A method for unmanned aerial vehicle data collection in a forest fire scenario

Abstract

The present application relates to unmanned aerial vehicle and internet of things field, especially relates to a kind of unmanned aerial vehicle data collection method under forest fire scene, comprising the following steps: obtaining unmanned aerial vehicle data, node data, establish regional model and forest fire model;According to unmanned aerial vehicle data, node data, regional model and forest fire model, constraint condition is established;According to constraint condition, further optimization problem is established;Unmanned aerial vehicle path planning algorithm based on double depth Q network solves optimization problem, and the optimal parameter obtained by solving is used as the initial parameter of double depth Q network, to obtain trained model;Model is deployed to unmanned aerial vehicle, and model gives the action instruction of unmanned aerial vehicle next time according to real-time data captured by unmanned aerial vehicle, and unmanned aerial vehicle executes data collection task according to action instruction.The present application enhances the rationality of decision, avoids unmanned aerial vehicle to collect invalid node, improves resource utilization;Through new communication scheduling method, data transmission efficiency is improved, and communication stability is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicles and Internet of Things, and particularly relates to a method for collecting data of unmanned aerial vehicles in a forest fire scenario. BACKGROUND

[0002] Forest fires pose a significant risk to forest resource protection, and firefighters face serious life safety threats during each forest fire extinguishing operation. Timely overall situational awareness of the fire scene can minimize such threats.

[0003] The continuous and rapid development of wireless communication technology and sensor performance has made the deployment and use of wireless sensors simpler and more reliable, which has promoted the progress of the Internet of Things and wireless sensor networks. By reasonably deploying wireless sensors, specific data can be quickly collected for specific areas in a convenient and efficient manner. This realizes on-demand data collection and adapts to dynamically changing complex environments while saving energy consumption for data transmission. This feature makes unmanned aerial vehicle-assisted Internet of Things and wireless sensor networks no longer limited by network connectivity, and the deployment of device nodes only needs to consider the application requirements and the geographical structure of the target area.

[0004] Prior art such as CN2024119045509A discloses a low-altitude environment multi-unmanned aerial vehicle efficient patrol method based on a terrain elevation map, which considers multi-unmanned aerial vehicle cooperation, directly collects forest fire scene information during each unmanned aerial vehicle patrol, and uniformly plans the initial path of the multi-unmanned aerial vehicle by maintaining a global patrol information map. In the planning process of a single machine, the safe path of a single unmanned aerial vehicle is obtained by combining environmental obstacle information. The technology has the following problems:

[0005] 1. Low data and resource utilization efficiency: The scheme in the document mainly analyzes and decides based on the current unmanned aerial vehicle detection data and limited prior information, without fully exploiting and utilizing a large amount of historical fire data and related data in different forest environments. This makes the system's adaptability and generalization ability relatively weak when facing new fire scenarios or complex situations, making it difficult to make optimal decisions quickly and accurately. For example, the forest vegetation, topography, and climate conditions in different regions differ greatly, and the scheme may need to adjust parameters and models for each new scenario, which is inefficient.

[0006] 2. The influence of wind speed on unmanned aerial vehicles is not considered: When unmanned aerial vehicles perform data collection tasks, wind speed will limit the flight of unmanned aerial vehicles.

[0007] Therefore, there is an urgent need for a method for collecting data of unmanned aerial vehicles in a forest fire scenario. SUMMARY

[0008] In view of this, the present application discloses a kind of unmanned plane data collection method under forest fire scene, to solve the above problems;Including:

[0009] S1, obtain unmanned plane data, node data;

[0010] S2, according to node data, establish regional model, adopt cellular automaton to describe forest fire spread process and establish forest fire model;

[0011] S3, according to unmanned plane data, establish unmanned plane flight constraint condition;According to regional model, establish obstacle avoidance constraint condition;According to failure node, unmanned plane data and node data, establish communication constraint condition;

[0012] S4, comprehensive unmanned plane flight constraint condition, obstacle avoidance constraint condition, communication constraint condition, establish optimization problem;

[0013] S5, based on the double depth Q network's unmanned plane path planning algorithm is solved to optimization problem, obtains the optimal parameter of double depth Q network, and optimal parameter is used as the initial parameter of path planning algorithm, obtains the trained unmanned plane path planning model;

[0014] S6, the trained model is deployed to unmanned plane, and real-time data is input into the trained model by unmanned plane, and the model gives the action instruction of next time of unmanned plane, and unmanned plane executes data collection task according to action instruction.

[0015] The beneficial effects of the present application include:

[0016] By obtaining the environmental wind speed collected by ground node, and adding wind speed factor into unmanned plane flight constraint condition, the stability and adaptability of unmanned plane in real-time execution task are increased;By considering the uncertainty of the life cycle of ground node caused by fire, the rationality of decision is enhanced, and the resource utilization rate is improved by avoiding unmanned plane to collect invalid node;By adopting new communication scheduling method, data transmission efficiency is improved, and communication stability is guaranteed. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is system model schematic diagram of the present application;

[0018] Figure 2 It is convergence process comparison chart after smoothing under different algorithms of the present application;

[0019] Figure 3 It is data collection rate comparison chart of unmanned plane in each task under different algorithms of the present application;

[0020] Figure 4 It is success rate comparison chart of safe path planning of unmanned plane under different algorithms of the present application; DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme, characteristics and advantages of the present application more clear, the present application is further described below in combination with the drawings and examples.

[0022] As Figure 1 shown is a corresponding scene schematic diagram of the present embodiment, the unmanned aerial vehicle data collection method in the forest fire scene described in the present embodiment, comprising steps of:

[0023] S1, acquiring unmanned aerial vehicle data and node data.

[0024] S2, establishing a regional model and a forest fire model according to the node data, the forest fire model being used to calculate the failed node.

[0025] Specifically, in the forest fire site, wind is an important influencing factor, wind will affect the spreading process of forest fire, and also affect the air density around the unmanned aerial vehicle rotor blade, resulting in the change of motor power. The environmental wind speed v w [n] represents wherein represents the data of the unmanned aerial vehicle wind speed sensor, n represents the nth time slot obtained by uniformly dividing the maximum task time, and Δv w [n] represents the uncertainty of wind speed, which is determined by the difference between meteorological data and on-site wind speed. The greater the difference, the greater the uncertainty, the faster the change of wind, and the greater the maximum uncertainty of wind speed, which is obtained from the ground node.

[0026] Further, the forest fire model adopts a cellular automaton to describe the forest fire spreading process.

[0027] Specifically, the cellular automaton divides the task area into closely adjacent square grids, each grid being a cell, the properties of the cell being determined by the space terrain, vegetation, meteorological conditions, ignition time and burning state and other factors inside the cell, and the state of the cell can be divided into five stages: unburned, burning, completely burned, unextinguished and completely extinguished, which are represented as F0, F1, F2, F3 and F4 respectively.

[0028] The initial rate R F0 of forest fire spreading is calculated by the formula:

[0029] R F0 =a*T F +b*V F +c*(100-H F )+d

[0030] wherein, T F is the temperature, V F is the wind grade, and H F ​For relative humidity, a, b, c, d are all empirical parameters, and their preferred values are 0.03, 0.05, 0.01, -0.3 respectively. Further considering the influence of fuel type, terrain slope and wind vector on the initial spread rate, the modified forest fire spread rate is R = R F0 *K f *K s *K w wherein R is the modified forest fire spread rate; K f is the fuel type correction coefficient; K s is the terrain slope correction coefficient, K s describes the influence of different slopes on the forest fire spread rate; K w is the wind vector correction coefficient, K w describes the influence of different wind force levels and wind direction conditions on the spread rate.

[0031] Further, the coordinate index of each cell is defined as u and v, and the forest fire spread rate of cell (u, v) is defined as The distance between the cell and the adjacent cell is the edge length L of the cell. The state transition rule of the cell is: when the unburned cell is ignited, F0 becomes F1; F1 is converted to F2 after the internal burning time Δn, and ignites F0 in the direction parallel to the fire line within Δn time, wherein R IN is the internal forest fire spread rate of the cell, and L represents the edge length of the cell; F2 ignites all F0 around it, and is converted to F3 when there is no F0 around the cell in F2 state, and consumes Δn time; F3 enters F4 state after Δn time. The set represents the area where the forest fire spreads, which is composed of cells in F1 and subsequent states. If the coordinate q k of node k is in , then node k is a failed node.

[0032] S3, establishing a UAV flight constraint condition according to the UAV data; establishing an obstacle avoidance constraint condition according to the area model; and establishing a communication constraint condition according to the failed node, the UAV data and the node data.

[0033] Specifically, regarding the UAV flight constraint condition: the spatial position of the UAV is:

[0034] q[n] = (x[n], y[n], h u )

[0035] wherein h u is the flight height of the UAV, and the projection coordinates of the UAV in the flight plane are:

[0036] q u [n] = (x[n], y[n])

[0037] Drone heading Furthermore, the system has K wireless sensor nodes, where the position of the k-th node ∈ {1,2,…,K} is q. k =(x k ,y k ,z k Limited by the performance of the drone itself, there are constraints. and Due to the influence of wind speed in the regional model, there are constraints:

[0038] Specifically, regarding obstacle avoidance constraints: the mountain terrain is represented as follows:

[0039]

[0040] Where (x,y) represents the geographic coordinates, z is the altitude of the location, and h i The terrain correction parameter controls the overall height of the i-th mountain, x. si With y si Let $\mathbf$ be the attenuation of mountain body $i$ along the x-axis and y-axis, respectively, controlling the slope of the mountain. During drone flight, it needs to avoid both known terrain obstacles and unknown environmental obstacles. The region of obstacles in the drone's flight altitude plane is represented by the set $D = {D1, D2, ... D\mathbff}. j}, where D j Let D represent the j-th obstacle. j That is, z(x,y)>h u The set of points whose altitude is higher than the drone's flight altitude; the obstacle avoidance constraints for the drone are...

[0041] Specifically, regarding communication constraints: communication constraints include scheduling constraints and data transmission constraints. In forest fire scenarios, the channel between the UAV and nodes is mainly affected by vegetation. This invention adopts the path loss model PL. k The formula for [n] is:

[0042] PL k [n] = 20log 10 (f c )+10n1log 10 (θ k [n])+10n2log 10 (d k [n])+X σ

[0043]

[0044] Among them, f cdenotes the carrier frequency, θ k [n] denotes the elevation angle of the UAV relative to the ground node; X σ denotes the shadow fading, which follows a Gaussian distribution; d k [n] denotes the distance between the UAV and the node k; the spatial position of the UAV is (x[n], y[n], h u ); the position of the kth node is (x k , y k , z k ); n1 and n2 are the elevation angle influence factor and path loss exponent, respectively.

[0045] Further, the node transmit power is defined as p0, the UAV receiver noise power is defined as p n , and the signal-to-noise ratio γ k [n] of the UAV receiver is:

[0046] γ k [n] = p0- PL k [n] - p n

[0047] Further, the communication rate R K [n] between the UAV and the node is:

[0048]

[0049] where B denotes the channel bandwidth.

[0050] Further, in order to support stable data transmission, it is necessary to start the data collection task when the received signal-to-noise ratio γ k [n] of the UAV is greater than or equal to γ0, where γ0 is a signal-to-noise ratio threshold set by humans, and the preferred range is -5dB to -10dB, and the preferred value in the embodiment is -5dB. The UAV performs data collection tasks with at most one node each time, and the connection between the UAV and the kth node is denoted by , and the scheduling constraint is denoted by:

[0051] Specifically, the UAV preferentially performs data transmission tasks with nodes having good and stable communication channels, and the formula of the scheduling strategy is:

[0052]

[0053] where active nodes denote nodes with a signal-to-noise ratio γk[n] greater than or equal to γ0, and otherwise denotes nodes with a signal-to-noise ratio lower than the threshold γ0 and failed nodes.

[0054] Further, denotes the total amount of data collected from the node k at time slot n, and the superscript L is used to distinguish and Q K The data transmission constraint represented by the amount of data collected by the UAV from the node k is: The total amount of data that can be collected by the UAV during the data collection process of all nodes is The maximum task time is uniformly discretized into N time slots, so that the position of the UAV in each time slot can be approximately regarded as invariable, and the time slot length of the nth time slot is δ[n] = T max / N. In this embodiment, the number of time slots N is preferably 300.

[0055] S4, comprehensive UAV flight constraint condition, obstacle avoidance constraint condition, communication constraint condition, establish optimization problem.

[0056] Specifically, the total amount of data collected quickly from all nodes is maximized, and the optimization problem is established as follows:

[0057]

[0058] subject to

[0059]

[0060] C7:T≤T max

[0061] Wherein, C1 is the flight speed performance constraint of the UAV, C2 is the steering performance constraint of the UAV, C3 is the uncertainty constraint of the environmental wind, C4 is the obstacle avoidance constraint, C5 is the scheduling constraint, C6 is the data transmission constraint, and C7 represents the task time constraint.

[0062] S5, the UAV path planning algorithm based on double depth Q network solves the optimization problem, obtains the optimal parameters of the double depth Q network, takes the optimal parameters as the initial parameters of the path planning algorithm, and obtains the trained UAV path planning model.

[0063] Specifically, the sequence decision problem is expressed by Markov decision process, that is, the formula is expressed, and the tuple is used to represent, and the specific meaning of the decision process is as follows:

[0064] The state space is represented by S [n] = (S u [n], S o [n], S s [n], S t ); wherein, S u [n] is the state information vector of the UAV, v[n] represents the flight speed of the UAV, and the formula is v[n] = v u[n] + λv w [n], v u [n] is the speed of the UAV in the windless environment, λ is the discount factor of the wind speed influence, preferably in the range of 0.005-0.01, and the preferred value in the embodiment is 0.01; S o [n] is the surrounding environment information vector detected by the UAV; S o [n] = (d1[n], d2[n], …, d8[n]), where d i [n], i ∈ {1, 2, …, 8} represents the distance between the UAV and the obstacle detected by the sensor in a certain direction, and the sensor has a maximum detection distance limit; S s [n] is the global node information vector, S t is the task execution time consumption; further, s k [n] represents the information vector of a single node k.

[0065] represents an action, and the action of the agent is limited by the mechanical performance of the UAV, and only executable actions sampled from the legal action set can be taken to and define a safety rule that when the next action of the UAV will fly towards the obstacle, the action is cancelled and a penalty is given.

[0066] represents the transition of the state in the whole system.

[0067] represents the reward function set according to the optimization problem.

[0068] A double deep Q network reinforcement learning algorithm is used to solve the problem, and the following settings are made when solving:

[0069] The data collection reward is wherein α1 represents a data collection amount scaling coefficient, which is used to reduce the data collection amount to a suitable range, and in the embodiment, α1 is preferably 10 -6 , represents the total data collection amount collected from node k at time n+1, and k represents the number of nodes.

[0070] The remaining task time reward is α2 represents a time amount scaling coefficient, which is used to reduce the time amount to a suitable range, and in the embodiment, α2 is preferably 0.1.

[0071] The dynamic obstacle avoidance penalty is d = min(d1[n], d2[n], ..., d8[n]), α3 represents the obstacle avoidance penalty coefficient, used to amplify the shortest detection distance to a suitable range. In this embodiment, α3 is preferably 10. d represents the minimum distance between the UAV and the obstacle detected by the sensor in all directions. s This indicates the maximum detection distance of the drone from the sensor.

[0072] For known obstacles, there is a static obstacle collision penalty. α4 represents the collision penalty value, and in this embodiment, α4 is preferably 1.

[0073] To reduce invalid actions, the penalty for invalid actions is R. ia =-α5, where α5 represents the penalty value for invalid actions. In this embodiment, α5 is preferably 1.

[0074] The global reward function is R = R dc +R tk +R uo +R so +R ia .

[0075] Furthermore, the initial parameters for solving the Markov decision process are set as follows:

[0076] Map size 300*300*50 Number of ground nodes 10 Maximum flight speed of UAV 12 m / s Maximum turning angle π Flight height of UAV 30m Maximum mission time 300s Node transmit power 0.1W Signal-to-noise ratio threshold -5 dB

[0077] Furthermore, the optimization problem is solved to obtain the optimal parameters of the dual-depth Q-network. The solution steps include:

[0078] Step 1: Set the signal-to-noise ratio threshold γ0 and the maximum task time T. max The maximum flight speed of the drone, v max and maximum steering angle

[0079] Step 2: Initialize the replay buffer D, estimate the network ξ, and initialize the target network ξ by copying the estimated network parameters. - .

[0080] Step 3, from v max , Calculate the legal action space.

[0081] Step 4: Initialize the experimental environment by setting all environmental parameters to their initial values.

[0082] Step 5: Obtain the environmental information vector S[n], input S[n] into the network to obtain the action a[n], and the agent interacts with the environment to obtain a reward. And the new environmental information vector S[n+1].

[0083] Step 6: Plot the tuples Add to replay cache D.

[0084] Step 7: Sample a certain number of tuples from D and calculate the gradient using the loss function to update the network parameters. Further, the target Q-value function in the dual-depth Q-network is as follows:

[0085] S[n+1] is a non-final state.

[0086] Where S[n+1] is the condition for determining the final state: all The corresponding node signal disappears, or the task execution time T is greater than the maximum task time constraint T. max .

[0087] Step 8, every N r The next update copies the parameters in ξ to ξ. - In, N r In this embodiment, N is a threshold number of training iterations set manually for the neural network. r The preferred value is 600.

[0088] Step 9: Determine if S[n+1] is the final state; if yes, proceed to step 4; otherwise, proceed to step 5.

[0089] Step 10: If the data collection rate and path planning success rate reach the expected values, stop the training loop, obtain the optimal neural network parameters, and set the optimal parameters as the initial values ​​of each parameter in the neural network to obtain the trained model. The expected value for the data collection rate is preferably 98%, and the expected value for the path planning success rate is preferably 96%.

[0090] Furthermore, the optimal parameters are used as the initial parameters for the path planning algorithm to obtain a trained UAV path planning model.

[0091] S6. Deploy the trained model onto the drone. The drone captures real-time data and inputs it into the trained model. The model provides the drone with action instructions for the next moment. The drone then performs the data collection task according to the action instructions.

[0092] Furthermore, such as Figure 2 As shown, Figure 2 The convergence process of the proposed algorithm and the comparative algorithm after smoothing is demonstrated. The proposed algorithm is a UAV path planning algorithm based on dual-depth Q-networks, and the comparative algorithm is the Dueling DQN algorithm. It can be seen that the reward value of the Dueling DQN algorithm fluctuates drastically after rising and does not reach the maximum reward value; the proposed algorithm has significantly improved performance and exhibits smaller fluctuations during the convergence process, demonstrating stability in random environments.

[0093] As Figure 3 shown, Figure 3 The data collection rate of the unmanned aerial vehicle in each task under different algorithms is shown, which is affected by environmental factors and the fire spread process, and the collection rate of the Dueling DQN algorithm rises at the beginning and then hovers around 90%; the unmanned aerial vehicle path planning algorithm based on the double deep Q network adopted by the application rises at the beginning and then reaches 100%, and then stabilizes at a level close to 100%, which shows the effectiveness of the algorithm in the data collection task.

[0094] As Figure 4 shown, Figure 4 The success rate of the unmanned aerial vehicle planning a safe path under different algorithms is shown, and it can be seen that the success rates of the comparative algorithm and the algorithm adopted by the application can reach a level of 80% at the beginning, while the path planning success rate of the Dueling DQN algorithm continuously decreases and fluctuates sharply; the success rate of the algorithm adopted by the application converges to more than 95%, which shows strong adaptability to the complex environment of forest fires.

[0095] Finally, it should be noted that the above only describes some embodiments of the application, and for those skilled in the art, various changes, modifications, replacements and deformations of these embodiments can be made without departing from the principles and spirit of the application, the protection scope of the application is defined by the appended claims and their equivalents, and the above behaviors should be covered within the protection scope of the application.

Claims

1. A method for collecting data from drones in a forest fire scenario, characterized in that, include: S1. Acquire drone data and node data; S2. Establish a regional model and a forest fire model based on node data. The forest fire model is used to calculate failed nodes. The forest fire model uses cellular automata to describe the forest fire spread process. The cellular automata includes states: not burning, burning, completely burning, not extinguished, and completely extinguished, respectively represented as follows: The transition rules between states are as follows: When unburned cells are ignited, Become ; After internal combustion time Later converted to , Ignition in the direction parallel to the fire line within a time limit ,in It is the rate of forest fire spread within a cell. Indicates the side length of a cell; Ignite everything around it ,when The state of the cell does not exist around it Time conversion ,time consuming ; experience Enter after time state; Using sets This indicates the area that the forest fire has spread to, from the point of entry. and the subsequent state's cell composition, if the node coordinates exist In the middle, then the node This is a failed node; S3. Establish UAV flight constraints based on UAV data; establish obstacle avoidance constraints based on the region model. Establish communication constraints based on failed nodes, UAV data, and node data; S4. Based on the constraints of UAV flight, obstacle avoidance, and communication, establish an optimization problem. The established optimization problems include: UAV flight speed performance constraints, UAV turning performance constraints, environmental wind uncertainty constraints, obstacle avoidance constraints, scheduling constraints, data transmission constraints, and task time constraints. S5. The UAV path planning algorithm based on dual-depth Q-network solves the optimization problem, obtains the optimal parameters of the dual-depth Q-network, and uses the optimal parameters as the initial parameters of the dual-depth Q-network to obtain the trained UAV path planning model. S6. Deploy the trained model onto the drone. The drone captures real-time data and inputs it into the trained model. The model provides the drone with action instructions for the next moment. The drone then performs the data collection task according to the action instructions.

2. The method for collecting unmanned aerial vehicle (UAV) data in a forest fire scenario according to claim 1, characterized in that, The formula for the communication scheduling strategy used in the scheduling constraints is as follows: in, Indicates drones and the first The connection of each node Indicates the first Signal-to-noise ratio of each node, active nodes Defined as signal-to-noise ratio The node, Indicates the signal-to-noise ratio threshold. This indicates nodes with a signal-to-noise ratio below the threshold and failed nodes.

3. The method for collecting unmanned aerial vehicle (UAV) data in a forest fire scenario according to claim 2, characterized in that, Signal-to-noise ratio threshold The value range is -5dB to -10dB.

4. The method for collecting drone data in a forest fire scenario according to claim 1, characterized in that, Discount factor for wind speed effect The value range is 0.005 ~ 0.01.

Images