Low-orbit aircraft remote control instruction generation method and device based on deep learning and time series data modeling, and storage medium
Through a comprehensive solution that combines deep learning and time series data modeling with dynamic scheduling and blockchain verification, the problems of time series data modeling, dynamic task scheduling and multi-objective optimization in the generation of remote control commands for low-orbit aircraft have been solved, achieving efficient and safe remote control command generation and improving the mission execution efficiency and safety of low-orbit aircraft.
Patent Information
- Application Number
- CN202510786889.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-12
- Publication Date
- 2025-09-26
AI Technical Summary
Traditional remote control command generation technology has significant bottlenecks in time series data modeling, dynamic task scheduling, multi-objective optimization and safety assurance, and is difficult to meet the efficiency, accuracy and safety requirements of low-orbit aircraft. In particular, it lacks the ability to capture long-term dependencies in time series data modeling, lacks flexibility in dynamic task scheduling, has weak multi-objective optimization capabilities and an imperfect safety assurance mechanism.
A comprehensive solution combining deep learning and time series data modeling with dynamic scheduling, multi-objective optimization, and blockchain security verification is adopted. Time series data modeling is performed through LSTM and Transformer models, task scheduling optimization is performed by combining particle swarm optimization and genetic algorithm, and blockchain technology is used to ensure the integrity and authenticity of instructions.
It has achieved efficient time series data modeling, dynamic task scheduling, and multi-objective optimization, improved the accuracy and safety of remote control command generation, ensured the integrity and authenticity of the commands, and met the complex mission requirements of low-orbit aircraft.
Smart Images
Figure CN120706764A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of low-orbit aircraft control technology, and in particular to a method, device, and storage medium for generating remote control instructions for a low-orbit aircraft based on deep learning and time series data modeling. Background Art
[0002] With the widespread application of low-orbit spacecraft in communications, remote sensing, meteorological monitoring and other fields, the complexity of their missions and the demand for real-time performance have increased dramatically. However, traditional remote control command generation technology has significant bottlenecks in time series data modeling, dynamic task scheduling, multi-objective optimization and safety assurance, making it difficult to meet the efficiency, accuracy and safety requirements of modern satellite missions. Specifically, 1. The efficiency of time series data modeling is insufficient; the state data generated during the operation of low-orbit aircraft (such as orbital parameters, attitude information, battery power, etc.) has strong time-varying and nonlinear characteristics. Existing technologies mostly rely on traditional time series analysis methods (such as sliding window statistics, autoregressive integrated moving average model (ARIMA)), which can only capture short-term local features and cannot model long-term dependencies. For example, satellite orbit adjustment requires the prediction of multiple future states, but traditional methods ignore the global correlation of time series, resulting in accumulated prediction errors, which in turn affects the accuracy of command generation. In addition, static models are difficult to adapt to changes in the satellite's dynamic environment (such as sudden interference and sensor noise), further exacerbating the problem of disconnection between command generation and real-time status; 2. Dynamic task scheduling lacks flexibility; low-orbit spacecraft missions often involve multiple tasks in parallel, resource competition (such as energy and communication bandwidth), and complex dependencies (such as task A must be executed after task B is completed). Existing scheduling algorithms (such as fixed-priority scheduling and heuristic rules) rely on manually preset logic and are unable to dynamically adjust task priorities and resource allocation based on real-time status. For example, when a sudden high-priority task is inserted, traditional methods are prone to delays in critical tasks or resource conflicts due to scheduling rigidity. In addition, existing technologies do not fully combine task dependency reasoning and hierarchical optimization, making it difficult to maximize the global efficiency of task execution; 3. Weak multi-objective and multi-constraint optimization capabilities; satellite missions need to optimize multiple objectives simultaneously (such as shortest mission execution time, lowest energy consumption, and highest resource utilization) and meet strict constraints (such as battery power threshold and mission deadline). Traditional optimization methods (such as linear programming and single-objective heuristic algorithms) are usually designed only for a single objective and cannot achieve global optimization in complex multi-objective scenarios. For example, single time optimization may lead to energy overruns, while resource optimization may extend the mission cycle. Existing technologies lack a collaborative optimization framework that integrates multiple objectives and multiple constraints, making it difficult for task scheduling and instruction generation to balance efficiency and resource consumption; 4. Incomplete command security mechanisms: Remote control commands are subject to security threats such as tampering and forgery during transmission and execution. Existing security solutions often rely on one-way encryption (such as AES and RSA) or digital signature technologies. While these technologies can encrypt commands, they lack full traceability and tamper-proof verification of command integrity. For example, if a key is leaked or the transmission link is attacked, malicious commands could be injected and untraceable. Furthermore, traditional methods fail to address the issue of non-repudiation after command execution, making them difficult to meet the requirements of high-security scenarios such as military satellites.
[0003] In recent years, deep learning techniques (such as LSTM and Transformer) have demonstrated a powerful ability to capture long-term dependencies and nonlinear relationships in time series modeling. Dynamic scheduling algorithms (such as particle swarm optimization and genetic algorithms) have provided new approaches for multi-objective optimization. Blockchain technology ensures data credibility through decentralized evidence storage. However, existing research still faces several technical limitations: Technical fragmentation: Time series modeling and dynamic scheduling are not deeply integrated, resulting in a disconnect between prediction and scheduling; single optimization dimension: the lack of a collaborative optimization framework that considers task layering, resource allocation, and multi-objective constraints; and isolated security verification: encryption technology and task execution links do not form a closed-loop verification mechanism.
[0004] To address the above problems, there is an urgent need for a comprehensive solution that integrates efficient timing modeling, dynamic intelligent scheduling, multi-objective and multi-constraint collaborative optimization, and blockchain security verification. Summary of the Invention
[0005] The present invention proposes a method, device, and storage medium for generating remote control commands for low-orbit aircraft based on deep learning and time series data modeling, which solves the problems mentioned in the background art. The technical solution of the present invention is achieved as follows: A method for generating remote control instructions for a low-orbit aircraft based on deep learning and time series data modeling, characterized by comprising the following steps: Step S1: Time series data modeling and dynamic scheduling: collect real-time time series data of low-orbit spacecraft, perform time series data modeling through long short-term memory (LSTM) networks, and predict future satellite states; combine dynamic scheduling algorithms to generate task scheduling sequences based on task priority and resource competition analysis; Step S2: Deep learning command generation optimization: Use the Transformer model to train the preprocessed time series data to generate remote control commands that match the satellite status; Step S3: Task dependency reasoning and hierarchical optimization: construct a task graph model, determine the task execution order through topological sorting, and use the particle swarm optimization (PSO) algorithm to optimize the resource allocation of hierarchical tasks; Step S4: Multi-objective and multi-constraint optimization: Use genetic algorithm (GA) to perform multi-objective optimization on task execution time and resource consumption to generate a global optimal scheduling solution; Step S5: Blockchain-based instruction security verification; encrypt the generated remote control instruction and store the instruction hash value through the blockchain to ensure the integrity and authenticity of instruction transmission and execution.
[0006] As a preferred technical solution, the time series data modeling in step S1 specifically includes: The satellite's orbital data, attitude data, battery status, and mission execution status are cleaned, standardized, and normalized. The long-term dependencies of time series data are captured through the input gate, forget gate, and output gate structure of the LSTM unit, and the prediction error is minimized based on the backpropagation algorithm.
[0007] As a preferred technical solution, the dynamic scheduling algorithm in step S1 includes: The real-time priority is calculated based on the task urgency, resource consumption and dependency complexity, and the maximum flow algorithm is used to solve the resource contention problem and dynamically adjust the task execution order; The priority formula is: , in, is the priority of the task, The urgency of the task, For resource consumption, is the complexity of task dependencies, , , is the weight coefficient.
[0008] As a preferred technical solution, in step S2, the Transformer model processes time series data through a self-attention mechanism and performs model training based on a cross-entropy loss function and an Adam optimizer, wherein the training objective of the model is to minimize the following loss function: , in, and are the actual output and the predicted output respectively, and cross_entropy is the cross entropy loss function; During the optimization process, the Adam optimizer is used to adjust the parameters of the model: , Where α is the learning rate, and are momentum and second moment estimates, respectively, is a constant that prevents division by zero.
[0009] As a preferred technical solution, in step S2, after training, the Transformer model can generate corresponding remote control commands based on the input satellite status data. The command generation process can be formalized as follows: =f ( ,θ), in, Represents the remote control command generated in step t, is the input satellite status data, and θ is the trained model parameter.
[0010] As a preferred technical solution, the fitness function of the particle swarm optimization (PSO) algorithm in step S3 is defined as minimizing the task execution cost, including a weighted combination of time and resource consumption, wherein the fitness function of the particle swarm optimization is: , in, is the execution cost of task i, and x is the position of the particle.
[0011] As a preferred technical solution, the fitness function of the genetic algorithm GA in step S4 is the weighted sum of the total execution time and resource consumption, and the scheduling scheme is optimized through selection, crossover and mutation operations, wherein the objective function is: , in, is the total execution time, It is resource consumption, Is the weight coefficient; the fitness function of the genetic algorithm GA is: .
[0012] As a preferred technical solution, the blockchain verification in step S5 specifically includes: Use encryption algorithm to perform hash calculation on remote control command to generate unique hash value
[0013] The hash value is distributed and stored in the blockchain nodes to ensure that the instructions cannot be tampered with; A low-orbit aircraft, comprising a body, a memory, a processor, and a computer program stored in the memory, wherein when the processor executes the program, the steps of the method for generating remote control instructions for the low-orbit aircraft are implemented; A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for generating remote control instructions for a low-orbit aircraft based on deep learning and time series data modeling; Compared with the existing technology, this solution has the following beneficial effects: (1) Efficient time series data modeling: Through the LSTM model and Transformer model, the time-varying characteristics of satellite missions can be effectively captured, improving the accuracy of remote control command generation; (2) Dynamic task scheduling: Combining particle swarm optimization algorithm and priority scheduling, it can adjust the task execution order in real time, optimize resource allocation, and improve the execution efficiency of satellite missions; (3) Multi-objective and multi-constraint optimization: Multi-objective optimization is achieved through genetic algorithms, which can fully consider multiple factors such as task execution time and resource consumption in task scheduling to achieve the global optimal solution; (4) Security assurance: Remote control instructions are encrypted and stored based on blockchain technology to ensure the integrity and authenticity of the instructions. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0015] Figure 1 This is a flow chart of the method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling of the present invention. DETAILED DESCRIPTION
[0016] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.
[0017] Reference Figure 1 This paper proposes a method for generating remote control commands for low-orbit aircraft based on deep learning and time-series data modeling. This method relies on deep learning models, dynamic scheduling, task dependency reasoning, and blockchain verification. This technical solution improves the efficiency, accuracy, and security of remote control command generation. The following details the technical details and implementation of each step.
[0018] 1. Step 1: Time Series Data Modeling and Dynamic Scheduling
[0019] During the execution of a satellite mission, its state changes over time and in response to external circumstances. Therefore, time-series data modeling is crucial. Combined with dynamic scheduling, this allows for real-time optimization of remote control command generation and resource allocation.
[0020] 1.1 Data collection and preprocessing: Time series data is collected mainly from various satellite sensors and mission monitoring systems. Specific data types include but are not limited to: Orbital data: satellite position, velocity, and orbit adjustment requirements; Attitude data: satellite attitude control information, including pitch angle, yaw angle, etc.; Battery status: remaining battery power and charging status; Task execution status: the execution status and progress information of each task; After data collection, preprocessing is performed, including data cleaning, standardization, and normalization. Data cleaning removes missing values and outliers, while standardization converts all data to a uniform format (for example, timestamps to ISO format). Numerical data is normalized and scaled to the [0, 1] range for use in deep learning model training.
[0021] 1.2 Time Series Data Modeling: In order to capture the time-varying characteristics of task execution status, this paper uses a long short-term memory network (LSTM) to model time series data. The specific steps are as follows: Data input: The input data includes the historical status data of the satellite. Assume that the input data is a time series, X= Each x t represents the satellite status data at time step t, LSTM model training: The LSTM network consists of multiple LSTM units. Each LSTM unit contains structures such as input gate, forget gate, and output gate. The gating mechanism is used to capture long-term dependencies in time series. (hidden layer output), (cell status), in, is the output of the LSTM unit, is the activation function, is the cell state, is the activation value of the input gate, is the activation value of the forget gate, is a candidate cell state, Training goal: Minimize the loss function through the back-propagation algorithm to ensure that LSTM can accurately predict the future task state. The loss function is defined as: , in, is the true label, is the output predicted by the LSTM model.
[0022] 1.3 Dynamic Scheduling: Task scheduling needs to consider factors such as task priority and resource consumption. The implementation process of dynamic scheduling is as follows: Real-time priority calculation: Calculate the real-time priority of each task based on the urgency of the task, resource consumption, task completion dependencies, etc. The priority setting formula is: , in, is the priority of the task, The urgency of the task, For resource consumption, is the complexity of task dependencies, , , is the weight coefficient, Task Scheduling Algorithm: A priority-based scheduling algorithm is used to dynamically adjust the order of task execution based on the calculated priority. In situations where there is resource contention, a resource contention resolution algorithm, such as the maximum flow algorithm, is used for scheduling.
[0023] 2. Step 2: Deep Learning Instruction Generation Optimization
[0024] 2.1 Data Preparation: After preprocessing, the task data is further processed into the input format required by the deep learning model. The data is divided into training and test sets, and the model performance is evaluated through cross-validation.
[0025] 2.2 Model training and tuning: To optimize remote control command generation, we used the Transformer model for training to improve its performance on long-sequence tasks. The Transformer model processes input time series data through a self-attention mechanism.
[0026] The training goal of the model is to minimize the following loss function: , in, and are the actual output and the predicted output respectively, and cross_entropy is the cross entropy loss function.
[0027] During the optimization process, the Adam optimizer is used to adjust the parameters of the model: , Where α is the learning rate, and are momentum and second moment estimates, respectively, is a constant that prevents division by zero.
[0028] 2.3 Instruction Generation: After training, the Transformer model can generate corresponding remote control commands based on the input satellite status data. The command generation process can be formalized as follows: =f ( ,θ), in, Represents the remote control command generated in step t, is the input satellite status data, and θ is the trained model parameter.
[0029] 3. Step 3: Task Dependency Reasoning and Hierarchical Optimization
[0030] 3.1 Task dependency analysis: The dependencies between tasks are used to build a task graph model. Each task is a node and the dependencies are directed edges. By analyzing the task graph, the topological sorting algorithm is used to infer the execution order of tasks to ensure acyclic dependencies. The steps of the topological sorting algorithm are as follows: Enqueue all nodes with zero in-degree in the task graph, Process the tasks in the queue in sequence, reduce the in-degree of the successor task of the corresponding task by one, and if the in-degree of the successor task becomes zero, add the task to the queue.
[0031] 3.2 Task layer optimization: According to the topological structure of the task graph, tasks are processed in layers. After layering, tasks in each layer can be executed simultaneously through parallel processing. The resource allocation in task scheduling uses the particle swarm optimization (PSO) algorithm to optimize resource allocation. The fitness function of particle swarm optimization is: , in, is the execution cost of task i (time, resource consumption, etc.), and x is the position of the particle.
[0032] 4. Step 4: Multi-objective and multi-constraint optimization algorithm
[0033] 4.1 Objective function and constraints: In task scheduling, goals include minimizing task execution time, minimizing resource consumption, etc. Constraints include: Resource limitations: such as satellite battery power, bandwidth, etc. Time constraints: For example, a task must be completed before a certain time; The objective function is: , in, is the total execution time, It is resource consumption, is the weight coefficient.
[0034] 4.2 Optimization algorithm: Genetic algorithm (GA) is used for optimization to find the optimal solution through operations such as selection, crossover and mutation. The fitness function is: .
[0035] 5. Step 5: Blockchain-based instruction security verification
[0036] After each remote control command is generated, it will be protected using an encryption algorithm. The encrypted command is stored on the blockchain to ensure that the command cannot be tampered with during transmission and execution. The command hash value stored on the blockchain is as follows: , in, For instructions The hash value of .
[0037] Compared with the existing technology, this solution has the following beneficial effects: Efficient time series data modeling: LSTM and Transformer models can effectively capture the time-varying characteristics of satellite missions and improve the accuracy of remote control command generation; Dynamic task scheduling: Combining particle swarm optimization algorithm and priority scheduling, it can adjust the task execution order in real time, optimize resource allocation, and improve the execution efficiency of satellite missions; Multi-objective and multi-constraint optimization: Multi-objective optimization is achieved through genetic algorithms, which can fully consider multiple factors such as task execution time and resource consumption in task scheduling to achieve the global optimal solution; Security assurance: Remote control instructions are encrypted and stored based on blockchain technology to ensure the integrity and authenticity of the instructions.
[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for generating remote control commands for low-orbit aircraft based on deep learning and time series data modeling, characterized in that: The following steps are involved: Step S1: Time series data modeling and dynamic scheduling; Collect real-time status time series data of low-orbit spacecraft, model the time series data using a long short-term memory (LSTM) network, and predict future satellite status. Combined with a dynamic scheduling algorithm, generate a task scheduling sequence based on task priority and resource competition analysis. Step S2: Deep learning command generation optimization: Use the Transformer model to train the preprocessed time series data to generate remote control commands that match the satellite status; Step S3: Task dependency reasoning and hierarchical optimization; Build a task graph model, determine the task execution order through topological sorting, and use the particle swarm optimization (PSO) algorithm to optimize the resource allocation of hierarchical tasks; Step S4: Multi-objective and multi-constraint optimization: Use genetic algorithm (GA) to perform multi-objective optimization on task execution time and resource consumption to generate a global optimal scheduling solution; Step S5: Blockchain-based instruction security verification; encrypt the generated remote control instruction and store the instruction hash value through the blockchain to ensure the integrity and authenticity of instruction transmission and execution.
2. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: The time series data modeling described in step S1 specifically includes: data cleaning, standardization and normalization of the satellite's orbital data, attitude data, battery status and mission execution status; capturing the long-term dependency of the time series data through the input gate, forget gate and output gate structure of the LSTM unit, and minimizing the prediction error based on the backpropagation algorithm.
3. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: The dynamic scheduling algorithm in step S1 includes: The real-time priority is calculated based on the task urgency, resource consumption and dependency complexity, and the maximum flow algorithm is used to solve the resource contention problem and dynamically adjust the task execution order; The priority formula is: in, is the priority of the task, The urgency of the task, For resource consumption, is the complexity of task dependencies, , , is the weight coefficient.
4. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: In step S2, the Transformer model processes the time series data through the self-attention mechanism and performs model training based on the cross entropy loss function and the Adam optimizer. The training goal of the model is to minimize the following loss function: in, and are the actual output and the predicted output respectively, and cross_entropy is the cross entropy loss function; During the optimization process, the Adam optimizer is used to adjust the parameters of the model: Where α is the learning rate, and are momentum and second moment estimates, respectively, is a constant that prevents division by zero.
5. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: In step S2, after training, the Transformer model can generate corresponding remote control commands based on the input satellite status data. The command generation process can be formalized as follows: ， in, Represents the remote control command generated in step t, is the input satellite status data, and θ is the trained model parameter.
6. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: The fitness function of the particle swarm optimization (PSO) algorithm in step S3 is defined as minimizing the task execution cost, including a weighted combination of time and resource consumption, where the fitness function of the particle swarm optimization is: in, is the execution cost of task i, and x is the position of the particle.
7. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: The fitness function of the genetic algorithm GA in step S4 is the weighted sum of the total execution time and resource consumption, and the scheduling scheme is optimized through selection, crossover and mutation operations, where the objective function is: The fitness function of the genetic algorithm GA is: in, is the total execution time, It is resource consumption, is the weight coefficient.
8. The method for generating remote control instructions for low-orbit aircraft based on deep learning and time series data modeling according to claim 1, characterized in that: The blockchain verification in step S5 specifically includes: Use an encryption algorithm to perform hash calculation on the remote control command to generate a unique hash value: The hash value is distributed and stored in the blockchain nodes to ensure that the instructions cannot be tampered with.
9. A low-orbit aircraft, comprising a body, a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, the steps of a method for generating remote control instructions for a low-orbit aircraft based on deep learning and time series data modeling as described in any one of claims 1 to 8 are implemented.