CN-122021338-A - Design method of mine safety inspection robot

CN122021338ACN 122021338 ACN122021338 ACN 122021338ACN-122021338-A

Abstract

The invention relates to the technical field of mine safety monitoring and discloses a mine safety inspection robot design method which comprises the steps of obtaining roadway space data and an environment characteristic monitoring data set, determining a space weight attenuation factor by using a section resistance coefficient and a wind flow distribution vector, constructing an integrated learning prediction model, calibrating the monitoring data set by using the space weight attenuation factor to generate a calibration characteristic subspace with physical topological constraint, determining optimal parameters of the model by using a parallel subspace evolution algorithm, inputting the calibration characteristic subspace into the model, and outputting an environment parameter prediction result.

Inventors

SONG JIALING
WANG GUANGCHI
Pan Yuyue
ZHAO YUHUI
BAO JUNHAO
ZHOU XU

Assignees

华北理工大学

Dates

Publication Date: 20260512
Application Date: 20260224

Claims (10)

1. The design method of the mine safety inspection robot is characterized by comprising the following steps of: Step S1, acquiring mine tunnel space data and an environmental characteristic monitoring data set of an area to be inspected, wherein the mine tunnel space data comprises tunnel section geometric parameters, section resistance coefficients and wind flow distribution vectors; S2, constructing an integrated learning prediction model, and determining a spatial weight attenuation factor by using a section resistance coefficient and a wind flow distribution vector, wherein the integrated learning prediction model adopts a Stacking integrated frame, and performs asymmetric adjustment on a weight matrix of a base learner in the integrated learning prediction model through the spatial weight attenuation factor so as to compensate energy loss of environmental parameters in a physical space diffusion process; step S3, performing calibration processing on the environmental characteristic monitoring data set by using a space weight attenuation factor to generate a calibration characteristic subspace, wherein the method comprises the steps of calculating a phase lag characteristic value in a parameter diffusion process according to a tunnel section geometric parameter, and introducing the phase lag characteristic value as a time bias term into an input characteristic vector of a base learner so as to realize logic alignment of a calculation model and a mine physical topology; Step S4, performing super-parameter search on the integrated learning prediction model through a parallel subspace evolution algorithm to determine optimal model parameters, wherein the parallel subspace evolution algorithm extracts candidate solutions in parallel in a plurality of independent sampling spaces and dynamically reduces search boundaries by utilizing nonlinear control factors; And S5, inputting the calibration feature subspace into an integrated learning prediction model with optimal model parameters, executing regression operation by using the integrated learning prediction model, and outputting an environment parameter prediction result.
2. The method for designing the mine safety inspection robot according to claim 1, wherein the step S2 of determining the spatial weight attenuation factor by using the section resistance coefficient and the wind flow distribution vector comprises determining spatial topological association degrees among monitoring sampling points according to the wind flow distribution vector, calculating weight penalty components corresponding to the spatial topological association degrees by using the section resistance coefficient, and defining the weight penalty components as the spatial weight attenuation factor for adjusting the neuron connection strength of the base learner.
3. The method according to claim 1, wherein the calibration process performed by using the spatial weight attenuation factor in step S3 involves calculating the amplitude attenuation rate by using the roadway geometry parameters Amplitude decay rate The calculation of (a) follows the following logical rules: , wherein, Is the dimensionless amplitude attenuation rate, Is a preset tunnel wall resistance characteristic value, An equivalent physical distance (in m) that propagates for an environmental feature, Is the equivalent diameter of the roadway (unit is m).
4. The method for designing the mine safety inspection robot according to claim 1, wherein the step S4 of performing the super-parametric search through the parallel subspace evolution algorithm includes initializing a population composed of a plurality of search units, dividing the population into a global detection subgroup and a local development subgroup, establishing a data sharing link between the search units, enabling the global detection subgroup to dynamically adjust a search step length according to a deviation gradient of an environmental parameter prediction result, and performing attenuation processing on a displacement increment of the search unit by using a nonlinear control factor in an iterative process.
5. The mine safety inspection robot design method according to claim 1 is characterized by further comprising the step of executing linkage control by using an environmental parameter prediction result after the step S5, wherein the step S501 is used for calculating fire source coordinate probability distribution of a target area based on the environmental parameter prediction result and extracting semantic features of the fire source area in inspection path image data, the step S502 is used for constructing a linkage control closed loop covering environmental parameter time sequence prediction, semantic feature recognition and execution instruction feedback, and the step S503 is used for generating an intervention control signal and outputting the intervention control signal to an execution module of the robot when the environmental parameter prediction result meets a preset alarm threshold.
6. The method according to claim 1, wherein the sampling frequency of the environmental characteristic monitoring data set obtained in step S1 is set to be not lower than 10Hz, and the sampling frequency is dynamically adjusted according to the magnitude of the flow rate of the flue gas.
7. The mine safety inspection robot design method is characterized in that a primary learning level of a Stacking integrated framework adopted in the step S2 is composed of at least three heterogeneous decision tree models, a secondary learning level adopts a ridge regression model, and a space weight attenuation factor realizes weight adjustment by modifying node splitting gain of the decision tree model.
8. The method according to claim 1, wherein the step S3 of generating the calibration feature subspace comprises performing a self-attention calculation on the environmental feature monitoring dataset, extracting local space-time correlation features in the feature vectors, and performing weighted fusion of the local space-time correlation features with the output of the spatial weight attenuation factor.
9. The method according to claim 5, wherein the step S501 of extracting semantic features includes performing coordinate mapping on the inspection path image data using a position coding technique, and mapping confidence of predicted coordinates to classification weights using a quality focus loss function.
10. The method according to claim 5, wherein the step S503 of generating the intervention control signal includes establishing three-stage discrimination logic based on the trend of the fire evolution, switching the execution parameters according to the change slope of the environmental parameter prediction result, and generating the safe movement track data of the inspection robot.

Description

Design method of mine safety inspection robot Technical Field The invention belongs to the technical field of mine safety monitoring, and particularly relates to a design method of a mine safety inspection robot. Background The method comprises the steps of constructing an environment parameter prediction framework by combining a multisource sensor with an integrated learning model, selecting carbon monoxide concentration, temperature and smoke concentration in a roadway as characteristic data, building a prediction model by utilizing a nonlinear regression mechanism, generating a data fitting result under a stable working condition, wherein a mine roadway belongs to a typical limited branch space, the evolution rule of the environment parameter is limited by the physical topology of the roadway and damping effect generated by a wind flow field, the main stream calculation model regards sensing data as independent time sequence components, ignoring energy loss and phase lag of the parameter in the physical propagation process and the loss of physical causal logic, generating prediction deviation and numerical oscillation when the model faces to ventilation network fluctuation or multipoint burst risk, selecting an improved path for increasing network depth or stacking sample scale, increasing the calculation load of a main control unit, and being incapable of correcting internal contradiction between the statistical rule and the physical diffusion logic. The utility model patent with the authority of bulletin number of CN222319334U discloses an ARM-based embedded mine inspection robot control system, which integrates various sensing modules through an ARM controller to realize multi-path control and threshold alarming of the inspection robot, and optimizes the integration level and reliability of the control system through hardware redundancy and wireless communication, but regards the sensing signals as isolated time sequences or statistical components when processing the sensing signals, and does not consider wind flow field damping effect and section resistance in limited branch space of mine roadways, so that energy loss and phase lag are generated due to physical propagation of environmental parameters. Therefore, how to determine a calculation model which is fused with space topology constraint logic and has physical causal sensing capability, and solve the problems of prediction deviation and intervention delay caused by the disconnection of the calculation model and mine physical topology logic, become the technical problems to be solved by the invention. Disclosure of Invention The invention provides a design method of a mine safety inspection robot, which comprises the following steps: Step S1, acquiring mine tunnel space data and an environmental characteristic monitoring data set of an area to be inspected, wherein the mine tunnel space data comprises tunnel section geometric parameters, section resistance coefficients and wind flow distribution vectors; S2, constructing an integrated learning prediction model, and determining a spatial weight attenuation factor by using a section resistance coefficient and a wind flow distribution vector, wherein the integrated learning prediction model adopts a Stacking integrated frame, and performs asymmetric adjustment on a weight matrix of a base learner in the integrated learning prediction model through the spatial weight attenuation factor so as to compensate energy loss of environmental parameters in a physical space diffusion process; step S3, performing calibration processing on the environmental characteristic monitoring data set by using a space weight attenuation factor to generate a calibration characteristic subspace, wherein the method comprises the steps of calculating a phase lag characteristic value in a parameter diffusion process according to a tunnel section geometric parameter, and introducing the phase lag characteristic value as a time bias term into an input characteristic vector of a base learner so as to realize logic alignment of a calculation model and a mine physical topology; Step S4, performing super-parameter search on the integrated learning prediction model through a parallel subspace evolution algorithm to determine optimal model parameters, wherein the parallel subspace evolution algorithm extracts candidate solutions in parallel in a plurality of independent sampling spaces and dynamically reduces search boundaries by utilizing nonlinear control factors; And S5, inputting the calibration feature subspace into an integrated learning prediction model with optimal model parameters, executing regression operation by using the integrated learning prediction model, and outputting an environment parameter prediction result. Preferably, the step S2 of determining the space weight attenuation factor by using the section resistance coefficient and the wind flow distribution vector comprises determining the space topology associatio