Fire extinguishing control method and system for unmanned aerial vehicle

Abstract

The invention relates to the technical field of fire extinguishing control, in particular to an unmanned aerial vehicle fire extinguishing control method and system. The method comprises the following steps: carrying out hot-flow wind field real-time data acquisition on a high-rise building fire scene through a sensor carried on an unmanned aerial vehicle, carrying out approaching hot-flow wind field intensity difference analysis, and then carrying out unmanned aerial vehicle flight attitude instability numerical value quantification to obtain flight attitude instability numerical value quantification data; time-varying attitude spraying strategy adjustment is carried out according to the flight attitude instability numerical value quantization data, and a time-varying attitude spraying adjustment strategy is obtained; building a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy based on a strategy gradient algorithm to obtain a fire extinguishing spraying control model; and sending the fire extinguishing spraying control model to the terminal. The fire extinguishing control technology is optimized, so that the fire extinguishing control technology is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of fire extinguishing control, and particularly to a method and system for controlling a drone to extinguish a fire. Background Art

[0002] Fires in high-rise buildings are characterized by rapid spread of the fire, serious accumulation of smoke, and difficult evacuation of people. Previous fire-fighting methods often have difficulty in quickly and effectively extinguishing the fire source in the initial stage of the fire. Therefore, developing an efficient and intelligent fire extinguishing control method has become an important direction in the development of modern fire-fighting technology. As an emerging emergency response means, the drone fire extinguishing technology has gradually attracted attention due to its advantages such as high mobility, rapid deployment, and flexible operation, and has to a certain extent made up for the deficiencies of traditional fire-fighting methods. The drone fire extinguishing system mainly realizes remote and precise fire extinguishing operations by carrying equipment such as water spray and fire extinguishing agents. However, there is a problem in a traditional method for controlling a drone to extinguish a fire that the rising hot air flow generated at the fire site causes unstable air flow, which affects the accuracy of flight attitude control and thus affects the accuracy of fire extinguishing agent delivery. Summary of the Invention

[0003] Based on this, it is necessary to provide a method and system for controlling a drone to extinguish a fire to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for controlling a drone to extinguish a fire, the method includes the following steps: Step S1: Collect real-time data of the heat flow wind field at the high-rise building fire site through sensors carried on the drone to obtain real-time data of the heat flow wind field; perform an analysis of the intensity difference of the heat flow wind field in the vicinity to obtain grid intensity difference data of the heat flow wind field; Step S2: Quantify the numerical value of the flight attitude instability according to the grid intensity difference data of the heat flow wind field to obtain numerical value quantification data of the flight attitude instability; adjust the spraying strategy with time-varying attitude according to the numerical value quantification data of the flight attitude instability to obtain a time-varying attitude spraying adjustment strategy; Step S3: Construct a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy based on the policy gradient algorithm to obtain a fire extinguishing spraying control model; send the fire extinguishing spraying control model to the terminal to execute the method for controlling a drone to extinguish a fire.

[0005] Preferably, step S1 includes the following steps: Step S11: Collect real-time data of the heat flow wind field at the high-rise building fire site through sensors carried on the drone to obtain real-time data of the heat flow wind field; Step S12: Fill in the missing values of the real-time data of the heat flow wind field to obtain real-time filled data of the heat flow wind field; Step S13: Perform an analysis of the difference in the intensity of the approaching heat flow wind field on the real-time filled data of the heat flow wind field to obtain the difference data of the approaching intensity of the heat flow wind field; Step S14: Perform a grid structure conversion on the difference data of the approaching intensity of the heat flow wind field to obtain the difference data of the grid intensity of the heat flow wind field.

[0006] Preferably, step S2 includes the following steps: Step S21: Analyze the turbulent force domain of the difference data of the grid intensity of the heat flow wind field to obtain the turbulent force domain data of the heat flow wind field; Step S22: Quantify the numerical value of the instability of the UAV flight attitude based on the turbulent force domain data of the heat flow wind field to obtain the numerical quantification data of the flight attitude instability; Step S23: Analyze the amplitude limit of the lift constraint based on the numerical quantification data of the flight attitude instability to obtain the amplitude limit data of the flight lift constraint; Step S24: Adjust the time-varying attitude spraying strategy according to the numerical quantification data of the flight attitude instability, the amplitude limit data of the flight lift constraint, and the turbulent force domain data of the heat flow wind field to obtain the time-varying attitude spraying adjustment strategy.

[0007] Preferably, step S22 includes the following steps: Step S221: Analyze the spatial heterogeneity of the turbulent force domain data of the heat flow wind field to obtain the spatial heterogeneity data of the turbulent force domain; Step S222: Analyze the spectral density of the horizontal / vertical azimuth disturbing force based on the spatial heterogeneity data of the turbulent force domain and the turbulent force domain data of the heat flow wind field to obtain the spectral density of the horizontal / vertical azimuth disturbing force; Step S223: Analyze the disordered change of the UAV's rotation and swing according to the spectral density of the horizontal / vertical azimuth disturbing force to obtain the disordered change data of the rotation and swing; Step S224: Perform a regression analysis of the tilt moment fluctuation on the disordered change data of the rotation and swing to obtain the regression data of the tilt moment fluctuation; Step S225: Analyze the lateral airflow pressure rollover limit based on the disordered change data of the rotation and swing to obtain the lateral airflow rollover limit pressure data; Step S226: Quantify the numerical value of the instability of the UAV flight attitude according to the regression data of the tilt moment fluctuation and the lateral airflow rollover limit pressure data to obtain the numerical quantification data of the flight attitude instability.

[0008] Preferably, step S223 includes the following steps: Obtain the aerodynamic layout of the UAV; calculate the time-sequence pressure difference of the airflow disturbance between the horizontal / vertical azimuths for the spectral density of the horizontal / vertical azimuth disturbing force to obtain the time-sequence pressure difference of the airflow disturbance between the horizontal / vertical azimuths; Analyze the centroid repeated offset trajectory of the aerodynamic layout of the UAV according to the time - series pressure difference of airflow disturbance to obtain the flight centroid repeated offset trajectory; Calculate the average difference of the relative angle of horizontal offset of the flight centroid repeated offset trajectory to obtain the average difference of the relative angle of centroid offset; Analyze the disordered changes of the rotation and swing of the UAV based on the average difference of the relative angle of centroid offset and the flight centroid repeated offset trajectory to obtain the data of disordered changes of rotation and swing.

[0009] Preferably, step S24 includes the following steps: Step S241: Obtain the data of the fire - fighting spraying material carried by the UAV; Step S242: Analyze the amplitude range of attitude jitter of the numerically - quantified data of flight attitude instability to obtain the amplitude range of attitude jitter; Step S243: Adjust the flight time - series attitude control according to the amplitude range of attitude jitter and the data of flight lift - limit constraint amplitude to obtain the flight time - series attitude control data; Step S244: Control the spraying pressure of the fire - fighting spraying material data according to the amplitude range of attitude jitter and the data of heat - flow wind - field turbulence force domain to obtain the flight spraying pressure control data; Step S245: Adjust the time - varying attitude spraying strategy based on the flight time - series attitude control data and the flight spraying pressure control data to obtain the time - varying attitude spraying adjustment strategy.

[0010] Preferably, step S243 includes the following steps: Analyze the jitter lateral tilt angle of the amplitude range of attitude jitter to obtain the jitter lateral tilt angle data; Conduct a regression analysis of the lift - limit attitude angle margin based on the jitter lateral tilt angle data and the flight lift - limit constraint amplitude data to obtain the regression data of the lift - limit attitude angle margin; Match the axial torque output according to the jitter lateral tilt angle data and the flight lift - limit constraint amplitude data to obtain the axial torque output matching data; Adjust the flight time - series attitude control according to the regression data of the lift - limit attitude angle margin and the axial torque output matching data to obtain the flight time - series attitude control data.

[0011] 8. The UAV fire - fighting control method according to claim 7, wherein step S244 includes the following steps: Perform a spatial vector decomposition process on the heat - flow wind - field turbulence force domain data to obtain the turbulence direction component data; Identify and process the particle size distribution and viscosity characteristics of the fire - fighting spraying material data to obtain the spraying material particle size distribution data and the spraying material viscosity characteristics data respectively; Based on the data of the spoiler direction component and the amplitude range of attitude jitter, the simulation evaluation of the kinetic energy fluctuation range of the spraying air flow barrier is carried out to obtain the kinetic energy fluctuation range of the spraying air flow barrier; The approximate integration of the flow resistance fluctuation of the kinetic energy fluctuation range of the spraying air flow barrier is carried out to obtain the approximate data of the flow resistance fluctuation; According to the approximate data of the flow resistance fluctuation, the spraying pressure is controlled for the particle size distribution data and the viscosity characteristic data of the spraying material to obtain the flight spraying pressure control data.

[0012] Preferably, step S3 includes the following steps: Step S31: Normalize the time-varying attitude spraying adjustment strategy to obtain the normalized data of the time-varying attitude spraying adjustment; Step S32: Perform logical iterative learning on the normalized data of the time-varying attitude spraying adjustment to obtain the logical iterative data of the time-varying attitude spraying; Step S33: Based on the policy gradient algorithm, construct a fire extinguishing spraying control model for the logical iterative data of the time-varying attitude spraying to obtain the fire extinguishing spraying control model; Step S34: Send the fire extinguishing spraying control model to the terminal to execute the UAV fire extinguishing control method.

[0013] Preferably, the present invention also provides a UAV fire extinguishing control system for executing the UAV fire extinguishing control method as described above. The UAV fire extinguishing control system includes: A heat flow wind field intensity difference analysis module, which is used to collect real-time data of the heat flow wind field at the high-rise building fire site through sensors carried on the UAV to obtain real-time data of the heat flow wind field; perform an in-depth analysis of the intensity difference of the heat flow wind field on the real-time data of the heat flow wind field to obtain the intensity difference data of the in-depth grid of the heat flow wind field; A time-varying attitude spraying strategy adjustment module, which is used to numerically quantify the instability of the UAV flight attitude according to the intensity difference data of the in-depth grid of the heat flow wind field to obtain the numerically quantified data of the flight attitude instability; adjust the time-varying attitude spraying strategy according to the numerically quantified data of the flight attitude instability to obtain the time-varying attitude spraying adjustment strategy; A fire extinguishing spraying control model construction module, which is used to construct a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy based on the policy gradient algorithm to obtain the fire extinguishing spraying control model; send the fire extinguishing spraying control model to the terminal to execute the UAV fire extinguishing control method.

[0014] The beneficial effects of the present invention are as follows. By carrying sensors on the drone, heat flux and wind field data at the high-rise building fire scene can be obtained in real time. The acquisition of this data can comprehensively understand the heat flux distribution and wind field changes in the fire environment, thereby providing accurate data support for subsequent fire extinguishing operations. By analyzing the differences in heat flux and wind field intensity, the heat distribution and wind direction changes in the fire source area can be identified, providing a basis for the flight path planning and spraying strategy formulation of the drone. Real-time acquisition and differential analysis can improve the efficiency of dynamic response at the fire scene, making the fire extinguishing operation more accurate and timely. By further processing the data of the differences in heat flux and wind field grid intensity and quantifying the numerical instability of the flight attitude, the flight stability of the drone in a complex fire scene can be evaluated in a timely manner. This step can effectively quantify the flight attitude problems of the drone under different environmental conditions, such as stability problems when affected by heat flux or wind field changes. Based on the analysis of the flight attitude instability data, the time-varying attitude spraying strategy of the drone can be adjusted to ensure that the drone maintains the best attitude during flight, thereby improving the fire extinguishing efficiency and safety. Based on the policy gradient algorithm, a fire extinguishing spraying control model is constructed. This algorithm can continuously optimize the time-varying attitude spraying strategy and adjust the spraying path and spraying amount of the drone in real time, making the fire extinguishing process more refined and intelligent. By establishing an optimized spraying control model, the fire extinguishing accuracy and effect can be greatly improved, especially in the complex high-rise building fire environment, avoiding the problem of incomplete fire extinguishing caused by human judgment errors in traditional fire extinguishing methods. Finally, this control model is sent to the terminal to ensure that the drone can perform the fire extinguishing task according to the best strategy, thereby achieving efficient fire extinguishing. Therefore, the present invention makes an optimized treatment for a traditional drone fire extinguishing control method, solves the problem that a traditional drone fire extinguishing control method has a low accuracy in controlling the flight attitude due to the unstable air flow caused by the rising hot air generated at the fire scene, thus affecting the accuracy of the fire extinguishing agent delivery, improves the accuracy of flight attitude control, and enhances the accuracy of the fire extinguishing agent delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the step flow of a drone fire extinguishing control method; Figure 2 is Figure 1 a detailed implementation step flow diagram of step S2 in; Figure 3 is Figure 1 a detailed implementation step flow diagram of step S3 in. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Please refer to Figures 1 to 3 , a drone fire extinguishing control method and system, the method includes the following steps: Step S1: Use the sensors carried on the drone to collect real-time data on the heat flow wind field at the high-rise building fire scene, obtaining real-time heat flow wind field data; perform an analysis of the intensity differences of the approaching heat flow wind field on the real-time heat flow wind field data to obtain heat flow wind field grid intensity difference data; Step S2: Quantify the numerical values of the unstable flight attitude of the drone according to the heat flow wind field grid intensity difference data, obtaining numerical quantification data of the unstable flight attitude; adjust the time-varying attitude spraying strategy according to the numerical quantification data of the unstable flight attitude, obtaining a time-varying attitude spraying adjustment strategy; Step S3: Based on the policy gradient algorithm, construct a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy, obtaining a fire extinguishing spraying control model; send the fire extinguishing spraying control model to the terminal to execute the drone fire extinguishing control method.

[0017] In the embodiment of the present invention, refer to Figure 1 As described, it is a schematic diagram of the step flow of a drone fire extinguishing control method of the present invention. In this example, the drone fire extinguishing control method includes the following steps: Step S1: Use the sensors carried on the drone to collect real-time data on the heat flow wind field at the high-rise building fire scene, obtaining real-time heat flow wind field data; perform an analysis of the intensity differences of the approaching heat flow wind field on the real-time heat flow wind field data to obtain heat flow wind field grid intensity difference data; In the embodiment of the present invention, by installing a high-resolution infrared thermal imager (such as the FLIR Tau2 series, with a thermal resolution less than 0.05 °C) and a Doppler wind speed measurement lidar module (such as the RIEGL VUX-1 series, with a ranging accuracy of ±10 mm) on the drone platform, data collection is carried out in the actual scene of a high-rise building fire. The drone slowly rises vertically from a position 10 meters away from the building facade, and collects one data frame containing heat radiation distribution and airflow vector information every 1 second, with a sampling frequency of 1 Hz, and the flight speed is stably controlled at 0.5 m / s. During the 10-minute flight sampling process, a total of 600 frames of real-time heat flow wind field data are obtained. The collected heat maps and wind speed vector fields are registered according to the time stamps, and a four-dimensional tensor reconstruction is performed on the time-series heat flow data using a space-time reconstruction method. Based on the tensor data, a multi-scale Gaussian filter is used for background heat field fitting and elimination operations, the temperature change gradient of the mutation region is extracted, and a two-dimensional heat flow perturbation field difference function is constructed in combination with the wind direction and wind speed. Use the cubic spline interpolation method to construct a discrete vector field pointing from the heat flow mutation boundary to the perturbation propagation path, perform intensity difference calculations along the anisotropic heat flow path, and reconstruct the obtained difference results in a 10 cm × 10 cm grid cell in the spatial dimension to generate heat flow wind field grid intensity difference data.

[0018] Step S2: Numerically quantify the instability of the UAV flight attitude based on the heat flow wind field grid intensity difference data to obtain the numerically quantified data of the flight attitude instability; adjust the time-varying attitude spraying strategy according to the numerically quantified data of the flight attitude instability to obtain the time-varying attitude spraying adjustment strategy. In the embodiment of the present invention, the obtained heat flow wind field grid intensity difference data is input into the disturbing force analysis unit, and the disturbing flow field is reconstructed by using the disturbing flow force domain reconstruction algorithm based on the prediction of the vortex viscosity number. In this method, the heat disturbance intensity of the grid points is used as the leading variable, and the local disturbing force matrix of each grid point is calculated. The disturbing force matrix is six-dimensional vector field data, which respectively represent the lateral, longitudinal, and vertical aerodynamic disturbing forces and the disturbing force moments corresponding to the three axial directions. Further, based on the disturbing flow force domain data, the principal component analysis method is used to extract the first three principal disturbing modes and calculate their spatial variance ratios to obtain the spatial heterogeneity data of the disturbing flow distribution. The spatial heterogeneity data of this disturbing flow distribution is combined with the components of the disturbing flow force domain (where the components of the disturbing flow force domain refer to the six components in the above-mentioned six-dimensional vector field data, specifically including the lateral aerodynamic disturbing force moment, the longitudinal aerodynamic disturbing force moment, the vertical aerodynamic disturbing force moment, the disturbing force moment around the transverse axis, the disturbing force moment around the longitudinal axis, and the disturbing force moment around the vertical axis. Among them, the longitudinal aerodynamic disturbing force moment refers to the disturbing force along the front-back axis of the UAV, usually the disturbing force in the fuselage length direction; the vertical aerodynamic disturbing force refers to the disturbing force along the direction perpendicular to the ground, usually the disturbing force in the up-down direction), and the Fourier transform is used to obtain the disturbing force spectral density in the horizontal and vertical directions. The disturbing force spectral density is an energy distribution function obtained by converting the disturbing forces in the horizontal and vertical directions (including longitudinal and vertical directions in the vertical direction) in the time domain to the frequency domain, which represents the proportion or intensity of the disturbing components of different frequencies in the overall disturbance. The disturbance frequency domain range of the current disturbance area where the UAV is located is calculated through the spectral density function, and combined with the actual structural parameters of the UAV, its rotational response is simulated within the disturbance frequency domain range, and the disordered swing intensity in the roll, pitch, and yaw directions is analyzed. The intensity data is decomposed by the wavelet packet energy density to quantitatively obtain the attitude instability risk index in the rotational direction of the UAV. After being normalized in combination with the historical stability data of the UAV flight attitude, the numerically quantified data of the flight attitude instability is obtained, and the numerical range is limited to [0,1]. Subsequently, based on this instability quantification data, the current flight ceiling limit interval is extracted through the dynamic ceiling response model. The input of this model is the ratio of the attitude angle change rate to the disturbing force moment limit, and the output is the attitude ceiling margin. By constructing the ceiling amplitude difference function and the time-series jitter angle error distribution function, a flight time-series attitude control sequence is generated, and a control path function in polynomial form is constructed by integrating the predicted value of the wind direction change trend (given by the recursive convolutional neural network model). Finally, with the current flight attitude, the control path function, and the material spraying response threshold (such as the nozzle opening threshold pressure of 0.2 MPa) as the constraint conditions, a time-varying attitude spraying adjustment strategy for adjusting the attitude and spraying synchronization control is generated.

[0019] Step S3: Based on the policy gradient algorithm, construct a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy to obtain the fire extinguishing spraying control model; send the fire extinguishing spraying control model to the terminal to execute the UAV fire extinguishing control method.

[0020] In the embodiment of the present invention, the time-varying attitude spraying adjustment strategy output in step S2 is standardized. The min-max normalization method is used to uniformly map each control variable (attitude angle control value, spraying pressure control value, direction adjustment amplitude, etc.) to the interval [0,1][0,1][0,1], and the normalized data sequence is used as the policy representation vector and input into the policy learning module. The policy learning module uses the policy gradient optimization method in the temporal difference method to construct a neural network estimator to iteratively update the gradient relationship between the current policy and the fire extinguishing efficiency feedback. The selected policy gradient algorithm is a variant form of Proximal Policy Optimization (PPO). The neural network structure is a two-layer fully connected structure, with 128 nodes in the first layer and 64 nodes in the second layer. The activation function uses ReLU, and the loss function is the expected policy return difference function. The fire extinguishing efficiency feedback takes the temperature drop value per unit time and the fire extinguishing spraying area as inputs, and calculates the immediate reward signal using a weighted linear combination. During the training iteration process, every time a policy gradient update is completed, evaluate the coverage of the current control policy for spraying in the high-temperature area and the average fire suppression time in the simulation environment, and compare it with the historical average performance. When the current policy control performs better than the average performance policy of the previous five times within ten rounds, it is determined as the optimal policy and output as the fire extinguishing spraying control model. Finally, the constructed fire extinguishing spraying control model is uploaded to the UAV control terminal in sequence form, and the model is applied to the trigger module of the real-time attitude control and spraying system synchronization instruction through the terminal scheduling subsystem to achieve the precise execution of the fire extinguishing task.

[0021] Step S1 includes the following steps: Step S11: Collect real-time data of the heat flow wind field at the high-rise building fire site through the sensors carried on the UAV to obtain real-time data of the heat flow wind field; Step S12: Fill in the missing values in the real-time data of the heat flow wind field to obtain real-time filled data of the heat flow wind field; Step S13: Conduct an analysis of the approaching heat flow wind field intensity difference on the real-time filled data of the heat flow wind field to obtain approaching intensity difference data of the heat flow wind field; Step S14: Convert the approaching intensity difference data of the heat flow wind field into a grid structure to obtain grid intensity difference data of the heat flow wind field.

[0022] In the embodiments of the present invention, a quadcopter UAV platform equipped with a FLIR Tau2 infrared thermal imager (thermal sensitivity better than 0.05 °C) and a RIEGL miniVUX-1UAV lidar (ranging error less than ±15 mm, scanning frequency of 100 kHz) is selected to perform real-time data acquisition of the heat flux wind field in the fire area on the outer facade of high-rise buildings. The UAV takes off from a position 6 meters away from the building surface at the bottom of the building and slowly ascends vertically. The flight speed is controlled at 0.3 m / s. The radar echo sampling period is 10 ms, and the infrared image frame rate is 10 Hz. During the 10-minute sampling period, a total of 6000 frames of thermal map data and 36000 sets of wind speed and direction vector point cloud data are collected. All sensor data are aligned with timestamps. High-frequency noise is eliminated by a low-pass filter. The infrared images are enhanced with thermal gradient edge features through cubic interpolation, and the wind speed point cloud is densified through spherical linear interpolation. Finally, a real-time data triple of the heat flux wind field is generated, representing temperature, velocity vector, and wind direction angle respectively. The missing value data points in the above real-time heat flux wind field data caused by occlusion, signal interruption, or strong light interference are filled. First, the missing points in each frame of image or point cloud data are detected. The missing judgment criterion is that the sampling value of any thermal image or wind vector point is empty or there is an abnormal point with a numerical drift exceeding 5σ. The Kriging spatial interpolation method is used for data filling. The Kriging method constructs an optimal unbiased estimator based on the spatial covariance matrix, calculates the data covariance of five known points in the neighborhood, constructs a weight matrix, and calculates the value of the point to be estimated. To improve time consistency, a time sliding window constraint is added to construct a spatio-temporal joint covariance function, and the data at adjacent times are jointly involved in the estimation process. After filling, the mean deviation of all filled points is checked, and the points with a deviation exceeding 3 °C or a wind speed difference exceeding 0.8 m / s are filled again. Finally, the complete real-time filled data of the heat flux wind field after filling is output. The difference analysis of the heat flux wind field intensity is carried out for the real-time filled data of the heat flux wind field. Taking the heat flux data in the area where the UAV is 2 m to 4 m away from the building surface as the extraction object, first, a continuous sequence of thermal image frames in this area is selected. The temperature gradient and the wind speed direction gradient are calculated for each frame of image. The Sobel operator is used for local gradient extraction, and a 3×3 convolution kernel is introduced for high-frequency feature enhancement. Subsequently, a heat flux perturbation tensor is constructed. The gradient vector in each frame is subtracted from the gradient difference at the corresponding position in the adjacent frame to obtain the perturbation change intensity field. Through the main direction statistics and distribution analysis, the thermal perturbation direction volatility and the air flow direction mutation amplitude are calculated, and the average perturbation intensity estimation is performed on multiple frame sequences. The perturbation intensity is defined as the product of the change rate of the temperature vector per unit time and the change angle of the wind direction mutation, and then the difference data of the heat flux wind field approaching intensity is obtained. The grid structure conversion operation is performed on the obtained difference data of the heat flux wind field approaching intensity. The conversion target is to construct a spatial intensity difference distribution grid under a fixed structure for subsequent quantitative analysis and strategy input.First, set the spatial grid resolution to 10 cm × 10 cm. Construct an equally spaced grid index matrix in the original three-dimensional space coordinates, and map all intensity values to the corresponding grid centers according to the spatial positions of their original sampling points. The inverse distance weighted (IDW) interpolation method is used to unify the scattered point perturbation values within each grid into the representative value of the grid center point. The IDW weight exponent is set to 2, and the neighborhood is set to all valid perturbation value points within a radius of 0.2 m. After completing the mapping of all time frames, the final output is the heat flux wind field grid intensity difference data.

[0023] Step S2 includes the following steps: Step S21: Analyze the turbulent force domain of the heat flux wind field grid intensity difference data to obtain the heat flux wind field turbulent force domain data; Step S22: Numerically quantify the instability of the UAV flight attitude based on the heat flux wind field turbulent force domain data to obtain the numerical quantification data of the flight attitude instability; Step S23: Analyze the lift constraint amplitude based on the numerical quantification data of the flight attitude instability to obtain the flight lift constraint amplitude data; Step S24: Adjust the time-varying attitude spraying strategy according to the numerical quantification data of the flight attitude instability, the flight lift constraint amplitude data, and the heat flux wind field turbulent force domain data to obtain the time-varying attitude spraying adjustment strategy.

[0024] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: Analyze the turbulent force domain of the heat flux wind field grid intensity difference data to obtain the heat flux wind field turbulent force domain data; In the embodiment of the present invention, in the process of analyzing the turbulent force domain of the heat flux wind field grid intensity difference data, a three-dimensional heat flux perturbation tensor analysis method is adopted. First, map the temperature field data obtained by the infrared thermal imaging sensor to a three-dimensional grid structure in the Cartesian coordinate system centered on the UAV flight trajectory. The grid resolution is set to 20 cm per side. The temperature gradient field within the unit grid is constructed by the bilinear interpolation method, and then the heat flux velocity field is converted from the temperature gradient and combined with the local density field to calculate the distribution of the perturbation force using the momentum conservation theorem. In this process, the UAV flight speed is set to 6 m per second, the sampling frequency is 20 Hz, and the three-axis components of each perturbation force vector are integrated by the control volume integral method to obtain the magnitude of the thermal dynamic perturbation within the unit volume. The turbulent force domain tensor distribution map within the current flight area is obtained by superimposing the three-axis perturbation forces. The tensor records the perturbation direction, amplitude, and its action radius centered on the grid point. The perturbation direction represents the angle in spherical coordinates, the amplitude unit is Newton, and the action radius is calculated based on the radius of the heat flux isointensity line as the reference length scale.

[0025] Step S22: Numerically quantify the instability of the UAV flight attitude based on the thermal flow wind field turbulence force domain data to obtain the numerically quantified data of the flight attitude instability; In the embodiment of the present invention, during the process of numerically quantifying the instability of the UAV flight attitude according to the turbulence force domain data, a six-degree-of-freedom rigid body kinematic analysis method is adopted. The offset trends of the UAV in the roll, pitch, yaw, and three-axis linear displacement directions after being affected by the disturbing force are integrally estimated respectively. The Kalman Filter is used to fuse the data of the aircraft inertial measurement unit (IMU) and the disturbance tensor data. The filter parameters are set as follows: the diagonal elements of the state transition noise covariance matrix are 10^-3, and the diagonal elements of the observation noise covariance matrix are 10^-2. The disturbing force is converted into linear acceleration in the aircraft centroid coordinate system and the Euler angle increment integration method is introduced to calculate its influence on the attitude change rate. Further, the attitude change rates in each direction are compared with the maximum stable attitude change threshold of the aircraft structure. The pitch angle limit is set to 15 degrees per second, the roll angle limit is set to 20 degrees per second, and the yaw angle limit is set to 10 degrees per second. The quantization output result represents the strength of the instability tendency in the form of a percentage, and the data is saved in a numerical matrix in the three-axis direction. The matrix takes the time series as the horizontal axis and the ratio of the attitude change rate to the threshold as the vertical axis.

[0026] Step S23: Analyze the lift constraint amplitude based on the numerically quantified data of the flight attitude instability to obtain the flight lift constraint amplitude data; In the embodiment of the present invention, during the process of analyzing the lift constraint amplitude based on the numerically quantified data of the flight attitude instability, the parameters of the UAV lift generating components, flight speed, angle of attack change rate, and mass distribution are introduced into the lift analysis model. The difference between the actual generated lift and the designed maximum lift under the disturbance condition is estimated through the lift non-linear response function. The total mass of the aircraft is set to 3 kg, the rotor diameter is set to 0.45 m, and the maximum output power is set to 850 W. The change trend of the aerodynamic coefficient in the lift equation is inversely calculated to analyze the change of the lift margin under the disturbance. The one-dimensional regression analysis method is used to fit the relationship between the change of the turbulence-induced air flow velocity and the lift response delay. The flight lift constraint amplitude data is output, and the lift loss ratio at each time point is recorded in Newtons and sorted according to the disturbance intensity in the upward direction. The lift constraint peaks in the three main disturbance directions are retained, and the output structure is a three-dimensional vector array, where each element contains a time stamp, a percentage of the lift margin, and an identification number of the main disturbance direction.

[0027] Step S24: Adjust the time-varying attitude spraying strategy according to the numerically quantified data of the flight attitude instability, the flight lift constraint amplitude data, and the thermal flow wind field turbulence force domain data to obtain the time-varying attitude spraying adjustment strategy.

[0028] In the embodiment of the present invention, during the process of adjusting the time-varying attitude spraying strategy according to the flight attitude instability numerical quantization data, the flight lift limit amplitude data, and the heat flow wind field turbulence force domain data, a weighted shortest path search mechanism in the multi-objective optimization method is used to construct a real-time spraying angle adjustment path. The historical spraying path and the current attitude change path are fitted by a three-dimensional B-spline function, and the control amount of the spraying angle is adjusted with the turbulence direction as a constraint. The spraying radius is set to 1.2 meters, the spraying flow rate is 0.8 liters per second, and the spraying angle range is between 45 degrees and 90 degrees. The spraying angle is preferentially adjusted in the opposite direction of the turbulence source according to the turbulence direction. The spraying angle is discretely adjusted in combination with the attitude change rate and the lift amplitude change rate. The three-point difference method is used to calculate the coverage probability of the spraying fan-shaped area corresponding to each adjustment step, and then the next spraying direction is adjusted based on the average effect of the spraying coverage area on the heat flow disturbance of the fire source. The spraying direction and amplitude are refreshed once per second through a loop structure, and finally a set of time series spraying parameter matrices are formed. The matrix contains data on the time point, the spraying angle, the three-axis vector of the spraying direction, and the spraying flow rate per unit time. The response update period of the spraying strategy adjustment process is 10 milliseconds, ensuring that the system has real-time feedback capabilities.

[0029] Step S22 includes the following steps: Step S221: Perform spatial heterogeneity analysis on the heat flow wind field turbulence force domain data to obtain turbulence force domain spatial heterogeneity data; Step S222: Based on the turbulence force domain spatial heterogeneity data and the heat flow wind field turbulence force domain data, perform horizontal / vertical azimuth turbulence power spectral density analysis to obtain the horizontal / vertical azimuth turbulence power spectral density; Step S223: According to the horizontal / vertical azimuth turbulence power spectral density, perform analysis on the disordered changes of the UAV's rotation and swing to obtain the disordered change data of the rotation and swing; Step S224: Perform inclined moment fluctuation regression analysis on the disordered change data of the rotation and swing to obtain the inclined moment fluctuation regression data; Step S225: Based on the disordered change data of the rotation and swing, perform lateral airflow pressure rollover limit analysis to obtain the lateral airflow rollover limit pressure data; Step S226: According to the inclined moment fluctuation regression data and the lateral airflow rollover limit pressure data, perform numerical quantization of the UAV's flight attitude instability to obtain the flight attitude instability numerical quantization data.

[0030] In the embodiments of the present invention, when performing spatial heterogeneity analysis on the data of the turbulent flow force domain of the heat flow wind field, a hierarchical heterogeneity index matrix is constructed based on the three-dimensional grid dataset of the turbulent flow force field. First, the turbulent flow force field data is divided into three dimensions by 3-meter cube units. Each cube contains the distribution of turbulent flow force vectors and the parameters of heat flow gradient changes. Then, the coefficient of variation is used to statistically analyze the magnitude of the turbulent flow force of each spatial unit, and the spatial heterogeneity index is calculated for all units in sequence. The threshold CV > 0.4 is set as the discrimination criterion for high heterogeneity regions. Then, a spatial weighted model based on clustering is used to perform aggregation analysis on the same type of turbulent flow regions. The DBSCAN density clustering method is used to set the minimum number of samples to 5 and the radius distance to 2.5 meters. Finally, the spatial heterogeneity data of the turbulent flow force domain is generated, and the output format is a GeoTIFF file, accompanied by the average turbulent flow intensity, heterogeneity index, and central coordinates of each clustering block, which are used for subsequent spectral density resolution processes. When performing horizontal and vertical azimuth turbulent flow force spectral density analysis based on the spatial heterogeneity data of the turbulent flow force domain and the data of the turbulent flow force domain of the heat flow wind field, the turbulent flow vectors in the high heterogeneity regions are selected as the input signals for spectral analysis. For the horizontal direction, the X-axis turbulent flow component is selected, and for the vertical direction, the Z-axis turbulent flow component is selected. The length of the turbulent flow time series in each direction is set to 1024 points, and the sampling interval is 0.1 second. The fast Fourier transform (FFT) method is used for spectral decomposition, and each segment of the signal is processed with a Hanning window to avoid edge spectral leakage. Then, through the power spectral density (PSD) calculation module, the energy distribution of the turbulent flow force in each frequency component is output. The spectral interval is divided into 0.1 Hz to 5 Hz, with each 0.1 Hz as a band. The main frequency peaks and energy weights in each frequency band are statistically analyzed. Finally, the horizontal / vertical azimuth turbulent flow force spectral density data is generated, and the storage format is a two-dimensional matrix diagram, accompanied by the turbulent flow intensity corresponding to the main frequency point and the frequency band index value. When performing the analysis of the disordered changes in the rotation and swing of the unmanned aerial vehicle based on the horizontal and vertical azimuth turbulent flow force spectral density, the spectral density data is registered with the time series of the attitude changes of the unmanned aerial vehicle. The wavelet decomposition of the attitude angular rate changes in the corresponding time period is performed, and the Daubechies 6th-order wavelet basis is used for multi-scale analysis. The disturbance components above 0.5 Hz in the high-frequency coefficients are extracted as the response data of the swing disorder. Then, covariance analysis is combined with the frequency band where the main energy is concentrated in the PSD spectrum. When the energy concentration degree of the turbulent flow frequency band exceeds 60%, and at the same time, the variance of the high-frequency components of the attitude angular velocity exceeds the threshold of 0.3, it is defined as the section where the disordered rotation and swing occur. The maximum angular velocity change amount and the duration in this section are statistically analyzed, and the disordered change data of the rotation and swing is integrated and output in CSV format. Each row records the start and end times of the time period, the main disturbance frequency, the maximum attitude angular velocity change value, and the direction identifier, which are used for subsequent tilt moment analysis.When performing regression analysis on the tilt moment fluctuations of rotation and swing disorderly change data, the physical structure parameters of the aircraft body and the output model of the power system are called. The inputs include conventional mechanical parameters such as the center of mass position, inertia matrix, and motor thrust coefficient. Combining the attitude angle changes in the rotation and swing data, an estimation model of the tilt moment with pitch and roll angles as variables is constructed. A machine learning model based on Gradient Boosted Regression Tree is used to train the historical attitude disturbance and motor output response data. The learning rate is set to 0.1, and the number of iterations is 100 rounds. The goal is to fit the non-linear mapping relationship between the attitude angle changes and the actual moment output. Finally, the tilt moment fluctuation trend within each disorderly change segment is regressed, and the tilt moment fluctuation regression data is output, including the slope of the regression fitting curve, the average residual, and the error variance. The results are used in the subsequent pressure limit assessment and analysis. When performing the lateral airflow pressure rollover limit analysis based on the rotation and swing disorderly change data, the flight segments containing attitude angle mutation events are selected. The lateral disturbance acceleration and rotational inertia at each moment are calculated. The width of the UAV body, the center of gravity offset, and the motor layout radius are input. The rollover moment value generated by the lateral airflow is derived using the dynamic model. At the same time, the aerodynamic simulation module is called to simulate the influence of the wind field change on the lateral stability of the aircraft body. The threshold of the limit rollover angular velocity is set to 20 degrees per second. It is analyzed whether the rollover speed exceeds this threshold in the area where the disturbance intensity exceeds 200 Pa. When it continuously exceeds for more than 3 seconds, it is determined as the rollover critical state. The lateral airflow rollover limit pressure data is output, including the trigger time of each rollover, the maximum disturbance pressure value, the angular velocity curve graph, and the rollover duration, which is used in the instability quantification link. When performing the numerical quantification of the UAV flight attitude instability according to the tilt moment fluctuation regression data and the lateral airflow rollover limit pressure data, an attitude instability discrimination model is established. The input parameters include the change rate of the attitude angular velocity, the slope of the tilt moment fluctuation, the main frequency of the disturbance flow, the extreme value of the rollover angular velocity, etc. The principal component analysis method is used to reduce the dimension of all features, and the components with the top 95% contribution rate are retained for classification modeling. The support vector machine (SVM) algorithm is used to construct an instability classifier. The training set uses 1000 labeled flight instability data segments. The kernel function is set to the radial basis function (RBF), and the penalty coefficient C is set to 10. Finally, the model outputs the instability probability value for each time period and outputs the numerical quantification data of the flight attitude instability according to the probability distribution, including the time segment index, the instability probability value, the weight of the main control factor, and the model score, which is used for the subsequent attitude spraying adjustment strategy.

[0031] Step S223 includes the following steps: Obtain the aerodynamic layout of the UAV; calculate the airflow disturbance time series pressure difference of the horizontal / vertical azimuth disturbance power spectral density to obtain the airflow disturbance time series pressure difference between the horizontal / vertical azimuths; Analyze the centroid repeated offset trajectory of the aerodynamic layout of the UAV according to the time-sequence pressure difference of the airflow disturbance to obtain the repeated offset trajectory of the flight centroid; Calculate the average difference of the relative angles of the horizontal offset of the repeated offset trajectory of the flight centroid to obtain the average difference of the relative angles of the centroid offset; Analyze the disordered changes of the rotation and swing of the UAV based on the average difference of the relative angles of the centroid offset and the repeated offset trajectory of the flight centroid to obtain the disordered change data of the rotation and swing.

[0032] In the embodiment of the present invention, during the acquisition of the aerodynamic layout of the UAV, first, the aircraft three-dimensional modeling system is called to initialize and input the physical structure parameters of the UAV. The input parameters include the length of the UAV's arms, the motors are distributed at the four corners of the fuselage, the propeller diameter is 9 inches (about 22.86 cm), the battery is installed below the central symmetry axis, the payload is mounted at the center of the fuselage belly, the total mass of the whole machine is 2.6 kg, and the payload mass is 0.6 kg. The parameter modeling module is used to output data including the center of gravity position, the spatial coordinates of each motor relative to the center of gravity, the power distribution matrix, the moment of inertia matrix, the damping coefficients in the pitch and yaw directions, and the lift model coefficients. Finally, a complete aerodynamic layout data file is generated and stored in JSON format as the structural basis for subsequent disturbance simulation and center of gravity offset analysis. When calculating the time-series pressure difference of the airflow disturbance for the horizontal and vertical azimuth disturbance power spectral densities, the disturbance force data on the X-axis and Z-axis in the frequency spectrum are extracted respectively. The disturbance force density is inversely transformed in the time dimension to obtain the distribution function of the disturbance force in the time domain. The change rate of the disturbance force is calculated using the numerical differentiation method, and the influence of the disturbance force per unit area is converted into the form of pressure difference. The area weighted average algorithm is used to normalize the disturbances in different regions. The effective disturbance value is set when the change rate of the disturbance force density exceeds 12 Pa per second. The time-series of the disturbance pressure difference in the horizontal and vertical directions are constructed respectively, and the two series are analyzed side by side. A three-dimensional tensor matrix is constructed to represent the synthesis difference of the disturbances at different time points. The output format is a time-series array sampled 20 times per second, and each group of data contains the horizontal disturbance value, the vertical disturbance value, and the instantaneous difference. When analyzing the repeated center of gravity offset trajectory of the UAV's aerodynamic layout according to the time-series pressure difference of the airflow disturbance, first, the positions of each motor, the payload position, and the center of gravity coordinates in the aerodynamic layout structure are input into the simulation engine. The time-series pressure difference of the disturbance is superimposed and the instantaneous torque generated by the disturbance is calculated. The dynamic model based on the Newton-Euler equation is used to convert the torque generated at each disturbance moment into the center of gravity offset response. The relative displacement of the UAV's center of gravity in the three-dimensional space under each disturbance is recorded to generate the repeated center of gravity offset trajectory data. The time window is 60 seconds, and 20 data are recorded per second, for a total of 1200 groups of data. Each group of data contains the offset values of the center of gravity in the X, Y, and Z directions, as well as the corresponding disturbance force vector and time stamp. The output is in CSV format and a three-dimensional offset trajectory curve graph is plotted as the visualization basis.When calculating the average relative angle difference of horizontal offset for the repeated offset trajectory of the flight center of gravity, extract the X-axis and Y-axis center of gravity offset components at all time points, calculate the change in the angle between the offset vectors on the horizontal plane respectively, unify the angle unit to radians, calculate the change in the angle within every two adjacent seconds once, calculate the root mean square of all the differences between adjacent offset angles, set the continuous angle difference change greater than 0.15 radians as the significant disturbance identifier, obtain the average relative angle change data through full-window traversal analysis, and finally output the average relative angle difference of the center of gravity offset. The output form is a single-value index, accompanied by the angle variance and the maximum offset angle change information, which is used to identify the occurrence frequency and the angle fluctuation range of the rotational swing behavior. When analyzing the disordered change of the UAV's rotational swing based on the average relative angle difference of the center of gravity offset and the repeated offset trajectory of the flight center of gravity, use the above two data as input features to construct a rotational disorder recognition model. The model adopts a fast classification method based on the Extreme Learning Machine. The dimension of the input layer is 4, including the standard deviation of the X-axis offset, the standard deviation of the Y-axis offset, the average relative angle difference, and the maximum instantaneous offset value. The output is a binary classification result to judge whether it enters the disordered swing state. Use the pre-constructed labeled sample set for model training. The sample size is 10,000 flight data segments. When training, set the activation function as the sigmoid function and the number of hidden layer nodes as 80. After the model training is completed, detect the current flight segment second by second, mark all the time periods of disordered changes and record the duration of each segment, the maximum angular velocity change, and the change trend of the offset angle. Finally, form the disordered change data of the rotational swing and output it in the form of a structured time series for the next-stage tilt moment fluctuation regression analysis and processing operation.

[0033] Step S24 includes the following steps: Step S241: Obtain the fire spraying material data carried by the UAV; Step S242: Conduct an analysis of the attitude jitter amplitude interval for the numerically quantified data of the flight attitude instability to obtain the attitude jitter amplitude interval; Step S243: Adjust the flight time-series attitude control according to the attitude jitter amplitude interval and the flight lift constraint amplitude data to obtain the flight time-series attitude control data; Step S244: Control the spraying pressure of the fire spraying material data according to the attitude jitter amplitude interval and the heat flow wind field turbulence force domain data to obtain the flight spraying pressure control data; Step S245: Adjust the time-varying attitude spraying strategy based on the flight time-series attitude control data and the flight spraying pressure control data to obtain the time-varying attitude spraying adjustment strategy.

[0034] In the embodiments of the present invention, fire sprinkler material data is obtained from fire sprinkler material manufacturers. Additionally, a high-precision weighing module, a liquid volume sensor, and a pressure monitoring chip integrated on the payload platform at the tail of the drone are used to collect multi-dimensional data on the status of the sprinkler system. The system uses a modular liquid tank container made of carbon fiber material, with a tank volume of 3.0 liters. The tank is filled with an ammonium phosphate foam fire extinguishing agent that has passed high-temperature stability tests. The viscosity of the fire extinguishing agent is 132 mPa·s, and the operating temperature range is from -10 degrees Celsius to 60 degrees Celsius. The initial pressure in the tank is set to 3.2 MPa. The drive mechanism used is a combined structure of an electronically controlled micro-turbine pump and a high-frequency solenoid valve. An embedded control chip is used to collect the remaining mass of the fire extinguishing agent as 1.82 kg, and it outputs the current injection pressure as 2.7 MPa, the injection diameter as 6 mm, the measured nozzle outlet flow rate as 360 ml per second, and the nozzle angle as 15 degrees to the left. When analyzing the attitude jitter amplitude interval of the numerically quantified data of the flight attitude instability, the time series of the three-axis angular velocity change curves in the quantified data is divided. Each 5 seconds is set as an analysis period. The maximum amplitude extraction and mean deviation calculation of the angular velocity fluctuation values within the period are performed to determine whether they exceed the reference threshold. It is set that when the attitude change amplitude of the X-axis exceeds 8 degrees per second, the attitude change amplitude of the Y-axis exceeds 6 degrees per second, and the yaw change amplitude of the Z-axis exceeds 5 degrees per second, it is marked as unstable. The fluctuation amplitude interval is calculated based on the difference between the maximum and minimum fluctuations within each period. The fluctuation interval values within all periods are arranged along the time axis to generate an attitude jitter amplitude interval sequence. The structure includes multiple fields such as the start timestamp of each period, the upper and lower limits of the fluctuation, the fluctuation mean, and the fluctuation trend direction, and is output as a multi-dimensional matrix structure for use in the subsequent flight time series attitude control adjustment stage. When performing flight time series attitude control adjustment based on the attitude jitter amplitude interval and the flight lift constraint amplitude data, the aforementioned attitude fluctuation interval data is feature-matched with the lift constraint amplitude data. The lift constraint amplitude data includes the change range of the drone's lift margin, the maximum safe deflection angle of the propulsion direction, and the attitude feedback delay time. Among them, the lift margin is 1.3 Newtons, the maximum safe deflection angle is 12 degrees, and the attitude feedback delay is 160 milliseconds. A control adjustment mapping model is constructed to adjust the flight control instruction parameters for each attitude jitter period. A flight control adjustment module based on a stabilized PID (Proportional Integral Derivative) structure is used to set dynamic gain factors for the pitch, roll, and yaw directions respectively. When the current fluctuation exceeds the threshold, the proportion of the D term is increased, and the maximum gain adjustment amplitude is set to 1.8 times the original value. At the same time, the response time is compensated. The refresh frequency of the control output within each attitude period is adjusted through a frequency modulation control method, and the default refresh frequency is increased from 50 Hz to 80 Hz to enhance the response ability. Finally, a flight attitude control data set controlled by the time axis is formed.When controlling the spraying pressure of fire - fighting spraying material data according to the attitude jitter amplitude range and the turbulent flow force domain data of the heat flux wind field, first project the wind speed vector in the turbulent flow force domain data onto a three - dimensional space, extract the wind force vector with an angle relative to the nozzle direction less than 45 degrees, accumulate and fit its instantaneous wind pressure, and extract the average counter - pressure per unit area, which is 2.1 kPa. At the same time, synchronize the attitude fluctuation amplitude to the spraying control beat, construct a spraying pressure adjustment formula based on a two - way pressure - regulating model, calculate that the target injection pressure correction value is a 15% increase from the original set value, increase the output voltage of the nozzle electric control pump from 7.4 V to 9.1 V, drive the injection flow rate to increase from 300 ml per second to 390 ml per second. Detect the change of the injection pressure through real - time closed - loop feedback. If it exceeds the target by ±0.3 MPa, automatically adjust the opening and closing angle of the nozzle valve, correct the spraying angle to 15 degrees in the direction of the opposite region of the disturbance, and compress the nozzle cone angle from the default 45 degrees to 35 degrees to improve the beam stability. Finally, generate a flight spraying pressure control data file, including multi - dimensional data such as injection pressure, spraying flow rate, nozzle direction, and real - time wind resistance vector. When adjusting the time - varying attitude spraying strategy based on the flight - time - series attitude control data and the flight spraying pressure control data, input the two types of data into a dynamic strategy matching engine. This engine uses a multi - objective optimization algorithm to coordinately adjust the attitude response efficiency and the spraying accuracy coverage rate. Set the flight attitude control weight to 0.65 and the spraying coverage accuracy weight to 0.35 using the weight ratio control parameter. The matching strategy model outputs the preset attitude adjustment angle, nozzle direction deflection angle, and spraying throttle rate within each flight cycle. The output structure is a five - dimensional control vector, which respectively represents the pitch adjustment amplitude, roll correction deviation, nozzle offset angle, spraying flow rate adjustment ratio, and spraying cycle refresh frequency. This strategy is sent down to the flight control system execution module, and the flight control system makes a directional correction for the current cycle. Finally, form a complete time - varying attitude spraying adjustment strategy data sequence, which is used to guide the UAV to synchronously operate the dynamic response and spraying parameter control under different turbulent flow states and flight postures during the fire - fighting operation.

[0035] Step S243 includes the following steps: Analyze the jitter lateral tilt angle of the attitude jitter amplitude range to obtain the jitter lateral tilt angle data; Conduct a ceiling attitude angle margin regression analysis based on the jitter lateral tilt angle data and the flight ceiling constraint amplitude data to obtain the ceiling attitude angle margin regression data; Conduct an axial torque output matching according to the jitter lateral tilt angle data and the flight ceiling constraint amplitude data to obtain the axial torque output matching data; Conduct a flight - time - series attitude control adjustment according to the ceiling attitude angle margin regression data and the axial torque output matching data to obtain the flight - time - series attitude control data.

[0036] In the embodiments of the present invention, during the analysis of the jitter lateral tilt angle in the jitter amplitude interval, continuous attitude sensor angle time series are used to extract statistical envelopes. The original data source is the pitch angle, roll angle, and yaw angle output by the triaxial gyroscope and attitude solution unit carried by the unmanned aerial vehicle. The sampling frequency is 100 Hz. Every 100 samples form a frame of data. The envelope amplitudes of the roll angle and pitch angle are calculated, and the maximum lateral tilt angle change value in the jitter period is extracted therefrom. It is set that the roll angle change amplitude is greater than 5 degrees and the change period is less than 0.8 seconds as an obvious jitter feature. The lateral tilt angle is defined as the angle between the combined direction of the roll angle and pitch angle and the gravity direction, and is calculated using the inverse cosine function. The output format includes time index, maximum roll angle value, duration, angle mutation rate, and the identification of the attitude jitter interval. Each jitter sequence is registered in the time domain with the wind field disturbance sequence and numbered for subsequent calls. When performing ceiling attitude angle margin regression analysis based on the jitter lateral tilt angle data and flight ceiling constraint amplitude data, the support vector regression method (SVR, support vector regression model) is used to fit and train the relationship between the roll angle and the ceiling constraint amplitude. The maximum lift tolerance angle and instantaneous lift direction offset angle included in the ceiling constraint amplitude are used as the output targets, and the roll angle, angle change rate, and jitter duration are used as input features. The training set size is set to 3000 historical data. The radial basis kernel function is selected for modeling, the penalty parameter C is adjusted to 10, and the loss tolerance ε is 0.01. After training, sliding window prediction is performed on the existing data set. Each window contains 5 seconds of continuous attitude data to obtain the predicted value of the attitude angle margin. The prediction result is marked as a valid fitting when the sum of squared residuals is less than 0.05. The output format includes the current roll angle, predicted ceiling margin, lift direction angle tolerance, and model fitting score. When performing axial torque output matching operation based on the jitter lateral tilt angle data and flight ceiling constraint amplitude data, first, the first-order differential of the attitude angle change output by the attitude solution unit is taken to obtain the attitude angular velocity change rate as the dynamic input variable. Then, combined with the maximum allowable lift change rate and attitude angle response ratio corresponding to each attitude provided in the ceiling constraint amplitude data, an axial expected torque output curve is constructed. By establishing a linear regression relationship between the attitude angle change rate and the motor output torque, the fitting slope is obtained as the matching gain factor. The motor response time delay is set to 0.15 seconds, and the maximum controllable torque is 0.28 N·m. The target torque value is calculated for each attitude angle change segment and the difference is matched with the actual motor output value. When the difference exceeds 0.03 N·m, it is marked as a torque output mismatch section, and the time index, mismatch peak, attitude angle change speed, and required gain coefficient of this section are recorded, and the axial torque output matching data is output.During the flight time-sequence attitude control adjustment process based on the regression data of ceiling attitude angle margin and the matching data of axial moment output, the model predictive control method (Model Predictive Control) is used to optimize and adjust the flight attitude changes within the next 5-second time window. The control variable is the motor torque output distribution within each 0.1-second time period. The system dynamic model is discretized using the six-degree-of-freedom flight dynamics equation. The state variables include attitude angle, attitude angular velocity, and torque distribution ratio. The constraint conditions are provided by the maximum allowable angle change and the axial torque response range in the regression data of ceiling attitude angle margin. The optimization objective function is set to minimize the weighted sum of the attitude jitter amplitude and the torque response error, and the weight ratio is set to 0.6 and 0.4. The solver uses the interior point method for iterative convergence accuracy. Finally, the output flight time-sequence attitude control adjustment data format includes control time index, motor torque adjustment value, target attitude angle change curve, maximum allowable offset angle, and attitude control error curve, forming an executable attitude time-sequence control signal sequence for driving the control unit to perform real-time adjustment.

[0037] Step S244 includes the following steps: Perform spatial vector decomposition processing on the heat flow wind field turbulence force domain data to obtain turbulence direction component data; Perform particle size distribution and viscosity characteristic identification processing on the fire sprinkling material data to obtain sprinkling material particle size distribution data and sprinkling material viscosity characteristic data respectively; Based on the turbulence direction component data and the attitude jitter amplitude interval, simulate and evaluate the kinetic energy fluctuation interval of the sprinkling air flow barrier to obtain the kinetic energy fluctuation interval of the sprinkling air flow barrier; Perform approximate integration of the fluidity resistance fluctuation on the kinetic energy fluctuation interval of the sprinkling air flow barrier to obtain approximate data of the fluidity resistance fluctuation; According to the approximate data of the fluidity resistance fluctuation, control the sprinkling pressure for the sprinkling material particle size distribution data and the sprinkling material viscosity characteristic data to obtain flight sprinkling pressure control data.

[0038] In the embodiments of the present invention, during the process of spatially vector decomposing the data of the turbulent force domain of the heat flow wind field, a turbulent force direction decomposition matrix is constructed by using the included angle between the three-dimensional wind speed vector and the flight heading. The coordinate transformation method is adopted to project the wind speed vector onto the roll direction, pitch direction, and yaw direction in the UAV's own coordinate system. The disturbance wind speed vector in each time frame is decomposed and calculated, and the output disturbance direction component data format is a time series vector sequence, including the roll-direction wind speed disturbance value, pitch-direction wind speed disturbance value, and yaw-direction wind speed disturbance value. At the same time, the variance and maximum value of the wind speed disturbance in each direction are recorded as the amplitude characterization quantity of the disturbance direction. The particle size distribution data and viscosity characteristic data of the spraying material are obtained from fire-fighting spraying material manufacturers. It is also possible to use the image recognition module on the UAV to perform high-speed imaging analysis on the droplets formed at the nozzle outlet of the spraying material. The image acquisition rate is 4000 frames per second, and the image resolution is 1280×720 pixels. After extracting the droplet boundary through the image segmentation algorithm, the droplet diameter in each frame of the image is calculated and statistically analyzed to generate a particle size distribution histogram. The small particle size is defined as less than 0.1 mm, the medium particle size is 0.1 to 0.3 mm, and the large particle size is greater than 0.3 mm. At the same time, a rotational viscometer is used to perform dynamic viscosity tests on the spraying material under different shear rate conditions. The shear rate gradient is set to 50 to 400 s⁻¹, and the test temperature is set to 25 °C. The viscosity values at each shear rate are recorded, and the cubic spline interpolation method is used to fit them into a continuous viscosity response curve. The particle size distribution data and viscosity characteristic data are output in tabular form. When simulating and evaluating the kinetic energy fluctuation range of the spraying air flow barrier based on the disturbance direction component data and the attitude jitter amplitude range, first, the disturbance direction component data and the attitude jitter amplitude are time-aligned, a coupling response model of the disturbance vector and the attitude disturbance is established, the Euler method is used to solve the differential equation, the local kinetic energy loss caused by the disturbance per unit time is calculated, and the difference between this kinetic energy loss and the kinetic energy of the spraying particles is calculated to obtain the kinetic energy fluctuation value of the spraying flow under the action of disturbances in different directions. The fluctuation value is used as the intensity index of the disturbance action window. The time period in which the kinetic energy fluctuation caused by continuous disturbance action is greater than 30% and the duration exceeds 0.5 s is defined as the kinetic energy fluctuation range. The disturbance intensity in each range is integrally segmented, and the included angle value between the dominant vector of the disturbance direction and the spraying axis is marked. The output data format of the kinetic energy fluctuation range of the spraying air flow barrier includes the range number, average disturbance intensity, percentage of kinetic energy loss, action duration, and corresponding flight time index.When performing approximate integration of the fluidity resistance fluctuation of the spraying airflow blocking kinetic energy fluctuation range, use the kinematic viscosity model of the spraying particles, combine the aforementioned spraying particle size and viscosity characteristics, calculate the change in the flow resistance of the spraying particles per unit volume under the action of turbulent flow, define the unit resistance fluctuation as the difference between the resistance value and the static spraying resistance value, perform discrete integration on the resistance fluctuation within each kinetic energy fluctuation range, set the integration time step to 0.02 seconds, use the trapezoidal integration method to solve the total resistance fluctuation, define the interval with a total resistance fluctuation amplitude greater than 0.5 Pa as the significantly disturbed fluidity section, record the maximum, minimum, average, and change slope of the resistance for each section, and at the same time match with the angle between the turbulent flow direction and mark the characteristics of the disturbed direction vector, output the approximate data of the fluidity resistance fluctuation for use as the basis for controlling and adjusting the input parameters. When performing spraying pressure control on the spraying material particle size distribution data and the spraying material viscosity characteristic data according to the approximate data of the fluidity resistance fluctuation, establish a spraying nozzle pressure control model, adopt a double-loop pressure regulator structure, the inner loop controls the instantaneous pressure stability of the nozzle, and the outer loop controls the adaptive adjustment of the total spraying pressure with the turbulent flow resistance fluctuation. Take the resistance fluctuation values in each time period in the approximate data of the fluidity resistance fluctuation as the expected input of the outer loop, use the output of the nozzle pressure sensor as the feedback variable, adjust the proportional gain Kp to 1.8 and the integral coefficient Ki to 0.05, derive the minimum compensation pressure increment required for each spraying cycle, calculate the pressure requirements for different particle size sections according to the particle size distribution, the pressure increment corresponding to the large particle size is 0.3 MPa, the pressure increment corresponding to the medium particle size is 0.2 MPa, and the small particle size increment is 0.1 MPa. Combine the viscosity characteristic curve to calculate the non-linear adjustment coefficient, and finally output the flight spraying pressure control data including the nozzle number, target pressure value, adjustment period, upper and lower pressure boundaries, and real-time pressure change curve to complete the definition of the spraying pressure closed-loop adjustment logic.

[0039] Step S3 includes the following steps: Step S31: Normalize the time-varying attitude spraying adjustment strategy to obtain the normalized time-varying attitude spraying adjustment data; Step S32: Perform logical iterative learning on the normalized time-varying attitude spraying adjustment data to obtain the logical iterative data of the time-varying attitude spraying; Step S33: Based on the policy gradient algorithm, construct a fire extinguishing spraying control model for the logical iterative data of the time-varying attitude spraying to obtain the fire extinguishing spraying control model; Step S34: Send the fire extinguishing spraying control model to the terminal to execute the unmanned aerial vehicle fire extinguishing control method.

[0040] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: Normalize the time-varying attitude spraying adjustment strategy to obtain the normalized time-varying attitude spraying adjustment data; In the embodiment of the present invention, when normalizing the time-varying attitude spraying adjustment strategy, first collect a multi-dimensional control strategy sequence including flight attitude angle change data, spraying angle adjustment instruction data, and target fire source heat intensity response data. The original data sampling frequency is set to 10 Hz, the data time span is 120 seconds, and the Z-score normalization method is used to normalize the data of each dimension. Before normalization, the flight attitude angle range is roll angle ±30 degrees, pitch angle ±20 degrees, yaw angle ±45 degrees, the spraying angle range is 0 to 90 degrees, and the target fire source heat intensity value range is 200 to 900 kilowatts per square meter. During the normalization process, the mean and standard deviation of each dimension sample are used as the calculation basis, and the normalization result is limited to the interval [-1, 1]. The processed data format is uniformly in the form of a 128×6 matrix, where 128 is the total number of normalization time frames, and 6 is the dimension of the control vector included in each frame.

[0041] Step S32: Perform logical iterative learning on the normalized data of the time-varying attitude spraying adjustment to obtain the time-varying attitude spraying logical iterative data; In the embodiment of the present invention, when performing logical iterative learning on the normalized data of the time-varying attitude spraying adjustment, a neural network model with a three-layer fully connected structure is constructed. The number of input layer nodes is set to 6, the middle two layers are 64 nodes and 128 nodes respectively, the ReLU function is selected as the activation function, and the number of output layer nodes is 1 to output the current spraying control response value. The neural network is trained using the supervised learning method. The dataset used is the previously normalized control strategy data, and label data is constructed to represent the residual value between the actual spraying cooling effect and the attitude adjustment execution error. The training set data division ratio is 80%, the validation set is 10%, and the test set is 10%. The loss function uses the mean square error, the Adam optimizer is selected as the optimizer, the initial learning rate is set to 0.001, each round of training iterates 200 times, and the total number of training rounds is set to 300 rounds. The output time-varying attitude spraying logical iterative data is the iterative response output value corresponding to the input data at each time point.

[0042] Step S33: Based on the policy gradient algorithm, construct a fire extinguishing spraying control model for the time-varying attitude spraying logical iterative data to obtain the fire extinguishing spraying control model; In an embodiment of the present invention, when constructing a fire extinguishing spraying control model for time-varying attitude spraying logic iterative data based on the policy gradient algorithm, the Deep Deterministic Policy Gradient (DDPG) algorithm is adopted to construct a dual-channel reinforcement learning architecture including an Actor network and a Critic network. The Actor network is used to output the spraying angle and attitude adjustment actions under different input states, and the Critic network is used to evaluate the value function of the actions. The state input is the normalized time-varying attitude control feature vector, and the action output space includes the adjustment range of the spraying angle, the attitude change speed, and the attitude target angle. The reward function is designed as the increase in the fire extinguishing coverage rate minus the attitude deviation amount and the energy consumption factor, where the fire extinguishing coverage rate is obtained from the comparison calculation result of the actual droplet action area coverage image returned by the spraying control simulation module. During the training process, the capacity of the experience replay buffer is 100,000, the batch size is 64, the update frequency is to update the Actor network twice for every one-time update of the Critic network, the number of training steps is 50,000 steps, and finally the spraying control policy function is output as the core parameter of the model.

[0043] Step S34: Send the fire extinguishing spraying control model to the terminal to execute the drone fire extinguishing control method.

[0044] In an embodiment of the present invention, when sending the fire extinguishing spraying control model to the terminal, first, the model is sent to the flight control terminal of the drone through the wireless communication module (such as Wi-Fi or 5G communication) of the flight control system, and an encryption protocol is used to ensure the security of the model data transmission. The data transmission adopts the JSON format, and the fields include: [model ID, control strategy, spraying parameters, timestamp]. After the transmission is completed, after the flight control system receives the control model, it will automatically verify the integrity and validity of the model. After verification, the control strategy and spraying parameters are loaded into the spraying system and the attitude control system in real time. In the flight control system, the system will dynamically adjust the flight attitude and spraying angle according to the real-time sensor data. During the flight of the drone, the spraying system will adjust the spraying pressure and flow rate according to the real-time data to ensure the best fire extinguishing effect. At the same time, the flight control system will update the flight attitude in a timely manner according to the current flight state to complete the fire extinguishing task of the drone.

[0045] The present invention also provides a drone fire extinguishing control system for executing the above-mentioned drone fire extinguishing control method. The drone fire extinguishing control system includes: A heat flow wind field intensity difference analysis module for collecting real-time heat flow wind field data of a high-rise building fire site through sensors carried on the drone to obtain real-time heat flow wind field data; performing close-range heat flow wind field intensity difference analysis on the real-time heat flow wind field data to obtain close-range grid intensity difference data of the heat flow wind field; A time-varying attitude spraying strategy adjustment module, which is used to numerically quantify the instability of the UAV flight attitude according to the difference data of the heat flow wind field approaching grid intensity, so as to obtain the numerically quantified data of the flight attitude instability; and adjust the time-varying attitude spraying strategy according to the numerically quantified data of the flight attitude instability, so as to obtain the time-varying attitude spraying adjustment strategy. A fire extinguishing spraying control model construction module, which is used to construct a fire extinguishing spraying control model for the time-varying attitude spraying adjustment strategy based on the policy gradient algorithm, so as to obtain the fire extinguishing spraying control model; and send the fire extinguishing spraying control model to the terminal to execute the UAV fire extinguishing control method.

[0046] The above are only specific embodiments of the present invention, which enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A UAV fire extinguishing control method, characterized in that: The following steps are involved: Step S1: using sensors carried by the drone to collect real-time data of the heat flow and wind field at the fire scene of the high-rise building, and obtaining real-time data of the heat flow and wind field; The real-time data of the thermal flow wind field is analyzed for the intensity difference of the thermal flow wind field, and the grid intensity difference data of the thermal flow wind field is obtained; Step S2: numerically quantify the instability of the UAV's flight attitude according to the grid intensity difference data of the heat flow wind field, and obtain numerical quantification data of the instability of the flight attitude; adjust the time-varying attitude spraying strategy according to the numerical quantification data of the instability of the flight attitude, and obtain the time-varying attitude spraying adjustment strategy; Step S3: construct a fire extinguishing spraying control model for the time-varying posture spraying adjustment strategy based on the policy gradient algorithm to obtain a fire extinguishing spraying control model; send the fire extinguishing spraying control model to the terminal to execute the UAV fire extinguishing control method.

2. The UAV fire extinguishing control method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting real-time data of heat flow and wind field at the fire scene of a high-rise building through sensors carried by the drone to obtain real-time data of heat flow and wind field; Step S12: Fill missing values ​​in the real-time data of the heat flow wind field to obtain real-time filled data of the heat flow wind field; Step S13: performing an approaching heat flow wind field intensity difference analysis on the real-time filling data of the heat flow wind field to obtain approaching heat flow wind field intensity difference data; Step S14: Performing grid structure conversion on the approaching intensity difference data of the heat flow wind field to obtain the grid intensity difference data of the heat flow wind field.

3. The UAV fire extinguishing control method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing disturbance force domain analysis on the thermal flow wind field grid intensity difference data to obtain thermal flow wind field disturbance force domain data; Step S22: numerically quantifying the instability of the UAV's flight attitude according to the thermal flow wind field disturbance force domain data to obtain numerical quantification data of the instability of the flight attitude; Step S23: performing lift limit amplitude analysis based on the flight attitude instability numerical quantification data to obtain flight lift limit amplitude data; Step S24: adjusting the time-varying attitude spraying strategy according to the flight attitude instability numerical quantification data, the flight lift limit amplitude data and the thermal flow wind field disturbance force domain data to obtain the time-varying attitude spraying adjustment strategy.

4. The UAV fire extinguishing control method according to claim 3, characterized in that: Step S22 includes the following steps: Step S221: performing spatial heterogeneity analysis on the thermal flow wind field disturbance force domain data to obtain disturbance force domain spatial heterogeneity data; Step S222: performing horizontal / vertical azimuth disturbance force spectral density analysis based on the disturbance force domain spatial heterogeneity data and the thermal flow wind field disturbance force domain data to obtain the horizontal / vertical azimuth disturbance force spectral density; Step S223: analyzing the disordered changes in the rotation and swing of the UAV according to the horizontal / vertical azimuth disturbance force spectrum density to obtain the disordered changes in the rotation and swing; Step S224: performing a tilt moment fluctuation regression analysis on the rotation swing disordered change data to obtain tilt moment fluctuation regression data; Step S225: performing lateral airflow pressure tumbling limit analysis based on the rotational swing disordered variation data to obtain lateral airflow tumbling limit pressure data; Step S226: numerically quantify the flight attitude instability of the UAV according to the tilt moment fluctuation regression data and the lateral airflow tumbling limit pressure data to obtain the flight attitude instability numerical quantification data.

5. The UAV fire extinguishing control method according to claim 4, characterized in that: Step S223 includes the following steps: Obtain the aerodynamic layout of the UAV; calculate the airflow disturbance time-series pressure difference of the horizontal / vertical azimuth disturbance force spectrum density to obtain the airflow disturbance time-series pressure difference between the horizontal / vertical azimuths; According to the pressure difference of airflow disturbance time sequence, the repeated deviation trajectory of the center of gravity of the UAV is analyzed to obtain the repeated deviation trajectory of the flight center of gravity. The relative angle average difference of horizontal deviation is calculated for the repeated deviation trajectory of the flight center of gravity, and the relative angle average difference of the center of gravity deviation is obtained; The disordered changes of the rotation and swing of the UAV are analyzed based on the relative angle difference of the center of gravity offset and the repeated offset trajectory of the flight center of gravity, and the disordered changes of the rotation and swing are obtained.

6. The UAV fire extinguishing control method according to claim 3, characterized in that: Step S24 includes the following steps: Step S241: Acquire the fire spray material data carried by the drone; Step S242: performing attitude jitter amplitude interval analysis on the flight attitude instability numerical quantification data to obtain an attitude jitter amplitude interval; Step S243: performing flight sequence attitude control adjustment according to the attitude jitter amplitude interval and the flight lift limit amplitude data to obtain flight sequence attitude control data; Step S244: performing spraying pressure control on the fire spraying material data according to the attitude jitter amplitude interval and the thermal flow wind field disturbance force domain data to obtain flight spraying pressure control data; Step S245: adjusting the time-varying attitude spraying strategy based on the flight timing attitude control data and the flight spraying pressure control data to obtain the time-varying attitude spraying adjustment strategy.

7. The UAV fire extinguishing control method according to claim 6, characterized in that: Step S243 includes the following steps: Performing jitter lateral tilt angle analysis on the posture jitter amplitude interval to obtain jitter lateral tilt angle data; Based on the jitter lateral tilt angle data and the flight ceiling limit amplitude data, the ceiling attitude angle margin regression analysis is performed to obtain the ceiling attitude angle margin regression data; The axial torque output is matched according to the jitter lateral tilt angle data and the flight lift limit amplitude data to obtain the axial torque output matching data; The flight sequence attitude control is adjusted according to the ceiling attitude angle margin regression data and the axial torque output matching data to obtain the flight sequence attitude control data.

8. The UAV fire extinguishing control method according to claim 7, characterized in that: Step S244 includes the following steps: Perform spatial vector decomposition processing on the disturbance force domain data of the thermal flow wind field to obtain the disturbance direction component data; Perform particle size distribution and viscosity characteristic identification processing on the fire spray material data to obtain the spray material particle size distribution data and the spray material viscosity characteristic data respectively; Based on the disturbance direction component data and the attitude jitter amplitude range, the spray airflow blocking kinetic energy fluctuation range is simulated and evaluated to obtain the spray airflow blocking kinetic energy fluctuation range; The approximate integration of the flow resistance fluctuation is performed on the spray airflow blocking kinetic energy fluctuation range to obtain the approximate data of the flow resistance fluctuation; The spraying pressure is controlled based on the particle size distribution data of the spraying material and the viscosity characteristic data of the spraying material according to the approximate data of the fluidity resistance fluctuation, so as to obtain the flight spraying pressure control data.

9. The UAV fire extinguishing control method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: normalizing the time-varying posture spraying adjustment strategy to obtain time-varying posture spraying adjustment normalized data; Step S32: performing logical iterative learning on the time-varying posture spraying adjustment normalization data to obtain the time-varying posture spraying logical iterative data; Step S33: constructing a fire extinguishing spraying control model for the time-varying posture spraying logic iteration data based on a policy gradient algorithm to obtain a fire extinguishing spraying control model; Step S34: Send the fire extinguishing spray control model to the terminal to execute the drone fire extinguishing control method.

10. A UAV fire extinguishing control system, characterized in that: Used to execute the drone fire extinguishing control method as claimed in claim 1, the drone fire extinguishing control system comprises: The heat flow wind field intensity difference analysis module is used to collect real-time data of the heat flow wind field at the high-rise building fire scene through the sensors carried by the drone to obtain the real-time data of the heat flow wind field; the real-time data of the heat flow wind field is analyzed for the difference in the intensity of the heat flow wind field close to the grid to obtain the difference in the intensity of the heat flow wind field close to the grid; The time-varying attitude spraying strategy adjustment module is used to numerically quantify the instability of the UAV's flight attitude according to the difference data of the intensity of the heat flow wind field approaching the grid, and obtain the numerical quantification data of the flight attitude instability; adjust the time-varying attitude spraying strategy according to the numerical quantification data of the flight attitude instability, and obtain the time-varying attitude spraying adjustment strategy; The fire extinguishing spraying control model construction module is used to construct a fire extinguishing spraying control model for the time-varying posture spraying adjustment strategy based on the policy gradient algorithm to obtain the fire extinguishing spraying control model; the fire extinguishing spraying control model is sent to the terminal to execute the UAV fire extinguishing control method.

Images