Method for unmanned aerial vehicle flight control path planning in power plant scenario

Abstract

This invention discloses a method for drone flight control path planning in a power plant setting, comprising: collecting power plant environmental data through multiple sensors to construct a real-time updated environmental model; processing sensor data to form multimodal data input into an AI model to improve the accuracy of environmental understanding; generating and optimizing an initial flight path based on the multimodal data using an improved A* algorithm combined with deep learning; and dynamically adjusting the path during flight based on real-time sensor feedback to cope with obstacles and environmental changes. Dynamic adjustment includes: introducing a reinforcement learning-based path optimization strategy, combining a genetic algorithm to balance flight time, energy consumption, and obstacle avoidance effectiveness, calculating path costs, adjusting the optimization objective through weight coefficients, and performing state transitions to ensure real-time adaptation to environmental changes. This invention, through multimodal data fusion and dynamic optimization algorithms, enables safe and efficient drone flight in complex power plant environments, improves the real-time performance and adaptability of path planning, reduces risks, and increases operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning technology, and more specifically to a method for UAV flight control path planning in a power plant scenario. Background Technology

[0002] With the development of the power industry, the requirements for refined operation and maintenance of power plants are constantly increasing. Traditional manual inspections have shortcomings such as low efficiency, high safety risks, and insufficient coverage. Although drone inspections have been widely used, existing technologies still have many bottlenecks. Existing solutions mostly adopt fixed-path operations, which cannot be dynamically optimized, resulting in problems such as flight redundancy, high energy consumption, and low efficiency. The multi-modal sensor data fusion capability is insufficient, making it difficult to perceive the complex environment of power plants in real time and avoid safety hazards, and the environmental adaptability is poor. The autonomous operation capability is lacking, relying heavily on manual operation, and lacking an algorithm iteration optimization mechanism based on historical data, making it difficult to adapt to the complex and ever-changing operation and maintenance scenarios of power plants. There is an urgent need for intelligent drone inspection technology to overcome the above pain points and improve the intelligence level, operational efficiency, and safety assurance capabilities of power plant operation and maintenance. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method for UAV flight control path planning in power plant scenarios. This method solves the problems of fixed path redundancy, insufficient data fusion, and lack of autonomous operation capabilities by constructing a real-time environment model, fusing multimodal data, using an improved A* algorithm and deep learning to generate paths, and combining reinforcement learning and genetic algorithms to dynamically optimize paths.

[0004] To achieve the above objectives, the present invention provides the following technical solution, comprising the following steps: Step 1: Collect environmental data of the power plant and its surroundings using various sensors carried by the drone, and build and update the environmental model in real time; Step two involves processing the data obtained from different sensors in step one to form unified multimodal data, which is then input into the AI ​​model to improve the accuracy of environmental understanding. Step 3: Based on the multimodal data generated in Step 2, the improved A* algorithm combined with a deep learning model is used to generate the initial flight path of the UAV, and the path is adjusted by an optimization algorithm. Step 4: Take off the drone based on the initial flight path generated in Step 3. During the drone's flight, its various onboard sensors provide real-time feedback of multimodal data, and the drone dynamically adjusts its flight path based on this multimodal data to cope with possible obstacles and environmental changes. The specific steps for dynamically adjusting the flight path in step four are as follows: Step 41: Introduce reinforcement learning algorithms to continuously optimize the path selection strategy based on the UAV's historical flight data and environmental feedback; Step 42: Introduce a genetic algorithm to further optimize the path, especially when multiple objectives need to be traded, such as flight time, energy consumption, and obstacle avoidance performance. Step 43: Calculate the cost of the optimized path, evaluate the merits of the optimized path, and flexibly change the drone's preference for the shortest distance, obstacle avoidance, or time when selecting a path by adjusting the weight coefficients. Step 44 involves state transition, which dynamically updates the drone's state to ensure that path adjustments reflect environmental changes in real time, enabling the drone to safely and flexibly cope with the complex environment of the power plant.

[0005] As a further improvement to the present invention, the specific method of introducing the reinforcement learning algorithm in step four-one is as follows: A Q-learning model is adopted, and the update strategy is implemented through a state-action value function, enabling the UAV to better select the optimal path when facing complex environments. The formula is as follows: in: Current status Take action value, The immediate reward gained from the current action; α: learning rate, which determines the degree of influence of new information on old information; γ: discount factor, which measures the importance of future rewards.

[0006] As a further improvement to the present invention, the specific method for introducing a genetic algorithm to further optimize the path in step four-two is as follows: Define the fitness function of the genetic algorithm: Where: F(p): the fitness value of path p, , , The weight of each objective can be adjusted by the user according to the specific task requirements.

[0007] As a further improvement to the present invention, the specific method for calculating the optimized path cost in step four-three is as follows: The cost function for path planning is set as: in: The total cost of path p. Path length, representing the straight-line distance traveled by the drone. Obstacle avoidance cost indicates the level of risk a path faces when passing near obstacles. Time cost: Represents the estimated flight time required for the route; α, β, γ: Weighting coefficients used to adjust the impact of different costs on the total cost.

[0008] As a further improvement to the present invention, the specific method for state transition in step four is as follows: the state transition equation is defined as: in: Current status The actions currently being taken, : Current environment model, f: State transition function, which describes how to transition to the next state after taking a specific action in the current state.

[0009] The beneficial effects of this invention are as follows: Real-time environmental modeling and dynamic path adjustment solve the flight redundancy problem of traditional fixed-path operations, reducing energy consumption and improving operational efficiency; the combination of multimodal data fusion and AI models enhances environmental perception capabilities and improves adaptability to complex power plant scenarios; the introduction of reinforcement learning and genetic algorithms enables autonomous optimization of path selection, reducing reliance on manual operation and significantly improving the intelligence level and safety assurance capabilities of UAV inspections. Subsequent improvements further enhance the real-time performance, accuracy, and robustness of path planning through Q-learning algorithm optimization of decision-making strategies, genetic algorithms for multi-objective trade-offs, dynamic adjustment of cost functions, and state transition mechanisms. Together, these improvements construct an intelligent flight control system adapted to the complex environment of power plants, achieving efficient and safe path planning in complex environments such as power plants through multimodal data fusion and deep learning optimization. The innovation of this method not only improves the accuracy of path planning but also enhances the autonomous decision-making capabilities of UAVs, demonstrating broad application prospects. Detailed Implementation

[0010] The present invention will be further described in detail below with reference to the given embodiments.

[0011] The method for drone flight control path planning in a power plant scenario in this embodiment includes the following steps: Step 1: Collect environmental data of the power plant and its surroundings using various sensors carried by the drone, and build and update the environmental model in real time; Step two involves processing the data obtained from different sensors in step one to form unified multimodal data, which is then input into the AI ​​model to improve the accuracy of environmental understanding. Step 3: Based on the multimodal data generated in Step 2, the improved A* algorithm combined with a deep learning model is used to generate the initial flight path of the UAV, and the path is adjusted by an optimization algorithm. Step 4: Take off the drone based on the initial flight path generated in Step 3. During the drone's flight, its various onboard sensors provide real-time feedback of multimodal data, and the drone dynamically adjusts its flight path based on this multimodal data to cope with possible obstacles and environmental changes. The specific steps for dynamically adjusting the flight path in step four are as follows: Step 41: Introduce reinforcement learning algorithms to continuously optimize the path selection strategy based on the UAV's historical flight data and environmental feedback; Step 42: Introduce a genetic algorithm to further optimize the path, especially when multiple objectives need to be traded, such as flight time, energy consumption, and obstacle avoidance performance. Step 43: Calculate the cost of the optimized path, evaluate the merits of the optimized path, and flexibly change the drone's preference for the shortest distance, obstacle avoidance, or time when selecting a path by adjusting the weight coefficients. Step 44 involves state transition, which dynamically updates the drone's state to ensure that path adjustments reflect environmental changes in real time, enabling the drone to safely and flexibly cope with the complex environment of the power plant.

[0012] During operation, the drone first constructs an environmental model using sensors, addressing the insufficient coverage of traditional inspections. Multimodal data fusion enhances the accuracy of environmental understanding, overcoming data processing limitations. Initial path generation and dynamic adjustment mechanisms eliminate fixed path redundancy, reducing energy consumption and improving efficiency. Reinforcement learning and genetic algorithms optimize and enhance autonomous decision-making capabilities, reducing reliance on manual intervention. Through these processes, intelligent, efficient, and safe drone inspections in power plant scenarios are achieved, effectively addressing the core pain points in the background technologies.

[0013] Furthermore, the specific method for introducing reinforcement learning algorithms in step four-one is as follows: A Q-learning model is adopted, and the strategy is updated through a state-action value function, enabling the UAV to better select the optimal path when facing complex environments. The formula is as follows: in: Current status Take action value, The immediate reward gained from the current action; α: learning rate, which determines the degree of influence of new information on old information; γ: discount factor, which measures the importance of future rewards.

[0014] This process uses historical data to iteratively optimize strategies, enabling drones to quickly adapt to changes in obstacle distribution in the complex environment of power plants, improving the real-time performance and accuracy of path selection, assisting in the main path planning to solve the problem of poor environmental adaptability, and additionally realizing the self-evolution of decision-making capabilities.

[0015] Furthermore, the specific method for introducing a genetic algorithm to further optimize the path in step four-two is as follows: Define the fitness function of the genetic algorithm: Where: F(p): the fitness value of path p, , , The weight of each objective can be adjusted by the user according to the specific task requirements.

[0016] Genetic algorithms, by simulating the processes of natural selection and genetic mutation, can explore optimal solutions among multi-dimensional objectives, effectively improving the robustness and flexibility of path planning. By adjusting the weights of multiple objectives, they can achieve a dynamic balance between flight time, energy consumption, and obstacle avoidance performance, overcoming the limitations of single-objective optimization in traditional schemes and further enhancing the flexibility and task adaptability of path planning.

[0017] Furthermore, the specific method for calculating the optimized path cost in step four-three is as follows: The cost function for path planning is defined as: in: The total cost of path p. Path length, representing the straight-line distance traveled by the drone. Obstacle avoidance cost indicates the level of risk a path faces when passing near obstacles. Time cost: Represents the estimated flight time required for the route; α, β, γ: Weighting coefficients used to adjust the impact of different costs on the total cost.

[0018] This function quantifies multi-dimensional costs to comprehensively evaluate the merits of a path. By adjusting the weighting coefficients, it can flexibly change whether the UAV prioritizes the shortest distance, the optimal obstacle avoidance, or the shortest time when selecting a path. This provides an objective basis for path evaluation, assists the genetic algorithm in achieving precise optimization, and further enhances the scientific nature and interpretability of path selection.

[0019] Furthermore, the specific method for state transition in step four is as follows: Define the state transition equation as: in: Current status The actions currently being taken, : Current environment model, f: State transition function, which describes how to transition to the next state after taking a specific action in the current state.

[0020] By updating the mapping relationship between the UAV status and the environment model in real time, the UAV status is dynamically updated to ensure that the path adjustment can reflect the environmental changes in real time. This enables the UAV to safely and flexibly cope with the complex environment of the power plant, ensures that the path adjustment is synchronized with the environmental changes, solves the response lag problem in dynamic scenarios, and further improves the robustness and safety of the system.

[0021] In summary, this invention provides a UAV flight control path planning scheme that integrates multimodal data fusion, improved A* algorithm, reinforcement learning, and genetic algorithm. Through real-time environment modeling, dynamic path optimization, and multi-objective cost trade-offs, it effectively solves the problems of fixed path redundancy, poor environmental adaptability, and insufficient autonomous capability in traditional power plant inspections. It achieves technical effects such as improved inspection efficiency, reduced energy consumption, and enhanced safety assurance, providing key technical support for intelligent operation and maintenance of power plants.

[0022] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for unmanned aerial vehicle (UAV) flight control path planning in a power plant scenario, characterized in that: Includes the following steps: Step 1: Collect environmental data of the power plant and its surroundings using various sensors carried by the drone, and build and update the environmental model in real time; Step two involves processing the data obtained from different sensors in step one to form unified multimodal data, which is then input into the AI ​​model to improve the accuracy of environmental understanding. Step 3: Based on the multimodal data generated in Step 2, the improved A* algorithm combined with a deep learning model is used to generate the initial flight path of the UAV, and the path is adjusted by an optimization algorithm. Step 4: Take off the drone based on the initial flight path generated in Step 3. During the drone's flight, its various onboard sensors provide real-time feedback of multimodal data, and the drone dynamically adjusts its flight path based on this multimodal data to cope with possible obstacles and environmental changes. The specific steps for dynamically adjusting the flight path in step four are as follows: Step 41: Introduce reinforcement learning algorithms to continuously optimize the path selection strategy based on the UAV's historical flight data and environmental feedback; Step 42: Introduce a genetic algorithm to further optimize the path, especially when multiple objectives need to be traded, such as flight time, energy consumption, and obstacle avoidance performance. Step 43: Calculate the cost of the optimized path, evaluate the merits of the optimized path, and flexibly change the drone's preference for the shortest distance, obstacle avoidance, or time when selecting a path by adjusting the weight coefficients. Step 44 involves state transition, which dynamically updates the drone's state to ensure that path adjustments reflect environmental changes in real time, enabling the drone to safely and flexibly cope with the complex environment of the power plant.

2. The method for UAV flight control path planning in a power plant scenario according to claim 1, characterized in that: The specific method for introducing the reinforcement learning algorithm in step four-one is as follows: A Q-learning model is used, and the strategy is updated through a state-action value function, enabling the UAV to better select the optimal path when facing complex environments. The formula is as follows: in: Current status Take action value, The immediate reward gained from the current action; α: learning rate, which determines the degree of influence of new information on old information; γ: discount factor, which measures the importance of future rewards.

3. The method for UAV flight control path planning in a power plant scenario according to claim 2, characterized in that: The specific method for introducing a genetic algorithm to further optimize the path in step four-two is as follows: Define the fitness function of the genetic algorithm: Where: F(p): the fitness value of path p, , , The weight of each objective can be adjusted by the user according to the specific task requirements.

4. The method for UAV flight control path planning in a power plant scenario according to claim 3, characterized in that: The specific method for calculating the optimized path cost in step four-three is as follows: The cost function for path planning is defined as: in: The total cost of path p. Path length, representing the straight-line distance traveled by the drone. Obstacle avoidance cost indicates the level of risk a path faces when passing near obstacles. Time cost: Represents the estimated flight time required for the route; α, β, γ: Weighting coefficients used to adjust the impact of different costs on the total cost.

5. The method for UAV flight control path planning in a power plant scenario according to claim 4, characterized in that: The specific method for state transition in step four is as follows: The state transition equation is defined as: in: Current status The actions currently being taken, : Current environment model, f: State transition function, which describes how to transition to the next state after taking a specific action in the current state.

Images