Smoke detection early warning device and control system thereof

Abstract

The invention relates to the technical field of smoke detection early warning devices, in particular to a smoke detection early warning device and a control system thereof.The smoke detection early warning device is characterized in that an optical sensor, an ionic sensor and a gas sensor are installed in an early warning device body, and a wind speed sensor, a green LED, a yellow LED and a red LED are installed outside the early warning device body; the control system specifically comprises a multi-source fusion sensing module, an intelligent signal processing module, a smoke type identification module, a smoke concentration dynamic estimation and environment correction module, a multi-level dynamic early warning decision module, a multi-mode communication and data transmission module, a self-adaptive power supply management module and a multi-dimensional fault diagnosis and self-repairing module. In the control system, the limitation of a single sensor is broken through through optical, ionic and gas three-source cooperation of the multi-source fusion sensing module, and the multi-dimensional characteristics of smoke are comprehensively captured.

Description

Technical Field

[0001] This invention relates to the field of smoke detection and warning technology, and in particular to a smoke detection and warning device and its control system. Background Technology

[0002] With the increasing complexity of building spaces, the need for fire safety protection in homes, industrial plants, shopping malls, and other settings is becoming increasingly urgent. The initial stages of a fire are often accompanied by the generation of large amounts of smoke, which spreads much faster than flames. Timely detection and early warning can buy crucial time for evacuation and initial firefighting, significantly reducing casualties and property damage. Therefore, smoke detection and early warning devices have become core equipment for ensuring safety in various scenarios.

[0003] However, existing smoke detection and warning devices have significant technical shortcomings, making it difficult to meet the high requirements of complex scenarios. At the perception level, most products rely on a single sensor, which can only capture one feature of the smoke. They have weak identification capabilities against interference sources such as cooking fumes and industrial dust, and are prone to false alarms. The signal processing stage uses simple threshold comparisons, without considering the dynamic changes of smoke and the influence of environmental factors, resulting in high response delays and missing the best warning opportunity. Summary of the Invention

[0004] To address the significant technical shortcomings of existing smoke detectors and their inability to meet the high requirements of complex scenarios, this invention provides a smoke detector and its control system.

[0005] The technical solution adopted in this invention is: a smoke detection and warning device, including a warning device body, an optical sensor, an ionization sensor, and a gas sensor installed in the warning device body, a wind speed sensor, a green LED, a yellow LED, and a red LED installed on the outside of the warning device body, a microcontroller and a power supply assembly fixedly installed inside the warning device body, and a buzzer fixedly installed on the outside of the warning device body.

[0006] In one embodiment, the control system specifically includes a multi-source fusion sensing module, an intelligent signal processing module, a smoke type identification module, a smoke concentration dynamic estimation and environmental correction module, a multi-level dynamic early warning decision module, a multi-mode communication and data transmission module, an adaptive power management module, and a multi-dimensional fault diagnosis and self-repair module. in: The multi-source fusion sensing module is used to sense the optical characteristics, ionization characteristics and gas composition characteristics of smoke; The intelligent signal processing module is used to extract the dynamic features of the smoke; The smoke type identification module constructs a multi-dimensional identification model based on "spectral features, temporal features, and gas features" to distinguish between fire smoke and non-fire smoke. The smoke concentration dynamic estimation and environmental correction module is used to construct a dynamic concentration estimation model and an environmental correction formula. The multi-level dynamic early warning decision module is used to construct a multi-level early warning model and decision-making mechanism; The multi-mode communication and data transmission module dynamically selects the communication mode based on the warning level and the strength of the environmental signal; The adaptive power management module constructs an adaptive power supply strategy based on the working status of each module and the remaining battery power. The multi-dimensional fault diagnosis and self-repair module constructs a multi-dimensional fault diagnosis model, identifies fault types, and triggers a self-repair mechanism.

[0007] In one embodiment, the multi-source fusion sensing module is based on a three-source collaborative sensing architecture of optical sensors, ionization sensors, and gas sensors. It complementaryly senses the optical characteristics, ionization characteristics, and gas composition characteristics of smoke, providing multi-dimensional raw data for subsequent signal processing, as detailed below: The scattering optical sensor, based on the Lambert-Beer law, shows that the intensity of light scattered by smoke particles at a specific wavelength is positively correlated with the smoke concentration; Its sensing signal output formula is as follows: ; in: The intensity of the scattered light output by the optical sensor; The intensity of light emitted by the light source; This refers to the sensitivity coefficient of the optical sensor. Smoke concentration; The light absorption coefficient of smoke particles; The distance between the light source and the receiver; Inherent noise of optical sensors Ionization sensors utilize the blocking effect of smoke particles on alpha particles in the ionization chamber to change the ionization current. The formula for its output current is as follows: ; in: The output current for the ion sensor; The reference current for the ionization chamber in the absence of smoke; The response coefficient of the ion sensor; denoted as smoke concentration; 0.8 represents the nonlinear exponent of the smoke particle blocking effect. The gas sensor is a CO sensor. The formula for the relationship between the output resistance of the CO sensor and the CO concentration is as follows: ; in: The output resistance of the gas sensor; This is the reference resistance of the gas sensor when there is no CO. This refers to the sensitivity coefficient of the gas sensor. The concentration of CO in the environment; Multi-source data synchronous control and acquisition uses an STM32H743 microcontroller as the main controller, and a 16-bit ADC module to synchronously acquire the output signals of the optical sensor, ion sensor, and gas sensor. The sampling frequency is set to 1kHz, and the acquisition period is... .

[0008] In one embodiment, the intelligent signal processing module sequentially performs noise suppression, feature extraction, and data fusion processing on the raw signal output by the multi-source fusion sensing module to eliminate the influence of environmental noise and extract the dynamic features of smoke. Specifically as follows: Wavelet transform noise suppression: The original signal is decomposed into three levels of wavelet using the db4 wavelet basis function, and noise components are eliminated by thresholding to reconstruct a clean signal; The wavelet decomposition formula is as follows: ; ; in: These are the low-frequency approximation coefficients for the j-th layer; Here are the high-frequency detail coefficients for the j-th layer; These are the low-pass filter coefficients; These are the high-pass filter coefficients; The number of decomposition layers; For coefficient index; Index of signal sampling points; These are the signal coefficients of the (j-1)th layer; Noise thresholding: An adaptive threshold formula is used. ; Where: λ is the noise threshold; σ is the noise standard deviation of the high-frequency detail components; N is the number of signal sampling points; Smoke dynamic feature extraction: Core dynamic features are extracted from the denoised multi-source signal to reflect the changing trend and stability of smoke concentration, as detailed below: Concentration change rate: ; in: The rate of change in smoke concentration; Let be the smoke concentration at time t; Let be the smoke concentration at time t-Δt; For time intervals; Signal fluctuation variance: ; in: The variance of the output signal of the optical sensor; This represents the number of sampling points within the statistics window; This is the sensor output for the m-th sampling point; This represents the average value of the sampled points within the window. CO concentration gradient: ; in: For CO concentration gradient; Let be the CO concentration at time t; The initial CO concentration; For detection time; Multi-source data fusion: A fuzzy neural network is used to fuse feature data from optical, ionization, and gas sensors to output the fused smoke concentration. With confidence level ; The input layer of the FNN consists of feature values ​​from three sensors (optical sensor, ion sensor, and gas sensor). , , ,in , , The hidden layer uses a Gaussian membership function to represent the denoised signal value. ; in: For the i-th input feature The membership degree of the j-th fuzzy set; Let i be the i-th input feature value, i=1,2,3, which corresponds to... , , ; The center value of the j-th fuzzy set of the i-th input feature; The width of the j-th fuzzy set of the i-th input feature; The output layer of FNN uses a weighted summation formula: ; ; in: Let be the weight of the k-th fuzzy rule; The number of fuzzy rules; Confidence coefficient; This represents the smoke concentration threshold.

[0009] In one embodiment, the smoke type recognition module is based on the output fusion features. A multi-dimensional identification model based on "spectral features, temporal features, and gas features" is constructed to distinguish between fire smoke and non-fire smoke, as detailed below: Spectral feature extraction: By using the multi-wavelength scattering signals from optical sensors, the spectral scattering ratio of smoke is calculated, reflecting the refractive index and particle size distribution of smoke particles; Spectral scattering ratio formula: ; in: Spectral scattering ratio; wavelength The intensity of the scattered light; wavelength The intensity of the scattered light; Temporal feature analysis: based on extracted and Construct time-domain feature vectors The temporal features are trained and recognized using a support vector machine classifier; The kernel function of SVM uses radial basis functions: ; in: The kernel function value; This represents the temporal feature vector of the sample to be identified. The time-domain feature vector of the training samples; These are kernel function parameters; The square of the Euclidean distance between the two vectors; SVM decision function: ; in: The Lagrange multipliers for support vectors; Labels for the training samples; The number of support vectors; Let be the temporal feature vector of the i-th support vector; For SVM bias terms; It is a symbolic function; Gas Feature Verification: Based on gas sensor data, CO concentration determination criteria are constructed as a method for smoke type identification. then else ; in: The CO concentration threshold; For gas characteristic tags, 1 indicates a fire and 0 indicates no fire. Multi-feature joint identification: DS evidence theory is used to identify spectral feature results and temporal feature results. Gas characteristic labels The data is fused to output the final recognition result. ; The basic probability assignment function of the DS evidence theory is as follows: ; ; in: Let K be the basic probability assignment value for proposition A; K is the conflict coefficient. , , The propositions are respectively based on spectral, temporal, and gas characteristics. , , BPA value; Final identification rule: If and ,but ;otherwise .

[0010] In one embodiment, the smoke concentration dynamic estimation and environmental correction module is based on the fused smoke concentration. By considering the effects of temperature and humidity on the sensor, a dynamic concentration estimation model and environmental correction formula are constructed: Environmental parameters were collected using temperature and humidity sensors to measure ambient temperature. With relative humidity The sampling frequency is synchronized with the multi-source fusion sensing module, and the raw temperature and humidity signals are output. and Processed by moving average filtering: ; ; in: This is the size of the filtering window; The original temperature value of the nth sampling point This represents the original humidity value at the nth sampling point. Temperature affects the sensitivity of sensors and the motion state of smoke particles. Increased temperature leads to decreased sensitivity of optical sensors and increased ionization current of ionization sensors, thus necessitating the construction of a temperature correction coefficient. : ; in: This is the temperature correction factor; , , These are the fitting coefficients; The ambient temperature; Temperature-corrected concentration formula: ; in: This is the temperature-corrected smoke concentration; Increased humidity causes smoke particles to absorb moisture and increase in weight, altering their optical scattering properties and ionization blocking ability. A humidity correction coefficient is then constructed. : ; in: This is the humidity correction factor; Humidity influence coefficient; The relative humidity of the environment; Reference humidity; Humidity-corrected concentration formula: ; in: The smoke concentration after temperature and humidity correction; Smoke diffusion is a dynamic process. Based on the diffusion equations of fluid mechanics, a dynamic estimation model of smoke concentration over time and space is constructed: ; in: Let x be the smoke concentration at time t. The smoke diffusion coefficient; The speed of airflow; The intensity of the smoke source; The horizontal distance from the smoke source; For time; The concentration estimation formula is obtained by solving the diffusion equation using the finite difference method: ; in: ; ; for time The concentration value at that location.

[0011] In one embodiment, the multi-level dynamic early warning decision module constructs a multi-level early warning model and decision mechanism based on the smoke type identification result T and the dynamic concentration C(x,t), combined with the smoke diffusion rate, as follows: Smoke diffusion rate calculation: Based on the smoke concentration dynamic estimation and environmental correction module, the smoke diffusion rate is calculated. This reflects the speed at which the smoke spreads, as detailed below: ; in, ; in: , The positions of two adjacent grid points; , To achieve the concentration Time; The low-alert concentration threshold; Early warning level evaluation indicators: Construct two core evaluation indicators: concentration exceeding the standard. With diffusion risk : Concentration exceeding the standard: ; in: The dynamic concentration at time t, where the warning device is located; The threshold for no alarms; The high alert threshold; Propagation hazard level: ; in: The rate at which smoke diffuses; This refers to the rate of spread without police intervention. To ensure the speed of high-alert spread; Multi-level early warning decision-making: Calculating the comprehensive early warning index using a weighted summation formula. ,according to Classification of warning levels: ; in: Weights for concentration exceeding the scale; Weights for the degree of diffusion hazard; This is a comprehensive early warning index; Warning level classification rules: No police presence: or Non-fire smoke, no audible or visual alarms, only data recording; Low alert: and The green LED light flashes and the buzzer sounds an intermittent alarm. Chinese police: and The yellow LED light flashes and the buzzer sounds a continuous alarm. High Police: and The red LED light flashes and the buzzer sounds a high-decibel alarm. Early warning decisions are dynamically updated; The alert level is updated every 0.5 seconds, based on the latest data. and Recalculate To achieve dynamic adjustment; when At any time, regardless Both sizes are forcibly set to no alarm to avoid false alarms caused by non-fire smoke.

[0012] In one embodiment, the multi-mode communication and data transmission module adopts a multi-mode communication architecture of LoRa, NB-IoT, and Bluetooth. It dynamically selects the communication mode based on the warning level and environmental signal strength to achieve reliable transmission of warning information and data interaction, as detailed below: The core parameters for the three communication modes are as follows: Communication mode selection: The model is based on the early warning level. With communication signal strength Construct a communication mode selection function : ; in: Weighting of early warning levels; Signal strength weights; For mode m, the warning level Adaptability; For signal strength adaptation of mode m, , The minimum received signal strength for mode m; Data transmission protocol settings: A custom transmission protocol is used, including frame header, data segment, check segment, and frame trailer, to ensure data integrity; The data segment format is as follows: ; in: A unique identifier for the device; It is a type of smoke; Dynamic concentration; For diffusion rate; It is at the warning level; Confidence level; Signal strength; The check segment uses CRC16 verification: ; in: This represents the length of the data segment. For the first Bytes of data; 65536 is the modulus of CRC16; Communication energy consumption optimization: Dynamically adjust the communication cycle according to the warning level. Reduce power consumption in non-warning states: .

[0013] In one embodiment, the adaptive power management module constructs an adaptive power supply strategy based on the working status of each module and the remaining battery power, and dynamically adjusts the module power supply voltage and working cycle to achieve low power consumption operation. Battery remaining power estimation uses lithium batteries as the power source, by collecting battery voltage data. With discharge current Estimating the remaining power based on the coulomb method : ; in: Remaining battery level; This is the initial charge level; This refers to the battery's nominal capacity. for The discharge current at any given moment; Discharge time; The fitting formula for the relationship between battery voltage and SOC is as follows: ; Module power consumption model establishment: power consumption models for each core module, calculating power consumption under different operating conditions. : Multi-source fusion sensing module: ;in: To sense the operating current of the module; This refers to the power supply voltage. Intelligent signal processing module: ;in: This is the operating current of the microcontroller; This refers to the power supply voltage. Multimode communication and data transmission module: ,in: For the communication current of mode m; This refers to the power supply voltage. Multi-level dynamic early warning decision-making module: ;in: This refers to the LED operating current. LED voltage; This refers to the buzzer current. This is the buzzer voltage; Total system power consumption: ; Adaptive power supply strategy: based on SOC and warning level Dynamically adjust the working cycle of each module With supply voltage The details are as follows: Work cycle adjustment: ; Power supply voltage adjustment: A DC-DC converter is used to dynamically adjust the voltage according to the module requirements. ; in: This is the module's reference voltage; Low power protection mechanism: When SOC≤10%, low power protection is activated: Bluetooth module is turned off, only LoRa / NB-IoT communication is retained; LED brightness and buzzer volume are reduced, power consumption is reduced by 50%; low power alarm information is pushed to remind the user to replace the battery; In one embodiment, the multi-dimensional fault diagnosis and self-repair module constructs a multi-dimensional fault diagnosis model, identifies fault types, and triggers a self-repair mechanism: Sensor fault diagnosis: For optical sensors, ion sensors, and gas sensors in multi-source fusion sensing modules, fault diagnosis is performed through signal consistency and deviation analysis. Optical sensor failure: If or This is determined to be a fault; in: This is the historical average output of the optical sensor; Output the standard deviation of the historical data; Ion sensor malfunction: If or This is determined to be a fault; Gas sensor malfunction: If or This is determined to be a fault; Circuit fault diagnosis: Monitoring key parameters of the microcontroller, the ADC module within the microcontroller, and the DC-DC converter within the microcontroller. Microcontroller failure: If the watchdog timer times out, it is determined to be a program crash failure; Microcontroller ADC module fault: If the signal values ​​are the same for 10 consecutive acquisitions. The problem was determined to be an ADC malfunction. DC-DC converter fault in microcontroller: If the output voltage With set voltage deviation The problem was determined to be a converter malfunction. Communication fault diagnosis: Based on the ACK signal and received signal strength indication of the communication module, a communication fault judgment model is constructed to diagnose whether the communication link is abnormal in real time; The specific determination formula is as follows: ; in: For communication fault labels; ACK is for data transmission acknowledgment; RSSI is for received signal strength indication. For communication fault threshold; " is the logical "OR" operator.

[0014] The beneficial effects of this invention are as follows: Compared with the prior art, the control system of this invention comprehensively captures the multi-dimensional features of smoke through a multi-source fusion sensing module, improving detection accuracy and anti-interference capability from the source; the intelligent signal processing module converts the original signal into high-quality feature data, reducing processing latency and enhancing reliability; the smoke type identification module distinguishes between fire and non-fire smoke, solving the problem of false alarms; the concentration dynamic estimation and environmental correction module achieves accurate concentration detection and trend prediction under different environments; the multi-level early warning decision module dynamically adjusts the response strategy, taking into account both early warning targeting and emergency efficiency; the multi-mode communication module dynamically selects the communication mode according to the scenario, ensuring reliable information transmission and optimizing power consumption; the adaptive power management module intelligently regulates power supply, significantly extending battery life and reducing maintenance costs; and the multi-dimensional fault diagnosis and self-repair module monitors faults in real time and triggers repairs, reducing the risk of system failure. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a schematic diagram of the structure of the early warning device body in this invention; Figure 3 This is a schematic diagram of the structure of the optical sensor, ion sensor, and gas sensor in this invention; Figure 4 This is a schematic diagram of the framework structure of the control system in this invention.

[0016] The following components are marked in the diagram: 1. Warning device body; 2. Green LED; 3. Yellow LED; 4. Red LED; 5. Wind speed sensor; 6. Buzzer; 7. Optical sensor; 8. Ionization sensor; 9. Gas sensor; 10. Microcontroller; 11. Power supply assembly. Detailed Implementation

[0017] In the description of this invention, it should be noted that the terms "front", "up", "down", "left", "right", "vertical", "horizontal", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0018] refer to Figures 1-4In order to solve the problems existing in the background technology, this application proposes the following technical solution: a smoke detection and warning device, including a warning device body 1, wherein an optical sensor 7, an ionization sensor 8, and a gas sensor 9 are installed in the warning device body 1, and a green LED 2, a yellow LED 3, and a red LED 4 are installed on the outside of the warning device body 1. A microcontroller 10 and a power supply component 11 are fixedly installed inside the warning device body 1, and a buzzer 6 is also fixedly installed on the outside of the warning device body 1.

[0019] The control system specifically includes a multi-source fusion sensing module, an intelligent signal processing module, a smoke type identification module, a smoke concentration dynamic estimation and environmental correction module, a multi-level dynamic early warning decision module, a multi-mode communication and data transmission module, an adaptive power management module, and a multi-dimensional fault diagnosis and self-repair module.

[0020] The above technical solution is explained as follows: The multi-source fusion sensing module adopts a three-source collaborative sensing architecture consisting of "optical sensor 7, ionization sensor 8, and gas sensor 9". By complementary sensing of the optical characteristics, ionization characteristics, and gas composition characteristics of smoke, it solves the problem of weak anti-interference capability of single sensors and provides multi-dimensional raw data for subsequent signal processing. Using a scattering-type optical sensor 7, and based on the Lambert-Beer law, the intensity of the scattered light from smoke particles at a specific wavelength (650nm red light) is positively correlated with the smoke concentration. Its sensing signal output formula is as follows: ; in: The intensity of scattered light output by optical sensor 7 (unit: μA); The intensity of light emitted by the light source (a fixed value, taken as 50 μA); The sensitivity coefficient of optical sensor 7 (obtained through calibration, ranging from 0.02 to 0.05 μA / (mg / m²)) 3 )); Smoke concentration (unit: mg / m³) 3 ); The light absorption coefficient of smoke particles (based on experimental measurements of fire smoke). Values ​​range from 0.15 to 0.3m -1 cooking fumes Values ​​range from 0.05 to 0.1m -1 ); The distance between the light source and the receiver is a fixed value, taken as 0.05m. The inherent noise of optical sensor 7 (obtained through dark box testing, with a value ≤0.5μA).

[0021] In the above technical solution, the intensity of the scattered light output by the optical sensor 7 is calculated to reflect the scattering effect of smoke particles on light, providing a basis for preliminary estimation of smoke concentration.

[0022] Based on the ionization sensor 8, the blocking effect of smoke particles on alpha particles in the ionization chamber is used to change the ionization current. The formula for its output current is as follows: ; in: The output current of the ion sensor 8 (unit: nA); The reference current for the ionization chamber in the absence of smoke (fixed value, taken as 100nA); The response coefficient of the ion sensor is 8 (experimental calibration value, ranging from 0.003 to 0.006 m). 3 / mg) 0 . 8 ); For smoke concentration (as in optical sensor 7) Consistent, unit: mg / m 3 ); 0.8 is the nonlinear exponent of the smoke particle blocking effect (obtained through extensive experimental fitting, reflecting the correlation between particle size and blocking ability).

[0023] In the above technical solution, the interference of smoke particles on the ionization process is reflected by the change of ionization current, which supplements the optical sensor 7's ability to sense tiny particles (particle size ≤ 0.1 μm).

[0024] Gas sensor 9 is a semiconductor CO sensor. The CO concentration in fire smoke is typically ≥50ppm, while the CO concentration in non-fire smoke is ≤10ppm. The formula relating its output resistance to the CO concentration is as follows: ; in: The output resistance of gas sensor 9 (unit: kΩ); The reference resistance for gas sensor 9 when there is no CO (fixed value, taken as 100kΩ); The sensitivity coefficient of gas sensor 9 (experimental calibration value, ranging from 0.015 to 0.02 ppm) -1 ); The concentration of CO in the environment (unit: ppm).

[0025] In the above technical solution, the CO concentration is detected by the change in resistance, which serves as the key basis for distinguishing between fire smoke (containing CO) and non-fire smoke (without CO).

[0026] Multi-source data synchronous control and acquisition is implemented using an STM32H743 microcontroller 10 as the main controller, and a 16-bit ADC module to synchronously acquire the output signals of the optical sensor 7, ion sensor 8, and gas sensor 9. The sampling frequency is set to 1kHz, and the acquisition period is [not specified]. This ensures the consistency of multi-source data over time, laying the foundation for subsequent fusion processing.

[0027] The above technical solution comprehensively captures the multidimensional properties of smoke particles of different sizes by simultaneously acquiring the optical scattering characteristics, ionization blocking effect, and CO gas composition information of the smoke. This effectively resists the influence of interference sources such as cooking fumes and industrial dust. Compared with traditional single sensors, it has a more comprehensive sensing dimension and more accurate data coverage, providing rich and reliable raw data support for subsequent signal processing. This fundamentally improves the basic accuracy and anti-interference capability of smoke detection, solving the core problems of limited sensing and susceptibility to interference in existing technologies.

[0028] Intelligent signal processing module: For the raw signal output by the multi-source fusion sensing module, noise suppression, feature extraction and data fusion processing are performed in sequence to eliminate the influence of environmental noise (such as light changes and electromagnetic interference), extract the dynamic features of smoke, and provide high-quality feature data for the recognition module, thus solving the problem of coarse signal processing in existing technologies.

[0029] Specifically as follows: Wavelet transform noise suppression: The original signal is decomposed into three levels using the db4 wavelet basis function, and noise components are eliminated through thresholding to reconstruct a clean signal.

[0030] The wavelet decomposition formula is as follows: ; ; in: These are the low-frequency approximation coefficients for the j-th layer; Let be the high-frequency detail coefficients of the j-th layer.

[0031] These are the low-pass filter coefficients; (the filter coefficients corresponding to the db4 wavelet basis, with values ​​of: [-0.1830,0.3153,-0.0352,-0.0854,0.0352,0.3153,-0.1830,0.6309]).

[0032] These are the high-pass filter coefficients; (obtained by shifting and flipping h(n), with values ​​as follows:) [0.6309,0.1830,-0.3153,-0.0352,0.0854,-0.0352,0.3153,-0.1830]).

[0033] The number of decomposition levels (taken as 3); For coefficient index; Index of signal sampling points; Let be the signal coefficients of the (j-1)th layer.

[0034] In the above technical solution, the original signal is decomposed into a low-frequency approximate component (S(j,k)) and a high-frequency detail component (D(j,k)). The high-frequency component contains noise. After removing the high-frequency noise through thresholding, the signal is reconstructed to achieve noise suppression.

[0035] Noise thresholding: An adaptive threshold formula is used. ; Where: λ is the noise threshold; σ is the noise standard deviation of the high-frequency detail components (estimated through the high-frequency coefficients of the third layer). N is the number of signal sampling points (take 1000 points, corresponding to 1 second of sampling data).

[0036] In the above technical solution, the noise threshold of the high-frequency detail component is calculated. When the absolute value of the high-frequency coefficient is less than λ, it is determined to be noise and set to 0; when it is greater than λ, it is retained, thereby achieving adaptive noise suppression.

[0037] Smoke dynamic feature extraction: Core dynamic features are extracted from the denoised multi-source signal to reflect the changing trend and stability of smoke concentration, as detailed below: Concentration change rate: ; in: The rate of change in smoke concentration (unit: mg / (m³)) 3 ·s)); Let be the smoke concentration at time t; for The smoke concentration at any given time; The time interval is 0.2s.

[0038] In the above technical solution, the rate of change of smoke concentration per unit time is calculated, and the fire smoke... Usually ≥0.5mg / (m 3 ·s), non-fire smoke ≤0.1mg / (m 3 ·s).

[0039] Signal fluctuation variance: ; in: Variance of the output signal of optical sensor 7 (unit: μA); This represents the number of sampling points within the statistical window (50 points, corresponding to 0.05s). This is the sensor output for the m-th sampling point; This represents the average value of the sampled points within the window.

[0040] In the above technical solution, the stability of the sensor output signal is reflected, and the fluctuation of the fire smoke signal is small. ≤0.3μA), interference sources (such as cooking fumes) have large signal fluctuations ( ≥0.8μA).

[0041] CO concentration gradient: ; In the above technical solution, the gradient of CO concentration over time is calculated under fire scenarios. ≥0.5ppm / s, in non-fire scenarios ≈0.

[0042] in: CO concentration gradient (unit: ppm / s); Let be the CO concentration at time t; The initial CO concentration (the ambient CO concentration in the absence of smoke, typically ≤5ppm); The detection time is expressed in seconds.

[0043] Multi-source data fusion: A fuzzy neural network (FNN) is used to fuse feature data from optical, ionization, and gas sensors 9, outputting the fused smoke concentration. With confidence level .

[0044] The input layer of the FNN consists of the feature values ​​of three sensors (optical sensor 7, ion sensor 8, and gas sensor 9). , , ,in , , (The signal value is denoised), and the hidden layer uses a Gaussian membership function: ; In the above technical solution, the input feature value is fuzzified and transformed into fuzzy membership degree, which reflects the degree to which the input value belongs to a certain fuzzy set (such as "low concentration", "medium concentration" and "high concentration").

[0045] in: For the i-th input feature The membership degree of the j-th fuzzy set.

[0046] For the i-th input feature value (i=1,2,3, respectively corresponding to...) , , ).

[0047] The center value of the j-th fuzzy set of the i-th input feature (obtained through optimization of training samples, such as...) , , ).

[0048] The width of the j-th fuzzy set for the i-th input feature (training optimization value, such as...) , , ).

[0049] The output layer of FNN uses a weighted summation formula: ; ; in: The weight of the k-th fuzzy rule (training optimization value, summed to 1); The number of fuzzy rules (take 27, 3 inputs × 3 fuzzy sets); , which is the confidence coefficient (set to 5 to control the steepness of the confidence curve). The smoke concentration threshold (taken as 5 mg / m³) 3 (This is the lower limit of the warning specified in the national standard GB50116-2013).

[0050] In the above technical solution, The fused smoke concentration, which integrates information from multiple sensors, has improved accuracy compared to a single sensor. The confidence level for concentration estimation is 0 ≤ P ≤ 1, with data considered valid when P ≥ 0.8. Wavelet transform adaptive noise suppression technology accurately removes environmental noise interference from the original signal, ensuring signal purity. Subsequent extraction of dynamic features such as concentration change rate and signal fluctuation variance profoundly reflects the temporal variation of smoke. Fuzzy neural networks then enable deep fusion of multi-source data, significantly improving the quality and effectiveness of the feature data. This approach changes the traditional, extensive signal processing model, achieving efficient transformation from raw signals to high-quality features, significantly reducing signal processing latency, enhancing data reliability, and providing a high-quality data foundation for subsequent identification and decision-making modules.

[0051] Smoke type recognition module: based on output fusion features A multi-dimensional identification model based on "spectral features, temporal features, and gas features" is constructed to distinguish between fire smoke (such as wood burning smoke and plastic burning smoke) and non-fire smoke (such as cooking fumes, moxibustion smoke, and industrial dust), thus solving the problem that existing technologies cannot distinguish smoke types.

[0052] Spectral feature extraction: The spectral scattering ratio of smoke is calculated by using the multi-wavelength (650nm red light, 450nm blue light) scattering signal of optical sensor 7, which reflects the refractive index and particle size distribution of smoke particles (fire smoke particles refractive index ≈ 1.5, non-fire smoke particles ≈ 1.3).

[0053] Spectral scattering ratio formula: ; in: Spectral scattering ratio; wavelength Intensity of scattered light at (650nm) (unit: μA); wavelength The intensity of scattered light at (450nm) (unit: μA).

[0054] In the above technical solution, the optical properties of smoke particles are distinguished by the ratio of the intensities of scattered light at different wavelengths, thus differentiating the optical characteristics of fire smoke. (650nm / 450nm) is typically ≥1.2, and for non-fire smoke it is ≤0.8.

[0055] Temporal feature analysis: based on extracted and Construct time-domain feature vectors The temporal features are trained and identified using a support vector machine (SVM) classifier.

[0056] The kernel function of SVM uses radial basis functions (RBF): ; in: The kernel function value; This represents the temporal feature vector of the sample to be identified. The time-domain feature vector of the training samples; The kernel function parameters (optimized through cross-validation, with values ​​ranging from 0.1 to 10). The square of the Euclidean distance between the two vectors.

[0057] In the above technical solution, the time-domain feature vector is mapped to a high-dimensional space to solve the problem of linear inseparability and to classify the trend and stability of smoke concentration changes.

[0058] SVM decision function: ; In the above technical solution, the classification results of the time-domain features are output. When the value is 1, it is determined to be a fire smoke temporal feature. When the value is -1, it is determined to be a non-fire smoke temporal feature.

[0059] in: The Lagrange multipliers for the support vectors (training optimization values, only those corresponding to the support vectors) ); Labels for training samples (fire smoke) non-fire smoke ); The number of support vectors; Let be the temporal feature vector of the i-th support vector; This refers to the bias term (training optimization value) of the SVM. It is a symbolic function.

[0060] Gas characteristic verification: Based on data from gas sensor 9, CO concentration determination criteria are constructed as a "veto" item for smoke type identification: then else ; In the above technical solution, CO concentration ≥ threshold is determined to be fire smoke. Otherwise, it is determined to be non-fire smoke. This mitigates the risk of misjudgment based on spectral and temporal characteristics.

[0061] in: The CO concentration threshold is 20 ppm, referring to GB / T20284-2006 "Single Combustion Test of Building Materials or Products"; For gas characteristic labels (1 for fire, 0 for non-fire).

[0062] Multi-feature joint identification: DS evidence theory is used to identify spectral features. ( hour ,otherwise Temporal feature recognition results Gas characteristic labels The data is fused to output the final recognition result. .

[0063] The basic probability assignment (BPA) function of the DS evidence theory is as follows: ; ; In the above technical solution, the identification results of the three features are used as independent evidence. The evidence is fused through the BPA function to calculate the confidence level between "fire smoke" (A={1}) and "non-fire smoke" (A={0}). The one with the higher confidence level is the final identification result.

[0064] in: is the basic probability assignment value (confidence level) for proposition A; K is the conflict coefficient (reflecting the degree of conflict between evidences, fusion is effective when K<0.5). , , The propositions are respectively based on spectral, temporal, and gas characteristics. , , BPA value (e.g.) , The confidence level for identifying a fire based on spectral characteristics is 0.9.

[0065] Final identification rule: If and ,but (Fire smoke); otherwise (Non-fire smoke).

[0066] The above technical solution constructs a multi-feature joint identification system based on spectral, temporal, and gas characteristics, and combines DS evidence theory to achieve multi-dimensional evidence fusion.

[0067] By differentiating the optical properties of smoke particles through spectral scattering ratio, capturing concentration variation patterns through temporal feature analysis, and providing core criteria for determining fire smoke based on gas characteristics, this module works synergistically to accurately distinguish between fire smoke and non-fire smoke. Compared to existing technologies that cannot effectively identify smoke types, this module solves the problem of false alarms.

[0068] Smoke Concentration Dynamic Estimation and Environmental Correction Module: Based on Fuded Concentration By combining the effects of temperature and humidity on the sensor, a dynamic concentration estimation model and environmental correction formula are constructed to improve the accuracy of concentration estimation under different environments and solve the problem of concentration deviation caused by the failure of existing technologies to consider environmental factors.

[0069] Environmental parameters were collected using an SHT30 temperature and humidity sensor to measure ambient temperature. (Unit: °C) and relative humidity (Unit: %RH), sampling frequency synchronized with the multi-source fusion sensing module (1kHz), outputting raw temperature and humidity signals. and Processed by moving average filtering: ; ; The above technical solution eliminates high-frequency noise in temperature and humidity signals, obtains stable environmental parameter values, and provides a basis for subsequent calibration.

[0070] in: The filter window size is 20 points, corresponding to 0.02s. , This represents the original temperature and humidity values ​​at the nth sampling point.

[0071] Temperature affects the sensitivity of the sensor and the motion state of smoke particles. Increased temperature leads to decreased sensitivity of optical sensor 7 and increased ionization current of ionization sensor 8, necessitating the construction of a temperature correction coefficient. : ; In the above technical solution, the relationship between temperature and correction coefficient is fitted by a quadratic function to perform temperature compensation on the fusion concentration, so that the estimated concentration values ​​at different temperatures are consistent.

[0072] in: This is a temperature correction factor (range: 0.8-1.2). , , The fitting coefficient (determined experimentally) , , ); Ambient temperature (unit: °C). Temperature-corrected concentration formula: ; In the above technical solution, the fusion concentration is corrected to the temperature-compensated concentration, such as when At ℃, ,like mg / m 3 ,but mg / m 3 This aligns with the physical law that the actual concentration of smoke particles decreases due to diffusion at high temperatures.

[0073] in: Temperature-corrected smoke concentration (unit: mg / m³) 3 ).

[0074] Increased humidity causes smoke particles to absorb moisture and increase in weight, altering their optical scattering properties and ionization blocking ability. A humidity correction coefficient is then constructed. : ; In the above technical solution, the relationship between humidity and correction coefficient is linearly fitted to compensate for the influence of humidity on the characteristics of smoke particles.

[0075] in: This is the humidity correction factor (range: 0.9-1.1). The humidity effect coefficient (experimentally determined, taken as 0.002%RH) is used. -1 ); The relative humidity of the environment; The baseline humidity is 50%RH.

[0076] Humidity-corrected concentration formula: ; In the above technical solution, the temperature-corrected concentration is further corrected to the concentration after combined temperature and humidity correction, such as when RH=95%. ,like ,but This reflects the concentration perception bias caused by the increased weight of smoke particles under high humidity.

[0077] in: Smoke concentration after temperature and humidity correction (unit: ).

[0078] Considering the dynamic process of smoke diffusion, a dynamic estimation model of smoke concentration over time and space is constructed based on the diffusion equation of fluid mechanics: ; In the above technical solution, the smoke concentration at different locations (x) and at different times (t) is predicted to provide spatial information for early warning decision-making (such as determining whether the smoke has spread to key areas).

[0079] in: The smoke concentration at position x at time t (unit: mg / m³) 3 ); Smoke diffusion coefficient (the coefficient of dispersion of small particles in the air) (Measured through experiments) The airflow velocity (collected by wind speed sensor 5, unit: m / s; if there is no wind speed sensor 5, take 0.2 m / s, which is the average indoor wind speed). Smoke source intensity (unit: mg / (m²)) 3 •s), where fire smoke S≥10, and non-fire smoke S≤1). The horizontal distance from the smoke source (unit: m); Time (unit: seconds).

[0080] The concentration estimation formula is obtained by solving the diffusion equation using the finite difference method: ; In the above technical solution, grid points are obtained through iterative calculation. The concentration value at the specified location is used to achieve dynamic concentration prediction with a prediction error of ≤5%.

[0081] in: (i=0,1,…, , For spatial step size, (corresponding to a 5m range) (k=0,1,…, , (time step). for time The concentration value at that location.

[0082] In the above technical solution, the influence of temperature and humidity on detection accuracy is fully considered, and a temperature and humidity joint correction model is constructed to effectively compensate for the concentration estimation deviation caused by environmental factors.

[0083] Meanwhile, by combining the hydrodynamic diffusion equation to establish a dynamic concentration estimation model, accurate prediction of smoke concentration in both time and space dimensions is achieved. This ensures that the concentration detection results remain highly accurate under different environmental conditions. It can not only reflect the current concentration in real time, but also predict the smoke diffusion trend, providing more comprehensive and accurate concentration data support for early warning decision-making.

[0084] The multi-level dynamic early warning decision module is based on the smoke type identification result T and the dynamic concentration C(x,t), combined with the smoke diffusion speed, to construct a multi-level early warning model and decision mechanism, realize the dynamic switching of "no warning - low warning - medium warning - high warning", and solve the problems of single early warning level and crude decision-making in existing technologies.

[0085] Smoke diffusion rate calculation: Based on the smoke concentration dynamic estimation and environmental correction module, the smoke diffusion rate is calculated. This reflects the speed at which the smoke spreads, as detailed below: ; ; In the above technical solution, the diffusion rate is calculated by the location and time difference when the concentration reaches the low alarm threshold. The vs of fire smoke is usually ≥0.3m / s, and the vs of non-fire smoke is ≤0.1m / s.

[0086] in: , The position of two adjacent grid points (unit: m); , To achieve the concentration Time (unit: seconds); The low alarm concentration threshold (taken as 8 mg / m³) 3 ).

[0087] Early warning level evaluation indicators: Construct two core evaluation indicators: concentration exceeding the standard. With diffusion risk : Concentration exceeding the standard: ; in: The dynamic concentration at time t of the warning device location (unit: mg / m³) 3 ); The threshold for no alarm (taken as 5 mg / m³) 3 ); High alert threshold (taken as 20 mg / m³) 3 ).

[0088] Propagation hazard level: ; In the above technical solution, the degree of smoke diffusion speed relative to the danger threshold is quantified, with a value range of 0-1.

[0089] in: The smoke diffusion velocity (unit: m / s); The velocity of the virus without warning is 0.1 m / s. The high-alert spread velocity is taken as 0.5 m / s.

[0090] Multi-level early warning decision-making: Calculating the comprehensive early warning index using a weighted summation formula. ,according to Classification of warning levels: ; The above technical solution integrates two dimensions, concentration and diffusion rate, to output an early warning index and achieve multi-level dynamic early warning.

[0091] in: The weight for concentration exceeding the scale is set to 0.6, with concentration being the core indicator. The weight for the risk of spread is 0.4, with the spread rate as an auxiliary indicator. The comprehensive early warning index (value range 0-1).

[0092] Warning level classification rules: No police presence: or (Non-fire smoke) No audible or visual alarm, only data is recorded; Low alert: and The green LED2 light flashes (frequency 1Hz), and the buzzer (6) alarms intermittently (1 sound / 5s). Chinese police: and The yellow LED3 light flashes (frequency 2Hz), and the buzzer (6) sounds a continuous alarm (volume 60dB). High Police: and The red LED 4 lights flash (frequency 5Hz), the buzzer (6) alarms at a high decibel level (volume 85dB), and can also remotely push early warning information (APP), link fire-fighting equipment (such as sprinklers and fire extinguishers), make emergency calls (119, user mobile phone), etc.

[0093] Early warning decisions are dynamically updated; The alert level is updated every 0.5 seconds, based on the latest data. and Recalculate This allows for dynamic adjustment. When At any time, regardless Both sizes are forcibly set to no alarm to avoid false alarms caused by non-fire smoke.

[0094] The aforementioned technical solution constructs a multi-dimensional early warning evaluation system based on smoke type, dynamic concentration, and diffusion speed, innovatively dividing the warning levels into four levels: "no alarm," "low alarm," "medium alarm," and "high alarm." By dynamically updating the warning index, the warning level can be adjusted in real time, with different levels corresponding to differentiated audible and visual alarms, information push notifications, and equipment linkage strategies. Compared to traditional single-level warnings, this approach is more targeted. Low alarms alert users, while high alarms quickly link fire-fighting equipment and emergency communications. This avoids the confusion caused by excessive warnings and ensures timely and efficient emergency response in the event of a fire, maximizing the protection of personnel and property.

[0095] The multi-mode communication and data transmission module adopts a multi-mode communication architecture of "LoRa+NB-IoT+Bluetooth". It dynamically selects the communication mode according to the warning level and the strength of the environmental signal, so as to realize the reliable transmission of warning information and data interaction, and solve the problems of single communication mode and low transmission reliability of existing technologies.

[0096] The core parameters for the three communication modes are as follows: LoRa: Transmission distance 1-3km (line of sight), data rate 0.3-50kbps, power consumption 15-30mA (transmit current), suitable for long-distance industrial transmission; NB-IoT: Transmission distance 0.5-1.5km, data rate 0.1-20kbps, power consumption 5-15mA, suitable for wide area networks (operator coverage). Bluetooth: Transmission distance 10-50m, speed 1-2Mbps, power consumption 50-100mA, suitable for short-range local interaction (such as mobile APP configuration).

[0097] Communication mode selection: The model is based on the early warning level. (No alarm = 0, Low alarm = 1, Medium alarm = 2, High alarm = 3) and communication signal strength (Unit: dBm), Construct a communication mode selection function : ; In the above technical solution, the early warning level adaptability is achieved. Compatibility with signal strength We use a weighted summation to select the optimal communication mode.

[0098] in: The warning level weight is set to 0.7, with higher-level warnings prioritizing the reliable mode. The signal strength weight is set to 0.3. For mode m, the warning level Adaptability (e.g., during high alert periods) , , ).

[0099] Signal strength fit for mode m ( , The minimum received signal strength for mode m, such as LoRa. , ).

[0100] Data transmission protocol settings: A custom transmission protocol is used, including frame header, data segment, check segment, and frame trailer, to ensure data integrity.

[0101] The data segment format is as follows: ; The above technical solution defines the content of the transmitted data, which includes seven fields: device ID, smoke type, concentration, diffusion speed, warning level, confidence level, and signal strength, with a total length of 16 bytes.

[0102] in: This is a unique identifier for the device (4 bytes). Smoke type (1 byte, 0 = non-fire, 1 = fire); Dynamic concentration (2 bytes, precision 0.1 mg / m³) 3 ); Diffusion velocity (2 bytes, precision 0.01 m / s); Warning level (1 byte); Confidence level (1 byte, precision 0.01); Signal strength (1 byte, in dBm); The check segment uses CRC16 verification: ; In the above technical solution, cyclic redundancy check (CRC) is used to detect errors during data transmission, with a success rate of ≥99.99%. in: The data segment length (16 bytes); For the first Bytes of data; 65536 is the modulus of CRC16.

[0103] Communication energy consumption optimization: Dynamically adjust the communication cycle according to the warning level. Reduce power consumption in non-warning states: ; In the above technical solution, the communication frequency is reduced when there is no alarm and increased when there is a high alarm, ensuring real-time information transmission while optimizing energy consumption. For example, the communication power consumption when there is no alarm is only 1 / 300 of that when there is a high alarm, significantly improving battery life.

[0104] The above technical solution employs a multi-mode communication architecture combining LoRa, NB-IoT, and Bluetooth, dynamically selecting the optimal communication mode based on the warning level and signal strength to ensure reliable transmission of warning information in different scenarios. Simultaneously, a customized high-efficiency data transmission protocol and optimized communication cycle significantly reduce communication power consumption in non-warning states while ensuring real-time transmission of high-alert information.

[0105] The adaptive power management module constructs an adaptive power supply strategy based on the working status of each module in the system (such as sensing, processing, communication, and alarm) and the remaining battery power. It dynamically adjusts the module's power supply voltage and working cycle to achieve low-power operation and solve the problem of short battery life caused by fixed power supply in existing technologies.

[0106] The remaining battery capacity is estimated using a lithium battery (3.7V / 5000mAh) as the power supply component 11, by collecting battery voltage data. With discharge current Estimating the remaining power based on the coulomb method (StateofCharge): ; In the above technical solution, the remaining battery power percentage is calculated in real time, providing a basis for adjusting the power supply strategy.

[0107] in: Remaining battery power (in %); Initial charge (100%) This refers to the nominal capacity of the battery (5000mAh). for Discharge current at any given moment (unit: mA); Discharge time (in hours). The formula for fitting the relationship between battery voltage and SOC is: ; In the above technical solution, voltage-assisted calibration of SOC improves estimation accuracy, with an SOC error ≤3%. -Wherein: The fitting coefficients (taken as 5V) ); The current battery voltage (unit: V, 3.0V≤) ≤4.2V). This is the battery cutoff voltage (3.0V).

[0108] Module power consumption model establishment: power consumption models for each core module, calculating power consumption under different operating conditions. : Multi-source fusion sensing module: ;in: The operating current of the sensing module is 20mA during sampling and 0.1mA during sleep. The supply voltage is 3.3V.

[0109] Intelligent signal processing module: ;in: The operating current of the microcontroller 10 is 50mA during operation and 0.5mA during sleep. The supply voltage is 3.3V.

[0110] Multimode communication and data transmission module: ;in: The communication current for mode m is 25mA for LoRa transmission, 15mA for NB-IoT transmission, and 80mA for Bluetooth transmission. The supply voltage is 3.3V.

[0111] Multi-level dynamic early warning decision-making module: ;in: LED operating current (5mA for green, 8mA for yellow, 10mA for red); LED voltage (2.0V); The current for the buzzer is 6 (10mA at 60dB, 20mA at 85dB); The voltage for the buzzer (6) is 3.3V.

[0112] Total system power consumption: ; Adaptive power supply strategy: based on SOC and warning level Dynamically adjust the working cycle of each module With supply voltage The details are as follows: Work cycle adjustment: ; In the above technical solution, when the SOC decreases or the warning level decreases, the module's working cycle is extended and power consumption is reduced.

[0113] in: The baseline working cycle for the modules is 0.001s for the sensing module and 0.1s for the processing module. The warning level coefficient ( hour , hour ).

[0114] Power supply voltage adjustment: A DC-DC converter is used to dynamically adjust the voltage according to module requirements. For example, when SOC ≤ 20%, the power supply voltage of non-core modules (such as Bluetooth) is reduced from 3.3V to 3.0V to reduce power consumption. ; In the above technical solution, when the SOC is below 20%, the power supply voltage is linearly reduced to reduce power consumption while ensuring the normal operation of the module.

[0115] in: The module reference voltage is 3.3V.

[0116] Low battery protection mechanism: When SOC ≤ 10%, low battery protection is activated. Disable the Bluetooth module and retain only LoRa / NB-IoT communication; Reduce LED brightness and buzzer volume, and power consumption is reduced by 50%; The device can push low battery alerts to remind users to replace the battery, and a corresponding app can be developed to monitor the device.

[0117] In the above technical solution, a dynamic power supply strategy is constructed based on the remaining battery power and the warning level. By adjusting the working cycle and power supply voltage of each module, intelligent optimization of power consumption is achieved. At the same time, the remaining battery power is accurately estimated and a low-power protection mechanism is set to ensure that the system can efficiently utilize electrical energy under different operating conditions.

[0118] The multi-dimensional fault diagnosis and self-repair module monitors the working status of several key components such as sensors, circuits, communications, and power supplies in real time, builds a multi-dimensional fault diagnosis model, identifies fault types, and triggers a self-repair mechanism to avoid missed detections caused by system failures and solves the problem of missing fault detection in existing technologies.

[0119] Sensor fault diagnosis: For the three sensors (optical sensor 7, ion sensor 8, and gas sensor 9) of the multi-source fusion sensing module, faults are diagnosed through signal consistency and deviation analysis. Optical sensor 7 malfunction: If or If the fault is identified (e.g., damaged light source, blocked receiver), it is determined to be a fault.

[0120] In the above technical solution, sensor anomalies are identified by whether the deviation of the signal from the historical average exceeds three times the standard deviation.

[0121] in: The historical average output of optical sensor 7 (unit: μA); The historical output standard deviation (unit: μA).

[0122] Ion sensor 8 faults: If or The fault is identified as such (e.g., ionization chamber failure, circuit short circuit).

[0123] In the above technical solution, the fault is identified based on the normal current range of the ion sensor 8.

[0124] Gas sensor 9 malfunction: If or The fault is identified as a problem (such as sensor aging or poor lead contact).

[0125] In the above technical solution, the fault is identified based on the normal resistance range of the gas sensor 9.

[0126] Circuit fault diagnosis: Monitoring key parameters of microcontroller 10, ADC module, and DC-DC converter: Microcontroller 10 fault: If the watchdog timer times out (without resetting for more than 1 second), it is determined to be a program crash fault.

[0127] ADC module failure: If the signal values ​​are the same in 10 consecutive acquisitions. The problem was determined to be an ADC malfunction.

[0128] DC-DC converter malfunction: If the output voltage... With set voltage deviation The problem was determined to be a converter malfunction.

[0129] Communication fault diagnosis: Based on the ACK (acknowledgment) signal and Received Signal Strength Indication (RSSI) of the communication module, a communication fault determination model is constructed to diagnose whether the communication link is abnormal in real time. The specific determination formula is as follows: ; The above technical solution accurately identifies communication faults through dual judgment conditions (ACK loss and low signal strength). A communication fault is determined when three consecutive data transmissions fail to receive an ACK response from the receiver (indicating data transmission failure), or when the signal strength remains below the fault threshold for five consecutive seconds (indicating extremely poor link quality). Otherwise, it is considered that the communication is normal. ).

[0130] in: The following are the communication fault labels (1 indicates a fault, 0 indicates normal); ACK is the data transmission acknowledgment signal (an acknowledgment signal returned by the receiving end after successfully receiving data); RSSI is the received signal strength indicator (unit: dBm, reflecting the degree of signal attenuation in the communication link). Communication fault threshold (set according to communication mode: LoRa mode) dBm, NB-IoT mode dBm, in Bluetooth mode dBm); The "OR" operator is used (a fault is determined when either condition is met). This diagnostic model avoids misjudgments caused by instantaneous signal fluctuations by combining a time window (5 seconds) and a threshold of 3 times. The fault identification accuracy is ≥99%, providing a reliable triggering basis for subsequent self-repair mechanisms.

[0131] The above technical solution comprehensively monitors the working status of key components such as sensors, circuits, communications, and power supplies, and establishes a multi-dimensional fault diagnosis model, which can quickly and accurately identify various fault types.

[0132] In summary, this embodiment utilizes a multi-source fusion sensing module that combines optical, ionization, and gas sources to overcome the limitations of a single sensor, comprehensively capturing multi-dimensional smoke features and improving detection accuracy and anti-interference capabilities from the source. The intelligent signal processing module transforms the original signal into high-quality feature data through wavelet transform denoising, dynamic feature extraction, and fuzzy neural network fusion, reducing processing latency and enhancing reliability. The smoke type identification module relies on multi-feature joint identification and evidence fusion to accurately distinguish between fire and non-fire smoke, solving the false alarm problem. The concentration dynamic estimation and environmental correction module combines temperature and humidity correction and a diffusion model to achieve accurate concentration detection and trend prediction under different environments. The multi-level early warning decision module constructs a four-level early warning system, dynamically adjusting response strategies to balance early warning targeting and emergency efficiency. The multi-mode communication module dynamically selects the communication mode according to the scenario, ensuring reliable information transmission and optimizing power consumption. The adaptive power management module intelligently regulates power supply, significantly extending battery life and reducing maintenance costs. The multi-dimensional fault diagnosis and self-repair module monitors faults in real time and triggers repairs, reducing the risk of system failure.

[0133] Although embodiments of the invention have been shown and described, the scope of the invention will be defined by the appended claims and their equivalents by those skilled in the art.

Claims

1. A smoke detection and warning device, comprising a warning device body (1), characterized in that, The warning device body (1) is equipped with an optical sensor (7), an ion sensor (8), and a gas sensor (9). The exterior of the warning device body (1) is equipped with a wind speed sensor (5), a green LED (2), a yellow LED (3), and a red LED (4). The warning device body (1) is fixedly equipped with a microcontroller (10) and a power supply assembly (11). The exterior of the warning device body (1) is also fixedly equipped with a buzzer (6).

2. A control system for a smoke detection and early warning device, characterized in that, The control system specifically includes a multi-source fusion sensing module, an intelligent signal processing module, a smoke type identification module, a smoke concentration dynamic estimation and environmental correction module, a multi-level dynamic early warning decision module, a multi-mode communication and data transmission module, an adaptive power management module, and a multi-dimensional fault diagnosis and self-repair module. in: The multi-source fusion sensing module is used to sense the optical characteristics, ionization characteristics and gas composition characteristics of smoke; The intelligent signal processing module is used to extract the dynamic features of the smoke; The smoke type identification module constructs a multi-dimensional identification model based on "spectral features, temporal features, and gas features" to distinguish between fire smoke and non-fire smoke. The smoke concentration dynamic estimation and environmental correction module is used to construct a dynamic concentration estimation model and an environmental correction formula. The multi-level dynamic early warning decision module is used to construct a multi-level early warning model and decision-making mechanism; The multi-mode communication and data transmission module dynamically selects the communication mode based on the warning level and the strength of the environmental signal; The adaptive power management module constructs an adaptive power supply strategy based on the working status of each module and the remaining battery power. The multi-dimensional fault diagnosis and self-repair module constructs a multi-dimensional fault diagnosis model, identifies fault types, and triggers a self-repair mechanism.

3. The control system for a smoke detector and early warning device according to claim 2, characterized in that, The multi-source fusion sensing module is based on a three-source collaborative sensing architecture of optical sensor (7), ionization sensor (8), and gas sensor (9). It complementarily senses the optical characteristics, ionization characteristics, and gas composition characteristics of smoke, providing multi-dimensional raw data for subsequent signal processing, as detailed below: The scattering optical sensor (7) shows that, according to the Lambert-Beer law, the intensity of the scattered light of smoke particles at a specific wavelength is positively correlated with the smoke concentration; Its sensing signal output formula is as follows: ； in: The intensity of the scattered light output by the optical sensor (7); The intensity of light emitted by the light source; The sensitivity coefficient of the optical sensor (7); Smoke concentration; The light absorption coefficient of smoke particles; The distance between the light source and the receiver; The inherent noise of the optical sensor (7); The ion sensor (8) utilizes the blocking effect of smoke particles on α particles in the ionization chamber to change the ionization current. Its output current formula is as follows: ； in: Output current for the ion sensor (8); The reference current for the ionization chamber in the absence of smoke; The response coefficient of the ion sensor (8); denoted as smoke concentration; 0.8 represents the nonlinear exponent of the smoke particle blocking effect. The gas sensor (9) is a CO sensor. The formula for the relationship between the output resistance of the CO sensor and the CO concentration is as follows: ； in: The output resistance of the gas sensor (9); The reference resistance of the gas sensor (9) is when there is no CO. The sensitivity coefficient of the gas sensor (9); The concentration of CO in the environment; Multi-source data synchronous control and acquisition uses an STM32H743 microcontroller (10) as the main controller, and then a 16-bit ADC module to synchronously acquire the output signals of the optical sensor (7), ion sensor (8), and gas sensor (9). The sampling frequency is set to 1kHz, and the acquisition period is [not specified]. .

4. The control system for a smoke detector and early warning device according to claim 3, characterized in that, The intelligent signal processing module sequentially performs noise suppression, feature extraction, and data fusion processing on the raw signal output by the multi-source fusion sensing module to eliminate the influence of environmental noise and extract the dynamic features of the smoke, as detailed below: Wavelet transform noise suppression: The original signal is decomposed into three levels of wavelet using the db4 wavelet basis function, and noise components are eliminated by thresholding to reconstruct a clean signal; The wavelet decomposition formula is as follows: ； ； in: These are the low-frequency approximation coefficients for the j-th layer; Here are the high-frequency detail coefficients for the j-th layer; These are the low-pass filter coefficients; These are the high-pass filter coefficients; The number of decomposition layers; For coefficient index; Index of signal sampling points; These are the signal coefficients of the (j-1)th layer; Noise thresholding: An adaptive threshold formula is used. ； Where: λ is the noise threshold; σ is the noise standard deviation of the high-frequency detail components; N is the number of signal sampling points; Smoke dynamic feature extraction: Core dynamic features are extracted from the denoised multi-source signal to reflect the changing trend and stability of smoke concentration, as detailed below: Concentration change rate: ; in: The rate of change in smoke concentration; Let be the smoke concentration at time t; Let be the smoke concentration at time t-Δt; For time intervals; Signal fluctuation variance: ; in: The variance of the output signal of the optical sensor (7); This represents the number of sampling points within the statistics window; This is the sensor output for the m-th sampling point; This represents the average value of the sampled points within the window. CO concentration gradient: ; in: For CO concentration gradient; Let be the CO concentration at time t; The initial CO concentration; For detection time; Multi-source data fusion: A fuzzy neural network is used to fuse the feature data of the optical sensor (7), the ionization sensor (8), and the gas sensor (9), and the fused smoke concentration is output. With confidence level ; The input layer of the FNN consists of the feature values ​​of the optical sensor (7), the ion sensor (8), and the gas sensor (9). , , ,in , , The hidden layer uses a Gaussian membership function to represent the denoised signal value. ； in: For the i-th input feature The membership degree of the j-th fuzzy set; Let i be the i-th input feature value, i=1,2,3, which corresponds to... , , ; The center value of the j-th fuzzy set of the i-th input feature; The width of the j-th fuzzy set of the i-th input feature; The output layer of FNN uses a weighted summation formula: ； ； in: Let be the weight of the k-th fuzzy rule; The number of fuzzy rules; Confidence coefficient; This represents the smoke concentration threshold.

5. The control system for a smoke detector and early warning device according to claim 4, characterized in that, The smoke type recognition module is based on the output fusion features. A multi-dimensional identification model based on "spectral features, temporal features, and gas features" is constructed to distinguish between fire smoke and non-fire smoke, as detailed below: Spectral feature extraction: The spectral scattering ratio of smoke is calculated by using the multi-wavelength scattering signal of the optical sensor (7), which reflects the refractive index and particle size distribution of the smoke particles; Spectral scattering ratio formula: ； in: Spectral scattering ratio; wavelength The intensity of the scattered light; wavelength The intensity of the scattered light; Temporal feature analysis: based on extracted and Construct time-domain feature vectors The temporal features are trained and recognized using a support vector machine classifier; The kernel function of SVM uses radial basis functions: ； in: The kernel function value; This represents the temporal feature vector of the sample to be identified. The time-domain feature vector of the training samples; These are kernel function parameters; The square of the Euclidean distance between the two vectors; SVM decision function: ； in: The Lagrange multipliers for support vectors; Labels for the training samples; The number of support vectors; Let be the temporal feature vector of the i-th support vector; For SVM bias terms; It is a symbolic function; Gas characteristic verification: Based on the data from the gas sensor (9), CO concentration determination conditions were constructed as a method for smoke type identification: then else ; in: The CO concentration threshold; For gas characteristic tags, 1 indicates a fire and 0 indicates no fire. Multi-feature joint identification: DS evidence theory is used to identify spectral feature results and temporal feature results. Gas characteristic labels The data is fused to output the final recognition result. ; The basic probability assignment function of the DS evidence theory is as follows: ； ； in: Let K be the basic probability assignment value for proposition A; K is the conflict coefficient. For spectral characteristics of the proposition BPA value, For the time domain characteristics of the proposition BPA value, For the characteristics of gases, the proposition BPA value; Final identification rule: If and ,but ;otherwise .

6. The control system for a smoke detector and early warning device according to claim 5, characterized in that, The smoke concentration dynamic estimation and environmental correction module is based on the fused smoke concentration. By considering the effects of temperature and humidity on the sensor, a dynamic concentration estimation model and environmental correction formula are constructed, as follows: Environmental parameters were collected using temperature and humidity sensors to measure ambient temperature. With relative humidity The sampling frequency is synchronized with the multi-source fusion sensing module, and the raw temperature and humidity signals are output. and The result is processed using a moving average filter, as follows: ； ； in: This is the size of the filtering window; The original temperature value of the nth sampling point This represents the original humidity value at the nth sampling point. Temperature affects the sensitivity of the sensor and the motion state of smoke particles. Increased temperature leads to decreased sensitivity of the optical sensor (7) and increased ionization current of the ionization sensor (8), thus requiring the construction of a temperature correction coefficient. : ； in: This is the temperature correction factor; , , These are the fitting coefficients; The ambient temperature; Temperature-corrected concentration formula: ； in: This is the temperature-corrected smoke concentration; Increased humidity causes smoke particles to absorb moisture and increase in weight, altering their optical scattering properties and ionization blocking ability. A humidity correction coefficient is then constructed. : ； in: This is the humidity correction factor; Humidity influence coefficient; The relative humidity of the environment; Reference humidity; Humidity-corrected concentration formula: ； in: The smoke concentration after temperature and humidity correction; Smoke diffusion is a dynamic process. Based on the diffusion equations of fluid mechanics, a dynamic estimation model of smoke concentration over time and space is constructed: ； in: Let x be the smoke concentration at time t. The smoke diffusion coefficient; The speed of airflow; The intensity of the smoke source; The horizontal distance from the smoke source; For time; The concentration estimation formula is obtained by solving the diffusion equation using the finite difference method: ； in: , =0, 1..., , =0.5m is the spatial step size. =10 corresponds to a range of 5m; ; =0, 1..., , =0.1s is the time step; for time The concentration value at that location.

7. The control system for a smoke detector and early warning device according to claim 6, characterized in that, The multi-level dynamic early warning decision module, based on the smoke type identification result T and the dynamic concentration C(x,t), combined with the smoke diffusion rate, constructs a multi-level early warning model and decision mechanism, as detailed below: Smoke diffusion rate calculation: Based on the smoke concentration dynamic estimation and environmental correction module, the smoke diffusion rate is calculated. This reflects the speed at which the smoke spreads, as detailed below: ； in, ; in: , The positions of two adjacent grid points; , To achieve the concentration Time; The low-alert concentration threshold; Early warning level evaluation indicators: Construct two core evaluation indicators: concentration exceeding the standard. With diffusion risk : Concentration exceeding the standard: ； in: The dynamic concentration at time t, where the warning device is located; The threshold for no alarms; The high alert threshold; Propagation hazard level: ； in: The rate at which smoke diffuses; This refers to the rate of spread without police intervention. To ensure the speed of high-alert spread; Multi-level early warning decision-making: Calculating the comprehensive early warning index using a weighted summation formula. ,according to Classification of warning levels: ； in: Weights for concentration exceeding the scale; Weights for the degree of diffusion hazard; This is a comprehensive early warning index; Warning level classification rules: No police presence: or Non-fire smoke, no audible or visual alarms, only data recording; Low alert: and The green LED (2) flashes, and the buzzer (6) sounds an intermittent alarm; Chinese police: and The yellow LED (3) flashes and the buzzer (6) continues to sound an alarm; High Police: and The red LED (4) flashes, and the buzzer (6) sounds a high-decibel alarm; Early warning decisions are dynamically updated; The alert level is updated every 0.5 seconds, based on the latest data. and Recalculate To achieve dynamic adjustment; when At any time, regardless Both sizes are forcibly set to no alarm to avoid false alarms caused by non-fire smoke.

8. The control system for a smoke detector and early warning device according to claim 7, characterized in that, The multi-mode communication and data transmission module adopts a multi-mode communication architecture of LoRa, NB-IoT, and Bluetooth. It dynamically selects the communication mode based on the warning level and environmental signal strength to achieve reliable transmission of warning information and data interaction, as detailed below: The core parameters for the three communication modes are as follows: Communication mode selection: The model is based on the early warning level. With communication signal strength Construct a communication mode selection function : ； in: Weighting of early warning levels; Signal strength weights; For mode m, the warning level Adaptability; For signal strength adaptation of mode m, , The minimum received signal strength for mode m; Data transmission protocol settings: A custom transmission protocol is used, including frame header, data segment, check segment, and frame trailer; The data segment format is as follows: ; in: A unique identifier for the device; It is a type of smoke; Dynamic concentration; For diffusion rate; It is at the warning level; Confidence level; Signal strength; The check segment uses CRC16 verification: ; in: This represents the length of the data segment. For the first Bytes of data; 65536 is the modulus of CRC16; Communication energy consumption optimization: Dynamically adjust the communication cycle according to the warning level. Reduce power consumption in non-warning states: 。 9. The control system for a smoke detector and early warning device according to claim 8, characterized in that, The adaptive power management module constructs an adaptive power supply strategy based on the operating status of each module and the remaining battery power, dynamically adjusting the module power supply voltage and duty cycle to achieve low-power operation, as detailed below: The remaining battery power is estimated using a lithium battery as the power supply component (11), by collecting battery voltage data. With discharge current Estimating the remaining power based on the coulomb method : ； in: Remaining battery level; This is the initial charge level; This refers to the battery's nominal capacity. for The discharge current at any given moment; Discharge time; The fitting formula for the relationship between battery voltage and SOC is as follows: ； Module power consumption model establishment: power consumption models for each core module, calculating power consumption under different operating conditions. : Multi-source fusion sensing module: ;in: To sense the operating current of the module; This refers to the power supply voltage. Intelligent signal processing module: ;in: This is the operating current of the microcontroller (10); This refers to the power supply voltage. Multimode communication and data transmission module: ,in: For the communication current of mode m; This refers to the power supply voltage. Multi-level dynamic early warning decision-making module: ;in: This refers to the LED operating current. LED voltage; The current for the buzzer (6); Voltage for buzzer (6); Total system power consumption: ; Adaptive power supply strategy: based on SOC and warning level Dynamically adjust the working cycle of each module With supply voltage The details are as follows: Work cycle adjustment: ; Power supply voltage adjustment: A DC-DC converter is used to dynamically adjust the voltage according to the module requirements. ； in: This is the module's reference voltage; Low power protection mechanism: When SOC≤10%, low power protection is activated: turn off the Bluetooth module and retain only LoRa / NB-IoT communication; reduce LED brightness and buzzer (6) volume.

10. The control system for a smoke detector and early warning device according to claim 9, characterized in that, The multi-dimensional fault diagnosis and self-repair module constructs a multi-dimensional fault diagnosis model, identifies fault types, and triggers a self-repair mechanism, as detailed below: Sensor fault diagnosis: For the optical sensor (7), ion sensor (8), and gas sensor (9) of the multi-source fusion sensing module, faults are diagnosed through signal consistency and deviation analysis. Optical sensor (7) fault: If or This is determined to be a fault; in: The historical average output of the optical sensor (7); Output the standard deviation of the historical data; Ion sensor (8) fault: If or This is determined to be a fault; Gas sensor (9) fault: If or This is determined to be a fault; Circuit fault diagnosis: Monitoring key parameters of the microcontroller (10), the ADC module in the microcontroller (10), and the DC-DC converter in the microcontroller (10): Microcontroller (10) fault: If the watchdog timer times out, it is determined to be a program crash fault; ADC module fault in microcontroller (10): If the signal values ​​collected 10 times consecutively are the same The problem was determined to be an ADC malfunction. DC-DC converter fault in microcontroller (10): If the output voltage With set voltage deviation The problem was determined to be a converter malfunction. Communication fault diagnosis: Based on the ACK signal and received signal strength indication of the communication module, a communication fault judgment model is constructed to diagnose whether the communication link is abnormal in real time; The specific determination formula is as follows: ； in: For communication fault labels; ACK is for data transmission acknowledgment; RSSI is for received signal strength indication. For communication fault thresholds; "is the logical OR operator".

Images