Fire preventing and extinguishing monitoring and positioning system in coal mine tunnel

Abstract

The invention discloses a fire prevention and extinguishment monitoring and positioning system in a coal mine tunnel, and relates to the field of fire prevention and extinguishment monitoring and positioning, the fire prevention and extinguishment monitoring and positioning system comprises a deployment module, a purification module, a prediction module, a route optimization module and a positioning module, a sensor range is determined through a transient thermal shock experiment, and a tunnel two-dimensional point cloud is obtained through three-dimensional laser scanning; constructing a topological potential field to deploy a main node, balancing coverage and energy consumption through a quantum algorithm, obtaining a final sensor deployment scheme, collecting original sensor data flow data, performing chaotic modulation and optimization, obtaining chaotic spectrum time domain purification data, calculating information entropy of each node, and establishing a game utility function to optimize a routing strategy. The method comprises the following steps: obtaining an energy-data co-occurrence strategy routing table, constructing a three-dimensional tensor, extracting spatial-temporal characteristics, obtaining a data-object modified and optimized fire probability value through model training and probability correction, triggering a positioning algorithm signal, constructing an implicit fire source field, and obtaining verified real fire source coordinates through multi-agent and positioning topology analysis, so that the positioning accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fire prevention and extinguishing monitoring and positioning, in particular to a coal mine roadway fire prevention and extinguishing monitoring and positioning system. BACKGROUND

[0002] Coal mine roadway fires have the characteristics of strong concealment, rapid spread and complex environment, and the existing monitoring and positioning technology faces the following core challenges:

[0003] 1. Extensive sensor deployment: Traditional methods rely on experiential node layout, making it difficult to balance monitoring coverage completeness and energy efficiency, especially in complex roadways (such as high curvature, multiple branch areas), resulting in monitoring blind spots and missing data at the early stage of fire.

[0004] 2. Insufficient data reliability: Strong electromagnetic interference and dust environment in the mine can cause sensor data distortion, traditional filtering methods are difficult to effectively suppress unstructured noise, and there is a lack of dynamic encryption mechanism in the data transmission process, which poses a security risk.

[0005] 3. Lack of physical constraints in fire prediction: Existing models rely on single data-driven (such as temperature threshold), without incorporating thermodynamic laws (such as energy conservation, heat diffusion characteristics), which can easily cause false positives or false negatives due to sensor errors.

[0006] 4. Poor topological consistency of fire source positioning: Traditional positioning algorithms (such as weighted average) do not consider the spatial topological characteristics of the roadway (such as connectivity, void distribution), and the positioning results may deviate from the true fire source area, resulting in deviation in the direction of rescue.

[0007] 5. Poor routing basis: Only using physical layer indicators such as signal strength and hop count as the basis for routing, without quantifying the information value of the data itself (such as the spatiotemporal variability of fire precursor data), which cannot distinguish between high-value data (such as sudden temperature changes, CO concentration anomalies) and regular monitoring data, which may cause critical warning information to be delayed or discarded due to node energy consumption priority strategies.

[0008] In order to solve the above-mentioned defects, a technical solution is provided. SUMMARY

[0009] In order to solve the technical problems raised in the above background, the present application is proposed. The embodiments of the present application provide a coal mine roadway fire prevention and extinguishing monitoring and positioning system.

[0010] The purpose of the present application can be achieved by the following technical solution: a coal mine roadway fire prevention and extinguishing monitoring and positioning system, comprising a deployment module, a purification module, a prediction module, a routing optimization module, a positioning module,

[0011] The deployment module determines the sensor range through a transient thermal shock experiment, obtains a two-dimensional point cloud of the roadway by using a three-dimensional laser scanner, constructs a topological potential field deployment master node, identifies a gap area to generate a candidate redundant node, balances coverage and energy consumption through a quantum algorithm, and obtains a final sensor deployment scheme {xi, x'j};

[0012] The purification module collects original sensor data stream data s(t) based on the final sensor deployment scheme {xi, x'j}, performs chaos modulation and optimization, and obtains chaos spectrum time domain purification data;

[0013] The routing optimization module performs calculation based on the chaos spectrum time domain purification data, calculates the information entropy of each node, establishes a game utility function to optimize the routing strategy, and obtains a symbiotic strategy routing table {pi j};

[0014] The prediction module constructs a three-dimensional tensor based on the chaos spectrum time domain purification data and the final sensor deployment scheme, extracts spatiotemporal features, performs model training and probability correction, obtains a fire probability value, and determines a threshold value to trigger a positioning algorithm signal;

[0015] The positioning module constructs an implicit fire source field based on the fire probability value and the roadway point cloud data, performs multi-agent and positioning topology analysis, and obtains verified real fire source coordinates (x**, y**).

[0016] Further, the final sensor deployment scheme {xi, x'j} analysis step is as follows:

[0017] The sensor node energy consumption parameter set and the sensor setting range Rsen are obtained, and the energy consumption value of the candidate redundant node position {xj} is calculated through the power consumption energy consumption model;

[0018] The candidate redundant node position {xj} is mapped to a state vector of a quantum bit based on quantum bit encoding, wherein the polar angle is mapped to the candidate node coordinates, the potential field coverage rate and the node energy consumption Ej are obtained, and the chromosome fitness is calculated;

[0019] The polar angle is adjusted by a learning rate η according to the chromosome fitness gradient, the polar angle is adjusted through a quantum rotation gate, and the polar angle is adjusted in the direction of π / 2. If the fitness F is greater than a set threshold value, the adjustment is stopped. The adjusted quantum state is measured and decoded to obtain the final redundant node coordinates {x'j};

[0020] The final redundant node coordinates {x'j} and the deployment coordinates set {xi} of the master sensor form the final sensor deployment scheme {xi, x'j}.

[0021] Further, the candidate redundant node position {xj} analysis step is as follows:

[0022] A transient thermal shock test was conducted on the surrounding rock of the roadway and the temperature response curve was fitted to obtain the thermal relaxation time of the surrounding rock material of the roadway. The type of burning material in the coal mine roadway was obtained. The fuel calorific value, combustion rate and combustion efficiency were obtained from the corresponding database, and the product was processed to obtain the fire source power.

[0023] A non-Fourier heat conduction model is established based on the thermal relaxation time of the tunnel surrounding rock material and the fire source power in the tunnel surrounding rock. The model includes:

[0024] in represents the Laplace operator, the second-order spatial derivative, describing the spatial distribution of temperature, β represents the thermal diffusivity, qs represents the fire source power in the tunnel surrounding rock, τq represents the thermal relaxation time, and after solving the equation, the maximum temperature gradient is obtained. The sensor setting range Rsen is obtained through the formula;

[0025] Based on the sensor range Rsen, the point cloud twin topology potential field function is constructed using the two-dimensional point cloud data of the roadway. By solving the extreme point of the point cloud twin topology potential field, if the gradient value of the point cloud twin topology potential field is And Cur(x)>κth, where κth is the curvature threshold, then the point is an extreme point, and the extreme points constitute the deployment coordinate set {xi} of the main sensor;

[0026] Based on the deployment coordinate set {xi} of the main sensor, coverage is performed with the main node xi as the center and the sensor range Rsen as the radius. The uncovered area in the lane is identified and marked as the gap area Ω. The candidate redundant node position {xj} is generated in the gap area Ω.

[0027] Furthermore, the chaotic spectrum time domain purification data analysis steps are as follows:

[0028] Based on the final sensor deployment plan {xi, x′j}, the original sensor data stream data s(t) is collected and the node residual energy is calculated through the node energy consumption Ej;

[0029] Based on the original sensor data stream and the node residual energy, the Lorenz chaotic system is dynamically modulated, and the chaotic modulation is performed according to the residual energy ratio of the node. The differential equation is discretized and the state variables are iteratively updated to obtain the energy-coupled chaotic encrypted data stream senc(t);

[0030] Based on the energy-coupled chaotic encrypted data stream, the energy-coupled chaotic carrier spectrum is obtained through Fourier transform. The non-carrier frequency band noise is suppressed by convolutional neural differential equations on the energy-coupled chaotic encrypted data stream and the energy-coupled chaotic carrier spectrum. The loss function training is established through clean data, and the decoding network decodes the hidden state to obtain the chaotic spectrum time domain purification data.

[0031] Further, the energy-dependent symbiotic strategy routing table {pi j The analysis steps are as follows:

[0032] Based on the chaos spectrum time domain purification data and the node residual energy, the information entropy of each node is calculated, the game utility function is established based on the information entropy of each node, the routing strategy of each node is independently updated, the probability distribution of the next hop is selected, and the iteration is stopped when all nodes meet the set conditions, and the energy-dependent symbiotic strategy routing table {pi j} is obtained.

[0033] Further, the fire probability value analysis step is as follows:

[0034] The space-time feature vector Fte is input into the machine learning model, the model is trained with the real fire label ytrue as supervision, the model weight and bias are adjusted, the space-time feature vector Fte input by the model is linearly transformed, and the fire probability value Pfir is obtained by outputting the mapping to the [0, 1] interval through the sigmoid function.

[0035] The space-time feature vector Fte is input into the machine learning model, the model is trained with the real fire label ytrue as supervision, the model weight and bias are adjusted, the space-time feature vector Fte input by the model is linearly transformed, and the fire probability value Pfir is obtained by outputting the mapping to the [0, 1] interval through the sigmoid function.

[0036] Based on the real fire label ytrue and the fire occurrence probability value Pfir, the cross-entropy loss function is obtained, and the true fire probability value cross-entropy CroEntr is obtained.

[0037] Based on the true fire probability value cross-entropy CroEntr, the heat accumulation prediction value and the heat flow prediction value The real fire label ytrue is corrected and optimized by the meta-learning early warning model through data physical double constraints, and the object-optimized fire probability value Pfir1 is obtained.

[0038] Based on the object-optimized fire probability value Pfir1 and the set threshold, if it is greater than the set threshold, the positioning algorithm signal is triggered.

[0039] Further, the fire occurrence probability value Pfir analysis step is as follows:

[0040] Based on the chaos spectrum time domain purification data P(x) and the final sensor deployment scheme, a three-dimensional tensor Xsp is constructed by a Dirac function, core features are extracted by high-order singular value decomposition, and a space-time feature vector Fte is obtained.

[0041] The space-time feature vector Fte is input into the machine learning model, the model is trained with the real fire label ytrue as supervision, the model weight and bias are adjusted, the space-time feature vector Fte input by the model is linearly transformed, and the fire probability value Pfir is obtained by outputting the mapping to the [0, 1] interval through the sigmoid function.

[0042] Furthermore, the analysis steps of the verified real fire source coordinates (x**, y**) are as follows:

[0043] Based on the object-optimized fire probability value Pfir1 and the roadway point cloud data, the neural radiation field model constructs an implicit fire source field through a multi-layer perceptron. The model is trained by training the rendering loss function to obtain the implicit 3D fire source field σ(x).

[0044] Based on the implicit 3D fire source field σ(x) and the final sensor deployment plan {xi, x′j}, the local state sj of each node j is defined. The coverage quality and communication cost are combined through the MA-DDPG reward function rj. The Actor network combines the reward function through the MA-DDPG framework and outputs the node weight wj.

[0045] The collaborative positioning coordinates (x*, y*) are obtained by weighted average calculation based on the node weight wj;

[0046] Based on the collaborative positioning coordinates (x*, y*), a Vietoris-Rips complex is constructed with the maximum distance between sensor nodes as the scale. Its H1 persistence graph is calculated to extract topological void features, including the birth radius and death radius of the void. It is verified whether there is a void feature that satisfies the condition that the difference between the birth and death radius exceeds the set topological threshold, and the distance between the preliminary positioning coordinates and the geometric center of the void is less than the average radius of the void. If the conditions are met, the geometric center of the void is output as the verified true fire source coordinates (x**, y**); if not, the positioning calculation is repeated until the conditions are met.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention determines the sensor range through a transient thermal shock experiment through a deployment module, obtains a two-dimensional point cloud of the tunnel using three-dimensional laser scanning, constructs a topological potential field to deploy the main node, identifies the gap area to generate candidate redundant nodes, and balances coverage and energy consumption through a quantum algorithm to obtain the final sensor deployment plan {xi, x′j}. The purification module collects the original sensor data stream data s(t) based on the final sensor deployment plan {xi, x′j}, performs chaotic modulation and optimization, and obtains chaotic spectrum time domain purification data. The routing optimization module calculates the information entropy of each node based on the chaotic spectrum time domain purification data, establishes a game utility function to optimize the routing strategy, and obtains the energy according to the symbiotic strategy routing table {π j}, the prediction module constructs a three-dimensional tensor based on the chaos spectrum time domain purification data and the final sensor deployment scheme, extracts space-time features, obtains the fire probability value after model training and probability correction, and determines the threshold value to trigger the positioning algorithm signal, which can balance the monitoring coverage integrity and energy efficiency, especially in complex tunnels (such as high curvature, multiple branch areas) with monitoring blind spots, ensuring that the initial fire data is not missing, and the sensor data is not distorted in the strong electromagnetic interference and dust environment, which can effectively suppress unstructured noise, and can dynamically encrypt the data transmission process, and can integrate thermodynamic laws (such as energy conservation and heat diffusion characteristics), and use physical layer indicators such as signal strength and hop count as routing basis, and quantify the information value of the data itself (such as the spatio-temporal variability of fire precursor data), and can distinguish high-value data (such as sudden temperature changes and CO concentration anomalies) from regular monitoring data.

[0049] 2、The present application constructs an implicit fire source field by the fire probability value and the roadway point cloud data, and obtains the verified real fire source coordinates (x**, y**) through multi-agent and positioning topology analysis, considers the roadway space topology characteristics (such as connectivity and cavity distribution), and the positioning result is more accurate and close to the real fire source area, and the rescue time is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. The following drawings are not deliberately drawn according to the actual size and proportion, and the focus is on showing the main idea of the present application.

[0051] Figure 1 The system block diagram of the present application is shown in the figure.

[0052] Figure 2 The positioning module flow chart of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor also belong to the scope of protection of the present application.

[0054] As shown in the figure, a coal mine roadway fire prevention and extinguishing monitoring and positioning system includes a deployment module, a purification module, a prediction module, a routing optimization module, and a positioning module. Figure 1

[0055] ​The deployment module determines the sensor range through transient thermal shock experiments, uses 3D laser scanning to obtain a 2D point cloud of the roadway, constructs a topological potential field to deploy the main node, identifies gap areas and generates candidate redundant nodes, and uses a quantum algorithm to balance coverage and energy consumption to obtain the final sensor deployment plan {xi, x′j}.

[0056] The purification module collects the original sensor data stream s(t) based on the final sensor deployment plan {xi, x′j}, performs chaotic modulation and optimization, and obtains chaotic spectrum time domain purification data;

[0057] The routing optimization module is based on the chaotic spectrum time domain purification data, calculates the information entropy of each node, establishes the game utility function to optimize the routing strategy, and obtains the routing table {π j};

[0058] The prediction module constructs a three-dimensional tensor and extracts spatiotemporal features based on the chaotic spectrum time-domain purification data and the final sensor deployment plan. After model training and probability correction, it obtains an optimized fire probability value based on the object, and determines the trigger signal of the positioning algorithm based on this threshold.

[0059] The positioning module constructs an implicit fire source field based on the object-optimized fire probability value and the roadway point cloud data. After multi-agent and positioning topology analysis, the verified real fire source coordinates (x**, y**) are obtained.

[0060] Specifically, the deployment module analysis steps are as follows:

[0061] The transient thermal shock test of the tunnel surrounding rock was carried out and the temperature response curve was fitted, T(t)=T0+ΔT(1-e -t / τt ), where T(t) represents the temperature change value of the tunnel surrounding rock material, T0 represents the initial temperature change value, ΔT represents the amplitude of the temperature change, t represents time, and τq represents the thermal relaxation time. The thermal relaxation time of the tunnel surrounding rock material is obtained, and the type of combustion material in the coal mine tunnel is obtained. The fuel calorific value, combustion rate, and combustion efficiency are obtained from the corresponding database, and the product is processed to obtain the fire source power;

[0062] A non-Fourier heat conduction model is established based on the thermal relaxation time of the tunnel surrounding rock material and the fire source power in the tunnel surrounding rock. The model includes:

[0063] in represents the Laplace operator, the second-order spatial derivative, which describes the spatial distribution of temperature. β represents the thermal diffusivity, β = k / (ρc), where k is the thermal conductivity, ρ is the density, c is the specific heat capacity, and qs represents the power of the fire source in the surrounding rock of the tunnel. After solving the equation, the model obtains the maximum temperature gradient. By formula Get the sensor setting range Rsen;

[0064] Based on the sensor setting range Rsen and using the roadway two-dimensional point cloud data, a point cloud twin topology potential field function is constructed:

[0065] Wherein Cur(x) represents the roadway curvature, which is calculated by point cloud data, a represents the curvature penalty coefficient, d(x, xi) represents the distance from point x to the i-th sensor, I represents the number of sensors, and Φ(x) represents the point cloud twin topology potential field function. By solving the point cloud twin topology potential field extreme point, if the point cloud twin topology potential field gradient value i Cur(x) > κth, wherein κth is the curvature threshold, the point is the extreme point, and the extreme point constitutes the deployment coordinate set {xi} of the main sensor;

[0066] Specifically, a three-dimensional laser scanner (such as LiDAR) is used to collect three-dimensional point cloud (containing x, y, z coordinates) along the roadway, and an inertial measurement unit (IMU) and an odometer are combined to ensure position accuracy; then the three-dimensional data is sliced along a fixed height (such as the center line) or orthogonally projected onto the xy plane, and is converted into two-dimensional point cloud; after noise reduction filtering (removing outliers), ICP registration and splicing, the effective points are purified; then the geometric features are extracted: the roadway curvature Cur(x) is calculated by local curve fitting (for high curvature area identification), and the distance d(x, xi) between points is calculated; finally, the horizontal geometric information (width, curvature, distance) of the two-dimensional point cloud is used as input to support the construction of the topology potential field function (such as the distance term and curvature penalty in formula Φ(x)), to provide roadway planar form data for sensor deployment optimization, and to realize node layout calculation based on heat shock monitoring requirements. This process integrates three-dimensional scanning, point cloud processing and engineering geometric analysis to ensure that the data accurately reflects the horizontal features of the roadway, serving the model construction and optimization of the coal mine monitoring system.

[0067] Based on the deployment coordinate set {xi} of the main sensor, a coverage is performed with the main node xi as the center and the sensor range Rsen as the radius, the uncovered area in the roadway is identified, and is marked as the gap area Ω. The candidate redundant node positions {xj} are generated in the gap area Ω. Specifically, Ω = roadway area \ U i B(x i , Rsen), wherein U i B(x i , Rsen) represents the union set of all main node coverage areas, i.e. the total coverage range under the current sensor deployment scheme, B(x i , Rsen) represents the coverage area with the main node xi as the center and the sensor range Rsen as the radius, and \ represents the difference set operation, i.e. the remaining part after removing the right set elements from the left set;

[0068] ​The sensor node energy consumption parameter set and the sensor setting range Rsen are obtained, and the energy consumption value of the candidate redundant node position {xj} is calculated through the power consumption energy consumption model. The specific sensor node energy consumption parameter set includes the transmission and reception energy consumption per unit data, the sleep state power consumption, the sleep time, and the data amount processed by the node. The specific power consumption energy consumption model includes:

[0069] wherein Ej represents the node energy consumption, dj represents the distance from the node to the base station, Cj represents the data amount processed by the node, Eid represents the sleep state power consumption, t id represents the sleep time, Etx and Erx represent the transmission and reception energy consumption per unit data, respectively;

[0070] The candidate redundant node position {xj} is mapped to the state vector of the quantum bit based on the quantum bit coding, wherein the polar angle is mapped to the candidate node coordinate. The potential field coverage rate and the node energy consumption Ej are obtained to calculate the chromosome fitness;

[0071] The polar angle is adjusted by the learning rate η according to the chromosome fitness gradient, the polar angle is adjusted by the quantum rotation gate, and the polar angle is adjusted in the direction of π / 2. If the fitness F is greater than the set threshold value, the adjustment is stopped. The adjusted quantum state is measured and decoded to obtain the final redundant node coordinate {x′j};

[0072] The final redundant node coordinate {x′j} and the deployment coordinate set {xi} of the main sensor form the final sensor deployment scheme {xi, x′j};

[0073] Specifically, the above-mentioned formula is as follows:

[0074]

[0075] xj′=θj×Lx / (π / 2)

[0076] F=B0+u∑e -Ej

[0077]

[0078] Wherein, |ψj> represents the state vector of the jth quantum bit, describes the quantum state of the candidate redundant node, corresponds to the candidate redundant node position {xj}, θj represents the polar angle, φj represents the azimuth angle, which can be mapped to the y coordinate to determine the y direction position, |0>, |1> represents the quantum bit ground state, the potential position of the node is represented by using superposition, i represents the imaginary unit, F represents the chromosome fitness, which measures the pros and cons of the deployment scheme, xj' represents the candidate node coordinate mapped by the polar angle, Lx represents the roadway length, B0 represents the potential field coverage rate = coverage volume / roadway volume, u represents the weight coefficient, balancing the influence of coverage and energy consumption, Δθj represents the polar angle adjustment amount, optimizing the node position, η represents the learning rate, controlling the angle update step, represents the chromosome fitness gradient, the partial derivative of fitness with respect to polar angle, guiding the adjustment direction of the angle, measuring the adjusted quantum state, collapsing into a classical bit string, and decoding to obtain the final redundant node coordinates {x'j}, under the condition of main node coverage, the coverage rate is improved through the redundant node, the candidate redundant position is generated first, θj is adjusted through the quantum rotation gate iteration, so that the node with high coverage and low energy consumption is retained, the total coverage rate meets, and the total energy consumption is minimized.

[0079] Specifically, the analysis steps of the purification module and the routing optimization module are as follows:

[0080] Based on the final sensor deployment scheme {xi, x'j}, the original sensor data stream data s(t) is collected, and the node residual energy is calculated based on the node energy consumption Ej;

[0081] Based on the original sensor data stream data and the node residual energy, the Lorenz chaotic system is dynamically modulated, and the chaotic modulation is performed through the proportion of the residual energy of the node, the differential equation is discretized, and the state variable is iteratively updated to obtain the energy coupling chaotic encryption data stream senc(t);

[0082] Based on the energy coupling chaotic encryption data stream, the energy coupling chaotic carrier spectrum is obtained through Fourier transform, the energy coupling chaotic encryption data stream and the energy coupling chaotic carrier spectrum are convolved to suppress the noise in the non-carrier frequency band, the loss function is trained through clean data, the hidden state is decoded by the decoding network, and the chaotic spectrum time domain purification data is obtained;

[0083] Based on the chaotic spectrum time domain purification data and the node residual energy, the information entropy of each node is calculated, the game utility function is established based on the information entropy of each node, each node independently updates the routing strategy, and the probability distribution of the next hop is selected, and the iteration is stopped when all nodes meet the set conditions to obtain the energy-based symbiotic strategy routing table {π j};

[0084] Specifically, the node residual energy is obtained by subtracting the cumulative node energy from the initial energy in Represents the initial energy of the jth node, that is, the total energy of the node when it is deployed. It should be noted that j includes all monitoring nodes, ∑ cycle Ej represents the cumulative energy consumption of the jth node in all working cycles, the ratio of the node's remaining energy to the node's maximum rated energy, and the modulation coefficient β of the Lorenz chaotic system is obtained.

[0085] Based on the final sensor deployment plan {xi, x′j}, the raw sensor data stream s(t) is collected, including time series signals of temperature, CO concentration, and vibration. The Lorenz chaotic system includes: Where σ,ρ are the parameters of the chaotic system, is the time derivative of the chaotic state variable, describing the dynamic changes of the system. x, y, and z are chaotic state variables. The modulation coefficient β of the high-energy consumption node increases, which enhances the chaotic characteristics of the state variable x. The state variable x is discretized through the differential equation and iteratively updated: k+1 =x k +Δt×[σ(y k -x k )+β×s(t k )], where x k+1 、x k represents the chaotic state variable x at the kth and k+1th time steps, Δt is the time step, s(t k ) is t k The original sensor data at the time is output to obtain the energy-coupled chaotic encrypted data stream senc(t)=x k , represents the sensor data at time tk; the energy-coupled chaotic encrypted data stream senc(t) is transformed into a frequency domain spectrum through Fourier transform, and its characteristic frequency distribution, such as the fractal spectrum of broadband chaos and the carrier frequency of energy concentration, is extracted to obtain the energy-coupled chaotic carrier spectrum F(ω);

[0086] Based on the energy-coupled chaotic encrypted data stream senc(t) and the energy-coupled chaotic carrier spectrum F(ω), the non-carrier frequency band noise is suppressed through convolution operation and neural differential equations:

[0087] Where fθ is a parameterized neural network such as LSTM, k(t) is the hidden state, For the convolution operation, through the clean data x clea (t) Establish the loss function ||x(t)-x clea (t)|| 2 Training, decoding network decodes hidden state and obtains chaotic spectrum time domain purification data P(x), P(x) = gφ(k(t)), gφ represents the decoding network output;

[0088] based on the chaos spectrum time domain purification data P(x) and the node residual energy By calculating the information entropy Hj = -∑ m p(x j，m )logp(x j，m ), p(x j，m ) represents the probability distribution of the node j chaos spectrum time domain purification data P(x), and m represents the index of the data sampling point, and the game utility function is established based on the information entropy of each node:

[0089] Wherein Uj represents the game utility function, represents the hop number of node j to the base station such as transmission cost, and γ represents the weight, and each node independently updates the probability distribution of the routing strategy to select the next hop: Wherein represents the routing strategy of the kth iteration, and ξ represents the learning rate, represents the gradient of the utility function, guiding the strategy to update in the increasing direction, and when all nodes satisfy the iteration is stopped, and the symbiotic strategy routing table {pi j} is obtained.

[0090] Specifically, the analysis steps of the prediction module are as follows

[0091] Based on the chaos spectrum time domain purification data P(x) and the final sensor deployment scheme, a three-dimensional tensor Xsp is constructed through the Dirac function, and the core features are extracted through high-order singular value decomposition, and the space-time feature vector is obtained;

[0092] Specifically, Wherein δ(*) represents the Dirac function (mapping discrete nodes to continuous space), J is the total number of nodes, and the three-dimensional tensor Xsp is the space s x parameter p x time t, X = G x 1U s x 2U p x 3U t , wherein X is the space-time parameter tensor of coal mine monitoring data, integrating the space, parameter and time dimensions, G is the core tensor, U s , U p , U t represent the factor matrix of space, parameter and time, and the core tensor G is obtained through the tensor vectorization operation to obtain the space-time feature vector Fte.

[0093] The spatiotemporal feature vector Fte is input into a machine learning model, the model is trained with a real fire label ytrue as supervision, the model weight and bias are adjusted, the model performs linear transformation on the input spatiotemporal feature vector Fte, and the probability value Pfir of fire occurrence is obtained through the sigmoid function output mapping to the [0, 1] interval, the real fire label is the temperature ≥ 75℃ and the temperature rise rate > 1.5℃ / min and the CO concentration ≥ 50ppm and the change rate > 5ppm / min in the chaos frequency spectrum time domain purification data P(x);

[0094] The spatiotemporal feature vector Fte is input into a machine learning model, the model is trained with a real fire label ytrue as supervision, the model weight and bias are adjusted, the model performs linear transformation on the input spatiotemporal feature vector Fte, and the probability value Pfir of fire occurrence is obtained through the sigmoid function output mapping to the [0, 1] interval, the real fire label is the temperature ≥ 75℃ and the temperature rise rate > 1.5℃ / min and the CO concentration ≥ 50ppm and the change rate > 5ppm / min in the chaos frequency spectrum time domain purification data P(x);

[0095] The spatiotemporal feature vector Fte is input into a machine learning model, the model is trained with a real fire label ytrue as supervision, the model weight and bias are adjusted, the model performs linear transformation on the input spatiotemporal feature vector Fte, and the probability value Pfir of fire occurrence is obtained through the sigmoid function output mapping to the [0, 1] interval, the real fire label is the temperature ≥ 75℃ and the temperature rise rate > 1.5℃ / min and the CO concentration ≥ 50ppm and the change rate > 5ppm / min in the chaos frequency spectrum time domain purification data P(x);

[0096] The above formula is:

[0097] Pfir=σ(W T ×Fte+b)

[0098]

[0099]

[0100] Specifically, σ represents a sigmoid activation function, w is a weight vector, b is a bias term, T represents the transpose of a matrix, condition 1 represents that the temperature ≥ 75℃ and the temperature rise rate > 1.5℃ / min in the chaos frequency spectrum time domain purification data P(x), condition 2 represents that the CO concentration ≥ 50ppm and the change rate > 5ppm / min, Jsensor represents the measured heat flow, k is the thermal conductivity of the material, is the temperature gradient (sensor directly measures);

[0101] Specifically, the model is trained with historical heat data (measured heat accumulation, heat flow), vQ and WJ are optimized (the prediction error is minimized, such as mean square error), and the energy conservation constraint is introduced as a physical loss (to ensure that the heat accumulation change rate and the heat flow divergence comply with the law of conservation of energy

[0102] Based on the real fire label ytrue and the probability value Pfir of fire occurrence, the cross-entropy loss function is obtained, and the true disaster probability value cross-entropy CroEntr is obtained.

[0103] Specifically, CroEntr=-[ytruelogPfir+(1-ytrue)log(1-Pfir)]

[0104] Based on the true disaster probability value cross-entropy CroEntr, the heat accumulation prediction value and heat flow prediction value The real fire label ytrue is corrected and optimized by the meta-learning early warning model through data physical double constraints to obtain a physical optimization fire probability value Pfir1;

[0105] Specifically, the meta-learning early warning model includes a loss function:

[0106]

[0107] Wherein, L represents the loss function, χ and ψ represent weight coefficients, and the importance of balancing data and physical constraints, ||*|| 2 represents the square of the Euclidean norm, represents the divergence operator, which is used to describe the divergence or convergence characteristics of the vector field;

[0108] Based on the comparison between the physical optimization fire probability value Pfir1 and the set threshold, if it is greater than the set threshold, a positioning algorithm signal is triggered.

[0109] Specifically, as shown in the following formula: Figure 2 The analysis of the positioning module is as follows:

[0110] Based on the physical optimization fire probability value Pfir1 and the roadway point cloud data, the neural radiation field model constructs an implicit fire source field through a multi-layer perception mechanism, trains the model through a rendering loss function, and obtains an implicit 3D fire source field σ(x);

[0111] Based on the implicit 3D fire source field σ(x) and the final sensor deployment scheme {xi, x′j}, the local state sj of each node j is defined, the coverage quality and communication cost are combined through the MA-DDPG reward function rj, and the Actor network outputs the node weight wj through the MA-DDPG framework combined with the reward function.

[0112] Based on the node weight wj, the collaborative positioning coordinates (x*, y*) are obtained through weighted average calculation;

[0113] Based on the collaborative positioning coordinates (x*, y*), the Vietoris-Rips complex is constructed with the maximum distance between sensor nodes as the scale, the H1 persistent diagram is calculated to extract the topological void feature, including the birth radius and death radius of the void, and it is verified whether the void feature satisfies the condition that the difference between the birth radius and the death radius exceeds the set topological threshold, and the distance between the preliminary positioning coordinates and the geometric center of the void is less than the average radius of the void. If the condition is met, the geometric center of the void is output as the verified real fire source coordinates (x**, y**); if not, the positioning calculation is performed again until the condition is met.

[0114] Specifically, the implicit fire source field includes: where x = (x, y, z) is the spatial coordinate (point cloud provides), is the wind speed vector (sensor measurement), t is the time (data timestamp), σ(x) represents the output fire source existence probability density, T represents the temperature value, C represents the CO gas concentration value, the model is trained by training the rendering loss function, and the implicit 3D fire source field σ(x) is output, such as the key output σ>0.5 area is the fire source candidate area, and the rendering loss function is trained:

[0115] Lrend = ||C 渲染 -C 传感器 || 2 +ν||σ(x)-Pfir1(x)|| 2 , where Lrend represents the rendering loss function, C 传感器 is the measured value of CO concentration, C 渲染 is the virtual prediction value of the neural radiation field model to the gas concentration CO, and the rendering result of the gas concentration of each point in the three-dimensional space is generated, and v is a weight coefficient. The construction of the implicit 3D fire source field is optimized so that it is consistent with the measured value and the global fire probability logic in the three-dimensional space;

[0116] Based on the implicit 3D fire source field σ(x) and the final sensor deployment scheme {xi, x'j}, the local state of each node j is defined where represents the gradient of σ(xj), and the neighbor state represents the state of the adjacent node (such as the near neighbor node in the sensor deployment), including: the fire source probability (σ) and the gradient of the neighbor

[0117] The MA-DDPG reward function rj combines the coverage quality and the communication cost to optimize the node weight wj, and the collaborative positioning result is the weighted average coordinate (x*, y*)

[0118] Specific MA-DDPG reward function where IoU represents the intersection over union of the predicted fire area Rj and the fire source candidate area σ(x)>0.5 area, Rj represents the fire area range predicted by the node j, Я represents the weight balance of coverage and overhead, ||wj||0 represents the communication frequency of the node j, L0 norm, communication cost, IoU(σ(x), Rj) represents the coverage quality, and the actor network outputs the node weight wj through the MA-DDPG framework combined with the reward function;

[0119] Based on the node weight wj, the collaborative positioning coordinate is obtained by weighted average calculation

[0120] Based on the collaborative positioning coordinates (x*, y*), the Vietoris-Rips complex is constructed with the maximum distance between sensor nodes as the scale, and the H1 persistent diagram is calculated to extract the topological hole feature. The birth radius and death radius of the hole are contained. It is verified whether there is a hole feature that satisfies the difference between the birth and death radius exceeding the set topological threshold, and the distance between the preliminary positioning coordinates and the geometric center of the hole is less than the average radius of the hole. If the conditions are met, the geometric center of the hole is output as the verified true fire source coordinates (x**, y**); if not, the positioning calculation is performed again until the conditions are met.

[0121] Specifically, the above formula is:

[0122]

[0123] PD1={(bn,dn)}

[0124]

[0125] Where ∈ is the maximum distance of the node, the diameter of the complex is ≤ ∈, S={(xj, yj)} is the sensor coordinate set, diam(σ) is the diameter of the simplex, bn is the birth radius of the hole, dn is the death radius, VR ε represents the Vietoris-Rips complex, which is a core concept in topological data analysis for constructing topological structures from point set data, PD1 represents a one-dimensional persistent diagram, τtop represents a threshold value, which is equal to 0.2 ∈, cn represents the geometric center of the hole, rn represents the average radius of the hole, the average of the birth radius and the death radius of the hole, s.t. represents the restriction, represents the existential quantifier, which means "there is at least one", ∈ represents belongs to.

[0126] Specifically, the deployment module realizes the optimal balance of sensor coverage and energy consumption based on transient thermal shock experiments and quantum algorithms, uses three-dimensional laser scanning to construct a tunnel point cloud twin topology potential field, combines a non-Fourier heat conduction model to determine the sensor range, and forms a precise layout of the main node and redundant node collaboration. The purification module uses the Lorenz chaotic system to perform energy coupling dynamic modulation on the original data stream, suppresses the non-carrier frequency band noise through the convolution neural differential equation, combines the information entropy game to optimize the routing strategy, realizes the encrypted transmission of data and the routing selection of energy perception. The prediction module constructs a three-dimensional tensor in space-time through the Dirac function, extracts core features through high-order singular value decomposition, combines the meta-learning early warning model to fuse data-driven and physical constraints (such as the law of conservation of energy), corrects the fire probability prediction result, and significantly improves the early warning accuracy. The positioning module constructs an implicit 3D fire source field based on the neural radiation field, optimizes the node weight through the MA-DDPG framework to realize collaborative positioning, and finally uses the H1 persistent graph analysis of Vietoris-Rips complex to verify and output the high-precision fire source coordinates. Through the cooperation of multiple modules, the system realizes the space-time multi-dimensional perception, data secure transmission, probability accurate prediction and fire source topology positioning of coal mine roadway fire monitoring, and provides full-process intelligent technical support for mine fire prevention decision-making.

[0127] The above is a description of the present application and should not be considered limiting. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application. Accordingly, all such modifications are intended to be included within the scope of the present application as defined in the claims. It should be understood that the above is a description of the present application and should not be considered limiting. The present application is limited by the claims and their equivalents.

Claims

1. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system, comprising a deployment module, a purification module, a prediction module, a routing optimization module, and a positioning module, characterized in that: The deployment module determines the sensor range through transient thermal shock experiments, uses 3D laser scanning to obtain a 2D point cloud of the roadway, constructs a topological potential field to deploy the main node, identifies gap areas and generates candidate redundant nodes, and uses a quantum algorithm to balance coverage and energy consumption to obtain the final sensor deployment plan {xi, x′j}. The purification module collects the original sensor data stream s(t) based on the final sensor deployment plan {xi, x′j}, performs chaotic modulation and optimization, and obtains chaotic spectrum time domain purification data; The routing optimization module is based on the chaotic spectrum time domain purification data, calculates the information entropy of each node, establishes the game utility function to optimize the routing strategy, and obtains the routing table {π j }; The prediction module constructs a three-dimensional tensor and extracts spatiotemporal features based on the chaotic spectrum time-domain purification data and the final sensor deployment plan. After model training and probability correction, it obtains an optimized fire probability value based on the object, and determines the trigger signal of the positioning algorithm based on this threshold. The positioning module constructs an implicit fire source field based on the object-optimized fire probability value and the roadway point cloud data. After multi-agent and positioning topology analysis, the verified real fire source coordinates (x**, y**) are obtained.

2. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 1, characterized in that: The analysis steps of the final sensor deployment plan {xi, x′j} are as follows: Obtain the sensor node energy consumption parameter set and sensor setting range Rsen, and calculate the energy consumption value of the candidate redundant node position {xj} through the power consumption model; Based on the candidate redundant node position {xj}, the quantum bit encoding is mapped to the quantum bit state vector, where the polar angle is mapped to the candidate node coordinates, and the potential field coverage and node energy consumption Ej are obtained to calculate the chromosome fitness; According to the chromosome fitness gradient, the polar angle is adjusted with the learning rate η. The polar angle is adjusted in the direction of π / 2 through the quantum rotating gate. When the fitness F is greater than the set threshold, the adjustment stops. The adjusted quantum state is measured and decoded to obtain the final redundant node coordinates {x′j}; The final redundant node coordinates {x′j} and the deployment coordinate set {xi} of the primary sensor constitute the final sensor deployment plan {xi, x′j}.

3. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 2, characterized in that: The analysis steps of the candidate redundant node position {xj} are as follows: A transient thermal shock test was conducted on the surrounding rock of the roadway and the temperature response curve was fitted to obtain the thermal relaxation time of the surrounding rock material of the roadway. The type of burning material in the coal mine roadway was obtained. The fuel calorific value, combustion rate and combustion efficiency were obtained from the corresponding database, and the product was processed to obtain the fire source power. A non-Fourier heat conduction model is established based on the thermal relaxation time of the tunnel surrounding rock material and the fire source power in the tunnel surrounding rock. The model includes: Among them 2 T represents the Laplace operator, the second-order spatial derivative, which describes the spatial distribution of temperature, β represents the thermal diffusivity, qs represents the fire source power in the tunnel surrounding rock, and τq represents the thermal relaxation time. After solving the equation, the maximum temperature gradient is obtained. The sensor setting range Rsen is obtained through the formula; Based on the sensor range Rsen, the point cloud twin topology potential field function is constructed using the two-dimensional point cloud data of the roadway. By solving the extreme point of the point cloud twin topology potential field, if the gradient value of the point cloud twin topology potential field is (x) = 0 and Cur(x) > κth, where κth is the curvature threshold, then this point is an extreme point, and the extreme points constitute the deployment coordinate set {xi} of the main sensor; Based on the deployment coordinate set {xi} of the main sensor, coverage is performed with the main node xi as the center and the sensor range Rsen as the radius. The uncovered area in the lane is identified and marked as the gap area Ω. The candidate redundant node position {xj} is generated in the gap area Ω.

4. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 1, characterized in that: The chaotic spectrum time domain purification data analysis steps are as follows: Based on the final sensor deployment plan {xi, x′j}, the original sensor data stream data s(t) is collected and the node residual energy is calculated through the node energy consumption Ej; Based on the original sensor data stream and the node residual energy, the Lorenz chaotic system is dynamically modulated, and the chaotic modulation is performed according to the residual energy ratio of the node. The differential equation is discretized and the state variables are iteratively updated to obtain the energy-coupled chaotic encrypted data stream senc(t); Based on the energy-coupled chaotic encrypted data stream, the energy-coupled chaotic carrier spectrum is obtained through Fourier transform. The non-carrier frequency band noise is suppressed by convolutional neural differential equations on the energy-coupled chaotic encrypted data stream and the energy-coupled chaotic carrier spectrum. The loss function training is established through clean data, and the decoding network decodes the hidden state to obtain the chaotic spectrum time domain purification data.

5. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 1, characterized in that: The symbiotic strategy routing table {π j The analysis steps are as follows: Based on the chaotic spectrum time domain purification data and node residual energy, the information entropy of each node is calculated, and the game utility function is established based on the information entropy of each node. Each node independently updates the routing strategy and selects the probability distribution of the next hop. When all nodes meet the set conditions, the iteration is stopped and the symbiotic strategy routing table {π j }.

6. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 1, characterized in that: The steps for analyzing the fire probability value based on the object are as follows: The spatiotemporal feature vector Fte is pre-trained with the fully connected layer weight vQ to output the predicted value of heat accumulation The spatiotemporal feature vector Fte is passed through the pre-trained weight matrix WJ to output the heat flow prediction value Based on the true fire label ytrue and the probability value of fire occurrence Pfir, the cross entropy loss function is used to obtain the true disaster probability cross entropy CroEntr; Based on the cross entropy CroEntr of the true disaster probability and the predicted value of heat accumulation and heat flow predictions Through the meta-learning early warning model, the real fire label ytrue is corrected and optimized with data-physics dual constraints to obtain the object-optimized fire probability value Pfir1; The fire probability value Pfir1 based on the object is compared with the set threshold. If it is greater than the set threshold, the positioning algorithm signal is triggered.

7. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 6, characterized in that: The analysis steps of the probability value Pfir of fire occurrence are as follows: Based on the chaotic spectrum time-domain purified data P(x) and the final sensor deployment plan, a three-dimensional tensor Xsp is constructed through the Dirac function, and the core features are extracted through high-order singular value decomposition to obtain the spatiotemporal feature vector Fte; The spatiotemporal feature vector Fte is input into the machine learning model. The model is trained with the real fire label ytrue as supervision, and the model weights and bias are adjusted. The model performs a linear transformation on the input spatiotemporal feature vector Fte and maps the output to the [0,1] interval through the sigmoid function to obtain the probability value Pfir of fire occurrence.

8. A coal mine tunnel fire prevention and extinguishing monitoring and positioning system according to claim 1, characterized in that: The analysis steps of the verified true fire source coordinates (x**, y**) are as follows: Based on the object-optimized fire probability value Pfir1 and the roadway point cloud data, the neural radiation field model constructs an implicit fire source field through a multi-layer perceptron. The model is trained by training the rendering loss function to obtain the implicit 3D fire source field σ(x). Based on the implicit 3D fire source field σ(x) and the final sensor deployment plan {xi, x′j}, the local state sj of each node j is defined. The coverage quality and communication cost are combined through the MA-DDPG reward function rj. The Actor network combines the reward function through the MA-DDPG framework and outputs the node weight wj. The collaborative positioning coordinates (x*, y*) are obtained by weighted average calculation based on the node weight wj; Based on the collaborative positioning coordinates (x*, y*), a Vietoris-Rips complex is constructed with the maximum distance between sensor nodes as the scale. Its H1 persistence graph is calculated to extract topological void features, including the birth radius and death radius of the void. It is verified whether there is a void feature that satisfies the condition that the difference between the birth and death radius exceeds the set topological threshold, and the distance between the preliminary positioning coordinates and the geometric center of the void is less than the average radius of the void. If the conditions are met, the geometric center of the void is output as the verified true fire source coordinates (x**, y**); if not, the positioning calculation is repeated until the conditions are met.

Images