Robot fire-fighting inspection early-warning supervision system

Abstract

The invention provides a robot fire-fighting inspection early-warning supervision system, and relates to the technical field of fire-fighting inspection, the robot fire-fighting inspection early-warning supervision system comprises a robot body and a master control system, the robot body is internally provided with a multi-mode sensing array and an autonomous navigation unit, and the master control system comprises a dynamic prediction engine, a data fusion platform and a remote communication terminal. The remote communication terminal is used for connecting the robot body and the master control system for data communication and supervision; the multi-modal sensing array integrates a plurality of sensing units and is used for collecting sensing data in a plurality of types; according to the invention, a multi-modal sensing array is integrated, multiple kinds of sensing data are collected, synchronous monitoring of multiple parameters is realized, a laser radar and visual SLAM fusion technology is adopted, a three-dimensional risk map is constructed, the precision of autonomous navigation is improved, a double-Kalman filtering algorithm is adopted, dynamic noise reduction is carried out on laser radar data, and the accuracy of autonomous navigation is improved. And the navigation precision can still be kept in an external complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of fire inspection, and particularly to a robot fire inspection early warning supervision system. Background Art

[0002] The fire inspection early warning supervision system is an intelligent solution integrating Internet of Things, artificial intelligence and robot technologies. By deploying multi-parameter environmental perception devices and autonomous mobile inspection robots, it realizes real-time monitoring of fire hazards, significantly improving the fire prevention and control capabilities and response efficiency in complex scenarios;

[0003] The existing systems mainly have the following disadvantages:

[0004] Most of the fire robots on the market are single-function devices, lacking the ability of multi-parameter comprehensive monitoring and having insufficient autonomous navigation accuracy;

[0005] The existing systems mostly adopt threshold-triggered early warning, lacking the ability of dynamic prediction of fire hazards and unable to realize the transformation from "post-disaster response" to "pre-disaster prevention";

[0006] Traditional robot systems rely on threshold-triggered alarms, are vulnerable to environmental interference (such as dust, steam), and the false alarm rate can reach more than 40%, increasing the operation and maintenance costs. Moreover, they fail to realize the fusion analysis of multi-sensor data, for example, the composite early warning of sudden temperature rise and CO concentration change is not associated;

[0007] Therefore, the present invention proposes a robot fire inspection early warning supervision system to solve the problems existing in the prior art. Summary of the Invention

[0008] In view of the above problems, the present invention proposes a robot fire inspection early warning supervision system. The robot fire inspection early warning supervision system integrates a multi-modal perception array, collects various types of sensing data, realizes synchronous monitoring of multiple parameters, adopts the fusion technology of lidar and visual SLAM to construct a three-dimensional risk map, improves the accuracy of autonomous navigation, and adopts a dual Kalman filtering algorithm to dynamically denoise the lidar data, still maintaining the navigation accuracy in the complex external environment.

[0009] To achieve the object of the present invention, the present invention is realized through the following technical solutions: A robot fire inspection early warning supervision system, including a robot body and a total control system. The robot body is internally provided with a multi-modal perception array and an autonomous navigation unit. The total control system includes a dynamic prediction engine, a data fusion platform and a remote communication terminal. The remote communication terminal is used to connect the robot body and the total control system for data communication and supervision;

[0010] The multi-modal perception array integrates multiple sensing units for collecting sensing data of various types. The autonomous navigation unit fuses IMU, visual SLAM, and an adaptive Kalman filtering algorithm to achieve three-dimensional spatial positioning. The dynamic prediction engine is based on a fire data model, introduces a fuzzy logic algorithm, and conducts fire hazard prediction and risk handling deduction through association rule mining. The data fusion platform is used to build a multi-sensor data fusion analysis model, correlate changes in sensing parameters, and conduct composite risk early warning.

[0011] A further improvement is that the multi-modal perception array includes a thermal imager, a temperature and humidity sensor, a combustible gas sensor, a smoke sensor, a spectral analyzer, an ultrasonic level gauge, a MEMS inertial unit, a lidar, and a camera. The spectral analyzer is used for flame feature recognition, and the ultrasonic level gauge is used to monitor the water level of the fire pool.

[0012] A further improvement is that the autonomous navigation unit adopts a visual SLAM fusion technology based on lidar and camera, fuses an inertial measurement unit IMU based on a MEMS inertial unit, constructs a three-dimensional risk map with a resolution of 0.05 meters to identify various details of the spatial orientation, and at the same time, automatically generates an inspection path based on the three-dimensional risk map, preferentially covering high-risk areas. The three-dimensional risk map is transmitted to the master control system through a remote communication terminal, stretched with vector parameters of points, lines, and surfaces to construct a three-dimensional risk model, and the data layer of various sensing data in the multi-modal perception array is superimposed.

[0013] A further improvement is that the autonomous navigation unit is built-in with a Kalman filtering algorithm for dynamically denoising lidar data, controlling the navigation accuracy to be ±0.15 meters in an environment with a dust concentration of 150 mg / m 3 The autonomous navigation unit is built-in with a dynamic obstacle detection unit that jointly identifies dynamic obstacles using the optical flow method and a deep learning classifier.

[0014] A further improvement is that the dynamic prediction engine is built-in with an XGBoost model containing N historical fire data. Through association rule mining, fire hazard prediction is carried out N hours in advance, and at the same time, a fuzzy logic algorithm is introduced to fuse composite indicators such as a sudden temperature rise > 5 °C / min, a CO concentration > 500 ppm, and smoke particles > 2 μm for prediction. The specific steps are as follows:

[0015] Reconstruct the three-dimensional risk model and fuse the real-time point cloud based on the data collected by the multi-modal perception array;

[0016] Use the Kriging spatial interpolation algorithm to generate a risk heat map and convert it into an early fire hazard prediction;

[0017] The fire spread is simulated based on the physical engine and risk management is deduced.

[0018] Further improvements are to generate risk heat maps and convert them into early fire hazard predictions, which specifically include the following steps:

[0019] Perform Delaunay triangulation on discrete sampling points;

[0020] Calculate the risk value of each vertex using the formula: R = α·T'+β·G+γ·P, where T' is the temperature change rate, G is the gas concentration gradient, and P is the particle concentration;

[0021] The radial basis function (RBF) is used for spatial interpolation to generate a dynamically updated pseudo-color three-dimensional thermal map for early fire hazard prediction.

[0022] Further improvements are: simulating the spread of fire based on the physical engine and conducting risk management deduction, including the following steps:

[0023] Identify spatial coordinates, initial temperature, temperature change rate, combustible type, and automatically match the combustion curve;

[0024] Set wind speed, ventilation system status, and external fire response time;

[0025] Adopt variable step-size integration algorithm to strike a balance between simulation speed and accuracy;

[0026] Dynamically generate evacuation restricted areas based on thermal radiation flux and smoke layer height;

[0027] The fire-fighting robot path is planned based on the ant colony optimization algorithm to ensure that 80% of high-risk areas are covered within 5 minutes and generate a set of equipment control instructions.

[0028] Further improvements are as follows: the multi-sensor data fusion analysis model adopts an LSTM neural network to learn the nonlinear correlation characteristics between sudden temperature rise and abnormal changes in gas concentration, identify coupling anomalies, and capture micro-change correlations in sensor data by setting dynamic thresholds, thereby judging the current firefighting specific situation.

[0029] Further improvements are as follows: the data fusion platform has a built-in multi-dimensional time series matrix, which extracts the characteristics of the current fire situation through the TCN time series convolutional network, calculates the risk probability using the Bayesian network, and triggers a graded warning when the probability value exceeds the adaptive threshold. The adaptive threshold is automatically adjusted according to the environmental baseline fluctuations and dynamically corrected in combination with historical false alarm records.

[0030] A further improvement is that the remote communication terminal provides the functions of viewing inspection images, receiving warning information, and remotely controlling the robot body, and supports multi-robot collaborative supervision function for connecting to several robot bodies at the same time.

[0031] The beneficial effects of the present invention are as follows:

[0032] 1. The present invention integrates a multi-modal perception array to collect various types of sensing data, realizes synchronous monitoring of multiple parameters, adopts the fusion technology of lidar and visual SLAM to construct a three-dimensional risk map, improves the accuracy of autonomous navigation, and uses a dual Kalman filtering algorithm to dynamically denoise lidar data, so as to maintain the navigation accuracy under complex external environments.

[0033] 2. Based on the XGBoost model containing a number of historical fire data, the present invention realizes early fire hazard prediction through association rule mining. The fuzzy logic algorithm is introduced to fuse composite indicators for fire hazard prediction, and based on the physical engine to simulate the spread of fire for risk treatment deduction, that is, early prediction of hazards and a braking treatment route, which is more reliable.

[0034] 3. The present invention adopts an LSTM neural network to learn the non-linear correlation features between sudden temperature rise and abnormal gas concentration changes, identify coupled anomalies, and capture the micro-change correlations in sensing data by setting dynamic thresholds, so as to judge the current specific fire situation, improve the accuracy of alarms, and avoid false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a composition diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] In order to deepen the understanding of the present invention, the following will further elaborate on the present invention in combination with embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation to the protection scope of the present invention.

[0037] Embodiment 1

[0038] According to Figure 1 as shown, this embodiment proposes a robot fire inspection warning and supervision system, including a robot body and a total control system. A multi-modal perception array and an autonomous navigation unit are built in the robot body. The total control system includes a dynamic prediction engine, a data fusion platform and a remote communication terminal. The remote communication terminal is used to connect the robot body and the total control system for data communication and supervision;

[0039] The multi-modal perception array integrates multiple sensing units for collecting various types of sensing data. The autonomous navigation unit fuses IMU, visual SLAM, and an adaptive Kalman filtering algorithm to achieve three-dimensional space positioning. The dynamic prediction engine is based on a fire data model, introduces a fuzzy logic algorithm, and through association rule mining, conducts fire hazard prediction and risk handling deduction. The data fusion platform is used to construct a multi-sensor data fusion analysis model, correlate changes in sensing parameters, and conduct composite risk early warning. In this invention, a multi-modal perception array is integrated to collect various types of sensing data, realizing synchronous monitoring of multiple parameters. The lidar and visual SLAM fusion technology is adopted to construct a three-dimensional risk map, improving the accuracy of autonomous navigation. The dual Kalman filtering algorithm is used to dynamically denoise lidar data, and the navigation accuracy can still be maintained in a complex external environment. And based on an XGBoost model containing several historical fire data, through association rule mining, early fire hazard prediction is achieved. The fuzzy logic algorithm is introduced to fuse composite indicators for fire hazard prediction, and based on a physical engine to simulate the spread of fire, risk handling deduction is carried out, that is, potential hazards are predicted in advance and there are braking treatment routes, which is more reliable.

[0040] The multi-modal perception array includes an infrared thermal imager, a temperature and humidity sensor, a combustible gas sensor, a smoke sensor, a spectral analyzer, an ultrasonic level gauge, a MEMS inertial unit, a lidar, and a camera. The spectral analyzer is used for flame feature recognition, and the ultrasonic level gauge is used to monitor the water level of the fire pool. The lidar scanning accuracy is ≥0.1 mm, and a three-dimensional space risk model is constructed by combining infrared thermal imaging data, with a temperature field resolution of ≤0.5 °C / m 3 , the working band of the infrared thermal imager is 8-14 μm, the temperature measurement accuracy is ±2 °C, and it supports early fire temperature anomaly detection. The combustible gas sensor uses the electrochemical principle, with a detection range of 0-100% LEL and a response time of ≤5 seconds, and has the characteristics of anti-poisoning and anti-high-concentration flooding. The smoke sensor uses the optoelectronic principle, with a sensitivity of 0.1 dB / m, supports visibility detection, and the minimum detectable particle size is 0.3 μm. Gas mass spectrometer (detection range 0~10 6 ppm), MEMS inertial unit (angular velocity range ±2000 ° / s); spectral analyzer (wavelength range 380~780 nm), used for flame feature recognition; ultrasonic level gauge (accuracy ±1 mm), used to monitor the water level of the fire pool.

[0041] The autonomous navigation unit adopts the vision SLAM fusion technology based on lidar and camera, integrates the inertial measurement unit IMU based on the MEMS inertial unit, constructs a three-dimensional risk map with a resolution of 0.05 meters to identify various details of the spatial orientation, and at the same time, automatically generates an inspection path based on the three-dimensional risk map, preferentially covering high-risk areas. The three-dimensional risk map is transmitted to the master control system through a remote communication terminal, stretched with vector parameters of points, lines and planes, and a three-dimensional risk model is constructed, superimposing the data layer of various sensing data in the multi-modal sensing array. The autonomous navigation unit is built-in with a Kalman filtering algorithm to dynamically denoise the lidar data and control the navigation accuracy to be ±0.15 meters under the environment of dust concentration of 150mg / m 3 Under the environment, the navigation accuracy is maintained at ±0.15 meters. The autonomous navigation unit is built-in with a dynamic obstacle detection unit, which jointly identifies dynamic obstacles by using the optical flow method and a deep learning classifier. Integrating the inertial measurement unit (IMU), vision SLAM and adaptive Kalman filtering algorithm, the navigation accuracy is improved to ±5cm, supporting three-dimensional space risk modeling, and also having the following functions: associating building BIM data when establishing a three-dimensional risk model, dynamically adjusting the inspection path according to the FDI value (dense scanning in high-risk areas), and generating hidden danger disposal suggestions (including ventilation strategies and evacuation paths). Establish the initial pose estimation of the visual inertial odometer (VIO); fuse the millimeter-wave radar point cloud to construct an obstacle map; introduce UWB anchor point constraints to accumulate errors; adopt factor graph optimization to realize multi-source data fusion.

[0042] The dynamic prediction engine is built-in with an XGBoost model containing N pieces of historical fire data, and through association rule mining, fire hazard prediction is carried out N hours in advance. At the same time, a fuzzy logic algorithm is introduced to fuse the composite indicators of temperature sudden rise > 5℃ / min, CO concentration > 500ppm, and smoke particles > 2μm for prediction. The specific steps are as follows:

[0043] Reconstruct the three-dimensional risk model and fuse the real-time point cloud based on the data collected by the multi-modal sensing array;

[0044] Adopt the Kriging spatial interpolation algorithm to generate a risk heat map and convert it into an early fire hazard prediction;

[0045] Based on the physical engine, simulate the spread of fire and conduct risk treatment deduction.

[0046] Generate a risk heat map and convert it into an early fire hazard prediction. The specific steps are as follows:

[0047] Perform Delaunay triangulation on the discrete sampling points;

[0048] Calculate the risk values of each vertex using the formula: R = α·T'+β·G+γ·P, where T' is the temperature change rate, G is the gas concentration gradient, and P is the particle concentration;

[0049] Use the Radial Basis Function (RBF) for spatial interpolation to generate a dynamically updated pseudo-color three-dimensional thermal map for early fire hazard prediction.

[0050] Simulate the fire spread based on a physics engine and conduct risk processing and deduction, including the following steps:

[0051] Identify the spatial coordinates, initial temperature, temperature change rate, and combustible type, and automatically match the combustion curve;

[0052] Set the wind speed, ventilation system status, and external fire response time;

[0053] Adopt a variable step-size integration algorithm to balance the simulation speed and accuracy;

[0054] Dynamically generate evacuation restricted areas based on the heat radiation flux and the height of the smoke layer;

[0055] Plan the path of the fire-fighting robot based on the ant colony optimization algorithm to ensure that 80% of the high-risk areas are covered within 5 minutes, and generate a set of equipment control instructions.

[0056] Implement real-time coupled calculations of combustion kinetics, fluid mechanics, and structural mechanics, with the error rate reduced by 42% compared to traditional single-field models. It also has a data assimilation and correction mechanism: compare the measured temperature / smoke data transmitted back by the robot with the simulation results every 30 seconds, and use the Kalman filter algorithm to dynamically correct the fire source parameters; automatically switch the deduction strategy according to the fire development stage: in the initial fire stage (<100kW): focus on simulating the smoke filling path and optimizing the evacuation route selection. In the intense stage (>1MW): calculate the effectiveness of the fire compartment and predict the flashover time point (temperature exceeds 600°C and radiation flux > 20kW / m 2 )

[0057] The multi-sensor data fusion analysis model uses an LSTM neural network to learn the non-linear correlation features between sudden temperature rises and abnormal changes in gas concentration, identify coupled anomalies, and capture micro-change correlations in sensor data by setting dynamic thresholds, thereby determining the specific current fire situation. The data fusion platform has a built-in multi-dimensional time series matrix, extracts the features of the specific current fire situation through a TCN temporal convolutional network, calculates the risk probability using a Bayesian network, and triggers a hierarchical warning when the probability value exceeds the adaptive threshold. The adaptive threshold is automatically adjusted according to environmental baseline fluctuations and dynamically corrected in combination with historical false alarm records. The data fusion platform is the core hub of this system, and it innovatively constructs a multi-sensor data fusion analysis model. The platform first integrates real-time data streams from temperature and humidity sensors, combustible gas sensors (focusing on monitoring CO concentration), smoke sensors, and infrared thermal imagers, and performs spatio-temporal alignment and noise removal on these heterogeneous data through the Kalman filter algorithm to ensure data quality. Then, an improved LSTM neural network is used to establish a multi-parameter correlation model, which can learn the non-linear correlation features between sudden temperature rises and abnormal changes in CO concentration - for example, when the temperature in a certain area rises by 5°C within a short time while the CO concentration increases by 0.02%, the model will immediately identify this coupled anomaly. By setting dynamic thresholds (instead of fixed thresholds), the system can capture micro-change correlations between parameters, combine Bayesian network inference of the fire occurrence probability, and finally issue a composite risk warning before the temperature reaches the preset threshold, achieving pre-disaster prevention in a true sense.

[0058] Embodiment 2

[0059] According to Figure 1 As shown, this embodiment proposes a robot fire inspection and early warning supervision system, including a robot body and a master control system. The robot body has a built-in multi-modal perception array and an autonomous navigation unit. The master control system includes a dynamic prediction engine, a data fusion platform, and a remote communication terminal. The remote communication terminal is used to connect the robot body and the master control system for data communication and supervision;

[0060] The multimodal perception array integrates multiple sensing units for collecting various types of sensing data. The autonomous navigation unit fuses IMU, visual SLAM, and an adaptive Kalman filter algorithm to achieve three-dimensional space positioning. The dynamic prediction engine is based on a fire data model, introduces a fuzzy logic algorithm, and through association rule mining, conducts fire hazard prediction and risk handling deduction. The data fusion platform is used to construct a multi-sensor data fusion analysis model, correlate changes in sensing parameters, and conduct composite risk early warning. In this invention, a multimodal perception array is integrated to collect various types of sensing data, realizing synchronous monitoring of multiple parameters. The lidar and visual SLAM fusion technology is adopted to construct a three-dimensional risk map, improving the accuracy of autonomous navigation. The dual Kalman filter algorithm is used to dynamically denoise lidar data, and the navigation accuracy can still be maintained in a complex external environment. And based on an XGBoost model containing several historical fire data, through association rule mining, early fire hazard prediction is achieved. The fuzzy logic algorithm is introduced to fuse composite indicators for fire hazard prediction, and based on a physical engine to simulate the spread of fire, risk handling deduction is carried out, that is, potential hazards are predicted in advance and there are braking treatment routes, which is more reliable.

[0061] The multi-sensor data fusion analysis model uses an LSTM neural network to learn the non-linear correlation features between sudden temperature rise and abnormal gas concentration changes, identify coupled anomalies, and capture micro-change correlations in sensing data by setting dynamic thresholds, thereby judging the current specific fire situation. The data fusion platform is built with a multi-dimensional time series matrix, extracts the features of the current specific fire situation through a TCN temporal convolutional network, and uses a Bayesian network to calculate the risk probability. When the probability value exceeds the adaptive threshold, a hierarchical early warning is triggered. The adaptive threshold is automatically adjusted according to environmental baseline fluctuations and dynamically corrected in combination with historical false alarm records. The data fusion platform is the core hub of this system, and it innovatively constructs a multi-sensor data fusion analysis model. The platform first integrates real-time data streams from temperature and humidity sensors, combustible gas sensors (focusing on monitoring CO concentration), smoke sensors, and infrared thermal imagers, and uses the Kalman filter algorithm to perform spatio-temporal alignment and noise removal on these heterogeneous data to ensure data quality. Then, an improved LSTM neural network is used to establish a multi-parameter correlation model, which can learn the non-linear correlation features between sudden temperature rise and abnormal CO concentration changes - for example, when the temperature in a certain area rises by 5°C within a short time and the CO concentration increases by 0.02%, the model will immediately identify this coupled anomaly. By setting dynamic thresholds (instead of fixed thresholds), the system can capture micro-change correlations between parameters, combine with Bayesian network reasoning to calculate the probability of fire occurrence, and finally issue a composite risk early warning before the temperature reaches the preset threshold, achieving true pre-disaster prevention.

[0062] The remote communication terminal provides functions such as viewing inspection screens, receiving early warning information, and remotely controlling the robot body, and supports the multi-robot collaborative supervision function for simultaneously accessing several robot bodies. It also has a communication function: locally deploy an MQTT proxy server (with a latency <50 ms), use slicing technology for the 5G link (ensuring 20% bandwidth priority), and the offline continuous transmission cache duration ≥72 hours.

[0063] The robot fire inspection early warning and supervision system integrates a multi-modal perception array to collect various types of sensing data, realizes synchronous monitoring of multiple parameters, adopts the fusion technology of lidar and visual SLAM to construct a three-dimensional risk map, improves the accuracy of autonomous navigation, and uses a dual Kalman filtering algorithm to dynamically denoise the lidar data, so as to maintain the navigation accuracy even in a complex external environment. Moreover, the present invention is based on an XGBoost model containing several historical fire data, and through association rule mining, realizes early prediction of fire hazards. Introduce a fuzzy logic algorithm to fuse composite indicators for fire hazard prediction, and simulate the spread of fire based on a physical engine for risk handling deduction, that is, predict hazards in advance and have a braking treatment route, which is more reliable. At the same time, use an LSTM neural network to learn the non-linear correlation characteristics between sudden temperature rise and abnormal gas concentration changes, identify coupled anomalies, and capture micro-change correlations in sensing data by setting dynamic thresholds, so as to judge the current fire situation and improve the accuracy of alarms and avoid false alarms.

[0064] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. The robot fire inspection warning and supervision system includes a robot body and a master control system, and is characterized in that: The robot body is built-in with a multi-modal perception array and an autonomous navigation unit. The master control system includes a dynamic prediction engine, a data fusion platform, and a remote communication terminal. The remote communication terminal is used to connect the robot body and the master control system for data communication and supervision; The multi-modal perception array integrates multiple sensing units for collecting various types of sensing data. The autonomous navigation unit fuses IMU, visual SLAM, and an adaptive Kalman filter algorithm to achieve three-dimensional space positioning. The dynamic prediction engine is based on a fire data model, introduces a fuzzy logic algorithm, and through association rule mining, conducts fire hazard prediction and risk handling deduction. The data fusion platform is used to build a multi-sensor data fusion analysis model, correlate changes in sensing parameters, and conduct composite risk early warning.

2. The robot fire inspection warning and supervision system according to claim 1, characterized in that: The multi-modal perception array includes an infrared thermal imager, a temperature and humidity sensor, a combustible gas sensor, a smoke sensor, a spectral analyzer, an ultrasonic level gauge, a MEMS inertial unit, a lidar, and a camera. The spectral analyzer is used for flame feature recognition, and the ultrasonic level gauge is used to monitor the water level of the fire pool.

3. The robot fire inspection warning and supervision system according to claim 2, characterized in that: The autonomous navigation unit adopts a visual SLAM fusion technology based on lidar and camera, and fuses an inertial measurement unit IMU based on a MEMS inertial unit to build a three-dimensional risk map with a resolution of 0.05 meters, which is used to identify various details of the spatial orientation. At the same time, based on the three-dimensional risk map, a patrol path is automatically generated, giving priority to covering high-risk areas. The three-dimensional risk map is transmitted to the master control system through the remote communication terminal, stretched with vector parameters of points, lines, and surfaces to build a three-dimensional risk model, and superimposed on the data layer of various sensing data in the multi-modal perception array.

4. The robot fire inspection, early warning, supervision and management system according to claim 3, characterized in that: The autonomous navigation unit incorporates a Kalman filtering algorithm for dynamically reducing noise in lidar data, controlling to maintain a navigation accuracy of ±0.15 meters under an environment with a dust concentration of 150 mg / m 3 . The autonomous navigation unit is built-in with a dynamic obstacle detection unit, which jointly identifies dynamic obstacles using the optical flow method and a deep learning classifier.

5. The robot fire inspection, early warning and supervision system according to claim 1, characterized in that: The dynamic prediction engine is built-in with an XGBoost model containing N pieces of historical fire data. Through association rule mining, it conducts fire hazard prediction N hours in advance. At the same time, it introduces a fuzzy logic algorithm and fuses a composite index of temperature sudden rise > 5°C / min, CO concentration > 500 ppm, and smoke particles > 2 μm for prediction. The specific steps are as follows: Reconstruct the three-dimensional risk model and fuse the real-time point cloud based on the data collected by the multi-modal perception array; Use the Kriging spatial interpolation algorithm to generate a risk heat map and convert it into an early fire hazard prediction; Simulate the spread of the fire based on the physical engine and conduct risk handling deduction.

6. The robot fire inspection early warning supervision system according to claim 5, characterized in that: Generate a risk heat map and convert it into an early fire hazard prediction. The specific steps are as follows: Conduct Delaunay triangulation on the discrete sampling points; Calculate the risk value of each vertex. The formula is: R = α·T'+β·G+γ·P, where T' is the temperature change rate, G is the gas concentration gradient, and P is the particle concentration; Use the radial basis function RBF for spatial interpolation to generate a dynamically updated pseudo-color three-dimensional heat map for early fire hazard prediction.

7. The robot fire inspection, early warning, supervision and management system according to claim 6, characterized in that: Simulate the spread of the fire based on the physical engine and conduct risk handling deduction, including the following steps: Identify the spatial coordinates, initial temperature, temperature change rate, and combustible type, and automatically match the combustion curve; Set the wind speed, ventilation system status, and external fire response time; Adopt a variable step - size integration algorithm to balance between simulation speed and accuracy; Dynamically generate evacuation restricted areas according to the heat radiation flux and the height of the smoke layer; Based on the ant colony optimization algorithm, plan the path of the fire - fighting robot to ensure that 80% of the high - risk areas are covered within 5 minutes and generate a set of equipment control instructions.

8. The robot fire inspection early warning supervision system according to claim 1, wherein: The multi - sensor data fusion analysis model uses an LSTM neural network to learn the non - linear correlation features between sudden temperature rise and abnormal gas concentration changes, identify coupled anomalies, and capture the micro - change correlations in the sensing data by setting dynamic thresholds, so as to judge the current fire - fighting situation.

9. The robot fire inspection, early warning, supervision and management system according to claim 8, wherein: The data fusion platform is built - in with a multi - dimensional time - series matrix, extracts the features of the current fire - fighting situation through a TCN temporal convolutional network, calculates the risk probability using a Bayesian network, triggers a hierarchical early warning when the probability value exceeds the adaptive threshold, and the adaptive threshold is automatically adjusted according to the environmental baseline fluctuations and dynamically corrected in combination with historical false alarm records.

10. The robot fire inspection, early warning, supervision and management system according to claim 1, characterized in that: The remote communication terminal provides functions such as viewing the inspection screen, receiving early warning information, and remotely controlling the robot body, and supports the multi - robot collaborative supervision function for simultaneously accessing several robot bodies.

Images