Intelligent fire smoke sensing detection system and method based on AI spectrum analysis

Abstract

The application provides an intelligent fire-fighting smoke sensing detection system and method based on AI spectrum analysis, comprising a surface area type active detection module, a composite sensing fusion module, an intelligent early warning decision module, a self-adaptive environment learning module and an Internet of Things communication module; aims to solve the shortcomings of the existing fire-fighting smoke sensing detection technology in monitoring range, early warning timeliness, environment adaptation, judgment accuracy and system compatibility, and provide an intelligent fire-fighting smoke sensing detection system and method capable of realizing ultra-early warning, high-precision fire identification, full-scene adaptation and remote upgrading, and meeting the needs of intelligent fire fighting.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent fire-fighting technology, and in particular to an intelligent fire-fighting smoke sensing detection system and method based on AI spectrum analysis. BACKGROUND

[0002] With the acceleration of urbanization and the increase in the complexity of industrial production, the demand for fire safety prevention and control has shifted from post-fire extinguishing to pre-warning, which has promoted the development of fire-fighting smoke sensing detection technology through multiple generations. Early traditional smoke sensing equipment mainly uses point ion smoke sensing technology, which can only detect smoke through a single ion sensor, and the monitoring range is limited to 1-2 meters around the sensor. Moreover, it is greatly disturbed by environmental humidity and dust, and the false alarm rate is as high as more than 20%, which makes it difficult to meet the monitoring needs of large spaces (such as factories and warehouses) and special environments (such as cold storage and boiler rooms).

[0003] In the 21st century, with the development of sensing technology, photoelectric smoke sensing technology gradually replaced ion smoke sensing technology, and through laser scattering principle, the sensitivity of smoke detection was improved, and some devices added temperature detection function to realize smoke+temperature dual parameter monitoring, but there were still obvious limitations, that is, the monitoring method was still point passive detection, which depended on the diffusion of smoke to the sensor to trigger the alarm, and the alarm time lagged behind the fire occurrence by 3-5 minutes, missing the key window for early disposal. In recent years, the rise of Internet of Things and AI technology has promoted the upgrading of fire-fighting smoke sensing technology to multi-parameter fusion + intelligent decision-making, and some manufacturers have introduced smoke+temperature parameter detection equipment and introduced simple algorithms to optimize the alarm logic, but they have not broken through the core technical bottleneck: first, there is a lack of active detection capability in the field, and most devices still rely on passive sampling, with a monitoring range of only 3-4 meters, which cannot cover large spaces; second, the algorithm design is simple, and most of them use single parameter threshold stacking to judge fire (such as smoke concentration exceeding the standard and temperature rising to alarm), without considering the correlation between parameters (such as the cooperative change trend of smoke and temperature), and the false alarm rate remains at 8%-10%; third, the environmental adaptability is poor, and the threshold is not dynamically adjusted combined with historical environmental data, and in scenes with large fluctuations in temperature and humidity, there are still problems of frequent false alarms or delayed false alarms; fourth, the compatibility and expandability are insufficient, and most devices only support a single communication protocol, making it difficult to access traditional fire-fighting systems, and the algorithm cannot be upgraded remotely, so the performance of the device during its life cycle cannot be optimized.

[0004] Therefore, the existing fire-fighting smoke sensing detection technology has obvious shortcomings in monitoring range, warning timeliness, environmental adaptation, judgment accuracy, and system compatibility, and an intelligent detection scheme with active detection in the field, multi-parameter AI fusion, adaptive environmental learning, and full-scene compatibility is urgently needed to meet the needs of modern intelligent fire-fighting for ultra-early, high-precision, and full-scene warning. SUMMARY

[0005] The present application aims to solve the short board of the existing fire smoke sensing technology in monitoring range, early warning timeliness, environmental adaptation, judgment accuracy and system compatibility, and provides an intelligent fire smoke sensing system and method capable of realizing ultra-early warning, high-precision fire identification, full-scene adaptation and remote upgrading, to meet the needs of intelligent fire protection.

[0006] In one aspect, the present application provides an intelligent fire smoke sensing system based on AI spectral analysis, comprising a surface domain active detection module, a composite sensing fusion module, an intelligent early warning decision module, an adaptive environment learning module and an Internet of Things communication module.

[0007] The surface domain active detection module is used to construct a three-dimensional monitoring field and extract air samples to realize smoke detection, and its output end is connected with the composite sensing fusion module to transmit smoke data.

[0008] The composite sensing fusion module is used to collect environmental temperature data and cooperatively integrate and process smoke data and environmental temperature data, and its output end is connected with the intelligent early warning decision module and the adaptive environment learning module respectively to realize data sharing.

[0009] The intelligent early warning decision module is used to receive the integrated data, calculate the temperature change rate and fire confidence, and trigger graded alarm combined with the alarm threshold.

[0010] The adaptive environment learning module is used to dynamically adjust the alarm threshold based on historical temperature and humidity data through an adaptive environment learning algorithm, and its output end is connected with the intelligent early warning decision module.

[0011] The Internet of Things communication module is used to realize protocol compatibility, data transmission and storage, and remote firmware upgrade, to ensure the system networking and remote management functions.

[0012] In another aspect, the present application provides an intelligent fire smoke sensing method based on AI spectral analysis, comprising the following steps:

[0013] Step one, the surface domain active detection module constructs a three-dimensional monitoring field, extracts air samples, and obtains smoke concentration through double light path detection.

[0014] Step two, the composite sensing fusion module collects environmental temperature, and performs time alignment, abnormality elimination and standardization processing on smoke data and temperature data to form multi-dimensional integrated data.

[0015] Step three, the intelligent early warning decision module calls the temperature change rate algorithm to operate the temperature data in the integrated data to obtain a dynamic temperature change rate result.

[0016] Step four, the intelligent early warning decision module inputs the temperature change rate result, smoke concentration data and data covariance parameters into the multi-parameter fusion model, and obtains the fire confidence through operation;

[0017] Step five, the adaptive environment learning module extracts and integrates the historical temperature and humidity data in the data, calculates the environmental characteristic parameters, and substitutes the adaptive algorithm to obtain the adjusted alarm threshold;

[0018] Step six, the intelligent early warning decision module compares the fire confidence with the adjusted alarm threshold, triggers the corresponding level alarm, transmits the alarm information through the Internet of Things communication module and stores the data.

[0019] Compared with the prior art, the beneficial effects of the present application are:

[0020] 1. The present application relies on the three-dimensional monitoring and double optical path detection of the surface area type active detection module, combines the temperature change rate algorithm and the multi-parameter fusion model, can synchronously capture smoke and temperature, and can advance the alarm time by 3-5 minutes compared with the traditional equipment, and the fire confidence calculation accuracy reaches 98%, which can accurately identify early hidden fire such as smoldering, and can gain key time for personnel evacuation and early fire extinguishing.

[0021] 2. The adaptive environment learning module dynamically adjusts the alarm threshold based on the historical temperature and humidity data, cooperates with the multi-parameter collaborative verification, can adapt to a wide temperature range of-40 DEG C to +85 DEG C and special scenes such as high humidity and dust, effectively solves the false alarm problem of the traditional fixed threshold in the extreme environment, reduces the false alarm rate by more than 85% compared with the prior art, and guarantees the monitoring reliability.

[0022] 3. The Internet of Things communication module supports ModbusRTU / 485, LoRa and other protocols, can seamlessly connect the traditional fire fighting system and the intelligent fire fighting platform, realizes real-time transmission of alarm information, cloud storage and backtracking of historical data, and can remotely optimize the algorithm model through the OTA upgrade function, without the need of on-site modification to improve the equipment performance, and meets the long-term intelligent fire fighting construction demand.

[0023] 4. The system adopts a modular hardware architecture and a flexible networking design, supports 256 devices in cascade, can flexibly select and match modules according to different scenes (residence, factory, cold storage and the like), reduces the customized cost; at the same time, the service life of the sensor is prolonged to 10 years, the equipment replacement frequency is reduced, the whole life cycle operation and maintenance cost is significantly reduced, and the system has wide popularization and application value.

[0024] Other features and advantages of the present application will be set forth in the following description of the application, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the present application can be achieved and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0025] The technical solutions of the present application are described in further detail below with reference to the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0026] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation on the present application. In the drawings:

[0027] Figure 1 is a structural schematic diagram of an intelligent fire smoke sensing detection system based on AI spectrum analysis provided by an embodiment of the present application;

[0028] Figure 2 is a flow schematic diagram of an intelligent fire smoke sensing detection method based on AI spectrum analysis provided by an embodiment of the present application. DETAILED DESCRIPTION

[0029] The preferred embodiments of the present application are described below in combination with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application.

[0030] Example 1:

[0031] An embodiment of the present application provides an intelligent fire smoke sensing detection system based on AI spectrum analysis, please refer to Figure 1 , which comprises a surface domain active detection module, a composite sensing fusion module, an intelligent early warning decision module, a self-adaptive environment learning module and an Internet of Things communication module;

[0032] The surface domain active detection module is used to construct a three-dimensional monitoring field and extract air samples, realize the detection of smoke, and its output end is connected with the composite sensing fusion module to transmit smoke data;

[0033] The surface domain active detection module comprises an array type laser scattering sensor unit, a micro vacuum pump unit and a double light path detection unit.

[0034] The array type laser scattering sensor unit adopts a plurality of groups of laser emitter and receiver arrays to construct a three-dimensional monitoring field within a radius of 5 meters, and real-time capture smoke particle signals. The signal output end thereof is connected with the control end of the micro vacuum pump unit, and the micro vacuum pump unit is triggered to start when the smoke particle signal is detected;

[0035] The micro vacuum pump unit adjusts the air suction rate according to the smoke particle signal strength, and actively extracts air samples in the three-dimensional monitoring field. The sample output end thereof is connected with the double light path detection unit.

[0036] The dual-optical-path detection unit calculates smoke concentration by analyzing the intensity of forward and backward scattered laser light (including but not limited to averaging calculations). The detection signal output of the dual-optical-path detection unit is connected to the composite sensor fusion module.

[0037] The composite sensor fusion module is used to collect ambient temperature data and integrate smoke data and ambient temperature data. Its output is connected to the intelligent early warning decision module and the adaptive environment learning module to achieve data sharing.

[0038] The composite sensing fusion module includes a temperature sensor unit and a data collaborative processing unit.

[0039] The temperature sensor unit collects ambient temperature data in real time, and its output is connected to the data processing unit.

[0040] The data collaborative processing unit is used to perform timestamp alignment, outlier removal, and standardization on smoke data and ambient temperature data to form an integrated dataset. Its output is connected to the intelligent early warning decision module and the adaptive environment learning module, respectively.

[0041] The intelligent early warning and decision-making module is used to receive the integrated data, calculate the temperature change rate and fire confidence level, and trigger graded alarms in combination with alarm thresholds. It communicates bidirectionally with the IoT communication module.

[0042] The intelligent early warning decision module includes a temperature change rate algorithm calculation unit, a multi-parameter fusion model calculation unit, and a hierarchical alarm unit.

[0043] The temperature change rate algorithm calculation unit is used to receive temperature data and calculate the dynamic temperature change rate through the temperature change rate algorithm. Its output is connected to the multi-parameter fusion model calculation unit.

[0044] The multi-parameter fusion model calculation unit is used to receive temperature change rate results and smoke concentration data, calculate fire confidence level through multi-parameter fusion model, and its output is connected to the graded alarm unit.

[0045] The graded alarm unit has preset alarm thresholds. By comparing the fire confidence level with the alarm thresholds, it triggers level one, level two, and level three alarms respectively. It is connected to the Internet of Things communication module to transmit alarm information.

[0046] Specifically, when the fire confidence level C meets... The alarm is classified as a Level 1 alarm when the fire confidence level C is satisfied. The alarm is classified as a Level 2 alarm when the fire confidence level C is satisfied. This is typically classified as a level three alarm. It will not be lower than 0.3 and will not be higher than 0.8. If it is lower than 0.3 or higher than 0.8, the management personnel will be notified to make corrections.

[0047] An adaptive environment learning module is configured to dynamically adjust the alarm threshold based on historical temperature and humidity data by using an adaptive environment learning algorithm, and an output end of the adaptive environment learning module is connected to the intelligent early warning decision module;

[0048] The adaptive environment learning module includes a historical data storage subunit, an environment feature extraction subunit, and a threshold adjustment operation subunit.

[0049] The historical data storage subunit is configured to store historical temperature and humidity data in the past 30-90 days, and support cyclic coverage storage, and an output end of the historical data storage subunit is connected to the environment feature extraction subunit.

[0050] The environment feature extraction subunit is configured to calculate environment feature parameters, and the environment feature parameters include historical temperature average, humidity average, temperature standard deviation, humidity standard deviation, temperature extreme value, humidity extreme value, and temperature and humidity covariance, and an output end of the environment feature extraction subunit is connected to the threshold adjustment operation subunit.

[0051] The threshold adjustment operation subunit is configured to calculate an adjusted alarm threshold based on the environment feature parameters by using an adaptive environment learning algorithm, and an output end of the threshold adjustment operation subunit is connected to the intelligent early warning decision module, and the adjusted alarm threshold is used to replace the original alarm threshold.

[0052] An Internet of Things communication module is configured to realize protocol compatibility, data transmission and storage, and remote firmware upgrade, and guarantee system networking and remote management functions.

[0053] The Internet of Things communication module includes a protocol compatibility unit, a dual-mode data transmission unit, a cloud data storage unit, and an OTA upgrade unit.

[0054] The protocol compatibility unit supports ModbusRTU / 485 and LoRa protocols, and realizes compatible communication with traditional fire fighting systems and Internet of Things gateways.

[0055] The dual-mode data transmission unit adopts 4G+WiFi dual-mode communication, and is configured to transmit alarm information and real-time data.

[0056] The cloud data storage unit is configured to store smoke, temperature historical change curves, and alarm logs, and support data backtracking query.

[0057] The OTA upgrade unit is configured to receive an algorithm model update package pushed by the cloud, and realize remote upgrade of the intelligent early warning decision module and the adaptive environment learning module.

[0058] The above embodiments rely on the three-dimensional monitoring and dual-optical-path detection of the area-based active detection module, combined with a temperature change rate algorithm and a multi-parameter fusion model, to simultaneously capture smoke and temperature. This allows for alarm timing 3-5 minutes earlier than traditional equipment, with a fire confidence calculation accuracy of 98%. It can accurately identify early smoldering and other concealed fires, buying crucial time for personnel evacuation and initial firefighting. The adaptive environment learning module dynamically adjusts the alarm threshold based on historical temperature and humidity data. Combined with multi-parameter collaborative verification, it can adapt to a wide temperature range of -40℃ to +85℃ and special scenarios such as high humidity and dust, effectively solving the false alarm problem of traditional fixed thresholds in extreme environments. The false alarm rate is reduced by more than 85% compared to existing technologies, ensuring monitoring reliability. The IoT communication module supports protocols such as Modbus RTU / 485 and LoRa, seamlessly connecting to traditional fire protection systems and smart fire protection platforms. This enables real-time transmission of alarm information, cloud storage and retrieval of historical data. The OTA upgrade function can also remotely optimize the algorithm model, improving equipment performance without on-site modifications and meeting long-term smart fire protection construction needs. The system adopts a modular hardware architecture and flexible networking design, supporting the cascading of 256 devices. Modules can be flexibly selected according to different scenarios (residential, factory, cold storage, etc.) to reduce customization costs. At the same time, the sensor lifespan is extended to 10 years, reducing the frequency of equipment replacement and significantly reducing the total life cycle maintenance cost, making it widely applicable.

[0059] In one specific embodiment, the temperature change rate algorithm calculation unit uses the following formula for the temperature change rate algorithm:

[0060]

[0061] in, For the temperature change rate results, The number of sampled data sets. The number of steps between adjacent sampling intervals. Let i be the temperature value from the i-th sampling. For the first The temperature value of the second sample. Let i be the sampling time. For the first Sampling time, This represents the average temperature from the previous n samplings. The standard deviation of the temperature from the first n samplings. The sign function outputs a value of 1 when the temperature rises, 0.3 when the temperature falls, and 0.1 when the temperature remains constant.

[0062] in, Representing the dynamic temperature change rate, its core function is to reflect the trend of ambient temperature change over time, and it is a key indicator for judging the temperature characteristics in the early stage of a fire. Its physical dimension is "℃ / s", and its value can be positive or negative. Positive values ​​correspond to temperature rise (the core characteristic in a fire scenario), and negative values ​​correspond to temperature drop. The larger the absolute value, the faster the temperature change rate, which can intuitively reflect the intensity of temperature change. The total number of temperature sampling data sets used in the temperature change rate calculation is a fundamental parameter to ensure calculation accuracy. This parameter is dimensionless and is a positive integer. In practical applications, it is usually set according to the system sampling frequency. For example, when the sampling frequency is 1 time / second, n is generally taken as 20-40 sets, which can cover a temperature change cycle of 20-40 seconds and avoid calculation deviations caused by too few sampling sets. This represents the "interval step" when calculating the temperature difference, i.e., selecting the i-th sample data and the i-th sample data. The temperature difference is compared using the sampled data. By using interval sampling, the interference of single sampling fluctuations on the temperature change rate calculation is reduced. This parameter has no unit, is a positive integer, and its value range is usually 1-6 steps. For example, when k=2, the i-th sampling data is compared with the i-2-th sampling data, which can achieve a balance between real-time performance and anti-interference. It is the raw ambient temperature data collected by the system at the i-th sampling time, which is the basic input for calculating the temperature change rate; its physical dimension is "℃", and it is collected by the temperature sensor unit in the composite sensing fusion module, which has high sampling accuracy and can ensure the accuracy of temperature data. The system in the first The ambient temperature data collected at each sampling time, and Used together, they calculate the temperature difference between adjacent time intervals, serving as a benchmark reference value in temperature difference calculations; their physical dimension is also "℃", and their data sources and accuracy standards are the same. Consistency is ensured to guarantee the reliability of temperature difference calculations; The timestamp corresponding to the i-th temperature sampling is used to calculate the interval between two samplings; its physical dimension is "s" (second), and it is generated by the system's unified clock module. The timestamp accuracy error is ≤1ms to ensure the accuracy of the sampling interval calculation and avoid the influence of time deviation on the temperature change rate result. It is the first The timestamp corresponding to the next temperature sampling, and In conjunction with the calculation of the sampling interval, a time dimension reference is provided for the calculation of the temperature change rate; its physical dimension is "s", and it adopts the same... Using the same time base ensures the consistency of data across sampling intervals; The arithmetic mean of the first n sets of temperature sampling data is obtained by the following formula: Its core function is to correct the interference of extreme temperature values ​​on the calculation of temperature change rate; its physical dimension is "℃", which can reflect the average temperature level of the recent environment and serve as a reference benchmark for judging whether the temperature of a single sampling deviates from the normal range. Through formula The calculation shows that the dispersion of the first n sets of temperature sampling data is reflected, which indirectly reflects the stability of the ambient temperature; its physical dimension is "℃", the larger the value, the more violent the fluctuation of the ambient temperature, and the smaller the value, the more stable the ambient temperature. The time-weighted temperature difference term has the numerator being the temperature difference between the i-th and ik-th samples, and the denominator being the square root of the time interval between the two samples. This overall approach achieves time-weighted temperature difference calculation. This design can reduce the impact of instantaneous temperature fluctuations in a short period on the temperature change rate calculation. For example, when the interval between two samples is short, the denominator is small, which can amplify the weight of the temperature difference; when the interval is long, the denominator is large, which can appropriately reduce the weight of the temperature difference, making the calculation results more consistent with the actual temperature change trend. The temperature deviation correction term corrects the absolute value of the temperature difference in the form of an exponential function. When the absolute value of a single temperature difference is large (which may be an extreme outlier), the value of this exponential term becomes smaller, reducing its contribution to the overall temperature change rate calculation result, thereby suppressing the interference of extreme temperature values ​​and ensuring the stability of the calculation result. For function correction terms, through The deviation of a single sampled temperature from the average value is calculated (based on the standard deviation), then converted into a value in the 0-1 range, and finally added to 1 to obtain a correction coefficient in the 1-2 range. This correction term can achieve non-linear correction of the degree of temperature deviation from the average value. When the temperature deviation is large, the correction coefficient approaches 2, amplifying the impact of this sampling on the temperature change rate; when the deviation is small, the correction coefficient approaches 1, reducing the impact and making the calculation results more reflective of abnormal temperature changes. For the sign function term, different weights are output according to the direction of temperature change. The specific logic is as follows: when When the temperature rises, output 1; when When the temperature drops, the output is 0.3; when... When the temperature remains constant, the output is 0.1. The core purpose of this design is to highlight the impact of temperature rise (the core feature of fire scenarios) on the temperature change rate results, while weakening the impact of temperature drop and no change, so as to ensure that the temperature change rate results can accurately point to the temperature change trend related to the fire.

[0063] The temperature change rate algorithm formula achieves accurate calculation of dynamic temperature change rate through multi-stage correction and weighted integration. Its core principle can be divided into three steps:

[0064] The first step is to calculate the relationship between the baseline temperature difference and time: through... Calculate the temperature difference between adjacent intervals, and combine time-weighted, to obtain a basic temperature difference term associated with the time dimension, which preliminarily reflects the change of temperature with time;

[0065] Second step, multi-dimensional deviation correction: introduce temperature deviation correction term (exponential function) and function correction term, respectively suppress extreme temperature difference and temperature deviation from the average interference, ensure that each sampling data on the contribution of temperature change rate is more in line with the actual environmental characteristics, avoid the calculation deviation caused by abnormal data;

[0066] Third step, direction weight and mean integration: give different weights according to the direction of temperature change through the sign function, highlight the influence of temperature rise, and then take the average value of the calculation results of n groups of sampling data, finally get the dynamic temperature change rate result which can truly reflect the trend of environmental temperature change, and provide accurate temperature change index for subsequent fire confidence calculation.

[0067] In a specific embodiment, the formula of the multi-parameter fusion model used by the multi-parameter fusion model operation unit is:

[0068]

[0069] wherein, the fire confidence is 0-1, the preset weight coefficient and , the actual smoke concentration, the maximum detectable smoke concentration, the temperature change rate result is the temperature change rate result, the maximum temperature change rate threshold, the covariance of smoke concentration and temperature, the smoke concentration standard deviation, the temperature standard deviation;

[0070] wherein, (representing fire confidence, which is the core quantitative index for judging whether to trigger alarm and alarm level, the value range is 0-1; the closer the value is to 1, the higher the probability that the current environment meets the fire characteristics; the closer the value is to 0, the higher the probability that the current environment is in a normal state, which provides a clear basis for grading alarm; all are preset weight coefficients, which give different influence weights to the four parameters of smoke concentration, temperature change rate, smoke-temperature covariance; in actual application, they can be flexibly adjusted according to different scenes (such as residence, factory, warehouse), for example, in the residential scene, the influence of smoke concentration is more critical, the value of can be increased, usually the value range is 0.35-0.55, 0.15-0.25, 0.15-0.25, It is 0.05-0.15; The actual smoke concentration in the current environment detected by the system through the area-based active detection module is one of the core and intuitive indicators for judging the fire situation. Its physical dimension is "mg / m³", which is collected and calculated by the forward scattering detection subunit in the dual-optical-path detection unit and can directly reflect the smoke content in the environment. The maximum smoke concentration detection limit designed for the system is a benchmark value that standardizes the actual smoke concentration; its physical dimension is "mg / m³", which is determined by the system hardware performance and represents the upper limit of the smoke concentration range that the system can accurately detect. If this value is exceeded, the system can still output an alarm, but the concentration data may be outside the accurate detection range. The calculation result of the above temperature change rate algorithm formula reflects the rate of change of the current ambient temperature, which is an important parameter for judging the early characteristics of a fire. Its physical dimension is "℃ / s". A positive value and a larger value indicate that the temperature rises faster and is more consistent with the temperature change characteristics in the early stage of a fire. The maximum temperature change rate threshold preset for the system is a benchmark value for standardizing the temperature change rate results; its physical dimension is "℃ / s". It is usually set according to the temperature change pattern of fire in different scenarios. For example, in a warehouse scenario with a lot of flammable materials, the threshold can be appropriately increased, while in an ordinary residential scenario, it can be appropriately decreased. It is used to determine whether the current temperature change rate is within the abnormal range related to the fire. The covariance of smoke concentration and temperature is calculated using statistical methods, reflecting the degree of linear correlation between the two parameters; its calculation formula is as follows: (in This represents the average smoke concentration from the previous n measurements. (This is the average temperature of the previous n times). A positive value indicates a positive correlation between the two (common in fire scenarios, where temperature rises as smoke concentration increases), while a negative value indicates a negative correlation. The larger the absolute value, the stronger the correlation. The formula reflects the dispersion of the first n smoke concentration sampling data and is as follows: Its physical dimensions are consistent with those of smoke concentration (mg / m³). The larger the value, the more drastic the fluctuation of smoke concentration, and the smaller the value, the more stable the smoke concentration. With the formula for the rate of temperature change The meaning is consistent, referring to the dispersion of the first n temperature sampling data, with the physical dimension "℃", reflecting the stability of the ambient temperature; For the smoke concentration standardization term, the actual smoke concentration S is compared with the maximum detectable concentration. By analogy, the smoke concentration is converted into a standardized value in the 0-1 range, eliminating the influence of differences in the absolute value of smoke concentration under different scenarios, thus making the smoke concentration data from different scenarios comparable. For example, in At this time, the standardization term is 0.5, representing that the current smoke concentration reaches 50% of the maximum detection value of the system; For the temperature change rate standardization term, the temperature change rate result is compared with the maximum temperature change rate threshold , the temperature change rate is converted into a standardized value in the 0-1 interval, and the influence of the absolute value difference of the temperature change rate in different scenes is also eliminated, so that the temperature change rate data can be compared across scenes; For the temperature change rate nonlinear enhancement term, the standardized temperature change rate is nonlinearly enhanced in the form of a sine function. When the standardized temperature change rate is 0, the sine term is 0; when the standardized temperature change rate is 1 (i.e., the temperature change rate reaches the maximum threshold), the sine term is ; when the standardized temperature change rate is between 0 and 1, the sine term nonlinearly rises with the increase of the temperature change rate; this design can amplify the influence of the temperature change rate on the fire confidence, especially when the temperature change rate is close to the maximum threshold, the enhancement effect is more significant, and the feature of rapid temperature rise under fire is more reflected; The smoke concentration-temperature change rate correlation term is established in the form of an exponential function to establish the correlation between smoke concentration and temperature change rate. When the temperature change rate rises, the exponential term value increases, thereby amplifying the contribution of smoke concentration to fire confidence; when the temperature change rate is low, the exponential term value tends to 1, and the contribution of smoke concentration remains relatively stable; the core logic of this design is that under fire, the temperature change rate and smoke concentration usually rise synchronously, and through the exponential term, the synergistic effect of the two can be strengthened to improve the accuracy of fire judgment; The smoke-temperature correlation term is essentially the correlation coefficient (value range: -1-1) of smoke concentration and temperature, which can reflect the synchronicity of smoke concentration and temperature change. Under the fire scene, the two are usually positively correlated (correlation coefficient close to 1), while under the environmental interference (such as smoke sensor false triggering) scene, the correlation between the two is low (correlation coefficient close to 0 or negative), and through this parameter, the authenticity of the fire can be further verified to reduce the false alarm probability;

[0071] The multi-parameter fusion model formula realizes the accurate calculation of fire confidence through the logic of dimension quantization, synergistic enhancement, and correlation verification, and the core principle can be divided into four steps:

[0072] First, the smoke concentration and the temperature change rate are standardized in the 0-1 interval to eliminate the influence of different parameter dimensions and absolute value differences;

[0073] Second, for the smoke concentration dimension, the nonlinear enhancement term (sine function) of the temperature change rate is combined to amplify the synergistic effect of the temperature change rate on the smoke concentration in judging fire; for the smoke concentration dimension, the temperature change rate is associated through the exponential term to strengthen the synchronous change feature of the two under fire, so that the fire indication effect of a single parameter is more accurate;

[0074] Third step, introduce smoke-temperature correlation coefficient From the perspective of parameter correlation, verify the authenticity of the fire, and exclude false positives caused by single parameter false triggering. For example, when the smoke concentration increases but is not positively correlated with the temperature, the correlation term value is low, which can reduce the overall fire confidence and avoid false positives;

[0075] Fourth step, weight fusion calculation: the results of the above standardization processing, synergistic enhancement, and correlation verification are weighted and summed through the weight coefficient to finally obtain the fire confidence C in the interval of 0-1, realizing the organic fusion of multi-dimensional parameters and ensuring the comprehensiveness and accuracy of fire judgment;

[0076] In one specific embodiment, the adaptive environmental learning algorithm formula used by the threshold adjustment operator unit is:

[0077]

[0078] wherein, is the adjusted alarm threshold, with a value range of 0-1, is the basic alarm threshold, is the number of historical temperature and humidity data sets, is the jth historical humidity value, is the historical humidity average value, is the historical humidity maximum and minimum value, is the jth historical temperature value, is the historical temperature average value, is the historical temperature maximum and minimum value, is the historical temperature standard deviation and historical humidity standard deviation, is the covariance of temperature and humidity.

[0079] wherein, represents the adjusted alarm threshold, which is the new judgment basis for the intelligent early warning decision module to trigger graded alarms, with a value range consistent with the fire confidence C (0-1); this threshold can dynamically change according to historical environmental characteristics, for example, in high humidity and large temperature fluctuation environments, it will be appropriately increased to avoid false positives caused by environmental interference; in stable environment scenarios, it can be appropriately reduced to improve the sensitivity of fire detection; is the initial alarm threshold preset by the system, which is the reference value for dynamic adjustment, usually with a value range of 0.5-0.7; its value is set based on the fire characteristics of common scenarios, for example, in ordinary residential scenarios, Can be set to 0.6, which represents triggering the corresponding level alarm when the fire confidence C reaches 0.6, providing an initial reference for subsequent adaptive adjustment; m (referring to the total number of historical temperature and humidity data groups participating in adaptive calculation, which is a basic parameter to ensure the accuracy of environmental feature analysis; the parameter has no unit and is a positive integer, with a value range of 150-600 groups, which can cover historical data of nearly 30-90 days (according to the system sampling frequency setting), ensuring that the long-term environmental characteristics of the application scenario can be fully reflected; refers to the jth historical humidity sampling data stored by the system, which is the basic raw data for analyzing the environmental humidity characteristics; the historical data collected by the composite sensor fusion module and stored in the adaptive environment learning module's historical data storage subunit reflect the environmental humidity state at a certain time in history; is the arithmetic mean of the m historical humidity data, representing the long-term average humidity level of the application scenario, which is the basis for judging whether a single historical humidity deviates from the normal range, and is also a reference for the standardization processing of humidity-related parameters; are the maximum and minimum values of the m historical humidity data, respectively, representing the extreme range of humidity in the application scenario, which is used to standardize the degree of deviation of a single historical humidity from the average value and eliminate the influence of the absolute value difference of humidity in different scenarios; refers to the jth historical temperature sampling data stored by the system, which is the basis for analyzing the environmental temperature characteristics; corresponding, is the basic raw data for analyzing the environmental temperature characteristics, and the collection source and storage method are consistent with ; it reflects the environmental temperature state at a certain time in history; is the arithmetic mean of the m historical temperature data, representing the long-term average temperature level of the application scenario, which is the basis for standardization processing and deviation correction of temperature-related parameters; are the maximum and minimum values of the m historical temperature data, respectively, representing the extreme range of temperature in the application scenario, which is used to standardize the degree of deviation of a single historical temperature from the average value, together with to form a reference for environmental extreme characteristics; reflects the dispersion degree of the m historical temperature data, and the larger the value, the more intense the historical temperature fluctuation, and the poorer the environmental temperature stability; the smaller the value, the more stable the environmental temperature; reflects the dispersion degree of the m historical humidity data, and the larger the value, the more intense the historical humidity fluctuation, and the poorer the environmental humidity stability; the smaller the value, the more stable the environmental humidity; reflects the linear correlation between historical temperature and humidity data, with the calculation formula being ; a positive value indicates a positive correlation, a negative value indicates a negative correlation, and the larger the absolute value, the stronger the correlation, which can reflect the collaborative characteristics of temperature and humidity changes in the application scenario, such as in the rainy season scenario, where temperature decreases and humidity increases, and the two may be negatively correlated; For the humidity deviation standardization term, the numerator is the deviation of the jth historical humidity from the average value, and the denominator is the difference between the maximum and minimum historical humidity values. The humidity deviation is standardized to a value in the interval of -1-1. When the value is positive, it represents that the humidity is higher than the average level, and vice versa. It can intuitively reflect the degree of single humidity deviation from the normal range; For the temperature deviation standardization term, the logic is consistent with the humidity deviation standardization term. The numerator is the deviation of the jth historical temperature from the average value, and the denominator is the difference between the maximum and minimum historical temperature values. The temperature deviation is standardized to a value in the interval of -1-1, which is used to reflect the degree of single temperature deviation from the normal range; For the temperature and humidity deviation correction term, the sum of the absolute values of the temperature and humidity deviations is corrected in the form of an exponential function, where is the sum of the average temperature and humidity. When the sum of the absolute values of the single temperature and humidity deviations is large (possibly an extreme abnormal value), the value of this exponential term becomes small, reducing the impact of this data on the overall adjustment. When the deviation is small, the value of the exponential term tends to 1, increasing the impact, thereby suppressing the interference of extreme historical data and ensuring that the adjustment result is based on normal environmental characteristics; For the environmental stability and correlation correction term, it contains three sub-terms, which together achieve fine correction of the alarm threshold, i.e. For the environmental stability correction sub-term, and are the coefficients of variation (relative dispersion) of temperature and humidity, respectively. The larger the value, the worse the environmental stability. This sub-term is negative, and the smaller the value of this sub-term, the smaller the overall exponential term value, which can appropriately reduce the adjustment range of the alarm threshold to avoid frequent changes in the threshold due to environmental fluctuations; For the temperature and humidity correlation correction sub-term, is the correlation coefficient of temperature and humidity. After being multiplied by 0.1, it has a small impact on the overall exponential term. Its main function is to fine-tune the correction range based on the correlation between temperature and humidity. For example, when they are strongly positively correlated, this sub-term is positive, which can slightly increase the value of the exponential term, making the threshold adjustment more consistent with the environmental characteristics;

[0080] This adaptive environmental learning algorithm achieves adaptive optimization of the alarm threshold through historical feature analysis, multi-dimensional correction, and dynamic adjustment logic. The core principle can be divided into three steps:

[0081] First, standardize the m sets of historical temperature and humidity data (humidity and temperature deviation standardization terms), suppress extreme data interference with the deviation correction term, and then calculate the mean value to obtain the mean value reflecting the long-term environmental temperature and humidity deviation characteristics, which provides the basic direction for threshold adjustment. If the mean value is positive, it tends to increase the threshold, and if the mean value is negative, it tends to decrease the threshold;

[0082] Second step, by environmental stability modifier Adjust the threshold adjustment range according to the stability degree of the historical environment to avoid frequent fluctuations of the threshold in unstable environment; by the temperature and humidity correlation modifier , combined with the characteristics of temperature and humidity, the correction range is fine-tuned to make the threshold adjustment more in line with the scene environment law;

[0083] Third step, taking the basic alarm threshold As the reference, the historical temperature and humidity deviation from the mean term is multiplied by the environmental stability-correlation correction term (exponential form) to obtain the adjusted alarm threshold ; this process realizes dynamic adaptation based on historical environmental characteristics, ensuring that the alarm threshold can be optimized in real time with the change of the environment in the application scene, avoiding false positives / misses of fixed thresholds in extreme environments, and ensuring the sensitivity of fire detection, ultimately improving the adaptability and reliability of the system in different scenes;

[0084] Embodiment 2:

[0085] The embodiment of the application provides an intelligent fire-fighting smoke sensing detection method based on AI spectrum analysis, please refer to Figure 2 , the method is based on the system of embodiment 1, including the following steps:

[0086] Step one, the surface area type active detection module constructs a three-dimensional monitoring field, extracts air samples, and obtains smoke concentration through double light path detection;

[0087] Step two, the composite sensing fusion module collects environmental temperature, and performs time alignment, abnormal rejection and standardization processing on smoke data and temperature data to form multi-dimensional integrated data;

[0088] Step three, the intelligent early warning decision module calls the temperature change rate algorithm to operate the temperature data in the integrated data to obtain a dynamic temperature change rate result;

[0089] Step four, the intelligent early warning decision module inputs the temperature change rate result, smoke concentration data and data covariance parameters into a multi-parameter fusion model to obtain a fire confidence through operation;

[0090] Step five, the self-adaptive environment learning module extracts historical temperature and humidity data from the integrated data, calculates environmental characteristic parameters, and inputs the self-adaptive algorithm to obtain an adjusted alarm threshold;

[0091] Step six, the intelligent early warning decision module compares the fire confidence with the adjusted alarm threshold to trigger an alarm of corresponding level, transmits alarm information through the Internet of Things communication module and stores data.

[0092] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent fire smoke detection system based on AI spectral analysis, characterized in that, It includes a surface-area active detection module, a composite sensor fusion module, an intelligent early warning and decision-making module, an adaptive environment learning module, and an IoT communication module; The area-based active detection module is used to construct a three-dimensional monitoring field and extract air samples to detect smoke. Its output end is connected to the composite sensor fusion module to transmit smoke data. The composite sensing fusion module is used to collect ambient temperature data, and to collaboratively integrate and process smoke data and ambient temperature data. Its output is connected to the intelligent early warning decision module and the adaptive environment learning module to achieve data sharing. The intelligent early warning decision module is used to receive the integrated data, calculate the temperature change rate and fire confidence level, and trigger graded alarms in combination with alarm thresholds. It communicates bidirectionally with the Internet of Things communication module. The adaptive environment learning module is used to dynamically adjust the alarm threshold based on historical temperature and humidity data through an adaptive environment learning algorithm, and its output is connected to the intelligent early warning decision module. The IoT communication module is used to achieve protocol compatibility, data transmission and storage, and remote firmware upgrade, ensuring system networking and remote management functions; The area-based active detection module includes an array-type laser scattering sensor unit, a miniature vacuum pump unit, and a dual-optical-path detection unit. The array-type laser scattering sensor unit uses multiple laser emitter and receiver arrays to construct a three-dimensional monitoring field within a radius of 5 meters, capturing smoke particle signals in real time. Its signal output terminal is connected to the control terminal of the micro vacuum pump unit, and the micro vacuum pump unit is triggered to start when smoke particle signals are detected. The miniature vacuum pump unit adjusts the pumping rate according to the signal intensity of smoke particles and actively extracts air samples in the three-dimensional monitoring field. Its sample output end is connected to the dual optical path detection unit. The dual-optical-path detection unit calculates the smoke concentration by analyzing the intensity of forward and backward scattered laser light, and the detection signal output terminal of the dual-optical-path detection unit is connected to the composite sensor fusion module.

2. The intelligent fire smoke detection system based on AI spectral analysis according to claim 1, characterized in that, The composite sensing fusion module includes a temperature sensor unit and a data collaborative processing unit. The temperature sensor unit collects ambient temperature data in real time, and its output is connected to the data processing unit. The data collaborative processing unit is used to perform timestamp alignment, outlier removal, and standardization on smoke data and ambient temperature data to form an integrated dataset. Its output is connected to the intelligent early warning decision module and the adaptive environment learning module, respectively.

3. The intelligent fire smoke detection system based on AI spectral analysis according to claim 1, characterized in that, The intelligent early warning decision-making module includes a temperature change rate algorithm calculation unit, a multi-parameter fusion model calculation unit, and a hierarchical alarm unit. The temperature change rate algorithm calculation unit is used to receive temperature data and calculate the dynamic temperature change rate through the temperature change rate algorithm. Its output is connected to the multi-parameter fusion model calculation unit. The multi-parameter fusion model calculation unit is used to receive temperature change rate results and smoke concentration data, calculate fire confidence level through multi-parameter fusion model, and its output is connected to the graded alarm unit. The graded alarm unit is preset with alarm thresholds. By comparing the fire confidence level with the alarm thresholds, it triggers level one, level two, and level three alarms respectively. It is connected to the Internet of Things communication module to transmit alarm information.

4. The intelligent fire smoke detection system based on AI spectral analysis according to claim 1, characterized in that, The adaptive environment learning module includes a historical data storage subunit, an environmental feature extraction subunit, and a threshold adjustment calculation subunit. The historical data storage subunit is used to store historical temperature and humidity data for the past 30-90 days, supports cyclic overwrite storage, and its output is connected to the environmental feature extraction subunit. The environmental feature extraction subunit is used to calculate environmental feature parameters, which include historical average temperature, average humidity, standard deviation of temperature, standard deviation of humidity, extreme temperature, extreme humidity, and temperature-humidity covariance. Its output is connected to the threshold adjustment calculation subunit. The threshold adjustment operation subunit is used to calculate the adjusted alarm threshold based on environmental feature parameters through an adaptive environment learning algorithm. Its output is connected to the intelligent early warning decision module, and the adjusted alarm threshold replaces the original alarm threshold.

5. The intelligent fire smoke detection system based on AI spectral analysis according to claim 1, characterized in that, The IoT communication module includes a protocol compatibility unit, a dual-mode data transmission unit, a cloud data storage unit, and an OTA upgrade unit; The protocol-compatible unit supports Modbus RTU / 485 and LoRa protocols, enabling compatible communication with traditional fire protection systems and IoT gateways; The dual-mode data transmission unit adopts 4G+WiFi dual-mode communication to transmit alarm information and real-time data; The cloud data storage unit is used to store historical change curves of smoke and temperature, as well as alarm logs, and supports data backtracking and querying. The OTA upgrade unit is used to receive algorithm model update packages pushed from the cloud, enabling remote upgrades of the intelligent early warning decision module and the adaptive environment learning module algorithms.

6. The intelligent fire smoke detection system based on AI spectral analysis according to claim 3, characterized in that, The formula for the temperature change rate algorithm used by the temperature change rate algorithm calculation unit is as follows: ； in, For the temperature change rate results, The number of sampled data sets. The number of steps between adjacent sampling intervals. Let i be the temperature value from the i-th sampling. For the first The temperature value of the second sample. Let i be the sampling time. For the first Sampling time, This represents the average temperature from the previous n samplings. The standard deviation of the temperature from the first n samplings. The sign function outputs a value of 1 when the temperature rises, 0.3 when the temperature falls, and 0.1 when the temperature remains constant. The multi-parameter fusion model used by the multi-parameter fusion model computation unit is as follows: ； in, The fire confidence level ranges from 0 to 1. For preset weighting coefficients and , This represents the actual smoke concentration. To the maximum detectable smoke concentration, For the temperature change rate results, The maximum temperature change rate threshold, Let be the covariance of smoke concentration and temperature. The standard deviation of smoke concentration. This represents the temperature standard deviation.

7. The intelligent fire smoke detection system based on AI spectral analysis according to claim 4, characterized in that, The formula for the adaptive environment learning algorithm used by the threshold adjustment operation subunit is as follows: ； in, The adjusted alarm threshold ranges from 0 to 1. Basic alarm threshold, The number of historical temperature and humidity data sets. Let j be the historical humidity value. This represents the historical average humidity. These are the historical maximum and minimum humidity values. Let j be the historical temperature value. This is the historical average temperature. These are the historical maximum and minimum temperatures. The historical temperature standard deviation and historical humidity standard deviation are given. Let be the covariance of temperature and humidity.

8. A method for intelligent fire smoke detection based on AI spectral analysis, based on the intelligent fire smoke detection system according to any one of claims 1-7, characterized in that, Includes the following steps: Step 1: The area-based active detection module constructs a three-dimensional monitoring field, extracts air samples, and obtains the smoke concentration through dual-optical-path detection; Step 2: The composite sensor fusion module collects ambient temperature data and performs time alignment, anomaly removal, and standardization on smoke and temperature data to form multi-dimensional integrated data. Step 3: The intelligent early warning decision module calls the temperature change rate algorithm to calculate the temperature data in the integrated data and obtain the dynamic temperature change rate result. Step four: The intelligent early warning decision module inputs the temperature change rate results, smoke concentration data and data covariance parameters into the multi-parameter fusion model to calculate the fire confidence level; Step 5: The adaptive environment learning module extracts historical temperature and humidity data from the integrated data, calculates environmental characteristic parameters, and substitutes them into the adaptive algorithm to obtain the adjusted alarm threshold. Step six: The intelligent early warning decision module compares the fire confidence level with the adjusted alarm threshold, triggers the corresponding level of alarm, and transmits the alarm information and stores the data through the IoT communication module.

Images