Fire-fighting early warning system and method based on artificial intelligence

Abstract

The invention discloses a fire-fighting early warning system and method based on artificial intelligence, and relates to the technical field of intelligent fire-fighting early warning, and the system comprises an environment sensing module which comprises a multispectral sensor, a temperature sensor, a smoke sensor and a high-definition camera, and is used for collecting fire characteristic spectral data and transmitting the data to a data analysis module; the data analysis module comprises an edge calculation unit and a data fusion unit, and is used for performing feature extraction and anomaly detection on the fire feature spectrum data through a deep learning model to generate a fire risk assessment result; the early warning decision module comprises a risk assessment unit, an early warning grading unit and a communication unit, and is used for determining an early warning grade according to a fire risk assessment result and triggering equipment of the execution feedback module; and the execution feedback module comprises a spraying system, an emergency lighting system and an alarm and is used for executing fire emergency measures according to the instruction of the early warning decision module. According to the invention, comprehensive improvement of fire prevention and control efficiency is realized through cooperative work of multiple modules.

Description

Technical Field

[0001] The present invention relates to the field of intelligent fire warning technology, and in particular to an artificial intelligence-based fire warning system and method. Background Art

[0002] Traditional fire protection systems primarily rely on single sensors for threshold determination. While these systems offer basic early warning capabilities, they are susceptible to environmental interference in complex scenarios, leading to false alarms or missed alerts. In recent years, the introduction of artificial intelligence (AI) technologies, particularly deep learning and multimodal data fusion methods, has significantly improved the accuracy and real-time nature of fire detection. However, existing technologies still suffer from issues such as insufficient data collaboration, limited edge computing capabilities, and low efficiency in linking early warning and execution, making them unable to meet the precise fire protection needs of highly dynamic environments such as modern buildings and industrial facilities.

[0003] CN116543538B proposes an IoT-based fire electrical early warning method. This method uses a trained fire warning network to detect anomalies in electrical monitoring data and optimizes the model's generalization capabilities using multi-type datasets to improve recognition accuracy. This technology's advantage lies in its use of AI to enhance the identification of electrical fires, making it suitable for scenarios such as power grids and distribution systems. However, its limitations include: it targets only electrical fires and does not cover multi-source fire characteristic spectral data, making it difficult to address non-electrical fires; it relies on centralized data processing, and its real-time performance is affected by network transmission delays; and it lacks an intelligent linkage mechanism with automated firefighting equipment, requiring manual intervention after an early warning.

[0004] CN110711332A* discloses an AI-based fire warning and extinguishing system. This system integrates multiple firefighting modules through a main console and utilizes network communications for remote command, optimizing evacuation efficiency. This system's advantage lies in its multi-module collaboration, which improves emergency response speed. However, its main issues include: reliance on a centralized control architecture, which can lead to system failure if the main console fails; underutilization of edge computing, resulting in high data processing latency; and environmental perception, which still primarily relies on traditional sensors and lacks new detection methods such as multispectral and visual analysis, making it difficult to identify early fire characteristics. Summary of the Invention

[0005] In view of the problems of existing fire warning systems, such as single environmental perception dimension, insufficient real-time data analysis, and low efficiency in linkage between warning and execution, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to achieve early and accurate identification of fire hazards, dynamic risk assessment and intelligent emergency linkage through multi-spectral sensor fusion, edge intelligent computing and adaptive early warning mechanism, so as to solve the problems of delayed response, high false alarm rate and low disposal efficiency of traditional fire protection systems.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides an artificial intelligence-based fire warning system, which includes an environment perception module, a data analysis module, a warning decision module, and an execution feedback module;

[0009] The environmental perception module includes a multispectral sensor, a temperature sensor, a smoke sensor and a high-definition camera, which is used to collect fire characteristic spectrum data in real time and transmit the fire characteristic spectrum data to the data analysis module;

[0010] The data analysis module includes an edge computing unit and a data fusion unit, which is used to perform feature extraction and anomaly detection on the fire characteristic spectrum data through a deep learning model to generate a fire risk assessment result;

[0011] The warning decision module includes a risk assessment unit, a warning classification unit and a communication unit, which are used to determine the warning level according to the fire risk assessment result and trigger the corresponding device of the execution feedback module;

[0012] The execution feedback module includes a sprinkler system, an emergency lighting system and an alarm, and is used to execute fire emergency measures according to the instructions of the early warning decision module.

[0013] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the multispectral sensor performs spectral detection within the near-infrared band and the mid-infrared band, and scans inside the building according to a preset detection area, and is equipped with a spectrum analyzer and a signal amplifier; the spectrum analyzer is used to collect flame characteristic spectrum information, smoke particle scattering spectrum information and combustible gas absorption spectrum information; the signal amplifier performs signal enhancement processing on the fire characteristic spectrum data, and uploads the enhanced fire characteristic spectrum data to the data analysis module for processing.

[0014] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, wherein: the edge computing unit adopts an improved convolutional neural network model for real-time data processing; the data fusion unit adopts a Bayesian fusion algorithm to fuse the fire characteristic spectral data; the Bayesian fusion algorithm includes prior probability calculation, likelihood function and posterior probability update; the credibility of each sensor data is determined by the prior probability calculation, the real-time monitoring data is converted into a fire probability distribution using the likelihood function, the fire risk is dynamically evaluated through the posterior probability update and the fused fire risk assessment result is output.

[0015] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the improved convolutional neural network model includes a multi-scale feature extraction layer, a spatiotemporal fusion layer and a weight adaptation layer; the multi-scale feature extraction layer performs multi-dimensional feature extraction on fire characteristic spectral data; the spatiotemporal fusion layer performs fusion analysis on feature data of different time series; and the weight adaptation layer dynamically adjusts the weight coefficient of each sensor data according to environmental changes.

[0016] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the risk assessment unit includes a fire risk assessment model; the fire risk assessment model is based on a fuzzy logic reasoning algorithm and calculates a risk level quantitative value by fusing fire characteristic spectral data; the warning classification unit executes a warning classification strategy based on the risk level quantitative value; and the communication unit uses a wireless ad hoc network communication protocol to realize information transmission between devices.

[0017] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the warning classification strategy performs classification judgment through a threshold comparison algorithm. When the quantitative value of the risk level is greater than the first threshold, a green warning is triggered, and the execution feedback module starts the environmental monitoring enhancement mode; when the quantitative value of the risk level is greater than the first threshold and less than the second threshold, a yellow warning is triggered, and the execution feedback module starts the pre-treatment measures; when the quantitative value of the risk level is greater than the second threshold and less than the third threshold, an orange warning is triggered, and the execution feedback module starts active intervention; when the quantitative value of the risk level is greater than the third threshold, a red warning is triggered, and the execution feedback module executes a full system emergency response.

[0018] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the wireless ad hoc network communication protocol includes route discovery, path maintenance and data transmission; the communication path between each fire equipment node is established through the route discovery, the communication link is monitored and dynamically adjusted in real time using the path maintenance, and the warning instructions and status information are reliably transmitted between devices through the data transmission.

[0019] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the sprinkler system executes a sprinkler strategy according to the warning level; and the emergency lighting system uses an intelligent path planning algorithm to generate an optimal evacuation path.

[0020] As a preferred solution of the artificial intelligence-based fire warning system described in the present invention, the intelligent path planning algorithm performs path search based on the A* algorithm; by acquiring real-time personnel distribution data and fire spread data inside the building, the A* algorithm is used to calculate the shortest safe path from the current position to the safe exit, and the optimal evacuation path is indicated by light guidance from emergency lighting equipment.

[0021] Compared with the existing technology, the beneficial effects of the present invention are as follows: the fire prevention and control efficiency is comprehensively improved through the collaborative work of multiple modules; the environmental perception module adopts multi-spectral sensing technology, breaking through the detection limitations of traditional single sensors, and can accurately identify multiple characteristic signals of early fires; the data analysis module realizes high-precision dynamic assessment of fire risks through improved convolutional neural networks and Bayesian fusion algorithms; the early warning decision module establishes a multi-level early warning mechanism based on fuzzy logic reasoning, which significantly improves the emergency response speed; the intelligent sprinkler system and dynamic path planning technology of the execution feedback module optimize the fire extinguishing efficiency and evacuation guidance effect; the entire system adopts edge computing architecture and adaptive communication mechanism, which enhances system reliability while ensuring real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0023] Figure 1 This is the overall framework diagram of the fire warning system based on artificial intelligence.

[0024] Figure 2 The figure is a flow chart of the fire warning method based on artificial intelligence. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0028] As mentioned in the background technology above, the accuracy of fire detection is limited by the perception dimension of a single sensor. For example, traditional smoke sensors have difficulty identifying early smoldering fires, while temperature sensors have a delayed response to rapidly spreading electrical fires. Judgment methods based on fixed thresholds lack the fusion analysis of multi-source environmental data, resulting in high false alarm and missed alarm rates. The data processing architecture of existing systems has limitations. For example, although CN116543538B uses artificial intelligence technology, it still relies on centralized cloud computing, resulting in insufficient real-time fire risk assessment; and although CN110711332A* realizes multi-module linkage, it does not fully utilize edge computing capabilities, making it difficult to respond to sudden fires in a timely manner. In response to the above problems, an artificial intelligence-based fire warning system and method are proposed.

[0029] Example 1, Figure 1 FIG is a system framework diagram of an artificial intelligence-based fire warning system according to an embodiment of the present invention. Figure 1 As shown, in an artificial intelligence-based fire warning system, it includes: an environmental perception module, a data analysis module, a warning decision module and an execution feedback module.

[0030] Preferably, the environmental perception module includes a multispectral sensor, a temperature sensor, a smoke sensor and a high-definition camera, which are used to collect fire characteristic spectral data in real time and transmit the fire characteristic spectral data to the data analysis module.

[0031] It should be noted that the fire characteristic spectrum data includes spectrum data, temperature data and smoke concentration data.

[0032] Specifically, the multispectral sensor performs spectral detection within the near-infrared and mid-infrared bands, and scans inside the building according to the preset detection area. It is equipped with a spectrum analyzer and a signal amplifier; the spectrum analyzer is used to collect flame characteristic spectrum information, smoke particle scattering spectrum information and combustible gas absorption spectrum information; the signal amplifier performs signal enhancement processing on the fire characteristic spectrum data, and uploads the enhanced fire characteristic spectrum data to the data analysis module for processing.

[0033] Furthermore, the temperature sensor is customized according to the temperature range of the detection environment, including a thermocouple, a signal conditioning circuit and a wireless transmission module; when the environmental perception module performs the temperature monitoring task, the edge computing unit of the data analysis module sends a temperature acquisition instruction to the temperature sensor; after receiving the instruction, the temperature sensor starts the thermocouple to perform temperature detection, converts the temperature change into an electrical signal, and amplifies and filters the electrical signal through the signal conditioning circuit, and uploads the processed temperature data to the data analysis module through the wireless transmission module; if the temperature sensor detects an abnormal temperature rise, the edge computing unit sends a continuous monitoring instruction to the temperature sensor, and the temperature sensor receives the instruction and increases the sampling frequency, so that the thermocouple continuously detects temperature changes, and at the same time the signal conditioning circuit processes and transmits the temperature data to the data analysis module in real time.

[0034] Furthermore, when the environmental perception module performs the temperature monitoring task and cannot accurately identify the ambient temperature, the wireless transmission module transmits the temperature change data in real time and transmits the temperature data of several groups of monitoring areas to the data analysis module through the temperature sensor;

[0035] It should be noted that the edge computing unit will transmit temperature data back to the early warning decision module; the early warning decision module assesses the fire risk based on the temperature change trend and abnormal conditions. If there is a risk, a warning instruction will be sent to the edge computing unit, and the edge computing unit will trigger the emergency response instruction of the execution feedback module; if there is no risk, the edge computing unit will send a normal monitoring instruction to the temperature sensor to maintain routine detection.

[0036] Preferably, the data analysis module includes an edge computing unit and a data fusion unit, which are used to perform feature extraction and anomaly detection on fire characteristic spectral data through a deep learning model to generate fire risk assessment results.

[0037] Specifically, the edge computing unit uses an improved convolutional neural network model for real-time data processing; the data fusion unit uses a Bayesian fusion algorithm to fuse the fire characteristic spectral data; the Bayesian fusion algorithm includes prior probability calculation, likelihood function and posterior probability update; the credibility of each sensor data is determined by prior probability calculation, the real-time monitoring data is converted into a fire probability distribution using the likelihood function, the fire risk is dynamically evaluated through posterior probability update and the fused fire risk assessment result is output.

[0038] Furthermore, the improved convolutional neural network model includes a multi-scale feature extraction layer, a spatiotemporal fusion layer and a weight adaptation layer; the multi-scale feature extraction layer performs multi-dimensional feature extraction on the fire characteristic spectral data; the spatiotemporal fusion layer fuses and analyzes the feature data of different time series; and the weight adaptation layer dynamically adjusts the weight coefficient of each sensor data according to environmental changes.

[0039] Furthermore, the data fusion unit loads the historical calibration data of each sensor and assigns initial prior probability values ​​to the multispectral sensor, temperature sensor, and smoke sensor respectively.

[0040] It should be noted that the historical calibration data includes the spectral detection accuracy of the multispectral sensor, the temperature measurement error range of the temperature sensor, and the concentration calibration curve of the smoke sensor; the initial prior probability value reflects the basic credibility of each sensor in the absence of real-time data.

[0041] Specifically, constructing a likelihood function includes: establishing likelihood functions corresponding to multispectral sensors, temperature sensors, and smoke sensors based on the physical characteristics of fire characteristic spectral data: the likelihood function of the multispectral sensor is constructed based on the matching degree between the flame characteristic spectral information and the standard flame spectral library; the likelihood function of the temperature sensor adopts a Gaussian distribution model, whose mean is determined by the temperature value detected by the thermocouple, and the variance is determined by the noise level of the signal conditioning circuit; the likelihood function of the smoke sensor is generated based on the smoke particle scattering spectral information and a preset concentration-scattering intensity relationship curve.

[0042] Furthermore, the edge computing unit receives the real-time fire characteristic spectral data uploaded by the environmental perception module, and normalizes the spectral data, temperature data and smoke concentration data respectively to eliminate dimensional differences; according to the signal-to-noise ratio of the real-time fire characteristic spectral data and the degree of deviation of historical data, the prior probability of each sensor is dynamically adjusted: if the change in spectral intensity detected by the multi-spectral sensor exceeds a preset multiple of its noise standard deviation, its prior probability is reduced; if the difference between the temperature value of the temperature sensor monitoring point and its temperature reference value continues to increase, its prior probability weight is increased; if the slope of the cumulative distribution function of the smoke concentration data increases sharply, the prior probability of the smoke sensor is nonlinearly corrected.

[0043] Furthermore, the pre-processed real-time fire characteristic spectral data is input into the likelihood function, and combined with the updated prior probability to calculate the posterior probability of each sensor: the spectral intensity change of the multi-spectral sensor, the temperature value of the temperature sensor and the cumulative distribution function of the smoke concentration data are converted into a joint fire probability distribution through the Bayesian fusion algorithm; the Markov chain Monte Carlo method is used to iteratively optimize the posterior probability to ensure probability convergence; based on the posterior probability results, the data fusion unit generates a fused fire risk assessment result: if the posterior probabilities of the multi-spectral sensor and the temperature sensor simultaneously exceed the dynamic threshold, it is determined to be an open fire risk; if only the posterior probability of the smoke sensor continues to rise and the temperature data does not exceed the limit, it is determined to be a smoldering risk; the risk assessment result is transmitted to the risk assessment unit of the early warning decision module.

[0044] Specifically, the edge computing unit reversely calibrates the prior probability of each sensor based on the final judgment result of the early warning decision module: if a fire is confirmed to have occurred and the posterior probability of a certain sensor does not match the actual situation, the weight of the historical calibration data of the sensor is lowered; if the early warning result is a false alarm, the parameters of the likelihood function are recalibrated.

[0045] Preferably, the early warning decision module includes a risk assessment unit, an early warning classification unit and a communication unit, which are used to determine the early warning level according to the fire risk assessment result and trigger the corresponding device of the execution feedback module.

[0046] Specifically, the risk assessment unit includes a fire risk assessment model; the fire risk assessment model is based on a fuzzy logic reasoning algorithm and calculates a quantitative value of the risk level by fusing fire characteristic spectral data; the fuzzy logic reasoning algorithm includes fuzzification processing, rule base matching and defuzzification output; the fire risk assessment result is converted into a fuzzy set through fuzzification processing, the fuzzy set is inferred and judged using rule base matching, and the quantitative value of the risk level is obtained through defuzzification output.

[0047] It should be noted that the specific calculation formula for the quantitative value of the risk level is as follows:

[0048]

[0049] Among them, R is the quantitative value of risk level, T is the time window length, N is the number of multispectral sensors, α i (t) is the time-varying attenuation coefficient of the i-th sensor, ΔS i (t) is the spectral intensity change of the i-th sensor, σ i is the noise standard deviation of the i-th sensor, M is the number of temperature monitoring points, β j is the weight coefficient of the jth temperature monitoring point, L j (t) is the temperature value of the jth monitoring point, μ j is the temperature reference value of the jth monitoring point, σ j is the temperature variance of the jth monitoring point, Φ(t) is the cumulative distribution function of smoke concentration, Q is the number of fuzzy rules, λ p is the steepness parameter of the pth fuzzy rule, θ p is the center point of the pth fuzzy rule, ξ p is the shape parameter of the pth fuzzy rule, η p is the scale parameter of the p-th fuzzy rule, ζp is the normalized exponent of the p-th fuzzy rule, and Γ(*) is the gamma function.

[0050] Furthermore, the value range of this formula is [0,1], 0≤R<0.25: indicates normal state, corresponding to the green warning level, and the system is in regular monitoring mode; 0.25≤R<0.5: indicates mild risk, corresponding to the yellow warning level, and pre-treatment measures need to be initiated; 0.5≤R<0.75: indicates moderate risk, corresponding to the orange warning level, and active intervention is required; 0.75≤R≤1: indicates high risk, corresponding to the red warning level, and a full system emergency response needs to be implemented; this formula processes time series data through integral operations, uses product functions to fuse multi-sensor information, uses hyperbolic tangent functions to process temperature anomalies, combines limit functions to describe the smoke diffusion process, and realizes mathematical modeling of fuzzy logic reasoning through fractional structures and gamma functions, which can accurately quantify the fire risk level.

[0051] Furthermore, the warning classification unit executes the warning classification strategy based on the quantitative value of the risk level; the warning classification strategy performs classification judgment through a threshold comparison algorithm. When the quantitative value of the risk level is greater than the first threshold, a green warning is triggered, and the execution feedback module starts the environmental monitoring enhancement mode; when the quantitative value of the risk level is greater than the first threshold and less than the second threshold, a yellow warning is triggered, and the execution feedback module starts the pre-disposal measures; when the quantitative value of the risk level is greater than the second threshold and less than the third threshold, an orange warning is triggered, and the execution feedback module starts active intervention; when the quantitative value of the risk level is greater than the third threshold, a red warning is triggered, and the execution feedback module executes a full system emergency response.

[0052] It should be noted that the first threshold is determined based on the statistical characteristics of historical environmental noise data. Specifically, it is determined by analyzing the fluctuation range of sensor data under normal working conditions for at least 6 months, and taking the upper limit of the 99% confidence interval of temperature, smoke concentration and spectral characteristic data; the second threshold is determined based on the calibration of initial fire experimental data. By simulating 12 typical initial fire conditions such as smoldering and electrical short circuit in a controlled environment, the median of the characteristic values ​​of each sensor in the budding stage of fire is extracted; the third threshold is determined based on the critical state parameters of the fire. The critical state parameters are obtained through the following steps: input the building structural parameters (space volume and ventilation area) and material combustion characteristics data; calculate the minimum heat release rate threshold required for the flame to spread to adjacent areas; convert the heat release rate threshold into a fusion monitoring value of the multi-spectral sensor and the temperature sensor as the benchmark value of the third threshold.

[0053] Specifically, the environmental monitoring enhancement mode includes: increasing the sensor sampling frequency to twice that of the normal mode, activating high-definition cameras for video review, and sending primary alarm signals to the fire control center through the communication unit; pre-treatment measures include: closing the fire dampers of the ventilation system, unlocking the electromagnetic access control of the emergency passage, prompting potential risks through voice broadcasts, and linking the sprinkler system to enter the pre-pressurization state; active intervention includes: activating the sprinkler system in the local area, starting the path guidance function of the emergency lighting system, triggering the sound and light alarm, and sending real-time fire data containing positioning information to the fire department; executing a full system emergency response, including: the sprinkler system starts to release fire extinguishing agents, forcibly opening all evacuation passages, and dynamically adjusting the evacuation route through the intelligent path planning algorithm to avoid the fire spread area, and pushing escape navigation information to the personnel terminal.

[0054] Furthermore, the communication unit adopts a wireless ad hoc network communication protocol to realize information transmission between devices; the wireless ad hoc network communication protocol includes route discovery, path maintenance and data transmission; through route discovery, a communication path is established between each fire-fighting equipment node, and path maintenance is used to monitor and dynamically adjust the communication link in real time, and through data transmission, warning instructions and status information are reliably transmitted between devices.

[0055] Preferably, the execution feedback module includes a sprinkler system, an emergency lighting system and an alarm, which is used to execute fire emergency measures according to the instructions of the early warning decision module.

[0056] Specifically, the sprinkler system executes a sprinkler strategy according to the warning level.

[0057] It should be noted that the spraying strategy includes local spraying, regional spraying and full spraying; when a yellow warning is received, the local spraying mode is activated, and the single area where the abnormality is detected is sprayed; when an orange warning is received, the regional spraying mode is activated, and the abnormal area and its adjacent areas are sprayed; when a red warning is received, the full spraying mode is activated, and the entire building is sprayed.

[0058] Furthermore, the emergency lighting system uses an intelligent path planning algorithm to generate the optimal evacuation path; the intelligent path planning algorithm performs path search based on the A* algorithm; by obtaining real-time personnel distribution data and fire spread data inside the building, the A* algorithm is used to calculate the shortest safe path from the current location to the safe exit, and the optimal evacuation path is indicated by the light guidance of the emergency lighting equipment.

[0059] Furthermore, the emergency lighting system obtains real-time personnel distribution data through infrared thermal imagers and wireless positioning beacons deployed in the building; at the same time, it receives fire spread data transmitted by the early warning decision module; discretizes the building's three-dimensional structural model into a two-dimensional grid map, with each grid containing access status attributes; based on the fire spread data, the flame-covered grid is marked as an inaccessible area, the smoke concentration exceeding the limit grid is marked as a high-risk area, and the temperature abnormality grid is marked as a dynamic obstacle.

[0060] It should be noted that the fire spread data includes the coordinates of the flame center, the smoke diffusion range, and the boundary coordinates of the temperature abnormality area.

[0061] Specifically, the wireless signal strength fingerprint in the personnel distribution data is aligned with the grid map coordinate system to determine the current position grid of each person; the position detection jitter error of the infrared thermal imager is eliminated through the Kalman filter algorithm.

[0062] Furthermore, a multi-factor weighted cost function is set for the A** algorithm, including: basic path length: the Manhattan distance from the current grid to the target exit; fire threat coefficient: the weighted sum of the temperature gradient and smoke concentration gradient of each grid calculated based on the fire spread data; congestion level: the population density within a preset range around each grid is statistically calculated based on the population distribution data.

[0063] Furthermore, the forward search starts from the personnel's current location grid and expands the nodes along the length of the basic path; the reverse search starts from the safety exit grid and gives priority to paths with low fire threat factors; when the open sets of the bidirectional search intersect, the paths are merged and the continuity is verified.

[0064] Specifically, after the path is generated, changes in fire spread data are continuously monitored: if the newly detected flame covers the key grids in the path, local replanning is triggered to retain the unaffected path segments; if the personnel distribution data shows that the target path is congested, the congestion degree weight is increased and the alternative path is recalculated.

[0065] Furthermore, the optimal evacuation path is converted into a control instruction sequence for emergency lighting equipment: for straight channel sections, the high-brightness LED light strips of the emergency lighting equipment are activated to form a continuous light flow to guide the direction; for turning nodes, the flashing frequency difference is used to indicate the turning direction; and red warning light sources are added in high-risk areas to enhance risk warnings.

[0066] Furthermore, the emergency lighting system works in conjunction with the voice broadcasting system of the execution feedback module: when a stationary person is detected in the personnel distribution data, the lighting equipment is controlled to project a pulsed light spot at the corresponding position; the voice broadcasting system is synchronously triggered to play a directional voice prompt, which includes a description of the direction of the optimal evacuation path.

[0067] In summary, the present invention achieves a comprehensive improvement in fire prevention and control efficiency through the collaborative work of multiple modules; the environmental perception module adopts multi-spectral sensing technology, breaking through the detection limitations of traditional single sensors, and can accurately identify multiple characteristic signals of early fires; the data analysis module uses an improved convolutional neural network and Bayesian fusion algorithm to achieve high-precision dynamic assessment of fire risks; the early warning decision-making module establishes a multi-level early warning mechanism based on fuzzy logic reasoning, which significantly improves the emergency response speed; the intelligent sprinkler system and dynamic path planning technology of the execution feedback module optimize the fire extinguishing efficiency and evacuation guidance effect; the entire system adopts edge computing architecture and adaptive communication mechanism, which enhances system reliability while ensuring real-time performance.

[0068] Example 2

[0069] Reference Figure 2 , which is an embodiment of the present invention, this embodiment also provides an artificial intelligence-based fire warning method, including: processing fire characteristic spectral data through a signal amplifier to form a multi-dimensional sensor data set; inputting the multi-dimensional sensor data set into an improved convolutional neural network model for feature extraction to obtain a feature vector, and using a Bayesian fusion algorithm to fuse the feature vector to generate a fire risk assessment result; using a fuzzy logic reasoning algorithm to quantify the risk level of the fire risk assessment result, executing a warning classification strategy, and determining a corresponding warning level based on the quantified risk level value; based on the warning level, sending a control instruction to an execution feedback module through a wireless ad hoc network communication protocol to activate a sprinkler strategy and an intelligent path planning algorithm, execute fire emergency measures of the corresponding level, and generate an evacuation guidance plan.

[0070] Specifically, a fire early warning system was deployed in a commercial complex (12,000 square meters, 4.5-meter-high, and encompassing dining, retail, and office areas). The environmental perception module includes: a multispectral sensor (model MS-2100), with detection wavelengths ranging from 850-1700 nm in the near-infrared (NIR) and 3-5 μm in the mid-infrared (MIR), distributed across a 10 m x 10 m grid with 36 monitoring points; a PT100 thermocouple temperature sensor with a range of 0-300°C and an accuracy of ±0.5°C, co-located with the multispectral sensor; and a laser scattering smoke sensor with a range of 0-20% obs / m and a resolution of 0.1% obs / m. Fire simulation scenarios involved setting up a standard fire source (heptane combustion pool with a heat release rate of 500 kW) in the dining area kitchen and a smoldering fire source (overheated electrical short circuit with a temperature gradient of 2°C / s) in the retail area.

[0071] Furthermore, the multispectral sensor collects flame characteristic spectra (CO2 absorption peak 4.26μm, H2O absorption peak 2.7μm) and smoke scattering spectra (Mie scattering intensity peak 1.5μm). After processing with a signal amplifier (gain 40dB, signal-to-noise ratio improvement 15dB), a multidimensional sensor dataset containing spectral intensity, temperature gradient and smoke concentration is formed; the dataset is input into a convolutional neural network structured with 4 layers of multi-scale feature extraction layers and 2 layers of spatiotemporal fusion layers, which outputs a 128×1 feature vector.

[0072] Furthermore, a Bayesian fusion algorithm (prior probabilities: 0.7 for the multispectral sensor, 0.6 for the temperature sensor, and 0.5 for the smoke sensor) was used to calculate the posterior probability distribution. In the case of a smoldering fire, the algorithm dynamically increased the weight of the smoke sensor from 0.5 to 0.8, significantly reducing the false negative rate. The risk assessment results were input into a fuzzy logic inference model (with a rule base consisting of 32 if-then rules) to output a quantitative risk level value, R. In a heptane fire experiment, the R value jumped from 0.15 to 0.82 within 30 seconds, triggering a red alert. A command was sent to the sprinkler system via a wireless ad hoc network communication protocol (ZigBee + LoRa dual-mode, packet loss rate <0.1%). An intelligent path planning algorithm also generated an evacuation route.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based fire warning system, characterized by: include, Environmental perception module, data analysis module, early warning decision module, and execution feedback module; The environmental perception module includes a multispectral sensor, a temperature sensor, a smoke sensor and a high-definition camera, which is used to collect fire characteristic spectrum data in real time and transmit the fire characteristic spectrum data to the data analysis module; The data analysis module includes an edge computing unit and a data fusion unit, which is used to perform feature extraction and anomaly detection on the fire characteristic spectrum data through a deep learning model to generate a fire risk assessment result; The warning decision module includes a risk assessment unit, a warning classification unit and a communication unit, which are used to determine the warning level according to the fire risk assessment result and trigger the corresponding device of the execution feedback module; The execution feedback module includes a sprinkler system, an emergency lighting system and an alarm, and is used to execute fire emergency measures according to the instructions of the early warning decision module.

2. The artificial intelligence-based fire warning system according to claim 1, characterized in that: The multispectral sensor performs spectral detection within the near-infrared and mid-infrared bands and scans within a building according to a preset detection area. It is equipped with a spectrum analyzer and a signal amplifier. The spectrum analyzer is used to collect flame characteristic spectrum information, smoke particle scattering spectrum information, and combustible gas absorption spectrum information. The signal amplifier performs signal enhancement processing on the fire characteristic spectrum data and uploads the enhanced fire characteristic spectrum data to the data analysis module for processing.

3. The artificial intelligence-based fire warning system according to claim 1, characterized in that: The edge computing unit uses an improved convolutional neural network model for real-time data processing; the data fusion unit uses a Bayesian fusion algorithm to fuse the fire characteristic spectral data; the Bayesian fusion algorithm includes prior probability calculation, likelihood function and posterior probability update; the credibility of each sensor data is determined by the prior probability calculation, the real-time monitoring data is converted into a fire probability distribution using the likelihood function, the fire risk is dynamically evaluated through the posterior probability update and the fused fire risk assessment result is output.

4. The artificial intelligence-based fire warning system according to claim 3, characterized in that: The improved convolutional neural network model includes a multi-scale feature extraction layer, a spatiotemporal fusion layer, and a weight adaptation layer; the multi-scale feature extraction layer performs multi-dimensional feature extraction on fire characteristic spectral data; the spatiotemporal fusion layer performs fusion analysis on feature data from different time series; and the weight adaptation layer dynamically adjusts the weight coefficient of each sensor data according to environmental changes.

5. The artificial intelligence-based fire warning system according to claim 1, characterized in that: The risk assessment unit includes a fire risk assessment model; the fire risk assessment model is based on a fuzzy logic inference algorithm and calculates a risk level quantitative value by fusing fire characteristic spectral data; the early warning classification unit executes an early warning classification strategy based on the risk level quantitative value; the communication unit uses a wireless ad hoc network communication protocol to realize information transmission between devices.

6. The artificial intelligence-based fire warning system according to claim 5, characterized in that: The warning classification strategy performs classification judgment through a threshold comparison algorithm. When the quantitative value of the risk level is greater than the first threshold, a green warning is triggered, and the execution feedback module starts the environmental monitoring enhancement mode; when the quantitative value of the risk level is greater than the first threshold and less than the second threshold, a yellow warning is triggered, and the execution feedback module starts the pre-treatment measures; when the quantitative value of the risk level is greater than the second threshold and less than the third threshold, an orange warning is triggered, and the execution feedback module starts active intervention; when the quantitative value of the risk level is greater than the third threshold, a red warning is triggered, and the execution feedback module executes a full system emergency response.

7. The artificial intelligence-based fire warning system according to claim 5, characterized in that: The wireless ad hoc network communication protocol includes route discovery, path maintenance and data transmission; the communication path between each fire equipment node is established through the route discovery, the communication link is monitored and dynamically adjusted in real time using the path maintenance, and the warning instructions and status information are reliably transmitted between devices through the data transmission.

8. The artificial intelligence-based fire warning system according to claim 1, characterized in that: The sprinkler system executes a sprinkler strategy according to the warning level; the emergency lighting system uses an intelligent path planning algorithm to generate an optimal evacuation path.

9. The artificial intelligence-based fire warning system according to claim 8, characterized in that: The intelligent path planning algorithm performs path search based on the A* algorithm; by acquiring real-time data on the distribution of people inside the building and fire spread data, the A* algorithm is used to calculate the shortest safe path from the current location to the safe exit, and the light guidance of emergency lighting equipment indicates the optimal evacuation path.

10. An artificial intelligence-based fire warning method, based on the artificial intelligence-based fire warning system according to any one of claims 1 to 9, characterized in that: include, The fire characteristic spectrum data is processed through a signal amplifier to form a multi-dimensional sensor data set; Inputting the multi-dimensional sensor data set into the improved convolutional neural network model for feature extraction to obtain feature vectors, and fusing the feature vectors using a Bayesian fusion algorithm to generate a fire risk assessment result; quantifying the fire risk assessment results into risk levels using a fuzzy logic inference algorithm, executing an early warning grading strategy, and determining a corresponding early warning level based on the quantified risk level values; Based on the warning level, control instructions are sent to the execution feedback module through the wireless ad hoc network communication protocol to start the sprinkler strategy and intelligent path planning algorithm, execute the corresponding level of fire emergency measures and generate an evacuation guidance plan.

Images