Beidou satellite positioning platform remote monitoring and scheduling system and method

Abstract

The invention discloses a Beidou satellite positioning platform remote monitoring and scheduling system and method. The method comprises the following steps: S1, dividing a scheduling task of Beidou satellite positioning data into a first layer, a second layer and a third layer; s2, the first layer generates a preliminary scheduling scheme; s3, a second layer utilizes a graph neural network to process dependency and resource conflicts among tasks, and generates an optimized scheduling scheme in combination with a near-end strategy optimization algorithm based on strategy gradient; s4, the third layer adopts a global graph neural network model to process the global dependency relationship of the tasks, and generates an execution instruction of a global scheduling scheme in combination with a soft actor commentator algorithm; s5, modeling task information into a directed graph, propagating messages layer by layer through a graph neural network, and generating an optimal scheduling scheme; s6, setting a feedback mechanism between the layers; and S7, carrying out graphical display on the finally generated optimal scheduling scheme through a visual tool. According to the invention, accurate dynamic scheduling of Beidou satellite data is realized through hierarchical scheduling and graph neural network optimization.

Description

Technical Field

[0001] The present invention relates to the field of satellite positioning and dispatching technology, and in particular to a Beidou satellite positioning platform remote monitoring and dispatching system and method. Background Art

[0002] With the continuous development of satellite positioning technology, the global navigation satellite system has been widely used in all walks of life. As a satellite navigation system independently developed by China, the Beidou satellite positioning system is rapidly becoming an important part of domestic and international positioning, navigation and timing services with its high precision, high reliability and global coverage. The Beidou satellite positioning system plays a key role in military, civil, scientific research and other fields, especially in logistics scheduling, emergency rescue, traffic management and other scenarios that require precise positioning and real-time scheduling. The application prospects of the Beidou system are extremely broad.

[0003] In the prior art, scheduling methods based on satellite positioning systems usually rely on rule-driven or simple priority sorting algorithms. These methods perform well in dealing with simple tasks or scenarios with simple task dependencies, but the limitations of existing methods gradually emerge when faced with large-scale tasks, complex dependencies, and multi-dimensional resource conflicts. Specifically, traditional scheduling methods have difficulty in handling complex dependencies between tasks, especially when resources are limited and multiple tasks compete for the same resources. Traditional methods often cannot achieve optimal resource allocation. In addition, most traditional scheduling systems lack the ability to adjust dynamically. When the execution of a task deviates from expectations, the system cannot respond and adjust in time, resulting in poor execution of the scheduling scheme. Existing scheduling methods are usually static and lack real-time feedback and optimization mechanisms, which makes them often show low robustness and adaptability when facing dynamic environments and emergencies.

[0004] With the rapid development of artificial intelligence and deep learning technologies, more and more studies have begun to try to apply graph neural networks and reinforcement learning to complex task scheduling. As a deep learning model for processing graph structured data, graph neural networks can effectively capture the dependencies and resource competition between tasks, while reinforcement learning learns and optimizes scheduling strategies through continuous interaction with the environment. The introduction of these technologies has brought new solutions to complex task scheduling. However, the existing scheduling systems based on graph neural networks and reinforcement learning still face challenges. First, the existing scheduling algorithms often only apply graph neural networks at a single level, while ignoring the hierarchical processing of task complexity and urgency, which leads to a lack of pertinence and efficiency in processing different types of tasks. Second, the existing systems usually lack an effective feedback mechanism and cannot monitor and adjust the scheduling strategy in real time during task execution, which makes it difficult for the scheduling system to adapt to dynamically changing task requirements and environmental conditions.

[0005] Therefore, how to provide a Beidou satellite positioning platform remote monitoring and dispatching system and method is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0006] One purpose of the present invention is to propose a Beidou satellite positioning platform remote monitoring and scheduling system and method. The present invention realizes the accurate dynamic scheduling of Beidou satellite positioning data in a complex mission environment through a hierarchical scheduling architecture combined with a graph neural network and a reinforcement learning algorithm. Through an effective feedback mechanism and local optimization, the system can adjust the scheduling strategy in a timely manner on the basis of real-time monitoring of the task execution status to ensure the optimal allocation of resources and the efficient execution of tasks. This method not only improves the accuracy and response speed of scheduling, but also significantly enhances the flexibility and robustness of the system in dealing with emergencies and environmental changes, making it perform well in variable and complex application scenarios.

[0007] The Beidou satellite positioning platform remote monitoring and scheduling method according to an embodiment of the present invention comprises the following steps:

[0008] S1. The scheduling tasks of Beidou satellite positioning data are divided into the first layer, the second layer and the third layer according to the complexity and urgency;

[0009] S2, the first layer uses a lightweight rule algorithm to process, generate a preliminary scheduling plan, and pass the preliminary scheduling plan to the second layer;

[0010] S3: The second layer uses graph neural network to process dependencies and resource conflicts between tasks, and optimizes the preliminary scheduling plan of the first layer in combination with the proximal policy optimization algorithm based on policy gradient, generates an optimized scheduling plan, and passes the optimized scheduling plan to the third layer;

[0011] S4, the third layer uses a global graph neural network model to process the global dependencies of the tasks, and combines the soft actor critic algorithm to adjust the optimization scheduling plan of the second layer in real time, monitor the overall scheduling execution, and finally generate execution instructions for the global scheduling plan of Beidou satellite positioning data;

[0012] S5. In the scheduling process of each layer, the scheduling task information of Beidou satellite positioning data is modeled as a directed graph, where nodes represent tasks and edges represent dependencies and resource allocation between tasks. The message is propagated layer by layer through the graph neural network to generate the optimal scheduling plan.

[0013] S6. There is a feedback mechanism between each layer. If resource conflicts or poor scheduling effects are detected in the second or third layer tasks, the feedback information will be passed to the upper layer to readjust the scheduling plan.

[0014] S7. The optimal scheduling plan finally generated is graphically displayed through a visualization tool.

[0015] Optionally, the S2 specifically includes:

[0016] S21, receiving Beidou satellite positioning data, parsing and filtering abnormal data;

[0017] S22. Based on the preset task requirements and resource conditions, the preliminary constraints for scheduling tasks are defined, including the time window for task execution, task priority, resource availability, and interdependencies between tasks. The defined constraints are expressed in matrix form:

[0018] C=[c ij ];

[0019] Among them, c ij Indicates the dependency or conflict relationship between the i-th task and the j-th task. The value range is 0 or 1, 0 means no dependency or conflict, and 1 means there is dependency or conflict;

[0020] S23. Based on the urgency, complexity and resource occupancy of tasks, tasks are classified and sorted. The priority dynamic adjustment formula is expressed as:

[0021]

[0022] Among them, P i represents the priority of the i-th task, U i represents the urgency of the i-th task, W i represents the complexity weight of the i-th task, R i represents the total amount of resources required for the i-th task, δ represents the resource occupancy scheduling factor, and C ik represents the dependency relationship between the i-th task and the k-th task, D k represents the resource occupancy rate of the kth task, and n represents the total number of tasks;

[0023] S24. Design an adaptive scheduling function, combine the dynamic change characteristics of Beidou satellite positioning data, and perform preliminary scheduling of tasks. The optimization objective function is expressed as:

[0024]

[0025] Among them, r ij represents the amount of resources of the jth type allocated to the i-th task, t ij represents the estimated execution time of the i-th task after resources are allocated, α i represents the urgency weight factor of the i-th task, λ represents the weight factor of the task dependency in the scheduling strategy, β represents the weight factor of the environmental factor on the scheduling, f(E j) represents the function value of the impact of the environmental dynamic data provided by the BeiDou satellite on the j-th type of resources;

[0026] S25. By optimizing the objective function, an adaptive adjustment algorithm is used to perform preliminary resource allocation and scheduling, and a preliminary scheduling plan S is generated:

[0027] S=[s1,s2,…,s n ];

[0028] Among them, s i represents the preliminary scheduling result of the ith task, including the allocated resources, the estimated execution time, and the priority adjusted for environmental factors;

[0029] S26, matching and verifying the preliminary scheduling plan with the task constraints, if a conflict or unreasonable situation is detected, readjusting the parameters of the adaptive scheduling function and repeating step S24;

[0030] S27. Pass the verified preliminary scheduling plan to the second layer.

[0031] Optionally, the S3 specifically includes:

[0032] S31, receiving the preliminary scheduling plan generated by the first layer, processing the scheduling tasks using a graph neural network, and modeling the tasks and dependencies as a directed graph G = (N, L), where N represents a set of task nodes and L represents a set of directed edges between tasks;

[0033] S32. In the graph neural network, the message passing mechanism is used to update the status of each task node layer by layer:

[0034]

[0035] in, represents the state vector of node n at the k+1th layer, represents the state vector of node m at layer k, Q (k) represents the weight matrix of the kth layer, M(n) represents the set of neighbor nodes of node n, and e mn represents the weight of the edge between node m and node n, p n represents the degree of node n, p m represents the degree of node m, For normalized message passing, q (k) represents the bias term, φ represents the activation function;

[0036] S33. After the state of the multi-layer graph neural network is updated, the final state representation of each task node is obtained Get the global representation Z of the entire graph task graph G G ;

[0037] S34. Based on the generated global representation Z G , combined with the proximal policy optimization algorithm based on policy gradient to optimize the preliminary scheduling plan, the objective function of the policy update is expressed as:

[0038]

[0039] in, represents the optimized objective function, δ1 represents the cropping coefficient, η represents the exploration weighting factor, represents the advantage function, E t represents the expected value symbol, T represents the total number of time steps, and π ω (b t |x t ) represents the state x under the current policy parameter ω t When selecting action b t The probability distribution of Represents the old strategy ω old The probability distribution of , T represents the total number of time steps;

[0040] S35. Use the optimized scheduling plan as output, generate an optimized scheduling plan, and pass the optimized scheduling plan to the third layer.

[0041] Optionally, the S4 specifically includes:

[0042] S41, receiving the optimized scheduling plan generated by the second layer, using the global graph neural network model to process the global dependencies of the tasks, and modeling all tasks and global dependencies as a directed graph G global =(N global ,L global ), where N global represents the global task node set, L global Represents the set of directed edges between global tasks;

[0043] S42. In the global graph neural network, an extended message passing mechanism is used to obtain the global state representation of each task node through multi-layer state updates;

[0044] S43. After the multi-layer state of the global graph neural network model is updated, the global task graph G is obtained. global The final global representation G final ;

[0045] S44. Final representation G based on the global graph final , combined with the soft actor critic algorithm to adjust the strategy in real time, define the state space Z t is the state of the current global task, action space W t represents the set of executable global scheduling actions, and the reward function Ut Used to evaluate the pros and cons of each step of global scheduling operation;

[0046] S45. In the soft actor-critic algorithm, optimization is performed through the soft Q function and the policy update function:

[0047]

[0048] Among them, J SAC (θ) represents the optimization objective function of the soft actor-critic algorithm, α2 represents the temperature parameter, log2π θ (w t |z t ) indicates the action w is selected under the current strategy t The logarithmic probability, Q θ (z t ,w t ) represents the soft Q function, evaluating t Next execute action w t The value of , β2 represents the learning rate weight factor of the policy gradient, and T represents the total number of time steps;

[0049] S46, generating execution instructions for a global scheduling solution for Beidou satellite positioning data.

[0050] Optionally, the S5 specifically includes:

[0051] S51, preprocessing the scheduling task information in the Beidou satellite positioning data and converting it into an input format suitable for the graph structure;

[0052] S52, assign an initial feature vector to each task node, the feature vector includes the time requirement, resource requirement and urgency information of the task, and uses vector x i =[t i ,r i ,u i ] indicates that t i represents the time requirement of the i-th task, r i represents the resource requirement of the ith task, u i Indicates the urgency of the i-th task;

[0053] S53. According to the dependencies and conflicts between tasks, each edge in the graph is assigned a weight w. ij :

[0054]

[0055] Among them, t j represents the time requirement of the jth task, r j represents the resource requirement of the jth task, u jrepresents the urgency of the jth task, ζ1 represents the coefficient of the influence of the control time difference on the weight, ζ2 represents the coefficient of the control resource difference on the weight, ζ3 represents the coefficient of the control urgency difference on the weight, |V T | represents the size of the task node set;

[0056] S54. Use the graph neural network to perform layer-by-layer message transmission and feature aggregation on the task graph to obtain a new feature representation for each task node.

[0057] S55. After being processed by a multi-layer graph neural network, an optimal scheduling solution is generated.

[0058] Optionally, the S6 specifically includes:

[0059] S61, real-time monitoring of task execution status and resource usage, continuously updating system status information through sensor data, Beidou satellite positioning data and task execution feedback information, and recording the actual start time, completion time, resource consumption and delay of all tasks;

[0060] S62. Calculate the difference between the actual scheduling effect and the expected scheduling effect based on the real-time feedback of the task execution data:

[0061]

[0062] Among them, Δ i represents the difference in scheduling effect of task i, represents the actual completion time of task i, represents the planned completion time of task i, represents the actual resource consumption of task i, represents the planned resource consumption of task i, represents the planning priority of task i, represents the actual priority of task i, ξ1 represents the weight coefficient of time, ξ2 represents the weight coefficient of resources, and ξ3 represents the weight coefficient of priority;

[0063] S63, when a large difference between tasks is detected in the second or third layer, the feedback mechanism is immediately triggered to pass the difference information to the upper layer to re-evaluate the scheduling plan;

[0064] S64. For the tasks triggered by the feedback mechanism, an enhanced local scheduling optimization algorithm is used for adjustment. The optimization objective function is expressed as:

[0065]

[0066] Among them, A represents the set of tasks that need to be adjusted. represents the new difference value of task i after adjustment, B represents the set of tasks affected by the adjustment, represents the adjusted completion time of task i, C represents the set of tasks that have a direct dependency on task A, represents the adjusted dependency strength between task i and task j, represents the actual dependency strength between task i and task j, V ij represents the dependency weight between task i and task j, ρ represents the weight factor for balancing time delay, and τ represents the weight factor for adjusting the impact of dependency adjustment on the overall optimization;

[0067] S65. After the local scheduling optimization, update the adjusted task plan and continue to monitor the task execution;

[0068] S66. If the adjusted solution still does not achieve the expected effect, continue the iterative feedback and adjustment process until the task difference metric Δ i until it falls below a preset threshold.

[0069] The Beidou satellite positioning platform remote monitoring and dispatching system according to an embodiment of the present invention includes the following modules:

[0070] The data receiving and preprocessing module is used to receive and parse Beidou satellite positioning data, filter abnormal data, convert scheduling task information into a graph structure input format, and define preliminary constraints;

[0071] The hierarchical scheduling module is used to divide tasks into three layers according to their complexity and urgency. The first layer generates preliminary scheduling plans, the second layer optimizes task dependencies, and the third layer processes global dependencies and makes real-time adjustments.

[0072] The graph neural network processing module is used to model the scheduling tasks as directed graphs at each level, update the task node status layer by layer through the graph neural network, and generate the optimal scheduling plan;

[0073] Feedback mechanism and local optimization module, used to monitor the task execution status in real time, calculate the scheduling effect difference, trigger the feedback mechanism, and adjust the scheduling plan through the local optimization algorithm;

[0074] The visualization display module is used to display the final generated scheduling plan through visualization tools, providing a graphical representation of task dependencies, resource allocation and scheduling paths.

[0075] The beneficial effects of the present invention are:

[0076] First, through the design of a hierarchical scheduling architecture, the present invention enables the system to hierarchically process scheduling tasks according to the complexity and urgency of the tasks. This design not only improves the efficiency of the system in processing tasks, but also enables the system to adopt the optimal scheduling strategy for different types of tasks. Lightweight rule algorithms are used to process simple tasks, graph neural networks are used to optimize dependencies and resource conflicts of tasks of medium complexity, and global graph neural network models are used to process global dependencies and real-time adjustment strategies. Such hierarchical processing enables the system to effectively reduce the computational burden and improve the overall scheduling efficiency when facing complex tasks.

[0077] Secondly, the present invention realizes dynamic optimization of complex task scheduling strategies by combining graph neural networks and reinforcement learning algorithms, especially the proximal policy optimization algorithm and the soft actor critic algorithm. Graph neural networks perform well in dealing with dependencies and resource conflicts between tasks, and can capture the complex relationships between tasks; while reinforcement learning algorithms continuously adjust and optimize scheduling strategies through continuous interaction with the environment. This combination enables the system to not only generate optimized scheduling solutions when processing complex tasks, but also adjust strategies in real time during task execution, thereby ensuring the high adaptability and robustness of the system in a dynamic environment.

[0078] In addition, the feedback mechanism and local optimization algorithm introduced in the present invention enable the system to monitor the task execution status in real time during the scheduling process, and promptly discover and adjust the deviations in the scheduling. When the system detects that the task execution effect is significantly different from the expected, the feedback mechanism can quickly trigger the adjustment process, and adjust the resource allocation and execution order of the task through the local optimization algorithm to ensure that the scheduling scheme can always maintain the optimal state in a complex and ever-changing environment. This real-time feedback and adjustment mechanism greatly enhances the system's responsiveness and flexibility in dealing with emergencies and environmental changes.

[0079] Finally, the present invention uses visualization tools to graphically display the final generated scheduling plan, so that the scheduling plan's dependencies, resource allocation, and scheduling paths can be intuitively presented to decision makers. Such a visualization not only improves the transparency of the scheduling plan, but also provides decision makers with more comprehensive and clear reference information, helping them make more accurate and efficient decisions in complex situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 A flowchart of the Beidou satellite positioning platform remote monitoring and dispatching system and method proposed by the present invention;

[0082] Figure 2 A schematic diagram of a hierarchical scheduling architecture of a Beidou satellite positioning platform remote monitoring and scheduling system and method proposed in the present invention;

[0083] Figure 3 This is a task scheduling flow chart of the feedback mechanism and local optimization algorithm of the Beidou satellite positioning platform remote monitoring and scheduling system and method proposed by the present invention. DETAILED DESCRIPTION

[0084] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0085] refer to Figure 1-3 ,The Beidou satellite positioning platform remote monitoring and scheduling method includes the following steps:

[0086] S1. The scheduling tasks of Beidou satellite positioning data are divided into the first layer, the second layer and the third layer according to the complexity and urgency;

[0087] S2, the first layer uses a lightweight rule algorithm to process, generate a preliminary scheduling plan, and pass the preliminary scheduling plan to the second layer;

[0088] S3: The second layer uses graph neural network to process dependencies and resource conflicts between tasks, and optimizes the preliminary scheduling plan of the first layer in combination with the proximal policy optimization algorithm based on policy gradient, generates an optimized scheduling plan, and passes the optimized scheduling plan to the third layer;

[0089] S4, the third layer uses a global graph neural network model to process the global dependencies of the tasks, and combines the soft actor critic algorithm to adjust the optimization scheduling plan of the second layer in real time, monitor the overall scheduling execution, and finally generate execution instructions for the global scheduling plan of Beidou satellite positioning data;

[0090] S5. In the scheduling process of each layer, the scheduling task information of Beidou satellite positioning data is modeled as a directed graph, where nodes represent tasks and edges represent dependencies and resource allocation between tasks. The message is propagated layer by layer through the graph neural network to generate the optimal scheduling plan.

[0091] S6. There is a feedback mechanism between each layer. If resource conflicts or poor scheduling effects are detected in the second or third layer tasks, the feedback information will be passed to the upper layer to readjust the scheduling plan.

[0092] S7. The optimal scheduling plan finally generated is graphically displayed through a visualization tool.

[0093] In this implementation, S2 specifically includes:

[0094] S21, receiving Beidou satellite positioning data, parsing and filtering abnormal data;

[0095] S22. Based on the preset task requirements and resource conditions, the preliminary constraints for scheduling tasks are defined, including the time window for task execution, task priority, resource availability, and interdependencies between tasks. The defined constraints are expressed in matrix form:

[0096] C=[c ij ];

[0097] Among them, c ij Indicates the dependency or conflict relationship between the i-th task and the j-th task. The value range is 0 or 1, 0 means no dependency or conflict, and 1 means there is dependency or conflict;

[0098] S23. Based on the urgency, complexity and resource occupancy of tasks, tasks are classified and sorted. The priority dynamic adjustment formula is expressed as:

[0099]

[0100] Among them, P i represents the priority of the i-th task, U i represents the urgency of the i-th task, W i represents the complexity weight of the i-th task, R i represents the total amount of resources required for the i-th task, δ represents the resource occupancy scheduling factor, and C ik represents the dependency relationship between the i-th task and the k-th task, D k represents the resource occupancy rate of the kth task, and n represents the total number of tasks;

[0101] S24. Design an adaptive scheduling function, combine the dynamic change characteristics of Beidou satellite positioning data, and perform preliminary scheduling of tasks. The optimization objective function is expressed as:

[0102]

[0103] Among them, r ij represents the amount of resources of the jth type allocated to the i-th task, t ij represents the estimated execution time of the i-th task after resources are allocated, α i represents the urgency weight factor of the i-th task, λ represents the weight factor of the task dependency in the scheduling strategy, β represents the weight factor of the environmental factor on the scheduling, f(E j ) represents the function value of the impact of the environmental dynamic data provided by the BeiDou satellite on the j-th type of resources;

[0104] S25. By optimizing the objective function, an adaptive adjustment algorithm is used to perform preliminary resource allocation and scheduling, and a preliminary scheduling plan S is generated:

[0105] S=[s1,s2,…,s n ];

[0106] Among them, s i represents the preliminary scheduling result of the ith task, including the allocated resources, the estimated execution time, and the priority adjusted for environmental factors;

[0107] S26, matching and verifying the preliminary scheduling plan with the task constraints, if a conflict or unreasonable situation is detected, readjusting the parameters of the adaptive scheduling function and repeating step S24;

[0108] S27. Pass the verified preliminary scheduling plan to the second layer.

[0109] In this implementation, S3 specifically includes:

[0110] S31, receiving the preliminary scheduling plan generated by the first layer, processing the scheduling tasks using a graph neural network, and modeling the tasks and dependencies as a directed graph G = (N, L), where N represents a set of task nodes and L represents a set of directed edges between tasks;

[0111] S32. In the graph neural network, the message passing mechanism is used to update the status of each task node layer by layer:

[0112]

[0113] in, represents the state vector of node n at the k+1th layer, represents the state vector of node m at layer k, Q (k) represents the weight matrix of the kth layer, M(n) represents the set of neighbor nodes of node n, and e mn represents the weight of the edge between node m and node n, p n represents the degree of node n, p m represents the degree of node m, For normalized message passing, q (k) represents the bias term, φ represents the activation function;

[0114] S33. After the state of the multi-layer graph neural network is updated, the final state representation of each task node is obtained Get the global representation Z of the entire graph task graph G G ;

[0115] S34. Based on the generated global representation Z G, combined with the proximal policy optimization algorithm based on policy gradient to optimize the preliminary scheduling plan, the objective function of the policy update is expressed as:

[0116]

[0117] in, represents the optimized objective function, δ1 represents the cropping coefficient, η represents the exploration weighting factor, represents the advantage function, E t represents the expected value symbol, T represents the total number of time steps, and π ω (b t |x t ) represents the state x under the current policy parameter ω t When selecting action b t The probability distribution of Represents the old strategy ω old The probability distribution of , T represents the total number of time steps;

[0118] S35. Use the optimized scheduling plan as output, generate an optimized scheduling plan, and pass the optimized scheduling plan to the third layer.

[0119] In this implementation manner, the S4 specifically includes:

[0120] S41, receiving the optimized scheduling plan generated by the second layer, using the global graph neural network model to process the global dependencies of the tasks, and modeling all tasks and global dependencies as a directed graph G global =(N global ,L global ), where N global represents the global task node set, L global Represents the set of directed edges between global tasks;

[0121] S42. In the global graph neural network, an extended message passing mechanism is used to obtain the global state representation of each task node through multi-layer state updates;

[0122] S43. After the multi-layer state of the global graph neural network model is updated, the global task graph G is obtained. global The final global representation G final ;

[0123] S44. Final representation G based on the global graph final , combined with the soft actor critic algorithm to adjust the strategy in real time, define the state space Z t is the state of the current global task, action space W t represents the set of executable global scheduling actions, and the reward function U t Used to evaluate the pros and cons of each step of global scheduling operation;

[0124] S45. In the soft actor-critic algorithm, optimization is performed through the soft Q function and the policy update function:

[0125]

[0126] Among them, J SAC (θ) represents the optimization objective function of the soft actor-critic algorithm, α2 represents the temperature parameter, log2π θ (w t |z t ) indicates the action w is selected under the current strategy t The logarithmic probability, Q θ (z t ,w t ) represents the soft Q function, evaluating t Next execute action w t The value of , β2 represents the learning rate weight factor of the policy gradient, and T represents the total number of time steps;

[0127] S46, generating execution instructions for a global scheduling solution for Beidou satellite positioning data.

[0128] In this implementation manner, S5 specifically includes:

[0129] S51, preprocessing the scheduling task information in the Beidou satellite positioning data and converting it into an input format suitable for the graph structure;

[0130] S52, assign an initial feature vector to each task node, the feature vector includes the time requirement, resource requirement and urgency information of the task, and uses vector x i =[t i ,r i ,u i ] indicates that t i represents the time requirement of the i-th task, r i represents the resource requirement of the ith task, u i Indicates the urgency of the i-th task;

[0131] S53. According to the dependencies and conflicts between tasks, each edge in the graph is assigned a weight w. ij :

[0132]

[0133] Among them, t j represents the time requirement of the jth task, r j represents the resource requirement of the jth task, u jrepresents the urgency of the jth task, ζ1 represents the coefficient of the influence of the control time difference on the weight, ζ2 represents the coefficient of the control resource difference on the weight, ζ3 represents the coefficient of the control urgency difference on the weight, |V T | represents the size of the task node set;

[0134] S54. Use the graph neural network to perform layer-by-layer message transmission and feature aggregation on the task graph to obtain a new feature representation for each task node.

[0135] S55. After being processed by a multi-layer graph neural network, an optimal scheduling solution is generated.

[0136] In this implementation manner, S6 specifically includes:

[0137] S61, real-time monitoring of task execution status and resource usage, continuously updating system status information through sensor data, Beidou satellite positioning data and task execution feedback information, and recording the actual start time, completion time, resource consumption and delay of all tasks;

[0138] S62. Calculate the difference between the actual scheduling effect and the expected scheduling effect based on the real-time feedback of the task execution data:

[0139]

[0140] Among them, Δ i represents the difference in scheduling effect of task i, represents the actual completion time of task i, represents the planned completion time of task i, represents the actual resource consumption of task i, represents the planned resource consumption of task i, represents the planning priority of task i, represents the actual priority of task i, ξ1 represents the weight coefficient of time, ξ2 represents the weight coefficient of resources, and ξ3 represents the weight coefficient of priority;

[0141] S63, when a large difference between tasks is detected in the second or third layer, the feedback mechanism is immediately triggered to pass the difference information to the upper layer to re-evaluate the scheduling plan;

[0142] S64. For the tasks triggered by the feedback mechanism, an enhanced local scheduling optimization algorithm is used for adjustment. The optimization objective function is expressed as:

[0143]

[0144] Among them, A represents the set of tasks that need to be adjusted. represents the new difference value of task i after adjustment, B represents the set of tasks affected by the adjustment, represents the adjusted completion time of task i, C represents the set of tasks that have a direct dependency on task A, represents the adjusted dependency strength between task i and task j, represents the actual dependency strength between task i and task j, V ij represents the dependency weight between task i and task j, ρ represents the weight factor for balancing time delay, and τ represents the weight factor for adjusting the impact of dependency adjustment on the overall optimization;

[0145] S65. After the local scheduling optimization, update the adjusted task plan and continue to monitor the task execution;

[0146] S66. If the adjusted solution still does not achieve the expected effect, continue the iterative feedback and adjustment process until the task difference metric Δ i until it falls below a preset threshold.

[0147] Beidou satellite positioning platform remote monitoring and dispatching system, including the following modules:

[0148] The data receiving and preprocessing module is used to receive and parse Beidou satellite positioning data, filter abnormal data, convert scheduling task information into a graph structure input format, and define preliminary constraints;

[0149] The hierarchical scheduling module is used to divide tasks into three layers according to their complexity and urgency. The first layer generates preliminary scheduling plans, the second layer optimizes task dependencies, and the third layer processes global dependencies and makes real-time adjustments.

[0150] The graph neural network processing module is used to model the scheduling tasks as directed graphs at each level, update the task node status layer by layer through the graph neural network, and generate the optimal scheduling plan;

[0151] Feedback mechanism and local optimization module, used to monitor the task execution status in real time, calculate the scheduling effect difference, trigger the feedback mechanism, and adjust the scheduling plan through the local optimization algorithm;

[0152] The visualization display module is used to display the final generated scheduling plan through visualization tools, providing a graphical representation of task dependencies, resource allocation and scheduling paths.

[0153] Embodiment 1:

[0154] In order to verify the feasibility of the present invention in implementation, the present invention is applied in a large logistics enterprise, where the Beidou satellite positioning system is widely used in logistics scheduling and vehicle management across the country. The enterprise handles tens of thousands of transportation tasks every day, involving the real-time scheduling of thousands of logistics vehicles. These tasks are not only distributed over a vast geographical area, but also there are complex dependencies between each task, such as priority, timeliness requirements and resource conflicts. Although the existing scheduling system can complete basic task allocation, with the increase in the amount of tasks and the complexity of the scheduling environment, the system gradually exposes problems such as low scheduling efficiency, unreasonable resource allocation and untimely response, resulting in reduced logistics efficiency and increased operating costs.

[0155] In order to meet these challenges, the company introduced the remote monitoring and dispatching system and method based on the Beidou satellite positioning platform proposed in this invention. The system effectively improves the efficiency and accuracy of logistics dispatching through the combination of hierarchical dispatching architecture, graph neural network, reinforcement learning algorithm, feedback mechanism and visualization tools.

[0156] In actual application, the company starts the dispatching system at 6 am every day according to the logistics needs of all parts of the country. First, the system pre-processes all the received transport task data, and uses the precise positioning information and real-time traffic data provided by Beidou satellites to divide the tasks into three layers according to complexity and urgency for processing. For simple and independent tasks, the system uses lightweight rule algorithms in the first layer to quickly generate preliminary dispatching plans; for tasks involving multi-task dependencies and resource competition, the system uses graph neural networks in the second layer to optimize the dependencies between tasks, and combines the proximal strategy optimization algorithm to adjust resource allocation and dispatching strategies; for the most complex tasks, such as urgent tasks that require global coordination, the system uses a global graph neural network model in the third layer combined with a soft actor critic algorithm for global optimization and real-time adjustment to ensure the rationality and efficiency of the overall dispatch.

[0157] On July 15, 2024, the company was faced with an urgent dispatching task: a batch of urgently needed medical supplies needed to be delivered from City A to City B within 12 hours due to heavy rains that closed many highways, and this batch of supplies needed to be coordinated with logistics tasks in 12 other cities. This task was extremely complex, not only because of the strict timeliness requirements, but also because it required coordinating multiple transportation routes within the disaster area, avoiding resource conflicts, and causing minimal interference with other ongoing tasks.

[0158] The system first performed preliminary resource allocation and time scheduling for all tasks at the first layer, but found that many tasks could not be carried out as planned due to road closures caused by heavy rain, and resource conflicts and task dependencies became more complicated. At this time, the system passed these complex tasks to the second layer, using graph neural networks to remodel all affected tasks and analyze the dependencies between tasks and possible resource conflicts. Through the proximal policy optimization algorithm, the system recalculated the optimal resource allocation plan and generated an optimized scheduling plan.

[0159] However, due to the continuous heavy rain and the changing traffic conditions, the system monitored at the third layer that the main highway in the direction of City B was closed again. The system immediately triggered the feedback mechanism, passed the task status update information back to the second layer, and made real-time adjustments to the route through the local scheduling optimization algorithm. After multiple iterations of optimization, the system finally generated a globally optimized scheduling plan and presented it to the operators at the dispatch center through visualization tools. The plan shows that the system recommends using other secondary roads instead and splitting the transportation task from City A to City B into multiple phased tasks, which are transferred through different routes and cities, thus avoiding major traffic jams. In the end, the batch of medical supplies was successfully delivered to City B at 6 pm, and the entire journey took only 11.5 hours, saving nearly 2 hours compared to traditional scheduling methods.

[0160] The application of this system not only ensures the smooth execution of logistics tasks, but also demonstrates significant beneficial effects at multiple levels. According to statistics, since the system was put into use on July 1, 2024, the overall logistics efficiency of the enterprise has increased by about 23%, the average response time of a single dispatch task has been reduced by 15%, and resource utilization has increased by about 18% in complex task scheduling. At the same time, through the application of feedback mechanisms and local optimization algorithms, the flexibility and emergency response capabilities of the system have been greatly improved, and the dispatch success rate in emergencies has increased from the original 78% to 92%. In addition, the visual display function of the system makes dispatch decisions more transparent and easy to understand, and the workload of dispatchers has been reduced by about 30%.

[0161] Table 1 Comparison of enterprise logistics scheduling performance indicators before and after the introduction of the present invention

[0162]

[0163]

[0164] Table 1 above clearly shows that after the introduction of the Beidou satellite positioning platform remote monitoring and dispatching system of the present invention, the enterprise has significantly improved in multiple key indicators such as logistics dispatching efficiency, task response time, resource utilization and emergency handling capabilities. The overall logistics efficiency has increased by 23%, the average response time of a single dispatching task has decreased by 15%, and the resource utilization rate of complex tasks has increased to 82.6%. In addition, the dispatching success rate of emergencies has increased from 78% to 92%, the total time of transportation tasks and the average delay time of logistics tasks have been shortened, the workload of dispatchers has been significantly reduced, and the transparency and comprehensibility of dispatching decisions have also been significantly improved. These data show that the implementation of the present invention effectively solves the problems of inefficiency and untimely response in the existing dispatching system, and greatly improves the operating efficiency of the enterprise and its ability to cope with complex dispatching scenarios.

[0165] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A Beidou satellite positioning platform remote monitoring and scheduling method, characterized in that: The steps include: S1. The scheduling tasks of Beidou satellite positioning data are divided into the first layer, the second layer and the third layer according to the complexity and urgency; S2, the first layer uses a lightweight rule algorithm to process, generate a preliminary scheduling plan, and pass the preliminary scheduling plan to the second layer; S3: The second layer uses graph neural networks to process dependencies and resource conflicts between tasks, and optimizes the preliminary scheduling plan of the first layer in combination with the proximal policy optimization algorithm based on policy gradient, generates an optimized scheduling plan, and passes the optimized scheduling plan to the third layer; S4, the third layer uses a global graph neural network model to process the global dependencies of the tasks, and combines the soft actor critic algorithm to adjust the optimization scheduling plan of the second layer in real time, monitor the overall scheduling execution, and finally generate execution instructions for the global scheduling plan of Beidou satellite positioning data; S5. In the scheduling process of each layer, the scheduling task information of Beidou satellite positioning data is modeled as a directed graph, where nodes represent tasks and edges represent dependencies and resource allocation between tasks. The message is propagated layer by layer through the graph neural network to generate the optimal scheduling plan. S6. There is a feedback mechanism between each layer. If resource conflicts or poor scheduling effects are detected in the second or third layer tasks, the feedback information will be passed to the upper layer to readjust the scheduling plan. S7. The optimal scheduling plan finally generated is graphically displayed through a visualization tool.

2. The Beidou satellite positioning platform remote monitoring and scheduling method according to claim 1, characterized in that: The S2 specifically includes: S21, receiving Beidou satellite positioning data, parsing and filtering abnormal data; S22. Based on the preset task requirements and resource conditions, the preliminary constraints for scheduling the task are defined, including the time window for task execution, task priority, resource availability, and interdependencies between tasks. The defined constraints are expressed in matrix form: C=[c ij ]; Among them, c ij Indicates the dependency or conflict relationship between the i-th task and the j-th task. The value range is 0 or 1, 0 means no dependency or conflict, and 1 means there is dependency or conflict; S23. Based on the urgency, complexity and resource occupancy of tasks, tasks are classified and sorted. The priority dynamic adjustment formula is expressed as: Among them, P i represents the priority of the i-th task, U i represents the urgency of the i-th task, W i represents the complexity weight of the i-th task, R i represents the total amount of resources required for the i-th task, δ represents the resource occupancy scheduling factor, and C ik represents the dependency relationship between the i-th task and the k-th task, D k represents the resource occupancy rate of the kth task, and n represents the total number of tasks; S24. Design an adaptive scheduling function, combine the dynamic change characteristics of Beidou satellite positioning data, and perform preliminary scheduling of tasks. The optimization objective function is expressed as: Among them, r ij represents the amount of resources of the jth type allocated to the i-th task, t ij represents the estimated execution time of the i-th task after resources are allocated, α i represents the urgency weight factor of the i-th task, λ represents the weight factor of the task dependency in the scheduling strategy, β represents the weight factor of the environmental factor on the scheduling, f(E j ) represents the function value of the impact of the environmental dynamic data provided by the BeiDou satellite on the j-th type of resources; S25. By optimizing the objective function, an adaptive adjustment algorithm is used to perform preliminary resource allocation and scheduling, and a preliminary scheduling plan S is generated: S=[s1,s2,…,s n ]; Among them, s i represents the preliminary scheduling result of the ith task, including the allocated resources, the estimated execution time, and the priority adjusted for environmental factors; S26, matching and verifying the preliminary scheduling plan with the task constraints, if a conflict or unreasonable situation is detected, readjusting the parameters of the adaptive scheduling function and repeating step S24; S27. Pass the verified preliminary scheduling plan to the second layer.

3. The Beidou satellite positioning platform remote monitoring and scheduling method according to claim 1, characterized in that: The S3 specifically includes: S31, receiving the preliminary scheduling plan generated by the first layer, processing the scheduling tasks using a graph neural network, and modeling the tasks and dependencies as a directed graph G = (N, L), where N represents a set of task nodes and L represents a set of directed edges between tasks; S32. In the graph neural network, the message passing mechanism is used to update the status of each task node layer by layer: in, represents the state vector of node n at the k+1th layer, represents the state vector of node m at layer k, Q (k) represents the weight matrix of the kth layer, M(n) represents the set of neighbor nodes of node n, and e mn represents the weight of the edge between node m and node n, p n represents the degree of node n, p m represents the degree of node m, For normalized message passing, q (k) represents the bias term, φ represents the activation function; S33. After the state of the multi-layer graph neural network is updated, the final state representation of each task node is obtained Get the global representation Z of the entire task graph G G ; S34. Based on the generated global representation Z G , combined with the proximal policy optimization algorithm based on policy gradient to optimize the preliminary scheduling plan, the objective function of the policy update is expressed as: in, represents the optimized objective function, δ1 represents the cropping coefficient, η represents the exploration weighting factor, represents the advantage function, E t represents the expected value symbol, T represents the total number of time steps, and π ω (b t |x t ) represents the state x under the current policy parameter ω t Select action b t The probability distribution of Represents the old strategy ω old The probability distribution of , T represents the total number of time steps; S35. Use the optimized scheduling plan as output, generate an optimized scheduling plan, and pass the optimized scheduling plan to the third layer.

4. The Beidou satellite positioning platform remote monitoring and scheduling method according to claim 1, characterized in that: The S4 specifically includes: S41, receiving the optimized scheduling plan generated by the second layer, using the global graph neural network model to process the global dependencies of the tasks, and modeling all tasks and global dependencies as a directed graph G global =(N global ,L global ), where N global represents the global task node set, L global Represents the set of directed edges between global tasks; S42. In the global graph neural network, an extended message passing mechanism is used to obtain the global state representation of each task node through multi-layer state updates; S43. After the multi-layer state of the global graph neural network model is updated, the global task graph G is obtained. global The final global representation G final ; S44. Final representation G based on the global graph final , combined with the soft actor critic algorithm to adjust the strategy in real time, define the state space Z t is the state of the current global task, action space W t represents the set of executable global scheduling actions, and the reward function U t Used to evaluate the pros and cons of each step of global scheduling operation; S45. In the soft actor-critic algorithm, optimization is performed through the soft Q function and the policy update function: Among them, J SAC (θ) represents the optimization objective function of the soft actor-critic algorithm, α2 represents the temperature parameter, log2π θ (w t |z t ) indicates the action w is selected under the current strategy t The logarithmic probability, Q θ (z t ,w t ) represents the soft Q function, evaluating t Next execute action w t The value of , β2 represents the learning rate weight factor of the policy gradient, and T represents the total number of time steps; S46, generating execution instructions for a global scheduling solution for Beidou satellite positioning data.

5. The Beidou satellite positioning platform remote monitoring and scheduling method according to claim 1, characterized in that: The S5 specifically includes: S51, preprocessing the scheduling task information in the Beidou satellite positioning data and converting it into an input format suitable for the graph structure; S52, assign an initial feature vector to each task node, the feature vector includes the time requirement, resource requirement and urgency information of the task, and uses vector x i =[t i ,r i ,u i ] indicates that t i represents the time requirement of the i-th task, r i represents the resource requirement of the i-th task, u i Indicates the urgency of the i-th task; S53. According to the dependencies and conflicts between tasks, each edge in the graph is assigned a weight w. ij : Among them, t j represents the time requirement of the jth task, r j represents the resource requirement of the jth task, u j represents the urgency of the jth task, ζ1 represents the coefficient of the influence of the control time difference on the weight, ζ2 represents the coefficient of the control resource difference on the weight, ζ3 represents the coefficient of the control urgency difference on the weight, |V T | represents the size of the task node set; S54. Use the graph neural network to perform layer-by-layer message transmission and feature aggregation on the task graph to obtain a new feature representation for each task node. S55. After being processed by a multi-layer graph neural network, an optimal scheduling solution is generated.

6. The Beidou satellite positioning platform remote monitoring and scheduling method according to claim 1, characterized in that: The S6 specifically includes: S61, real-time monitoring of task execution status and resource usage, continuously updating system status information through sensor data, Beidou satellite positioning data and task execution feedback information, and recording the actual start time, completion time, resource consumption and delay of all tasks; S62. Calculate the difference between the actual scheduling effect and the expected scheduling effect based on the real-time feedback of the task execution data: Among them, Δ i represents the difference in scheduling effect of task i, represents the actual completion time of task i, represents the planned completion time of task i, represents the actual resource consumption of task i, represents the planned resource consumption of task i, represents the planning priority of task i, represents the actual priority of task i, ξ1 represents the weight coefficient of time, ξ2 represents the weight coefficient of resources, and ξ3 represents the weight coefficient of priority; S63, when a large difference between tasks is detected in the second or third layer, the feedback mechanism is immediately triggered to pass the difference information to the upper layer to re-evaluate the scheduling plan; S64. For the tasks triggered by the feedback mechanism, an enhanced local scheduling optimization algorithm is used for adjustment. The optimization objective function is expressed as: Among them, A represents the set of tasks that need to be adjusted. represents the new difference value of task i after adjustment, B represents the set of tasks affected by the adjustment, represents the adjusted completion time of task i, C represents the set of tasks that have a direct dependency on task A, represents the adjusted dependency strength between task i and task j, represents the actual dependency strength between task i and task j, V ij represents the dependency weight between task i and task j, ρ represents the weight factor for balancing time delay, and τ represents the weight factor for adjusting the impact of dependency adjustment on the overall optimization; S65. After the local scheduling optimization, update the adjusted task plan and continue to monitor the task execution; S66. If the adjusted solution still does not achieve the expected effect, continue the iterative feedback and adjustment process until the task difference metric Δ i until it falls below a preset threshold.

7. Beidou satellite positioning platform remote monitoring and dispatching system, characterized in that: Includes the following modules: The data receiving and preprocessing module is used to receive and parse Beidou satellite positioning data, filter abnormal data, convert scheduling task information into a graph structure input format, and define preliminary constraints; Hierarchical scheduling module, which is used to divide tasks into three levels according to their complexity and urgency; The first layer generates preliminary scheduling plans, the second layer optimizes task dependencies, and the third layer processes global dependencies and makes real-time adjustments; The graph neural network processing module is used to model the scheduling tasks as directed graphs at each level, update the task node status layer by layer through the graph neural network, and generate the optimal scheduling plan; Feedback mechanism and local optimization module, used to monitor the task execution status in real time, calculate the scheduling effect difference, trigger the feedback mechanism, and adjust the scheduling plan through the local optimization algorithm; The visualization display module is used to display the final generated scheduling plan through visualization tools, providing a graphical representation of task dependencies, resource allocation and scheduling paths.

Images