Bidirectional approval method and system based on joint operation of UAV and one-button sequential control

Abstract

The present invention discloses a two-way approval method and system based on the joint operation of unmanned aerial vehicles and one-key sequential control, including the following steps: system modeling and data acquisition: constructing the dynamic model of the unmanned aerial vehicle and the logical model of the one-key sequential control system, and collecting the data of the unmanned aerial vehicle and the one-key sequential control system, including data preprocessing; distributed data processing and analysis: designing a distributed data processing framework, and performing feature extraction and data analysis on the preprocessed data in step S1; designing a two-way approval mechanism: based on the data analysis results in step S2, designing the unmanned aerial vehicle approval algorithm and the one-key sequential control system approval algorithm; formulating an adaptive learning and decision-making mechanism: designing a unmanned aerial vehicle scheduling strategy based on reinforcement learning, using a dual deep Q network algorithm, and using experience playback and target network technology to improve the learning efficiency and stability of the one-key sequential control system. This scheme can effectively improve the efficiency, accuracy and safety of the joint operation of unmanned aerial vehicles and one-key sequential control systems.

Description

Technical Field

[0001] The present invention relates to the field of power data processing, and in particular to a two-way verification method and system based on the joint operation of an unmanned aerial vehicle and one-button sequential control. Background Art

[0002] With the rapid development of drone technology and the continuous improvement of industrial automation, drones and one-button sequential control systems are increasingly used in multiple fields, especially in collaborative operations in complex environments, such as agriculture, logistics, emergency rescue, power inspection, etc. However, the joint operation of drones and one-button sequential control systems faces many challenges, including the real-time nature of data transmission, the accuracy of operating instructions, and the security and stability of the system.

[0003] As a flexible and efficient aerial work platform, drones have been widely used in many fields. Drones have the following advantages: Flexibility: drones can fly freely in complex environments and perform a variety of tasks; Efficiency: drones can quickly cover large areas and improve work efficiency; Safety: drones can replace humans to perform dangerous tasks and reduce the risk of casualties.

[0004] However, the application of drones also faces some challenges: Environmental perception: drones need to perceive the surrounding environment in real time to avoid collisions and obstacles; Mission planning: drones need to plan paths and schedule tasks according to mission requirements; Communication reliability: The communication between drones and ground control systems needs to ensure real-time and reliability.

[0005] The one-button sequential control system is an automated control system that achieves automated control of complex equipment through simple operating instructions. The one-button sequential control system has the following advantages: Easy operation: Users only need to press a button to complete complex operating procedures; High efficiency: The system can respond quickly to operating instructions and improve work efficiency; Safety: The system has automatic detection and protection functions to ensure the safety of operation.

[0006] However, the application of one-button sequential control system also faces some challenges: instruction accuracy: the system needs to ensure the accurate transmission and execution of operation instructions; system stability: the system needs to have high stability to avoid failures caused by operational errors; data processing capabilities: the system needs to process a large amount of real-time data to ensure the real-time and accuracy of operations.

[0007] In practical applications, the joint operation of drones and one-button sequential control systems faces the following challenges:

[0008] Real-time data transmission: Data transmission between the UAV and the one-button sequential control system needs to ensure real-time performance to ensure timely delivery and execution of operating instructions; Accuracy of operating instructions: The UAV needs to accurately execute the operating instructions generated by the one-button sequential control system to avoid operating errors; System security and stability: The joint operation system needs to have high security and stability to avoid operation interruptions or accidents caused by system failures; Environmental adaptability: The UAV needs to adapt to complex and changing environments, and the one-button sequential control system needs to dynamically adjust operating strategies according to environmental changes.

[0009] The existing joint operation technology of UAV and one-button sequential control system has the following shortcomings:

[0010] Insufficient data processing capabilities: When processing large amounts of real-time data, existing systems are prone to data delays and processing bottlenecks; Low accuracy of operating instructions: When generating and executing operating instructions, existing systems are prone to errors, affecting operational efficiency and safety; Insufficient system stability: Existing systems are prone to failures in complex environments, affecting operational continuity and safety; Poor environmental adaptability: When faced with complex and changing environments, existing systems find it difficult to dynamically adjust operational strategies, affecting operational results. Summary of the invention

[0011] The purpose of the invention is to provide a two-way verification method and system based on the joint operation of UAV and one-key sequential control to solve the above-mentioned problems existing in the prior art.

[0012] The technical solution is based on a two-way approval method for joint operation of drones and one-button sequential control, including the following steps:

[0013] Step S1, system modeling and data collection: constructing a dynamic model of the UAV and a logical model of the one-button sequential control system, and collecting data of the UAV and the one-button sequential control system, including data preprocessing;

[0014] Step S2, distributed data processing and analysis: design a distributed data processing framework, and perform feature extraction and data analysis on the data preprocessed in step S1;

[0015] Step S3: Design a two-way approval mechanism: Based on the data analysis results in step S2, design a drone approval algorithm and a one-button sequential control system approval algorithm;

[0016] Step S4, formulate adaptive learning and decision-making mechanism: design a UAV scheduling strategy based on reinforcement learning, adopt a dual deep Q network algorithm, and use experience replay and target network technology to improve the learning efficiency and stability of the one-button sequential control system.

[0017] According to another aspect of the present application, the step S1 is further:

[0018] Step S11: Establishing an integrated model of the drone and the one-button sequential control system, including:

[0019] UAV system modeling: Establish the UAV's dynamic model, including modules such as flight control, navigation, and sensor data acquisition; design the communication interface between the UAV and the one-button sequential control system to ensure the real-time and reliability of data transmission;

[0020] One-button sequential control system modeling: Establish a logical model of the one-button sequential control system, including modules such as operation command generation, execution, and feedback; design the interface between the one-button sequential control system and the UAV system to ensure the accurate transmission and execution of operation commands;

[0021] Step S12, data collection and preprocessing, includes:

[0022] UAV data collection: Use the sensors carried by the drone to collect on-site environmental data; and pre-process the collected data, including denoising, correction, and splicing operations to generate a high-precision three-dimensional environmental model;

[0023] One-key sequential control system data collection: collect the operation log and equipment status data of the one-key sequential control system; and pre-process the collected data to ensure the integrity and consistency of the data.

[0024] According to another aspect of the present application, step S2 is further:

[0025] Step S21: designing a distributed data processing framework, including:

[0026] Step S211, data sharding and distributed storage: sharding the collected data according to geographic location and time dimension, and storing it in a distributed database;

[0027] Step S212: Design data redundancy and backup mechanisms to ensure data security and reliability;

[0028] Step S213: Using the MapReduce distributed computing framework as the system framework to process the preprocessed data;

[0029] Step S214: design a scheduling strategy for data processing tasks to optimize the utilization of computing resources;

[0030] Step S22, data analysis and feature extraction, includes:

[0031] Step S221, environmental data analysis: analyzing the environmental data collected by the drone to extract key features, including terrain, obstacles, and lighting;

[0032] Step S222: using the SVM machine learning algorithm to classify and predict environmental features;

[0033] Step S223, operation data analysis: analyzing the operation data of the one-key sequential control system and extracting key features, including: operation frequency and device status change;

[0034] Step S224: Use time series analysis methods to predict and detect anomalies in the operation data.

[0035] According to another aspect of the present application, step S3 is further:

[0036] Step S31, the two-way approval process between the drone and the one-key sequential control system includes:

[0037] Step S311: The drone generates operation suggestions based on the collected environmental data and sends them to the one-key sequential control system;

[0038] Step S312: the one-key sequential control system verifies the operation suggestion to ensure the safety and feasibility of the operation;

[0039] Step S313: the one-key sequential control system generates an operation instruction according to the operation data and sends it to the drone;

[0040] Step S314: The drone verifies the operation instruction to ensure the accuracy and real-time performance of the instruction;

[0041] Step S32: designing an approval algorithm, including:

[0042] Step S321: Design an approval algorithm based on environmental characteristics to determine the feasibility of the operation suggestion;

[0043] Step S322: Optimize the approval strategy using a reinforcement learning algorithm to improve the accuracy and efficiency of approval;

[0044] Step S323: designing an approval algorithm based on the operation data to determine the rationality of the operation instruction;

[0045] Step S324: Use a probabilistic graphical model such as a Bayesian network to perform risk assessment and optimization on the operation instructions.

[0046] According to another aspect of the present application, step S4 is further:

[0047] S41. Design a reinforcement learning-based drone scheduling strategy using a dual deep Q-network algorithm, including:

[0048] Step S411, initialize the parameters of the dual deep Q network, including network weights, biases, and learning rates; initialize the experience replay pool for storing experience data; initialize the target network for stabilizing the learning process;

[0049] Step S412: normalize the preprocessed data to make it meet the input requirements of the dual deep Q network;

[0050] Step S413: Use the current state data to input into the dual deep Q network, evaluate the value of the current state, calculate the Q value of each possible action, and select the action with the highest Q value as the current action;

[0051] S42, and adopt experience playback and target network technology to improve the learning efficiency and stability of the one-key sequential control system, including:

[0052] Step S421, storing the current state, action, reward, and new state in the experience replay pool;

[0053] Step S422: randomly extract a batch of experience data from the experience replay pool for training the dual deep Q network;

[0054] Step S423, regularly updating the parameters of the target network to synchronize them with the parameters of the current dual deep Q network;

[0055] Step S424, using the target network to calculate the target Q value for training the dual deep Q network;

[0056] Step S425, calculate the loss function, use the gradient descent method to update the network parameters, and adjust the learning rate hyperparameters according to the change of the loss function during the training process.

[0057] According to another aspect of the present application, a two-way approval system based on the joint operation of a drone and one-button sequential control is also provided, including:

[0058] at least one processor; and

[0059] a memory communicatively connected to at least one of the processors; wherein,

[0060] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the two-way approval method based on the joint operation of the drone and one-button sequential control as described in any of the above technical solutions.

[0061] Beneficial effects: This solution solves many challenges in the joint operation of UAVs and one-button sequential control systems through system modeling, data processing, two-way approval mechanism and adaptive learning and decision-making mechanism. It has high efficiency, safety, stability, flexibility and broad application prospects. This solution can significantly improve the efficiency, accuracy and safety of the joint operation of UAVs and one-button sequential control systems, adapt to complex and changing environments, and has important practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1It is an overall flow chart of the scheme of the present invention.

[0063] Figure 2 It is a flow chart of step S1 of the present invention.

[0064] Figure 3 It is a flow chart of step S21 of the present invention.

[0065] Figure 4 It is a flow chart of step S22 of the present invention.

[0066] Figure 5 It is a flow chart of step S31 of the present invention.

[0067] Figure 6 It is a flow chart of step S32 of the present invention.

[0068] Figure 7 This is a flow chart of step S41 of the present invention.

[0069] Figure 8 This is the flow chart of step S42 of the present invention DETAILED DESCRIPTION

[0070] like Figure 1 As shown, according to one aspect of the present application, a two-way approval method based on the joint operation of a drone and a one-button sequential control includes the following steps:

[0071] Step S1, system modeling and data collection: constructing a dynamic model of the UAV and a logical model of the one-button sequential control system, and collecting data of the UAV and the one-button sequential control system, including data preprocessing;

[0072] Step S2, distributed data processing and analysis: design a distributed data processing framework, and perform feature extraction and data analysis on the data preprocessed in step S1;

[0073] Step S3: Design a two-way approval mechanism: Based on the data analysis results in step S2, design a drone approval algorithm and a one-button sequential control system approval algorithm;

[0074] Step S4, formulate adaptive learning and decision-making mechanism: design a UAV scheduling strategy based on reinforcement learning, adopt a dual deep Q network algorithm, and use experience replay and target network technology to improve the learning efficiency and stability of the one-button sequential control system.

[0075] This scheme proposes a two-way approval method based on the joint operation of UAV and one-button sequential control. Through system modeling, data processing, two-way approval mechanism and adaptive learning and decision-making mechanism, many problems in the existing technology are solved, and the dynamic model of the UAV and the logical model of the one-button sequential control system are constructed to ensure the comprehensiveness and accuracy of the system model; the collected data are pre-processed, including denoising, correction, splicing and other operations to generate a high-precision three-dimensional environmental model to improve the quality and availability of the data; a distributed data processing framework is designed, and the MapReduce distributed computing framework is used to improve data processing capabilities and real-time performance; data redundancy and backup mechanisms are designed to ensure data security and reliability; feature extraction and analysis are performed on the environmental data collected by the UAV and the operation data of the one-button sequential control system to extract key features and improve the accuracy and efficiency of data analysis; UAV approval algorithms and one-button sequential control system approval algorithms are designed to ensure the accuracy and security of operation instructions; reinforcement learning algorithms are used to optimize approval strategies to improve the accuracy and efficiency of approval; probabilistic graph models such as Bayesian networks are used to conduct risk assessment and optimization of operation instructions to further improve the security of the system; The UAV scheduling strategy of reinforcement learning adopts the dual deep Q network (DDQN) algorithm to improve the intelligence and adaptability of decision-making; the experience replay and target network technology are used to improve the learning efficiency and stability of the one-key sequential control system, ensuring the robustness of the system in complex environments; the parameters of the DDQN network are dynamically adjusted according to environmental changes and data feedback, so that it can adapt to different task requirements and improve the flexibility and adaptability of the system; the communication interface between the UAV and the one-key sequential control system is designed to ensure the real-time and reliability of data transmission and improve the response speed of the system; the one-key sequential control system realizes the automatic control of complex equipment through simple operation instructions, improving the simplicity and efficiency of operation; through distributed data processing, two-way approval mechanism and adaptive learning and decision-making mechanism, the high stability and security of the system in complex environments are ensured; the time series analysis method is used to predict and detect anomalies of operation data, timely discover and handle system faults, and improve the reliability and continuity of the system; this scheme is suitable for agriculture, logistics, emergency rescue, power inspection and other fields, and has broad application prospects; the system can adapt to complex and changing environments, dynamically adjust operation strategies, and improve operation effects and efficiency.

[0076] This solution solves many challenges in the joint operation of drones and one-button sequential control systems through system modeling, data processing, two-way approval mechanism and adaptive learning and decision-making mechanism. It has high efficiency, safety, stability, flexibility and broad application prospects. Through these innovations, this solution can significantly improve the efficiency, accuracy and safety of the joint operation of drones and one-button sequential control systems, adapt to complex and changing environments, and has important practical application value.

[0077] According to another aspect of the present application, the step S1 is further:

[0078] Step S11: Establishing an integrated model of the drone and the one-button sequential control system, including:

[0079] UAV system modeling: Establish the UAV's dynamic model, including modules such as flight control, navigation, and sensor data acquisition; design the communication interface between the UAV and the one-button sequential control system to ensure the real-time and reliability of data transmission;

[0080] One-button sequential control system modeling: Establish a logical model of the one-button sequential control system, including modules such as operation command generation, execution, and feedback; design the interface between the one-button sequential control system and the UAV system to ensure the accurate transmission and execution of operation commands;

[0081] Step S12, data collection and preprocessing, includes:

[0082] UAV data collection: Use the sensors carried by the drone to collect on-site environmental data; and pre-process the collected data, including denoising, correction, and splicing operations to generate a high-precision three-dimensional environmental model;

[0083] One-key sequential control system data collection: collect the operation log and equipment status data of the one-key sequential control system; and pre-process the collected data to ensure the integrity and consistency of the data.

[0084] According to another aspect of the present application, step S2 is further:

[0085] Step S21: designing a distributed data processing framework, including:

[0086] Step S211, data sharding and distributed storage: sharding the collected data according to geographic location and time dimension, and storing it in a distributed database;

[0087] Step S212: Design data redundancy and backup mechanisms to ensure data security and reliability;

[0088] Step S213: Using the MapReduce distributed computing framework as the system framework to process the preprocessed data;

[0089] Step S214: design a scheduling strategy for data processing tasks to optimize the utilization of computing resources;

[0090] Step S22, data analysis and feature extraction, includes:

[0091] Step S221, environmental data analysis: analyzing the environmental data collected by the drone to extract key features, including terrain, obstacles, and lighting;

[0092] Step S222: using the SVM machine learning algorithm to classify and predict environmental features;

[0093] Step S223, operation data analysis: analyzing the operation data of the one-key sequential control system and extracting key features, including: operation frequency and device status change;

[0094] Step S224: Use time series analysis methods to predict and detect anomalies in the operation data.

[0095] According to another aspect of the present application, step S213 is further:

[0096] Step S213a, configuring the MapReduce framework, including: configuring a MapReduce cluster, including a master node (MasterNode) and multiple worker nodes (WorkerNode), wherein the master node is responsible for task scheduling, and the worker node is responsible for specific data processing tasks;

[0097] Step S213b, Map phase, includes:

[0098] Task allocation: The master node allocates Map tasks to each worker node; each worker node reads the allocated data shards from the distributed database;

[0099] Data processing: The working nodes preprocess and extract features from data shards, convert and calculate data according to data analysis requirements, and generate intermediate results;

[0100] Intermediate result storage: Worker nodes store intermediate results in a distributed storage system;

[0101] Step S213c, Reduce phase, includes:

[0102] Task allocation: The master node allocates Reduce tasks to each worker node, and each worker node reads the intermediate results from the distributed storage system;

[0103] Data aggregation: The working node aggregates and summarizes the intermediate results to generate the final analysis results. According to the data analysis requirements, the aggregated results are further processed and calculated;

[0104] Final result storage: The worker node stores the final result in a distributed database for use in subsequent steps;

[0105] Step S213d, task scheduling and resource optimization, includes:

[0106] Task scheduling strategy: Adopt dynamic scheduling strategy to dynamically adjust task allocation according to node load conditions, optimize task allocation and execution order, and improve the utilization of computing resources;

[0107] Resource monitoring and optimization: Real-time monitoring of resource usage of each node in the cluster, including CPU, memory, network, etc., dynamically adjust task allocation and resource configuration according to monitoring results, and optimize system performance;

[0108] Step S213e, fault tolerance mechanism, includes:

[0109] Node fault detection: Real-time detection of the status of each node in the cluster, timely detection of node faults, and use of heartbeat mechanism and status report to ensure timely detection and handling of node faults;

[0110] Fault recovery: When a node fails, tasks are automatically reallocated to other normal nodes to ensure task continuity and integrity; data redundancy and backup mechanisms are used to ensure data reliability and consistency.

[0111] By adopting the MapReduce distributed computing framework, step S213 can efficiently process the preprocessed data, extract key features, and generate data analysis results. The system has the advantages of high efficiency, scalability, fault tolerance, and resource optimization, which can meet the needs of large-scale data processing and improve the overall performance and stability of the system.

[0112] According to another aspect of the present application, step S3 is further:

[0113] Step S31, the two-way approval process between the drone and the one-key sequential control system includes:

[0114] Step S311: The drone generates operation suggestions based on the collected environmental data and sends them to the one-key sequential control system;

[0115] Step S312: the one-key sequential control system verifies the operation suggestion to ensure the safety and feasibility of the operation;

[0116] Step S313: the one-key sequential control system generates an operation instruction according to the operation data and sends it to the drone;

[0117] Step S314: The drone verifies the operation instruction to ensure the accuracy and real-time performance of the instruction;

[0118] Step S32: designing an approval algorithm, including:

[0119] Step S321: Design an approval algorithm based on environmental characteristics to determine the feasibility of the operation suggestion;

[0120] Step S322: Optimize the approval strategy using a reinforcement learning algorithm to improve the accuracy and efficiency of approval;

[0121] Step S323: designing an approval algorithm based on the operation data to determine the rationality of the operation instruction;

[0122] Step S324: Use a probabilistic graphical model such as a Bayesian network to perform risk assessment and optimization on the operation instructions.

[0123] According to another aspect of the present application, step S323 is further:

[0124] Step S323a, feature extraction and analysis: extracting key features from the pre-processed operation data, including:

[0125] Operation frequency: the frequency of generating operation instructions per unit time;

[0126] Equipment status change: the status change of the equipment before and after operation, such as switch status, temperature change, etc.

[0127] Operation time: the time when the operation instruction is generated and executed;

[0128] Operation history: Historical operation records of the equipment for analyzing operation patterns and trends;

[0129] Step S323b, feature analysis: perform statistical analysis on the extracted features, calculate statistical indicators such as the mean, variance, and correlation of the features; use time series analysis methods to predict and detect anomalies in the operation data to identify potential operation anomalies.

[0130] Step S323c, approving the algorithm design, includes:

[0131] Algorithm framework: Design an approval algorithm framework based on operational data, including modules such as data input, feature extraction, model training, and instruction approval;

[0132] Model training: Use machine learning algorithms, such as support vector machine (SVM), random forest, neural network, etc., to train operation data and establish an approval model for operation instructions; use historical operation data as a training set to train the model and perform cross-validation to ensure the accuracy and generalization ability of the model;

[0133] Instruction approval: Input the current operation instructions and operation data into the approval model for approval judgment; the model outputs the approval results, including the rationality score and risk assessment of the operation instructions.

[0134] Step S323d, review and feedback of the approval results, including:

[0135] Approval result evaluation: Evaluate the approval result to determine the rationality and safety of the operation instruction. If the approval result shows that the operation instruction is unreasonable or there is a risk, the system will issue a warning or refuse to execute the operation instruction;

[0136] Feedback mechanism: Feedback the approval results to the one-key sequential control system to guide the system to generate more reasonable operation instructions. According to the approval results, adjust the operation strategy and parameters to optimize the system's operation process.

[0137] Step S323e, algorithm optimization and performance improvement, includes:

[0138] Algorithm optimization: Based on feedback and data from actual applications, the approval algorithm is continuously optimized to improve the accuracy and efficiency of the algorithm; online learning technology is used to update the approval model in real time to adapt to changes in operational data and new operational requirements.

[0139] Performance improvement: Optimize the computational complexity of the algorithm, increase the execution speed of the approval algorithm, and ensure the real-time response of the system; adopt parallel computing and distributed processing technology to improve the processing power and efficiency of the algorithm.

[0140] By designing an approval algorithm based on operation data, step S323 can effectively judge the rationality of the operation instructions generated by the one-key sequential control system, ensuring the safety and accuracy of the operation instructions. The algorithm has the advantages of accuracy, safety, efficiency and scalability, and can adapt to different types of operation instructions and operation data, improving the overall performance and stability of the system.

[0141] According to another aspect of the present application, step S4 is further:

[0142] S41. Design a reinforcement learning-based drone scheduling strategy using a dual deep Q-network algorithm, including:

[0143] Step S411, initialize the parameters of the dual deep Q network, including network weights, biases, and learning rates; initialize the experience replay pool for storing experience data; initialize the target network for stabilizing the learning process;

[0144] Step S412: normalize the preprocessed data to make it meet the input requirements of the dual deep Q network;

[0145] Step S413: Use the current state data to input into the dual deep Q network, evaluate the value of the current state, calculate the Q value of each possible action, and select the action with the highest Q value as the current action;

[0146] S42, and adopt experience playback and target network technology to improve the learning efficiency and stability of the one-key sequential control system, including:

[0147] Step S421, storing the current state, action, reward, and new state in the experience replay pool;

[0148] Step S422: randomly extract a batch of experience data from the experience replay pool for training the dual deep Q network;

[0149] Step S423, regularly updating the parameters of the target network to synchronize them with the parameters of the current dual deep Q network;

[0150] Step S424, using the target network to calculate the target Q value for training the dual deep Q network;

[0151] Step S425, calculate the loss function, use the gradient descent method to update the network parameters, and adjust the learning rate hyperparameters according to the change of the loss function during the training process.

[0152] According to another aspect of the present application, step S425 is to calculate the loss function, update the parameters of the dual deep Q network (DDQN) using the gradient descent method, and dynamically adjust the learning rate hyperparameter according to the change of the loss function during the training process. Through this step, the training process of the DDQN network is ensured to be efficient, stable, and able to adapt to different task requirements, including:

[0153] Step S425a, loss function calculation, includes:

[0154] Define the loss function: Select a suitable loss function. Common loss functions include mean square error (MSE) and cross-entropy loss. For DDQN network, mean square error (MSE) is usually used as the loss function. The formula is as follows:

[0155] L(θ)=E(s,a,r,s ′ )[(Q(s,a;θ)-y) 2 ];

[0156] Among them, Q(s,a;θ) is the Q value output by the current DDQN network, y is the target Q value, and the calculation formula is:

[0157] y=r+γmax a′ Q ′ (s ′ ,a ′ θ ′ );

[0158] Where r is the reward, γ is the discount factor, and Q ′ (s ′ ,a ′ θ ′ ) is the Q value output by the target network;

[0159] Calculate the loss: randomly extract a batch of experience data from the experience replay pool, calculate the Q value and target Q value output by the current DDQN network, and use the above formula to calculate the value of the loss function;

[0160] Step S425b, updating network parameters using the gradient descent method, includes:

[0161] Calculate gradient: Use the back propagation algorithm to calculate the gradient of the loss function to the network parameters. The gradient calculation formula is:

[0162]

[0163] Update network parameters: Use the gradient descent method to update network parameters. The formula is as follows:

[0164]

[0165] Here, α is the learning rate.

[0166] Step S425c, dynamically adjusting the learning rate, includes:

[0167] Learning rate initialization: Initialize the learning rate, usually set to a small value such as 0.001.

[0168] Loss function monitoring: During the training process, monitor the changes in the loss function in real time, record the loss function value of each training cycle, and calculate the change rate of the loss function;

[0169] Learning rate adjustment strategy: dynamically adjust the learning rate according to the rate of change of the loss function; commonly used adjustment strategies include:

[0170] Learning rate decay: When the loss function changes slowly, gradually reduce the learning rate to avoid overfitting;

[0171] Increase learning rate: When the loss function changes at a large rate, increase the learning rate appropriately to speed up convergence;

[0172] Adaptive learning rate: Use adaptive learning rate algorithms, such as Adam, RMSprop, etc., to automatically adjust the learning rate;

[0173] Learning rate adjustment formula: The learning rate adjustment formula is as follows:

[0174] α←α×1+1 / (β×epoch);

[0175] Among them, β is the decay factor, epoch is the current training cycle;

[0176] Step S425d, monitoring and optimizing the training process, includes:

[0177] Training process monitoring: Real-time monitoring of indicators such as loss function value, gradient change, and learning rate change during training. Use visualization tools such as TensorBoard to display various indicators of the training process for easy analysis and optimization.

[0178] Optimize the training process: According to the monitoring results, adjust the hyperparameters in the training process, such as batch size, discount factor (γ), etc.; regularly evaluate the performance of the model to ensure the generalization ability of the model in different environments.

[0179] By calculating the loss function, updating the network parameters using the gradient descent method, and dynamically adjusting the learning rate, step S425 can ensure that the training process of the DDQN network is efficient and stable, and can adapt to different task requirements. The system has the advantages of accuracy, efficiency, stability and adaptability, which can improve the training effect and performance of the DDQN network and ensure the robustness and reliability of the system in complex environments.

[0180] According to another aspect of the present application, a two-way approval system based on the joint operation of a drone and one-button sequential control is also provided, including:

[0181] at least one processor; and

[0182] a memory communicatively connected to at least one of the processors; wherein,

[0183] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the two-way approval method based on the joint operation of the drone and one-button sequential control as described in any of the above technical solutions.

[0184] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A two-way approval method based on the joint operation of UAV and one-button sequential control, characterized in that: The steps include: Step S1, system modeling and data collection: constructing a dynamic model of the UAV and a logical model of the one-button sequential control system, and collecting data of the UAV and the one-button sequential control system, including data preprocessing; Step S2, distributed data processing and analysis: design a distributed data processing framework, and perform feature extraction and data analysis on the data preprocessed in step S1; Step S3: Design a two-way approval mechanism: Based on the data analysis results in step S2, design a drone approval algorithm and a one-button sequential control system approval algorithm; Step S4, formulate adaptive learning and decision-making mechanism: design a UAV scheduling strategy based on reinforcement learning, adopt a dual deep Q network algorithm, and use experience replay and target network technology to improve the learning efficiency and stability of the one-button sequential control system.

2. The two-way approval method based on the joint operation of drone and one-button sequential control as claimed in claim 1 is characterized in that: The step S1 is further as follows: Step S11: Establishing an integrated model of the drone and the one-button sequential control system, including: UAV system modeling: Establish the UAV's dynamic model, including modules such as flight control, navigation, and sensor data acquisition; design the communication interface between the UAV and the one-button sequential control system to ensure the real-time and reliability of data transmission; One-button sequential control system modeling: Establish a logical model of the one-button sequential control system, including modules such as operation command generation, execution, and feedback; design the interface between the one-button sequential control system and the UAV system to ensure the accurate transmission and execution of operation commands; Step S12, data collection and preprocessing, includes: UAV data collection: Use the sensors carried by the drone to collect on-site environmental data; and pre-process the collected data, including denoising, correction, and splicing operations to generate a high-precision three-dimensional environmental model; One-key sequential control system data collection: collect the operation log and equipment status data of the one-key sequential control system; and pre-process the collected data to ensure the integrity and consistency of the data.

3. The two-way approval method based on the joint operation of drone and one-button sequential control as claimed in claim 1 is characterized in that: The step S2 is further as follows: Step S21: designing a distributed data processing framework, including: Step S211, data sharding and distributed storage: sharding the collected data according to geographic location and time dimension, and storing it in a distributed database; Step S212: Design data redundancy and backup mechanisms to ensure data security and reliability; Step S213: Using the MapReduce distributed computing framework as the system framework to process the preprocessed data; Step S214: design a scheduling strategy for data processing tasks to optimize the utilization of computing resources; Step S22, data analysis and feature extraction, includes: Step S221, environmental data analysis: analyzing the environmental data collected by the drone to extract key features, including terrain, obstacles, and lighting; Step S222: using the SVM machine learning algorithm to classify and predict environmental features; Step S223, operation data analysis: analyzing the operation data of the one-key sequential control system and extracting key features, including: operation frequency and device status change; Step S224: Use time series analysis methods to predict and detect anomalies in the operation data.

4. The two-way approval method based on the joint operation of drone and one-button sequential control as claimed in claim 1 is characterized in that: The step S3 is further as follows: Step S31, the two-way approval process between the drone and the one-key sequential control system includes: Step S311: The drone generates operation suggestions based on the collected environmental data and sends them to the one-key sequential control system; Step S312: the one-key sequential control system verifies the operation suggestion to ensure the safety and feasibility of the operation; Step S313: the one-key sequential control system generates an operation instruction according to the operation data and sends it to the drone; Step S314: The drone verifies the operation instruction to ensure the accuracy and real-time performance of the instruction; Step S32: designing an approval algorithm, including: Step S321: Design an approval algorithm based on environmental characteristics to determine the feasibility of the operation suggestion; Step S322: Optimize the approval strategy using a reinforcement learning algorithm to improve the accuracy and efficiency of approval; Step S323: designing an approval algorithm based on the operation data to determine the rationality of the operation instruction; Step S324: Use a probabilistic graphical model such as a Bayesian network to perform risk assessment and optimization on the operation instructions.

5. The two-way approval method based on the joint operation of drone and one-button sequential control as claimed in claim 1 is characterized in that: The step S4 is further as follows: S41. Design a reinforcement learning-based drone scheduling strategy using a dual deep Q-network algorithm, including: Step S411, initialize the parameters of the dual deep Q network, including network weights, biases, and learning rates; initialize the experience replay pool for storing experience data; initialize the target network for stabilizing the learning process; Step S412: normalize the preprocessed data to make it meet the input requirements of the dual deep Q network; Step S413: Use the current state data to input into the dual deep Q network, evaluate the value of the current state, calculate the Q value of each possible action, and select the action with the highest Q value as the current action; S42, and adopt experience playback and target network technology to improve the learning efficiency and stability of the one-key sequential control system, including: Step S421, storing the current state, action, reward, and new state in the experience replay pool; Step S422: randomly extract a batch of experience data from the experience replay pool for training the dual deep Q network; Step S423, regularly updating the parameters of the target network to synchronize them with the parameters of the current dual deep Q network; Step S424, using the target network to calculate the target Q value for training the dual deep Q network; Step S425, calculate the loss function, use the gradient descent method to update the network parameters, and adjust the learning rate hyperparameters according to the change of the loss function during the training process.

6. A two-way approval system based on the joint operation of drones and one-button sequential control, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the two-way approval method based on the joint operation of the drone and one-button sequential control as described in any one of claims 1 to 5.

Images