Intelligent safety monitoring and control method and system for gas stove

By using multi-source sensing and vision fusion technology and closed-loop verification, the problems of single identification and rigid control in gas stove monitoring methods have been solved, realizing intelligent management and control of the cooking process and improving the accuracy and efficiency of safety monitoring.

CN123065008APending Publication Date: 2026-11-10FOSHAN YIPINHUI ELECTRIC CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611229031.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing gas stove monitoring methods have limited identification dimensions, are prone to false alarms and missed alarms, and have rigid control strategies that cannot meet the safety requirements of complex cooking scenarios.

Method used

By simultaneously acquiring multi-source sensor signals and visual images, spatiotemporal consistency verification is performed. The characteristics of cookware presence, flame presence, temperature change, and humidity trend are extracted and combined with evidence constraints within a preset time window to achieve hierarchical control. A closed-loop verification and secondary intervention mechanism is also introduced.

Benefits of technology

It achieves full-process automated and intelligent control of the cooking process, reduces false alarm rate, improves the accuracy and efficiency of safety monitoring, and ensures the dynamic adaptability and reliability of control strategies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065008A_ABST
    Figure CN123065008A_ABST
Patent Text Reader

Abstract

The application discloses a kind of gas stove intelligent safety monitoring control method and system, it is related to safety monitoring technical field.Therein, the method includes: synchronously collecting the multi-source sensing signal of multiple sensors output of gas stove, visual image and user operation information;Multi-source sensing signal and visual image are checked in space-time consistency, and the characteristic of in-place cookware, flame existence, temperature change and humidity trend is fused and extracted;According to the above characteristics and operation information, the cooking process is divided into multiple states, and the transition between cooking states is judged based on multiple evidence condition constraints in continuous time window;According to the state currently being in, the corresponding hierarchical control strategy is triggered;Read valve position feedback, flame and gas concentration signal, close-loop verification execution result of hierarchical control strategy, and execute secondary intervention mechanism when verification fails.The application effectively reduces the false alarm rate of single sensor, and excludes the failure risk of open-loop control.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, and in particular to an intelligent safety monitoring and control method and system for gas stoves. Background Technology

[0002] With the development of smart home technology, the safety management of gas stoves has become a core aspect of home security. Gas stoves face complex application scenarios in daily cooking, and in order to improve safety protection, effective monitoring of their operating status, especially the timely identification of abnormal risks such as dry burning and leaks, is crucial.

[0003] In practice, the kitchen environment presents complex challenges such as temperature fluctuations, moisture evaporation, and cooking fumes. Currently, preventing risks associated with gas stoves primarily relies on user observation or basic mechanical safety devices. This approach is not only inefficient and subjective but also prone to oversights, failing to meet the safety demands of modern, fast-paced lifestyles.

[0004] To address the aforementioned drawbacks, several automated monitoring tools based on the Internet of Things (IoT) have emerged in the existing technology. However, these tools have significant technical limitations. First, most of them rely on a single modality for hazard assessment, such as using only temperature threshold comparisons or single gas concentration matching. This approach cannot comprehensively evaluate the overall semantics constituted by multimodal information including environmental, flame, and user operation information, leading to inaccurate judgments of abnormal states. Second, existing measurement methods are relatively rudimentary and struggle to identify subtle differences in characteristics between normal cooking and early hazardous states, resulting in limited accuracy. Furthermore, when executing control commands, these tools typically employ a static open-loop execution mode, failing to dynamically confirm the execution effect based on the actual movement of underlying valves. This makes the safety protection strategy rigid and may cause true hazards to be missed when mechanical jamming occurs.

[0005] In summary, existing monitoring methods, due to the limited functionality and low level of intelligence of their automated tools, are unable to accurately identify and handle complex cooking risks, potentially leading to false alarms or missed alarms, ultimately impacting overall safety protection effectiveness. Therefore, there is an urgent need for a technological solution capable of intelligently, dynamically, and comprehensively monitoring and controlling gas stoves. Summary of the Invention

[0006] This invention provides a method and system for intelligent safety monitoring and control of gas stoves, which solves the problems of existing technologies, such as the single identification dimension, the susceptibility to false alarms and missed alarms, and the potential for failure in one-way open-loop control.

[0007] The first aspect of this invention provides a method for intelligent safety monitoring and control of a gas stove, comprising: Simultaneously collect multi-source sensor signals, visual images, and user operation information output from multiple sensors of the gas stove; Spatiotemporal consistency verification is performed on multi-source sensor signals and visual images, and the in-situ features of cookware, flame presence features, temperature change features and humidity trend features are extracted and fused. Based on the characteristics of the cookware in place, the characteristics of the flame presence, the characteristics of temperature change, the characteristics of humidity trend, and the user operation information, the cooking process of the gas stove is divided into multiple cooking states, and the transition between cooking states is determined based on multiple evidence conditions within a preset continuous time window. Based on the current cooking state, trigger the corresponding hierarchical control strategy; After triggering the hierarchical control strategy, the gas valve position feedback signal, flame signal and gas concentration signal from the multi-source sensor signals are read to verify whether the hierarchical control strategy is executed successfully in a closed loop, and a secondary intervention mechanism is executed if the verification fails.

[0008] In one optional implementation, the multi-source sensing signals include flame signals, gas valve opening signals, gas valve position feedback signals, gas concentration signals, ambient humidity signals, and temperature signals; wherein, the gas concentration signals include gas concentration signals, carbon monoxide concentration signals, and carbon dioxide concentration signals. The method also includes effectiveness evaluation and baseline calibration of multiple sensors, specifically: Real-time acquisition of multi-source sensor signals, and extraction of signal fluctuation characteristics and temperature drift characteristics within a sliding time window; The effectiveness scores of multiple sensors are calculated based on signal fluctuation characteristics and temperature drift characteristics. When the effectiveness score of a sensor is lower than the preset failure threshold, the data input of the corresponding sensor is blocked. A lightweight anomaly model is used to extract the joint rate of change features of gas concentration signals, carbon monoxide concentration signals, and carbon dioxide concentration signals from the gas concentration signal as a multidimensional state feature vector. The multidimensional state feature vector is then compensated by combining pre-acquired kitchen environmental background data, the drift self-calibration baseline is dynamically updated, and the multi-source sensor signals are calibrated based on the drift self-calibration baseline.

[0009] In one optional implementation, spatiotemporal consistency verification is performed on the multi-source sensor signals and visual images, and the features of cookware presence, flame presence, temperature change, and humidity trend are extracted and fused, including: Visual images are identified using object detection algorithms, and features of cookware images and steam obstruction features are extracted. The position of the cookware image features in the image coordinate system is matched with the temperature signal acquisition area in the physical space through coordinate mapping. When vapor obstruction features are detected or the ambient illuminance is below a preset threshold, the confidence weight of the visual image in data fusion is reduced, while the confidence weight of the multi-source sensor signal is increased simultaneously. Based on the adjusted confidence weights, the spatiotemporally aligned multi-source sensor signals and visual images are fused to obtain the in-situ features of the cookware, the presence features of the flame, the temperature change features, and the humidity trend features.

[0010] In one alternative implementation, when determining the presence of flame characteristics, changes in carbon dioxide concentration signals are used as auxiliary evidence of combustion. The verification process includes: Simultaneously acquire the operating status signals of the kitchen ventilation equipment; Calculate the relative rate of change of carbon dioxide concentration signals and carbon monoxide concentration signals within a set time window; When the gas valve opening signal indicates that it is open and no flame is detected, but the gas concentration signal shows an upward trend, a combustion joint judgment matrix is ​​constructed based on the relative change rate and the working status signal. When the relative rate of change is lower than the normal ignition threshold and the gas concentration signal increases, the gas stove is identified as leaking. When the working status signal indicates that the kitchen ventilation equipment is on, the safety judgment threshold of the joint judgment matrix is ​​increased.

[0011] In one optional implementation, the multiple cooking states include at least a normal heating state and a suspected dry-burning state; based on multiple evidence constraints within a preset continuous time window, the transition of the cooking process between cooking states is determined, including: The normal heating state is divided into the ignition stage, the heating stage, the boiling stage, and the water evaporation stage, and the temporal evolution of the cooking process in the ignition stage, the heating stage, the boiling stage, and the water evaporation stage is monitored. When it is determined that the current stage is boiling and the humidity trend is increasing, it is determined to be a normal evolutionary state. When the transition from the moisture evaporation stage to the suspected dry burning state, the multiple evidence conditions within the continuous time window include at least the following: the humidity trend characteristics turn to a continuous decline after reaching the peak, the temperature change characteristics exceed the preset maximum safe slope, and the user operation information indicates no human intervention. The cooking process is shifted to a suspected dry-burning state if and only if multiple evidentiary constraints overlap sequentially within a consecutive time window.

[0012] In one alternative implementation, the hierarchical control strategy for the case where the cookware is removed is as follows: When the presence of flame is confirmed and the pot is determined to be in a removed state based on the pot's position, the transition holding phase of the short-term holding timer is triggered. During the transition holding phase, if the cookware presence indicator still indicates that the cookware has been removed, the gradual heat reduction strategy is activated. The gradual flame reduction strategy drives the gas stove's control element to reduce the gas flow, causing the flame level to gradually decrease over time, and simultaneously issuing a reminder command. When the cumulative duration of the short-term holding timer exceeds the preset protection limit and the cookware's presence characteristic has not been restored to the present state, the gas supply pipeline is cut off.

[0013] In one optional implementation, the multiple cooking states also include a flameout leakage state. The process of determining the transition between cooking states includes forced flameout leakage detection logic and state rollback correction logic. The forced flameout leakage judgment logic is as follows: when the presence of the cookware is determined based on the cookware's presence characteristics, but the flame presence characteristics change from presence to disappearance during continuous detection, and the gas concentration signal is detected to show an increasing trend, the system switches to the flameout leakage state and triggers emergency shut-off. The state rollback correction logic is as follows: If the system enters a suspected dry-burning state, and during the delayed warning observation period, it re-captures the manual intervention action based on the user operation information, and the temperature change characteristic changes from rising to flat or falling, then the previous state judgment is considered a false alarm, the suspected dry-burning state is canceled, and the cooking process is rolled back to the normal heating state.

[0014] In one alternative implementation, the lightweight anomaly model is built upon a multi-branch neural network to determine the evolution of the cooking process towards an abnormal state, including: Historical continuous sampling data of multi-source sensor signals are acquired and a two-dimensional matrix sequence is generated as the feature input of a lightweight anomaly model; The first feature extraction branch is used to extract local high-frequency oscillation features for the short-term transient change features of the temperature signal; The second feature extraction branch is used to extract long-term smoothing trend features for environmental humidity signals and carbon monoxide concentration signals. After fusing local high-frequency oscillation characteristics and long-term smooth trend characteristics, classification and prediction are performed. When the predicted probability exceeds the safety threshold, physical gas cut-off disaster prevention operations are triggered.

[0015] In one optional implementation, the closed-loop verification mechanism checks whether the hierarchical control strategy was successfully executed, and a secondary intervention mechanism is implemented when verification fails, including: When a command to close the gas valve is issued, the mechanical operating position of the physical valve is read from the position detection element and used as the gas valve position feedback signal. After confirming that the gas valve position feedback signal has reached the closed dead point, the flame signal is read after a set delay period to confirm that the flame has disappeared; At the same time, analyze whether the gas concentration signal shows the expected downward trend; If any of the following occurs: the gas valve position feedback signal fails to reach the closing dead point, the flame signal indicates that the flame has not disappeared, or the gas concentration signal does not show the expected downward trend, the graded control strategy is deemed to have failed and a secondary intervention mechanism is initiated. The secondary intervention mechanism includes shutting off the main intake shut-off valve at the front end and linking the ventilation equipment to exhaust the air.

[0016] A second aspect of this invention provides an intelligent safety monitoring and control system for gas stoves, comprising: The multi-source signal acquisition module is used to simultaneously acquire multi-source sensor signals, visual images, and user operation information output by multiple sensors of the gas stove; The consistency verification feature extraction module is used to perform spatiotemporal consistency verification on multi-source sensor signals and visual images, and to fuse and extract features such as cookware presence, flame presence, temperature change, and humidity trend. The state evolution constraint module is used to divide the cooking process of the gas stove into multiple cooking states based on the characteristics of the cookware in place, the characteristics of the flame presence, the characteristics of temperature change, the characteristics of humidity trend, and user operation information. Based on multiple evidence conditions within a preset continuous time window, it determines the transition of the cooking process between cooking states. The hierarchical control execution module is used to trigger the corresponding hierarchical control strategy based on the current cooking state of the cooking process; The closed-loop feedback verification module is used to read the gas valve position feedback signal, flame signal and gas concentration signal from the multi-source sensor signals after the hierarchical control strategy is triggered, to verify whether the hierarchical control strategy is executed successfully, and to execute a secondary intervention mechanism when the verification fails.

[0017] Beneficial effects: This invention provides an intelligent safety monitoring and control method and system for gas stoves. By performing spatiotemporal consistency verification on multi-source sensor signals and visual images to form a unified perception view, and combining it with a constraint mechanism based on multiple evidence conditions within a continuous time window, it achieves automated and intelligent control of the entire cooking process. This effectively overcomes the limitations of existing technologies, such as reliance on a single sensor, low prevention efficiency, difficulty in identifying complex risks, and rigid control strategies. By integrating multi-dimensional features to deeply analyze the cooking scenario, it transforms changes in physical parameters into specific cooking states and automatically triggers corresponding hierarchical control strategies. This reduces the false alarm rate caused by daily cooking disturbances, allowing users to obtain accurate safety protection without manually judging dangers, thereby improving the efficiency and accuracy of kitchen safety monitoring. The introduction of a closed-loop verification and secondary intervention mechanism based on multi-source signals enables the system to dynamically confirm control results based on the actual physical feedback from valves and flames. This avoids the command failure risks caused by traditional one-way open-loop control, ensuring that the monitoring and control system maintains a stable and reliable operating state when handling different abnormal risks. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 A flowchart illustrating the steps of an intelligent safety monitoring and control method for a gas stove provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the functional modules of an intelligent safety monitoring and control system for a gas stove provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to the accompanying drawings.

[0021] Reference Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0022] like Figure 1As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (Wi-Fi). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0023] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0024] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and computer programs.

[0025] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the intelligent safety monitoring and control system for gas stove stored in the memory 1005 through the processor 1001 and executes the intelligent safety monitoring and control method for gas stove provided in the embodiment of the present invention.

[0026] Combined with appendix Figure 2 This embodiment provides an intelligent safety monitoring and control method for gas stoves, which includes the following basic steps and detailed sub-steps for various optional implementation methods: S101, synchronously collects multi-source sensor signals, visual images and user operation information output by multiple sensors of the gas stove; In this embodiment, the edge monitoring hardware platform is equipped with a variety of high-precision sensors. The multi-source sensing signals specifically include: flame signals (acquired via K-type thermocouples or ion probes), gas valve opening signals (acquired via high-precision Hall absolute angle encoders), gas valve position feedback signals (acquired via microswitches to detect physical dead points), gas concentration signals, ambient humidity signals (acquired via capacitive humidity sensors), and temperature signals (including pot bottom temperature acquired via infrared thermometry matrix and stove surface temperature acquired via thermistors). The gas concentration signals further include gas concentration signals (e.g., methane, acquired using a catalytic combustion sensor), carbon monoxide concentration signals (acquired using an electrochemical sensor), and carbon dioxide concentration signals (acquired using a non-dispersive infrared (NDIR) sensor).

[0027] To ensure strict alignment of heterogeneous data on the timeline, the edge monitoring hardware platform employs a hardware timer interrupt mechanism with microsecond-level synchronization accuracy to trigger sampling for all sensors. The sensor sampling frequency is preferably set to 10 Hz, and the visual image sampling frequency is set to 30 frames per second (FPS).

[0028] To ensure the reliability of the underlying data and cope with interference from complex kitchen environments, one optional implementation method also includes effectiveness evaluation and baseline calibration of multiple sensors, specifically comprising the following sub-steps: S1011. Acquire multi-source sensor signals in real time and extract signal fluctuation characteristics and temperature drift characteristics within the sliding time window; S10111. For each type of sensor, allocate a space with a length of [length missing] in the microcontroller's Static Random-Access Memory (SRAM). A ring buffer is used as a sliding time window. In this embodiment, The value range is set from 3 to 8 seconds, with an optimal value of 5 seconds. The technical basis for this value range is that the physical processes of heat conduction and gas diffusion in the gas stove environment have inherent lag. If the time window is less than 3 seconds, due to insufficient sample size, it is impossible to effectively smooth out the transient high-frequency noise of airflow caused by the instantaneous flickering of the flame or the movement of people. If the time window is greater than 8 seconds, it will cause the algorithm to be severely delayed in responding to real gas leaks or sudden changes in dry burning temperature, thus losing the significance of early warning.

[0029] S10112. Within a 5-second time window (containing 50 discrete sampling points), calculate the statistical variance of the signal as a characteristic of signal fluctuation. The calculation formula is as follows:

[0030] In the formula, The total number of sampling points within the time window (here) ), For the first The sensor at the first Real-time measurement values ​​at each sampling point This is the arithmetic mean of the signal within this time window.

[0031] S10113. The least squares method is used to perform first-order linear regression fitting on the data points within the time window, and the slope of the fitted line and the absolute drift of the first and last data points of the sliding window are extracted. As a characteristic of temperature drift.

[0032] S1012. Calculate the effectiveness scores of multiple sensors based on signal fluctuation characteristics and temperature drift characteristics. When the effectiveness score of a certain sensor is lower than the preset failure threshold, block the data input of the corresponding sensor. S10121. The gas stove controller uses the following formula to calculate the sensor's effectiveness score:

[0033] In the formula, For the first The real-time effectiveness score of each sensor is a dimensionless relative score. The baseline score is set to 100. The variance penalty coefficient is set to a value range of [0.5, 1.5], and is preferably 1.0 in this embodiment. This coefficient is used to measure the degree to which high-frequency noise damages data reliability. Since semiconductor sensors exhibit severe, irregular output jitter in the early stages of aging, It can effectively suppress the weight of aging probes without amplifying normal thermal noise. The temperature drift penalty coefficient is set to a value range of [2.0, 5.0], and is preferably 3.5 in this embodiment. Kitchen environments often experience slow temperature rises; normal physical temperature drift should be gradual. An excessively high value usually indicates chemical poisoning or physical detachment of the reference electrode inside the sensor. (Settings) The basis for this is to increase the penalty for persistent abnormal drift at the zero point, causing the score of damaged sensors to drop rapidly.

[0034] S10122. Set the preset failure threshold to 60 points. When a failure is detected... If this occurs, it indicates that the sensor may have suffered physical damage, severe aging, or the probe may have been blinded by heavy oil contamination. The edge monitoring hardware platform will immediately cut off data reading from the analog-to-digital converter (ADC) channel at the Hardware Abstraction Layer (HAL) via register configuration to prevent the anomaly from contaminating the subsequent artificial intelligence decision-making network.

[0035] S1013. The joint rate of change features of gas concentration signal, carbon monoxide concentration signal and carbon dioxide concentration signal in the gas concentration signal are extracted using a lightweight anomaly model as a multidimensional state feature vector. The multidimensional state feature vector is compensated by combining the pre-acquired kitchen environment background data, the drift self-calibration baseline is dynamically updated, and the multi-source sensor signals are calibrated based on the drift self-calibration baseline.

[0036] S10131. Due to the evaporation of cooking wine, the heating of vinegar, and the use of strong alkaline cleaning agents, traditional semiconductor gas sensors may experience cross-sensitivity interference. This embodiment deploys a lightweight anomaly model based on a one-dimensional convolutional neural network (1D-CNN) architecture in the neural network processor. The input layer receives a tensor matrix with dimensions [50,3] (50 sampling steps, corresponding to three feature channels: methane, carbon monoxide, and carbon dioxide). The network contains two one-dimensional convolutional layers. The first layer has a kernel size of 3 and a stride of 1, used to extract small concentration gradients between three adjacent time steps. The activation function is a rectified linear unit (ReLU), with the formula: To address the gradient vanishing problem, the second convolutional kernel size is set to 5 to expand the receptive field and extract overall trend evolution features.

[0037] S10132. Extract the joint rate of change feature vector through convolution operation and input it into the fully connected layer. The gas stove controller pre-collects background data for up to 72 hours during non-cooking periods (such as nighttime standby mode) to generate an environmental background noise distribution matrix reflecting the current kitchen microenvironment (including temperature, humidity, and background gases). .

[0038] S10133, The gas stove controller compensates for bias through matrix subtraction: In the formula, To compensate for the calibrated gas concentration feature vector; The joint rate of change feature vector of the original input; This is the dynamic weight matrix output by the 1D-CNN; The aforementioned environmental background noise distribution matrix is ​​used to extract the true baseline concentration of combustion byproducts. This drift self-calibration baseline is set to be dynamically updated every 60 seconds, ensuring that this method always calculates subsequent gas concentration data based on a true zero-point reference.

[0039] S102, in one optional implementation, performs spatiotemporal consistency verification on the multi-source sensor signals and visual images, and fuses and extracts the in-situ features of the cookware, the presence features of the flame, the temperature change features, and the humidity trend features, including the following sub-steps: S1021. Visual images are identified using target detection algorithms, and features of cookware images and steam obstruction features are extracted. S10211. This method employs a YOLOv8-tiny object detection algorithm optimized for the computing power of edge devices through deep pruning. The backbone network uses a variant of the Cross-Stage Partial Network (CSPNet) structure to reduce redundant gradient flow computation. During offline training, supervised learning is performed using a dataset of over 100,000 images containing various cookware materials such as stainless steel, cast iron, and ceramic, under different lighting intensities, and with numerous scenes obscured by cooking fumes. The network prediction head output includes the two-dimensional pixel coordinates of the cookware bounding box, the class probability vector, and the confidence score.

[0040] S10212. The algorithm evaluates the attenuation degree of high-frequency edge information in the image using the Laplacian variance operator and calculates the vapor occlusion features of the global image using the Structural Similarity Index Measure (SSIM). When the proportion of pixel regions with an SSIM below 0.65 exceeds a set ratio, the vapor occlusion ratio factor is quantized and output. .

[0041] S1022. Perform coordinate mapping matching between the position of the cookware image features in the image coordinate system and the temperature signal acquisition area in the physical space; S10221. Due to perspective distortion in visual images, the gas stove controller utilizes the camera intrinsic matrix (denoted as...) pre-calibrated at the factory. ) and the extrinsic matrix (denoted as ) which contains rotation and translation vectors. This involves performing an inverse homogeneous coordinate transformation using a perspective projection model derived from a pinhole camera. (The physical world's three-dimensional coordinates are then used.) with pixel coordinates The mapping formula is:

[0042] In the formula, These are the pixel coordinates of the image; For the physical world's three-dimensional coordinates; This is the camera intrinsic parameter matrix; It is an extrinsic parameter matrix containing rotation and translation vectors; This provides depth information in the camera coordinate system. The depth information is calculated by inversely projecting the center point of the bottom edge of the cookware's bounding box onto the physical plane of the gas stove (assuming...). The actual physical coordinates of ) .

[0043] S10222, Calculate the physical coordinates. Center point of the temperature measurement field of view (FOV) with the built-in infrared temperature sensor array Euclidean distance Set the matching radius threshold to 100 mm. When When the data is measured in millimeters, it is confirmed that the target is the same in time and space, and the "visually extracted cookware" and the "heat source measured by infrared sensor" are bound and fused in the data structure.

[0044] S1023. When a vapor obstruction feature is detected or the ambient illuminance is lower than a preset threshold, the confidence weight of the visual image in data fusion is reduced, and the confidence weight of the multi-source sensor signal is increased simultaneously. S10231. Ambient illumination data is acquired from the automatic exposure (AE) gain value of the camera image sensor or from a separate photoresistor. The preset illumination threshold is set to 50 lux.

[0045] S10232. When the detected illuminance is below 50 Lux, or the vapor shading factor... When the accuracy exceeds 40%, visual target detection is prone to missed detections due to feature loss. In this case, the gas stove controller dynamically adjusts the observation noise covariance matrix in the Extended Kalman Filter (EKF). The specific operation is as follows: [The matrix is ​​then...] The diagonal element values ​​corresponding to the visual observation variables are exponentially amplified (equivalent to reducing the Kalman gain of visual information), while the matrix element values ​​corresponding to multi-source sensing signals such as infrared temperature and magnetic Hall probes are simultaneously reduced, making the filter more adaptively dependent on physical sensing data when updating its state.

[0046] S1024. Based on the adjusted confidence weights, the spatiotemporally aligned multi-source sensor signals and visual images are fused to obtain the in-situ features of the cookware, the presence features of the flame, the temperature change features, and the humidity trend features.

[0047] Spatiotemporal data fusion is performed using Kalman filtering. Its state prediction equation is: The state update equation is In the formula, for The prior state estimate vector at time t; This is the state transition matrix; for The posterior state estimate vector at time t; To control the input matrix; For control vectors; for The posterior state estimate vector updated by observation at each time step; The Kalman gain matrix; for The observation vector at any given time (containing the current observation values ​​extracted from multi-source sensor signals and visual images); The observation matrix is ​​defined by the above equations. The gas stove controller iteratively calculates and outputs the estimated state vector with the optimal posterior probability. This yields a highly robust feature output, effectively reducing the risk of misjudgment caused by the failure of a single sensor in a harsh kitchen environment.

[0048] S103, based on the characteristics of the cookware in place, the characteristics of the flame presence, the characteristics of temperature change, the characteristics of humidity trend and user operation information, divide the cooking process of the gas stove into multiple cooking states, and based on multiple evidence conditions within a preset continuous time window, determine the transition of the cooking process between cooking states. In one alternative implementation, when determining the presence of flame characteristics, changes in carbon dioxide concentration signals are used as auxiliary evidence of combustion. The verification process includes: S1031. The gas stove controller synchronously acquires the working status signal (including fan start / stop status and speed level) of kitchen ventilation equipment (such as range hood) through wireless local area network or short-range radio frequency communication protocol (such as Bluetooth, Zigbee).

[0049] S1032. Calculate the relative rate of change of carbon dioxide concentration signals and carbon monoxide concentration signals within the set time window. Due to methane ( When liquefied petroleum gas (LPG) or liquefied petroleum gas (LPG) is completely burned, according to the chemical reaction equation (taking methane as an example: This process is accompanied by the formation of carbon dioxide. Its relative rate of change... The calculation formula is as follows:

[0050] In the formula, The current carbon dioxide concentration value after baseline calibration following the aforementioned steps; This is the calibrated carbon dioxide concentration value from the previous sampling time. This is the time window step. Taking into account the transport delay of gas diffusion, The value range is set to 1.5 seconds to 3 seconds, preferably 2.0 seconds.

[0051] S1033. When the gas valve opening signal indicates that the valve is in the open state, and the thermocouple / ion probe does not detect the presence of flame due to local airflow, but the gas concentration (such as methane concentration) signal shows an upward trend for three consecutive sampling cycles, the gas stove controller constructs a combustion joint judgment matrix based on the calculated relative change rate and the working status signal.

[0052] S1034. Set the threshold for the relative change rate of carbon dioxide during normal ignition to 0.05 / second. When the actual calculated relative change rate of carbon dioxide is lower than 0.05 / second and the gas concentration signal continues to rise, it indicates that combustible gas has been discharged from the gas pipeline but no exothermic oxidation-reduction reaction has occurred, which helps to determine whether the gas stove has a substantial leak.

[0053] Specifically, when the operating status signal indicates that the kitchen ventilation equipment is on (such as when the strong exhaust setting is turned on), the air curtain effect created by the exhaust will quickly dilute the local gas concentration around the stove. At this time, the gas stove controller automatically adjusts the parameters of the joint judgment matrix, raising the safety judgment threshold (for example, relaxing the gas concentration trigger threshold for the unlit alarm to 1.5 times the default value), thereby preventing the normal physical phenomenon of gas dilution caused by strong airflow from being misjudged as flame extinguishing or gas leakage.

[0054] In one optional implementation, the multiple cooking states include at least a normal heating state and a suspected dry-burning state; based on multiple evidence constraints within a preset continuous time window, the transition of the cooking process between cooking states is determined, including: S1035, the gas stove controller calculates the first-order time derivative (i.e., the rate of temperature change) based on the temperature change characteristics, and divides the normal heating state into the ignition stage, the heating stage, the boiling stage, and the water evaporation stage according to the thermodynamic characteristics, and monitors the temporal evolution of the cooking process in the ignition stage, the heating stage, the boiling stage, and the water evaporation stage.

[0055] The characteristics of each stage are defined as follows: The ignition stage is characterized by a sudden rise in the stove surface temperature from room temperature to above 70℃; the heating stage is characterized by a stable slope of the temperature change rate of the pot bottom between 1.5℃ / second and 3.0℃ / second; the boiling stage is characterized by a phase transition isothermal plateau appearing near the standard boiling point of water (usually between 95℃ and 100℃ due to differences in altitude and air pressure), with the slope of the temperature change rate approaching 0; the water evaporation stage is characterized by a decrease in latent heat absorption and a break in the isothermal equilibrium before the free water in the pot is exhausted, causing the temperature to rise again.

[0056] S1036. When the gas stove controller determines that it is currently in the boiling stage, if the humidity trend characteristics acquired synchronously show an upward trend, the local humidity increase due to the large amount of water vaporization during boiling is in line with the normal principle of evaporation thermodynamics. This method determines that this is a normal evolution state, suppresses the triggering of any over-temperature alarm signal, and prevents it from disturbing the user's normal cooking.

[0057] S1037. When the state transitions from the moisture evaporation stage to a suspected dry-burning state, to reduce the false alarm rate, multiple evidentiary constraints within a continuous time window must be simultaneously met. These constraints include at least: First, after reaching a local maximum peak, the first derivative of the humidity trend characteristic turns negative and continues to decrease (physical meaning: the water in the pot has been completely boiled dry and can no longer provide water vapor). Second, the temperature change characteristics exceed the preset maximum safe slope. In this embodiment, the maximum safe slope is set to 2.5℃ / second (physical basis: the specific heat capacity of water is approximately 4200). The specific heat capacity of common pot bottom metals, iron and aluminum, is only 450~900. When the water boils away, all the heat is used to heat the metal pot, which has a very low specific heat capacity, causing the temperature rise rate to spike dramatically. Third, the user operation information extracted visually indicates that there is no human intervention (i.e., the target detection algorithm did not extract human arm features in the area above and around the cookware boundary box, and the gas valve opening sensor showed that the valve position remained constant without manual adjustment fluctuations).

[0058] S1038. The gas stove controller's state machine officially transfers the cooking process to the suspected dry-burning state only when the above three independent physical and operational evidence conditions overlap within a set continuous time window (ranging from 5 to 10 seconds, preferably 8 seconds in this embodiment to filter out temperature jumps caused by brief stirring actions). This multi-dimensional temporal coupling constraint effectively reduces false positive alarms caused by disturbances to a single sensor.

[0059] In one optional implementation, the multiple cooking states also include a flameout leakage state. The process of determining the transition between cooking states includes forced flameout leakage detection logic and state rollback correction logic. S1039, the forced flameout leakage judgment logic is as follows: When the fused pot's presence characteristics determine that the pot is actually present on the burner, but within a continuous 1.5-second detection cycle, the flame presence characteristic is detected to drop from the "present" logic state to "disappear" (this phenomenon usually corresponds to the physical extinguishing of the flame by overflowing liquid when cooking porridge or soup), and simultaneously, the gas concentration signal is detected to show an increasing trend, the gas stove controller determines that the situation has reached a high-risk level. At this time, bypassing all conventional smoothing filtering delay mechanisms, the state machine is forced to transition to the flameout leakage state and an emergency shut-off command is immediately issued.

[0060] S10310, the state rollback correction logic is as follows: To address the highly dynamic nature of Chinese cooking, such as stir-frying, if the gas stove controller has entered a suspected dry-burning state, within the delayed warning observation period before issuing the gas shut-off command (set to be within 10 seconds after the warning sound is issued), based on the visual and valve sensors re-capturing a clear human intervention action (e.g., recognizing continuous wok-tossing displacement, or detecting the addition of room-temperature ingredients to the pot causing heat loss), and the temperature change characteristic at this time changes from a sharp rise to a gradual decrease or a step drop, the controller's internal decision logic determines that the previous high temperature and dry-burning state were false alarms caused by normal cooking operations. The gas stove controller immediately cancels the warning flag for the suspected dry-burning state and safely rolls the cooking process state machine pointer back to the normal heating state (i.e., the heating stage or boiling stage).

[0061] In one alternative implementation, the lightweight anomaly model is constructed based on a multi-branch neural network to predict the evolution of the cooking process towards an abnormal state (especially a slowly accumulating dangerous state), including: S10311, The microcontroller periodically extracts historical continuous sampling data of multi-source sensor signals from the SRAM buffer, generating data with dimension [missing information]. A sequence of two-dimensional matrix tensors is used as the feature input for a lightweight anomaly model. Considering the SRAM capacity limitations of the edge NPU and the latency requirements of real-time inference, The preferred setting for the time window length is 60 sampling points (i.e., a time span of 6 seconds). The (number of feature channels) is set to 5, covering temperature, humidity, carbon monoxide concentration, gas concentration and valve opening.

[0062] S10312. The first feature extraction branch of the multi-branch network is used to extract local high-frequency oscillation features for short-term transient abrupt changes in temperature signals. This branch consists of three stacked 1D-CNN layers. To capture microsecond-level abrupt changes in the first and second derivatives of temperature (characterizing abrupt changes in the heat transfer properties of metal during dry burning or features of hot oil just before reaching its flash point), the kernel sizes of the three convolutional layers are set to 5, 3, and 3, respectively, with a stride of 1. The activation function used is the Gaussian Error Linear Unit (GELU), whose smooth and nonlinear characteristics can avoid gradient vanishing and retain weak temperature gradient information.

[0063] S10313. Utilizing the second feature extraction branch, long-term smoothing trend features are extracted for environmental humidity and carbon monoxide concentration signals. Considering that gas diffusion, exhaust gas accumulation, and water vapor propagation are typical slow-dynamic macroscopic physical processes, this branch employs a single-layer Long Short-Term Memory (LSTM) network. The LSTM's hidden state dimension is designed to be 32. Relying on the inherent forget gate, input gate, and output gate structure within the LSTM cell, the model can effectively remember and accumulate the slow upward trend of incomplete combustion byproducts such as carbon monoxide over the past tens of seconds.

[0064] S10314. In the feature fusion layer, the local high-frequency oscillation feature vector extracted from the first branch and the long-term smooth trend feature vector extracted from the second branch are concatenated using channel-dimensional tensors. The fused full features are input to a fully connected layer composed of two perceptron layers for nonlinear mapping. Finally, a multi-class prediction is performed using the Softmax activation function, outputting the probability distribution values ​​of the predicted cooking state evolving into various abnormal nodes. When the predicted probability corresponding to high-risk classification nodes such as "dry burning and fire evolution" or "precursor to excessive carbon monoxide" exceeds a preset safety threshold (preferably 0.85 in this embodiment), the hardware-level interrupt controller directly sends a highest-priority non-maskable interrupt to the microcontroller core, triggering a physical-level gas cut-off disaster prevention operation.

[0065] S104, trigger the corresponding hierarchical control strategy according to the current cooking state of the cooking process; In one alternative implementation, the hierarchical control strategy for situations where the pot is removed (such as when a user takes the pot to pour food or washes it, a typical scenario that is highly likely to cause the valve to close accidentally) is as follows: S1041. When it is confirmed that the bottom flame is present, but the cookware is determined to change from "in place" to "removed state" based on the multi-source fusion cookware presence characteristics, the gas stove controller determines that this behavior may be a brief operation. Therefore, it does not immediately trigger the valve shut-off control action to ensure the continuity of cooking, but instead starts the hardware timer and enters the transition holding stage with short-term holding time. S1042. During the transition holding phase, if continuous sampling shows that the pot's in-situ characteristics are still stably indicating the removed state (confirming that slight tossing motions at the level of several hundred milliseconds have been filtered out, with a duration of, for example, up to 3 seconds), then the microcontroller wakes up the closed-loop control unit and starts the step-by-step heat reduction strategy. S1043, a step-by-step flame reduction strategy drives the stepper motor-type electronic proportional control valve inside the gas stove to reduce the gas flow. This method uses a position-based discrete proportional-integral-derivative control algorithm, and its control law formula is:

[0066] In the formula, For the first The target number of pulses output to the stepper motor or the target valve opening angle; For the first The deviation between the current firepower status and the preset target value of the fire reduction curve step is recorded. This is the deviation from the previous beat; For the first Historical cumulative deviation of shooting; The gain constants are configured as proportional, integral, and derivative, respectively. These parameters are tuned using the critical proportional gain method to ensure a smooth flow regulation curve without overshoot, allowing the firepower to gradually and flexibly decrease over time, preventing flameout or backfire caused by a sudden, precipitous drop in gas flow. Simultaneously, the controller drives an external voice synthesis chip or a pulse-width modulation (PWM) buzzer via a universal asynchronous transceiver interface to issue an acoustic alert, informing the user that the system is currently in a standby, idling state. S1044. The gas stove controller continuously monitors the cumulative duration of the short-term hold timer. When this cumulative duration exceeds the preset protection limit and the cookware's presence characteristic has not yet been restored, the microcontroller pulls down the relay control pin level, cutting off the main solenoid valve of the gas supply pipeline. The preset protection limit is set to a range of 20 to 40 seconds, and in this embodiment, it is preferably 30 seconds. The technical basis for this is that ergonomic research shows that 30 seconds is the upper limit of time for most ordinary household users to complete the process of removing the pot, washing the cookware, and placing it back on the stove. Exceeding this upper limit can be confidently determined as a user forgetting to cook, posing a risk of dry burning.

[0067] S105, in one optional implementation, after triggering the hierarchical control strategy, a closed-loop verification is performed to determine whether the hierarchical control strategy was successfully executed, and a secondary intervention mechanism is executed if the verification fails, including the following sub-steps: If the valve core in the gas pipeline becomes mechanically stuck due to prolonged high temperature or heavy oil contamination, preventing the valve port from sealing properly, the one-way open-loop logic valve-closing command will fail to block the hazard. Therefore, this step introduces dual closed-loop verification at both physical and chemical levels: S1051. When the microcontroller issues an electronic command to close the gas valve, it immediately reads the actual operating mechanical angle position of the physical valve mechanical shaft from the position detection element (such as a high-resolution absolute Hall encoder) at the end of the valve body through a single bus or internal integrated circuit communication protocol, and inputs it as the gas valve position feedback signal into the gas stove controller. S1052. After confirming at the logic level that the gas valve position feedback signal has reached the physical dead point calibrated as fully closed, the gas stove controller calls the hardware delay module to delay the set period. The value of this delay period ranges from 1.0 seconds to 2.0 seconds, preferably 1.5 seconds. The technical basis is that it is necessary to wait for the trace amount of combustible gas remaining in the downstream pipeline of the shut-off valve to burn completely at the burner. After the delay ends, the microampere reading of the ionization flame probe is read. The leakage current value at level 1 is used as a flame signal to directly confirm, based on the physical fact of ionization, that the ionization zone has disappeared (i.e., the flame has been extinguished). S1053. Simultaneously, the ambient gas sensor module continuously analyzes whether the gas concentration signal exhibits an exponentially decreasing trend consistent with the expected diffusion thermodynamics. Based on the gas diffusion dissipation model under ideal conditions, the concentration after the leak stops... It should approximately follow an exponential decay law: In the formula, To stop the leak Real-time gas concentration at any given moment; To confirm the initial gas concentration at the moment the valve is closed; It is a natural constant; The attenuation constant is related to the kitchen ventilation rate; This is a time variable calculated from the time the valve is closed. The gas stove controller uses this time variable... The first-order time derivative was calculated to confirm that the concentration evolution trend was negative.

[0068] S1054, The gas stove controller performs a comprehensive logical AND operation on the above multi-dimensional feedback. When any of the following situations occur: The gas valve position feedback signal indicates that the mechanical shaft has not reached the predetermined dead-end position (indicating physical jamming); the flame signal amplitude after the delay period is still higher than the flameout threshold (indicating internal leakage due to rubber gasket damage, resulting in a persistent smoldering flame); the fitted evolution curve of the gas concentration signal does not show the expected exponential decreasing trend (indicating possible multi-point concurrent leakage at the pipeline front end or other nodes). Once any of these anomalies is triggered, the gas stove controller determines that the primary level control strategy has failed, entering a dangerous out-of-control state, and immediately activates a higher-level secondary intervention mechanism. S1055, the secondary intervention mechanism execution steps include: the main microcontroller, through a dedicated high-voltage optocoupler isolation relay drive circuit, sends a physical cut-off pulse to directly shut off the main air intake cut-off solenoid valve pre-installed at the front end of the indoor air intake pipe (near the main pipe), cutting off the gas supply from the physical source; simultaneously, it sends a broadcast message to the whole-house smart hub via dry contact or IoT protocol (such as Wi-Fi / Matter protocol), triggering the activation of all kitchen ventilation and exhaust equipment and forcibly adjusting it to the maximum exhaust volume level, forcibly performing air replacement to dilute any mixed gases that may have accumulated to the lower explosive limit or be toxic to humans. At the same time, it calls the communication module to push high-level red audible and visual warnings and fault diagnosis codes to the user's bound smart mobile terminal and the property management backend, achieving comprehensive closed-loop safety protection.

[0069] This invention also provides an intelligent safety monitoring and control system for gas stoves, as described in the following embodiments. Figure 3 The diagram shows a functional block diagram of an intelligent safety monitoring and control system 300 for a gas stove according to the present invention. The system may include the following modules: The multi-source signal acquisition module 301 is used to simultaneously acquire multi-source sensor signals, visual images, and user operation information output by multiple sensors of the gas stove; The consistency verification feature extraction module 302 is used to perform spatiotemporal consistency verification on multi-source sensor signals and visual images, and to fuse and extract features such as cookware presence, flame presence, temperature change, and humidity trend. The state evolution constraint module 303 is used to divide the cooking process of the gas stove into multiple cooking states based on the characteristics of the cookware in place, the characteristics of the flame presence, the characteristics of temperature change, the characteristics of humidity trend and user operation information, and to determine the transition of the cooking process between cooking states based on multiple evidence conditions within a preset continuous time window. The hierarchical control execution module 304 is used to trigger the corresponding hierarchical control strategy according to the current cooking state of the cooking process; The closed-loop feedback verification module 305 is used to read the gas valve position feedback signal, flame signal and gas concentration signal from the multi-source sensor signals after the hierarchical control strategy is triggered, to verify whether the hierarchical control strategy is executed successfully in a closed loop, and to execute a secondary intervention mechanism when the verification fails.

[0070] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. Memory, used to store computer programs; The processor, when executing the program stored in the memory, implements the intelligent safety monitoring and control method for gas stoves of the present invention.

[0071] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include RAM, or it can include NVM, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0072] The processors mentioned above can be general-purpose processors, including CPUs, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0073] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent safety monitoring and control method for gas stoves according to embodiments of the present invention.

[0074] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent safety monitoring and control of a gas stove, characterized in that, include: Simultaneously collect multi-source sensor signals, visual images, and user operation information output from multiple sensors of the gas stove; The spatiotemporal consistency of the multi-source sensor signals and the visual image is verified, and the in-situ features of the cookware, the presence of the flame, the temperature change features, and the humidity trend features are extracted and fused. Based on the characteristics of the cookware in place, the characteristics of the flame presence, the characteristics of temperature change, the characteristics of humidity trend, and the user operation information, the cooking process of the gas stove is divided into multiple cooking states, and the transition of the cooking process between the cooking states is determined based on multiple evidence conditions within a preset continuous time window. Based on the current cooking state of the cooking process, trigger the corresponding hierarchical control strategy; After triggering the hierarchical control strategy, the gas valve position feedback signal, flame signal and gas concentration signal in the multi-source sensor signals are read to verify whether the hierarchical control strategy is executed successfully in a closed loop, and a secondary intervention mechanism is executed if the verification fails.

2. The intelligent safety monitoring and control method for gas stoves according to claim 1, characterized in that, The multi-source sensor signals include the flame signal, the gas valve opening signal, the gas valve position feedback signal, the gas concentration signal, the ambient humidity signal, and the temperature signal; wherein, the gas concentration signal includes the gas concentration signal, the carbon monoxide concentration signal, and the carbon dioxide concentration signal. The method further includes effectiveness evaluation and baseline calibration of the plurality of sensors, specifically: The multi-source sensor signals are acquired in real time, and the signal fluctuation characteristics and temperature drift characteristics within the sliding time window are extracted. The effectiveness scores of the multiple sensors are calculated based on the signal fluctuation characteristics and the temperature drift characteristics. When the effectiveness score of a sensor is lower than a preset failure threshold, the data input of the corresponding sensor is blocked. A lightweight anomaly model is used to extract the joint rate of change features of the gas concentration signal, the carbon monoxide concentration signal, and the carbon dioxide concentration signal from the gas concentration signal as a multidimensional state feature vector. The multidimensional state feature vector is then compensated by combining pre-acquired kitchen environmental background data, the drift self-calibration baseline is dynamically updated, and the multi-source sensor signal is calibrated based on the drift self-calibration baseline.

3. The intelligent safety monitoring and control method for gas stoves according to claim 2, characterized in that, The step of performing spatiotemporal consistency verification between the multi-source sensor signals and the visual image, and fusing and extracting features such as cookware presence, flame presence, temperature change, and humidity trend, includes: The visual image is identified using a target detection algorithm, and the features of the cookware image and the steam obstruction feature are extracted. The position of the cookware image features in the image coordinate system is matched with the temperature signal acquisition area in the physical space by coordinate mapping. When the vapor obstruction feature is identified or the ambient illuminance is detected to be lower than a preset threshold, the confidence weight of the visual image in data fusion is reduced, and the confidence weight of the multi-source sensor signal is increased simultaneously. Based on the adjusted confidence weights, the spatiotemporally aligned multi-source sensor signals and the visual image are fused to obtain the in-situ features of the cookware, the presence features of the flame, the temperature change features, and the humidity trend features.

4. The intelligent safety monitoring and control method for gas stoves according to claim 2, characterized in that, When determining the presence of flame characteristics, the change in the carbon dioxide concentration signal is used as auxiliary evidence of combustion. The verification process includes: Simultaneously acquire the operating status signals of the kitchen ventilation equipment; Calculate the relative rate of change of the carbon dioxide concentration signal and the carbon monoxide concentration signal within a set time window; When the gas valve opening signal indicates that it is open and no flame characteristics are detected, but the gas concentration signal shows an upward trend, a combustion joint determination matrix is ​​constructed based on the relative change rate and the working status signal; When the relative change rate is lower than the normal ignition threshold and the gas concentration signal increases, the gas stove is identified as leaking; when the working status signal indicates that the kitchen ventilation equipment is on, the safety judgment threshold of the joint judgment matrix is ​​increased.

5. The intelligent safety monitoring and control method for gas stoves according to claim 1, characterized in that, The multiple cooking states include at least a normal heating state and a suspected dry-burning state; the determination of the transition of the cooking process between the cooking states based on multiple evidence constraints within a preset continuous time window includes: The normal heating state is divided into the ignition stage, the heating stage, the boiling stage, and the water evaporation stage, and the temporal evolution of the cooking process in the ignition stage, the heating stage, the boiling stage, and the water evaporation stage is monitored. When it is determined that the current stage is boiling and the humidity trend is increasing, it is determined to be a normal evolutionary state. When the transition from the moisture evaporation stage to the suspected dry burning state occurs, the multiple evidence conditions within the continuous time window include at least the following: the humidity trend characteristic turns to a continuous decline after reaching its peak, the temperature change characteristic exceeds the preset maximum safe slope, and the user operation information indicates no human intervention. The cooking process is transferred to the suspected dry-burning state if and only if multiple of the aforementioned evidentiary conditions overlap in time within the consecutive time window.

6. The intelligent safety monitoring and control method for gas stoves according to claim 1, characterized in that, The hierarchical control strategy for the case where the cookware is removed is as follows: When the presence of the flame is confirmed and the pot is determined to be in a removed state based on the pot's presence characteristics, a transition holding phase for short-term holding timer is triggered. During the transition holding phase, if the cookware presence feature still indicates the removed state, a step-by-step heat reduction strategy is initiated. The step-by-step flame reduction strategy drives the control element of the gas stove to reduce the gas flow, so that the flame level gradually decreases over time, and a reminder command is issued simultaneously. When the cumulative duration of the short-term holding time exceeds the preset protection limit and the cookware in-situ feature has not been restored to its present state, the gas supply pipeline is cut off.

7. The intelligent safety monitoring and control method for gas stoves according to claim 5, characterized in that, The multiple cooking states also include a flameout leakage state. In determining the transition between these cooking states, forced flameout leakage detection logic and state rollback correction logic are provided. The forced flameout leakage judgment logic is as follows: when the presence of the cookware is determined based on the cookware's presence characteristics, but during continuous detection, the flame presence characteristics are found to change from presence to disappearance, and the gas concentration signal is detected to show an increasing trend, the system is switched to the flameout leakage state and an emergency cut-off is triggered. The state rollback correction logic is as follows: if the system enters the suspected dry burning state, and during the delayed warning observation period, it re-captures the manual intervention action based on the user operation information, and the temperature change characteristic changes from rising to leveling off or falling, then the previous state judgment is determined to be a false alarm, the suspected dry burning state is revoked, and the cooking process is rolled back to the normal heating state.

8. The intelligent safety monitoring and control method for gas stoves according to claim 2, characterized in that, The lightweight anomaly model, built upon a multi-branch neural network, is used to determine the evolution of the cooking process towards an abnormal state, including: Historical continuous sampling data of the multi-source sensor signals are acquired, and a two-dimensional matrix sequence is generated as the feature input of the lightweight anomaly model. The first feature extraction branch is used to extract local high-frequency oscillation features for the short-term transient change features of the temperature signal; The second feature extraction branch is used to extract long-term smoothing trend features for the environmental humidity signal and the carbon monoxide concentration signal; The local high-frequency oscillation features and the long-term smooth trend features are fused and classified for prediction. When the predicted probability exceeds the safety threshold, a physical gas cut-off disaster prevention operation is triggered.

9. The intelligent safety monitoring and control method for gas stoves according to claim 1, characterized in that, The closed-loop verification confirms whether the hierarchical control strategy was successfully executed, and executes a secondary intervention mechanism when the verification fails, including: When a command to close the gas valve is issued, the mechanical operating position of the physical valve is read from the position detection element and used as the valve position feedback signal of the gas valve. After confirming that the gas valve position feedback signal has reached the closed dead point, the flame signal is read after a set delay period to confirm that the flame has disappeared; Simultaneously, analyze whether the gas concentration signal shows the expected downward trend; If any of the following occurs: the gas valve position feedback signal fails to reach the closing dead point, the flame signal indicates that the flame has not disappeared, or the gas concentration signal does not show the expected downward trend, the graded control strategy is determined to have failed and the secondary intervention mechanism is activated. The secondary intervention mechanism includes shutting off the main intake shut-off valve at the front end and coordinating with the ventilation equipment to exhaust air.

10. A smart safety monitoring and control system for a gas stove, characterized in that, include: The multi-source signal acquisition module is used to simultaneously acquire multi-source sensor signals, visual images, and user operation information output by multiple sensors of the gas stove; The consistency verification feature extraction module is used to perform spatiotemporal consistency verification between the multi-source sensor signals and the visual image, and to fuse and extract features such as cookware presence, flame presence, temperature change, and humidity trend. The state evolution constraint module is used to divide the cooking process of the gas stove into multiple cooking states based on the in-situ characteristics of the cookware, the presence characteristics of the flame, the temperature change characteristics, the humidity trend characteristics, and the user operation information, and to determine the transition of the cooking process between the cooking states based on multiple evidence conditions within a preset continuous time window. The hierarchical control execution module is used to trigger the corresponding hierarchical control strategy according to the current cooking state of the cooking process; The closed-loop feedback verification module is used to read the gas valve position feedback signal, flame signal and gas concentration signal from the multi-source sensor signals after the hierarchical control strategy is triggered, to verify whether the hierarchical control strategy is executed successfully in a closed loop, and to execute a secondary intervention mechanism when the verification fails.