Intelligent kitchen safety ventilation multi-mode alarm integrated system

Abstract

The invention discloses an intelligent kitchen safety ventilation multi-mode alarm integrated system, relates to the technical field of intelligent kitchen safety ventilation and multi-mode alarm, and is used for solving the problem that online judgment and establishment of an effective dilution path cannot be realized when a kitchen and a guest and restaurant are regarded as single ventilation areas. In the installation stage, the pollution source solidification area, the transition area and the occupancy area are distributed in three areas, and processed data frames are output at a fixed sampling rhythm. The data fusion processing module is used for carrying out equivalence on the multi-channel quantity, constructing a partition situation map, and combining quantitative indexes such as partition synchronization amplitude, delay correlation and arrival order with fan and opening feedback to form situation classification; a path effectiveness quantitative index and a differential pressure agent estimator are provided, an effective dilution path is established under the fence, orientation and acceleration strategies, and high air volume is entered only when the standard is reached and the direction is correct. According to the system, a unique control instruction set is generated by adopting double-track judgment and consistency constraint, and trace audit and self-adaptive optimization are realized by anti-fact evaluation and risk scoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent kitchen safety ventilation and multi-modal alarm, more specifically, the present application relates to an intelligent kitchen safety ventilation multi-modal alarm integrated system. BACKGROUND

[0002] In an open-plan house, the air flow in the kitchen and the dining room is affected by the superimposed influence of cooking heat, door and window opening and closing, air conditioning / fresh air supply and walking interference, and the pollution often appears out of sync fluctuation above the cooking table and in the occupied area; the wrong high air volume and window opening direction are easy to push the smell to the occupied area, resulting in poor body feeling, high energy consumption and alarm fatigue. The existing means mostly regard the kitchen and the dining room as a single ventilation area, and the control variables are concentrated in the fan speed and a single window opening angle, lacking the identification of the space zoning and the arrival order of the pollution source area-transition area-occupied area, and also lacking the online judgment and arrangement of the dilution path under the condition of no differential pressure sensor. The traditional threshold alarm is difficult to distinguish between real leakage and non-target stimulus source, and is easy to be frequently triggered and repeatedly reset under complex disturbance, affecting the user's stickiness and trust in the system. On the other hand, the remote intervention and local disposal lack consistency constraints and replayable evidence chain, making it difficult to audit and optimize parameters.

[0003] In view of the above problems, the present application provides a solution. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an intelligent kitchen safety ventilation multi-modal alarm integrated system to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] In a preferred embodiment, it comprises an environment perception module, a data fusion processing module, an active ventilation control module, a multi-modal alarm module, a remote interaction module and a power supply and self-checking module, and the modules are connected by signals;

[0007] The power supply and self-checking module is used for redundant switching standby power supply with mains and battery, and sequentially issues very short pulse to the fan and air valve according to short time on-off to check the driving and feedback, and then short chirp, short flash and low volume broadcast sound and light voice link for verification;

[0008] After the environmental perception module passes the self-test, it defines the S1 pollution source area, S2 transition area, and S3 occupied area. Then, it performs power-on stabilization, load calibration, and aligns the time base. It synchronously collects combustible gas volume fraction, CO concentration, smoke index, and temperature and humidity airflow according to S1, S2, and S3 and frames them with the local clock. Subsequently, it calculates the synchronization amplitude, delay correlation, and arrival order to form a situation after temperature and humidity compensation and noise reduction. It summarizes PV and sets quantile thresholds by combining the monotonicity and causal consistency of the fan stepping. At the same time, it constructs inlet dynamic pressure and window opening resistance proxy using power mapping and impedance table, and gives the enclosure zone and external pressure difference proxy by combining S2 airflow and gradient.

[0009] The data fusion processing module is used for temperature and humidity airflow compensation and resampling alignment of the time axis, and generates a unique instruction by taking the higher of the two-track judgment levels.

[0010] The active ventilation control module is used to establish a dilution path according to instructions and shut off the gas source at the danger level, and reset after a delay after verifying the dual conditions of concentration and decrease rate.

[0011] The active ventilation control module is used to trigger sound, light, and voice prompts according to the judgment level within the same judgment cycle and push them remotely with high priority;

[0012] The remote interaction module is used to aggregate status and approve permissions for shutdown and wind turbine control commands, generate transaction numbers, issue execution and receipts after consistency checks, and perform gray-scale updates to the model and prompts.

[0013] In a preferred embodiment, continuous power supply preparation is completed through a redundant switching circuit between mains power and backup battery. Then, extremely short pulses are sequentially sent to the exhaust fan actuator and the air source solenoid valve actuator according to the short-time turn-on process to verify that the drive stage and feedback channel are normal. Subsequently, the multimodal alarm module uses a short-sounding buzzer, a short-flashing indicator light, and plays a low-volume voice sample to confirm the integrity of the sound, light, and voice link.

[0014] In a preferred embodiment, after receiving the self-test pass signal from the power supply and self-test module, the baseline is first solidified through a power-on stabilization process, then the calibration coefficient is loaded and the time reference is aligned, and then the volume fraction of methane / propane is obtained from combustible gas, the CO concentration is obtained from carbon monoxide, the smoke index is calculated from smoke or particulate matter, and environmental quantities such as temperature, humidity and airflow are obtained.

[0015] In a preferred embodiment, points are deployed and fixed in three zones, defining S1 pollution source zone, S2 transition zone, and S3 occupied zone. The smoke or particulate matter index, local airflow velocity, temperature and humidity of the three zones are aligned with the local real-time clock before outputting the processed data frame. In each judgment cycle, temperature and humidity compensation and noise reduction are performed to obtain a cleaning sequence. The synchronization amplitude, delay correlation coefficient and arrival order consistency of the zones are calculated. A situation classifier is constructed by combining the target and feedback speed of the fan and the target and feedback opening of the ventilation components, and the threshold is adaptively updated according to the health day quantile.

[0016] The path effectiveness quantification index PV is summarized by the monotonicity of the decrease in S1 to S2 to S3 under the fan step trigger, the arrival order and causal consistency, and the threshold is fine-tuned according to the offline quantile and incorporated into the log; the inlet dynamic pressure agent is estimated by fan power mapping under the condition of no differential pressure sensor, the window opening resistance agent is calculated by combining the opening degree of ventilation components with the impedance table, and the transient relationship between the S2 airflow and the concentration gradients of S1 and S2 is given for the enclosure zone Δp and the external Δp.

[0017] In a preferred embodiment, when entering the closed-loop arrangement, if the situation is reverse flow or insufficient PV, the kitchen side is opened at a small angle and the living room side is suppressed according to the unique instruction sequence of the enclosure directional speed increase, and the fan is limited to low and medium speed to stabilize the enclosure belt Δp*.

[0018] Subsequently, the fan speed is increased in a step cycle and the kitchen side opening is slightly increased to drive the gradual monotonic reduction from S1 to S2. High air volume is only switched in when PV meets the standard, the external Δp direction is correct, and the intensity reaches the minimum agent intensity for entering high air volume. Otherwise, window sash swapping or short pulse strong exhaust is prioritized to break the stagnation. At the end of each cycle, the gain is adjusted based on Δp, PV, and zone reduction.

[0019] In a preferred embodiment, the actual trajectory and alternative trajectory are generated based on the same event, and the exposure integral of the occupied area, energy consumption proxy and noise proxy are calculated. The remote interaction module selects the minimum regret sequence as the preferred arrangement for the next event along the offline Pareto front under the user preference boundary. When the external disturbance causes the PV to drop but the pollution source does not expand, the fan micro-step probe is initiated and the three-zone reduction response signature is compared within a short observation window. If they do not match, the window is switched and retried. If the standard is not met twice in a row, the conservative arrangement is returned.

[0020] In a preferred embodiment, after receiving and processing the data frame, cross-channel synchronization compensation is first performed using temperature, humidity and airflow context, and then the time axis is aligned using short-window denoising and uniform step size resampling, and the values ​​of combustible gas, carbon monoxide and smoke are equivalent to the comparable concentrations and indices of the current operating conditions.

[0021] Subsequently, a dual-track approach of hierarchical threshold rules and pattern recognition is used to give the security level and leakage confidence. When the conclusions are inconsistent, a higher level of consistency constraint is applied and a unique set of control instructions is generated.

[0022] The active ventilation control module drives the exhaust fan and ventilation components to establish a dilution path according to the unique control instruction set. Under dangerous conditions, it then drives the gas source solenoid valve to cut off the gas supply. After that, it performs closed-loop verification based on both absolute concentration and drop rate and executes a forced ventilation delay with a fixed upper limit.

[0023] In a preferred embodiment, a multimodal alarm is triggered according to the alarm level output by the data fusion module within the same judgment period, causing the buzzer to sound rhythmically, the indicator light to flash at a frequency, and the voice command to be broadcast. Then, the alarm level, trigger factor, and handling timeline are pushed to the terminal with high priority through remote interaction. If the release condition is not met, the alarm is pushed at the set frequency until it is released or the authorized user resets it.

[0024] The remote interactive side encrypts and aggregates measurement and handling status, verifies remote shutdown and fan start / stop speed control commands according to role permissions, generates transaction numbers, submits them for consistency checks, and then hands them over to the active ventilation control module for execution. It then writes back success or failure, timestamp, parameters and equipment feedback, and distributes model and prompt update packages in a gray-scale manner.

[0025] The technical effects and advantages of the intelligent kitchen safety ventilation multimodal alarm integrated system of this invention are as follows:

[0026] This invention constructs a zoned situation map for the same judgment period through three-zone deployment and equivalent processing. PV transforms monotonicity, arrival order, and causal consistency into hard criteria for entering high-volume areas. Supplemented by Δp*, which represents a proxy estimation of the enclosure zone and outward direction, it achieves a staging logic of first enclosure, then orientation, and finally acceleration, thereby shortening the effective dilution path establishment time and reducing the exposure of occupied areas. Dual-track judgment combined with consistency constraints outputs a unique control instruction set, reducing multi-caliber conflicts and improving interpretability. Counterfactual J is used to make the least regret choice among exposure, energy consumption, and noise, and, in conjunction with remote interaction, enables controlled parameter updates and user preference constraints. Full-link event logs, judgment logs, handling logs, and maintenance records form replayable evidence, supporting auditing and rollback. The system exhibits robustness and adaptability under complex disturbances, alleviating alarm fatigue and improving user stickiness. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the integrated intelligent kitchen safety ventilation multimodal alarm system module of the present invention.

[0028] Figure 2 This is a timing diagram of the intelligent kitchen safety ventilation multimodal alarm integrated system of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example

[0031] This invention discloses an integrated multi-modal alarm system for intelligent kitchen safety ventilation, such as... Figure 1 As shown, it includes: an environmental sensing module, a data fusion processing module, an active ventilation control module, a multimodal alarm module, a remote interaction module, and a power supply and self-test module, with signal connections between each module.

[0032] like Figure 2 As shown, the power supply and self-test module first prepares for continuous power supply through the redundancy switching circuit of the mains power and the backup battery, and then confirms the availability of key components through a short-term turnaround process.

[0033] Specifically, the power supply and self-test module first sends extremely short pulses to the exhaust fan actuator and the air source solenoid valve actuator to verify that the drive stage and feedback channel are normal. Then, the multi-modal alarm module uses a short-beep buzzer and a short-flash indicator light and plays a low-volume voice sample to verify the integrity of the sound, light and voice links. Finally, the remote interaction module reads and displays the above self-test results.

[0034] After receiving the self-test pass signal from the power supply and self-test module, the environmental perception module first achieves baseline stabilization through the sensor power-on stabilization process, then achieves traceable sampling by loading calibration coefficients and aligning with the time reference, and then outputs processed data frames to the data fusion processing module at a fixed sampling rhythm.

[0035] Specifically, the environmental sensing module consists of a combustible gas sensing unit, a carbon monoxide sensing unit, a smoke / particulate matter sensing unit, and temperature, humidity, and airflow sensing units. The environmental sensing module first acquires the volume fraction signal of methane or propane through the metal oxide (MOX) or microelectromechanical system (MEMS) sensitive chip of the combustible gas sensing unit, and converts the analog quantity into a digital quantity through the isothermal and pressure drive and ΔΣ analog-to-digital conversion of the local analog front-end. Then, it acquires the carbon monoxide mass concentration through the electrochemical sensor of the carbon monoxide sensing unit, and obtains the digital quantity after temperature drift suppression through a built-in transimpedance amplification and temperature compensation network.

[0036] Subsequently, the scattering intensity of smoke / PM is collected by the optical scattering probe of the smoke / particulate matter sensing unit and converted into a smoke index. At the same time, the ambient temperature, relative humidity and local airflow velocity are obtained by the temperature and humidity and airflow sensing units.

[0037] After the environmental sensing module completes the synchronous acquisition of the above four types of physical quantities, it then uses two or more calibration coefficients loaded during the installation phase to achieve the engineering standardization of the numerical domain, and uses a local real-time clock to achieve unified encoding of the timestamp, ultimately forming a processed data frame containing quantitative values, units of measurement, timestamps, and channel numbers.

[0038] It should be noted that in open-plan apartments where the living room, dining room, and kitchen are connected, the combined effects of intense cooking during dinner, simultaneous foot traffic with open doors, and altered indoor airflow due to air conditioning / fresh air supply often result in a sawtooth pattern in the smoke / particulate matter index and local airflow velocity measured by the environmental sensing module above the stove. This pattern involves a rapid rise, a brief drop, and then another rise. Simultaneously, due to limitations in perceived height and return flow paths in the living room seating area, the relevant sensing channels exhibit a lagging rise. Furthermore, after the kitchen exhaust fan operates at high speed, backflow vortices easily form at doorways and corridors. Internal factors include the heat plume from the stove and the shear layer along the lower edge of the upper cabinets, as well as the dynamic pressure and structural impedance at the exhaust hood inlet. External factors include the opening and closing status of the living room side windows / balcony doors, and the direction and volume of the air conditioning supply. The superposition of internal and external factors often leads to asynchronous concentration fluctuations in the pollution source area, above the stove, in the occupied area, and in the living room breathing zone within the same sampling window. Moreover, the concentration in the occupied area may even temporarily increase during high-speed fan operation.

[0039] Current methods treat the kitchen and living / dining area as a single ventilation zone by default, which has two structural shortcomings:

[0040] First, the spatial zoning and arrival order of the pollution source area, transition area, and occupied area were not identified and verified, and there was a lack of online judgment on whether an effective ventilation path had been established.

[0041] Secondly, the control actions are centered on the fan speed and a single window opening angle, lacking the preliminary steps of differential pressure barriers and path guidance.

[0042] This leads to three possible outcomes in open-plan apartments:

[0043] High airflow combined with incorrect window opening direction causes odors to spread into the occupied area, resulting in a worse user experience.

[0044] Energy consumption per unit time increased significantly, but the concentration in the occupied area decreased slowly, and the delayed reset was triggered multiple times.

[0045] Multimodal alarms are repeatedly and passively re-sensed in occupied areas, leading to alarm fatigue. Users tend to manually downgrade or even shut down the system, reducing overall user stickiness.

[0046] Therefore, in this embodiment, firstly, during the installation phase, the environmental sensing module completes the three-zone layout and fixes the points: the measuring point 0.6–0.8m above the stove is defined as the pollution source zone S1, the measuring point at the kitchen doorway / passageway is defined as the transition zone S2, and the measuring point in the breathing zone of the living room seating area is defined as the occupied zone S3; the smoke / particulate matter index, local airflow velocity, and temperature and humidity of the three zones are time-aligned under the local real-time clock and form a processing data frame.

[0047] Based on this, for each judgment period, the data fusion processing module constructs a partitioned situation map for {S1,S2,S3}: first, a cleaning sequence is obtained through temperature and humidity compensation and noise reduction, then (i) partition synchronization amplitude, (ii) delay correlation coefficient, and (iii) consistency score of arrival order are calculated, and (iv) fan target / feedback speed and (v) ventilation component target / feedback opening are combined, where (iv) fan target / feedback speed and (v) ventilation component target / feedback opening are both from the active ventilation control module log. First, a situation classifier is constructed using these quantities to divide the current moment into three states: forward dissipation, reverse return flow, or stagnant flow without a path. Then, by robustly estimating the quantile interval of the most recent few healthy days, the synchronization and delay thresholds are adaptively updated so that the classifier can drift slowly with changes in apartment type and season without going out of bounds.

[0048] The most recent decision window w can be composed of several sampling windows, for example, selected according to the strength of scene interference; the partition synchronization amplitude is calculated based on the robust synchronization degree of the first-order difference within the window, and is defined as: ;

[0049] Where IQR is the interquartile range and ε is a minimum positive number used for zero-prevention. The delay correlation coefficient is obtained by peak cross-correlation to obtain the lag from S1 to S2 and from S2 to S3. Consistency of arrival order The value is 1 if the data falls within a reasonable range given during the installation phase or health day statistics; otherwise, it is 0. The corresponding threshold can be updated adaptively based on health days, for example, selecting days without alarms, with quantiles updated to ensure that A aligns with... The threshold shifts slowly with the type of apartment and the seasons without crossing the line.

[0050] Secondly, to avoid judging whether an effective path has been established solely at the qualitative level, a quantitative index of path effectiveness, PV, is constructed using the monotonicity of the unit-time decrease along the path from S1 to S2 and then to S3 within the same observation window, the consistency of arrival order, and the coupling degree of the turbine stepping response. Specifically, monotonicity is first measured by the temporal relationship between the turbine stepping, the decrease in S1, the decrease in S2, and the decrease in S3; then, arrival order is measured by the delay relationship between S1 leading S2 and S2 leading S3.

[0051] Subsequently, the causal consistency of the ventilator step-up to the zonal reduction is measured by the coupling degree; the three are aggregated into a PV score in the data fusion processing module with normalized weights and written to the event log. When the PV exceeds the threshold Ypv, it is considered that an effective ventilation path has been established. The threshold Ypv is adaptively fine-tuned in the offline window using a robust quantile method. The adjustment record is backed up by the remote interaction module for auditing and rollback.

[0052] Specifically, the wind turbine stepping event is defined as... The process is triggered when the relative change within a single sampling window reaches a preset step threshold; the step time is denoted as t*. Monotonicity score. After pressing t, w is inside Scoring is done for each zone without ascending, and the average of the three zones is taken; scores are awarded based on arrival order. exist Otherwise 0, its For the first time the reduction threshold is reached from t At that moment, The score can be adaptively determined by the historical quantiles of health days, for example, by taking the middle to high quantiles to enhance robustness; causal consistency score exist[ The direction of decrease in each sub-region and Δ 1 if consistent, 0 otherwise; The weights can be finely adjusted in a controlled manner within the remote interaction module.

[0053] Furthermore, to enable differential pressure programming even without differential pressure sensors, a differential pressure proxy estimator is constructed within the data fusion processing module: first, the inlet dynamic pressure is estimated by using the fan speed and motor current feedback, or PWM duty cycle and fan power mapping; then, the effective window opening resistance is calculated by using the ventilation component feedback opening degree and the impedance table calibrated before mass production; and finally, the local pressure difference trend is estimated by using the transient relationship between the S2 airflow velocity and the S1 and S2 concentration gradients, resulting in two proxy quantities: Δp* for the enclosure zone and Δp* for the external zone.

[0054] Among them, the inlet dynamic pressure agent Wind turbine power mapping table or The window opening resistance agent was found. Depend on via impedance meter Conversion; Local pressure difference trend estimated using the transient relationship between S2 airflow velocity and S1 and S2 concentration gradients: ,

[0055] Where α and β are dimensionless weights, for example, initial values ​​can be obtained by minimizing the false positive rate from daily health data offline and then periodically calibrated; then, based on these, calculations are performed. and Both are dimensionless substitutes; positive and negative indicate direction, and amplitude indicates relative intensity.

[0056] Then, it enters closed-loop orchestration. When the situation is classified as reverse backflow or PV is below the threshold, the original plan of directly increasing airflow is not executed. Instead, the active ventilation control module is first triggered through a unique control instruction set to execute three steps: enclosure, orientation, and acceleration.

[0057] The first step is to open the kitchen-side ventilation components at a small angle and suppress the opening on the living room side, limiting the fan to low and medium speed, and using the enclosure to stabilize Δp* within the target range;

[0058] The second step is to gradually increase the fan speed in a step cycle and slightly increase the opening of the kitchen side, so that the ΔC / Δt given by the three-zone curve of the environmental perception module along S1 to S2 gradually becomes monotonous.

[0059] The third step only enters the high air volume range when PV meets the standard and the external Δp* direction is correct.

[0060] To avoid accidental entry into high-airflow areas that could cause a temporary increase in the occupied area, the entry condition is quantified as: PV ≥ ,and The direction is consistent with the exhaust direction from the kitchen to the outside, where sign is positive, and its amplitude is not less than the minimum agent intensity γ used for high air volume. γ can be automatically generated by the installation phase or health day statistics and checked periodically. If any condition is not met, the enclosure should be maintained and window sash replacement or short pulse strong exhaust should be tried first. The pulse width and step amplitude can be set to small disturbances according to the equipment safety boundary to break the stagnation, and then checked again.

[0061] Subsequently, after each cycle, the data fusion processing module uses the enclosure band Δp*, PV, and zone reduction reading results to perform closed-loop gain adjustment: if PV is still insufficient, the enclosure is retained and window sash swapping or short pulse strong exhaust is attempted to break the stagnation; if PV significantly exceeds the standard and the occupancy area reduces too quickly, causing a feeling of dryness, the air volume can be reduced by one level in the next window to save energy.

[0062] Outside the closed loop, two counterfactual trajectories are constructed on the same historical segment of the event. One is the actual trajectory executed according to the current enclosure and acceleration sequence, and the other is an alternative trajectory, such as orientation before enclosure or short pulse strong placement and long weak placement curve. The input of both trajectories comes from existing logs, parameter tables and three-zone concentration curves. Any unobserved measurements are substituted with mass production calibration or values ​​recorded in the previous cycle. Then, three indicators are calculated on the two trajectories in the data fusion processing module: (i) the exposure integral of the occupied area, i.e., the S3 concentration integral over time; (ii) energy consumption proxy, such as: fan power, time integral and opening power consumption weighted, calculated by fan speed, sound power calibration and door gap attenuation table; (iii) noise proxy. The three together constitute J = {exposure, energy consumption, noise}.

[0063] Among them, the exposure score of the occupied area Determined by the trigger and deactivation timeline of the unique control instruction set; energy consumption agent Where λ is a trade-off coefficient, which can be automatically set by energy consumption calibration during the installation phase or by daily energy consumption distribution during healthy days; noise proxy ,in, The sound power of the wind turbine is calibrated and converted from the door gap attenuation table; under the constraint of user preference boundary, the remote interaction module selects the minimum regret sequence as the preferred arrangement for the next event on the Pareto front obtained offline in advance.

[0064] Furthermore, under the constraints of user preference boundaries configured in the remote interaction module, such as the upper limit of nighttime noise, the data fusion processing module selects the minimum regret sequence that best matches the current situation from the Pareto front obtained offline in advance, and uses it as the preferred arrangement for the next event; if the user is in the nighttime quiet zone, the noise constraint is reserved first on the Pareto surface, and a trade-off is made between exposure and energy consumption.

[0065] Subsequently, to enhance robustness to external disturbances such as air conditioning airflow direction switching and temporary opening of balcony doors, a disturbance probe micro-test was conducted to verify path judgment without user perception: when PV drops below the threshold but the pollution source has not yet expanded, the floor fan speed is briefly increased by a small step, and the response signature of the {S1,S2,S3} decrease is monitored within 2–3 sampling windows; if the signature does not meet the typical templates of enclosure, orientation, and speed increase, the alternative path of window sash swapping is automatically switched, and the micro-test is executed again.

[0066] Among them, the micro-stepping amplitude of the wind turbine Set to a small relative change and a very short observation window; the response signature and hoarding-orientation-acceleration template are evaluated using the Pearson correlation coefficient or an equivalent robust similarity index. The similarity threshold η is determined by the health day statistics and can be controlled and adjusted. If it is lower than η, the window sash will be automatically swapped and retried once; if it fails to meet the standard twice in a row, it will fall back to conservative arrangement and be written to the event log.

[0067] It should be noted that the amplitude and duration of such probes are constrained by the power supply and self-test module's power and health status. When the power is low or the fan temperature rises too high, the probes will automatically stop and revert to conservative programming.

[0068] After receiving the data frame, the data fusion processing module first achieves cross-channel synchronous compensation through temperature, humidity and airflow context, then achieves consistency in the time domain through short-window denoising and unified step-size resampling, and finally achieves security level determination through the parallel implementation of hierarchical threshold rules and pattern recognition algorithms.

[0069] To ensure comparability between different channels at the same decision time, a fixed sampling rhythm is defined as the sampling period. For example, a fixed value can be determined based on the self-inspection records during the installation phase, and a unified timestamp can be used. For each partition i∈{S1,S2,S3}, the volume fraction of combustible gas is... carbon monoxide mass concentration Smoke / Particulate Matter Index At the current temperature relative humidity With local airflow velocity Equivalence is performed within the ternary context to obtain The equivalent function loads the temperature, humidity, and sensitivity mapping tables that are fixed during the factory and installation stages; time-domain consistency can be achieved by median filtering and triple-σ limiting to suppress transient spikes, with a fixed step size. Align the samples, and for missing measurement points, use the most recent valid value to preserve or smooth extrapolate from the same sequence.

[0070] When discrepancies arise between the two tracks, a unique set of control instructions is generated with a consistency constraint that tends towards safety.

[0071] Specifically, the hierarchical threshold rule track is... Three threshold curves are set for safety, warning, and danger levels respectively; pattern recognition track is used. As a feature vector, the leakage confidence M(t)∈[0,1] and category candidates are output; when the conclusions of the two tracks are inconsistent, the higher alarm level is taken as the unique alarm level, and the triggering factor, evidence items and threshold / model version are written to the event log to improve interpretability.

[0072] Specifically, the data fusion processing module first compensates for the values ​​of combustible gas, carbon monoxide and smoke through the three-dimensional context of temperature, humidity and airflow. It first loads the temperature and humidity and sensitivity mapping table that is fixed at the factory and during the installation stage, and then corrects the response delay caused by diffusion hysteresis according to the current airflow speed, so as to obtain the equivalent concentration comparable under the current operating conditions.

[0073] Subsequently, the data fusion processing module unifies the samples from different channels to the same judgment time by time alignment with a fixed step size, and suppresses transient spikes with median filtering and triple σ limiting to form a clean sequence of combustible gas equivalent concentration, carbon monoxide equivalent concentration and smoke equivalent index.

[0074] Next, the data fusion processing module first completes a deterministic judgment by using preset safety, warning, and danger level three threshold curves that have been solidified through type testing and on-site debugging, thereby achieving rapid classification of obvious boundary violations or obviously safe scenarios. Then, by inputting a feature vector composed of combustible gas equivalent concentration, carbon monoxide equivalent concentration, smoke equivalent index, temperature, humidity, and airflow into a pattern recognition classifier trained offline, it completes the distinction between real leaks and non-target stimuli, such as cleaning agent vapors and cooking aerosols.

[0075] After the two-track determination is completed, the data fusion processing module uses consistency constraints to first lock in the minimum safety boundary with conservative determination, then improves interpretability with evidence item interpretation, and finally generates a unique alarm level and a unique set of control instructions.

[0076] Furthermore, to ensure consistency between internal interpretation and auditing, the data fusion processing module subsequently summarizes evidence through a one-time risk score, and then writes the triggering factors, evidence items, and control parameters into the event log using a parameterized template. The risk score is only used for record-keeping and does not replace threshold determination; furthermore, to ensure consistency between internal interpretation and auditing, the data fusion processing module summarizes evidence through a one-time risk score R(t); this score does not replace threshold determination and pattern recognition classification, and is only used for event replay and auditing. Let... Let M(t)∈[0,1] be the equivalent quantities normalized to [0,1] based on the percentage of the lower explosive limit / health limit, respectively, and M(t)∈[0,1] be the leakage confidence level of the pattern recognition output. ;

[0077] in, These are non-negative weights, which can be obtained through ROC objective optimization of the controlled dataset, and the parameter version number and effective timestamp can be recorded.

[0078] After receiving the unique set of control commands from the data fusion processing module, the active ventilation control module first establishes an effective gas dilution path through the exhaust fan actuator and the ventilation component actuator, then cuts off the gas supply at the hazardous level through the gas source solenoid valve actuator, and finally achieves closed-loop verification and delayed reset of the treatment through the dual conditions of absolute concentration and decrease rate.

[0079] The absolute concentration release threshold can be set according to a conservative proportion of the safety threshold for each channel, and the decrease rate release threshold can be set according to the linear regression slope of the most recent sampling windows being less than a certain negative threshold. The specific values ​​can be automatically generated from daily health statistics. Forced ventilation delay is executed after both conditions are met simultaneously. During the delay period, the exhaust fan and ventilation components maintain a medium speed and a medium opening until the delay window ends, and then gradually return to normal or shut down according to the strategy.

[0080] Specifically, the active ventilation control module first drives the frequency converter or PWM speed controller according to the target fan speed command in the control command set, so that the exhaust fan is increased from standstill or low speed to the target speed; then it drives the electric window or electric louver according to the target opening command of the ventilation component, so that the exhaust path forms a cross-interface flow that matches the fan suction; when the alarm level is dangerous, the active ventilation control module then cuts off the air supply according to the target status command of the air source solenoid valve, so that the source of concentration increase is eliminated at the same time.

[0081] Subsequently, the active ventilation control module does not immediately enter the reset state. Instead, it first calculates the concentration decrease rate by reading back the latest processed data frame from the environmental perception module. Then, it uses the absolute concentration falling back to the safe threshold and the decrease rate reaching the release threshold as the basis for dual-condition judgment. Only when both conditions are met simultaneously will it enter the delayed reset state.

[0082] To prevent residual gas from flowing back, the active ventilation control module first performs a forced ventilation delay with a fixed upper limit before resetting, allowing the exhaust fan and ventilation components to continue running at medium speed and medium opening until the delay window ends, and then gradually returns to normal or stops according to the strategy.

[0083] Meanwhile, throughout the process, the active ventilation control module writes the speed change, opening change, valve status change and timestamp into the handling log, and uses the event primary key to correspond one-to-one with the judgment log of the data fusion processing module and the measurement log of the environmental perception module, so as to ensure that the handling sequence of first detecting the boundary, then establishing the dilution path, then cutting off the gas source, then closing the loop for verification, and then delaying the reset can be completely replayed.

[0084] The multimodal alarm module and the active ventilation control module execute in parallel within the same judgment cycle: the multimodal alarm module first achieves immediate arrival through local sound and light signals, then provides semantically clear behavioral guidance through voice broadcast, and subsequently achieves synchronous perception for remote users through high-priority push from the remote interaction module.

[0085] Specifically, the multimodal alarm module first loads the corresponding prompt template based on the alarm level given by the data fusion processing module, causing the buzzer to sound at the corresponding rhythm and the indicator light to flash at the corresponding frequency. The voice broadcast unit then broadcasts Chinese instruction prompts, such as "Combustible gas detected, please ventilate immediately and stay away from the fire source." The alarm level, trigger factor, handling action, and timeline are then pushed to the mobile terminal or the monitoring terminal via a high-priority pop-up window through the remote interaction module. If the handling does not meet the conditions for deactivation, the multimodal alarm module continues to push at the set frequency until the data fusion processing module issues a deactivation decision or an authorized user completes the reset.

[0086] To prevent conflicts caused by multiple interpretations, the multimodal alarm module does not repeatedly set thresholds or make independent judgments locally. Instead, it uses the level of the data fusion processing module as the sole triggering basis and writes each trigger, duration, and termination point back to the event log, forming a parallel trajectory with the handling log.

[0087] The remote interaction module plays a closed-loop role in the above process, including data aggregation, permission verification, and manual intervention.

[0088] Specifically, the remote interaction module first aggregates the equivalent concentration curve of the environmental perception module, the alarm level and triggering factor of the data fusion processing module, the execution status of the active ventilation control module, and the health information of the power supply and self-test module into a unified interface through encrypted communication. Then, it realizes the compliance review of intervention instructions through roles and permission matrix. Subsequently, the approved instructions are packaged into controlled transactions and issued for execution. After the execution is completed, the receipt and the scope of impact are written back.

[0089] When an authorized user issues a command to remotely shut off the gas supply or remotely start / stop / adjust the speed of the exhaust fan, the remote interaction module first verifies the user's identity and the operation context, such as whether the current situation is dangerous. Then, it generates a transaction number with the timestamp, user identifier, and target device identifier and submits it to the data fusion processing module for consistency check. Only after the consistency check is passed will the active ventilation control module execute the command.

[0090] The execution receipt, including success / failure, timestamp, execution parameters, and device feedback, is written to the event log to ensure that every operation under human-machine co-governance can be audited and revoked. Simultaneously, the remote interaction module uses a phased rollout strategy to distribute model update packages and prompt templates in a controlled manner, enabling pattern recognition and prompts to adapt to different kitchen interference spectra and user characteristics without altering established safety boundaries.

[0091] The power supply and distribution module and self-test module are used throughout the entire system life cycle, and complete health diagnosis and maintenance prompts at a fixed rhythm without interrupting monitoring and handling.

[0092] Specifically, the power supply and self-test module first initiates a self-test during off-peak hours, and then writes the sensor channel availability, short-term connection of the fan and solenoid valve, and short sound and voice alarms into the self-test log in sequence. Subsequently, one-time events such as battery health, backup power intervention, and component replacement are written into the maintenance file with timestamps and responsibility fields, thereby providing comparative evidence of whether the equipment was healthy at the time of any alarm or handling.

[0093] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0094] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0095] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0096] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0098] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated multi-modal alarm system for intelligent kitchen safety ventilation, characterized in that: include: The system includes an environmental sensing module, a data fusion and processing module, an active ventilation control module, a multimodal alarm module, a remote interaction module, and a power supply and self-test module, with signal connections between each module. The power supply and self-test module is used to switch between mains power and battery redundant backup power. It sends extremely short pulses to the fan and air valve in sequence to check the drive and feedback, and then uses short beeps and flashes and low volume broadcasts to verify the audio-visual link. After the environmental perception module passes the self-test, it defines the S1 pollution source area, S2 transition area, and S3 occupied area. Then, it performs power-on stabilization, load calibration, and aligns the time base. It synchronously collects combustible gas volume fraction, CO concentration, smoke index, and temperature and humidity airflow according to S1, S2, and S3 and frames them with the local clock. Subsequently, it calculates the synchronization amplitude, delay correlation, and arrival order to form a situation after temperature and humidity compensation and noise reduction. It summarizes PV and sets quantile thresholds by combining the monotonicity and causal consistency of the fan stepping. At the same time, it constructs inlet dynamic pressure and window opening resistance proxy using power mapping and impedance table, and gives the enclosure zone and external pressure difference proxy by combining S2 airflow and gradient. The data fusion processing module is used for temperature and humidity airflow compensation and resampling alignment of the time axis, and generates a unique instruction by taking the higher of the two-track judgment levels. The active ventilation control module is used to establish a dilution path according to instructions and shut off the gas source at the danger level, and reset after a delay after verifying the dual conditions of concentration and decrease rate. The active ventilation control module is used to trigger sound, light, and voice prompts according to the judgment level within the same judgment cycle and push them remotely with high priority; The remote interaction module is used to aggregate status and approve permissions for shutdown and wind turbine control commands, generate transaction numbers, issue execution and receipts after consistency checks, and perform gray-scale updates to the model and prompts.

2. The intelligent kitchen safety ventilation multimodal alarm integrated system according to claim 1, characterized in that: The redundant switching circuit between mains power and backup battery completes the preparation for continuous power supply. Then, according to the short-time turn-on process, extremely short pulses are sent sequentially to the exhaust fan actuator and the air source solenoid valve actuator to verify that the drive stage and feedback channel are normal. Subsequently, the multi-modal alarm module sounds a buzzer for a short time, flashes an indicator light for a short time, and plays a low-volume voice sample to confirm that the sound, light and voice link is complete.

3. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 2, characterized in that: After receiving the self-test pass signal from the power supply and self-test module, the baseline is first solidified through the power-on stabilization process, then the calibration coefficient is loaded and the time reference is aligned. Then, combustible gas is collected to obtain the volume fraction of methane / propane, carbon monoxide is collected to obtain the CO concentration, smoke or particulate matter is converted into smoke index, and temperature, humidity and airflow are collected to obtain environmental quantities.

4. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 3, characterized in that: The points are deployed and fixed in three zones, defining the S1 pollution source zone, S2 transition zone, and S3 occupied zone. The smoke or particulate matter index, local airflow velocity, temperature and humidity of the three zones are aligned with the local real-time clock before the processed data frame is output. In each judgment cycle, temperature and humidity compensation and noise reduction are performed to obtain a cleaning sequence. The synchronization amplitude of the partition, the delay correlation coefficient and the consistency of the arrival order are calculated. The situation classifier is constructed by combining the target and feedback speed of the fan and the target and feedback opening of the ventilation components, and the threshold is adaptively updated according to the health day quantile. The path effectiveness quantification index PV is summarized by the monotonicity of the decrease in S1 to S2 to S3 under the step trigger of the wind turbine, the arrival order and the causal consistency, and the threshold is fine-tuned according to the offline quantile and incorporated into the log. In the absence of differential pressure sensors, the inlet dynamic pressure agent is estimated by fan power mapping, the window opening resistance agent is calculated by combining the opening degree of ventilation components with the impedance table, and the transient relationship between the S2 airflow and the concentration gradients of S1 and S2 is used to give the enclosure zone Δp and the external Δp.

5. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 4, characterized in that: When entering the closed-loop arrangement, if the situation is reverse flow or insufficient PV, the only instruction sequence of directional speed increase of the enclosure is to open the kitchen side at a small angle and suppress the living room side, and limit the fan to low and medium speed to stabilize the enclosure belt Δp*. Subsequently, the fan speed is increased in a step cycle and the kitchen side opening is slightly increased to drive the gradual monotonic reduction from S1 to S2. High air volume is only switched in when PV meets the standard, the external Δp direction is correct, and the intensity reaches the minimum agent intensity for entering high air volume. Otherwise, window sash swapping or short pulse strong exhaust is prioritized to break the stagnation. At the end of each cycle, the gain is adjusted based on Δp, PV, and zone reduction.

6. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 5, characterized in that: Based on the same event, the actual trajectory and alternative trajectory are generated, and the exposure integral of the occupied area, energy consumption proxy and noise proxy are calculated. The remote interaction module selects the minimum regret sequence as the preferred arrangement for the next event along the offline Pareto front under the user preference boundary. When the external disturbance causes the PV to drop but the pollution source does not expand, the fan micro-step probe is initiated and the three-zone reduction response signature is compared in a short observation window. If they do not match, the window is switched and retried. If the standard is not met twice in a row, the conservative arrangement is returned.

7. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 6, characterized in that: After receiving and processing the data frame, cross-channel synchronization compensation is first performed using temperature, humidity and airflow context. Then, short-window denoising and uniform step size resampling are used to align the time axis and convert the values ​​of combustible gas, carbon monoxide and smoke to comparable concentrations and indices under the current operating conditions. Subsequently, a dual-track approach of hierarchical threshold rules and pattern recognition is used to give the security level and leakage confidence. When the conclusions are inconsistent, a higher level of consistency constraint is applied and a unique set of control instructions is generated. The active ventilation control module drives the exhaust fan and ventilation components to establish a dilution path according to the unique control instruction set. Under dangerous conditions, it then drives the gas source solenoid valve to cut off the gas supply. After that, it performs closed-loop verification based on both absolute concentration and drop rate and executes a forced ventilation delay with a fixed upper limit.

8. The integrated intelligent kitchen safety ventilation multimodal alarm system according to claim 7, characterized in that: Within the same judgment period, a multimodal alarm is triggered according to the alarm level output by the data fusion module, causing the buzzer to sound rhythmically, the indicator light to flash at a frequency, and the voice command to be broadcast. Then, the alarm level, trigger factor, and handling timeline are pushed to the terminal with high priority through remote interaction. If the conditions for cancellation are not met, the alarm will continue to be pushed at the set frequency until it is cancelled or the authorized user resets it. The remote interactive side encrypts and aggregates measurement and handling status, verifies remote shutdown and fan start / stop speed control commands according to role permissions, generates transaction numbers, submits them for consistency checks, and then hands them over to the active ventilation control module for execution. It then writes back success or failure, timestamp, parameters and equipment feedback, and distributes model and prompt update packages in a gray-scale manner.

Images