The diagram displays the state trajectories of nodes 1, 2, and 3, as well as the optimal solution. The dark trajectory marked by circles represents the changes in the time-varying optimal solution, while the light trajectory represents the state changes of nodes 1, 2, and 3. It can be observed that active nodes can effectively track the optimal solution over long-term operation. Furthermore, nodes 1, 2, and 3 gradually converge under the algorithm's guidance. In summary, the distributed reliable optimization method for unmanned clusters in uncertain scenarios proposed in this invention can effectively overcome the challenge of reliable solution development for unmanned clusters where the number of nodes dynamically changes due to node failures.

[0105] Based on the same inventive concept, this invention provides an unmanned cluster distributed trusted optimization system for uncertain scenarios, comprising:

[0106] The problem construction module is used to map each unit of the unmanned cluster to a node. For uncertain scenarios such as sudden node failure and random access, a state awareness mechanism is introduced to capture the dynamic changes in the node state in real time. With the goal of minimizing the total cost, a distributed online optimization model based on time-varying topology is constructed.

[0107] The penalty model transformation module is used to transform the consistency constraints in the distributed online optimization model into dynamic penalty functions by adopting a penalty factor adjustment strategy. Through the real-time linkage between the penalty factor and the node state deviation, a penalty optimization model is constructed, which is an equivalent solvable form of the distributed online optimization model.

[0108] The algorithm design and solution module addresses the communication topology change problem caused by variations in the number of nodes. It designs a dynamic communication weight allocation mechanism based on topological correlation, combines historical interaction data with real-time connection quality optimization weight parameters, and proposes a distributed trusted optimization algorithm to solve the penalty optimization model, obtaining the optimized state variables of active nodes at each time step. The distributed trusted optimization algorithm first initializes the active node states, initial edge weights, step sizes, and penalty factor parameters; then it iteratively updates the active node set and communication weights at each time step, calculates state values ​​according to node type, and outputs the optimized state variables of active nodes at each time step after iterating to the maximum number of iterations.

[0109] Furthermore, the system also includes an algorithm evaluation module, which introduces dynamic regret and static regret as core evaluation indicators to quantitatively evaluate the effectiveness of the algorithm in uncertain environments.

[0110] It should be understood that the unmanned cluster distributed trusted optimization method system under uncertain scenarios in the embodiments of the present invention can implement all the technical solutions in the above method embodiments. The functions of each functional module can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can be referred to the relevant descriptions in the above embodiments, which will not be repeated here.

[0111] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the unmanned cluster distributed trusted optimization method under uncertain scenarios as described above.

[0112] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the unmanned cluster distributed trusted optimization method under uncertain scenarios as described above.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), computer devices, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that each step in the flowchart and combinations thereof can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

Claims

1. A distributed trusted optimization method for unmanned clusters in uncertain scenarios, characterized in that, Includes the following steps: Each unit of the unmanned swarm is mapped to a node. For uncertain scenarios such as sudden node failures and random access, a state awareness mechanism is introduced to capture the dynamic changes in the node's state in real time. With the goal of minimizing the total cost, a distributed online optimization model based on time-varying topology is constructed. A penalty factor adjustment strategy is adopted to transform the consistency constraints in the distributed online optimization model into a dynamic penalty function. By linking the penalty factor with the node state deviation in real time, a penalty optimization model is constructed, which is an equivalent solvable form of the distributed online optimization model. To address the communication topology change problem caused by variations in the number of nodes, a dynamic communication weight allocation mechanism based on topological correlation is designed. Combining historical interaction data with real-time connection quality optimization weight parameters, a distributed trusted optimization algorithm is proposed to solve the penalty optimization model, obtaining the optimized state variables of active nodes at each time step. The distributed trusted optimization algorithm first initializes the active node state, initial edge weights, step size, and penalty factor parameters; then it iteratively updates the active node set and communication weights at each time step, calculates the state value according to node type, and outputs the optimized state variables of active nodes at each time step after iterating to the maximum number of iterations.

2. The method according to claim 1, characterized in that, The process of constructing a distributed online optimization model includes: Define time interval The nodes that participate in the iterative calculation are active nodes. At any given time, the communication graph between active nodes is represented as an undirected connected graph. ,in Let be the set of active nodes at time t. Let be the set of edges connecting active nodes at time t; like Then the edge weight This is used to characterize the intensity of information interaction between nodes i and j; otherwise... ; Define the set of nodes to leave at time t as The set of nodes that can be reached is The set of retained nodes is Set the upper bound of the number of active nodes at any given time as . ,satisfy , , as well as ; remember Let i be the time when node i enters the system. Let i be the time node i leaves the system, satisfying Each node that is entered is considered a new node; At each time point t, node It has a private time-varying convex cost function Where R is the set of real numbers, and the state variables of node i satisfy the closed convex set constraint. ; The expression for the distributed online optimization model is: The local cost function subgradient Bounded, that is, for , There exists a positive constant M such that .

3. The method according to claim 2, characterized in that, The process of constructing the penalty optimization model includes: The distributed online optimization model can be equivalently represented as: The optimal solution to this optimization problem at time t is denoted as . ; Define penalty function ,in This is a column vector consisting of the state values ​​of the active nodes at time t; The expression for the penalty optimization model is: ,in This is a penalty factor.

4. The method according to claim 1, characterized in that, The design rules for dynamic communication weights include: If node i exits the system and becomes inactive, then the edges associated with node i... Disappearance, corresponding weight ; If node i is a newly added active node and is connected to node j Then the weight ; Communication weights satisfy any row sum , where N is a positive constant.

5. The method according to claim 3, characterized in that, The iterative process of the distributed trusted optimization algorithm includes: S31: Initialization: Set the total number of active nodes at the initial time. Number of nodes reached Number of departing nodes Number of remaining nodes Within the constraint set, the initial state vectors of all active nodes are randomly set as follows: The weights of the edges connecting active nodes Set as Step size Punishment factor ,in It is a positive number. It represents the minimum non-zero value of the edge weight between active nodes; S32: Iterative Update: Loop ,in Represents the maximum time iteration step, updating the set of left nodes. Arrive at the set of nodes and the remaining set of nodes Determine the set of active nodes at the current moment. And update the communication weights between active nodes according to the communication weight design rules. ; For each The state value of node i is calculated as follows: like ,but and with node Disconnect all related edges and set all edge weights to 0; like ,but Randomly set as constraint set Any value in; like ,remember for In the set The projection, then The update method is ; After the current loop ends, the loop variable t is incremented by 1, and the loop continues until the maximum number of iterations is reached, returning the state variables of the active nodes at each time step. .

6. The method according to claim 1, characterized in that, The method further includes: introducing dynamic regret and static regret as core evaluation indicators to quantitatively evaluate the effectiveness of the algorithm in uncertain environments; Among them, nodes The static regret is represented as: , It is a node The fixed optimal strategy during its active period; Node The dynamic regret is represented as: , It is a node At the current time The dynamic optimal strategy within.

7. A distributed, trusted optimization system for unmanned clusters in uncertain scenarios, characterized in that, include: The problem construction module is used to map each unit of the unmanned cluster to a node. For uncertain scenarios such as sudden node failure and random access, a state awareness mechanism is introduced to capture the dynamic changes in the node state in real time. With the goal of minimizing the total cost, a distributed online optimization model based on time-varying topology is constructed. The penalty model transformation module is used to transform the consistency constraints in the distributed online optimization model into dynamic penalty functions by adopting a penalty factor adjustment strategy. Through the real-time linkage between the penalty factor and the node state deviation, a penalty optimization model is constructed, which is an equivalent solvable form of the distributed online optimization model. The algorithm design and solution module addresses the communication topology change problem caused by variations in the number of nodes. It designs a dynamic communication weight allocation mechanism based on topological correlation, combines historical interaction data with real-time connection quality optimization weight parameters, and proposes a distributed trusted optimization algorithm to solve the penalty optimization model, obtaining the optimized state variables of active nodes at each time step. The distributed trusted optimization algorithm first initializes the active node states, initial edge weights, step sizes, and penalty factor parameters; then it iteratively updates the active node set and communication weights at each time step, calculates state values ​​according to node type, and outputs the optimized state variables of active nodes at each time step after iterating to the maximum number of iterations.

8. A computer device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the unmanned cluster distributed trusted optimization method in uncertain scenarios as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the distributed trusted optimization method for unmanned clusters in uncertain scenarios as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the distributed trusted optimization method for unmanned clusters in uncertain scenarios as described in any one of claims 1-6.

Images