Intelligent fire-fighting monitoring method and system based on multi-dimensional data
By performing benchmark alignment and feature interpolation calculations on multi-dimensional fire monitoring data, and combining it with a graph neural network model, the problem of spatiotemporal asynchrony of data in multi-dimensional fire monitoring was solved, enabling accurate identification and early prevention of fires.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 华能陕西子长发电有限公司
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-31
AI Technical Summary
In the existing three-dimensional fire monitoring system, the spatiotemporal asynchronous problem caused by the difference in data acquisition frequency and format of multi-dimensional sensing devices leads to delays in fire identification and low accuracy in comprehensive judgment.
By acquiring positioning information, timing information, three-dimensional wind field vectors, and fixed-point environmental parameters, a sensing data stream with absolute timestamps is generated. Benchmark alignment and physical feature interpolation calculations are performed to construct a thermodynamic directional mapping matrix. A graph neural network model is used for data fusion and feature extraction to output early warning signals and response commands.
It effectively solves the problem of asynchronous data in time and space from multi-dimensional sensing devices, enables accurate identification and early blocking of fires, eliminates the defects of information misalignment in time and space, and improves the timeliness and accuracy of fire early warning.
Smart Images
Figure CN122493584A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fire safety perception and multi-source data fusion technology, and in particular to an intelligent fire monitoring method and system based on multi-dimensional data. Background Technology
[0002] Fire sensing technology is developing towards a comprehensive and three-dimensional approach, with multi-dimensional collaborative monitoring systems becoming the mainstream technical solution for fire prevention in complex scenarios. In this multi-level sensing network, space satellite networks provide wide-area positioning and timing support, aerial drones perform mobile patrols and on-site image acquisition, and ground-based sensor terminals complete fixed-point data acquisition and command interaction. These three components together form a three-dimensional sensing network.
[0003] Existing three-dimensional fire monitoring systems suffer from insufficient multi-terminal coordination, specifically manifested in significant differences in data acquisition frequencies and formats across different dimensions of equipment, and the lack of standardized spatial coordinate systems across multiple terminals. These objective differences prevent the sensing platform from establishing an efficient collaborative mechanism, leading to delays in multi-source data interaction and spatiotemporal misalignment of information. For example, in the initial stages of a fire in a remote mountainous area, only trace amounts of smoke and localized temperature rise are produced. Due to physical coverage blind spots of fixed ground monitoring equipment, and the lack of spatiotemporal synchronization between mobile aerial patrol equipment and satellite networks, the scattered information collected by the sensing center cannot be deeply integrated. Ultimately, the significant time delays in multi-source information processing make it difficult for the integrated sensing center to output accurate fire warning signals in a timely manner. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides an intelligent fire monitoring method and system based on multi-dimensional data. This invention solves the technical problem that the failure of multi-dimensional sensing devices to achieve spatiotemporal synchronization of data results in delays in initial fire identification and low accuracy of comprehensive judgment.
[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the present invention provides an intelligent fire monitoring method based on multi-dimensional data, comprising: The system acquires positioning and timing information from a space network, collects three-dimensional wind field vectors and aerial environmental parameters from a mobile terminal, and collects fixed-point environmental parameters from a fixed terminal. It then aggregates the positioning information, timing information, three-dimensional wind field vectors, aerial environmental parameters, and fixed-point environmental parameters to generate a sensing data stream with an absolute timestamp. The absolute timestamp is extracted from the sensing data stream. The aerial environmental parameters and the fixed-point environmental parameters are aligned based on the absolute timestamp. Physical feature interpolation is performed on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence. A thermodynamic directed mapping matrix is constructed based on the positioning information and the three-dimensional wind field vector. The state feature compensation sequence and the thermodynamic directed mapping matrix are concatenated to output a comprehensive sensing data matrix. The comprehensive sensing data matrix is input into the graph neural network model, the state feature compensation sequence is mapped to network nodes and corresponding network node vectors, the thermodynamic directed mapping matrix is mapped to directed connection edges between the network nodes, the weight parameters of the directed connection edges are updated according to the three-dimensional wind field vector, node aggregation is performed to extract features and output the state sensing sequence. Physical parameters are extracted from the state-aware sequence, and the gradient rate of change of the physical parameters is calculated. An early warning signal is output when the gradient rate of change is greater than a preset threshold. The thermal radiation core coordinates are extracted from the state-aware sequence, and the physical thermal diffusion path is calculated by combining the three-dimensional wind field vector. The target spatial location obtained by the physical thermal diffusion path is extracted, and a disaster prevention action sequence matching the target spatial location is obtained. The target spatial location and the disaster prevention action sequence are combined to generate a response command, and the response command is sent.
[0006] Furthermore, the intelligent fire monitoring method based on multi-dimensional data described in this invention, wherein the step of performing physical feature interpolation calculation to generate a state feature compensation sequence by aligning the aerial environmental parameters and the fixed-point environmental parameters includes: The aligned aerial environmental parameters and the fixed-point environmental parameters are input into the thermodynamic physical model to calculate the physical decay gradient and physical diffusion gradient of environmental heat in the time dimension. Extract the time series gap between the aligned aerial environmental parameters and the fixed-point environmental parameters; Substitute the physical decay gradient and the physical diffusion gradient into the time series gap; The supplementary state characteristics within the time series gap are calculated based on the time step size of the time series gap. The aligned aerial environmental parameters, aligned fixed-point environmental parameters, and supplementary state features are merged and aligned, and the supplementary state features are output as a compensated sequence of the state features.
[0007] Furthermore, in the intelligent fire monitoring method based on multi-dimensional data described in this invention, the step of constructing a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector includes: The relative physical distance between the mobile terminal that collects the aerial environmental parameters and the fixed terminal that collects the fixed-point environmental parameters is extracted based on the positioning information. The relative physical space distance is multiplied by the three-dimensional wind field vector to obtain the wind field distance coupling coefficient. The normalized distribution calculation is performed on the wind field distance coupling coefficient to obtain the physical directed probability representing the heat transfer between the mobile terminal and the fixed terminal; Using the physical directed probabilities as matrix elements, a thermodynamic directed mapping matrix is constructed to indicate the direction of environmental thermal flow.
[0008] Furthermore, in the intelligent fire monitoring method based on multi-dimensional data described in this invention, updating the weight parameters of the directed connection edges according to the three-dimensional wind field vector includes: Extract the wind speed and wind direction components from the three-dimensional wind field vector; The wind speed component is set as a weighted step size, and the wind direction component is set as a weighted direction. Extract the matrix elements from the thermodynamic directed mapping matrix; Based on the weighted step size and the weighted direction, the matrix elements are subjected to vector dot product calculation to obtain the updated feature values; The updated feature values are set as the weight parameters.
[0009] Furthermore, the intelligent fire monitoring method based on multi-dimensional data described in this invention, wherein the execution node aggregates and extracts features and outputs a state-aware sequence, includes: According to the weight parameters, the network node vector is passed along the directed connection edge to the adjacent network node; The network node vectors passed to the adjacent network nodes are weighted and summed to obtain the node feature update variables; The network node vector and the node feature update variable are combined by feature concatenation to extract the physical features of thermal propagation with thermodynamic topological association. The physical characteristics of heat propagation are subjected to dimensionality reduction mapping calculation, and the dimensionality-reduced physical characteristics of heat propagation are arranged according to the time evolution order to output the state perception sequence.
[0010] Furthermore, in the intelligent fire monitoring method based on multi-dimensional data described in this invention, the step of extracting physical parameters from the state perception sequence and calculating the gradient rate of change of the physical parameters includes: Temperature and smoke concentration feature values are extracted from the state-sensing sequence as the physical parameters. Extract the numerical differences of the physical parameters at different time points; Divide the numerical difference by the time interval between the different time points to obtain the slope of the physical parameter's change in physical quantity over time. The slope of the change in the physical quantity is set as the gradient rate of change.
[0011] Furthermore, the intelligent fire monitoring method based on multi-dimensional data described in this invention, wherein extracting the core coordinates of thermal radiation from the state perception sequence and calculating the physical thermal diffusion path by combining the three-dimensional wind field vector includes: Extract the feature node with the largest thermal radiation value from the state-aware sequence and set it as the thermal radiation peak node; The thermal radiation peak nodes are mapped to a three-dimensional spatial coordinate system to obtain the coordinates of the thermal radiation core; Using the coordinates of the thermal radiation core as the spatial starting point, a time-step spatial coordinate offset accumulation calculation is performed along the direction indicated by the three-dimensional wind field vector to obtain a set of spatial offset coordinates. By fitting the set of spatial offset coordinates, the physical thermal diffusion path is obtained.
[0012] Furthermore, the intelligent fire monitoring method based on multi-dimensional data described in this invention, wherein the step of extracting the target spatial location calculated from the physical thermal diffusion path, obtaining a disaster prevention action sequence matching the target spatial location, and combining the target spatial location and the disaster prevention action sequence to generate a response command includes: Input the target spatial location into the emergency action database to query matching local cooling disaster prevention actions and spatial isolation disaster prevention actions; The local cooling disaster prevention actions and the spatial isolation disaster prevention actions obtained from the query are sorted in time according to the preset time execution priority, and the disaster prevention action sequence is obtained by combining them. The target spatial location and the disaster prevention action sequence are encapsulated into a standard instruction format, and the response instruction is output.
[0013] Furthermore, in the intelligent fire monitoring method based on multi-dimensional data described in this invention, sending the response command includes: The generated response instructions are parsed to obtain the disaster prevention action sequence and the target spatial location; Extract the local cooling disaster prevention action from the disaster prevention action sequence, compile the local cooling disaster prevention action into a physical suppression signal for the coordinates of the thermal radiation core, and send it directionally to the disaster prevention execution terminal corresponding to the target spatial location; The spatial isolation disaster prevention action is extracted from the disaster prevention action sequence, and the spatial isolation disaster prevention action is compiled into an isolation and cooling signal for the front end of the physical heat diffusion path and sent to the pipeline terminal in a directional manner.
[0014] Secondly, the present invention provides an intelligent fire monitoring system based on multi-dimensional data, applied to the intelligent fire monitoring method based on multi-dimensional data as described above, comprising: The perception acquisition module is used to acquire positioning information and timing information from the space network, collect three-dimensional wind field vectors and air environment parameters from the mobile terminal, collect fixed-point environmental parameters from the fixed terminal, and aggregate the positioning information, the timing information, the three-dimensional wind field vectors, the air environment parameters and the fixed-point environmental parameters to generate a perception data stream with an absolute timestamp. The spatiotemporal reconstruction module is used to extract the absolute timestamps from the sensing data stream to perform benchmark alignment, perform physical feature interpolation calculations on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence, construct a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector, and splice the state feature compensation sequence and the thermodynamic directed mapping matrix to output a comprehensive sensing data matrix. The topology fusion module is used to input the comprehensive sensing data matrix into the graph neural network model, map the state feature compensation sequence into network nodes and corresponding network node vectors, map the thermodynamic directed mapping matrix into directed connection edges between the network nodes, update the weight parameters of the directed connection edges according to the three-dimensional wind field vector, perform node aggregation to extract features and output the state sensing sequence. The intelligent judgment module is used to extract physical parameters from the state perception sequence and calculate the gradient change rate. When the gradient change rate is greater than a preset threshold, it outputs an early warning signal, extracts the core coordinates of thermal radiation from the state perception sequence, calculates the physical thermal diffusion path by combining the three-dimensional wind field vector, extracts the target spatial location calculated by the physical thermal diffusion path, obtains the disaster prevention action sequence matching the target spatial location and generates a response command, and sends the response command.
[0015] Beneficial effects of this invention: Heterogeneous terminals independently collect data in multidimensional space, resulting in physical time asynchrony errors exceeding preset thresholds. The data aggregation logic extracts absolute timestamps from the sensing data stream and uses these timestamps to perform benchmark alignment between aerial and fixed-point environmental parameters. Addressing the time discontinuity caused by the difference in sampling frequencies between mobile and fixed terminals, the aligned aerial and fixed-point environmental parameters are input into a thermodynamic physical model to calculate the physical attenuation and diffusion gradients of environmental heat over time. The computational logic substitutes these gradients into the time series gaps to calculate supplementary state features. The aligned environmental parameters and supplementary state features are then merged to output a smooth state feature compensation sequence. This time benchmark alignment, combined with physical feature interpolation, effectively overcomes the spatiotemporal misalignment caused by differences in data acquisition frequencies from multidimensional sensing devices in existing fire monitoring systems, eliminating the discontinuity of underlying data over time.
[0016] The complex geographical environment of remote mountainous areas and photovoltaic arrays creates objective physical sensing coverage blind spots for ground-based fixed terminals. Spatial correlation computation extracts the relative physical spatial distance between mobile and fixed terminals based on positioning information. This relative physical spatial distance is then multiplied by a 3D wind field vector to obtain a wind field distance coupling coefficient. Normalized distribution calculation transforms the wind field distance coupling coefficient into a physically directed probability representing heat transfer, thereby constructing a thermodynamic directed mapping matrix indicating the direction of environmental thermal flow. A graph neural network model maps the state feature compensation sequence to corresponding network node vectors and the thermodynamic directed mapping matrix to directed connecting edges. The weight parameters of the directed connecting edges are updated based on the dynamically changing 3D wind field vector, driving the network node vectors to propagate along the directed connecting edges and perform node aggregation. The thermodynamic directed mapping matrix construction process, in conjunction with graph neural network feature aggregation calculation, transforms a discrete 3D spatial coordinate matrix into a dynamic topological network reflecting objective airflow disturbances and thermal radiation flow patterns, compensating for the physical coverage blind spots of the ground-based fixed terminals.
[0017] The slight temperature rise and trace smoke accumulation in the early stages of concealed fires may be masked by interference signals from complex local environments. The intelligent assessment module extracts temperature and smoke concentration features from the state-aware sequence output by the graph neural network model as physical parameters. The logical calculation process extracts the numerical differences of these physical parameters at different time points and derives the slope of the physical quantity change. The system sets this slope as the gradient rate of change, and directly outputs a warning signal when the gradient rate of change exceeds a preset threshold. The disaster prevention tracking process extracts the feature node with the largest thermal radiation value from the state-aware sequence and maps it to the core coordinates of the thermal radiation. Combined with a three-dimensional wind field vector, it performs spatial coordinate offset accumulation calculations to obtain the physical thermal diffusion path. The target spatial location obtained from the physical thermal diffusion path calculation is extracted, and local cooling and spatial isolation disaster prevention actions are matched with those from the emergency action library to form a disaster prevention action sequence. The target spatial location and disaster prevention action sequence are encapsulated to generate response commands and sent in a targeted manner. The calculation of gradient change rate, combined with the deduction of physical and thermal diffusion paths and the sending of response commands, constructs a closed-loop data interaction link from the capture of weak abnormal energy to the response of pre-emptive physical suppression. This effectively solves the technical defects of time delay caused by multi-source information processing and the difficulty of timely output of accurate fire early warning signals by the comprehensive sensing center, and realizes accurate identification and early physical blocking of fires in complex scenarios. Attached Figure Description
[0018] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating an intelligent fire monitoring method based on multi-dimensional data according to the present invention. Detailed Implementation
[0020] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] Firstly, please refer to Figure 1 The present invention provides an intelligent fire monitoring method based on multi-dimensional data, comprising: Step 1: Obtain positioning and timing information from the space network, collect three-dimensional wind field vectors and aerial environmental parameters from the mobile terminal, collect fixed-point environmental parameters from the fixed terminal, and aggregate the positioning information, timing information, three-dimensional wind field vectors, aerial environmental parameters and fixed-point environmental parameters to generate a sensing data stream with absolute timestamps; Step 2: Extract the absolute timestamp from the sensing data stream; perform benchmark alignment between the aerial environmental parameters and the fixed-point environmental parameters based on the absolute timestamp; perform physical feature interpolation calculation on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence; construct a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector; and concatenate the state feature compensation sequence and the thermodynamic directed mapping matrix to output a comprehensive sensing data matrix. Step 3: Input the comprehensive sensing data matrix into the graph neural network model, map the state feature compensation sequence into network nodes and corresponding network node vectors, map the thermodynamic directed mapping matrix into directed connection edges between the network nodes, update the weight parameters of the directed connection edges according to the three-dimensional wind field vector, perform node aggregation to extract features and output the state sensing sequence. Step 4: Extract the physical parameters from the state perception sequence, calculate the gradient rate of change of the physical parameters, and output an early warning signal when the gradient rate of change is greater than a preset threshold; extract the core coordinates of thermal radiation from the state perception sequence, and calculate the physical thermal diffusion path by combining the three-dimensional wind field vector; extract the target spatial location obtained by the calculation of the physical thermal diffusion path, obtain the disaster prevention action sequence matching the target spatial location, combine the target spatial location and the disaster prevention action sequence to generate a response command, and send the response command.
[0022] In the complex geographical environment where large-scale photovoltaic power plants intersect with remote mountainous areas, environmental sensing elements exhibit multi-dimensional spatial distribution characteristics. During operation, the monitoring system acquires positioning and timing information from a spatial network. Positioning information includes longitude, latitude, and elevation data, while timing information provides a globally unified standard time reference. Simultaneously, mobile terminals perform mobile patrol flights within the monitored airspace, carrying multi-source sensor arrays to collect three-dimensional wind field vectors and aerial environmental parameters. The three-dimensional wind field vector includes wind direction deflection angles and absolute wind speed values in a three-dimensional geographic coordinate system, while aerial environmental parameters include upper-altitude temperature gradient values, upper-altitude suspended dust concentration indicators, and upper-altitude relative humidity values. Fixed terminals distributed within the photovoltaic panel arrays and at specific locations on the mountain surface synchronously collect fixed-point environmental parameters, covering surface temperature values of ground equipment, characteristics of underlying smoke concentration, and thermal radiation values of flammable vegetation surface. Subsequently, the perception acquisition module performs communication protocol parsing and data format integration on the collected positioning information, timing information, three-dimensional wind field vector, aerial environmental parameters, and fixed-point environmental parameters, and aggregates them to generate a perception data stream with an absolute timestamp. The absolute timestamp is used as a unified timing reference to eliminate the time asynchronous error caused by independent acquisition by heterogeneous perception hardware.
[0023] Given the discontinuous and discrete distribution of environmental monitoring data across multiple dimensions and time axes, the spatiotemporal reconstruction module extracts absolute timestamps from the sensing data stream and performs benchmark alignment between airborne and fixed-point environmental parameters based on these timestamps. Addressing the data gaps caused by differences in flight trajectory changes between mobile terminals and the hardware sampling periods of fixed terminals, the data processing logic inputs the aligned airborne and fixed-point environmental parameters into a thermodynamic physical model. Based on fluid dynamics laws and fundamental physical laws of heat conduction, the physical attenuation and diffusion gradients of environmental heat over time are calculated. The background processing program further extracts the time-series gaps between the aligned parameters caused by hardware sampling delays, incorporating the physical attenuation and diffusion gradients as dynamic compensation variables into these gaps. Following the microsecond-level time steps within the time-series gaps, supplementary state characteristics within these gaps are progressively calculated, reflecting the environmental physical evolution within the unsampled time window. The processing system merges the aligned airborne environmental parameters, aligned fixed-point environmental parameters, and supplementary state characteristics, outputting a state characteristic compensation sequence to fill in missing physical quantities, thus constructing a continuous panoramic mapping of physical features on a temporal scale.
[0024] Heat transfer in objective physical space is influenced by a combination of absolute geographical distance and airflow patterns. The topology fusion logic, based on location information, extracts the relative physical distance between mobile terminals collecting aerial environmental parameters and fixed terminals collecting fixed-point environmental parameters, reflecting the linear span between three-dimensional sensing nodes. To quantify the objective laws of environmental heat conduction under wind field influence, the mathematical operation program performs matrix multiplication on the relative physical distance and the three-dimensional wind field vector, calculating the wind field distance coupling coefficient, which integrates spatial span and wind potential energy. The higher the value, the stronger the physical tendency for heat exchange between corresponding spatial nodes. Subsequently, the matrix operation logic performs normalization distribution calculation on the wind field distance coupling coefficient, mapping the absolute value to a set standard value range, deriving the physical directed probability representing heat transfer between the mobile and fixed terminals. The operation system uses the physical directed probability as independent matrix elements to construct a thermodynamic directed mapping matrix indicating the physical direction of environmental thermal flow. After generating the thermodynamic directed mapping matrix, the splicing program merges the state feature compensation sequence with the thermodynamic directed mapping matrix, combining and outputting a comprehensive sensing data matrix that includes underlying node attributes and three-dimensional spatial relationships.
[0025] To deeply mine the potential features in multi-source sensing fusion data, the algorithm logic layer inputs the comprehensive sensing data matrix into a pre-constructed graph neural network model. The graph neural network model maps the state feature compensation sequence to network nodes and corresponding network node vectors, representing mobile and fixed terminals actually deployed in photovoltaic power plants and mountainous areas, carrying the collected and compensated physical feature values of temperature and smoke. Simultaneously, it maps the thermodynamic directed mapping matrix to directed connections between network nodes, representing implicit airflow disturbances and thermal radiation flow channels in the physical space. In scenarios of dynamic changes in the wind field, the extraction logic retrieves the wind speed and wind direction components from the three-dimensional wind field vector. The graph neural network model sets the wind speed component as the weighted step size and the wind direction component as the weighted direction. Based on the weighted step size and weighted direction, the matrix elements in the thermodynamic directed mapping matrix are obtained and subjected to vector dot product calculations with direction matching to obtain updated feature values that change in real time with the wind field. These updated feature values are set as weight parameters for the directed connections, ensuring that the underlying topology of the graph neural network model remains synchronously related to the external three-dimensional wind field environment.
[0026] The feature interaction and transmission logic between network communication nodes is controlled by wind field-driven weight parameters. The graph neural network model, based on dynamically updated weight parameters, transmits network node vectors along directed edges to adjacent network nodes, simulating the physical process of environmental heat energy and smoke particles diffusing to the surrounding environment with airflow. After receiving the transmitted feature information, the target network node performs a weighted summation calculation on the network node vectors transmitted to adjacent network nodes, extracting node feature update variables reflecting the comprehensive influence of the surrounding physical environment. Subsequently, the feature combination logic concatenates and combines the original network node vectors with the newly generated node feature update variables, extracting thermal propagation physical features with thermodynamic topological correlations through multi-dimensional feature cross-fusion. To reduce the matrix computation complexity of subsequent high-dimensional features, the model performs dimensionality reduction mapping calculations on the thermal propagation physical features while preserving the variance of the core data. The processing program arranges the dimensionality-reduced thermal propagation physical features in chronological order, outputting a state-aware sequence that dynamically reflects the overall physical evolution trend of the environment.
[0027] The state-aware sequence contains rich, dynamic, and temporal information. The intelligent judgment module extracts temperature and smoke concentration characteristics from this sequence as physical parameters characterizing changes in the underlying environmental state. To capture the objective evolution of initial concealed fires, the intelligent judgment module extracts the numerical differences of these physical parameters at different time points. The computational program divides these differences by the corresponding time interval to obtain the slope of the physical parameter's change over time, directly reflecting the gradient of local environmental temperature increases and smoke accumulation. The judgment logic sets the slope of the physical quantity change as the gradient rate of change. During numerical comparison calculations, when the calculation nodes distributed in the local photovoltaic array or forest boundary zone determine that the gradient rate of change exceeds a pre-defined safety threshold, it indicates an abnormal energy change in the underlying environmental physical characteristics. The system immediately generates and outputs a warning signal indicating the specific location of the initial hidden danger, triggering an alarm intervention.
[0028] After confirming the abnormal physical state of the environment, the intelligent judgment module enters the spatial thermal tracing and source-tracing stage. It traverses the state-sensing sequence of the algorithm, extracts the specific feature node with the highest thermal radiation value, and sets it as the thermal radiation peak node. The mapping program reverse-maps the thermal radiation peak node to the three-dimensional spatial coordinate system according to the underlying geographic mapping rules, calculating the core coordinates of thermal radiation in physical space, which serves as the absolute center location of the physical origin of the abnormal heat source. Using the core coordinates of thermal radiation as the spatial starting point, the intelligent judgment module loads a real-time three-dimensional wind field vector. The deduction logic performs time-step spatial coordinate offset accumulation calculations along the direction indicated by the three-dimensional wind field vector, deducing the specific geographical trajectory of the abnormal heat energy spreading with the wind, and obtaining a discrete set of spatial offset coordinates. Finally, a mathematical smoothing curve generation algorithm is applied to the set of spatial offset coordinates for curve fitting, outputting a continuous physical thermal diffusion path with directional attributes.
[0029] The physical thermal diffusion path clearly guides the macroscopic physical spread of potential fires. A geographic retrieval program searches a geographic information database along this path, extracting the coordinates of completely covered sensitive photovoltaic equipment components and flammable forest vegetation areas, and uniformly sets these as the target spatial location. The system calls the underlying emergency coordination processing program, inputting the target spatial location as a query index into a pre-set emergency action library. It queries for localized cooling and disaster prevention actions and spatial isolation actions that physically match the surrounding terrain. Localized cooling actions include activating fixed-point high-pressure water cannons, while spatial isolation actions include opening a water curtain barrier via a pipeline network. The scheduling system retrieves the built-in execution attributes of the queried actions, sorts them chronologically according to a preset time priority, and combines them to generate a disaster prevention action sequence with the underlying hardware execution logic order. The instruction encapsulation interface encapsulates the target spatial location and disaster prevention action sequence into a standard instruction format according to the communication protocol requirements of the underlying control equipment, generating response instructions that can be directly recognized by the underlying physical hardware drivers on-site.
[0030] The on-site hardware control facilities execute physical environment disaster prevention closed-loop actions based on the received standard protocol format data. The control center receives and sends response commands, parsing layer by layer to obtain the bottom-level executable disaster prevention action sequence and the corresponding target spatial location. The control logic extracts the local cooling disaster prevention action with the highest execution priority from the disaster prevention action sequence, and compiles it into a physical suppression signal targeting the core coordinates of the heat radiation core, combined with the bottom-level hardware signal driver. The physical suppression signal is then directionally sent to the disaster prevention execution terminal corresponding to the target spatial location, driving the high-pressure water cannon equipment or mobile fire extinguishing equipment deployed on-site to spray cooling and fire extinguishing media towards the core coordinates of the abnormal heat source. In the synchronous control link, the control logic extracts the spatial isolation disaster prevention action with anti-spread blocking attributes from the disaster prevention action sequence, compiles the spatial isolation disaster prevention action into an isolation cooling signal targeting the front end of the physical heat diffusion path, and directionally sends it to the pipeline terminal deployed in front of the fire spread path, directly triggering the physical water pipeline network to execute an environmental isolation cooling blocking action by actively opening the sprinkler valve.
[0031] Secondly, the present invention provides an intelligent fire monitoring system based on multi-dimensional data, applied to the intelligent fire monitoring method based on multi-dimensional data as described above, comprising: The perception acquisition module is used to acquire positioning information and timing information from the space network, collect three-dimensional wind field vectors and air environment parameters from the mobile terminal, collect fixed-point environmental parameters from the fixed terminal, and aggregate the positioning information, the timing information, the three-dimensional wind field vectors, the air environment parameters and the fixed-point environmental parameters to generate a perception data stream with an absolute timestamp. The spatiotemporal reconstruction module is used to extract the absolute timestamps from the sensing data stream to perform benchmark alignment, perform physical feature interpolation calculations on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence, construct a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector, and splice the state feature compensation sequence and the thermodynamic directed mapping matrix to output a comprehensive sensing data matrix. The topology fusion module is used to input the comprehensive sensing data matrix into the graph neural network model, map the state feature compensation sequence into network nodes and corresponding network node vectors, map the thermodynamic directed mapping matrix into directed connection edges between the network nodes, update the weight parameters of the directed connection edges according to the three-dimensional wind field vector, perform node aggregation to extract features and output the state sensing sequence. The intelligent judgment module is used to extract physical parameters from the state perception sequence and calculate the gradient change rate. When the gradient change rate is greater than a preset threshold, it outputs an early warning signal, extracts the core coordinates of thermal radiation from the state perception sequence, calculates the physical thermal diffusion path by combining the three-dimensional wind field vector, extracts the target spatial location calculated by the physical thermal diffusion path, obtains the disaster prevention action sequence matching the target spatial location and generates a response command, and sends the response command.
[0032] The specific formula for calculating physical feature interpolation is as follows:
[0033] In the formula, Represents supplementary state characteristics; This represents the aligned aerial environmental parameters or the aligned fixed-point environmental parameters. This represents the physical decay gradient of environmental heat over time. This represents the physical diffusion gradient of environmental heat over time. This represents the microsecond-level time step within the time series interval. The specific formula for calculating the product of the wind field distance coupling coefficient is as follows:
[0034] In the formula, Represents the wind field distance coupling coefficient; The three-dimensional wind field vector represents the data collected by the mobile terminal. This represents the unit direction vector between the mobile terminal and the fixed terminal. This represents the relative physical spatial distance between mobile and fixed terminals. The specific formula for calculating the physical directed probability normalized distribution is as follows:
[0035] In the formula, Represents physical directed probability; The wind field distance coupling coefficient represents the current target node being calculated. Represents the first in the surrounding adjacent space. Wind field distance coupling coefficient of each node; Represents the total number of nodes in the three-dimensional space within the surrounding adjacent spatial range; This represents an exponential function with the natural constant as its base. The updated formula for calculating the dot product of the feature vectors is as follows:
[0036] In the formula, This represents the updated feature value, which is then set as the weight parameter of the directed connection edge. Represents the matrix elements in the thermodynamic directed mapping matrix; This represents the wind direction component included in the three-dimensional wind field vector, and the wind direction component is set as a weighted direction; The physical topological direction vector representing the directed connection edges between network nodes; This represents the wind speed component included in the three-dimensional wind field vector, with the wind speed component set as a weighted step size. The specific formula for calculating the slope of the physical quantity change is as follows:
[0037] In the formula, This represents the slope of the change in a physical quantity, which is ultimately set as the gradient rate of change. This represents the difference in extracted temperature or smoke concentration values at different time points. It represents the time interval between different points in time.
[0038] The specific formula for calculating the cumulative spatial coordinate offset is as follows:
[0039] In the formula, Represents discrete three-dimensional offset coordinate points in a set of spatial offset coordinates; Represents the coordinates of the thermal radiation core extracted from the state-aware sequence; This represents the total number of clock cycles during the simulation process; The direction vector indicating the three-dimensional wind field vector; It represents the absolute time span value for each time step.
[0040] In a computational embodiment focusing on underlying hardware collaboration and specific numerical data transfer, the 64-bit central processing unit (CPU) within the data fusion node reads the underlying hardware registers of the mobile and fixed terminals via a high-speed peripheral component interconnect bus. The CPU extracts the aligned airborne environmental parameter temperature value. Degrees Celsius. The tensor processing physics unit loads a physical decay gradient value of [value missing] over a 120 MHz clock cycle. The physical diffusion gradient value is calculated at 1 degree Celsius per microsecond. Every microsecond, the time intervals recorded by the hardware timer are extracted, including microsecond-level time steps. Microseconds. The physics acceleration engine, which uses the multiply-accumulate operation of the tensor processing physics unit, substitutes the physical feature interpolation calculation formula and performs parallel calculations within the internal arithmetic logic unit to obtain the supplementary state feature temperature value. Temperature in Celsius. The relative physical distance for reading from the cache memory inside the central processing unit is... Meters, read the absolute wind speed value from the three-dimensional wind field vector. Meters per second, with the unit direction vector set to be perfectly parallel to the wind field vector. The arithmetic logic unit performs floating-point multiplication to obtain the wind field distance coupling coefficient. The hardware acceleration board extracts two adjacent absolute timestamps recorded by the hardware clock source and reads the time interval between different time points. Seconds, the difference between the extracted temperature feature values at different time points is read. Celsius. The floating-point arithmetic unit calls the underlying hardware division instruction to calculate the slope of the change in the physical quantity. Celsius per second.
[0041] The central processing unit sets the slope of the physical quantity change as the gradient rate of change, and compares it with the safety threshold value stored in the preset threshold register. The Celsius value is compared bit by bit every second. The conditional jump instruction determines the rate of gradient change. The temperature per second is greater than the preset threshold. At a temperature of 1 degree Celsius per second, the interrupt controller immediately sends a high-level interrupt warning signal to the digital logic controller of the linkage handling node. The digital logic controller parses the generated response instruction and sends a 16-bit pulse width modulation control signal to the pipeline terminal, activating the drive coil of the water pipeline solenoid valve to open the environmental isolation cooling blocking action.
[0042] In a specific embodiment of the present invention, the verification process of instruction coordination and data throughput of the underlying hardware system is described in detail below: In monitoring systems deployed in remote mountainous areas and large photovoltaic power plants, the underlying physical hardware units undertake the collaborative acquisition and throughput of multi-dimensional spatially distributed data. Fixed terminals deployed within the photovoltaic panel arrays embed 32-bit reduced instruction set microprocessors (RISC) units. The microprocessor unit's main frequency clock generator provides a 120 MHz reference operating frequency. The microprocessor unit establishes direct physical communication links with temperature and smoke sensors via an internal integrated circuit protocol bus. Within specific microsecond-level clock cycles, the microprocessor unit sends addressing control commands to a specific physical address in a 16-bit addressing space. The temperature and smoke sensors receive the addressing control commands and convert the detected fixed-point environmental parameters into analog electrical signals. The analog-to-digital converter (ADC) front-end physical chip converts the analog electrical signals into a 12-bit register physical width digital sequence. The microprocessor unit reads the digital sequence and stores it in a local static cache register. Mobile terminals performing mobile patrol flights within the monitored remote mountainous airspace are equipped with 64-bit multi-core heterogeneous coprocessors. The multi-core heterogeneous coprocessor establishes an electrical connection with a high-precision anemometer and meteorological probe via a high-speed serial computer expansion bus, and periodically reads the three-dimensional wind field vector and air environment parameters.
[0043] The wireless radio frequency transceiver antennas inside both fixed and mobile terminals transmit low-level data packets to the data fusion node. The data fusion node carries a central processing unit (CPU) and an independent memory physical controller. Upon receiving the low-level data packets, the independent memory physical controller triggers a hardware-level data bus interrupt request instruction. The CPU responds to the interrupt request instruction, suspends the main program pointer, and initiates a direct memory access physical transmission channel (DMI). The DMI bypasses the CPU's core computation unit and directly transports the low-level data packets to the synchronous dynamic random access memory (DRAM) physical address space, converging them to generate a sensing data stream with an absolute timestamp. The CPU's general-purpose input / output interface reads the sensing data stream and parses it to obtain the absolute timestamp. The CPU's internal high-level hardware timer resets the local counter register value based on the absolute timestamp, completing hardware clock-level reference alignment.
[0044] The data fusion node integrates a dedicated tensor processing physical unit to perform physical feature interpolation calculations. This tensor processing physical unit includes multiple multiply-accumulate physical acceleration engines. The central processing unit (CPU) sends the memory starting address and offset corresponding to the time series gaps to the control register of the tensor processing physical unit via the system bus. The tensor processing physical unit batch-loads physical decay gradients and physical diffusion gradients from synchronous dynamic random access memory (DRAM) to on-chip high-speed static random access memory (SRAM). The multiply-accumulate physical acceleration engines complete matrix-dot multiplication physical operations within specific nanosecond-level clock cycles, deriving supplementary state features and combining them to generate a state feature compensation sequence. Based on the positioning information and the three-dimensional wind field vector, the CPU generates a binary numerical stream corresponding to the thermodynamic directed mapping matrix within the arithmetic logic unit according to standard floating-point arithmetic rules. The CPU concatenates the state feature compensation sequence and the thermodynamic directed mapping matrix within a continuous address range in physical memory, outputting a comprehensive sensing data matrix.
[0045] The integrated sensing data matrix is input into the hardware acceleration board of the graph neural network model via the peripheral component interconnection physical bus. The hardware acceleration board performs parallel throughput calculations through multiple layers of hardware logic gates, outputting the state sensing sequence to main memory. The central processing unit (CPU) reads the state sensing sequence, extracts physical parameters, and calls the floating-point arithmetic unit to calculate the gradient rate of change of the physical parameters. The CPU compares the gradient rate of change with the fixed value in the preset threshold register. When the conditional jump instruction detects that the gradient rate of change is greater than the preset threshold, it triggers an external interrupt physical pin to generate a high-level warning signal. Based on the target spatial location calculated from the calculated physical thermal diffusion path, the CPU sends a response instruction, including a disaster prevention action sequence, to the peripheral control physical bus. The coordinated response node receives the response instruction. The digital logic controller inside the coordinated response node parses the response instruction. The digital logic controller compiles the disaster prevention action sequence in the response instruction into a physical suppression signal. The digital logic controller converts the physical suppression signal into a pulse-width modulated analog electrical signal with a specific duty cycle. A pulse-width modulated analog electrical signal is transmitted through the output pin to the servo motor driver of a high-pressure water cannon deployed between photovoltaic panels, driving the motor stator to physically rotate and spray cooling and extinguishing media. Simultaneously, the digital logic controller sends a general-purpose input / output pin level inversion control signal to the pipeline terminal. Upon receiving the signal, the pipeline terminal controls the physical relay to close, activating the physical power supply circuit of the solenoid valve drive coil, triggering the mechanical opening of the water pipeline valve to perform isolation and cooling actions.
Claims
1. A smart fire monitoring method based on multi-dimensional data, characterized in that, include: The system acquires positioning and timing information from a space network, collects three-dimensional wind field vectors and aerial environmental parameters from a mobile terminal, and collects fixed-point environmental parameters from a fixed terminal. It then aggregates the positioning information, timing information, three-dimensional wind field vectors, aerial environmental parameters, and fixed-point environmental parameters to generate a sensing data stream with an absolute timestamp. The absolute timestamp is extracted from the sensing data stream. The aerial environmental parameters and the fixed-point environmental parameters are aligned based on the absolute timestamp. Physical feature interpolation is performed on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence. A thermodynamic directed mapping matrix is constructed based on the positioning information and the three-dimensional wind field vector. The state feature compensation sequence and the thermodynamic directed mapping matrix are concatenated to output a comprehensive sensing data matrix. The comprehensive sensing data matrix is input into the graph neural network model, the state feature compensation sequence is mapped to network nodes and corresponding network node vectors, the thermodynamic directed mapping matrix is mapped to directed connection edges between the network nodes, the weight parameters of the directed connection edges are updated according to the three-dimensional wind field vector, node aggregation is performed to extract features and output the state sensing sequence. Extract physical parameters from the state-sensing sequence, calculate the gradient rate of change of the physical parameters, and output an early warning signal when the gradient rate of change is greater than a preset threshold; extract the core coordinates of thermal radiation from the state-sensing sequence, and calculate the physical thermal diffusion path by combining the three-dimensional wind field vector; Extract the target spatial location obtained from the physical thermal diffusion path calculation, obtain the disaster prevention action sequence matching the target spatial location, combine the target spatial location and the disaster prevention action sequence to generate a response command, and send the response command.
2. The intelligent fire monitoring method based on multi-dimensional data according to claim 1, characterized in that, The step of performing physical feature interpolation calculations on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence includes: The aligned aerial environmental parameters and the fixed-point environmental parameters are input into the thermodynamic physical model to calculate the physical decay gradient and physical diffusion gradient of environmental heat in the time dimension. Extract the time series gap between the aligned aerial environmental parameters and the fixed-point environmental parameters; Substitute the physical decay gradient and the physical diffusion gradient into the time series gap; The supplementary state characteristics within the time series gap are calculated based on the time step size of the time series gap. The aligned aerial environmental parameters, aligned fixed-point environmental parameters, and supplementary state features are merged and aligned, and the supplementary state features are output as a compensated sequence of the state features.
3. The intelligent fire monitoring method based on multi-dimensional data according to claim 2, characterized in that, The step of constructing a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector includes: The relative physical distance between the mobile terminal that collects the aerial environmental parameters and the fixed terminal that collects the fixed-point environmental parameters is extracted based on the positioning information. The relative physical space distance is multiplied by the three-dimensional wind field vector to obtain the wind field distance coupling coefficient. The normalized distribution calculation is performed on the wind field distance coupling coefficient to obtain the physical directed probability representing the heat transfer between the mobile terminal and the fixed terminal; Using the physical directed probabilities as matrix elements, a thermodynamic directed mapping matrix is constructed to indicate the direction of environmental thermal flow.
4. The intelligent fire monitoring method based on multi-dimensional data according to claim 3, characterized in that, The step of updating the weight parameters of the directed connection edges based on the three-dimensional wind field vector includes: Extract the wind speed and wind direction components from the three-dimensional wind field vector; The wind speed component is set as a weighted step size, and the wind direction component is set as a weighted direction. Extract the matrix elements from the thermodynamic directed mapping matrix; Based on the weighted step size and the weighted direction, the matrix elements are subjected to vector dot product calculation to obtain the updated feature values; The updated feature values are set as the weight parameters.
5. The intelligent fire monitoring method based on multi-dimensional data according to claim 4, characterized in that, The execution node aggregates and extracts features and outputs a state-aware sequence, including: According to the weight parameters, the network node vector is passed along the directed connection edge to the adjacent network node; The network node vectors passed to the adjacent network nodes are weighted and summed to obtain the node feature update variables; The network node vector and the node feature update variable are combined by feature concatenation to extract the physical features of thermal propagation with thermodynamic topological association. The physical characteristics of heat propagation are subjected to dimensionality reduction mapping calculation, and the dimensionality-reduced physical characteristics of heat propagation are arranged according to the time evolution order to output the state perception sequence.
6. The intelligent fire monitoring method based on multi-dimensional data according to claim 5, characterized in that, The step of extracting physical parameters from the state-aware sequence and calculating the gradient rate of change of the physical parameters includes: Temperature and smoke concentration feature values are extracted from the state-sensing sequence as the physical parameters. Extract the numerical differences of the physical parameters at different time points; Divide the numerical difference by the time interval between the different time points to obtain the slope of the physical parameter's change in physical quantity over time. The slope of the change in the physical quantity is set as the gradient rate of change.
7. The intelligent fire monitoring method based on multi-dimensional data according to claim 6, characterized in that, The step of extracting the core coordinates of thermal radiation from the state-sensing sequence and calculating the physical thermal diffusion path by combining them with the three-dimensional wind field vector includes: Extract the feature node with the largest thermal radiation value from the state-aware sequence and set it as the thermal radiation peak node; The thermal radiation peak nodes are mapped to a three-dimensional spatial coordinate system to obtain the coordinates of the thermal radiation core; Using the coordinates of the thermal radiation core as the spatial starting point, a time-step spatial coordinate offset accumulation calculation is performed along the direction indicated by the three-dimensional wind field vector to obtain a set of spatial offset coordinates. By fitting the set of spatial offset coordinates, the physical thermal diffusion path is obtained.
8. The intelligent fire monitoring method based on multi-dimensional data according to claim 7, characterized in that, The process of extracting the target spatial location calculated from the physical thermal diffusion path, obtaining a disaster prevention action sequence matching the target spatial location, and combining the target spatial location and the disaster prevention action sequence to generate a response command includes: Input the target spatial location into the emergency action database to query matching local cooling disaster prevention actions and spatial isolation disaster prevention actions; The local cooling disaster prevention actions and the spatial isolation disaster prevention actions obtained from the query are sorted in time according to the preset time execution priority, and the disaster prevention action sequence is obtained by combining them. The target spatial location and the disaster prevention action sequence are encapsulated into a standard instruction format, and the response instruction is output.
9. The intelligent fire monitoring method based on multi-dimensional data according to claim 8, characterized in that, Sending the response instruction includes: The generated response instructions are parsed to obtain the disaster prevention action sequence and the target spatial location; Extract the local cooling disaster prevention action from the disaster prevention action sequence, compile the local cooling disaster prevention action into a physical suppression signal for the coordinates of the thermal radiation core, and send it directionally to the disaster prevention execution terminal corresponding to the target spatial location; The spatial isolation disaster prevention action is extracted from the disaster prevention action sequence, and the spatial isolation disaster prevention action is compiled into an isolation and cooling signal for the front end of the physical heat diffusion path and sent to the pipeline terminal in a directional manner.
10. An intelligent fire monitoring system based on multi-dimensional data, applied to the intelligent fire monitoring method based on multi-dimensional data as described in any one of claims 1 to 9, characterized in that, include: The perception acquisition module is used to acquire positioning information and timing information from the space network, collect three-dimensional wind field vectors and air environment parameters from the mobile terminal, collect fixed-point environmental parameters from the fixed terminal, and aggregate the positioning information, the timing information, the three-dimensional wind field vectors, the air environment parameters and the fixed-point environmental parameters to generate a perception data stream with an absolute timestamp. The spatiotemporal reconstruction module is used to extract the absolute timestamps from the sensing data stream to perform benchmark alignment, perform physical feature interpolation calculations on the aligned aerial environmental parameters and the fixed-point environmental parameters to generate a state feature compensation sequence, construct a thermodynamic directed mapping matrix based on the positioning information and the three-dimensional wind field vector, and splice the state feature compensation sequence and the thermodynamic directed mapping matrix to output a comprehensive sensing data matrix. The topology fusion module is used to input the comprehensive sensing data matrix into the graph neural network model, map the state feature compensation sequence into network nodes and corresponding network node vectors, map the thermodynamic directed mapping matrix into directed connection edges between the network nodes, update the weight parameters of the directed connection edges according to the three-dimensional wind field vector, perform node aggregation to extract features and output the state sensing sequence. The intelligent judgment module is used to extract physical parameters from the state perception sequence and calculate the gradient change rate. When the gradient change rate is greater than a preset threshold, it outputs an early warning signal, extracts the core coordinates of thermal radiation from the state perception sequence, calculates the physical thermal diffusion path by combining the three-dimensional wind field vector, extracts the target spatial location calculated by the physical thermal diffusion path, obtains the disaster prevention action sequence matching the target spatial location and generates a response command, and sends the response command.