A method and system for stabilizing control of a UAV in a high-rise fire extinguishing scene

Abstract

This invention discloses a method and system for stabilizing unmanned aerial vehicles (UAVs) in a high-rise building firefighting scenario, relating to the field of UAV control technology. A UAV stabilization control system for a high-rise building firefighting scenario includes: a transient temperature module, a coupling compensation module, a shock wave prediction module, a feedforward control module, an attitude calculation module, a reliable estimation module, and an attitude estimation module. This invention achieves active compensation for sudden shock disturbances by acquiring a joint thermal-mechanical shock prediction factor and entering an shock feedforward control mode when the factor exceeds a preset threshold, thereby improving the response speed and stability of the attitude control system. Furthermore, by dynamically adjusting the attitude control loop parameters under the shock feedforward control mode, the damping capability of the control under shock disturbances is improved, reducing the attitude oscillation amplitude.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for stable control of UAVs in high-rise firefighting scenarios. Background Technology

[0002] With the increasing number of high-rise buildings in cities, the difficulty of fire rescue in high-rise buildings has significantly increased. Drones, due to their advantages such as high mobility, rapid deployment, and ability to access complex and dangerous areas, are widely used in high-rise fire reconnaissance and firefighting support operations. In actual firefighting scenarios, drones need to hover stably or maneuver precisely in complex environments such as high-temperature smoke, explosive impacts, strong airflow disturbances, and visual obstruction. Their flight stability and attitude control accuracy directly affect the safety and effectiveness of firefighting operations. Therefore, how to maintain the attitude stability of drones in extreme fire environments has become an important research direction in the current field of drone firefighting applications.

[0003] Existing technologies often employ attitude control methods based on inertial measurement units (IMUs), correcting IMU bias through temperature compensation algorithms or improving attitude estimation accuracy through multi-sensor fusion. However, these approaches typically treat temperature compensation, shock disturbance detection, and multi-source fusion as independent functional modules, relying primarily on reactive error feedback for correction. In high-rise fires, sudden disturbances such as glass shattering shock waves or drastic thermal gradient changes can cause nonlinear drift in the IMU bias within a short time. Existing methods struggle to proactively predict and compensate for these shocks before they arrive, leading to attitude estimation lag, control oscillations, and even short-term instability. Furthermore, the dynamic reliability changes of multi-source sensors under extreme environments, without a unified health assessment mechanism, can also affect the stability of the fusion results. Summary of the Invention

[0004] This invention proposes a UAV stability control method that achieves active feedforward compensation based on temperature field changes and shock wave prediction, and combines sensor health assessment for dynamic fusion in high-rise fire fighting scenarios and under the coupled effects of complex thermal environments and shock disturbances.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for stabilizing and controlling a drone in a high-rise firefighting scenario includes: Temperature data is collected in real time by temperature sensors located in the area where the IMU is located, a spatial transient temperature field model is constructed, and temperature characteristics, including temperature gradient vector, temperature rise rate, and temperature gradient change rate, are calculated. Establish a coupling compensation model between IMU bias and temperature data and temperature characteristics to obtain the IMU gyroscope bias correction amount; Acoustic data is collected by an acoustic sensor array and air pressure data is collected by a barometer. The energy spectrum, time difference of arrival and rate of change of air pressure in the preset frequency band are extracted to construct a spatiotemporal prediction model of shock wave and output the arrival time and intensity of shock wave. Based on the temperature gradient change rate and the arrival time and intensity of the shock wave, the joint predictor of thermal-mechanical shock is obtained. When the joint predictor exceeds the preset threshold, the shock feedforward control mode is entered, and the predictive compensation term is injected into the coupled compensation model and the attitude control loop parameters are adjusted. The attitude estimate is obtained by correcting the IMU angular velocity based on the IMU gyroscope zero bias correction and performing attitude calculation, and then outputting attitude control commands. When the shock wave actually arrives or the visual characteristics degrade, the health status is assessed based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. A dynamic weight allocation matrix is ​​generated based on the credibility index, and the visual positioning data and the corrected inertial data are nonlinearly fused to output the final attitude estimation result.

[0006] As a preferred technical solution of the present invention, the construction of the spatial transient temperature field model includes: mapping the installation positions of temperature sensors located at different spatial locations in the IMU area to three-dimensional coordinate points; collecting temperature data in real time through each temperature sensor within a preset sampling period; and establishing a spatial transient temperature field model in the IMU area with spatial coordinates and time as independent variables and temperature as dependent variable through interpolation based on the temperature data; in the spatial transient temperature field model, calculating the temperature gradient vector by performing the first-order partial derivative of temperature in the spatial direction, calculating the temperature rise rate by performing the first-order derivative of temperature in the time direction, and obtaining the temperature gradient change rate by differentiating the temperature gradient vector within a preset time window.

[0007] As a preferred technical solution of the present invention, the acquisition of the IMU gyroscope zero bias correction includes: during the calibration stage, collecting the measured values ​​of the IMU gyroscope under known zero bias conditions and the corresponding temperature data and temperature characteristics under different temperature distribution conditions, using the temperature data and temperature characteristics as input variables, and using the actual zero bias of each gyroscope axis as the output variable to construct a sample dataset; based on the sample dataset, establishing a zero bias coupling compensation function with temperature data and temperature characteristics as input and basic zero bias correction as output through a multivariate fitting algorithm, as a coupling compensation model; during actual flight, inputting the temperature data and temperature characteristics at the current moment into the zero bias coupling compensation function to obtain the basic zero bias correction, and superimposing a prediction compensation term obtained by the thermal-mechanical shock joint prediction factor on the basic zero bias correction to obtain the IMU gyroscope zero bias correction.

[0008] As a preferred embodiment of the present invention, the output of the shock wave arrival time and intensity includes: determining the propagation direction of the shock wave front relative to the UAV based on the arrival time difference of acoustic data received by each acoustic sensor in the acoustic sensor array, and estimating the propagation distance from the shock wave front to the UAV based on the spatial distance between each acoustic sensor; performing spectral analysis on the acoustic data within a preset frequency band to obtain an energy spectrum index characterizing the acoustic energy of the shock wave, and calculating the air pressure change rate from the air pressure data collected by the air pressure sensor; calculating the shock wave arrival time based on the propagation distance and a preset shock wave propagation speed, wherein the shock wave arrival time is equal to the ratio of the propagation distance to the preset shock wave propagation speed; and calculating the shock wave intensity based on the energy spectrum index and the air pressure change rate using a weighted summation method, wherein the shock wave intensity is obtained by multiplying the energy spectrum index and the air pressure change rate by preset weighting coefficients and then summing them.

[0009] As a preferred embodiment of the present invention, the acquisition of the thermal-mechanical shock joint prediction factor includes: performing time-accumulation calculation on the temperature gradient change rate within a preset time window to obtain a first thermal disturbance index characterizing the degree of drastic temperature change in the area where the IMU is located; based on the arrival time and intensity of the shock wave, normalizing the arrival time of the shock wave to reflect its proximity to the current moment, and normalizing the intensity of the shock wave to reflect the magnitude of the shock wave's impact energy, to obtain a second mechanical disturbance index; and combining the first thermal disturbance index and the second mechanical disturbance index according to preset weights to obtain the thermal-mechanical shock joint prediction factor.

[0010] As a preferred technical solution of the present invention, the shock feedforward control mode includes: determining the action time window and amplitude of the prediction compensation term based on the arrival time and intensity of the shock wave; superimposing the prediction compensation term with the basic zero bias correction amount determined by the coupled compensation model within the action time window to obtain the target zero bias correction amount used to correct the IMU angular velocity; and adjusting the attitude control loop parameters according to the shock wave intensity within the action time window.

[0011] As a preferred technical solution of the present invention, the attitude calculation includes: determining an initial attitude estimate based on the stationary state of the UAV at the initial moment; during the flight, in each attitude calculation cycle, subtracting the IMU gyroscope zero bias correction obtained by the coupling compensation model and the prediction compensation term from the gyroscope angular velocity in the IMU measurement value to obtain the corrected IMU angular velocity, and using the corrected IMU angular velocity as the angular rate input, updating the attitude estimate of the previous cycle through the quaternion-based attitude calculation algorithm to obtain the current attitude estimate result.

[0012] As a preferred technical solution of the present invention, the reliability index is obtained by: calculating the degree to which the temperature characteristics of the area where the IMU is located deviate from the preset safety range based on the transient temperature field model when the shock wave actually arrives or the visual characteristics degrade, and obtaining the thermal environment anomaly index through normalization processing; extracting the shock wave intensity and duration based on the measured shock wave data, and obtaining the mechanical disturbance index by comparing the shock wave intensity with the preset shock threshold and then normalizing it; calculating the residual mean square value based on the residual between the IMU measurement value and the attitude estimation, and obtaining the measurement consistency index by converting it through a monotonically decreasing function according to the relationship between the residual mean square value and the preset error threshold; calculating the IMU reliability index by weighted averaging the thermal environment anomaly index, the mechanical disturbance index, and the measurement consistency index; and calculating the reliability of the sensors that acquire visual positioning data, the acoustic sensor array, and the barometric pressure sensor based on the measurement consistency of each sensor and its corresponding environmental disturbance index to obtain the reliability index of each sensor.

[0013] As a preferred technical solution of the present invention, the nonlinear fusion includes: converting the reliability index of each sensor into weight coefficients for visual positioning data and corrected inertial data in a dynamic weight allocation matrix according to a preset nonlinear mapping relationship; within each attitude estimation update cycle, using the dynamic weight allocation matrix to weight and combine the visual positioning data and corrected inertial data to output the final attitude estimation result.

[0014] A drone stabilization control system for high-rise building firefighting scenarios includes: Transient temperature module: Real-time acquisition of temperature data, construction of spatial transient temperature field model, and calculation of temperature characteristics; Coupling compensation module: Establishes a coupling compensation model between IMU zero bias and temperature data and temperature characteristics, and obtains the IMU gyroscope zero bias correction amount; Shock wave prediction module: Collects acoustic and air pressure data, constructs a spatiotemporal prediction model for shock waves, and outputs the arrival time and intensity of shock waves; Feedforward control module: acquires the joint predictor of thermal-mechanical shock, and enters the shock feedforward control mode when the joint predictor exceeds the preset threshold; Attitude calculation module: Corrects the IMU angular velocity based on the IMU gyroscope zero bias correction and performs attitude calculation to obtain attitude estimate, and outputs attitude control command; Reliable estimation module: When the shock wave actually arrives or the visual features degrade, the module assesses the health of each sensor based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. Attitude estimation module: Generates a dynamic weight allocation matrix based on the confidence index, performs nonlinear fusion of visual positioning data and corrected inertial data, and outputs the final attitude estimation result.

[0015] The present invention has the following advantages: This invention establishes a coupling compensation model between temperature data and temperature characteristics, and combines it with a prediction compensation term to achieve zero-bias dynamic correction, thereby reducing the impact of high-temperature environment on IMU measurement accuracy and reducing the accumulation of attitude estimation error. By constructing a shock wave spatiotemporal prediction model and calculating the arrival time and intensity of the shock wave, it achieves early prediction of explosive shock disturbances and provides a time window for shock feedforward control.

[0016] This invention achieves active compensation for sudden impact disturbances by acquiring the joint predictive factor of thermal-mechanical shock and entering the impact feedforward control mode when it exceeds a preset threshold, thereby improving the response speed and stability of the attitude control system. By dynamically adjusting the parameters of the attitude control loop under the impact feedforward control mode, the damping capability of the control under impact disturbances is improved, and the attitude oscillation amplitude is reduced.

[0017] This invention assesses health based on a transient temperature field model, measured shock wave data, and IMU measurement residuals, enabling dynamic determination of the reliability of each sensor. By generating a dynamic weight allocation matrix and performing nonlinear fusion of visual positioning data and corrected inertial data, the robustness and anti-interference capability of attitude estimation results are improved. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort. Figure 1 This is a schematic diagram of the structure of a drone stability control system for a high-rise firefighting scenario, as used in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] Example 1: A method for stabilizing and controlling a drone in a high-rise firefighting scenario. The drone approaches the exterior of a high-rise building fire to perform firefighting operations: the drone is close to the flames and high-temperature smoke plumes, the temperature in the area where the IMU is located rises rapidly and there is spatial inhomogeneity; simultaneously, "shock waves" such as deflagration / pressure waves occur, causing transient mechanical disturbances; the visual sensor experiences feature degradation due to refraction / blurring caused by smoke and hot plumes. The method involves data acquisition, modeling, prediction, compensation, and fusion according to steps S1–S7, including the following steps: Step S1: Collect temperature data in real time using temperature sensors located in the area where the IMU is located, construct a spatial transient temperature field model, and calculate temperature characteristics, including temperature gradient vector, temperature rise rate, and temperature gradient change rate. The construction of the spatial transient temperature field model includes: mapping the installation positions of temperature sensors located at different spatial locations in the IMU area to three-dimensional coordinate points; collecting temperature data in real time through each temperature sensor within a preset sampling period; and establishing a spatial transient temperature field model in the IMU area with spatial coordinates and time as independent variables and temperature as the dependent variable through interpolation based on the temperature data. In the spatial transient temperature field model, the temperature gradient vector is obtained by calculating the first-order partial derivative of temperature in the spatial direction, the temperature rise rate is obtained by calculating the first-order derivative of temperature in the time direction, and the temperature gradient change rate is obtained by differentiating the temperature gradient vector within a preset time window.

[0021] The IMU (Indoor Temperature Detector) is installed inside the flight control unit or in its vicinity. To characterize the thermal state of this area, multiple (at least three) temperature sensors are deployed in the IMU area. Temperature data is collected synchronously within the same sampling period, forming a "multi-point, same-timestamp" temperature observation set. Each sampling moment corresponds to a set of temperature values, including the instantaneous readings and sampling timestamps of each temperature sensor, collected in real time from temperature sensors located at different spatial positions within the IMU area. For example, four temperature sensors are set around the IMU, located in front, behind, to the left, and to the right of the IMU, with a fixed sampling period (e.g., sampling at millisecond intervals). When the UAV approaches a flame and a hot plume, the temperature readings of the sensors closer to the heat source rise continuously, while the temperature readings on the side farther from the heat source rise more slowly, thus creating a spatial temperature difference and temporal variation.

[0022] In this method, the spatial transient temperature field model is defined as a continuous temperature distribution expression within the IMU region, with the three-dimensional spatial coordinates and time corresponding to the sensor installation location as independent variables and temperature as the dependent variable. It is used to reconstruct the temperature state of "any location and any sampling time within the IMU region" based on "a finite number of temperature sampling points", and organize discrete multi-point temperature data into a unified "spatial-temporal temperature description", avoiding the description loss caused by using only a single point temperature to represent the heating environment of the IMU.

[0023] Mapping the installation location of temperature sensors to three-dimensional coordinate points refers to assigning a unique spatial coordinate identifier to each temperature sensor within the UAV's system or IMU's local coordinate system, ensuring a one-to-one correspondence between temperature readings and spatial locations. Each temperature sensor corresponds to a three-dimensional coordinate point, which is bound to and stored with the sensor number. A record structure is created at each sampling moment: "timestamp + sensor number set + corresponding coordinate point set + temperature reading set". For example, four temperature sensors correspond to four coordinate points; the coordinate point closer to the heat source experiences a faster temperature rise, resulting in a higher temperature zone in the interpolated temperature field towards the heat source.

[0024] The purpose of interpolation is limited to generating a continuous temperature distribution representation based on a finite number of temperature sampling points within the IMU region, enabling consistent feature calculations in both spatial and temporal directions. The interpolation input consists of multiple temperature readings at the same sampling time and their spatial coordinates. The output is a continuous temperature distribution representation within the IMU region, further yielding a spatial transient temperature field model. The interpolation range is limited to a preset spatial boundary within the IMU region, preventing the temperature field from extending to areas unrelated to the IMU's thermal environment. This ensures, from a data validity perspective, that subsequent temperature gradients and temperature rise rates consistently reflect the "IMU region."

[0025] In this method, the temperature gradient vector is defined as: a vector obtained by calculating the first-order change of the temperature distribution along the spatial direction within the spatial transient temperature field model; it is used to characterize "the direction of the fastest temperature change in space" and "the intensity of the spatial temperature difference". The spatial change calculation results are derived from the spatial transient temperature field model.

[0026] In this method, the rate of temperature rise is defined as: a quantity obtained by calculating the first-order change of temperature along the time direction within a spatial transient temperature field model, used to characterize "how fast the temperature rises with time". It comes from the calculation results of the temperature field model's change in the time dimension, or from the temperature change at the same spatial location at consecutive sampling times.

[0027] In this method, the rate of change of the temperature gradient is defined as the magnitude of change obtained by differentiating the temperature gradient vector within a preset time window, used to characterize the "intensity of change in the spatial temperature difference distribution pattern over a short period of time". The temperature gradient vector sequence from the continuous attitude calculation cycle or temperature sampling cycle is obtained through time window differencing.

[0028] Step S2: Establish a coupling compensation model between IMU bias and temperature data and temperature characteristics, and obtain the IMU gyroscope bias correction amount; The acquisition of the IMU gyroscope zero-bias correction includes: during the calibration phase, collecting the measured values ​​of the IMU gyroscope under known zero-bias conditions, along with corresponding temperature data and temperature characteristics, under different temperature distribution conditions. The temperature data and characteristics are used as input variables, and the actual zero-bias of each gyroscope axis is used as the output variable to construct a sample dataset. Based on the sample dataset, a zero-bias coupling compensation function is established using a multivariate fitting algorithm, with temperature data and temperature characteristics as input and a basic zero-bias correction as output, serving as the coupling compensation model. During actual flight, the current temperature data and temperature characteristics are input into the zero-bias coupling compensation function to obtain the basic zero-bias correction. A prediction compensation term obtained from the thermal-mechanical shock joint prediction factor is then superimposed on the basic zero-bias correction to obtain the IMU gyroscope zero-bias correction.

[0029] The coupled compensation model is defined as a mapping model that describes the correspondence between "IMU gyroscope zero bias" and "thermal environment state of the IMU's location," where the thermal environment state is characterized by the temperature data and temperature characteristics obtained in step S1. The output of the coupled compensation model is used to correct the zero bias component in the angular velocity measured by the IMU gyroscope, ensuring stable attitude calculation input.

[0030] The zero-biased coupling compensation function is defined as the functional expression of the coupling compensation model, whose inputs are temperature data and temperature features, and whose output is the basic zero-biased correction. The zero-biased coupling compensation function is obtained by fitting the sample dataset during the calibration phase and is directly called during the flight phase.

[0031] The basic zero bias correction is defined as: the zero bias correction result output by the zero bias coupling compensation function under the current thermal environment conditions, which only reflects the influence of temperature and temperature characteristics on the zero bias, and does not include the additional compensation brought by the shock wave feedforward prediction.

[0032] The prediction compensation term is defined as: the feedforward compensation component generated by the thermal-mechanical shock joint prediction factor output by S4, which is used to make additional corrections to the zero bias correction amount within the preset action time window before the arrival of the shock wave, so that the zero bias correction amount simultaneously covers the "temperature coupling effect" and the "thermal-mechanical shock feedforward effect".

[0033] The sample dataset is constructed during the calibration phase. Each sample record in the sample dataset consists of "input items" and "output items" in its data structure, and the input items strictly follow the temperature data and temperature characteristics in S1.

[0034] Actual zero bias for each gyroscope axis: These correspond to the actual zero bias values ​​of the three axes of the IMU gyroscope under this operating condition, and are used as label data for supervised fitting.

[0035] In this method, "measured values ​​under known zero-bias conditions" are defined as: gyroscope measurement values ​​obtained during the calibration phase through a traceable zero-bias reference method, used to determine the actual zero-bias label under the corresponding operating conditions. This reference method, at the implementation level, is manifested as the condition that "zero-bias can be calculated, recorded, and used as an output variable," enabling the sample dataset to establish a definite correspondence between input and output.

[0036] The organization of different temperature distribution conditions: the "different temperature distribution conditions" in the calibration stage are organized according to two types of changes: spatially uneven heating and rapid temperature rise over time, and a corresponding relationship is established with the three types of temperature characteristics in S1.

[0037] The three scenarios are: **Non-uniform spatial heating:** A significant spatial temperature difference is created within the IMU region, causing the temperature gradient vector to have a significant amplitude in a certain direction, thus covering the difference between the heat source side and the back heat side. **Transient rapid heating:** The temperature rise rate reaches a high level within a short time, thus covering the heating rate change as the hot plume passes by. **Rapid change in spatial temperature difference morphology:** The temperature gradient vector changes significantly within a preset time window, thus covering the situation where the rate of change of the temperature gradient increases. These three scenarios correspond to three states of the UAV near the fire facade: hovering close to the flame side, crossing the boundary of the hot plume, and the hot plume swaying with the wind, causing the heat source side to alternate on both sides of the aircraft.

[0038] Based on the sample dataset, when establishing the zero-biased coupling compensation function using a multivariate fitting algorithm, the fitting process satisfies the following constraints: the input consists of temperature data and temperature features, where the temperature data represents "multi-point temperature readings" and the temperature features represent "gradient vector, temperature rise rate, and gradient change rate". The output includes the basic zero-biased correction amount corresponding to the three axes of the gyroscope, so that each axis obtains a correction result coupled with the thermal environment. The zero-biased coupling compensation function corresponds to a set of fitting parameters or lookup table parameters at the implementation level, which are solidified as the coupling compensation model and used for direct invocation during the flight phase.

[0039] A predictive compensation term, obtained from the joint predictor of thermal and mechanical shock, is superimposed on the basic zero-bias correction. This predictive compensation term originates from S4 and is generated when the joint predictor exceeds a preset threshold. The superposition is performed within the time window determined by S4; outside the window, the output only contains the basic zero-bias correction. The predictive compensation term covers the feedforward compensation requirements before the shock and, together with the basic zero-bias correction, constitutes the IMU gyroscope zero-bias correction output by S2, without altering the core structure of the zero-bias coupling compensation function.

[0040] The output of S2 is the IMU gyroscope zero bias correction amount. Its data content remains consistent in subsequent embodiments, specifically including: three-axis zero bias correction amount values, corresponding to the three axes of the gyroscope; output timestamp, aligned with the temperature data sampling time; correction amount composition identifier, distinguishing between "basic zero bias correction amount" and "whether to superimpose prediction compensation terms (and the state within the superposition window)".

[0041] Step S3: Acoustic data is collected using an acoustic sensor array and air pressure data is collected using a barometric pressure sensor. The energy spectrum, arrival time difference, and air pressure change rate of the preset frequency band are extracted to construct a shock wave spatiotemporal prediction model and output the arrival time and intensity of the shock wave. The output of the shock wave arrival time and intensity includes: determining the propagation direction of the shock wave front relative to the UAV based on the arrival time difference of acoustic data received by each acoustic sensor in the acoustic sensor array, and estimating the propagation distance from the shock wave front to the UAV based on the spatial distance between each acoustic sensor; performing spectral analysis on the acoustic data within a preset frequency band to obtain an energy spectrum index characterizing the acoustic energy of the shock wave, and calculating the air pressure change rate from the air pressure data collected by the air pressure sensor; calculating the shock wave arrival time based on the propagation distance and a preset shock wave propagation speed, where the shock wave arrival time is equal to the ratio of the propagation distance to the preset shock wave propagation speed; and calculating the shock wave intensity based on the energy spectrum index and the air pressure change rate using a weighted summation method, where the shock wave intensity is obtained by multiplying the energy spectrum index and the air pressure change rate by preset weighting coefficients and then summing them.

[0042] As drones approached the exterior of a high-rise building during firefighting operations, events such as deflagration, glass breakage, and sudden pressure changes occurred at the fire site, creating disturbances in the air that propagated in the form of wavefronts.

[0043] The acoustic data includes: the time series of sound pressure signals from each acoustic sensor in the acoustic sensor array, and the corresponding sampling timestamps. It is acquired in real time by the acoustic sensor array. The sensors within the array have a fixed spatial geometric relationship.

[0044] The barometric pressure data includes: the barometric pressure time series from the barometric pressure sensor and the corresponding sampling timestamps. It is acquired in real time by the barometric pressure sensor, and the sampling period is aligned with the attitude calculation period on the same time base or through timestamps.

[0045] For example, when a deflagration occurs in a window of a high-rise building, the acoustic array receives the sudden acoustic waveforms, and the barometric pressure sensor simultaneously records the short-term pressure change. This event is then converted into "time difference of arrival, energy spectrum index, and rate of pressure change," and the arrival time and intensity are output.

[0046] An acoustic sensor array is defined as a data acquisition structure consisting of multiple acoustic sensors fixedly installed according to a preset geometric layout. The relative spatial distance between each acoustic sensor is a known quantity. It is used to extract wavefront propagation direction and distance information through the time difference of multi-point reception. In this method, this array serves as one of the main sources of shock wave precursor information.

[0047] The time difference of arrival (TDOA) is defined as the time difference between the arrival of an acoustic waveform caused by the same impact event at different acoustic sensors within the array. TDOA reflects the incident direction information of the wavefront relative to the array and, together with the array geometry, is used to estimate the propagation distance of the wavefront to the UAV.

[0048] The preset frequency band energy spectrum (energy spectrum index) is defined as: an energy characterization index obtained by performing spectral analysis on acoustic data within a preset frequency band, used to characterize the concentration of acoustic energy of an impact event within that frequency band. The preset frequency band is set to eliminate low-frequency background noise or high-frequency transient spike interference unrelated to the impact event, ensuring that the energy spectrum index is consistent with the characteristics related to the impact event.

[0049] The rate of change of air pressure is defined as: the speed of air pressure change calculated from the air pressure time series between adjacent sampling times. It is used to characterize the strength and trend of pressure disturbances and serves as a mechanical lateral measurement for the calculation of shock wave intensity.

[0050] The shock wave spatiotemporal prediction model is defined as follows: taking the arrival time difference of the acoustic array, array geometric information, preset frequency band energy spectrum index, and air pressure change rate as inputs, and outputting the arrival time and intensity of the shock wave. In this method, this model is used to provide available predictive quantities of arrival time and intensity before the actual arrival of the shock wave, for use in step S4 for feedforward mode determination and compensation term calculation.

[0051] The estimation of propagation direction and propagation distance is organized in the order of "direction first, then distance". The data dependencies are clearly defined as follows: the time series of acoustic data of each acoustic sensor in the data array on which the direction is determined; the arrival time difference extracted from the acoustic data; the arrival time difference is used to determine the propagation direction of the wavefront relative to the UAV, and the propagation direction is characterized in the array coordinate system or the machine system.

[0052] The data relied upon for distance estimation includes: array geometry information (spatial distances between each acoustic sensor); obtained propagation direction information; the propagation direction and array geometry are used together to estimate the propagation distance from the wavefront to the UAV, providing distance input for subsequent time of arrival calculations.

[0053] For example, when the sound wave of a deflagration propagates from the direction of the window, the array sensor closer to the window receives the sudden waveform first, and the sensor farther away from the window receives it later, forming a stable time difference sequence of arrival, which in turn indicates that the wavefront propagates from the direction of the window to the direction of the UAV.

[0054] The arrival time of the shock wave is calculated from the propagation distance and the preset shock wave propagation speed. The inputs are the propagation distance and the preset shock wave propagation speed. The arrival time is based on the current sampling time and the output is under the same timestamp system. The propagation speed is taken as a preset value and used as a unified task parameter in the calculation.

[0055] The shock wave intensity is calculated by a weighted sum of the energy spectrum index and the rate of change of air pressure. This structure ensures that the intensity reflects both acoustic and mechanical observations, avoiding a one-sided judgment of the shock strength based solely on a single sensor characteristic. Its inputs include the energy spectrum index, the rate of change of air pressure, and preset weighting coefficients; the weighting coefficients are preset parameters that remain consistent throughout the same mission. The data content of "shock wave arrival time and intensity" includes at least: shock wave arrival time, the relative arrival time with the current time or the arrival time expressed as a unified timestamp; shock wave intensity, the intensity value obtained by weighting the energy spectrum index and the rate of change of air pressure; propagation direction and propagation distance (intermediate quantity), traceable information used for arrival time calculation, which facilitates the consistency verification and health assessment data interpretation in step S6 when the shock wave actually arrives.

[0056] Step S4: Based on the temperature gradient change rate and the arrival time and intensity of the shock wave, obtain the joint predictor of thermal-mechanical shock. When the joint predictor exceeds the preset threshold, enter the shock feedforward control mode, inject the predictive compensation term into the coupled compensation model and adjust the attitude control loop parameters. The acquisition of the thermal-mechanical shock joint prediction factor includes: performing time-accumulation calculation on the temperature gradient change rate within a preset time window to obtain a first thermal disturbance index to characterize the degree of drastic temperature change in the area where the IMU is located; based on the arrival time and intensity of the shock wave, normalizing the arrival time of the shock wave to reflect its proximity to the current moment, and normalizing the intensity of the shock wave to reflect the magnitude of the shock wave's impact energy to obtain a second mechanical disturbance index; and combining the first thermal disturbance index and the second mechanical disturbance index according to preset weights to obtain the thermal-mechanical shock joint prediction factor.

[0057] The shock feedforward control mode includes: determining the action time window and amplitude of the prediction compensation term based on the arrival time and intensity of the shock wave; superimposing the prediction compensation term with the basic zero bias correction amount determined by the coupled compensation model within the action time window to obtain the target zero bias correction amount used to correct the IMU angular velocity; and adjusting the attitude control loop parameters according to the shock wave intensity within the action time window.

[0058] The thermal-mechanical shock joint prediction factor is defined as a unified quantitative index used to characterize the combined risk level of "thermal disturbance intensity in the IMU area" and "the degree of approximation and energy level of shock wave mechanical disturbance". This factor uses the temperature gradient change rate obtained from S1 as the thermal input and the shock wave arrival time and intensity output from S3 as the force input. The output is used to determine whether to enter the shock feedforward control mode.

[0059] The preset threshold is defined as a judgment threshold set for the thermal-mechanical shock joint predictor factor, used to divide the flight state into "normal compensation control state" and "shock feedforward control state". The threshold serves as a fixed criterion during mission execution, ensuring that the control logic has consistent triggering conditions.

[0060] The shock feedforward control mode is defined as a control state that adjusts the zero-bias correction and attitude control loops in advance based on the arrival time and intensity of the shock wave before its actual arrival. The goal of this mode is to complete compensation and parameter preparation before the shock event occurs, thereby reducing the transient impact of the shock on attitude estimation and attitude control.

[0061] Predictive compensation term injection is defined as follows: the predicted compensation term calculated in S4 is superimposed as an additional component onto the basic zero-bias correction in the output link of S2 to form the target zero-bias correction, which is then used in S5 to correct the IMU angular velocity. In practice, this injection is implemented by superimposing the zero-bias correction within a time window, without altering the input-output structure of the coupled compensation model.

[0062] The first thermal disturbance index is obtained by "time accumulation calculation of the temperature gradient change rate within a preset time window". To facilitate the proper use of data in subsequent writing, its meaning is supplemented as follows: Input: The temperature gradient change rate sequence within the preset time window, which is formed by the difference of temperature gradient vectors at consecutive sampling times. Meaning of time accumulation calculation: Accumulating the temperature gradient change rate within a fixed time window yields an index reflecting the total change in the spatial morphology of the thermal field within that window. This makes the index insensitive to transient spikes at individual sampling points and more sensitive to persistent thermal disturbances. Output: The first thermal disturbance index is a single-value index; the larger the value, the more drastic the change in the spatial temperature distribution in the IMU's region within that time window. For example, when a thermal plume sweeps past the left side of the drone and rapidly moves to the right side under the influence of wind, the direction and amplitude of the temperature gradient vector change continuously within a short window, causing the overall temperature gradient change rate sequence to rise, resulting in a higher first thermal disturbance index after time accumulation.

[0063] The second mechanical disturbance index is obtained by normalizing the arrival time and intensity of the shock wave. It maps "proximity" and "energy magnitude" to a unified scale, ensuring compatibility with thermal indicators. Arrival time normalization converts the shock wave arrival time into a quantitative result reflecting "proximity to the current moment"; the shorter the arrival time, the higher the proximity, and the higher the normalized output value. Intensity normalization converts the shock wave intensity into a quantitative result reflecting "impact energy magnitude"; the greater the intensity, the higher the normalized output value. The single-value index obtained by combining the normalized proximity and intensity information ensures that the mechanical disturbance exhibits a high value when "imminent and of high intensity." For example, after a deflagration event, the propagation distance estimated by the acoustic array shortens, and the predicted shock wave arrival time continuously decreases; simultaneously, the energy spectrum index and the rate of change of air pressure increase, leading to an increase in intensity output. After normalization, both the proximity and intensity values ​​increase simultaneously, and the second mechanical disturbance index increases accordingly.

[0064] The first thermal disturbance index and the second mechanical disturbance index are combined according to preset weights to obtain the joint thermal-mechanical shock predictor factor. The combination method is further explained as follows: The preset weights represent the relative contribution ratios of thermal and mechanical disturbance information in the joint prediction; the weights remain consistent during the same task execution to ensure the comparability of the joint predictor factor and the consistency of threshold determination. The joint predictor factor output is a single-value quantized result, used for comparison with the preset threshold and triggering control mode switching. The calculation cycle of the joint predictor factor is aligned with the temperature gradient change rate update cycle, and the shock wave arrival time and intensity update cycle in timestamps to ensure that the joint prediction is based on thermal and mechanical information at the same moment.

[0065] The amplitude of the predicted compensation term is determined by the shock wave intensity. The input data for the amplitude is the shock wave intensity. The higher the shock wave intensity, the greater the additional compensation amplitude for the zero-bias correction, allowing the angular velocity correction to cover the disturbance trend caused by the shock in advance. The predicted compensation term data must include at least three-axis compensation components, an action time window identifier, and an amplitude identifier. The predicted compensation term is only superimposed on the basic zero-bias correction within the action time window; outside the window, the output remains the basic zero-bias correction, ensuring that the scope of the compensation term corresponds consistently to the shock event.

[0066] Within the action time window, the attitude control loop parameters are adjusted based on the shock wave intensity. At the method level, this parameter adjustment is further explained as follows: the attitude control loop parameters refer to the set of control parameters used to generate attitude control commands in the attitude control calculation. In this method, the parameter set participates in the adaptive adjustment within the shock window as an adjustable object. The shock wave intensity is used as the sole intensity reference; the higher the intensity, the greater the corresponding parameter adjustment range, ensuring that the control loop's ability to suppress sudden disturbances matches the shock level.

[0067] Step S5: Correct the IMU angular velocity based on the IMU gyroscope zero bias correction amount and perform attitude calculation to obtain attitude estimation, and output attitude control command; The attitude calculation includes: determining an initial attitude estimate based on the UAV's stationary state at the initial moment; during flight, in each attitude calculation cycle, subtracting the IMU gyroscope zero bias correction obtained by the coupled compensation model and the prediction compensation term from the gyroscope angular velocity in the IMU measurement to obtain the corrected IMU angular velocity, and using the corrected IMU angular velocity as the angular rate input, updating the attitude estimate of the previous cycle through a quaternion-based attitude calculation algorithm to obtain the current attitude estimate result.

[0068] The attitude calculation cycle is defined as the cycle in which the attitude calculation algorithm performs an update once at a fixed time interval. This cycle is used to specify the time step of gyro angular velocity sampling, zero bias correction, attitude integral update, and attitude estimation output, so that the attitude estimation is continuous and traceable in the time series.

[0069] The corrected IMU angular velocity is defined as follows: the three-axis angular velocity obtained by subtracting the IMU gyroscope zero-bias correction from the IMU gyroscope angular velocity measurement output by S2, which is used as the sole angular rate input for attitude calculation. This angular velocity includes the additional correction effect brought by the prediction compensation term within the impact feedforward control mode window.

[0070] Initial attitude estimation is defined as: the initial attitude value determined based on the stationary state of the UAV at the initial moment, which is used to provide initial conditions for quaternion attitude calculation, so that subsequent periodic updates have a definite starting point.

[0071] Attitude estimation is defined as the attitude result output by the quaternion-based attitude calculation algorithm in each attitude calculation cycle, which serves as the attitude information of the inertial side.

[0072] The attitude control command is defined as a control output determined jointly by the attitude estimation and the attitude control loop parameters, used to drive the attitude control process to adjust the current attitude state. Within the operating time window of the shock feedforward control mode, the attitude control loop parameters used to generate the attitude control command are the parameter set adjusted by S4 based on the shock wave intensity.

[0073] Initial attitude estimation is determined based on the UAV's stationary state at the initial moment. The initial attitude estimation is explained as follows: The stationary state is determined because the UAV is initially stationary or equivalently stationary, and the gyro angular velocity measurement maintains a low variation amplitude for a short period, allowing the initial attitude estimation to serve as a stable starting point for subsequent quaternion updates. The initial attitude estimation output consists of the initial attitude values ​​stored as quaternions, associated with an initial timestamp, ensuring continuity of subsequent attitude estimations over time.

[0074] Within each attitude calculation cycle, the following fixed sequence is executed: acquire gyro angular velocity measurements, read the three-axis angular velocity measurements and timestamps output by the IMU within that cycle; acquire the zero-bias correction amount, read the IMU gyro zero-bias correction amount aligned with the timestamp of that cycle; this correction amount is output by S2 and includes a prediction compensation term within the S4 action time window; perform angular velocity correction, subtracting the zero-bias correction amount from the gyro angular velocity measurements to obtain the corrected IMU angular velocity; perform quaternion attitude update, using the corrected IMU angular velocity as the angular rate input, and updating the current attitude estimation result based on the attitude estimation of the previous cycle; output attitude control commands, using the current attitude estimation as input, and generating attitude control commands according to the attitude control loop parameters; when in the impact feedforward control mode window, the attitude control loop parameters are the parameters adjusted by S4. This sequence ensures that the input for attitude estimation is always "angular velocity corrected by temperature coupling compensation and impact feedforward compensation," and ensures that the attitude control commands remain synchronized with the control parameter adjustments caused by the impact intensity.

[0075] Step S6: When the shock wave actually arrives or the visual characteristics degrade, a health assessment is conducted based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. The reliability index is obtained by: calculating the degree to which the temperature characteristics of the area where the IMU is located deviate from the preset safety range based on the transient temperature field model when the shock wave actually arrives or the visual characteristics degrade, and obtaining the thermal environment anomaly index through normalization; extracting the shock wave intensity and duration based on the measured shock wave data, and obtaining the mechanical disturbance index by comparing the shock wave intensity with the preset shock threshold and then normalizing it; calculating the residual mean square value based on the residual between the IMU measurement value and the attitude estimation, and obtaining the measurement consistency index by converting it through a monotonically decreasing function according to the relationship between the residual mean square value and the preset error threshold; calculating the IMU reliability index by weighted averaging the thermal environment anomaly index, the mechanical disturbance index, and the measurement consistency index; and calculating the reliability of the sensors that acquire visual positioning data, the acoustic sensor array, and the barometric pressure sensor based on the measurement consistency of each sensor and its corresponding environmental disturbance index to obtain the reliability index of each sensor.

[0076] The actual arrival of the shock wave is defined as: the moment when measured characteristic changes consistent with the shock event occur based on acoustic and air pressure data, and meet preset judgment rules. This marker is different from the "predicted arrival time" in step S3, and is used to characterize that the shock disturbance has acted on the environment near the UAV and has a real impact on sensor measurements.

[0077] Visual feature degradation is defined as an event that indicates a decrease in the number of visual features used to generate visual positioning data, a decrease in matching consistency, or unstable observation within the same pose estimation update cycle. This event indicates that the reliability of the visual positioning data needs to be assessed for health and its weights need to be adjusted during that period.

[0078] Health assessment is defined as: a process of quantitatively assessing the degree of environmental disturbance and measurement self-consistency of the sensor during abnormal phases such as the actual arrival of the shock wave or the degradation of visual features, and outputting a set of credibility indicators for fusion weight adjustment.

[0079] The credibility index is defined as a quantitative indicator used to characterize the reliability of the measurement results of the corresponding sensor at the current moment or within the current time window. The credibility index is based on the environmental disturbance index and the measurement consistency index, obtained through a preset calculation rule, and serves as the input for the S7 weight mapping.

[0080] The thermal environment anomaly index is defined as: the normalized result of the deviation of the temperature characteristics calculated based on the spatial transient temperature field model from the preset safe range, which is used to quantify the "threat level of the thermal environment to the reliability of the sensor".

[0081] The mechanical disturbance index is defined as: an index obtained by extracting the measured shock wave intensity and duration, comparing the intensity with a preset shock threshold, and then normalizing it. It is used to quantify the "threat level of shock disturbance to sensor reliability".

[0082] The measurement consistency index is defined as follows: based on the mean square value of the residuals between the IMU measurements and the attitude estimates, compared with a preset error threshold, and transformed by a monotonically decreasing function, this index is used to quantify the "degree of consistency between measurement and estimation". This index decreases as the mean square value of the residuals increases, thus giving a lower consistency evaluation for abnormal measurements.

[0083] "Measured shock wave data" is used to extract shock wave intensity and duration, and based on this, to obtain mechanical disturbance indices. To avoid confusion, the following additional information is provided: The measured shock wave data source consists of acoustic and air pressure time series data near the actual arrival time of the shock wave. Shock wave intensity extraction involves extracting an intensity measure reflecting the shock amplitude from the measured data. This intensity is then compared to a preset shock threshold and normalized. Shock wave duration extraction involves extracting the length of time during which the shock disturbance is significantly present from the measured data. This duration characterizes the range of the disturbance's effect and is used in the health assessment phase to interpret the disturbance's impact within a time window.

[0084] The mechanical disturbance index is obtained by "normalizing the measured shock wave intensity after comparing it with a preset shock threshold". It is used to determine the intensity threshold at which the shock disturbance enters the significant impact range, serving as a benchmark for the health assessment stage. When the measured intensity is higher than the threshold, the corresponding disturbance threat level is higher; when the measured intensity is close to or lower than the threshold, the corresponding threat level is lower. The comparison results are mapped to a unified scale to ensure that the mechanical disturbance index and the thermal environment anomaly index participate in the credibility calculation under the same dimensional system.

[0085] The residual mean square value is calculated based on the residuals between IMU measurements and attitude estimates, and a measurement consistency index is obtained through a monotonically decreasing function transformation. The following points are added: Source of residuals: Within the same attitude estimation update cycle, the residuals represent the difference between the IMU measurements and the attitude estimates output in step S5. The residuals reflect the "degree of matching between the current measurement and the estimation model." Meaning of residual mean square value: Mean square statistics are performed on the residual amplitude within a preset sample length or time window to obtain a stable representation that is more sensitive to abnormal fluctuations and less sensitive to sign changes. Meaning of preset error threshold: A baseline threshold used to determine whether measurement consistency has entered the abnormal range. Meaning of monotonically decreasing function transformation: The relationship between the residual mean square value and the threshold is mapped to a consistency index, so that the larger the residual mean square value, the lower the consistency index, thus giving a lower confidence basis to abnormal measurements.

[0086] The weighted average structure combines three types of indicators with preset weights to output a single-value IMU reliability index, ensuring that the reliability simultaneously reflects information from the thermal environment, mechanical disturbances, and measurement self-consistency. The preset weights reflect the relative importance of the three types of indicators in the reliability assessment, and these weights remain consistent throughout the same task execution.

[0087] The visual sensor reliability index, representing the consistency of visual positioning data observations within the current update cycle, directly corresponds to the stability changes during the "visual feature degradation" trigger period. The thermal environment anomaly index and mechanical disturbance index serve as background environmental disturbance quantities for visual observation, used to downgrade visual reliability when smoke, thermal plumes, and impact disturbances coexist. For example, smoke obstruction and thermal plumes lead to a decrease in feature matching consistency, resulting in a lower visual reliability index.

[0088] The reliability index of the acoustic sensor array includes the stability of the arrival time difference between the acoustic sensors within the array and the consistency of the energy spectrum index, which reflects the self-consistency of the acoustic observations. The mechanical disturbance index serves as a background quantity for the effectiveness of acoustic observations and is used to evaluate the consistency of acoustic measurement reliability during the impact arrival period. For example, when the impact arrival leads to a significant enhancement of acoustic observations and the array time difference relationship is stable, the reliability of the acoustic array remains at a level matching the consistency of the observations.

[0089] The reliability indicators for barometric pressure sensors include the consistency of the barometric pressure sequence's change pattern and the stability of its rate of change during the impact period. These indicators reflect whether there is abnormal drift or noise distortion in the barometric pressure observation. Mechanical disturbance indicators and thermal environment anomaly indicators together serve as background parameters for barometric pressure measurement, used to determine the consistency of barometric pressure reliability when high temperatures and impacts are superimposed. For example, if the rate of change of barometric pressure at the arrival of the impact exhibits a change process consistent with the duration of the impact, the reliability of the barometric pressure sensor corresponds to this consistency.

[0090] Step S7: Generate a dynamic weight allocation matrix based on the credibility index, perform nonlinear fusion of visual positioning data and corrected inertial data, and output the final attitude estimation result.

[0091] The nonlinear fusion includes: converting the reliability index of each sensor into weight coefficients for visual positioning data and corrected inertial data in a dynamic weight allocation matrix according to a preset nonlinear mapping relationship; within each attitude estimation update cycle, using the dynamic weight allocation matrix to weight and combine the visual positioning data and corrected inertial data to output the final attitude estimation result.

[0092] The dynamic weight allocation matrix is ​​defined as follows: within each attitude estimation update cycle, it is expressed as a set of weights calculated based on the confidence indices of each sensor, used to dynamically allocate the contribution ratios of visual positioning data and corrected inertial data during fusion. This matrix updates as the confidence indices change, enabling the fusion result to suppress low-confidence measurements during abnormal phases.

[0093] The preset nonlinear mapping relationship is defined as: a set of mapping rules that convert credibility indicators into weight coefficients. The mapping relationship reflects the "non-proportional impact of credibility changes on weight allocation" in a nonlinear way, so that the weight decreases more significantly when credibility decreases, thereby enhancing the ability to suppress abnormal measurements.

[0094] Nonlinear fusion is defined as the fusion process that combines visual positioning data with corrected inertial data and outputs a fused posture under the constraint of a dynamic weight allocation matrix. The nonlinearity of nonlinear fusion is introduced by the "nonlinear mapping of weights". The fusion ontology maintains a weighted combination as its core structure, making the output interpretable and clearly correlated with the credibility index.

[0095] The final attitude estimation result is defined as the fused attitude result output in each attitude estimation update cycle, which serves as the optimal combination of visual and inertial information under the current environmental disturbance and measurement consistency conditions, and is used as the data basis for subsequent flight control and health assessment closed loop.

[0096] The reliability indices of each sensor are converted into weighting coefficients according to a preset nonlinear mapping relationship. The mapping input includes at least visual reliability indices and IMU reliability indices, corresponding to visual positioning data and corrected inertial data, respectively. The reliability indices of the acoustic array and barometric pressure sensor serve as supporting information for environmental disturbance assessment, explaining the reasons for changes in visual and IMU reliability. However, the weighting coefficient allocation directly affects both visual and inertial data. The mapping output is a set of weighting coefficients, including at least visual and inertial weighting coefficients. These two coefficients form paired weights within the same update cycle for weighted combination. When the reliability index is in a high value range, the weight changes remain smooth; when the reliability index decreases and enters an abnormal range, the weight decreases more significantly, rapidly reducing the impact of low-reliability data on the fusion result.

[0097] The dynamic weight allocation matrix in the method description includes the following fixed contents: visual weight coefficient, which corresponds to the fusion contribution ratio of visual positioning data in the update cycle; inertial weight coefficient, which corresponds to the fusion contribution ratio of corrected inertial data in the update cycle; timestamp, which is aligned with the confidence output timestamp and attitude estimation timestamp; and trigger identifier inheritance, which inherits the trigger event identifier of S6 to explain the source of weight changes and ensure the traceability of the data chain.

[0098] The method specifies the use of a dynamic weighting matrix to weight and combine visual positioning data with corrected inertial data. The combination objects are the attitude observations provided by the visual positioning data and the attitude estimates from the inertial side. Within the same update cycle, the two types of attitude information are weighted and combined according to the visual weight coefficients and inertial weight coefficients given by the dynamic weighting matrix to obtain the fused attitude. The weights are derived from a nonlinear mapping of the reliability index, making the fusion result closer to a high-reliability source under abnormal environments. When visual degradation occurs, the fusion result converges towards the inertial side; when inertial consistency decreases, the fusion result reduces its dependence on the inertial side while maintaining its contribution to the visual side, thereby avoiding attitude estimation distortion caused by anomalies in a single data source.

[0099] Example 2: A drone stabilization control system for high-rise building firefighting scenarios, see [link / reference]. Figure 1 As shown, it includes the following modules: Transient temperature module: Real-time acquisition of temperature data, construction of spatial transient temperature field model, and calculation of temperature characteristics; Coupling compensation module: Establishes a coupling compensation model between IMU zero bias and temperature data and temperature characteristics, and obtains the IMU gyroscope zero bias correction amount; Shock wave prediction module: Collects acoustic and air pressure data, constructs a spatiotemporal prediction model for shock waves, and outputs the arrival time and intensity of shock waves; Feedforward control module: Based on the temperature gradient change rate and the arrival time and intensity of the shock wave, the joint predictor of thermal-mechanical shock is obtained. When the joint predictor exceeds the preset threshold, the shock feedforward control mode is entered, the predictive compensation term is injected into the coupled compensation model and the attitude control loop parameters are adjusted. Attitude calculation module: Corrects the IMU angular velocity based on the IMU gyroscope zero bias correction and performs attitude calculation to obtain attitude estimate, and outputs attitude control command; Reliable estimation module: When the shock wave actually arrives or the visual features degrade, the module assesses the health of each sensor based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. Attitude estimation module: Generates a dynamic weight allocation matrix based on the confidence index, performs nonlinear fusion of visual positioning data and corrected inertial data, and outputs the final attitude estimation result.

[0100] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for stabilizing and controlling a drone in a high-rise firefighting scenario, characterized in that, include: Temperature data is collected in real time by temperature sensors located in the area where the IMU is located, a spatial transient temperature field model is constructed, and temperature characteristics, including temperature gradient vector, temperature rise rate, and temperature gradient change rate, are calculated. Establish a coupling compensation model between IMU bias and temperature data and temperature characteristics to obtain the IMU gyroscope bias correction amount; Acoustic data is collected by an acoustic sensor array and air pressure data is collected by a barometer. The energy spectrum, time difference of arrival and rate of change of air pressure in the preset frequency band are extracted to construct a spatiotemporal prediction model of shock wave and output the arrival time and intensity of shock wave. Based on the temperature gradient change rate and the arrival time and intensity of the shock wave, the joint predictor of thermal-mechanical shock is obtained. When the joint predictor exceeds the preset threshold, the shock feedforward control mode is entered, and the predictive compensation term is injected into the coupled compensation model and the attitude control loop parameters are adjusted. The attitude estimate is obtained by correcting the IMU angular velocity based on the IMU gyroscope zero bias correction and performing attitude calculation, and then outputting attitude control commands. When the shock wave actually arrives or the visual characteristics degrade, the health status is assessed based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. A dynamic weight allocation matrix is ​​generated based on the credibility index, and the visual positioning data and the corrected inertial data are nonlinearly fused to output the final attitude estimation result.

2. The method for stable control of a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The construction of the spatial transient temperature field model includes: mapping the installation positions of temperature sensors located at different spatial locations in the IMU area to three-dimensional coordinate points; collecting temperature data in real time through each temperature sensor within a preset sampling period; and establishing a spatial transient temperature field model in the IMU area with spatial coordinates and time as independent variables and temperature as the dependent variable through interpolation based on the temperature data. In the spatial transient temperature field model, the temperature gradient vector is obtained by calculating the first-order partial derivative of temperature in the spatial direction, the temperature rise rate is obtained by calculating the first-order derivative of temperature in the time direction, and the temperature gradient change rate is obtained by differentiating the temperature gradient vector within a preset time window.

3. The method for stabilizing and controlling a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The acquisition of the IMU gyroscope zero-bias correction includes: during the calibration phase, collecting the measured values ​​of the IMU gyroscope under known zero-bias conditions, along with corresponding temperature data and temperature characteristics, under different temperature distribution conditions. The temperature data and characteristics are used as input variables, and the actual zero-bias of each gyroscope axis is used as the output variable to construct a sample dataset. Based on the sample dataset, a zero-bias coupling compensation function is established using a multivariate fitting algorithm, with temperature data and temperature characteristics as input and a basic zero-bias correction as output, serving as the coupling compensation model. During actual flight, the current temperature data and temperature characteristics are input into the zero-bias coupling compensation function to obtain the basic zero-bias correction. A prediction compensation term obtained from the thermal-mechanical shock joint prediction factor is then superimposed on the basic zero-bias correction to obtain the IMU gyroscope zero-bias correction.

4. The method for stabilizing and controlling a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The output of the shock wave arrival time and intensity includes: determining the propagation direction of the shock wave front relative to the UAV based on the arrival time difference of acoustic data received by each acoustic sensor in the acoustic sensor array, and estimating the propagation distance from the shock wave front to the UAV based on the spatial distance between each acoustic sensor; performing spectral analysis on the acoustic data within a preset frequency band to obtain an energy spectrum index characterizing the acoustic energy of the shock wave, and calculating the air pressure change rate from the air pressure data collected by the air pressure sensor; calculating the shock wave arrival time based on the propagation distance and a preset shock wave propagation speed, where the shock wave arrival time is equal to the ratio of the propagation distance to the preset shock wave propagation speed; and calculating the shock wave intensity based on the energy spectrum index and the air pressure change rate using a weighted summation method, where the shock wave intensity is obtained by multiplying the energy spectrum index and the air pressure change rate by preset weighting coefficients and then summing them.

5. The method for stable control of a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The acquisition of the thermal-mechanical shock joint prediction factor includes: performing time-accumulation calculation on the temperature gradient change rate within a preset time window to obtain a first thermal disturbance index to characterize the degree of drastic temperature change in the area where the IMU is located; based on the arrival time and intensity of the shock wave, normalizing the arrival time of the shock wave to reflect its proximity to the current moment, and normalizing the intensity of the shock wave to reflect the magnitude of the shock wave's impact energy to obtain a second mechanical disturbance index; and combining the first thermal disturbance index and the second mechanical disturbance index according to preset weights to obtain the thermal-mechanical shock joint prediction factor.

6. The method for stabilizing and controlling a drone in a high-rise firefighting scenario according to claim 3, characterized in that, The shock feedforward control mode includes: determining the action time window and amplitude of the prediction compensation term based on the arrival time and intensity of the shock wave; superimposing the prediction compensation term with the basic zero bias correction amount determined by the coupled compensation model within the action time window to obtain the target zero bias correction amount used to correct the IMU angular velocity; and adjusting the attitude control loop parameters according to the shock wave intensity within the action time window.

7. The method for stabilizing and controlling a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The attitude calculation includes: determining an initial attitude estimate based on the UAV's stationary state at the initial moment; during flight, in each attitude calculation cycle, subtracting the IMU gyroscope zero bias correction obtained by the coupled compensation model and the prediction compensation term from the gyroscope angular velocity in the IMU measurement to obtain the corrected IMU angular velocity, and using the corrected IMU angular velocity as the angular rate input, updating the attitude estimate of the previous cycle through a quaternion-based attitude calculation algorithm to obtain the current attitude estimate result.

8. The method for stable control of a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The reliability index is obtained by: calculating the degree to which the temperature characteristics of the area where the IMU is located deviate from the preset safety range based on the transient temperature field model when the shock wave actually arrives or the visual characteristics degrade, and obtaining the thermal environment anomaly index through normalization; extracting the shock wave intensity and duration based on the measured shock wave data, and obtaining the mechanical disturbance index by comparing the shock wave intensity with the preset shock threshold and then normalizing it; calculating the residual mean square value based on the residual between the IMU measurement value and the attitude estimation, and obtaining the measurement consistency index by converting it through a monotonically decreasing function according to the relationship between the residual mean square value and the preset error threshold; calculating the IMU reliability index by weighted averaging the thermal environment anomaly index, the mechanical disturbance index, and the measurement consistency index; and calculating the reliability of the sensors that acquire visual positioning data, the acoustic sensor array, and the barometric pressure sensor based on the measurement consistency of each sensor and its corresponding environmental disturbance index to obtain the reliability index of each sensor.

9. The method for stabilizing and controlling a drone in a high-rise firefighting scenario according to claim 1, characterized in that, The nonlinear fusion includes: converting the reliability index of each sensor into weight coefficients for visual positioning data and corrected inertial data in a dynamic weight allocation matrix according to a preset nonlinear mapping relationship; within each attitude estimation update cycle, using the dynamic weight allocation matrix to weight and combine the visual positioning data and corrected inertial data to output the final attitude estimation result.

10. A drone stabilization control system for high-rise building firefighting scenarios, characterized in that, The system employs a drone stabilization control method for high-rise firefighting scenarios as described in any one of claims 1 to 9, comprising: Transient temperature module: Real-time acquisition of temperature data, construction of spatial transient temperature field model, and calculation of temperature characteristics; Coupling compensation module: Establishes a coupling compensation model between IMU zero bias and temperature data and temperature characteristics, and obtains the IMU gyroscope zero bias correction amount; Shock wave prediction module: Collects acoustic and air pressure data, constructs a spatiotemporal prediction model for shock waves, and outputs the arrival time and intensity of shock waves; Feedforward control module: acquires the joint predictor of thermal-mechanical shock, and enters the shock feedforward control mode when the joint predictor exceeds the preset threshold; Attitude calculation module: Corrects the IMU angular velocity based on the IMU gyroscope zero bias correction and performs attitude calculation to obtain attitude estimate, and outputs attitude control command; Reliable estimation module: When the shock wave actually arrives or the visual features degrade, the module assesses the health of each sensor based on the transient temperature field model, measured shock wave data, and the residual between IMU measurements and attitude estimation to obtain the reliability index of each sensor. Attitude estimation module: Generates a dynamic weight allocation matrix based on the confidence index, performs nonlinear fusion of visual positioning data and corrected inertial data, and outputs the final attitude estimation result.

Images