Method and system for accurately monitoring oil fume emission based on multi-sensor fusion

Abstract

The present application relates to the technical field of trace detection, and specifically discloses a method and system for accurately monitoring oil fume emission based on multi-sensor fusion, which comprises the following steps: preprocessing original multi-source sensor data to obtain effective multi-source sensor data, then extracting the step change characteristics of the working conditions of the fan and the purification device from the effective multi-source sensor data, identifying the control action disturbance period and marking the disturbance period and the non-disturbance period; then generating a wet-state interference risk indicator, and in the non-disturbance period, the wet-state interference risk indicator is used to gate constraint to perform dynamic time warping, so that the concentration and the air volume are aligned in time, and then the basic emission of oil fume is calculated; further determining the health score of the monitoring link, and using the wet-state interference risk indicator and the health score of the monitoring link as constraints to dynamically correct the basic emission; when in the disturbance period, the control decoupling is performed on the emission output, thereby reducing false alarms, false treatments and completing traceable oil fume emission monitoring.

Description

Technical Field

[0001] This invention relates to the field of trace detection technology, specifically to a method and system for accurate monitoring of oil fume emissions based on multi-sensor fusion. Background Technology

[0002] As restaurant fume treatment has shifted from simply installing purification devices to a closed-loop system of online monitoring and operation and maintenance based on emission standards, the control side not only needs to determine whether the purification facilities are turned on or malfunctioning, but also needs to conduct continuous, comparable, and traceable precise monitoring of the amount of fume emitted from the emission outlet after treatment by the purification devices, in order to support early warning of exceeding standards and linkage control (such as power strategy for exhaust fans / purification devices and triggering of cleaning and maintenance).

[0003] Against this backdrop, existing technologies generally adopt a technical approach of deploying sensors at the back end of the purification device, collecting data online, and performing statistics in the cloud: oil fume concentration sensors (with optional NMHC / particulate matter sensors) are installed in the flue downstream of the purification device to simultaneously collect flue gas parameters such as temperature, pressure, and humidity. Often, the current / speed or on / off status of the exhaust fan is also collected as evidence of operating conditions. The data is then uploaded to the platform after being denoised, limited, averaged, and filtered for validity by the acquisition terminal. The platform calculates the emission rate based on the concentration monitoring value and in combination with the cross-sectional area of ​​the flue and the flow rate / volume (if the flow rate is not directly measured, it is often indirectly estimated using the exhaust fan current, frequency converter, etc.) and accumulates the emission amount over time. At the same time, it is linked with threshold rules to output alarms for exceeding standards, offline alarms, maintenance reminders, or control commands, thereby forming a monitoring-judgment-disposal control process.

[0004] Therefore, in actual catering scenarios, the intensity of cooking fluctuates drastically, exhaust fans need frequent speed adjustments, and the flue gas itself has high humidity, easily carrying a large amount of water mist and droplets. Furthermore, sensors and sampling pipelines inevitably become coated with oil after long-term operation, reducing their sensitivity. These complex conditions directly amplify the shortcomings of existing technical solutions in terms of detection timing and data reliability: In the concentration detection stage, due to factors such as transmission delays in the sampling pipeline, water vapor condensation obstruction, and filter adsorption losses, readings exhibit a significant lag effect. Meanwhile, the airflow uses real-time instantaneous estimates, making it impossible to accurately match the two in the time dimension, directly leading to deviations in emission calculations, especially under sudden changes in operating conditions such as stir-frying and exhaust fan speed adjustments. At the same time, water mist and droplets in the flue gas can also interfere with the concentration sensor, causing it to mistakenly interpret humidity changes as fluctuations in oil fume concentration. This not only leads to irregular sharp increases in readings but also a false trend of slow decline after the operating conditions stabilize, significantly increasing the difficulty for the monitoring system to distinguish between genuine excessive emissions and interference fluctuations.

[0005] More importantly, as operating time accumulates, oil stains gradually adhere to the surface of the sensor's light-transmitting window, causing a decrease in sampling volume. Such systematic errors caused by equipment contamination are often treated as normal data and continuously output over a long period of time. Based on these unreliable emission data, after the monitoring system triggers control actions such as power switching and cleaning, the sudden changes in equipment operating conditions will in turn disturb the distribution of droplets and airflow in the flue, further disrupting the detection readings of concentration and airflow, forming a vicious cycle. Ultimately, this interferes with the effective judgment of the true emission level of oil fumes, causing either missed detection of true over-emissions or false over-emissions. It also spurs a large number of ineffective cleaning, power adjustment, and other disposal actions, forcing a sharp increase in the frequency of equipment maintenance. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for accurate monitoring of oil fume emissions based on multi-sensor fusion, which can effectively solve the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a method for accurate monitoring of oil fume emissions based on multi-sensor fusion, comprising: collecting raw multi-source sensor data of a calibrated monitoring target area based on a set monitoring node; performing basic validity screening and hard anomaly processing to obtain valid multi-source sensor data; extracting step change characteristics of the valid multi-source sensor data; identifying control action disturbance periods during oil fume monitoring and marking them as disturbance periods and non-disturbance periods; generating a wet interference risk indication by calibrating the temperature and humidity state of the calibrated monitoring target area; performing dynamic time alignment of the valid multi-source sensor data based on the wet interference risk indication constraint during non-disturbance periods; calculating the basic oil fume emission result of the calibrated monitoring target area based on the alignment result; extracting a link health assessment set to determine the monitoring link health score; dynamically correcting the basic oil fume emission result based on the wet interference risk indication and the monitoring link health score; and performing control decoupling on the oil fume emission output during disturbance periods to complete oil fume emission monitoring.

[0008] The second aspect of this invention provides a precise monitoring system for oil fume emissions based on multi-sensor fusion, comprising: a sensor data processing module, used to collect raw multi-source sensor data of a calibrated monitoring target area based on set monitoring nodes, perform basic validity screening and hard anomaly processing, and obtain valid multi-source sensor data; a disturbance period determination module, used to extract step change characteristics of valid multi-source sensor data, identify disturbance periods of control actions during oil fume monitoring, and mark them as disturbance periods and non-disturbance periods; an oil fume basic emission calculation module, used to generate a wet interference risk indication based on the temperature and humidity state of the calibrated monitoring target area, and perform dynamic time alignment of valid multi-source sensor data based on the wet interference risk indication constraint during non-disturbance periods, and calculate the oil fume basic emission result of the calibrated monitoring target area based on the alignment result; and an oil fume real-time emission output module, used to extract a link health assessment set to determine the monitoring link health score, dynamically correct the oil fume basic emission result based on the wet interference risk indication and the monitoring link health score, and perform control decoupling on the oil fume emission output if it is within the disturbance period, thus completing the oil fume emission monitoring.

[0009] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) This invention provides a method and system for accurate monitoring of oil fume emissions based on multi-sensor fusion. It preprocesses raw multi-source sensor data to obtain effective multi-source sensor data, thereby avoiding misjudgments caused by outliers directly entering the emission calculation link. Subsequently, it extracts the step change characteristics of the fan and purification device operating conditions from the effective multi-source sensor data, identifies the control action disturbance period, and marks the disturbance period and non-disturbance period. Then, it generates a wet interference risk indication based on the temperature and humidity state of the calibrated monitoring target area, and performs dynamic time alignment by gating constraints of the wet interference risk indication during the non-disturbance period, aligning the concentration and airflow in time before... The system calculates the baseline emission of cooking fumes to mitigate the systematic bias in emission calculation caused by sampling lag and wet trailing. It further extracts a link health assessment set to determine the monitoring link health score, and uses the wet interference risk indicator and the monitoring link health score as constraints to dynamically correct the baseline emission, ensuring reliable and conservative outputs even under conditions of sensor contamination, sampling link degradation, or airflow extrapolation. During periods of disturbance, the emission output is decoupled from control, outputting only the disturbance period estimate and recovery trend, and downgrading strong judgments and strong linkages, thereby reducing false alarms and mishandling, and achieving traceable monitoring of cooking fume emissions.

[0010] (2) By identifying the step changes caused by fan start-up and shutdown, speed regulation, purification device gear switching and cleaning trigger, and marking the disturbance period and non-disturbance period, this invention enables the monitoring system to shield or downgrade the judgment of exceeding the standard and linkage control during the disturbance period, and only output the disturbance period estimate and recovery trend, thereby significantly reducing the probability of misjudging the short-term fluctuations caused by control actions as real over-emission, and improving the stability and verifiability of emission statistics in operation and maintenance.

[0011] (3) This invention summarizes evidence such as zero-point drift, stable operating noise, air volume estimation uncertainty and operating response consistency into a monitoring link health score, enabling the monitoring system to provide a quantitative judgment and cause label on whether the current data can be used as a reliable basis, and upgrades the emission result from a single value to an interpretable output that combines numerical values ​​with credible prompts; when the link health declines, it can provide early prompts to clean the optical window, check the sampling pump and pipeline or verify the air volume calibration, avoiding the long-term accumulation of hidden drift that leads to report distortion and incorrect operation and maintenance decisions.

[0012] (4) Compared with the common method of averaging by concentration and air volume after fixed time window in existing technologies, this method explicitly introduces disturbance identification, dynamic time alignment under wet risk gating, and dynamic correction and decoupling of output during disturbance period under link health constraints in the calculation link. It can apply executable suppression mechanisms to key error sources such as sampling lag, wet entrainment, long-term pollution drift and control action disturbance in catering scenarios one by one, thereby improving the accuracy, stability and traceability of emission results without increasing the hardware complexity, and reducing false alarms, mishandling and maintenance costs. Attached Figure Description

[0013] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0015] Figure 2 This is a schematic diagram of the system module connections of the present invention.

[0016] Figure 3 This is a schematic diagram of the model structure for the fusion model.

[0017] Figure 4 This is a graph showing the monitoring results of oil fume emissions. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] This invention is applied to the treatment of oil fume in catering establishments. The oil fume purification device operates under real-world conditions where it alternates between peak and low-load periods of stir-frying, the exhaust fan frequently adjusts its speed using a frequency converter, the flue gas is highly humid and easily approaches the dew point, and carries water mist and droplets, and the sensors and sampling links are gradually contaminated by oil during operation. The monitoring system needs to continuously and reliably monitor the amount of oil fume emitted from the downstream outlet of the purification device without affecting the normal operation of the store, and provide a basis for exceeding the standard warning, cleaning and maintenance reminders, and linkage control.

[0021] To this end, monitoring nodes synchronously collect data such as oil fume concentration, temperature, humidity, pressure, exhaust fan frequency or current, and purification device status near the emission outlet at second-level intervals. They first identify periods of disturbance such as start-up / shutdown, speed adjustment, power switching, or cleaning triggers to avoid fluctuations caused by control actions being misjudged as actual over-emissions. During periods of non-disturbance and low humidity risk, a sliding time window is used to match airflow changes with concentration changes, estimating and aligning sampling lag to ensure that concentration and airflow are aligned. Simultaneously, water mist condensation / entrainment interference is identified by combining temperature and humidity conditions with concentration change characteristics. Link health and reliability indicators are generated based on zero-adjustment or low-load references, stable operating noise, and operating condition consistency. After standardization, the baseline emission is calculated first, and then, under gating of humidity risk, link health, and airflow uncertainty, the fusion model outputs a compensation correction to compensate the baseline value, obtaining the corrected real-time emission and short-term prediction. The results are then used for monitoring records and the linkage control of the purification device.

[0022] Example 1: Refer to Figure 1 As shown, the first aspect of the present invention provides a method for accurate monitoring of oil fume emissions based on multi-sensor fusion, comprising: collecting raw multi-source sensor data of a calibrated monitoring target area based on a set monitoring node, performing basic validity screening and hard anomaly processing, and obtaining effective multi-source sensor data.

[0023] In this embodiment, a monitoring node is set up near the downstream emission outlet of the catering fume purification device to collect and process monitoring data on the purified emission side. To ensure that the subsequent emission calculation results have a consistent calculation caliber across different time periods and equipment states, and to facilitate verification, the caliber parameters used for emission calculation are configured and permanently saved during the installation and commissioning cycle. The caliber parameters include at least the flue cross-sectional dimensions (used to determine the conversion basis for exhaust volume flow rate), emission statistical period (used to determine the statistical time scale for real-time values, average values, and cumulative amounts), unit conversion rules (used to uniformly convert monitoring and calculation results into preset output units), zeroing cycle and zeroing duration (used to constrain the execution frequency and duration of zero-point calibration, supporting measurement stability under long-term operation). After being permanently saved, the above caliber parameters are consistently invoked during the operation of the monitoring system to ensure that the emission calculation process is traceable and verifiable.

[0024] The raw multi-source sensor data includes equivalent concentration data of oil fume particles, flue gas temperature data, flue gas humidity data, flue gas pressure data, exhaust fan operating status data, air volume related data used to characterize exhaust intensity, and operating condition data.

[0025] Basic validity screening includes at least range verification, negative and unreachable value verification, continuous constant value verification, and communication missing measurement marking.

[0026] Range verification involves comparing the values ​​of each sampling point with the upper and lower limits of the range fixed during the installation and commissioning cycle of each sensor. When the value exceeds the upper limit or falls below the lower limit, the sampling point is marked as invalid and the corresponding channel and time are recorded. Negative value and unreachable value verification involves determining that negative values ​​or extreme values ​​(such as concentration, humidity, air volume, etc.) that should not physically occur are unreachable according to preset rules and marked as invalid. Continuous constant value verification involves statistically analyzing whether the data of the same channel remains unchanged for a long time or the change amplitude is lower than the noise lower limit within a rolling time window. When multiple consecutive sampling points are exactly the same or approximately the same for a preset duration, it is determined that the channel may have sampling stagnation, jamming, or communication freeze, and the data segment is marked as "constant anomaly". Communication missing measurement marking involves recording the missing measurement duration and proportion of the channel when no data is received from a certain channel during the sampling period or when an abnormal placeholder code / empty packet is received. At the same time, it is prohibited to use interpolation to impersonate the original value for statistical calculation.

[0027] Hard anomaly handling includes at least performing median replacement on isolated spikes and retaining the replacement marker. Specifically, if a sampling point experiences a sudden jump relative to its preceding and following sampling points, and this jump only lasts for one or a few sampling cycles, it is identified as an isolated spike, provided that multiple adjacent sampling points are valid and the trend remains continuous. The median of the spike is calculated using a local window composed of several preceding and following sampling points, and the replacement marker and the original value are retained for auditing and tracing.

[0028] The specific steps for maintaining a continuous trend are as follows: Take several sampling points before and after the sampling point to be judged (e.g., three to five points before and after each, the number of which is fixed by the configuration); calculate the representative value of the preceding data (e.g., the median or average) and the representative value of the following data; determine whether the difference between the preceding and following representative values ​​is less than a preset continuity threshold, and whether the fluctuation range of the preceding and following data is within the normal noise range; if the above conditions are met, it is considered that the overall level before and after the point is consistent and the change is stable, that is, the trend remains continuous.

[0029] Extract the step change characteristics of effective multi-source sensor data, identify the control action disturbance period in the oil fume monitoring process, and mark it as the disturbance period and the non-disturbance period.

[0030] The aforementioned operational data is used to record the operating status data of the exhaust fan and the operating status data of the purification device. The exhaust fan operating status data includes at least one or more of the following: exhaust fan start / stop status, exhaust fan inverter frequency, and exhaust fan current. The purification device operating status data includes at least one or more of the following: purification device operating status, cleaning trigger status, and power switching status.

[0031] The start / stop status of the exhaust fan is obtained from the switch signal of the exhaust fan control circuit or the inverter operation signal. The frequency of the exhaust fan is obtained from the inverter communication interface or frequency feedback signal. The exhaust fan current is obtained from the current detection device of the exhaust fan power supply circuit or the current parameters reported by the inverter. The operating status data of the purification device is obtained by at least one of the following methods: output by the purification device controller through the communication interface and read by the monitoring node; output by the operation contact, fault contact or cleaning contact of the purification device electrical control box and collected by the monitoring node; or obtained by collecting and inferring the electrical parameters of the purification device power supply circuit and cleaning execution circuit.

[0032] The above air volume data are either direct measurements of the smoke exhaust fan flow rate or the smoke exhaust fan speed, or estimated air volume results.

[0033] Based on the step change characteristics of operating condition data and air volume related data, the system detects whether there is a triggering event. Specifically, when any of the following events are detected, such as a change in the start / stop status of the exhaust fan, the frequency of the exhaust fan reaching a preset frequency change threshold, the current of the exhaust fan reaching a preset current transition threshold, the air volume related data reaching a preset change threshold, the power setting of the purification device switching, or cleaning triggering, the interval from the time of the event to the preset duration is marked as a disturbance time period. When the event persists, the disturbance time period is extended in a rolling manner.

[0034] Specifically, the frequency change threshold of the exhaust fan variable frequency reaches the preset frequency change threshold when the cumulative change of the variable frequency within a preset short time window reaches the preset frequency change threshold, or when the variable frequency maintains the same direction of change for multiple consecutive sampling cycles and the cumulative change reaches the preset frequency change threshold.

[0035] Specifically, the smoke exhaust fan current step reaches the preset current threshold when the cumulative change in the smoke exhaust fan current within a preset short time window reaches the preset current threshold, or when the smoke exhaust fan current maintains the same direction of change within multiple consecutive sampling periods and the cumulative change reaches the preset current threshold.

[0036] When the air volume related data is direct measurement data of air volume or air speed, it is detected that the cumulative change of air volume or air speed within a preset short time window reaches the preset air volume change threshold; when the air volume related data is the frequency of the exhaust fan, the current of the exhaust fan, the static pressure of the exhaust fan, or the differential pressure data of the exhaust fan used to estimate the air volume, it is detected that the cumulative change of the air volume estimation result calculated by the air volume estimation relationship within a preset short time window reaches the preset air volume change threshold.

[0037] The above air volume estimation results are as follows: During the installation and commissioning cycle, the current, frequency conversion frequency, static pressure, or differential pressure of the exhaust fan were collected at the working point group of the exhaust fan to measure the stable operating state of the exhaust fan. The stable operating state of the exhaust fan means that after the exhaust fan has been switched to the target gear or target frequency conversion frequency, the fluctuations of the fan frequency conversion frequency and the fan current are less than the corresponding stable fluctuation thresholds within at least the preset stable observation period, and the fluctuations of the static pressure or differential pressure are less than the corresponding thresholds, thus indicating that the exhaust condition has entered a relatively stable stage.

[0038] Simultaneously, a portable reference anemometer is used to measure the reference wind speed. This portable reference anemometer is a portable instrument that can be used on-site and provide wind speed readings within a short time; preferably, it is a thermal anemometer or a rotor-type anemometer. Multi-point wind monitoring is conducted according to a preset pattern of characteristic measuring points. These characteristic measuring points can be the center point of the duct cross-section and several evenly distributed representative points, or representative points equidistantly distributed along the width and height of the cross-section. The average or median wind speed readings from each measuring point are taken to obtain the reference wind speed at that working point. The reference wind speed is the reference calibration value of the actual airflow velocity within the flue.

[0039] The reference air volume is calculated by converting the reference wind speed with the effective cross-sectional size of the solidified flue. The reference air volume is the volume of flue gas passing through the cross-section of the flue per unit time at the operating point. It serves as the benchmark calibration value for air volume estimation, thereby forming a sample dataset containing multiple sets of fan current, frequency conversion frequency, static pressure or pressure difference and the reference air volume.

[0040] Based on the aforementioned sample dataset, an airflow estimation relationship is generated according to a preset estimation method. This estimation method can be multiple regression, piecewise linear relationship, lookup table interpolation, or a combination thereof. This establishes a mapping relationship between fan current, frequency converter frequency, static pressure or differential pressure, and reference airflow. The generation of the airflow estimation relationship aims to minimize the deviation of the estimated airflow from the sample set. It outputs a set of parameters or lookup table entries that can be called by the monitoring system, and writes this set of parameters or lookup table entries into the monitoring system's configuration file or program parameter area for solidification. Simultaneously with solidification, the debugging coverage range is determined. The debugging coverage range represents the actual range of values ​​and their combinations covered by each input parameter in the sample dataset. For example, the minimum to maximum value of the frequency converter frequency, the minimum to maximum value of the current, the minimum to maximum value of the static pressure or differential pressure, and the typical combination range of these values ​​appearing in the sample. The coverage range is also solidified along with the configuration and used to determine online whether the current operating condition is within the calibration applicable range.

[0041] For each sample, the difference between the estimated air volume and the reference air volume is calculated. The difference can be expressed as an absolute error or a relative error. The root mean square error of these differences is used as the basic uncertainty of the air volume estimation for that store. The basic uncertainty is mapped to the basic uncertainty level according to the preset threshold classification rule, which is used to reflect the inherent error of the estimation relationship itself within the debugging coverage area.

[0042] During online operation, the real-time collected data on the current of the exhaust fan, the frequency of the exhaust fan, and the static pressure or pressure difference of the exhaust fan are substituted into the air volume estimation relationship to obtain real-time air volume estimation data. Based on the basic uncertainty and whether the substituted parameters exceed the debugging coverage range, an air volume estimation uncertainty indication is generated.

[0043] When the real-time input parameter is within the debugging coverage range, a first uncertainty indication corresponding to the basic uncertainty is output; when the real-time input parameter is close to or exceeds the debugging coverage range, a second uncertainty indication is output, wherein the first uncertainty indication is less than the second uncertainty indication. When direct airflow measurement data is available, the directly measured airflow is preferentially used as the real-time airflow result.

[0044] The wet interference risk indicator is generated by calibrating the temperature and humidity of the monitoring target area. During non-disturbance periods, dynamic time alignment of effective multi-source sensor data is performed based on the wet interference risk indicator constraint. The basic emission of oil fume in the calibrated monitoring target area is calculated based on the alignment result.

[0045] The wet disturbance risk indication is as follows: In each sampling cycle, flue gas temperature and relative humidity are acquired, and dew point temperature is obtained according to a preset dew point estimation rule. The dew point estimation rule is a well-known estimation rule for those skilled in the art. The temperature difference between flue gas temperature and dew point temperature is used as a criterion for the degree of condensation proximity. When the temperature difference is less than a preset temperature difference threshold and continues to reach a preset minimum duration, it is determined to be close to the dew point. At the same time, relative humidity is used as a high humidity criterion. When the relative humidity is greater than a preset high humidity threshold and continues to reach a preset minimum duration, it is determined to have reached the high humidity threshold. In order to avoid instantaneous noise triggering, it is also required that the relative humidity continuously sampled within the minimum duration exceeds the high humidity threshold.

[0046] Secondly, within a preset short time window, when the increase in concentration relative to the short-time baseline reaches a preset increment threshold and is completed within the short time window, a short-term surge is determined to exist. After the short-term surge reaches its peak, when the time required for the concentration to fall back from the peak to the interval close to the short-time baseline is greater than a preset fallback duration threshold, or when it has not fallen back to the interval close to the baseline by the end of the preset fallback observation window, a delayed fallback is determined to exist. When both short-term surge and delayed fallback are satisfied, the concentration data is determined to exhibit a wet interference pattern.

[0047] The wet interference risk level is set to multiple preset discrete levels, and the wet interference risk level is determined based on the number of criteria that are met in the near dew point determination result, the high humidity threshold determination result, and the wet interference morphology determination result: when only one criterion is met, it is determined as the first comprehensive risk score; when two criteria are met simultaneously, it is determined as the second comprehensive risk score; and when all three criteria are met simultaneously, it is determined as the third comprehensive risk score. At the same time, different risk contribution values ​​are configured for the near dew point determination, the high humidity threshold determination, and the wet interference morphology determination, where the risk contribution value corresponding to the wet interference morphology is greater than the risk contribution value corresponding to the near dew point, and the risk contribution value corresponding to the near dew point is greater than the risk contribution value corresponding to the high humidity threshold. When any criterion is met, the risk contribution value corresponding to that criterion is added to the comprehensive risk score. The comprehensive risk score is compared with the preset classification threshold to determine the wet interference risk level. By comparing the comprehensive risk score with the preset first and second level thresholds, a low-risk level is determined when the comprehensive risk score is less than the first level threshold, a medium-risk level is determined when the comprehensive risk score is not less than the first level threshold and less than the second level threshold, and a high-risk level is determined when the comprehensive risk score is not less than the second level threshold.

[0048] Furthermore, the corresponding wet interference risk level is only confirmed and output when the determined risk level is maintained for a preset minimum duration; otherwise, the wet interference risk level of the previous moment remains unchanged.

[0049] Perform dynamic time alignment of valid multi-source sensor data, specifically as follows: In this embodiment, to avoid the control action disturbance and wet interference from misleading the time alignment results, the monitoring system first performs gating screening on the data to be analyzed and extracts evidence that can be aligned before performing the correlation search of the alignment relationship, so as to update the alignment relationship only within the time slice that meets the conditions.

[0050] Specifically, data within the disturbance period is marked as not participating in the alignment update; based on the wet interference risk level, when the wet interference risk level reaches or exceeds the preset gate threshold, the monitoring system freezes the alignment relationship update or only allows it to be adjusted within a preset small range, thereby avoiding the trailing phenomenon caused by water mist condensation or water droplets being mistakenly regarded as concentration channel lag.

[0051] For time slices that meet the non-disturbance period and whose wet interference risk level is below the preset gate threshold, the monitoring system further extracts the changing trends from the air volume and concentration data as evidence for alignment. The changing trends are preferably expressed as the increase or decrease of adjacent sampling points, short-term slope, or incremental sequences after short-term smoothing, to highlight the dynamic response caused by changes in operating conditions. Then, it is determined whether the evidence for alignment meets the minimum change amplitude condition. The minimum change amplitude condition is: within a predetermined sliding window, the fluctuation of the air volume change trend reaches the air volume change threshold and the fluctuation of the concentration change trend reaches the concentration change threshold, and the changing trends show stable positive or negative correlation characteristics within the window. When the minimum change amplitude condition is not met, the alignment relationship of the previous moment remains unchanged and the current update is skipped. Only when the minimum change amplitude condition is met does the system perform a correlation search on the air volume change trend and the concentration change trend within the predetermined sliding window.

[0052] The monitoring system performs a correlation search within a predetermined sliding window to obtain alignment relationships. Specifically, the monitoring system constructs multiple candidate offset values ​​one by one within a configured and fixed candidate time offset range. That is, it assumes that the concentration has different time lags relative to the air volume, and shifts and aligns the concentration change trend sequence according to each candidate offset value so that it corresponds point by point with the air volume change trend sequence under the same time index. For each candidate offset value, a similarity calculation is performed on the two change trend sequences within the sliding window. The similarity calculation is a correlation evaluation. It is preferred to use the normalized correlation coefficient or an equivalent correlation scoring method to obtain the correlation score corresponding to the candidate offset value.

[0053] The correlation scores of each candidate offset amount are compared and the candidate offset amount with the highest score is selected as the alignment relationship of the current window. As time progresses, the sliding window scrolls forward by a preset step size and repeats the above process, thereby realizing the online updating and tracking of the alignment relationship.

[0054] The concentration data is resampled based on the alignment relationship to establish a temporal correspondence with the air volume-related data at the same time.

[0055] The results of the basic emissions of cooking fumes are as follows: Under the condition of uniform caliber, the equivalent concentration of oil fume particles after dynamic time alignment is converted to the preset concentration caliber, and the real-time air volume result is extracted and determined as the volumetric flow rate of flue gas discharged per unit time.

[0056] The above-mentioned unified caliber conditions mean that under the pre-fixed calculation caliber, the calculation caliber includes: converting the concentration data to the preset concentration caliber, determining the air volume data as the volumetric flow rate discharged per unit time, corresponding the concentration and air volume at the same time and unifying the sampling and statistical rules, as well as the unit conversion rules.

[0057] At the same time, the concentration and volumetric flow rate of the preset caliber are multiplied to obtain the mass of oil fume emission per unit time, and the mass of oil fume emission per unit time is accumulated according to the sampling period to obtain the basic emission result of oil fume within the statistical period.

[0058] Extract the link health assessment set to determine the monitoring link health score. Use the wet interference risk indicator and the monitoring link health score as constraints to dynamically correct the basic emission results of oil fume. If it is within the disturbance period, control decoupling is performed on the output of oil fume emission to complete the monitoring of oil fume emission.

[0059] The monitoring link health score is as follows: Within each preset scoring period, the monitoring system first performs segment screening on the monitoring data within that period. The screening process involves selecting time periods that meet the conditions from the continuous time-series data according to preset rules as input for subsequent evaluation. Specifically, within that period, time periods in which the number of sampling points marked as missing, exceeding the range limit, or replaced by isolated spikes are removed, and the proportion of these sampling points to the total number of sampling points in the corresponding time period exceeds a preset threshold is further removed. Time periods with control action disturbances are then removed. In the remaining time periods, segments of stable operation of the exhaust fan are selected and recorded as stable operating condition segments.

[0060] When the monitoring link has a zero-point calibration function, a zero-adjustment event is generated. Specifically, the monitoring system triggers valve switching according to a preset zero-adjustment cycle, so that the concentration sensor samples zero gas or clean air and performs zero-point calibration for a short period of time. This short period of time is the zero-adjustment event.

[0061] After a zeroing event begins, the monitoring system extracts concentration data from the stable zeroing period over a continuous time interval and calculates the low-level representative value of the concentration. The stable zeroing period is defined as a period in which the concentration reading has recovered from the switching transient and the fluctuation amplitude does not exceed a preset fluctuation threshold, and this fluctuation threshold is continuously met for a preset minimum duration. The low-level representative value of the concentration is a representative value used to characterize the zero-point level at this time. It is preferably obtained by taking the median of the concentration data in this period, or by taking the statistical value of the lower quantile of the concentration data in this period. The monitoring system compares the current low-level representative value of the concentration with representative values ​​obtained from multiple historical zeroing events. If there is a continuous increase, a continuous shift, or a difference between two adjacent values ​​exceeding the threshold, a zero-point drift conclusion is formed.

[0062] When the monitoring node does not have a zero-point calibration period, the monitoring system selects continuous time intervals within a preset observation window that meet the low-load or low-concentration criteria. The low-load criteria are that the cooking load-related operating conditions are continuously lower than the preset low-load threshold and remain at the preset minimum duration, and are not in a disturbance period. The low-concentration criteria are that the sliding median of the oil fume concentration is not higher than the preset low-concentration threshold and remains at the preset minimum duration, while the wet interference risk level is lower than the preset risk threshold level and is not in a disturbance period. The low-end representative value of the concentration data within the continuous time interval is extracted as a quasi-zero point reference and compared with the historical quasi-zero point reference to determine the zero-point drift trend.

[0063] It should be explained that the above-mentioned cooking load-related operating condition indicators are one or more of the following: stove power consumption, gas valve opening, instantaneous gas flow, stove ignition status, or fume hood suction intensity.

[0064] The concentration fluctuation level is calculated within a stable operating condition segment. This is achieved by statistically analyzing the fluctuation range or dispersion of the concentration around its representative value within the stable operating condition segment. For example, the difference between the maximum and minimum values, or the standard deviation or median absolute deviation are statistically analyzed and compared with the normal fluctuation range fixed during the installation and commissioning phase to obtain a noise conclusion.

[0065] Check the consistency of concentration response direction based on fan frequency events or current change events: After the change event occurs and within the preset response observation window, extract the concentration data and determine its change direction. That is, when the absolute value of the difference between the representative concentration value at the end of the observation window and the representative concentration value at the beginning of the observation window reaches the preset minimum change threshold, if the difference is positive, it is determined to be an overall increase; if the difference is negative, it is determined to be an overall decrease. When the absolute value of the difference does not reach the minimum change threshold, it is determined to be basically unchanged.

[0066] The direction of concentration change is compared with the direction of frequency or current change events for consistency: when the direction of frequency or current event is increasing and the direction of concentration change is decreasing or basically unchanged, or when the direction of frequency or current event is decreasing and the direction of concentration change is increasing or basically unchanged, the event is recorded as a direction inconsistency event; the number of direction inconsistency events and the total number of valid change events are counted within a preset statistical window, and the direction inconsistency ratio is calculated. When the direction inconsistency ratio exceeds the preset ratio threshold, the direction consistency is determined to be insufficient and a consistency conclusion is formed.

[0067] The zero-point drift conclusion, noise conclusion, airflow estimation uncertainty indication, and consistency conclusion are combined to form a monitoring link health score. The deviation is extracted from each conclusion. d1 = |Low-end representative value of concentration - Historical baseline representative value|, d2 = Concentration jitter statistics within a stable operating segment / Baseline jitter statistics fixed during installation and commissioning, d3 = max (Amplitude of frequency exceeding coverage range, amplitude of current exceeding coverage range, amplitude of static pressure or differential pressure exceeding coverage range), d4 = Number of events with inconsistent response directions / Total number of valid change events. Among these, the concentration jitter statistics (e.g., maximum / minimum difference, standard deviation, or median absolute deviation) and the baseline jitter statistics are reference values ​​statistically fixed during the installation and commissioning cycle under normal link conditions.

[0068] The deviation is segmented into discrete levels based on a threshold. This can be represented by the following piecewise function: Let the four sub-item levels be: zero-point drift level g1, noise level g2, air volume uncertainty level g3, and consistency level g4, ​​denoted as: Where g1~g4∈{0,1,2,3}, representing normal, mild, moderate, and severe, respectively; threshold T i，1 T i，2 T i，3 The installation, debugging, statistics, and solidification are based on an index number, i, which is used to distinguish which category of sub-item / which category of deviation corresponds to the threshold group.

[0069] A health score is obtained by combining the maximum value operator with the counting operator: Maximum value operator and counting operators ,in This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. Then, a health score is defined: The health score components decrease monotonically: S0 > S1 > S2. The S0 component corresponds to a high-confidence state of the monitoring link, the S1 component corresponds to a medium-confidence state of the monitoring link, and the S2 component corresponds to a low-confidence state of the monitoring link.

[0070] The monitoring link health score is used to quantitatively characterize the availability and reliability of the oil fume emission monitoring link. In catering scenarios where equipment operates for a long time, oil pollution gradually worsens, sampling links may be blocked, air volume estimation may be extrapolated, and control actions are frequently disturbed, the score continuously integrates evidence such as zero-point drift, stable operating noise level, air volume estimation uncertainty, and operating response consistency to automatically determine that the current monitoring link is in a reliable state, and uses this as a quality annotation and gating basis for the emission results.

[0071] The baseline emissions of cooking fumes will be dynamically revised, specifically as follows: In this embodiment, the multi-sensor data fusion model is used to output the compensation correction amount and the prediction amount of the oil fume emission. The model structure of the fusion model can be an integrated learning model, a temporal neural network model or a combination of both, but its model parameters are obtained by data training or calibration during the installation and debugging cycle and the online operation phase and are solidified into a deployable version.

[0072] The training data sources include: aligned concentration data collected during the installation and commissioning cycle, air volume or air volume estimation input data, temperature, humidity and pressure data, fan operating status and purification device operating status data, and reference emission or reference air volume data obtained during the preset verification period. The preset verification period is a period in which the fan is operating stably and the wet interference risk indication meets the low-risk condition and the monitoring link health score meets the high-confidence condition, using portable reference wind measurement or other reference methods to form a verification data segment.

[0073] The input and output process of the multi-sensor data fusion model is as follows: Figure 3 As shown, Figure 3 The diagram illustrates the structure of the fusion model. The inputs to the fusion model are constructed according to a preset input domain: first, a sequence of basic emissions of cooking fumes; and second, a sequence of operational action data. During internal model processing, the operational action data is first converted into a computable vector sequence through entity embedding. After temporal encoding using an attention mechanism and a bidirectional long short-term memory network, action labels are output via softmax. The basic emissions sequence and its associated multi-source sensor sequences are constructed into a temporal feature tensor in the time dimension (h, f, t, and T in the diagram represent the latent representation dimension, feature dimension, time index, and total window duration, respectively). The temporal representation of key time slices is extracted using an attention mechanism, denoted as I_T in the diagram. Simultaneously, constraint components such as wet interference risk indication, monitoring link health score, and airflow estimation uncertainty indication enter the temporal health assessment module to form gating constraints, denoted as I_S in the diagram.

[0074] Subsequently, I_T and I_S are fused to form a comprehensive temporal representation I_N, which, along with action labels, participates in disturbance discrimination in the temporal fusion strategy, outputting disturbance flags, emission compensation corrections, and short-term predicted emissions. Thus, the corrected real-time oil fume emissions are obtained by superimposing the compensation corrections onto the base emission result. When the disturbance flag is "yes," the model output enters the control decoupling branch before being output externally; when the disturbance flag is "no," the corrected real-time oil fume emissions and short-term predicted oil fume emissions are directly output.

[0075] It should be explained that, in this embodiment, regardless of whether it is in a disturbed or non-disturbed phase, the monitoring system first calculates the basic emission based on effective multi-source sensor data, and generates a compensation correction and short-term predicted emission under the gating constraints of wet interference risk indication, monitoring link health score, and air volume estimation uncertainty indication, thereby obtaining the corrected real-time emission and short-term predicted emission. When it is determined to be in a non-disturbed phase, the system directly outputs the corrected real-time emission and short-term predicted emission as normal monitoring results and uses them for statistics and judgment. When it is determined to be in a disturbed phase, the system does not directly use the instantaneous results of that phase as a strong judgment basis, but instead performs control decoupling on the output of oil fume emission before outputting it.

[0076] The comparison between real-time emissions and short-term predicted emissions is as follows: Figure 4 As shown, Figure 4 This graph shows the monitoring results of oil fume emissions, illustrating the comparison between actual and predicted values ​​during continuous monitoring. The horizontal axis represents time (minutes), and the vertical axis represents oil fume emissions (example unit). Two curves (with superimposed scatter points) represent the actual emission sequence and the short-term predicted emission sequence output by the model, respectively. By observing the degree of alignment between the two curves at their peaks, rising phases, and falling phases, the predictive ability to track sudden surges in emissions, sustained high emissions, and recovery processes in the catering scenario can be intuitively reflected. When the predicted curve can provide a similar trend before or simultaneously with the actual emission increase, it indicates that the prediction can be used for early warning and coordinated control. When the deviation between the two increases, it usually indicates factors such as wet interference, control action disturbances, or a decline in link health, requiring the triggering of gating and reliable alert mechanisms to avoid misjudgments and erroneous linkages.

[0077] The aforementioned constraints, such as the wet interference risk indicator, the monitoring link health score, and the air volume estimation uncertainty indicator, are entered into the time-series health assessment module to form gating constraints. Specifically, to ensure that the correction amount is controllable and verifiable under complex catering conditions, the wet interference risk indicator, the monitoring link health score, and the air volume estimation uncertainty indicator apply a two-layer constraint of update gating and amplitude gating to the compensation correction amount.

[0078] The update gate is set as follows: when the control action disturbance period is in effect, or the wet interference risk level reaches or exceeds the preset risk gate level, or the monitoring link health score reaches or falls below the preset low confidence gate level, or the airflow estimation uncertainty reaches or exceeds the preset high uncertainty gate level, the compensation correction amount is frozen and the compensation correction amount from the previous moment remains unchanged; when the following conditions are met simultaneously: not being in a control action disturbance period, the wet interference risk level is lower than the preset risk gate level, the monitoring link health score is higher than the preset low confidence gate level, and the airflow estimation uncertainty is lower than the preset high uncertainty gate level, the compensation correction amount is allowed to be updated.

[0079] The amplitude gating mechanism applies a preset upper limit constraint on the absolute value of the compensation correction amount, while allowing updates, and a preset upper limit constraint on the rate of change of the compensation correction amount in adjacent sampling periods. The preset upper limit is either a preset percentage of the baseline emissions or a preset upper limit of the absolute correction amount. The preset upper limit of the rate of change is the preset maximum increment allowed per unit time. These upper limit parameters are statistically obtained and fixed from verification samples with low humidity risk and high reliability of link health during the installation and commissioning cycle. Furthermore, when the humidity interference risk level is medium, the monitoring link health score is medium reliability, or the airflow estimation uncertainty is medium uncertainty, the upper limit of amplitude and the upper limit of the rate of change are replaced with the corresponding medium-level upper limit parameters. When the humidity interference risk level is high, the monitoring link health score is low reliability, or the airflow estimation uncertainty is high uncertainty, a freeze update rule is executed. This ensures that the compensation correction amount has clear executable gating boundaries and quantified amplitude boundaries under different data quality levels.

[0080] If the above-mentioned control is decoupled from the output of oil fume emissions during the disturbance period, specifically: Based on the real-time emissions and short-term predicted emissions generated by the model, the monitoring system does not directly use them as external outputs or as the sole basis for determining exceedances and implementing linkage control. Instead, it further introduces control decoupling processing to adapt to short-term disturbances caused by actions such as starting and stopping, speed adjustment, power gear switching, or cleaning triggering at the catering site.

[0081] The specific implementation method is as follows: When in a disturbance period, the monitoring system does not directly use the instantaneous value of the oil fume emission at that moment to conclude that the emission exceeds the standard or to trigger a strong linkage. Instead, it generates an estimated result of the disturbance period according to the preset disturbance output rules. The estimated result of the disturbance period is jointly determined by the baseline value of the steady-state oil fume emission before the start of the disturbance, the change range of the oil fume emission during the disturbance, and the duration of the disturbance. At the same time, the monitoring system starts the recovery period observation. Starting from the end of the disturbance, it enters the preset recovery period time window and continuously calculates the short-term average value and change trend of the emission within the recovery period time window.

[0082] The aforementioned trend is defined by a monotonically decreasing trend over multiple consecutive sampling periods, with the decrease reaching a preset threshold and remaining stable for a preset duration. When the convergence criteria are met, the output will switch from the disturbance period estimation result to the steady-state emission result, and the steady-state emission result will be updated to the new baseline value. If the convergence criteria are not met by the end of the recovery period window, the monitoring system will maintain the non-convergence mark during the recovery period and reduce the credibility warning level of the output for that period. At the same time, it will only allow triggering of review or maintenance reminders and will not trigger strong linkage measures.

[0083] In this embodiment, after obtaining the real-time oil fume emission and short-term predicted emission, the monitoring system performs closed-loop control on the purification device according to the pre-fixed linkage rules: First, in each control cycle, it reads the real-time emission, the corresponding short-term predicted emission, as well as the credible prompts such as the wet interference risk level, the monitoring link health score, and the air volume estimation uncertainty level. Only when the wet interference risk level is lower than the preset risk threshold level, the monitoring link health score is higher than the preset low credible threshold level, and the air volume estimation uncertainty is lower than the preset high uncertainty threshold level, is it allowed to enter the strong linkage judgment.

[0084] In the strong linkage determination, if the real-time emission continuously exceeds the preset exceedance threshold for a preset duration, or if the short-term predicted emission will exceed the exceedance threshold within the preset prediction window and the exceedance reaches the preset predicted exceedance threshold, the monitoring system generates a control command to perform one or more of the following actions on the purification device: power level adjustment, spray or cleaning process triggering, and fan exhaust intensity adjustment. After the command is issued, it enters a preset response observation window, in which the real-time emission is continuously monitored again to determine the control effect: if the real-time emission drops below the preset drop threshold within the observation window and remains stable for a preset duration, the linkage is recorded as successful and normal control is restored; if the real-time emission does not drop or exceeds the exceedance threshold again after dropping, the purification device power level is increased or enhanced cleaning is triggered according to the preset upgrade strategy, and a maintenance alarm is output.

[0085] When the reliable prompt does not meet the strong linkage gating conditions, the monitoring system will not perform strong actions such as power increase or strong cleaning, but will only output a verification or maintenance reminder and maintain the current operating strategy, thereby avoiding false triggering of control due to uncertainty of emission volume under wet interference, low link reliability or air volume extrapolation conditions.

[0086] Reference Figure 2 As shown, the second aspect of the present invention provides a precise monitoring system for oil fume emissions based on multi-sensor fusion, including: a sensor data processing module, a disturbance period determination module, an oil fume basic emission calculation module, and an oil fume real-time emission output module.

[0087] The sensor data processing module is connected to the disturbance period determination module, the disturbance period determination module is connected to the basic emission calculation module for oily fumes, and the basic emission calculation module for oily fumes is connected to the real-time emission output module for oily fumes.

[0088] The sensor data processing module is used to collect raw multi-source sensor data of the calibrated monitoring target area based on the set monitoring nodes, perform basic validity screening and hard anomaly processing, and obtain effective multi-source sensor data.

[0089] The disturbance period determination module is used to extract the step change characteristics of effective multi-source sensor data, identify the disturbance period of control actions in the oil fume monitoring process, and mark it as a disturbance period and a non-disturbance period.

[0090] The basic emission calculation module for cooking fumes is used to generate a wet interference risk indication by calibrating the temperature and humidity state of the monitoring target area, and to perform dynamic time alignment of effective multi-source sensor data based on the wet interference risk indication constraint during non-disturbance periods, and to calculate the basic emission result of cooking fumes in the calibrated monitoring target area based on the alignment result.

[0091] The real-time emission output module for cooking fumes is used to extract the link health assessment set to determine the health score of the monitoring link. The basic emission result of cooking fumes is dynamically corrected based on the wet interference risk indication and the health score of the monitoring link. If the emission is within the disturbance period, the control decoupling is performed on the output of cooking fumes to complete the monitoring of cooking fumes. Example 2:

[0092] Without changing other aspects of Embodiment 1, this embodiment can also be applied to the environmental emission requirement verification and supervision of the regulatory platform.

[0093] Specifically, the regulatory platform pre-fixes the emission outlet codes corresponding to the emission outlets, basic store information, and environmental emission requirement parameter sets. The parameter sets include at least the applicable emission concentration limits or emission volume limits, statistical caliber and statistical period requirements, duration requirements for exceeding standards, and data quality requirements. The platform receives real-time emissions, short-term predicted emissions, wet interference risk levels, monitoring link health scores, air volume estimation uncertainty levels, and disturbance time period markers reported by each store. It then generates minute-level, ten-minute-level, and hour-level statistical results and cumulative amounts according to the fixed statistical period. At the same time, it statistically analyzes the proportion of missing measurements, low reliability, high wet risk, and extrapolation application and writes them into the report as data quality notes.

[0094] When performing exceedance verification, the platform adopts a dual-gating and dual-caliber strategy: During the time period when the wet interference risk level is lower than the preset risk gating level, the monitoring link health score reaches the preset confidence level, and the air volume estimation uncertainty is lower than the preset uncertainty gating level, exceedance judgment is made based on the real-time emission volume and its statistical average value and environmental protection limit. The exceedance state is required to be maintained continuously for a preset duration before an exceedance event is triggered. During the time period that does not meet the gating conditions, the platform does not directly include the instantaneous emission volume in the exceedance conclusion. Instead, it marks the data segment as quality-limited and transfers it to the review queue. The platform uses short-term predicted emission volume to indicate potential exceedance risks and trigger on-site review or maintenance verification, thereby avoiding misjudgments caused by water mist condensation, link contamination, or air volume extrapolation.

[0095] For triggered exceedance events, the platform will archive key evidence packages for the period in which the event occurred. The evidence packages will include at least: the start and end times of the exceedance, statistical caliber, corresponding emission sequence and mean, alignment relationship and update status, air volume source type and uncertainty level, wet risk level, health score level, disturbance marker, and equipment operating condition receipts and historical maintenance record indexes to support subsequent regulatory review.

[0096] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for accurate monitoring of oil fume emissions based on multi-sensor fusion, characterized in that, include: Based on the set monitoring nodes, the raw multi-source sensor data of the calibrated monitoring target area is collected, and basic validity screening and hard anomaly processing are performed to obtain valid multi-source sensor data. Extract the step change characteristics of effective multi-source sensor data, identify the control action disturbance period in the oil fume monitoring process, and mark it as the disturbance period and the non-disturbance period; By calibrating the temperature and humidity status of the monitoring target area, a wet interference risk indication is generated. During non-disturbance periods, dynamic time alignment of effective multi-source sensor data is performed based on the wet interference risk indication constraint. The basic emission of oil fume in the calibrated monitoring target area is calculated based on the alignment result. The wet interference risk indication is specifically: The dew point approach state and the high humidity threshold state are derived from flue gas temperature data and flue gas humidity data. By maintaining a short-term baseline using equivalent concentration data of oil fume particles, the short-term baseline is the median or moving average of the concentration data within a set short-term window. Based on the short-term baseline, the existence of short-term upward surges and the existence of delayed declines are determined. When both a short-term surge and a delayed fall are simultaneously satisfied, the concentration data is determined to exhibit a wet interference pattern. The wet interference risk level is defined based on the number of criteria that are met, namely, the state of approaching dew point, the state of reaching high humidity threshold, and the state of wet interference. The wet interference risk level is used as an indicator of wet interference risk. The dynamic time alignment of the effective multi-source sensor data is specifically performed as follows: Only within non-disturbance periods and time slices where the wet disturbance risk level is below the preset gate threshold, the concentration change trend and exhaust intensity change trend are extracted as alignable evidence. The alignment relationship is obtained by searching the correlation between air volume and concentration within a predetermined sliding window, and the alignment relationship is updated when the alignable evidence meets the minimum change amplitude condition. Based on the alignment relationship, the concentration data is resampled to establish a temporal correspondence with the air volume-related data at the same time. Extract the link health assessment set to determine the monitoring link health score, and use the wet interference risk indicator and the monitoring link health score as constraints to dynamically correct the basic emission result of oil fume. If it is within the disturbance period, control decoupling is performed on the oil fume emission output to complete the oil fume emission monitoring. The health score of the monitoring link is specifically as follows: Within each preset scoring period, stable operating condition segments are selected, and the concentration fluctuation degree within the stable operating condition segments is statistically analyzed and compared with the normal fluctuation range fixed during the installation and commissioning period to obtain noise conclusions. When a zeroing event occurs in the monitoring link, the low-level representative value of the concentration in the zeroing stable segment is extracted and compared with the historical zeroing representative value to determine the zero-point drift conclusion. The consistency of the concentration response direction is statistically analyzed based on the frequency event of the exhaust fan or the current step event of the exhaust fan to obtain a consistency conclusion. The zero-point drift conclusion, noise conclusion, air volume estimation uncertainty indication, and consistency conclusion are combined into a monitoring link health score.

2. The method for accurate monitoring of oil fume emissions based on multi-sensor fusion according to claim 1, characterized in that, Includes the following steps: The original multi-source sensor data includes equivalent concentration data of oil fume particles, flue gas temperature data, flue gas humidity data, flue gas pressure data, exhaust fan operating status data, air volume related data used to characterize exhaust intensity, and operating condition data. The basic validity screening includes at least range verification, negative value and unreachable value verification, continuous constant value verification, and communication missing measurement marker. The hard anomaly handling includes at least performing a median replacement on isolated spikes and retaining the replacement marker.

3. The method for accurate monitoring of oil fume emissions based on multi-sensor fusion according to claim 2, characterized in that, Includes the following steps: The operating condition data is used to record the operating status data of the exhaust fan and the operating condition data of the purification device. The operating status data of the exhaust fan includes at least one or more of the following: exhaust fan start / stop status, exhaust fan frequency conversion frequency and exhaust fan current. The operating condition data of the purification device includes at least one or more of the following: purification device working status, cleaning trigger status and power switching status. The air volume related data are direct measurement results of the smoke exhaust fan flow rate or smoke exhaust fan speed, or air volume estimation results; Based on the step change characteristics of operating condition data and air volume related data, the system detects whether there are triggering events and identifies the control action disturbance period during the monitoring process based on the event triggering time.

4. The method for accurate monitoring of oil fume emissions based on multi-sensor fusion according to claim 3, characterized in that: The air volume estimation results are as follows: During the installation and commissioning cycle, the current of the exhaust fan, the frequency of the exhaust fan, and the static pressure or differential pressure of the exhaust fan were collected at the working point group of the exhaust fan under stable operation. At the same time, the reference wind speed was measured by a portable reference wind measurement tool and the reference air volume was calculated by combining it with the cross-sectional size of the flue to form a dataset containing multiple sets of corresponding samples. Based on the dataset, the air volume estimation relationship and the debugging coverage range are obtained and solidified, and the fitting residuals are statistically analyzed to form the basic uncertainty. During online operation, the real-time collected data on the current of the exhaust fan, the frequency of the exhaust fan, and the static pressure or pressure difference of the exhaust fan are substituted into the air volume estimation relationship to obtain real-time air volume estimation data. Based on the basic uncertainty and whether the substituted parameters exceed the debugging coverage range, an air volume estimation uncertainty indicator is generated and used as the air volume estimation result.

5. The method for accurate monitoring of oil fume emissions based on multi-sensor fusion according to claim 1, characterized in that: The specific results of the basic emissions of oily fumes are as follows: Under the condition of uniform caliber, the equivalent concentration of oil fume particles after dynamic time alignment is converted to the preset concentration caliber, and the volumetric flow rate of flue gas discharged per unit time is determined based on air volume related data. At the same time, the concentration and volumetric flow rate of the preset caliber are multiplied to obtain the mass of oil fume emission per unit time. The mass of oil fume emission per unit time is accumulated according to the sampling period to obtain the basic emission result of oil fume within the statistical period.

6. The method for accurate monitoring of oil fume emissions based on multi-sensor fusion according to any one of claims 4 or 5, characterized in that: The dynamic correction of the baseline emissions of cooking fumes is specifically as follows: A multi-sensor data fusion model for error correction is constructed based on multi-source sensor data. The fusion model takes the basic emission result of oil fume and the operating condition data as input, and outputs the oil fume emission compensation correction amount and short-term predicted emission amount, thereby obtaining the corrected real-time oil fume emission amount and short-term predicted emission amount. The updates and magnitude constraints of the oil fume emission compensation corrections are based on wet interference risk indicators, monitoring link health scores, and air volume estimation uncertainty indicators.

7. A precise monitoring system for oil fume emissions based on multi-sensor fusion, employing the precise monitoring method for oil fume emissions based on multi-sensor fusion as described in any one of claims 1-6, characterized in that: include: The sensor data processing module is used to collect raw multi-source sensor data of the calibrated monitoring target area based on the set monitoring nodes, perform basic validity screening and hard anomaly processing, and obtain effective multi-source sensor data. The disturbance period determination module is used to extract the step change characteristics of effective multi-source sensor data, identify the disturbance period of control actions in the oil fume monitoring process, and mark it as the disturbance period and the non-disturbance period; The basic emission calculation module for oil fume is used to generate a wet interference risk indication by calibrating the temperature and humidity state of the monitoring target area, and to perform dynamic time alignment of effective multi-source sensor data based on the wet interference risk indication constraint during non-disturbance periods, and to calculate the basic emission result of oil fume in the calibrated monitoring target area based on the alignment result. The real-time emission output module for cooking fumes is used to extract the link health assessment set to determine the health score of the monitoring link. The module dynamically corrects the basic emission result of cooking fumes based on the wet interference risk indication and the health score of the monitoring link. If the emission is within the disturbance period, the module performs control decoupling on the output of cooking fumes to complete the monitoring of cooking fumes.

Images