Unmanned aerial vehicle cluster task scheduling management system based on space-air-ground cloud cooperation

Abstract

The invention relates to the field of space-air-ground integrated network communication, in particular to an unmanned aerial vehicle cluster task scheduling management system based on space-air-ground cloud coordination. The system comprises a potential energy mapping modeling module, a signaling interaction module, a gradient circulation scheduling module and a back pressure consistency module. The system combines a network state and task attributes to calculate node dynamic potential energy, and a local potential energy gradient table is constructed; the core of the method is that a transmission path is determined by following a rule of flowing from high potential energy to low potential energy, and a reverse suppression signaling is backtracked and sent to clear an upstream copy when a task is locked; according to the method, the whole network topology maintenance is abandoned, the problem of routing failure caused by high-speed movement is solved by utilizing local interaction, the signaling storm is effectively inhibited, and the survivability of the system is improved.

Description

Technical Field

[0001] This invention relates to the field of integrated air-space-ground network communication and edge computing technology, specifically to a drone swarm task scheduling and management system based on air-space-ground-cloud collaboration. Background Technology

[0002] With the rapid development of integrated air-space-ground network technology, collaborative operations of UAV swarms in heterogeneous network environments are becoming increasingly common; such environments are typically characterized by high-speed node movement, weak connections, and frequent changes in topology.

[0003] Currently, task scheduling generally relies on traditional TCP / IP protocols or routing algorithms that require maintaining the entire network topology. These methods typically require global clock synchronization and maintain routing table consistency through flooding link state announcements. However, in highly dynamic scenarios, traditional methods face severe challenges. Frequent topology reconfiguration not only causes existing routes to fail but also triggers severe signaling storms, resulting in a waste of valuable idle communication bandwidth. In addition, the increased control signaling transmission latency and scheduling deadlock caused by the lag in network state convergence severely restrict the system's survivability and response speed.

[0004] Therefore, how to solve routing failures and resource scheduling delays in highly dynamic heterogeneous networks without maintaining the entire network topology and global clock synchronization has become an urgent problem to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration. Specifically, the technical solution of this invention includes:

[0006] The potential energy mapping modeling module is used to obtain the communication link state parameters and computing resource parameters of heterogeneous network nodes, and combine them with the task attribute parameters of the task to be scheduled to calculate the dynamic potential energy value of the network nodes.

[0007] The signaling interaction module is used to generate a potential energy beacon frame based on the dynamic potential energy value, and to establish and update the local potential energy gradient table in response to the potential energy beacon frame broadcast by the neighboring node.

[0008] The gradient flow scheduling module is used to encapsulate the task to be scheduled into a task data packet carrying initial potential energy, and determine the next hop transmission path of the task data packet according to the local potential energy gradient table and the rule of flowing from high potential energy nodes to low potential energy nodes.

[0009] The backpressure consistency module is used to generate a reverse potential energy suppression signaling in response to the destination node's confirmation of locking the scheduled task, and send it back along the transmission path to trigger the upstream node to clear the scheduled task stored locally.

[0010] Preferably, the potential energy mapping modeling module calculates the dynamic potential energy value of the network node based on the communication link state parameters, the computing resource parameters, and the task attribute parameters, including:

[0011] The task urgency parameter in the task attribute parameters, the remaining computing power of the node in the computing resource parameters, and the link signal-to-noise ratio in the communication link status parameters are used.

[0012] A multidimensional potential energy function model is established, wherein the multidimensional potential energy function model is configured as follows: the product of the task urgency and the first weight coefficient is calculated as the positive component, and the product of the node's remaining computing power and the second weight coefficient and the link signal-to-noise ratio and the third weight coefficient are calculated as the negative component.

[0013] The dynamic potential energy value is calculated by subtracting the negative component from the positive component.

[0014] Preferably, the signaling interaction module generates a potential energy beacon frame based on the dynamic potential energy value, and in response to the received potential energy beacon frame broadcast by a neighboring node, establishes and updates a local potential energy gradient table, including:

[0015] The dynamic potential energy value at the current moment is periodically written into the beacon frame payload and broadcast to neighboring nodes within the single-hop communication range;

[0016] Parse the received potential energy beacon frames broadcast by the neighboring nodes and extract the neighboring potential energy values;

[0017] Calculate the potential energy difference between the local area network node and the neighboring node, and map the potential energy difference into a gradient vector to construct the local potential energy gradient table.

[0018] Preferably, the gradient flow scheduling module determines the next-hop transmission path of the task data packet based on the local potential energy gradient table and according to the rule of flow from high potential energy nodes to low potential energy nodes, including:

[0019] The local potential energy gradient table is retrieved to identify a set of candidate neighbor nodes whose potential energy values ​​are lower than the current node's potential energy value.

[0020] If the candidate neighbor node set is not empty, then the node with the lowest potential energy value is selected from the candidate neighbor node set as the target node of the next hop transmission path, and the task data packet is forwarded to the target node;

[0021] If the set of candidate neighbor nodes is empty, the current node is determined to be in a local potential low point, and the forwarding of the task data packet is suspended.

[0022] Preferred options also include:

[0023] The potential energy accumulation excitation module is configured to linearly increase the accumulated potential energy value inside the task data packet over time in response to the current node being in a local potential energy trough.

[0024] If the accumulated potential energy value is greater than the preset barrier threshold, then the neighbor node with the second lowest potential energy value is forcibly selected as the target node for diffusion transmission.

[0025] If the accumulated potential energy value is less than or equal to the potential barrier threshold, the task data packet will continue to wait at the current node.

[0026] Preferably, the backpressure consistency module, in response to the destination node's confirmation of locking the scheduled task, generates a backpressure suppression signaling and sends it back along the transmission path, including:

[0027] When the task to be scheduled enters the execution state at the destination node, a backpressure data packet carrying a negative potential energy value is generated;

[0028] Based on the routing history in the header of the task data packet, the link connectivity of the previous hop node is checked; if the link connection is normal, the backpressure data packet is sent to the previous hop node; if the link is broken, a clearing instruction carrying the task ID is broadcast within the current single hop range to trigger neighboring nodes to check and clean up any remaining task copies.

[0029] In response to receiving the backpressure data packet, the network node removes the corresponding scheduled task from its local task queue and continues to forward the backpressure data packet to its previous hop node until it traces back to the source node.

[0030] Preferred options also include:

[0031] An adaptive parameter evolution module is used to obtain task execution time and energy consumption data after the scheduled task has been executed.

[0032] Based on the task execution time and energy consumption data, update the historical reputation weight of the network node;

[0033] Based on the updated historical reputation weights, the first weight coefficient, the second weight coefficient, and the third weight coefficient in the potential energy mapping modeling module are corrected.

[0034] Preferably, the network node includes:

[0035] Satellite nodes deployed at high altitudes, drone nodes deployed at low altitudes, control station nodes deployed on the ground, and cloud computing nodes accessed through the control station nodes;

[0036] The communication link status parameters include satellite link bandwidth, air-to-air communication bit error rate, and ground-to-air link delay.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This system constructs a potential energy field model based on local interaction, thus eliminating the limitations of traditional routing protocols that require maintaining the entire network topology and global clock synchronization. By utilizing the signaling interaction module to broadcast potential energy beacons only within a single hop, the complex maintenance of the entire network routing is transformed into lightweight local state updates. This mechanism significantly solves the routing failure problem caused by the high-speed movement of UAVs and frequent topology reconstruction in the background technology, effectively suppresses the signaling storm caused by flooding broadcasts, greatly saves valuable air-to-air communication bandwidth, and improves the system's survivability in weak connection environments.

[0039] 2. This system proposes a multi-dimensional potential energy mapping modeling method, which realizes cross-layer fusion of task urgency at the business layer and computing resources and communication link status at the network layer. By converting task urgency into positive potential energy components and node computing power and link signal-to-noise ratio into negative potential energy components, the scheduling tasks can automatically flow to the low-potential energy nodes with the most abundant computing power and the most stable links, like water flowing, according to the rules of physical field theory. The system can achieve adaptive resource matching and intelligent load balancing in heterogeneous networks without complex cross-layer service quality negotiation.

[0040] 3. To address the problem of traditional gradient descent algorithms easily getting trapped in local optima, this system designs an escape mechanism based on time-varying potential energy. When a task is in a local potential energy trough and cannot flow, the system linearly increases the accumulated potential energy value inside the task packet over time, simulating the tunneling effect in thermodynamics. When the accumulated potential energy exceeds a threshold, a suboptimal node is forcibly selected for diffusion. This mechanism uses time to trade for space, ensuring that the task can break through the potential barrier when the network is congested or the topology is restricted, avoiding the deadlock phenomenon common in distributed scheduling, and guaranteeing the global reachability of the task.

[0041] 4. This system constructs a closed-loop control system through a backpressure consistency module and an adaptive parameter evolution module; it utilizes reverse potential energy suppression signaling to backtrack and clear replicas along the original path after a task is locked for execution, preventing redundant computation and resource waste in a distributed environment; simultaneously, the system can dynamically adjust the weight coefficients in the potential energy model based on the actual time and energy consumption feedback after task execution; this self-learning mechanism based on historical reputation endows the system with evolutionary capabilities, enabling it to more accurately reflect the true carrying characteristics of the physical network as its running time increases, thereby achieving optimization of long-term scheduling strategies. Attached Figure Description

[0042] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0043] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0045] Example 1:

[0046] Please see Figure 1 A drone swarm mission scheduling and management system based on air-space-ground-cloud collaboration includes:

[0047] The potential energy mapping modeling module is used to obtain the communication link state parameters and computing resource parameters of heterogeneous network nodes, and combine them with the task attribute parameters of the task to be scheduled to calculate the dynamic potential energy value of the network nodes.

[0048] The signaling interaction module is used to generate potential energy beacon frames based on dynamic potential energy values, and to establish and update the local potential energy gradient table in response to the potential energy beacon frames broadcast by neighboring nodes.

[0049] The gradient flow scheduling module is used to encapsulate the task to be scheduled into a task data packet carrying initial potential energy, and determine the next hop transmission path of the task data packet according to the local potential energy gradient table and the rule of flowing from high potential energy nodes to low potential energy nodes.

[0050] The backpressure consistency module is used to generate a reverse potential energy suppression signaling in response to the destination node's confirmation of locking the scheduled task, and send it back along the transmission path to trigger the upstream node to clear the scheduled task stored locally.

[0051] This embodiment details the overall architecture and core operating mechanism of a UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration. This system is primarily applied to highly dynamic, weakly connected heterogeneous air-space-ground network environments, aiming to solve routing failures and signaling storms caused by frequent topology changes. The potential energy mapping modeling module, as the system's perception center, acquires the communication link state parameters and computing resource parameters of heterogeneous network nodes in real time. Combined with the task attribute parameters of the task to be scheduled, it uses a preset field theory mapping algorithm to calculate the dynamic potential energy value of each network node at the current moment. This value characterizes the thermodynamic state of the node as a task container; the lower the potential energy, the more suitable the node is as a task executor. The signaling interaction module, as the system's pulse, does not maintain a full network routing table but generates lightweight potential energy beacon frames based on the aforementioned dynamic potential energy values.

[0052] The system responds to the potential energy beacon frames broadcast by neighboring nodes by parsing the potential energy values ​​and establishing and updating a local potential energy gradient table in real time. This table only records the potential energy distribution of neighbors within a single hop, thus reducing the communication overhead of network maintenance from network-wide broadcasting to local interaction. The gradient flow scheduling module, as the system's execution engine, encapsulates the tasks to be scheduled into task data packets carrying initial potential energy. Unlike the traditional TCP / IP protocol that searches for specific IP addresses, this module uses a greedy gradient descent strategy based on the local potential energy gradient table, that is, it determines the next-hop transmission path of the task data packet according to the rule of flowing from high-potential-energy nodes to low-potential-energy nodes. This mechanism makes the task data packets flow naturally to the nodes with the most abundant computing power and the most stable links, just like water. The backpressure consistency module is used to ensure the state consistency of the distributed system. In response to the event that the destination node confirms the locking of the task to be scheduled, it generates a back pressure suppression signaling. This signaling is sent back along the original path of task transmission, triggering the upstream node to clear the locally cached copy of the task.

[0053] This embodiment introduces the concept of discrete spatiotemporal entropy field to construct a self-organizing liquid network that does not require global clock synchronization or maintenance of the entire network topology. When the air-space-ground network is disconnected or congested, tasks can automatically find low-impedance flow paths, significantly reducing the transmission delay of control signaling and solving the scheduling deadlock problem caused by high-frequency topology changes in coordination.

[0054] Example 2:

[0055] The potential energy mapping modeling module calculates the dynamic potential energy values ​​of network nodes based on communication link state parameters, computational resource parameters, and task attribute parameters, including:

[0056] The task urgency parameter in the task attribute parameters, the remaining computing power of the node in the computing resource parameters, and the link signal-to-noise ratio in the communication link status parameters are called;

[0057] A multidimensional potential energy function model is established, configured as follows: the product of task urgency and the first weight coefficient is used as the positive component, and the product of the node's remaining computing power and the second weight coefficient, as well as the product of the link signal-to-noise ratio and the third weight coefficient, are used as the negative components.

[0058] The dynamic potential energy value is calculated by subtracting the negative component from the positive component.

[0059] This embodiment further defines the specific algorithm logic for the potential energy mapping modeling module to calculate the dynamic potential energy value. Addressing the inconsistency in dimensions caused by direct calculation of decibels (dB) and linear physical quantities in the original model, this embodiment introduces a linearization mechanism to ensure the physical rigor of multidimensional heterogeneous resource quantification. The model design follows the thermodynamic principle that demand equates to high potential energy and resources equate to low potential energy. The system calls the task urgency parameter from the task attribute parameters, the remaining computing power of the node from the resource parameters, and the link signal-to-noise ratio from the communication link status parameters. The following multidimensional potential energy function model is established for calculation: ;

[0060] in, Originating from real-time computing, its physical meaning is the index in the network. The dynamic potential energy value of the node is dimensionless; : Derived from task metadata, its physical meaning is task urgency, and its value range is... It should be noted that, in order to adapt to the periodic beacon broadcasting mechanism of Embodiment 3, this system defines a reference urgency level. When nodes perform periodic potential energy broadcasts to maintain network topology, the system will... The value is assigned to a preset reference value. Substituting into the above formula, the fundamental potential energy value can be calculated. Broadcast; when a node processes a specific task to be scheduled, it reads the actual task information. Substitute the values ​​into the formula to calculate the task-specific potential energy value;

[0061] : Derived from the node hardware monitoring driver, its physical meaning is the node's remaining computing power, and the unit is . This translates to millions of floating-point operations per second; it should be noted here that this is to adapt to the weighting coefficients. The normalized order of magnitude, if the original computing power unit returned by the underlying hardware is The system will execute Preprocessing operations are performed to prevent the negative component values ​​from overflowing and causing the model to fail. Derived from physical layer channel estimation, its physical meaning is the linear ratio of the link signal-to-noise ratio, dimensionless, and its calculation formula is: ,in, The original signal-to-noise ratio, in units of ; : Derived from a preset constant, its physical meaning is the first weighting coefficient, used to adjust the positive contribution of task urgency to potential energy; : Derived from a preset constant, its physical meaning is the second weighting coefficient, used to adjust the negative contribution of computing power resources to potential energy; Derived from a preset constant, its physical meaning is the third weighting coefficient, used to adjust the negative contribution of link quality to potential energy; based on this, This constitutes a positive component; the more urgent the task, the higher the node's potential energy, manifesting as a repulsion of other tasks or an urgent need to unleash its own tasks. Simultaneously, This constitutes the negative component. The stronger the node's computing power and the better the communication link (i.e., the higher the linear signal-to-noise ratio), the larger the calculated negative value, leading to a greater total potential energy. The lower the value, the greater the potential energy trough that can attract tasks to flow in.

[0062] This embodiment achieves cross-layer fusion of business layer requirements and network layer status by performing adversarial calculations on positive and negative components and correcting the parameter dimension benchmark. The system does not require a complex QoS negotiation process. It can automatically achieve intelligent load balancing by prioritizing emergency tasks to high-quality nodes simply by adjusting the rise and fall of potential energy values. It is particularly suitable for resource-constrained UAV edge computing scenarios.

[0063] To verify the technical effectiveness and disclose the basis for parameter settings, this embodiment supplements the following experimental data and parameter ranges: The range of values ​​for the preset constants is set as follows: To ensure that urgent tasks are given high priority, Using normalized computing power values, To adapt to the numerical magnitude of the linearized signal-to-noise ratio; in a specific operational instance, to ensure a balanced contribution from each component, set... When the task is to be scheduled A certain network node The unit is Original signal-to-noise ratio The unit is Corresponding linear value When the value is obtained, the dynamic potential energy is calculated by substituting it into the formula: ;

[0064] The value shows that the computational power term and the signal-to-noise ratio term contribute equally, both at 1.5, and the calculated significant negative potential energy value indicates that the node exerts a strong attraction on the task, verifying the effectiveness and balance of the model under heterogeneous parameters.

[0065] Based on this, this embodiment further discloses the logic for setting the range of weight coefficient values: for the first weight coefficient Its value This is to adhere to the principle of urgency-driven decision-making, ensuring that the positive potential energy generated by high-urgency tasks is sufficient to counteract the attraction from low-computing-power nodes; for the second weighting coefficient... Its value This is to follow the principle of magnitude normalization, reducing the computational power of massive numbers. , usually Magnitude, mapped to The interval should be set to prevent it from overwhelming other parameters; for the third weighting coefficient... Its value For compatibility The linear amplification characteristic prevents the link quality factor from becoming too heavy, which could cause the computing power index to fail.

[0066] Example 3:

[0067] The signaling interaction module generates potential energy beacon frames based on dynamic potential energy values, and in response to the potential energy beacon frames broadcast by neighboring nodes, establishes and updates the local potential energy gradient table, including:

[0068] The dynamic potential energy value at the current moment is periodically written into the beacon frame payload and broadcast to neighboring nodes within the single-hop communication range;

[0069] Parse the received potential energy beacon frames broadcast by neighboring nodes and extract the neighboring potential energy values;

[0070] Calculate the potential energy difference between local network nodes and their neighboring nodes, and map the potential energy difference to a gradient vector to construct a local potential energy gradient table.

[0071] This embodiment further defines the specific communication mechanism for the signaling interaction module to establish a local potential energy gradient table; in order to achieve low-overhead state synchronization in a weak network environment, this embodiment adopts a periodic beacon broadcast mechanism; each network node periodically reads the dynamic potential energy value calculated at the current moment. The beacon frame is written into a specific field of the beacon frame payload. The node broadcasts this beacon frame to all neighboring nodes within its single-hop communication range via an omnidirectional antenna. This beacon frame is a short frame structure and does not contain complex routing information; it only contains the source node ID and potential energy value. During this period, the node activates listening mode, parses the received potential energy beacon frames broadcast by neighboring nodes, and extracts the neighboring potential energy values. ;

[0072] The system will measure the potential energy value of the local area network nodes. Potential energy value of neighbors Potential energy difference between : ; in response This indicates that the local potential energy is higher than that of its neighbors, suggesting a tendency for flow to that neighbor; at this point, the system performs a vectorized mapping: The magnitude of the gradient vector is defined as the value that points to neighboring nodes. The logical link is defined as the direction of the gradient vector; the system maps the potential energy difference to a gradient vector and stores it in the local potential energy gradient table. The data structure of this table is neighbor ID, potential energy value and gradient magnitude.

[0073] This embodiment abandons the traditional routing protocol's practice of flooding link state announcements and only maintains the potential state of single-hop neighbors. This enables the system to achieve millisecond-level state convergence when facing frequent network topology fragmentation caused by the high-speed movement of UAVs, greatly saving valuable air-to-air communication bandwidth and ensuring survivability in tactical edge networks.

[0074] Example 4:

[0075] The gradient flow scheduling module determines the next-hop transmission path of the task data packet based on the local potential energy gradient table and the rule of flow from high potential energy nodes to low potential energy nodes, including:

[0076] Search the local potential energy gradient table to identify the set of candidate neighbor nodes whose potential energy value is lower than that of the current node;

[0077] If the candidate neighbor node set is not empty, then select the node with the lowest potential value from the candidate neighbor node set as the target node of the next hop transmission path, and forward the task data packet to the target node;

[0078] If the set of candidate neighbor nodes is empty, the current node is determined to be in a local potential low point, and the forwarding of task data packets is suspended.

[0079] This embodiment further defines the specific decision-making logic of the gradient flow scheduling module in determining the next-hop transmission path; the module executes a gradient-based greedy forwarding strategy; when a node holds a task to be scheduled, it searches the local potential energy gradient table and identifies all neighboring nodes whose potential energy values ​​are lower than the current node's potential energy value, thus forming a candidate neighbor node set. The system executes the optimal forwarding logic, responding to... Not empty, traversing the set Select the node with the lowest potential energy as the target node. : Then, the task data packet is forwarded to the target node. This simulates the physical process of water flowing downhill; in response to If the value is empty, it means that the potential energy of all surrounding neighbors is higher than or equal to that of the current node. The system determines that the current node is in a local potential energy trough. This means that the task cannot find a better executor through simple gradient descent for the time being. The system suspends forwarding the task data packet and places it in the local cache queue to wait.

[0080] The greedy strategy in this embodiment ensures that tasks are always transmitted towards areas with better resources and links, avoiding resource waste caused by blind broadcasting. At the same time, the identification of local low points provides the triggering conditions for the subsequent potential energy activation mechanism, forming a complete adaptive flow closed loop.

[0081] Example 5:

[0082] Also includes:

[0083] The potential energy accumulation excitation module is configured to linearly increase the accumulated potential energy value inside the task data packet over time in response to the current node being in a local potential energy trough.

[0084] If the accumulated potential energy value is greater than the preset barrier threshold, the neighbor node with the second lowest potential energy value will be forcibly selected as the target node for diffusion transmission.

[0085] If the accumulated potential energy value is less than or equal to the barrier threshold, the task data packet will continue to wait at the current node.

[0086] This embodiment further defines a potential energy accumulation and excitation module for scenarios with local potential energy lows. To prevent tasks from remaining stuck in local optimal nodes and failing to reach the global optimal node, i.e., to solve the trap problem of gradient algorithms, this embodiment introduces a time-varying potential energy excitation mechanism. In response to the current node being in a local potential energy low, the system starts a timer to accumulate potential energy. As time progresses, the accumulated potential energy value inside the task data packet increases linearly. The calculation formula is as follows: ;

[0087] in, : Derived from real-time calculation, subscript Represents accumulation, with the physical meaning of time. The cumulative potential energy value, to distinguish it from the nodal potential energy. , here adopt Indicates that it is dimensionless; : Derived from the mission header, its physical meaning is the initial potential energy when the mission arrives, and it is dimensionless; : Derived from a preset positive number, its physical meaning is the potential energy growth coefficient, which is dimensionless; : Derived from system parameters, its physical meaning is the time normalization constant, and its unit is . , used to eliminate the dimensional effects of the time term; Derived from the system clock, its physical meaning is the current time, and the unit is... ; Derived from the system clock, its physical meaning is the time when a task arrives at the current node, and the unit is... ;

[0088] During this period, the system monitors in real time. Execute barrier breakdown judgment; respond to ,in If the system determines that the task is overheated based on a preset barrier threshold, it forcibly breaks the gradient descent rule and executes a suboptimal escape strategy. Specifically, the system checks the number of elements in the current candidate neighbor node set. ;like If the next lowest potential energy value is selected as the target node for diffusion and transmission, then the neighboring node with the second lowest potential energy value is chosen. If so, then select the unique neighbor node as the target node; if If no path is found, the system determines there is no way forward and continues to wait; this logic is similar to the tunneling effect in thermodynamics, but with added constraints on topological boundary conditions; conversely, in response to... If so, the task data packet will continue to wait in the current node, waiting for new neighbors brought about by network topology changes;

[0089] This embodiment uses time to trade off space, ensuring that even if a task falls into a local dead end, it can forcibly break through the barrier and find a new transmission path by increasing internal energy over time; thus, it probabilistically guarantees the global reachability of task scheduling and effectively avoids deadlock in distributed systems.

[0090] This embodiment further illustrates the logic for setting the threshold and coefficient: The setting is based on the network's maximum tolerable latency boundary, and the value range is [value range missing]. ; The setting is based on the task urgency decay rate, and the value range is [value range missing]. ; The setting is based on the system heartbeat cycle, and is usually set to a value of In the experimental test, the following settings were made: Assume the task remains stuck at a local low point for a certain period of time. initial potential energy The cumulative potential energy value is: because This triggers the barrier breakdown mechanism, forcibly spreading the task to the suboptimal node. Assuming there are multiple neighbors, experimental results show that this mechanism reduces the deadlock probability by more than 90%.

[0091] Example 6:

[0092] The backpressure consistency module responds to the destination node's confirmation of locking the scheduled task, generates a backpressure suppression signaling, and sends it back along the transmission path, including:

[0093] When a task to be scheduled enters the execution state at the destination node, a backpressure data packet carrying a negative potential energy value is generated.

[0094] Based on the routing history in the task data packet header, check the link connectivity of the previous hop node; if the link connection is normal, send the backpressure data packet to the previous hop node; if the link is broken, broadcast a clear command carrying the task ID within the current single hop range to trigger neighboring nodes to check and clean up any remaining task copies.

[0095] In response to receiving a backpressure data packet, the network node removes the corresponding scheduled task from its local task queue and continues to forward the backpressure data packet to its previous hop node until it traces back to the source node.

[0096] This embodiment further defines the backtracking and cleanup mechanism of the backpressure consistency module. In a multi-path transmission network, to prevent the same task from being executed repeatedly by multiple nodes, this embodiment designs a reverse potential energy suppression signaling. When the task to be scheduled is successfully locked and enters the execution state at the destination node, i.e., the point of lowest potential energy, the destination node immediately generates a backpressure data packet. This data packet carries a special negative potential energy value, which is a logical identifier rather than a physical negative value, used to neutralize the task potential energy in the network. Based on the routing history recorded in the packet header during the transmission of the task data packet, i.e., the path stack, the destination node sends the backpressure data packet to the previous hop node. Then, in response to receiving the backpressure data packet, the network node checks the local task queue. If a corresponding copy of the task to be scheduled exists, it is removed to free up storage space. The node continues to forward the backpressure data packet to its previous hop node until it backtracks to the source node that originally generated the task.

[0097] This embodiment implements a distributed task atomicity locking; the back pressure signal returns along the original path like a shock wave, quickly calming the potential energy disturbance in the network caused by the task, effectively preventing redundant calculations and channel congestion, and ensuring the consistency of the task status across the entire network. It is particularly suitable for high-concurrency task scheduling scenarios.

[0098] Example 7:

[0099] Also includes:

[0100] The adaptive parameter evolution module is used to obtain task execution time and energy consumption data after the scheduled task has been executed.

[0101] Update the historical reputation weights of network nodes based on task execution time and energy consumption data;

[0102] Based on the updated historical reputation weights, the first weight coefficient, second weight coefficient, and third weight coefficient in the potential energy mapping modeling module are corrected.

[0103] This embodiment further defines the adaptive parameter evolution module; in order to enable the system to adapt to different mission scenarios, such as reconnaissance missions that emphasize latency and strike missions that emphasize reliability, this embodiment introduces a weight adaptive correction mechanism based on historical reputation.

[0104] After the scheduled task is completed, the system performs closed-loop data acquisition and records the actual task execution time. Energy consumption data Based on the deviation between the measured data and the expected target, the historical reputation weights of the network nodes are updated. In response to a node's actual performance exceeding expectations, the system increases its reputation weight; based on the updated... The first weighting coefficient in the dynamic correction potential energy mapping modeling module Second weighting coefficient and the third weighting coefficient For example, if a high signal-to-noise ratio link is found not to deliver the expected low latency, it may be due to bandwidth limitations, and the latency will be reduced in the next round of calculation. The value of is used to reduce the weight of link quality on potential energy calculation;

[0105] This embodiment endows the system with evolutionary capabilities; through continuous feedback learning, the potential energy model can increasingly accurately reflect the true carrying capacity of the physical network, thereby achieving long-term optimization of the scheduling strategy and solving the problem that static models are difficult to cope with dynamic environmental changes.

[0106] This embodiment specifically discloses the algorithm details of the adaptive parameter evolution module; before executing the following update formula, the system estimates the time consumption. Compared with the estimated energy consumption Perform a numerical stability check to ensure the denominator is not zero: ;

[0107] in, This is the machine's minimum value; in this embodiment, it is taken as... Historical reputation weight The update follows the following exponential smoothing model: ;

[0108] in, Iteration rounds; Initial reputation weight, set during system initialization. ; Learning rate, range of values The value used in this embodiment ; : Estimated time and energy consumption; To address the issues of missing dimensional definitions and deviations in calculation principles in existing technologies, this embodiment discloses in detail its physical definition and generation logic: The adaptive parameter evolution module constructs and maintains a historical task feature library in local memory. This library uses a hash mapping data structure, where the key is the task type identifier. The value is a historical statistical vector: ;

[0109] in, The historical average time is represented here, using symbols. To distinguish it from the time normalization constant in Example 5 , Historical average energy consumption Counting historical samples; before calculation begins, the module uses the current number of tasks to be scheduled. For index retrieval If the search is successful, that is... Then directly read historical data and assign values: If the search does not find a match, that is... The new task type is based on the task data packet size. The unit is MB, and the task computation density. The unit is MFLOPS / MB, derived from task attributes, and linear estimation is performed. ; ;

[0110] The coefficients are defined as follows to ensure dimensional balance: : Time slope coefficient, physically meaning the time consumed per unit of computing power, measured in seconds per MFLOPS, is set to... ; The time-based constant, physically representing startup overhead, is set to [value in seconds]. ; Energy consumption slope coefficient, physically meaning energy consumption per unit of computing power, unit: J / MFLOPS, set as follows: ; The fundamental constant of energy consumption, physically meaning static energy consumption, is measured in joules (J). ; All units are in seconds (s). All units are in J; this formula introduces the calculation of density. This corrects the fundamental flaw of relying solely on packet size to estimate execution time; furthermore, after task execution, the library update rules follow an incremental averaging formula: ; ;

[0111] in, This is the updated average value, to achieve adaptive data accumulation; Actual measured time and energy consumption; based on the updated... The algorithm for correcting the weight coefficients follows the principle of same-direction suppression and reverse-direction decay to prevent logical reversal: for the positive component weights This involves exclusion: ;

[0112] For negative component weights and Since this involves attraction, to prevent the resource weight from excessively decaying to zero due to extremely high node reputation (i.e., parameter collapse), this embodiment introduces a weight lower bound protection mechanism, and the modified formula is as follows: ; ;

[0113] in, and The preset minimum weight threshold is set as follows in this embodiment: and This ensures that even under high-reputation conditions, computing power and link quality parameters retain minimal scheduling influence. To adjust the step size, the range of values ​​is... The value used in this embodiment The hyperbolic tangent function is introduced here. This is to utilize its saturation property within its domain to mitigate potentially divergent reputation values. Mapping to interval Internally, this algorithm prevents system oscillations caused by excessive adjustments to weight coefficients due to fluctuations in the historical data of a single node; it ensures that the higher the node's reputation, the more the system increases its weighting coefficients. and reduce This approach moderately suppresses the task absorption capacity of the node, i.e., increases its potential energy, thereby preventing high-quality nodes from collapsing due to overload and achieving adaptive optimization of system parameters.

[0114] Example 8:

[0115] Network nodes include:

[0116] Satellite nodes deployed at high altitudes, drone nodes deployed at low altitudes, control station nodes deployed on the ground, and cloud computing nodes accessed through the control station nodes;

[0117] Communication link status parameters include satellite link bandwidth, air-to-air communication bit error rate, and ground-to-air link delay.

[0118] This embodiment specifies the network node and communication link state parameters in a concrete scenario. In this system, physical entities are given clear field theory roles. Satellite nodes deployed at high altitudes act as high-potential macroscopic schedulers, characterized by wide coverage but high latency. Their satellite link bandwidth parameters are assigned a low potential energy reduction weight, meaning they are less attractive to tasks unless the task is extremely large. UAV nodes deployed at low altitudes act as the main executors and relays of tasks, and their air-to-air communication bit error rate is a key parameter affecting potential energy. Furthermore, control station nodes deployed on the ground typically act as potential energy black holes with unlimited computing power, i.e., absolute low potential energy points, used to handle ultra-large-scale tasks. Ground-to-air link latency is a key parameter in potential energy calculation; the greater the latency, the higher the potential energy, which manifests as task repulsion. This embodiment clarifies the roles of different physical entities in the potential energy field, ensuring the accurate correspondence between the mathematical model and physical devices. This makes the system engineering feasible for implementation in a real integrated air-space-ground network, effectively addressing the heterogeneity challenges in cross-domain collaboration.

[0119] Specifically, for the aforementioned ground control station nodes with unlimited computing power, when performing the formula calculation in Example 2, the system will use the node's remaining computing power parameters. The value is assigned to the preset saturation threshold. ,For example This approach avoids floating-point overflow in computers while ensuring that the negative component generated by this node in the mathematical model is preserved. The value is much larger than other nodes, thus correctly constructing the physical properties of a potential energy black hole numerically; similarly, for cloud computing nodes accessed through the control station, the system calculates their computational resource parameters. Set to the same saturation threshold as the ground station. However, when calculating the potential energy, a high-weight penalty coefficient is assigned to the corresponding air-to-ground link delay parameter to accurately reflect the characteristics of high latency and high computing power coexisting in the cloud offloading process.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration, characterized in that, include: The potential energy mapping modeling module is used to obtain the communication link state parameters and computing resource parameters of heterogeneous network nodes, and combine them with the task attribute parameters of the task to be scheduled to calculate the dynamic potential energy value of the network nodes. The signaling interaction module is used to generate a potential energy beacon frame based on the dynamic potential energy value, and to establish and update the local potential energy gradient table in response to the potential energy beacon frame broadcast by the neighboring node. The gradient flow scheduling module is used to encapsulate the task to be scheduled into a task data packet carrying initial potential energy, and determine the next hop transmission path of the task data packet according to the local potential energy gradient table and the rule of flowing from high potential energy nodes to low potential energy nodes. The backpressure consistency module is used to generate a reverse potential energy suppression signaling in response to the destination node's confirmation of locking the scheduled task, and send it back along the transmission path to trigger the upstream node to clear the scheduled task stored locally.

2. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration as described in claim 1, characterized in that, The potential energy mapping modeling module calculates the dynamic potential energy value of the network node based on the communication link state parameters, the computing resource parameters, and the task attribute parameters, including: The task urgency parameter in the task attribute parameters, the remaining computing power of the node in the computing resource parameters, and the link signal-to-noise ratio in the communication link status parameters are used. A multidimensional potential energy function model is established, wherein the multidimensional potential energy function model is configured as follows: the product of the task urgency and the first weight coefficient is calculated as the positive component, and the product of the node's remaining computing power and the second weight coefficient and the link signal-to-noise ratio and the third weight coefficient are calculated as the negative component. The dynamic potential energy value is calculated by subtracting the negative component from the positive component.

3. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration as described in claim 1, characterized in that, The signaling interaction module generates a potential energy beacon frame based on the dynamic potential energy value, and in response to the received potential energy beacon frame broadcast by a neighboring node, establishes and updates a local potential energy gradient table, including: The dynamic potential energy value at the current moment is periodically written into the beacon frame payload and broadcast to neighboring nodes within the single-hop communication range; Parse the received potential energy beacon frames broadcast by the neighboring nodes and extract the neighboring potential energy values; Calculate the potential energy difference between the local area network node and the neighboring node, and map the potential energy difference into a gradient vector to construct the local potential energy gradient table.

4. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration according to claim 1, characterized in that, The gradient flow scheduling module determines the next-hop transmission path of the task data packet based on the local potential energy gradient table and according to the rule of flow from high potential energy nodes to low potential energy nodes, including: The local potential energy gradient table is retrieved to identify a set of candidate neighbor nodes whose potential energy values ​​are lower than the current node's potential energy value. If the candidate neighbor node set is not empty, then the node with the lowest potential energy value is selected from the candidate neighbor node set as the target node of the next hop transmission path, and the task data packet is forwarded to the target node; If the set of candidate neighbor nodes is empty, the current node is determined to be in a local potential low point, and the forwarding of the task data packet is suspended.

5. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration according to claim 4, characterized in that, Also includes: The potential energy accumulation excitation module is configured to linearly increase the accumulated potential energy value inside the task data packet over time in response to the current node being in a local potential energy trough. If the accumulated potential energy value is greater than the preset barrier threshold, then the neighbor node with the second lowest potential energy value is forcibly selected as the target node for diffusion transmission. If the accumulated potential energy value is less than or equal to the potential barrier threshold, the task data packet will continue to wait at the current node.

6. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration according to claim 1, characterized in that, The backpressure consistency module, in response to the destination node's confirmation of locking the scheduled task, generates a backpressure suppression signaling message and sends it back along the transmission path, including: When the task to be scheduled enters the execution state at the destination node, a backpressure data packet carrying a negative potential energy value is generated; Based on the routing history in the header of the task data packet, the link connectivity of the previous hop node is checked; if the link connection is normal, the backpressure data packet is sent to the previous hop node; if the link is broken, a clearing instruction carrying the task ID is broadcast within the current single hop range to trigger neighboring nodes to check and clean up any remaining task copies. In response to receiving the backpressure data packet, the network node removes the corresponding scheduled task from its local task queue and continues to forward the backpressure data packet to its previous hop node until it traces back to the source node.

7. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration according to claim 1, characterized in that, Also includes: An adaptive parameter evolution module is used to obtain task execution time and energy consumption data after the scheduled task has been executed. Based on the task execution time and energy consumption data, update the historical reputation weight of the network node; Based on the updated historical reputation weights, the first weight coefficient, the second weight coefficient, and the third weight coefficient in the potential energy mapping modeling module are corrected.

8. The UAV swarm task scheduling and management system based on air-space-ground-cloud collaboration according to claim 1, characterized in that, The network nodes include: Satellite nodes deployed at high altitudes, drone nodes deployed at low altitudes, control station nodes deployed on the ground, and cloud computing nodes accessed through the control station nodes; The communication link status parameters include satellite link bandwidth, air-to-air communication bit error rate, and ground-to-air link delay.

Images