Multi-stage fire smoke early warning method and system based on taylor expansion feature extraction

Abstract

The application discloses a multi-stage fire smoke early warning method and system based on Taylor expansion feature extraction, and belongs to the technical field of intelligent fire monitoring and early warning.The application collects smoke concentration time series data through a sensor and completes noise reduction preprocessing, constructs a cubic polynomial fitting model based on the Taylor expansion continuous function approximation principle, combines the least square method to iteratively train and optimize fitting parameters, stably extracts a four-dimensional Taylor feature vector that can represent smoke static concentration, change rate, acceleration and mutation degree, builds a four-stage progressive early warning threshold system and a multi-feature joint determination logic, matches a hierarchical response mechanism of remote prompting, on-site verification, sound and light alarm, automatic fire extinguishing and fire alarm reporting according to different development stages of a fire, effectively solves the technical problems of traditional device early warning lag, weak anti-interference capability and single scene adaptation, can realize early trend prediction of a fire, and has the characteristics of lightweight overall algorithm, low deployment cost and strong working condition adaptability.

Description

Technical Field

[0001] This invention belongs to the field of fire safety monitoring and intelligent alarm technology, specifically involving a multi-level fire smoke early warning method and system based on Taylor expansion time series feature fitting training. It can be widely applied to various indoor and outdoor conventional and complex working conditions such as residences, shops, factories, and warehouses, and is suitable for early smoke dynamic monitoring, trend prediction and graded linkage early warning and disposal work. Background Technology

[0002] Fires spread rapidly and develop unpredictably. Early hazard identification and proactive warning intervention are core means to reduce fire losses and ensure site safety. Currently, mainstream civilian and industrial smoke warning devices on the market generally adopt a single judgment mechanism with a fixed concentration threshold. They determine the alarm status solely based on the instantaneous concentration value sampled by the sensor. This approach has a limited monitoring dimension and low utilization of time-series data, resulting in significant technical deficiencies in practical engineering applications. Traditional warning modes can only identify the open flame stage after a significant surge in smoke concentration, failing to capture the dynamic time-series characteristics of the slow rise and continuous growth of smoke during the early stages of a fire. Warning triggering is delayed, making it difficult to cover the golden window for fire response. Furthermore... The single static threshold discrimination mechanism lacks the ability to identify environmental interference and cannot effectively distinguish the temporal differences between conventional interferences such as cooking fumes, environmental dust, and air moisture fluctuations and actual fire conditions. The overall probability of false alarms and misreporting is relatively high. Frequent invalid alarms can reduce users' trust in early warning information and pose a safety hazard of overlooking actual fire conditions. In addition, traditional equipment only has two working states: alarm and non-alarm. It cannot output graded early warning instructions based on the development of the fire and is difficult to adapt to the stage evolution of fire from its inception, development, spread to loss of control. The overall level of intelligence and precision is insufficient and cannot meet the current demand for multi-scenario, highly reliable intelligent fire monitoring. Summary of the Invention

[0003] The purpose of this invention is to overcome the aforementioned defects in the existing technology. Addressing the engineering problems of weak dynamic sensing capabilities, high false alarm rates, lack of tiered response logic, and poor early warning effects in existing smoke warning devices, this invention proposes a multi-level fire smoke warning method and system based on Taylor expansion feature extraction. This invention relies on the continuous nonlinear characteristics of the temporal variation of smoke concentration. Through Taylor expansion temporal fitting modeling combined with least squares iterative optimization, it extracts multi-order temporal features that characterize the static state and dynamic evolution of smoke. It then constructs a four-level tiered early warning mechanism adapted to the full-cycle development characteristics of fire, achieving early fire trend identification, dynamic fire tracking, and tiered differentiated response, effectively improving the real-time performance and scenario adaptability of fire warnings.

[0004] To achieve the above technical objectives, this invention constructs a complete technical solution for time-series data processing and multi-level early warning determination, covering the entire process of data acquisition and preprocessing, Taylor polynomial fitting training, multi-order dynamic feature extraction, multi-level threshold joint determination, feature filtering optimization, and hierarchical response execution. The specific technical steps are as follows: Step 1: Data Acquisition and Preprocessing Photoelectric or ionization smoke sensors are used to continuously collect smoke concentration signals in the monitoring area at a fixed sampling frequency, constructing a continuous time series dataset of smoke concentration. The time series data is segmented by a fixed-length sliding window, and data preprocessing is performed by combining a moving average filtering algorithm to remove pulse anomalies and random noise generated during the acquisition process, ensuring the data stability and effectiveness of the input fitting algorithm, and reducing the interference of abnormal data on the accuracy of feature extraction.

[0005] Step 2: Taylor polynomial fitting training and multi-order dynamic feature extraction This step is the core improvement of this invention. By combining cubic polynomial modeling with least squares iterative fitting, it achieves accurate analysis of high-order features of discrete time series data. The specific execution process is as follows: 1. Model Construction: For the time-series data within the current sliding window, a cubic polynomial fitting model P(τ) = p0 + p1τ + p2τ is constructed to adapt to the continuous variation of smoke concentration. 2 +p3τ 3 In the formula, τ is the relative sampling time within the window, and p0, p1, p2, and p3 are the model coefficients to be fitted and solved.

[0006] 2. Iterative Fitting Training: Import all preprocessed valid sampled data within the sliding window, and use the minimum overall model fitting error as the iterative optimization objective. Continuously correct the coefficients of each model group until the fitting error converges to the preset accuracy threshold, and output the optimal combination of fitting parameters to effectively suppress the fitting deviation caused by environmental disturbances and sampling errors.

[0007] 3. Taylor Feature Mapping Conversion: Based on the converged optimal fitting coefficients, a higher-order Taylor feature mapping conversion is performed, transforming the single-dimensional concentration sampling data into four-dimensional dynamic features. The specific physical correspondences of each feature are as follows: The zero-order feature a0=p0 corresponds to the real-time smoke concentration at the current moment and is used to characterize the static smoke concentration level in the monitoring area; The first-order feature a1=p1 corresponds to the first derivative of the smoke concentration, which is used to characterize the real-time increase or decrease rate of smoke concentration and intuitively reflects the speed of fire development. The second-order feature a2=2p2 corresponds to the second derivative of smoke concentration, which is used to characterize the acceleration of concentration change and can accurately identify the trend of increasing or slowing fire rate. The third-order feature a3=6p3 corresponds to the third derivative of smoke concentration and is used to characterize the abruptness of concentration changes. It can effectively capture abnormal abrupt changes such as sudden escalation of fire and instantaneous increase in smoke. Step 3: Four-level progressive threshold system and hierarchical judgment logic This invention combines the phased evolution of fires from their inception, development, spread to out-of-control conditions, with the dynamic characteristics of multi-order Taylor features to establish a four-level progressive early warning threshold system. The threshold parameters at each level increase progressively, with clear hierarchical logic, adaptable to different fire severity levels. Through a combined judgment of static concentration characteristics and dynamic trend characteristics, it achieves refined and highly reliable fire identification. Specific judgment conditions and system response methods are shown in the table below:

[0008] Step 4: Extended Kalman Filter Feature Optimization To adapt to complex and highly interfering real-world monitoring scenarios, this invention adds a feature optimization filtering step. The extracted four-dimensional Taylor feature vector is used to construct the system state vector. An extended Kalman filter algorithm is then employed to perform iterative calculations for state prediction and observation correction. This compensates for feature shifts caused by fitting residuals, environmental noise, and sampling errors, further improving the stability and detection accuracy of the multi-dimensional feature output. Step 5: Execution of hierarchical response logic The system periodically executes the entire process of data acquisition, preprocessing, feature extraction, and threshold determination, and follows the principle of prioritizing response to the highest warning level to perform corresponding disposal operations, avoiding multi-level command conflicts and realizing continuous tracking and dynamic hierarchical disposal of fire status.

[0009] Beneficial effects Compared to traditional fixed-threshold smoke warning technology, this invention has significant advantages and practical value in engineering applications. Firstly, it overcomes the limitations of traditional equipment that relies solely on static instantaneous concentration for judgment. By utilizing Taylor series fitting and least squares iterative optimization mechanisms, it deeply mines the dynamic temporal characteristics of low-concentration, slowly growing smoke in the early stages of a fire. This allows for the early detection of fire development trends before open flames form, effectively extending the early warning time compared to traditional equipment and providing ample window time for on-site personnel to conduct hazard investigation and initial fire response. Secondly, this invention employs a four-dimensional feature joint judgment mode, distinguishing between actual fire conditions and environmental interference from both static numerical values ​​and dynamic change patterns. This invention effectively suppresses false alarms caused by interference factors such as dust, water vapor, and short-term oil fumes, significantly improving the long-term reliability and stability of the equipment. Simultaneously, the four-level graded early warning mode accurately matches the phased development characteristics of the fire situation, and can output differentiated prompts and handling instructions based on the severity of the hazard. This solves the shortcomings of traditional equipment's single alarm mode, which is crude and cannot adapt to multi-level risk management, thus improving the refined management level of fire safety monitoring. Furthermore, the overall algorithm of this invention is lightweight and has low computational overhead, allowing it to be directly integrated into conventional embedded main control chips without the need for high-performance computing equipment. This results in low hardware modification costs, wide scenario adaptability, and excellent engineering feasibility and market promotion prospects. Attached Figure Description

[0010] Figure 1 : Overall system architecture diagram of this invention; Figure 2 The overall flowchart of the early warning method based on Taylor expansion algorithm training in this invention; Detailed Implementation

[0011] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention. On the contrary, the embodiments of the present invention include all variations, modifications and equivalents falling within the spirit and scope of the appended claims.

[0012] To facilitate a complete understanding and reproduction of the technical solution of this invention by those skilled in the art, and in conjunction with the actual deployment and commissioning conditions of the equipment, the characteristics of real fire development, and the interference characteristics of different application scenarios, four sets of specific embodiments that can be reproduced in engineering are provided below. Each embodiment has its functions superimposed step by step, corresponding to conventional civilian standard deployment, multi-interference commercial and residential scenarios, industrial smoldering hazard monitoring, and complex and harsh outdoor conditions, respectively. These embodiments comprehensively verify the practical application effects and adaptability of the present invention in terms of Taylor time series feature extraction, iterative fitting optimization, multi-level threshold determination, and multi-source fusion early warning.

[0013] Example 1: Suitable for typical civilian scenarios such as homes and small shops with stable temperature and humidity, low dust, and no strong interference. It is mainly designed for early-stage and minor fires such as overloaded home sockets, slight smoldering of fabric materials, and overheating of small appliances. The hardware adopts a low-cost embedded terminal architecture, with the core controller being an STM32F407 microcontroller, paired with a standard civilian photoelectric smoke sensor. This sensor is only sensitive to smoke aerosol particles and has no obvious response to pure water vapor or light dust, which can meet the basic monitoring needs of civilian scenarios. The peripheral accessories include a WiFi communication module, LED status indicator lights, and a passive buzzer. There is no need to connect complex external fire extinguishing equipment, which is suitable for the basic security needs of ordinary residential buildings for early warning of potential hazards and timely on-site investigation.

[0014] This embodiment adopts fixed engineering parameters adapted for lightweight operation of embedded devices: the system sampling frequency is set to 1Hz, which matches the slow change characteristics of smoke in civilian scenarios and avoids the computing power redundancy problem caused by high-frequency sampling; the time-series sliding window length is set to 10 sampling points, which can stably cover the complete smoke change process within 10 seconds; the least squares method is fixed to 20 iterations, and the fitting convergence threshold is set to 0.001. This set of parameters has been tested and debugged multiple times in engineering, and can balance the fitting accuracy and the real-time operation of the device under the limited computing power of the microcontroller, without the occurrence of abnormal operation such as program lag or data delay.

[0015] Based on real-world fire test results in civilian scenarios, four levels of early warning thresholds were calibrated to adapt to common fire patterns, corresponding to different stages of fire development: L1 nascent warning threshold is suitable for scenarios with low-concentration, slowly rising smoke caused by slight overheating of lines and minor charring of materials; L2 action threshold is suitable for scenarios with continuous smoke diffusion and steady growth in the early stages of open flame; L3 and L4 high-risk thresholds are suitable for high-risk scenarios with continuous open flame and rapid increase in smoke. The specific calibrated threshold parameters are as follows: L1: a0=0.3%obs / m, a1=0.05%obs / m / s, a2=0.01%obs / m / s 2 ;L2: a0=0.6%obs / m, a1=0.15%obs / m / s, a2=0.03%obs / m / s 2 ;L3: a0=1.2%obs / m, a1=0.3%obs / m / s, a2=0.08%obs / m / s 2 a3 = 0.02%obs / m / s 3 ;L4: a0=2.0%obs / m, a1=0.5%obs / m / s, a2=0.15%obs / m / s 2 a3 = 0.05%obs / m / s 3 High-risk confirmation takes 2 seconds.

[0016] During actual operation, the equipment continuously collects environmental smoke time-series data. After filtering out single-sample fluctuation errors through a moving average filter, the data is fed into a cubic polynomial fitting model to complete iterative convergence calculations, outputting complete Taylor features from order 0 to 3. The system abandons the traditional single-concentration judgment logic and combines static smoke concentration, growth rate, acceleration trend, and degree of abrupt change for multi-dimensional comprehensive judgment. The test results show that for early and weak fires such as those caused by overload of ordinary sockets, when no dense smoke is visible to the naked eye and traditional early warning devices cannot trigger alarms, this system can accurately capture the continuous growth trend of weak smoke and trigger L1 remote alerts in advance. If the hidden danger is not dealt with in time and the smoke continues to intensify, the system upgrades the warning level step by step, eventually triggering on-site audible and visual alarms, which is fully adapted to the gradual development law of fires in civil scenarios from weak to strong.

[0017] Example 2: Applicable to mixed-use commercial and residential shops and open-plan offices where there is constant environmental interference, which is a common scenario for false alarms of traditional smoke detectors. In such scenarios, there are daily interference factors such as airflow disturbances caused by opening windows for ventilation, air moisture generated by floor humidification, short-term catering fumes, and dust from daily cleaning. These factors can easily cause instantaneous fluctuations in sensor values, leading to frequent false alarms from traditional fixed threshold devices. This example, based on the hardware architecture and basic algorithm framework of Example 1, adds extended Kalman filter feature optimization logic to accurately filter interference signals with short-term fluctuations and no continuous growth trend.

[0018] Engineering measurements can clearly distinguish the temporal characteristics of interference signals and real fires: interference signals such as water vapor, dust, and short-term oil fumes generally have the characteristics of sudden numerical increases, rapid declines, and no sustained positive growth, and the corresponding first-order and second-order Taylor features have no sustained positive gain; while the smoke generated by real fires has the core characteristics of continuous concentration increase, stable growth rate, and irreversible trend. In this embodiment, the four-dimensional Taylor feature vector is constructed as the system state vector, and iterative correction is completed through extended Kalman filtering to smooth short-term sudden noise, retain the sustained growth trend characteristics, and accurately distinguish the temporal differences between instantaneous environmental fluctuations and real fires.

[0019] Actual test results show that under normal interference conditions such as open window ventilation, short-term cooking fumes, and indoor dust, the system will not generate any invalid warnings after filtering and optimization. For weak early fires such as appliance overheating and aging wiring, the system can still accurately identify low-amplitude and continuous smoke growth trends, fully retaining the ultra-early warning capability. This embodiment significantly reduces the frequency of invalid false alarms in commercial and residential scenarios without reducing the sensitivity of fire detection, and is suitable for the actual operational needs of commercial venues with high personnel flow and long-term unattended areas.

[0020] Example 3: Applicable to high-risk industrial scenarios such as factories, warehouses, and computer rooms, where the ignition characteristics differ significantly from civilian open flame fires. These fires are mostly insidious, such as smoldering of accumulated materials, smoldering of aging cable insulation, and smoldering of overheated equipment. In the early stages of such fires, the smoke concentration is extremely low, changes slowly, and there is no obvious open flame or dense smoke. A single smoke sensor can only collect weak fluctuations, which traditional equipment cannot effectively identify, easily leading to missed or delayed reporting. By the time the dense smoke spreads and the fire becomes visible, it has already caused significant safety hazards and economic losses. To address this industrial pain point, this example adds temperature and CO dual-dimensional auxiliary monitoring based on the optimized algorithm of Example 2, and improves the identification capability of insidious fires through a multi-source data fusion mechanism.

[0021] On the hardware side, a high-precision temperature probe and an electrochemical CO sensor were added to achieve synchronous acquisition and time-series matching of three types of data: smoke concentration, ambient temperature, and carbon monoxide concentration. Engineering tests verified that the core characteristic of industrial smoldering fires is that the smoke changes are weak, but accompanied by a slow rise in ambient temperature and a continuous release of trace amounts of CO gas. In contrast, industrial dust and routine ventilation disturbances only cause instantaneous fluctuations in smoke values, without synchronous increases in temperature and CO concentration. This embodiment uses Taylor's four-dimensional dynamic smoke characteristics as the core judgment criterion and relies on DS evidence theory to integrate temperature and CO change characteristics to weight and confirm the weak growth trend of smoldering fires, effectively correcting the identification bias of single sensor data.

[0022] Actual operating condition test results show that, for typical hidden industrial hazards such as smoldering of cardboard boxes, smoldering of cable trays, and overheating and fire of circuit boards, this system can accurately locate the fire in the early stages when the smoke has not spread significantly, through multi-dimensional feature fusion. At the same time, it can effectively distinguish between normal operating conditions such as production dust, equipment heat dissipation, and ventilation and fire conditions, completely solving the industry problem of both missed and false alarms in industrial scenarios, and meeting the high reliability and zero missed alarm safety monitoring requirements of high-risk industrial sites.

[0023] Example 4: Applicable to outdoor storage yards, open-air material warehouses, semi-open factories, and temporary construction sites, which are typical high-interference and harsh working conditions. The environment is characterized by complex interference factors such as variable wind speed, large temperature difference between day and night, frequent sandstorms, and drastic fluctuations in air humidity. Traditional fixed-parameter algorithms cannot adapt to the dynamically changing outdoor environment. Strong winds can dilute the smoke concentration and cause data jumps. Sandstorms and moisture can interfere with the sampling accuracy of sensors, which can easily lead to problems such as false alarms, missed alarms, and unstable responses. This example adopts a dual optimization scheme of hardware protection upgrades and algorithm adaptive optimization to address the dynamic interference characteristics of complex working conditions, which greatly improves the system's environmental adaptability.

[0024] At the hardware level, a dustproof and waterproof protective shell and signal amplification and conditioning circuit are added to the multi-sensor acquisition architecture to filter high-frequency sampling noise and block the direct interference of sand and moisture to the sensor probe, ensuring the integrity and stability of the original sampling data. At the algorithm level, the fixed iteration parameter mode is abandoned and an adaptive iteration optimization mechanism is added: the system monitors the fluctuation variance of the data within the sliding window in real time. When the environmental conditions are stable, it maintains 20 basic iteration operations. When the environmental fluctuations are severe and the noise interference is strong, it automatically adaptively increases the number of iterations to 20-30 times, dynamically tightens the fitting convergence accuracy, enhances the algorithm's ability to extract effective time series trends, and accurately filters irregular environmental random disturbances.

[0025] Meanwhile, to address the issue of instantaneous changes in outdoor data, a continuous verification mechanism is added to the L3 and L4 high-risk fire alarm levels. Only when the high-order Taylor mutation feature continues to maintain the preset confirmation time can it be determined as a real fire, effectively avoiding instantaneous false triggers caused by gusts of wind and instantaneous dust. After verification by outdoor multi-scenario tests, this embodiment can operate stably in strong winds, diurnal temperature fluctuations, and light dust weather. It can accurately identify real fires such as open-air material fires and weed fires, and can completely shield various conventional environmental interferences, greatly expanding the engineering application scenarios and adaptation scope of this invention.

[0026] It should be noted that in the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-level fire smoke early warning method based on Taylor expansion feature extraction, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect the raw time-series signal of smoke concentration in the monitoring area, construct a smoke concentration time-series dataset, and use a sliding window filtering method to perform noise reduction preprocessing on the raw time-series data to remove abnormal sampling data and obtain stable and effective time-series input data. S2. Taylor high-order dynamic feature solution: Based on the Taylor expansion continuous function approximation principle, a cubic polynomial fitting model is constructed for the discrete time series data within the sliding window. The model iterative fitting operation is performed by the least squares method to obtain the 0th, 1st, 2nd and 3rd order Taylor feature coefficients corresponding to the smoke concentration time series function, and a four-dimensional dynamic feature vector is constructed. S3. Multi-level threshold comparison: The preset four-level progressive alarm threshold system is invoked, and the Taylor characteristic coefficients of each order output in real time are compared with the corresponding level thresholds. The highest warning level corresponding to the current fire situation is determined through the priority determination mechanism. S4. Tiered Emergency Response Execution: Based on the highest warning level determined, the corresponding differentiated emergency response mechanism is triggered to achieve real-time dynamic monitoring, trend prediction, and tiered handling of fire smoke, thereby improving the timeliness of fire warnings and the stability of scenario adaptation.

2. The multi-level fire smoke early warning method based on Taylor expansion feature extraction according to claim 1, characterized in that, The specific implementation methods of Taylor expansion fitting and feature solving in step S2 include: S21. Model Construction: Within the sliding window time domain, construct a cubic polynomial fitting model that adapts to the continuous variation characteristics of smoke concentration. The model expression is P(τ)=p0+p1τ+p2τ. 2 +p3τ 3 Where τ is the relative sampling time within the sliding window, and p0, p1, p2, and p3 are the fitting parameters to be solved for the model; S22, Iterative Fitting Convergence: Import all preprocessed effective time series data within the sliding window, take minimizing the overall model fitting error as the iterative objective, iteratively correct each set of fitting parameters until the fitting error converges to the preset accuracy threshold, and output the optimal combination of fitting parameters. S23. Feature Mapping Output: Based on the converged optimal fitting parameters, high-order feature mapping conversion is completed to obtain four-dimensional time-series features: zero-order feature a0=p0, representing the real-time static smoke concentration; first-order feature a1=p1, representing the rate of change of smoke concentration; second-order feature a2=2p2, representing the acceleration of change of smoke concentration; third-order feature a3=6p3, representing the abrupt change of smoke concentration.

3. The multi-level fire smoke early warning method based on Taylor expansion feature extraction according to claim 2, characterized in that: In the least squares iterative fitting process, the preset fitting error convergence threshold is 0.001, and the number of time series data iterations corresponding to a single sliding window is no less than 20. Through multiple rounds of iterative operations, sampling deviation and environmental noise interference are reduced, ensuring the stability and consistency of Taylor high-order feature extraction.

4. The multi-level fire smoke early warning method based on Taylor expansion feature extraction according to claim 1, characterized in that: The four-level progressive alarm threshold system corresponds one-to-one with each order of Taylor characteristics, and the thresholds at each level exhibit a hierarchical correlation with each level increasing progressively. The zero-order concentration characteristic threshold satisfies Th0. 1 < Th0 2 < Th0 3 < Th0 4 The characteristic thresholds corresponding to the first-order velocity, second-order acceleration, and third-order jerkiness increase synchronously and progressively. The alarms at all levels adopt a joint judgment mechanism of static concentration characteristics and dynamic trend characteristics, with judgment conditions becoming progressively stricter to adapt to the full-time evolution of fire smoke from its inception, development, spread to loss of control.

5. The multi-level fire smoke early warning method based on Taylor expansion feature extraction according to claim 1, characterized in that, The specific criteria and response mechanisms for determining each level of early warning are as follows: L1 warning level: The zero-order Taylor feature exceeds the preset warning threshold, and the first-order Taylor feature or the second-order Taylor feature meets the corresponding warning threshold condition. It is determined to be a fire in the early stage and a remote message push warning operation is executed. L2 Action Level: The zero-order Taylor feature, first-order Taylor feature, and second-order Taylor feature simultaneously exceed the corresponding action level threshold, which is judged as an escalation of the smoke diffusion hazard, triggering the on-site low-frequency prompt and light warning mechanism; L3 Fire Alarm Level 1: The zero-order Taylor feature exceeds the standard, and any first-order dynamic Taylor feature meets the sudden growth threshold condition, which is determined to be the initial formation of a fire, and the on-site high-frequency audible and visual alarm mechanism is activated. Level 4 Fire Alarm Level 2: The zero-order Taylor feature reaches the high-risk threshold, and the high-risk feature judgment condition maintains the preset confirmation time, which is judged as the fire continuing to spread, and the fire extinguishing device and fire alarm reporting mechanism are activated in conjunction.

6. The multi-level fire smoke early warning method based on Taylor expansion feature extraction according to claim 1, characterized in that: After the four-dimensional dynamic feature vector is constructed, a feature optimization filtering step is added to construct the four-dimensional Taylor feature vector into a system state vector. The vector is then input into an extended Kalman filter to complete the iterative calculation of state prediction and observation correction, which offsets the feature deviation caused by sampling error and fitting residual, and improves the accuracy and stability of feature output in complex scenarios.

7. A multi-level fire smoke early warning system based on Taylor expansion feature extraction, characterized in that, The method for implementing the multi-level fire smoke early warning method according to any one of claims 1 to 6 includes a hardware acquisition module and an algorithm processing module working in concert. The overall system architecture relies on Taylor high-order feature extraction and multi-level threshold comparison mechanism to achieve dynamic prediction and graded early warning of fire smoke, specifically including: Smoke sensor module: used to collect raw time-series signals of smoke concentration in the monitored area, providing raw data support for backend algorithm calculations; Data acquisition and preprocessing module: used to normalize, remove outliers and perform sliding window filtering on raw time series data, and output low-noise standardized time series data; Taylor Algorithm Fitting Training and Feature Calculation Module: Built-in cubic polynomial fitting model and least squares iterative operation logic, used to complete the convergence of time series data fitting and solve the dynamic features of 0 to 3rd order Taylor. Multi-level threshold comparison module: Pre-stores a four-level progressive Taylor feature threshold system to complete the progressive comparison between multi-dimensional features and hierarchical thresholds, and outputs the corresponding fire warning level; Tiered response execution module: Used to match different warning levels and execute corresponding tiered response operations such as remote prompts, on-site warnings, audible and visual alarms, automatic fire extinguishing, and fire alarm reporting.

8. The multi-level fire smoke early warning system according to claim 7, characterized in that: The system is also equipped with a multi-sensor fusion decision module and a Kalman filter optimization module. The multi-sensor fusion decision module receives multi-source monitoring data such as temperature and CO concentration, and uses DS evidence theory to complete multi-dimensional data cross-validation and fusion decision-making, assisting in correcting Taylor feature discrimination results and suppressing the risk of misjudgment caused by single sensor detection errors. The Kalman filter optimization module is used to iteratively correct the Taylor multi-dimensional dynamic features, suppress fitting residuals and environmental noise interference, and improve the stability and accuracy of the system's early warning under complex working conditions.

Images