Lampblack interference identification and compensation method in combustible gas monitoring network

Abstract

The invention discloses a lampblack interference identification and compensation method in a combustible gas monitoring network, which comprises the following steps: deploying a combustible gas monitoring network comprising a main sensor and an auxiliary sensor in a target place, the main sensor being used for detecting the concentration of combustible gas, and the auxiliary sensor being used for detecting the concentration of PM2.5 particles; historical data are collected, and a lampblack interference characteristic model is established based on a multivariate time sequence analysis method; collecting real-time data, and identifying lampblack interference in combustible gas monitoring data by using the established model; and carrying out self-adaptive compensation treatment on the identified oil smoke interference to obtain a compensated combustible gas concentration value. According to the method, the accuracy and reliability of combustible gas monitoring are improved, the method is particularly suitable for places with more lampblack such as kitchens and restaurants, the method does not depend on a large number of labeled samples and complex deep learning models, the system is easy to implement, and the resource demand is low.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of gas monitoring, and particularly relates to a cooking fume interference identification and compensation method in a combustible gas monitoring network. BACKGROUND

[0002] In the field of combustible gas monitoring, especially in places with more cooking fumes such as kitchens and restaurants, traditional combustible gas sensors often face the problem of cooking fume interference. Particulate matters in cooking fumes can be adsorbed on the surface of the sensor or enter the internal structure of the sensor, causing the performance of the sensor to decline, the reading to be inaccurate, and even false positives or false negatives, which seriously affects the reliability of gas safety monitoring.

[0003] Existing technologies usually use physical isolation, filtering devices or fixed threshold adjustment methods to solve the problem of cooking fume interference, but these methods are often difficult to adapt to complex and variable actual environments, and have a negative impact on the sensitivity and response speed of the sensor. Some advanced methods use deep learning models for interference identification, but such methods rely on a large number of labeled samples and require high computing resources, making it difficult to run in real time on resource-constrained sensor nodes.

[0004] Therefore, there is a need for an efficient and adaptive method to identify and compensate for cooking fume interference in combustible gas monitoring to improve the accuracy and reliability of the monitoring system. SUMMARY

[0005] The purpose of the present application is to provide a cooking fume interference identification and compensation method in a combustible gas monitoring network, based on multivariate time series analysis and a primary and secondary sensor cooperative working mechanism, without relying on a large number of labeled samples and complex deep learning models, to achieve the purpose of cooking fume interference identification and compensation diagnosis.

[0006] To achieve the above-mentioned purpose of the application, the specific technical solution is: a cooking fume interference identification and compensation method in a combustible gas monitoring network, the method comprising the following steps:

[0007] Step S1, deploying a combustible gas monitoring network in a target place, the combustible gas monitoring network comprising at least one primary sensor and one auxiliary sensor, the primary sensor being used to detect the concentration of combustible gas, and the auxiliary sensor being used to detect the concentration of PM2.5 particles.

[0008] Step S2, collecting historical data of the primary sensor and the auxiliary sensor of each node, and based on the historical data, establishing a cooking fume interference feature model according to a multivariate time series analysis method.

[0009] Step S3, collecting real-time combustible gas concentration data of the primary sensor and real-time PM2.5 particle concentration data of the auxiliary sensor, and using the feature model to identify cooking fume interference in the combustible gas monitoring data.

[0010] Step S4: Perform adaptive compensation processing on the identified oil fume interference to obtain the compensated combustible gas concentration value.

[0011] In step S2, the specific process of establishing the oil fume interference feature model is as follows:

[0012] Step S21: Collect the combustible gas concentration data sequence G(t) from the main sensor and the PM2.5 particle concentration data sequence P(t) from the auxiliary sensor;

[0013] t represents the sampling time point, t = 1, 2, ..., T, where T is the total number of samples;

[0014] Step S22: Perform time synchronization and normalization on G(t) and P(t) to obtain normalized data sequences G′(t) and P′(t);

[0015] Where, μ G and μ P Let σ be the mean of G(t) and P(t), respectively. G and σ P Let G(t) and P(t) be the standard deviations, respectively.

[0016] Step S23: Based on the normalized data sequence, establish a multivariate time series model;

[0017] Where α0 is the model constant term, representing the baseline value; α i β is the autoregressive coefficient, representing the weight of the influence of historical combustible gas concentration on the current value, i = 1, 2, ..., p, where p is the order of the autoregressive term; j is the external input coefficient, representing the influence weight of PM2.5 particle concentration on the reading of combustible gas sensor, j = 0, 1, ..., q, and Q is the order of the external input term.

[0018] The formula for calculating the model constant term α0 is as follows:

[0019] Where, μ G′ and μ P′ These are the normalized average values ​​of combustible gas concentration and PM2.5 concentration, respectively.

[0020] The method for determining the order p of the autoregressive term and the order q of the external input term is as follows:

[0021] Model order selection is performed using the Bayesian Information Criterion (BIC).

[0022] in, The residual variance is used to search for different combinations of p and q, where p∈[1,5] and q∈[0,5], and the combination that minimizes the BIC value is selected as the final model order.

[0023] The autoregressive coefficient α i and external input coefficient β j The method for determining α is as follows: The least squares method is used to estimate the parameters of the model; the autoregressive coefficient and the external input coefficient that minimize the sum of squared residuals are α. i and β j The possible values ​​of:

[0024]

[0025] In step S3, the specific process for identifying oil fume interference in the combustible gas monitoring data is as follows:

[0026] Step S31: Collect real-time combustible gas concentration data G 0(t) and real-time PM2.5 particulate concentration data P 0(t) ;

[0027] Step S32: Normalize the real-time data to obtain G′ 0(t) and P′ 0(t) :

[0028] Where, μ G σ G μ P and σ P Statistical values ​​from historical data;

[0029] Step S33: Calculate the estimated value of the current combustible gas concentration based on the established multivariate time series model;

[0030] Among them, G′ e(t) G′ is the normalized combustible gas concentration value estimated by the model. 0(t-i) Let P′ be the normalized combustible gas concentration at time i in the past. 0(t-j) The normalized PM2.5 particulate concentration at time j in the past;

[0031] Step S34: Calculate the residual between the actual value and the estimated value;

[0032] e(t)=G′0(t)-G′ e (t), where the residual represents the deviation between the actual observed value and the model prediction;

[0033] Step S35: When the residual e(t) meets the following condition, it is determined that there is oil fume interference:

[0034] |e(t)|>k·σ e , where σ e is the standard deviation of historical residuals, used to measure the normal fluctuation range of model fitting error; k is the threshold coefficient, with a value range of 2.5-3.5, used to determine the sensitivity of anomaly detection. The larger the value of k, the more conservative the detection.

[0035] The standard deviation σ of the historical residuals e The calculation method is as follows:

[0036] The historical data were fitted using the established multivariate time series model, and the residual e at each time point was calculated. hist (t):

[0037] e hist (t)=G′(t)-G e ′(t); Calculate the standard deviation of all residuals: Where, μ e Let N be the mean of the historical residuals, and N be the total number of historical data.

[0038] In step S4, the specific process of adaptive compensation for oil fume interference is as follows:

[0039] Step S41, establish a compensation model for oil fume interference:

[0040] Among them, G c ′(t) is the compensated normalized combustible gas concentration; γ(t) is the adaptive compensation coefficient, used to adjust the compensation intensity; The term represents the interference component caused by PM2.5 particle concentration;

[0041] The compensation coefficient γ(t) is determined using an adaptive adjustment method:

[0042] Where γ0 is the basic compensation coefficient, λ is the adjustment factor that controls the sensitivity of the compensation intensity to changes in PM2.5 concentration; P′ th The normalized PM2.5 concentration threshold indicates that the PM2.5 concentration level of the combustible gas sensor begins to significantly affect the signal.

[0043] Step S42, the compensated normalized value G′ c(t) Converting back to the original scale yields the final compensated combustible gas concentration value G. c(t) :

[0044] G c (t)=G′ c (t)·σ G +μ G .

[0045] The method for determining the basic compensation coefficient γ0 is as follows: by analyzing historical data, the response deviation of the combustible gas sensor under different PM2.5 concentration levels is calculated;

[0046] The basic compensation coefficient γ0 ranges from 0.7 to 0.9, and its specific value is determined according to the following formula:

[0047] Only |e(t)|>k·σ is calculated. e and Data points.

[0048] The normalized PM2.5 concentration threshold P′ th Based on the relationship between PM2.5 concentration and combustible gas sensor readings in historical data, the PM2.5 concentration threshold that significantly affects combustible gas sensor readings was determined: P′ th =μ P′ +η·σ P′ Wherein, η is an adjustment parameter, with a value range of 1.0-1.5, determined according to the interference sensitivity required by the target location.

[0049] Compared with the prior art, the beneficial effects of this invention are:

[0050] This invention deploys a monitoring network with primary and secondary sensors working in tandem, and employs a multivariate time-series analysis method to establish a characteristic model of oil fume interference. This enables accurate identification and adaptive compensation of oil fume interference in combustible gas monitoring. It not only eliminates the reliance on a large number of labeled samples and complex deep learning models, reducing system complexity and resource requirements, but also, through an adaptive compensation mechanism, dynamically adjusts the compensation intensity according to different oil fume concentration levels, significantly improving the accuracy and reliability of combustible gas monitoring. It is particularly suitable for places with a lot of oil fume, such as kitchens and restaurants, providing a more reliable technical guarantee for gas safety monitoring. Attached Figure Description

[0051] Figure 1 This is a flowchart of the method for identifying and compensating for oil fume interference in the combustible gas monitoring network of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of this invention, not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0053] like Figure 1The diagram shown is a flowchart of the oil fume interference identification and compensation method in the combustible gas monitoring network of the present invention. The method includes the following steps:

[0054] Step S1: Deploy a combustible gas monitoring network at the target location. The combustible gas monitoring network includes at least one main sensor and one auxiliary sensor. The main sensor is used to detect the concentration of combustible gas, and the auxiliary sensor is used to detect the concentration of PM2.5 particles.

[0055] The deployment of a gas monitoring network requires careful consideration of the rational arrangement of primary and secondary sensors. For example, in a typical commercial kitchen environment, primary sensors (detecting combustible gases such as methane and propane) can be installed near gas pipes and approximately 1.5-2 meters above the stove, while secondary sensors (detecting PM2.5 particles) can be installed in close proximity but away from direct heat sources, with a distance of 0.3-0.5 meters between them. This ensures that the collected data has spatial correlation while reflecting the physical quantities being measured. In large restaurant kitchens, it is recommended to deploy a set of primary and secondary sensors every 50-80 square meters to form a networked monitoring system. Each node transmits data to the central processing unit via wired or wireless means.

[0056] Step S2: Collect historical data from the main and auxiliary sensors of each node, and establish a characteristic model of oil fume interference based on the historical data and the multivariate time series analysis method.

[0057] In step S2, the specific process of establishing the oil fume interference feature model is as follows:

[0058] Step S21: Collect the combustible gas concentration data sequence G(t) from the main sensor and the PM2.5 particle concentration data sequence P(t) from the auxiliary sensor;

[0059] Let t represent the sampling time point, t = 1, 2, ..., T, where T is the total number of samples. During data sequence collection, it is crucial to ensure that the timestamps of G(t) and P(t) are perfectly aligned. Hardware synchronization triggering or software time compensation methods can be used to ensure data time consistency. For example, when two sensors have different response times and sampling frequencies (gas sensor response times are typically 10-30 seconds, while PM2.5 sensor response times may be 1-5 seconds), interpolation or downsampling techniques can be used to unify the two sets of data to the same time series. In a certain test environment, two types of sensors operated with sampling intervals of 2 seconds and 5 seconds, respectively. By using linear interpolation, all data were resampled to a unified 5-second interval time series, resolving the data synchronization problem.

[0060] Step S22: Perform time synchronization and normalization on G(t) and P(t) to obtain normalized data sequences G′(t) and P′(t);

[0061] Where, μ G and μ P Let σ be the mean of G(t) and P(t), respectively. G and σ P These are the standard deviations of G(t) and P(t), respectively. Using the Z-score standardization method not only makes data of different magnitudes comparable but also reduces the impact of outliers on the model. For example, in a certain test, the original range of combustible gas concentration data was 0-100 ppm, with a mean of approximately 12 ppm and a standard deviation of approximately 8 ppm; while the original range of PM2.5 concentration was 0-500 μg / m³. 3 The average value is approximately 75 μg / m 3 The standard deviation is approximately 65 μg / m 3 After normalization, both sets of data were converted to a standard normal distribution with a mean of 0 and a standard deviation of 1, which greatly improved the stability and accuracy of model training. It is important to note that the normalization parameters (mean and standard deviation) should be saved for subsequent real-time data processing.

[0062] Step S23: Based on the normalized data sequence, establish a multivariate time series model;

[0063] Where α0 is the model constant term, representing the baseline value; α i β is the autoregressive coefficient, representing the weight of the influence of historical combustible gas concentration on the current value, i = 1, 2, ..., p, where p is the order of the autoregressive term; j is the external input coefficient, representing the influence weight of PM2.5 particle concentration on the reading of combustible gas sensor, j = 0, 1, ..., q, where q is the order of the external input term.

[0064] This model captures the influence of two factors on combustible gas sensor readings: the continuity of its own historical readings (represented by an autoregressive term) and the interference effect of PM2.5 particulate concentration (represented by an external input term). For example, in a real-world test case of a commercial kitchen, when cooking activity suddenly resumed from an idle state, the PM2.5 concentration rapidly increased from 30 μg / m³. 3 Rise to 250 μg / m 3 Simultaneously, the combustible gas sensor reading abnormally increased from 5 ppm to 28 ppm, even though the actual combustible gas concentration (verified by calibration equipment) only increased to 7 ppm. In this situation, the model can learn the correlation pattern between PM2.5 concentration and abnormal combustible gas sensor readings, providing a basis for subsequent interference identification.

[0065] During the historical data collection phase, at least 14 days of continuous monitoring data are collected, covering different working conditions of the target location (such as peak cooking periods, idle periods, etc.).

[0066] For example, in a test at a restaurant, the system collected sensor data from 6:00 AM to 10:00 PM daily, with a sampling frequency of once per minute, accumulating over 15,000 data points as the basis for model training. After data collection, data cleaning should be performed to remove obvious outliers (such as data anomalies caused by sensor malfunctions or power fluctuations) to improve modeling quality.

[0067] In a real-world application case, box plots were used to identify and remove approximately 2% of outlier data points, which greatly improved the accuracy of subsequent modeling.

[0068] The formula for calculating the model constant term α0 is as follows:

[0069] Where, μ G′ and μ P′ These are the normalized average concentrations of combustible gases and PM2.5, respectively.

[0070] The model constant α0 represents the system's baseline value. Its calculation ensures that when all input variables are at their respective means, the output variable is also at its mean. In a practical application, modeling with 2000 sets of historical data yielded an α0 value of 0.137, indicating a small positive deviation when the system is at its "average state." This constant allows the model to more accurately represent the system's performance under different operating conditions, especially considering potential zero-point drift and gain variations inherent in the sensor itself. Furthermore, during long-term operation, it is recommended to recalculate the α0 value every 3-6 months to address the issue of sensor performance changing over time.

[0071] The method for determining the order p of the autoregressive term and the order q of the external input term is as follows:

[0072] Model order selection is performed using the Bayesian Information Criterion (BIC).

[0073] in, The residual variance is used to search for different combinations of p and q, where p∈[1,5] and q∈[0,5], and the combination that minimizes the BIC value is selected as the final model order.

[0074] The Bayesian Information Criterion (BIC) takes into account both model fit and complexity in the selection of model order, thus avoiding overfitting.

[0075] For example, in a restaurant kitchen implementation case, the system iterated through 25 different combinations of p and q (p from 1 to 5, q from 0 to 5), calculated the BIC value corresponding to each combination, and finally determined that the BIC value was the smallest when p=2 and q=1 (BIC=-4218.6). This means that the current combustible gas concentration is mainly affected by the historical concentrations of the previous two moments and the current PM2.5 concentration. This result is consistent with the cooking characteristics of this location—the impact of short-term oil fume bursts is mainly manifested at the current moment, while the combustible gas concentration has a certain continuity. In different application scenarios, the optimal p and q values ​​may be different, reflecting the differences in sensor response characteristics under different environments.

[0076] The autoregressive coefficient α i and external input coefficient β j The method for determining α is as follows: The least squares method is used to estimate the parameters of the model; the autoregressive coefficient and the external input coefficient that minimize the sum of squared residuals are α. i and β j The possible values ​​of:

[0077]

[0078] Step S3: Collect real-time combustible gas concentration data from the main sensor and real-time oil PM2.5 particle concentration data from the auxiliary sensor, and use the feature model to identify oil fume interference in the combustible gas monitoring data.

[0079] In step S3, the specific process for identifying oil fume interference in the combustible gas monitoring data is as follows:

[0080] Step S31: Collect real-time combustible gas concentration data G 0(t) and real-time PM2.5 particulate concentration data P 0(t) ;

[0081] Step S32: Normalize the real-time data to obtain G′ 0(t) and P′ 0(t) :

[0082] Where, μ G σ G μ P and σ P Statistical values ​​from historical data;

[0083] Step S33: Calculate the estimated value of the current combustible gas concentration based on the established multivariate time series model;

[0084] Among them, G′ e(t) G′ is the normalized combustible gas concentration value estimated by the model. 0(t-i)Let P′ be the normalized combustible gas concentration at time i in the past. 0(t-j) The normalized PM2.5 particulate concentration at time j in the past;

[0085] Step S34: Calculate the residual between the actual value and the estimated value;

[0086] e(t)=G′0(t)-G′ e (t), where the residual represents the deviation between the actual observed value and the model prediction;

[0087] Step S35: When the residual e(t) meets the following condition, it is determined that there is oil fume interference:

[0088] |e(t)|>k·σ e , where σ e is the standard deviation of historical residuals, used to measure the normal fluctuation range of model fitting error; k is the threshold coefficient, with a value range of 2.5-3.5, used to determine the sensitivity of anomaly detection. The larger the value of k, the more conservative the detection.

[0089] When determining the threshold setting for oil fume interference, the choice of the threshold coefficient k should balance detection sensitivity with the risk of false alarms. For example, in a densely populated commercial kitchen, k can be set to 2.5-2.8 to improve the system's sensitivity to potential risks; while in well-ventilated, sparsely populated areas, k can be set to 3.2-3.5 to reduce unnecessary false alarms. In a case study of a chain restaurant, k=2.7 was used during peak daytime hours, and automatically switched to k=3.3 during off-peak nighttime hours. This dynamic threshold strategy ensures that the system maintains appropriate sensitivity under different operating conditions. Furthermore, the system can gradually optimize the k value based on actual false alarms / missed alarms using an adaptive learning algorithm. For example, a Bayesian optimization method based on historical alarm accuracy can be used to make the k value automatically approach its optimal value over time.

[0090] The standard deviation σ of the historical residuals e The calculation method is as follows:

[0091] The historical data were fitted using the established multivariate time series model, and the residual e at each time point was calculated. hist (t):

[0092] e hist (t)=G′(t)-G e ′(t); Calculate the standard deviation of all residuals: Where, μ e Let N be the mean of the historical residuals, and N be the total number of historical data.

[0093] Outliers in historical data should be considered for σ eThe impact of extreme outliers can be mitigated. For example, truncated statistics can be used to calculate the standard deviation using only the middle 95% of the residual data, thus removing the influence of extreme outliers. In a case study of an office building cafeteria, the originally calculated σ... e The initial value was 0.328, which was corrected to 0.273 using a truncated statistical method, improving the accuracy of anomaly detection. Additionally, σ varies depending on the type of location. e The values ​​can vary significantly, reflecting differences in environmental complexity. For example, the σ value for an open kitchen... e Typically, σ is 20%-30% higher than that of a closed kitchen, which needs to be considered during system configuration. During long-term operation, it is recommended to recalculate σ every 3-6 months. e The value is adjusted to accommodate possible environmental changes or changes in sensor performance.

[0094] Step S4: Perform adaptive compensation processing on the identified oil fume interference to obtain the compensated combustible gas concentration value.

[0095] In step S4, the specific process of adaptive compensation for oil fume interference is as follows:

[0096] Step S41, establish a compensation model for oil fume interference:

[0097] Among them, G c ′(t) is the compensated normalized combustible gas concentration; γ(t) is the adaptive compensation coefficient, used to adjust the compensation intensity; The term represents the interference component caused by PM2.5 particle concentration;

[0098] The compensation coefficient γ(t) is determined using an adaptive adjustment method:

[0099] Where γ0 is the basic compensation coefficient, λ is the adjustment factor that controls the sensitivity of the compensation intensity to changes in PM2.5 concentration; P′ th The normalized PM2.5 concentration threshold indicates that the PM2.5 concentration level of the combustible gas sensor begins to significantly affect the signal.

[0100] Step S42, the compensated normalized value G′ c(t) Converting back to the original scale yields the final compensated combustible gas concentration value G. c(t) :

[0101] G c (t)=G′ c (t)·σ G +μ G .

[0102] The method for determining the basic compensation coefficient γ0 is as follows: by analyzing historical data, the response deviation of the combustible gas sensor under different PM2.5 concentration levels is calculated;

[0103] The basic compensation coefficient γ0 ranges from 0.7 to 0.9, and its specific value is determined according to the following formula:

[0104] Only |e(t)|>k·σ is calculated. e and Data points.

[0105] The normalized PM2.5 concentration threshold P′ th Based on the relationship between PM2.5 concentration and combustible gas sensor readings in historical data, the PM2.5 concentration threshold that significantly affects combustible gas sensor readings was determined: P′ th =μ P′ +η·σ P′ Wherein, η is an adjustment parameter, ranging from 1.0 to 1.5, determined according to the required interference sensitivity of the target location. For special environments (such as high humidity locations), the value of η needs to be specially adjusted to cope with the combined effects of humidity on the PM2.5 sensor and the combustible gas sensor.

[0106] Taking the actual application of this invention in the central kitchen of a large chain restaurant as an example, the technical effects of this invention are illustrated. This central kitchen covers an area of ​​approximately 400 square meters and is equipped with multiple high-powered woks and steaming / cooking equipment, with a daily output exceeding 3,000 meals. Due to the high intensity of cooking and drastic fluctuations in oil fume concentration, traditional combustible gas alarm systems frequently produce false alarms, triggering an average of 5-7 unnecessary alarms per week. This severely impacts the normal operation of the kitchen and reduces employees' sensitivity to alarm signals, creating potential safety hazards.

[0107] After applying the method of this invention, the kitchen deployed eight sets of monitoring nodes with primary and secondary sensors working in tandem, covering all gas usage areas. Data analysis after one month of system operation showed that the false alarm rate decreased by 92%, with only two false alarms occurring (both under extreme cooking conditions). Simultaneously, the system successfully identified and triggered alarms for two actual gas leak events (one a minor leak caused by a loose valve, and the other a short-term leak caused by improper employee operation). Particularly noteworthy is that in the second leak event, the leak location was precisely in an area with a large amount of cooking fumes. Traditional systems would likely misinterpret this as fumes and ignore it, while this system, through multivariate time-series analysis, accurately distinguished the superimposed signals of fumes interference and actual leaks, triggering an alarm in a timely manner and preventing potential safety accidents.

[0108] From a technical perspective, before applying the method of this invention, the combustible gas monitoring system in this kitchen could only detect fumes exceeding 200 μg / m³.3 Previously, the average deviation of combustible gas concentration readings was 62%, with the highest deviation reaching 115%. After applying the method of this invention, under the same oil fume concentration conditions, the average deviation of the compensated combustible gas concentration readings was reduced to 7.5%, with the maximum deviation not exceeding 15%, significantly improving measurement accuracy. Furthermore, the interference identification and compensation algorithm of this system has a computational latency of no more than 200 milliseconds on the embedded processor, fully meeting the requirements of real-time monitoring.

[0109] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and compensating for oil fume interference in a combustible gas monitoring network, characterized in that, The method includes the following steps: Step S1: Deploy a combustible gas monitoring network at the target location. The combustible gas monitoring network includes at least one main sensor and one auxiliary sensor. The main sensor is used to detect the concentration of combustible gas, and the auxiliary sensor is used to detect the concentration of PM2.5 particles. Step S2: Collect historical data from the main and auxiliary sensors of each node, and establish a characteristic model of oil fume interference based on the historical data and the multivariate time series analysis method. Step S3: Collect real-time combustible gas concentration data from the main sensor and real-time oil PM2.5 particle concentration data from the auxiliary sensor, and use the feature model to identify oil fume interference in the combustible gas monitoring data. Step S4: Perform adaptive compensation processing on the identified oil fume interference to obtain the compensated combustible gas concentration value.

2. The method according to claim 1, characterized in that, In step S2, the specific process of establishing the oil fume interference feature model is as follows: Step S21: Collect the combustible gas concentration data sequence G(t) from the main sensor and the PM2.5 particle concentration data sequence P(t) from the auxiliary sensor; t represents the sampling time point, t = 1, 2, ..., T, where T is the total number of samples; Step S22: Perform time synchronization and normalization on G(t) and P(t) to obtain normalized data sequences G′(t) and P′(t); Where, μ G and μ P Let σ be the mean of G(t) and P(t), respectively. G and σ P Let G(t) and P(t) be the standard deviations, respectively. Step S23: Based on the normalized data sequence, establish a multivariate time series model; Where α0 is the model constant term, representing the baseline value; α i β is the autoregressive coefficient, representing the weight of the influence of historical combustible gas concentration on the current value, i = 1, 2, ..., p, where p is the order of the autoregressive term; j is the external input coefficient, representing the influence weight of PM2.5 particle concentration on the reading of combustible gas sensor, j = 0, 1, ..., q, where q is the order of the external input term.

3. The method according to claim 2, characterized in that, The formula for calculating the model constant term α0 is as follows: Where, μ G′ and μ P′ These are the normalized average values ​​of combustible gas concentration and PM2.5 concentration, respectively.

4. The method according to claim 2, characterized in that, The method for determining the order p of the autoregressive term and the order q of the external input term is as follows: Model order selection is performed using the Bayesian Information Criterion (BIC). in, The residual variance is used to search for different combinations of p and q, where p∈[1,5] and q∈[0,5], and the combination that minimizes the BIC value is selected as the final model order.

5. The method according to claim 4, characterized in that, The autoregressive coefficient α i and external input coefficient β j The method for determining α is as follows: The least squares method is used to estimate the parameters of the model; the autoregressive coefficient and the external input coefficient that minimize the sum of squared residuals are α. i and β j The possible values ​​of:

6. The method according to claim 5, characterized in that, In step S3, the specific process for identifying oil fume interference in the combustible gas monitoring data is as follows: Step S31: Collect real-time combustible gas concentration data G 0(t) and real-time PM2.5 particulate concentration data P 0(t) ; Step S32: Normalize the real-time data to obtain G′ 0(t) and P′ 0(t) : Where, μ G σ G μ P and σ P Statistical values ​​from historical data; Step S33: Calculate the estimated value of the current combustible gas concentration based on the established multivariate time series model; Among them, G′ e(t) G′ is the normalized combustible gas concentration value estimated by the model. 0(t-i) Let P′ be the normalized combustible gas concentration at time i in the past. 0(t-j) The normalized PM2.5 particulate concentration at time j in the past; Step S34: Calculate the residual between the actual value and the estimated value; e(t)=G′0(t)-G′ e (t), where the residual represents the deviation between the actual observed value and the model prediction; Step S35: When the residual e(t) meets the following condition, it is determined that there is oil fume interference: |e(t)|>k·σ e , where σ e is the standard deviation of historical residuals, used to measure the normal fluctuation range of model fitting error; k is the threshold coefficient, with a value range of 2.5-3.5, used to determine the sensitivity of anomaly detection. The larger the value of k, the more conservative the detection.

7. The method according to claim 6, characterized in that, The standard deviation σ of the historical residuals e The calculation method is as follows: The historical data were fitted using the established multivariate time series model, and the residual e at each time point was calculated. hist (t): e hist (t)=G′(t)-G e ′(t); Calculate the standard deviation of all residuals: Where, μ e Let N be the mean of the historical residuals, and N be the total number of historical data.

8. The method according to claim 6, characterized in that, In step S4, the specific process of adaptive compensation for oil fume interference is as follows: Step S41, establish a compensation model for oil fume interference: Among them, G′ c (t) represents the compensated normalized combustible gas concentration; γ(t) is the adaptive compensation coefficient, used to adjust the compensation intensity; The term represents the interference component caused by PM2.5 particle concentration; The compensation coefficient γ(t) is determined using an adaptive adjustment method: Where γ0 is the basic compensation coefficient, λ is the adjustment factor that controls the sensitivity of the compensation intensity to changes in PM2.5 concentration; P′ th The normalized PM2.5 concentration threshold indicates that the PM2.5 concentration level of the combustible gas sensor begins to significantly affect the signal. Step S42, the compensated normalized value G′ c(t) Converting back to the original scale yields the final compensated combustible gas concentration value G. c(t) : G c (t)=G′ c (t)·s G +m G 。 9. The method according to claim 8, characterized in that, The method for determining the basic compensation coefficient γ0 is as follows: by analyzing historical data, the response deviation of the combustible gas sensor under different PM2.5 concentration levels is calculated; The basic compensation coefficient γ0 ranges from 0.7 to 0.9, and its specific value is determined according to the following formula: Only |e(t)|>k·σ is calculated. e and Data points.

10. The method according to claim 8, characterized in that, The normalized PM2.5 concentration threshold P′ th Based on the relationship between PM2.5 concentration and combustible gas sensor readings in historical data, the PM2.5 concentration threshold that significantly affects combustible gas sensor readings was determined: P′ th =μ P′ +η·σ P′ Wherein, η is an adjustment parameter, with a value range of 1.0-1.5, determined according to the interference sensitivity required by the target location.

Images