CN-122024432-A - Smoke anomaly monitoring method based on sensing data of Internet of things

CN122024432ACN 122024432 ACN122024432 ACN 122024432ACN-122024432-A

Abstract

The invention discloses a smoke anomaly monitoring method based on sensing data of the Internet of things, and particularly relates to the field of intelligent security and fire disaster early warning, which comprises the steps of collecting multidimensional environmental data through deployed sensor nodes of the Internet of things, preprocessing at an edge end, extracting core discrimination characteristics, constructing a feature vector and uploading the feature vector to a cloud; the method comprises the steps of fusing characteristics by a cloud, inputting a pre-trained smoke anomaly discrimination model to calculate anomaly probability, outputting a discrimination result, dynamically adjusting a discrimination threshold based on environmental parameters and historical data, restraining false alarm by combining time consistency check, determining a final smoke anomaly state, performing hierarchical response from high-frequency sampling, platform early warning to local warning, remote notification and multi-system linkage according to a final state matching early warning level, and performing online self-optimization and iterative updating on the discrimination model based on newly added data. The invention combines multidimensional sensing and intelligent analysis, and improves the accuracy, early warning capability and environmental adaptability of smoke monitoring.

Inventors

LIU ZELONG
LIU BINGHUI
ZHU JIANG
REN XINBO

Assignees

峥峰数防科技(山西)股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (9)

1. The smoke anomaly monitoring method based on the sensing data of the Internet of things is characterized by comprising the following steps of: S1, acquiring multidimensional environmental parameter data comprising smoke concentration, temperature, humidity, carbon monoxide concentration, volatile organic compound concentration and fine particulate matter concentration through an Internet of things sensor node deployed in a monitoring area; s2, receiving the feature vector uploaded by the edge end, performing integrity verification, then performing self-adaptive weighted feature fusion, inputting the fused feature into a pre-trained smoke anomaly discrimination model, calculating anomaly probability and outputting a smoke anomaly state discrimination result; s3, dynamically calculating and adjusting an abnormal judgment threshold value based on the current environmental parameters and the historical feature vector data, and carrying out false alarm suppression on the judgment result by combining time consistency verification to output a final smoke abnormal state; s4, according to the final smoke abnormal state matching corresponding early warning grade, performing grading response measures including edge end sampling frequency adjustment, platform early warning pushing, local audible and visual alarm, remote notification and fire extinguishing system linkage; And S5, based on the newly added monitoring data and the manual labeling result, constructing an increment training set to perform online increment learning on the judging model, optimizing the feature weight and the judging threshold value, and completing iterative updating and deployment of the smoke anomaly judging model.
2. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1 is characterized in that the construction of the feature vector comprises the steps of extracting and calculating five core distinguishing features including at least a smoke concentration standardized value, a smoke concentration instantaneous change rate, a temperature and humidity correction coefficient for eliminating high-humidity environment interference, a multi-gas synergistic feature for fusing carbon monoxide concentration and volatile organic matter concentration and a standardized particulate matter concentration based on the standardized multi-dimensional sensing data, and integrating the five core distinguishing features to form a five-dimensional feature vector.
3. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1, wherein the smoke anomaly discrimination model comprises a discrimination model trained based on LightGBM machine learning algorithm, wherein the model is input into a matrix formed by combining the fused features and the original feature vectors, and is output into a probability value representing the possibility of smoke anomaly.
4. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1 is characterized in that the calculation of anomaly probability comprises the steps of inputting an input matrix generated by feature fusion into a smoke anomaly discrimination model for reasoning calculation to obtain a smoke anomaly probability value at the current moment, and mapping the probability value into a specific smoke anomaly state based on a preset fixed threshold or a dynamically adjusted discrimination threshold.
5. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1, wherein the smoke anomaly state discrimination results comprise a normal state, a suspected smoke anomaly state and a smoke anomaly state confirmation, and the state discrimination basis is a comparison result of the anomaly probability value and a preset or dynamic discrimination threshold value.
6. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1, wherein the anomaly discrimination threshold comprises a basic threshold dynamically calculated based on a weighted average of historical environmental features and a final dynamic threshold after floating correction in combination with whether the current real-time environment is a high humidity or high dust scene.
7. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1, wherein the final smoke anomaly state comprises the steps of further introducing time consistency check on the basis of comparing the current anomaly probability with a dynamic threshold value, judging that the smoke anomaly state is valid only when a plurality of continuous sampling periods meet anomaly conditions, and otherwise correcting the smoke anomaly state to be normal and marking the smoke anomaly state as potential false alarm.
8. The smoke anomaly monitoring method based on the sensing data of the internet of things according to claim 1, wherein the early warning level comprises a first early warning corresponding to a normal state, a second early warning corresponding to a suspected smoke anomaly state, and a third early warning corresponding to a confirmation smoke anomaly state.
9. The smoke anomaly monitoring method based on the sensing data of the Internet of things according to claim 1 is characterized in that the step response measures comprise the steps of maintaining a conventional monitoring flow during primary early warning, lifting edge end sampling frequency and pushing yellow early warning information to a monitoring platform during secondary early warning, triggering local audible and visual warning, pushing warning information to a remote terminal and starting a fire-fighting smoke discharging, spraying and emergency channel control system in parallel during tertiary early warning.

Description

Smoke anomaly monitoring method based on sensing data of Internet of things Technical Field The invention relates to the technical field of intelligent monitoring and fire disaster early warning, in particular to a smoke anomaly monitoring method based on sensing data of the Internet of things. Background Smoke monitoring and fire early warning are key technologies for guaranteeing the life and property safety of personnel. Currently, common smoke monitoring approaches rely primarily on freestanding smoke detectors mounted in specific locations. Such detectors are typically based on photoelectric or ion principles, with single-dimensional detection of smoke particle concentration in the air, triggering a local audible and visual alarm when the concentration exceeds a fixed threshold. However, when the detector is actually used, the detector still has some defects, such as frequent false alarm caused by interference of non-fire particles such as water vapor, cooking fume, dust and the like in the environment, influence on normal production and living order, possibly cause a 'wolf' effect, reduce the trust degree of people on alarms, only rely on a single smoke concentration parameter, are difficult to effectively identify in the very early stage of fire, have high missed alarm risk, cannot meet the early warning requirement of high-safety places, adopt a fixed detection threshold value, cannot be adaptively adjusted according to environmental changes, have unstable detection performance in different environments, and mainly provide local warning after alarm, cannot carry out intelligent linkage and cooperative response with a fire protection system, a security protection system and a remote management platform, and have low emergency disposal efficiency. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a smoke anomaly monitoring method based on sensing data of the Internet of things, and solves the problems in the background art through the following scheme. In order to achieve the above purpose, the invention provides a smoke anomaly monitoring method based on sensing data of the Internet of things, which comprises the following steps: S1, acquiring multidimensional environmental parameter data comprising smoke concentration, temperature, humidity, carbon monoxide concentration, volatile organic compound concentration and fine particulate matter concentration through an Internet of things sensor node deployed in a monitoring area; s2, receiving the feature vector uploaded by the edge end, performing integrity verification, then performing self-adaptive weighted feature fusion, inputting the fused feature into a pre-trained smoke anomaly discrimination model, calculating anomaly probability and outputting a smoke anomaly state discrimination result; s3, dynamically calculating and adjusting an abnormal judgment threshold value based on the current environmental parameters and the historical feature vector data, and carrying out false alarm suppression on the judgment result by combining time consistency verification to output a final smoke abnormal state; s4, according to the final smoke abnormal state matching corresponding early warning grade, performing grading response measures including edge end sampling frequency adjustment, platform early warning pushing, local audible and visual alarm, remote notification and fire extinguishing system linkage; And S5, based on the newly added monitoring data and the manual labeling result, constructing an increment training set to perform online increment learning on the judging model, optimizing the feature weight and the judging threshold value, and completing iterative updating and deployment of the smoke anomaly judging model. The invention has the technical effects and advantages that: 1. According to the invention, through fusing multidimensional sensing data of smoke, temperature, humidity, various gases and particulate matters, a more comprehensive environmental situation portrait is constructed, and a cloud intelligent model is utilized for comprehensive discrimination, so that the defect that a traditional single sensor is easy to be interfered is overcome; 2. By extracting and fusing early-stage characteristic gas information of fires such as carbon monoxide, volatile organic compounds and the like and analyzing by combining dynamic change trend of smoke concentration, the system can find out potential fire conditions at the extremely early stage that traditional smoke detectors such as smoldering cannot respond, thereby gaining precious time for emergency treatment and reducing the risk of missing report; 3. the system not only supports preset parameters according to different scenes such as industry, warehouse, building and the like, but also continuously adapts to the change of specific environments through dynamic threshold adjustment and on-line self-optimization functions of the model, and ensures the