Intelligent judgment and switching method for saturation state of activated carbon adsorption bed for polyolefin waste gas treatment

Abstract

This invention provides an intelligent method for judging and switching the saturation state of activated carbon adsorption beds for polyolefin waste gas treatment. By collecting key operating parameters of the adsorption bed in real time, a six-dimensional basic time-series dataset is constructed. After synchronous filtering, signal conditioning, and data alignment, the outlet concentration curvature, bed pressure difference growth characteristics, and disturbance intensity factor are extracted. Based on a condition-adaptive multi-scale weighting mechanism, multiple phase transition criteria are integrated to dynamically judge the operating state of the adsorption bed. Under abnormal disturbances, the criterion weights are automatically adjusted, and short-time hysteresis loop verification ensures the stability and reliability of the judgment results. Finally, the timing of adsorption bed switching is strictly controlled in the form of structured tags, effectively reducing the risk of misjudgment, improving operational safety, and resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of industrial waste gas treatment and intelligent monitoring technology of adsorption processes, and particularly to a method for intelligent judgment and switching of saturation state of activated carbon adsorption bed for polyolefin waste gas treatment. Background Technology

[0002] Currently, activated carbon adsorption beds are widely used as the core unit for VOCs (volatile organic compounds) treatment in the field of polyolefin industrial waste gas treatment. Accurate determination of the adsorption bed saturation state is directly related to pollutant emission compliance and system safety operation. Existing mainstream technologies monitor the adsorption bed state in the following ways: one approach uses the outlet concentration threshold method, considering an excess of VOCs concentration at the adsorption bed outlet as a critical saturation state, triggering a switching operation; another approach uses statistical modeling of the relationship between the cumulative change trend of bed pressure difference and adsorption capacity to achieve coarse classification of the operating phase; some studies combine gas dynamics models and adsorption isotherm theory, using offline trained fitting functions to predict the adsorption process state and estimate capacity. In recent years, industrial automation and intelligent identification technologies have gradually penetrated the field of adsorption bed monitoring. Some systems have introduced multi-sensor fusion technology, empirical rule bases, and data-driven black-box models to improve the intelligence and adaptability of saturation determination. However, the following shortcomings and unmet needs still exist in the existing technology: (1) The traditional concentration threshold method cannot distinguish between the penetration of the adsorption bed caused by physical saturation and the false penetration caused by short-term abnormal conditions. Its misjudgment rate is high, which can easily lead to the delay or over-switching of switching instructions and affect the continuous operation efficiency of the system. (2) The empirical statistical model is limited by offline data calibration and static parameter setting, lacks physical mapping of the real kinetic behavior of the adsorption process, and is difficult to reflect the long-term degradation effects such as activated carbon aging and changes in environmental humidity. It lacks adaptability to long-term operating systems. (3) Traditional solutions have a high failure rate of criteria under abnormal disturbances (such as sudden changes in airflow, production line fluctuations, etc.), and the model is not robust enough. They are prone to false triggering or missed alarms, which increases the risk of VOCs penetration emissions and affects the emission compliance rate and process safety assurance. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, this invention provides an intelligent method for judging and switching the saturation state of activated carbon adsorption beds for the treatment of polyolefin waste gas.

[0004] The technical solution of this invention is implemented as follows: A method for intelligent judgment and switching of the saturation state of an activated carbon adsorption bed for polyolefin waste gas treatment, comprising: S1: Real-time collection of volatile organic compound concentrations at the inlet and outlet of the adsorption bed, bed pressure difference, cumulative operating time, instantaneous air volume, and inlet concentration constitutes a six-dimensional basic time series dataset, which serves as the raw input data for saturation state judgment; S2: Calculate the outlet concentration curvature feature sequence based on the six-dimensional basic time series dataset, extract the curvature value through the second difference of the sliding window, and calibrate the dynamic curvature threshold according to the historical penetration samples to generate the initial judgment basis for the concentration response phase change. S3: Based on the integral of cumulative adsorption load, calculate the slope characteristics of the pressure difference growth rate, model the correlation between the bed pressure difference increment and the cumulative adsorption amount, identify the slope transition point and generate the pressure difference growth phase transition criterion. S4: Based on the instantaneous air volume change rate and the fluctuation amplitude of the intake air concentration, a disturbance intensity factor is constructed, and the fusion weight coefficients of the concentration curvature criterion, the pressure difference growth slope criterion, and the remaining capacity ratio criterion are dynamically adjusted to form an adaptive weight set for the operating condition. S5: Input the initial judgment criteria for phase change based on concentration response, the judgment criteria for phase change based on pressure difference growth, and the feature of remaining capacity ratio into the multi-scale phase change threshold judgment model, and combine the working condition adaptive weight set to perform phase change flag activation judgment to generate a comprehensive phase change confirmation event. S6: When the disturbance intensity factor exceeds the preset threshold, a 15-second short-time hysteresis loop verification mechanism is activated to verify the continuous validity of the phase transition flag within the hysteresis time window and output a verification confirmation signal. S7: Generate a structured saturation state label based on the verification confirmation signal and the phase transition flag activation state, which includes the operation stage identifier, confidence index, trigger reason code and switching time window parameters; S8: Input the structured saturation state label into the switching control unit. When the operating stage in the label is identified as the critical zone of penetration and the confidence index exceeds the safety threshold, trigger the adsorption bed switching command and update the status of the queue to be desorbed.

[0005] The intelligent judgment and switching method for saturation state of activated carbon adsorption bed for polyolefin waste gas treatment provided by this invention has the following beneficial effects: (1) This invention significantly improves the accuracy and physical interpretability of state division by constructing a three-stage state identification mechanism based on the phase transition characteristics of adsorption kinetics. The sudden change in outlet concentration curvature is used as the initial criterion for "concentration response phase transition." The "pressure difference growth phase transition" is captured by combining the trend of pressure difference increment slope with the cumulative adsorption amount. Furthermore, the remaining capacity ratio after inverting the effective adsorption capacity is used for multi-dimensional cross-validation, achieving refined identification of the adsorption bed operation stages. This method not only avoids dependence on a large number of labeled samples but also achieves millisecond-level response and hundreds of millisecond-level delay warnings for key phase transition nodes without the need for deep learning inference. This is significantly superior to the judgment sensitivity and stability of traditional fixed threshold methods or statistical models in dynamic scenarios. (2) This invention introduces a disturbance intensity factor and a dynamic weight adjustment mechanism, along with short-time lag loop verification logic, which greatly enhances the robustness and fault tolerance of the judgment system. When a drastic change in flow rate or intake concentration is detected, the system automatically reduces the dependence weight on curvature index and instead strengthens the decision-making proportion of pressure difference evolution trend and remaining capacity assessment to prevent false phase transition signals caused by transient disturbances. At the same time, by setting continuous satisfaction requirements within a 15-second lag window, the interference caused by random noise and short-term fluctuations is effectively filtered out. In addition, all core parameters are updated and calibrated in real time using the online recursive least squares method, enabling the model to adaptively track long-term degradation effects such as activated carbon performance decay and environmental temperature and humidity drift, completely eliminating the strong dependence of offline training mode on initial assumptions, and ensuring that the system maintains a high confidence judgment capability throughout the operating cycle of several months or even several years. Attached Figure Description

[0006] Figure 1 This is a flowchart of the intelligent judgment and switching method for saturation state of activated carbon adsorption bed for polyolefin waste gas treatment according to the present invention. Figure 2 This is a sub-flowchart of the intelligent judgment and switching method for saturation state of activated carbon adsorption bed for polyolefin waste gas treatment of the present invention. Figure 3 This is another sub-flowchart of the intelligent judgment and switching method for saturated state of activated carbon adsorption bed for polyolefin waste gas treatment of the present invention. Figure 4 This is another sub-flowchart of the intelligent judgment and switching method for saturation state of activated carbon adsorption bed for polyolefin waste gas treatment of the present invention. Detailed Implementation

[0007] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0008] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0009] like Figure 1As shown, this invention provides an intelligent method for judging and switching the saturation state of an activated carbon adsorption bed for polyolefin waste gas treatment, specifically including: S1: Real-time collection of volatile organic compound concentrations at the inlet and outlet of the adsorption bed, bed pressure difference, cumulative operating time, instantaneous air volume, and inlet concentration constitutes a six-dimensional basic time series dataset, which serves as the raw input data for saturation state judgment; S2: Calculate the outlet concentration curvature feature sequence based on the six-dimensional basic time series dataset, extract the curvature value through the second difference of the sliding window, and calibrate the dynamic curvature threshold according to the historical penetration samples to generate the initial judgment basis for the concentration response phase change. S3: Based on the integral of cumulative adsorption load, calculate the slope characteristics of the pressure difference growth rate, model the correlation between the bed pressure difference increment and the cumulative adsorption amount, identify the slope transition point and generate the pressure difference growth phase transition criterion. S4: Based on the instantaneous air volume change rate and the fluctuation amplitude of the intake air concentration, a disturbance intensity factor is constructed, and the fusion weight coefficients of the concentration curvature criterion, the pressure difference growth slope criterion, and the remaining capacity ratio criterion are dynamically adjusted to form an adaptive weight set for the operating condition. S5: Input the initial judgment criteria for phase change based on concentration response, the judgment criteria for phase change based on pressure difference growth, and the feature of remaining capacity ratio into the multi-scale phase change threshold judgment model, and combine the working condition adaptive weight set to perform phase change flag activation judgment to generate a comprehensive phase change confirmation event. S6: When the disturbance intensity factor exceeds the preset threshold, a 15-second short-time hysteresis loop verification mechanism is activated to verify the continuous validity of the phase transition flag within the hysteresis time window and output a verification confirmation signal. S7: Generate a structured saturation state label based on the verification confirmation signal and the phase transition flag activation state, which includes the operation stage identifier, confidence index, trigger reason code and switching time window parameters; S8: Input the structured saturation state label into the switching control unit. When the operating stage in the label is identified as the critical zone of penetration and the confidence index exceeds the safety threshold, trigger the adsorption bed switching command and update the status of the queue to be desorbed.

[0010] Step S1: Real-time acquisition of volatile organic compound concentrations at the inlet and outlet of the adsorption bed, bed pressure difference, cumulative operating time, instantaneous air volume, and inlet air concentration to construct a six-dimensional basic time-series dataset, which serves as the raw input data for saturation state determination. Specifically, this includes: S1.1: Synchronously acquire and process the raw analog signals output by the volatile organic compound concentration sensor, bed pressure differential sensor, instantaneous air volume sensor, cumulative running time counter, and inlet air concentration sensor at the inlet and outlet of the adsorption bed to collect the initial analog signal data of each parameter as the starting input for data acquisition; The raw analog signals output by the volatile organic compound concentration sensors at the inlet and outlet of the adsorption bed, the bed pressure differential sensor, the instantaneous air volume sensor, the cumulative running time counter, and the inlet air concentration sensor are synchronously acquired using a method (parameter settings: sampling frequency). =200Hz, with the time base of each channel unified as the system master clock), to achieve time consistency capture of multi-channel signals; Furthermore, by constructing a multi-channel data stream buffer (parameters: buffer depth N=1024 points, FIFO structure), the original analog signal is buffered in parallel within the acquisition hardware, and a signal data frame that retains the original amplitude and phase is obtained. Furthermore, a phase-locked acquisition algorithm is adopted (parameters: the lock reference signal is the main clock synchronization pulse, and the phase-locked accuracy is ±1 sampling point) to achieve phase synchronization of all sensor analog signals, ensuring that the alignment accuracy of different signals in the time domain reaches the millisecond level; Furthermore, the channel calibration coefficient insertion algorithm is invoked (parameters based on: sensor factory calibration slope and zero-point deviation) to achieve first-order correction of the original analog signal amplitude and generate the initial analog value matrix corresponding to the physical quantity; By using a multi-source signal synchronous acquisition and calibration processing method, the results of the previous step are transformed into time-consistent and amplitude-corrected initial analog signal data, thereby achieving the high-precision initial input required for subsequent signal conditioning; For example, in a polyolefin waste gas treatment system, the output signals of the inlet and outlet VOCs concentration sensors are connected to a 6-channel synchronous acquisition module, with a sampling frequency of 200Hz and an instantaneous air volume sensor channel calibration coefficient of 0.0025 m. 3 The calibration coefficient for the bed pressure differential sensor is 1.1 kPa / V, the cumulative running time counter directly outputs the system master clock pulse, and the calibration coefficient for the inlet air concentration sensor is 0.98 ppm / V. A phase-locked acquisition algorithm is used to ensure that all channel signals are sampled on the rising edge of the system master clock, achieving a phase-locked accuracy of ±1 point. In the first data frame acquired, after calibration coefficient insertion processing, an initial analog value matrix is ​​obtained. For example, the inlet and outlet VOCs concentrations are 45.2 ppm and 1.25 ppm, respectively, the bed pressure differential is 3.52 kPa, and the instantaneous airflow is 12.8 m³ / s. 3 The cumulative runtime is 36,000 seconds. This matrix serves as a high-precision input for subsequent S1.2 signal conditioning operations, maintaining data temporal consistency and physical quantity accuracy in the subsequent processing chain. S1.2: Perform signal conditioning operations based on the initial analog signal data to eliminate electromagnetic interference and random noise in the industrial field, and generate a filtered analog signal sequence as an intermediate output for signal quality improvement; S1.3: The filtered analog signal sequence is input into the analog-to-digital converter for digitization to generate a discrete digital signal data stream, which serves as a key intermediate product for multi-source data conversion; S1.4: Perform timestamp alignment processing on the generated digital signal data stream to achieve precise synchronization of inlet and outlet volatile organic compound concentrations, bed pressure difference, cumulative operating time, instantaneous air volume and intake concentration, forming a time-aligned digital dataset as a prerequisite for data integration; S1.5: Perform dimensional integration processing on the time-aligned digital dataset to construct a six-dimensional basic time-series dataset containing complete time-series information of six-dimensional parameters, which serves as the original input data output for saturation state judgment.

[0011] Step S2: Calculate the outlet concentration curvature feature sequence based on the six-dimensional basic time-series dataset, extract curvature values ​​through second-order difference using a sliding window, and calibrate the dynamic curvature threshold based on historical penetration samples to generate preliminary judgment criteria for concentration response phase transition. Specifically, this includes: S2.1: Based on the historical penetration sample dataset, statistical analysis is performed on the outlet concentration response curves under different air volume conditions. The mean and standard deviation of the curvature threshold for each air volume range are calculated, and a dynamic curvature threshold function that is adaptive to air volume is generated to provide a criterion for determining the phase change of concentration response. Based on the time-series VOCs concentration curves of the outlet recorded in the historical penetration sample data, the statistical analysis method of air volume intervals (parameter setting basis: air volume interval division standard and sample coverage) is adopted to achieve preliminary classification and data grouping of the morphological characteristics of the concentration curves under different air volume conditions. Furthermore, by using a curvature calculation algorithm (parameters: sampling period Δt, difference order 2), the curvature sequence of the concentration response curve within each air volume range is extracted, and a list of curvature values ​​is obtained as the basic data for statistical analysis. Furthermore, by using the mean and standard deviation calculation algorithm (parameter: sample set of curvature values ​​for each air volume interval), the central tendency and dispersion of the curvature distribution of each air volume interval are quantified, and interval curvature threshold reference data are generated. Furthermore, by constructing an adaptive airflow function (parameters: mean and standard deviation of interval curvature), the fitting and generation of a dynamic curvature threshold function for adaptive airflow are achieved, and the dynamic curvature threshold function is obtained. The fitting formula can be expressed as: in, For coefficient adjustment factor, The mean curvature, The standard deviation of curvature; By using the aforementioned airflow adaptive curvature threshold function, the curvature statistics results from the previous step are transformed into a dynamic threshold benchmark that can be called for real-time phase change determination, thereby achieving the fusion of physical interpretability and operating condition adaptability of the concentration response phase change criterion. For example, in the operating data of the polyolefin exhaust gas treatment system, the air volume range is selected as [3200, 3400] m. 3 / h、[3400, 3600] m 3 / h、[3600, 3800] m 3 / h, extract the exit concentration curvature value sequence of historical penetration samples, sampling period The time is 2 seconds. For [3400, 3600] m 3 For the interval / h, the mean curvature is calculated to be 0.042, and the standard deviation is 0.008. The coefficient adjustment factor λ is set to 1.2. Substituting these values ​​into the formula, the dynamic curvature threshold for this interval is obtained as follows: In actual verification, when the real-time curvature value exceeds this threshold and continues for more than 3 sampling cycles, the system will generate a phase change initial judgment signal, which accurately reflects the state of the adsorption bed entering the nonlinear transition region, thereby significantly improving the stability and accuracy of saturation judgment. S2.2: Perform sliding window second-order difference processing on the outlet VOCs concentration time series data in the six-dimensional basic time series dataset to calculate the curvature feature sequence in order to quantify the local curvature of the concentration response curve; Time series data of outlet VOCs concentration in a six-dimensional basic time series dataset The sliding window second-order difference algorithm is adopted (parameters: window length w, sampling interval). This enables the extraction of local second-order variation features from concentration sequences; Furthermore, through curvature calculation methods (parameter: first-order difference sequence) Second-order difference sequences This allows for the measurement of local curvature in the concentration response curve and the acquisition of curvature values. ); Furthermore, the curvature formula is used: in, Represents the first-order difference value. The rate of change over time ensures the physical interpretability of the curvature metric; Furthermore, by statistically analyzing the curvature values ​​within the sliding window, a curvature feature sequence κ(t) is generated, and boundary condition detection is performed on it to eliminate outliers caused by incomplete window calculations. By generating curvature feature sequences, the second-order difference result from the previous step is transformed into a quantitative index that reflects the local curvature of the concentration response curve, thereby enabling sensitive capture of changes in the shape of the adsorption process curve. For example, VOCs concentration data at the outlet of the activated carbon adsorption bed were recorded for 10 minutes at a sampling frequency of 1 Hz, with the sliding window length set to 5 seconds and the sampling interval... Given a time interval of 1 second, the second-order difference derivation yields a first-order difference value with a mean of 0.0025 mg / m². 3 / s, the second-order difference peak value is 0.0008 mg / m 3 / s 2 Substituting this data into the curvature formula, the peak curvature is calculated. =0.045, the curvature sequence fluctuates within the window from 0.012 to 0.045, significantly higher than the maximum curvature of 0.010 in the linear region of the historical penetration curve, indicating a significant bending change in the curve. This curvature feature sequence was effectively identified as the initial judgment signal of the concentration response phase transition in the subsequent criterion model, verifying the early capture capability of this step in abnormal operating conditions for penetration trends; S2.3: Compare the current curvature feature sequence with the threshold of the dynamic curvature threshold function under the corresponding air volume in real time, and generate a curvature exceeding flag to identify potential phase transition points; S2.4: Perform continuous periodic verification on the curvature exceeding the standard flag. When the curvature exceeding the standard flag remains active for three consecutive sampling periods, output a continuous exceeding signal to eliminate instantaneous noise interference. S2.5: Generate a preliminary judgment criterion for the concentration response phase transition based on the continuous exceeding signal. This preliminary judgment criterion includes the phase transition trigger timestamp and confidence index, which can be used by the subsequent multi-scale phase transition threshold judgment model.

[0012] like Figure 2 As shown, step S3 involves calculating the slope characteristics of the pressure difference growth rate based on the cumulative adsorption load integral, modeling the correlation between the bed pressure difference increment and the cumulative adsorption amount, identifying the slope transition point, and generating a pressure difference growth phase transition criterion. Specifically, this includes: S3.1: Instantaneous adsorption load calculation is performed on the instantaneous air volume, inlet concentration, and outlet concentration in the six-dimensional basic time series data. First, the product of instantaneous air volume and the difference between inlet and outlet concentrations is obtained as the instantaneous adsorption load. Then, a time cumulative summation operation is performed on the instantaneous adsorption load to generate a cumulative adsorption load sequence. Next, a sliding window smoothing filter is applied to the cumulative adsorption load sequence to eliminate noise interference. Then, the increment of unit adsorption capacity is calculated based on the smoothed cumulative adsorption load sequence. Finally, the dynamic cumulative adsorption load characteristic sequence is output to quantify the trend of total adsorption capacity change during the operation of the adsorption bed. The instantaneous air volume, inlet concentration, and outlet concentration data in the six-dimensional basic time series dataset are processed using a product operation method (parameter: instantaneous air volume). Intake concentration Export concentration This allows for the calculation of instantaneous adsorption load. The formula for calculating instantaneous adsorption load is: in, For instantaneous adsorption load; Furthermore, the time integration method is used (parameters: lower limit of integration 0, upper limit of integration at current time t, integration step size). ), to achieve cumulative adsorption load sequence The generation of . The integral formula is: in, For integration variables; Furthermore, a sliding window smoothing filter method (parameters: window length N, filter kernel type: mean kernel) is used to suppress noise interference in the cumulative adsorbed load sequence, resulting in a smoothed cumulative adsorbed load sequence. This is to ensure the stability of subsequent slope calculations; Furthermore, through the difference calculation method (parameter: difference interval) ), to achieve an increase in unit adsorption capacity The calculation function. The formula for the unit adsorption increment is: in, This indicates the change in adsorption amount within a sampling period; The incremental adsorption capacity per unit volume is transformed into a dynamic cumulative adsorption load characteristic sequence using a differential sequence normalization method (parameter: normalization benchmark is the current maximum cumulative adsorption load). This enables a quantitative characterization of the trend of total adsorption capacity during the operation of the adsorption bed. Through the above algorithm or processing method, the result of the previous step is transformed into a smooth and quantitative dynamic cumulative adsorption load characteristic sequence, thereby improving the interpretability of the adsorption load change trend in terms of physical mechanism and the robustness under abnormal working conditions. For example, in an industrial site treating polyolefin waste gas, the instantaneous air volume Sensor range set to 0~2000m 3 / h, intake air concentration The detection range is set to 0~500 ppm, and the outlet concentration is... The detection range was set to 0~500 ppm. During a certain operating cycle, an instantaneous airflow of 1500 m³ / h was collected. 3With an inlet concentration of 300 ppm and an outlet concentration of 180 ppm, the instantaneous adsorption load was calculated using a formula. for: =1500×120=180000 (unit: m) 3 (ppm / h).

[0013] Integrating this adsorption load sequence yields the cumulative adsorption load at 120 minutes: = 21,600,000 (unit: m) 3 (ppm).

[0014] A sliding window smoothing filter with a window length of 5 significantly reduces the fluctuation range of the sequence after smoothing, effectively avoiding the influence of instantaneous outliers on trend judgment. Difference calculation. A typical value obtained in a smoothed sequence is 8000 (unit: m). 3 The dynamic cumulative adsorption load characteristic value after normalization is 0.37 (dimensionless), calculated as ppm / sampling cycle. This result effectively characterizes the medium-term trend of adsorption capacity changes in the adsorption bed during operation, providing accurate input features for modeling the correlation between pressure difference growth slope; S3.2: Differential processing is performed on the bed pressure difference time series data in the six-dimensional basic time series data. First, the bed pressure difference value between the current time and the previous time is extracted. Then, the difference between the two is calculated as the pressure difference increment per unit time. Next, median filtering is performed on the pressure difference increment sequence per unit time to suppress abnormal fluctuations. Then, a time-normalized pressure difference change rate is constructed based on the filtered pressure difference increment sequence. Finally, a stable pressure difference increment feature sequence is output to characterize the instantaneous dynamic characteristics of bed resistance change. S3.3: Correlation modeling is performed on the dynamic cumulative adsorption load characteristic sequence and the stable pressure difference increment characteristic sequence. First, the stable pressure difference increment characteristic sequence and the dynamic cumulative adsorption load characteristic sequence are aligned to form a two-dimensional correlation dataset. Then, a local linear fitting operation is performed on the two-dimensional correlation dataset to calculate the change slope. Next, the sliding standard deviation analysis is applied to the change slope sequence to identify local fluctuation characteristics. Then, the slope transition threshold is dynamically updated based on the local fluctuation characteristics. Finally, the pressure difference growth rate slope characteristic sequence is output to establish a nonlinear mapping relationship between the pressure difference increment and the cumulative adsorption amount. S3.4: The slope transition point identification process is performed on the slope feature sequence of differential pressure growth rate. First, candidate points in the slope feature sequence that exceed the dynamically updated slope transition threshold are detected. Then, the continuous validity of the candidate points within the preset time window is verified. Next, the slope transition intensity is quantized for the continuously valid candidate points. Then, based on the quantization result, it is determined whether the slope transition intensity meets the preset transition condition. Finally, the slope transition point identification signal is output to accurately locate the abrupt change moment of differential pressure growth behavior. S3.5: Perform pressure difference growth phase transition criterion generation processing on the slope transition point identification signal. First, combine the slope transition point identification signal with the cumulative running time to verify whether the transition duration exceeds the preset time threshold. Then, fuse the transition duration verification result with the slope transition intensity quantification value. Based on the fusion result, generate a binary phase transition state flag. Next, perform confidence weighted calculation on the binary phase transition state flag. Finally, output a structured pressure difference growth phase transition criterion, including phase transition activation state, confidence index and duration parameters, to provide an interpretable basis for physical phase transition judgment. A joint verification algorithm is performed on the slope transition point identifier signal and the cumulative running time (parameter: transition time threshold). This enables the function of judging the excessive duration of slope transitions; Furthermore, the slope intensity quantification method (parameter: slope change amplitude) is used. Slope increment per unit time This enables numerical evaluation of slope transition strength and yields a sequence of slope transition strength indices. Furthermore, a binary state generation method (parameters: time limit verification result, intensity index value) is adopted to construct binary phase transition state markers and generate a structured state marker matrix containing time and intensity dimensions. Furthermore, a confidence-weighted algorithm (parameter: historical slope fluctuation statistics) is used. Real-time slope stability index This enables the weighted fusion calculation of binary phase transition state markers and outputs the confidence index of the pressure difference growth phase transition criterion. By constructing a phase transition criterion, the slope transition identification result of the previous step is transformed into a structured differential pressure growth phase transition criterion that includes phase transition activation state, confidence index and duration parameters, so as to realize a physically interpretable basis for judging saturation state. For example, in the operation scenario of the adsorption bed for polyolefin waste gas treatment, the cumulative operating time collected in real time is 540 minutes. The slope transition point marker signal is continuously activated within the continuous time window. When using the joint verification algorithm, a preset transition time threshold is used. for seconds, the verification result is the duration. The threshold is exceeded within seconds to determine validity. In the slope transition intensity quantification method, the measured slope change amplitude is... for kPa / (m 3 ·h), the slope increment per unit time for kPa / s, the intensity index value after normalization is In the state flag matrix output by the binary state generation method, the time dimension represents activation, and the intensity dimension represents activation. The confidence weighting calculation formula is as follows: ,in For real-time slope stability index , Standard deviation of historical slope fluctuation The confidence index was calculated. for The final output structured pressure difference growth phase transition criterion includes the phase transition activation state "Active", confidence index = 0.78, and duration parameter = 145 seconds. The model achieves high stability and interpretability in saturation state physical determination even under abnormal operating conditions in this scenario.

[0015] like Figure 3 As shown, step S4 involves: constructing a disturbance intensity factor based on the instantaneous airflow change rate and the intake air concentration fluctuation amplitude; dynamically adjusting the fusion weight coefficients of the concentration curvature criterion, the pressure difference growth slope criterion, and the remaining capacity ratio criterion to form an adaptive weight set for the operating condition. Specifically, this includes: S4.1: First, perform sliding window differentiation on the raw time-series data of instantaneous air volume and intake air concentration to calculate the instantaneous air volume change rate and intake air concentration change rate. Then, take the absolute values ​​of the calculated instantaneous air volume change rate and intake air concentration change rate to generate a sequence of absolute values ​​of instantaneous air volume change rate and a sequence of absolute values ​​of intake air concentration fluctuation amplitude. Next, based on the average air volume value and average intake air concentration value of historical operating data, perform normalization processing on the sequence of absolute values ​​of instantaneous air volume change rate. Then, perform normalization processing on the sequence of absolute values ​​of intake air concentration fluctuation amplitude. Finally, perform weighted summation of the normalized instantaneous air volume change rate and the normalized intake air concentration fluctuation amplitude according to a preset weight coefficient of 0.5 to generate a disturbance intensity factor quantification value, which serves as the input basis for abnormal operating condition identification. For the instantaneous air volume raw time series data and the intake air concentration raw time series data in the six-dimensional basic time series data, a sliding window differential algorithm is used (parameters: window length w=5 seconds, sampling period). =1 second), to perform first-order difference operation on time series and output instantaneous rate of change feature sequence; Furthermore, by using the absolute value operation method (parameter: feature sequence input), the sign of the rate of change is removed, and the absolute value sequence of the instantaneous air volume change rate and the absolute value sequence of the intake air concentration fluctuation amplitude are obtained; Furthermore, a normalization algorithm is employed (parameters: the normalization reference is the average air volume value of historical operating data). With average intake air concentration value This process achieves dimensionless transformation of the absolute value sequence and generates a normalized instantaneous airflow change rate sequence and a normalized intake air concentration fluctuation amplitude sequence. The formula for calculating the normalized instantaneous airflow change rate is as follows: in, Instantaneous air volume Historical average air volume This represents the instantaneous rate of change of air volume. Furthermore, a weighted summation algorithm is adopted (parameter: preset weight coefficients). =0.5), achieving linear fusion of the normalized instantaneous airflow change rate and the normalized intake air concentration fluctuation amplitude, and generating a quantitative value for the disturbance intensity factor. Disturbance Intensity Factor The calculation formula is as follows: in, Intake concentration, This represents the historical average intake concentration. The above linear fusion algorithm transforms the normalization result of the previous step into a quantized value of the disturbance intensity factor, thereby constructing the input basis for identifying abnormal working conditions. For example, in a polyolefin exhaust gas treatment system, the instantaneous air volume sensor has a sampling period of 1 second and a historical average air volume. =500 m 3 / h, the intake air concentration sensor sampling period is 1 second, historical average intake air concentration =150 ppm. In the sliding window differential calculation at a certain operating time point, the instantaneous air volume change rate is 20 m³ / s. 3 The instantaneous airflow rate change was 7 ppm / s, with an airflow rate of 7 h / s. After absolute value processing, the absolute value of the instantaneous airflow rate change was 20, and the absolute value of the airflow concentration fluctuation was 7. Normalization yielded a normalized instantaneous airflow rate change of 0.04 and a normalized airflow concentration fluctuation value of 0.0467. The two values ​​were then weighted and summed with a weight of 0.5 to obtain the quantified value of the disturbance intensity factor. =0.04335. When this value is used as the input to S4.2, the system will subsequently compare it with the preset disturbance threshold of 0.3. When the value is below a threshold, the operating status is determined to be normal; when it is above the threshold, the abnormal operating condition identification process is triggered. In the above embodiment, the disturbance intensity factor value is significantly lower than the threshold, the system determines that the current operating condition is stable, maintains the existing criterion weight allocation, and ensures the accuracy and stability of the state judgment model under normal operating conditions. S4.2: First, obtain the quantized value of the disturbance intensity factor; then compare the quantized value of the disturbance intensity factor with the preset disturbance threshold of 0.3; next, determine whether the quantized value of the disturbance intensity factor exceeds the preset disturbance threshold; if it exceeds, generate an abnormal operating condition status indicator activation signal; finally, output the abnormal operating condition status indicator to clearly distinguish between normal operating conditions and abnormal disturbance operating conditions such as sudden airflow changes or concentration pulses. The digital data input for obtaining the quantized value of the disturbance intensity factor is used, employing a fixed-point comparison algorithm (parameters: quantized value of the disturbance intensity factor, preset disturbance threshold). This enables numerical comparison functionality; Furthermore, an interval threshold detection method is used (parameter: lower limit is...). The upper limit is This allows for the validation of the range of quantized values ​​for the disturbance intensity factor, and the acquisition of judgment flag data. Furthermore, a conditional discrimination algorithm (parameters: judgment flag data, disturbance intensity factor quantization value) is adopted to realize the logical judgment of whether the disturbance intensity factor exceeds the preset disturbance threshold and generate a Boolean result of the working condition status; Furthermore, by using a state mapping table lookup method (parameters: Boolean result of working condition state, encoding matrix of abnormal working condition state), the activation signal of abnormal working condition state is generated, and the signal trigger timestamp data is obtained. The abnormal status signal output module converts the Boolean result of the previous step and the trigger timestamp into a structured abnormal status identifier, thereby achieving the expected technical effect of clearly distinguishing between normal operation status and abnormal disturbance status such as airflow change or concentration pulse. For example, under the on-site operating conditions of a polyolefin waste gas treatment system, the normalized value of the absolute value of the instantaneous air volume change rate is: The normalized value of the absolute value of the intake air concentration fluctuation is: According to weighting coefficient The perturbation intensity factor is obtained by weighted summation. Applying the fixed-point comparison algorithm will... With preset disturbance threshold The comparison showed the result exceeded the limit; the interval threshold detection confirmed that it was within the range. to Within the valid range, the conditional discrimination algorithm generates a Boolean result of true for the abnormal operating condition; the state mapping table is used to look up the abnormal operating condition state activation signal and record the trigger timestamp. Seconds. The final abnormal operating condition status identifier contains the fields {Status: "Abnormal Disturbance", Trigger Reason: "Disturbance Intensity Exceeds Threshold", Timestamp:} Under this condition, the model will enter a weight adjustment process to improve the robustness and stability of subsequent state judgments. S4.3: First, obtain the abnormal operating condition status identifier; then determine whether the abnormal operating condition status identifier is activated; if activated, adjust the concentration curvature criterion weight coefficient from the initial value to 0.4; at the same time, adjust the pressure difference growth slope criterion weight coefficient to 0.35; and adjust the remaining capacity ratio criterion weight coefficient to 0.25; finally, generate a dynamically adjusted fusion weight coefficient set to improve the robustness of the model to abnormal disturbances. Based on the dynamic weight adjustment processing of abnormal working condition status indicators, a weight remapping algorithm (parameters: initial weight coefficient set, abnormal working condition status indicator) is adopted to realize the rapid correction of the weight coefficients of the three types of criteria. Furthermore, by using a conditional judgment algorithm (parameters: abnormal operating condition status identifier, judgment condition set), the activation status of abnormal operating conditions is detected, and the activation status signal is obtained. Furthermore, a substitution assignment algorithm (parameters: initial weight of concentration curvature, initial weight of pressure difference growth slope, initial weight of remaining capacity ratio) is used to correct the concentration curvature weight coefficient under abnormal operating conditions. The correction of the pressure difference growth slope weighting coefficient is as follows: The remaining capacity is adjusted to the weighting coefficient. And generate an adjusted set of weight coefficients; Furthermore, a weight set generation algorithm (parameter: adjusted weight coefficient set) is used to dynamically combine the weight coefficients of the three types of criteria to form a fused weight coefficient set; By using a dynamic remapping process triggered by abnormal operating conditions, the abnormal state identification results of the previous step are transformed into a set of fusion weight coefficients, thereby achieving the expected technical effect of model stabilization criterion fusion under abnormal disturbances. For example, in a polyolefin exhaust gas treatment system, the real-time acquired quantized value of the disturbance intensity factor is... Exceeding the preset disturbance threshold An abnormal operating condition activation signal is generated by a conditional decision-making algorithm. The original fusion weight set of the system consists of concentration curvature criterion weight coefficients. Weighting coefficient of pressure difference growth slope criterion Remaining capacity ratio criterion weighting coefficient Based on the substitution assignment algorithm, a dynamically adjusted set of weight coefficients is generated: concentration curvature criterion weight coefficients. Weighting coefficient of pressure difference growth slope criterion Remaining capacity ratio criterion weighting coefficient The weight set generation algorithm stores the adjustment result into the fused weight coefficient set for use by the multi-scale phase transition threshold criterion model. During operation, this adjustment enables the model to handle sudden changes in wind volume (rate of change reaching...). ), intake air concentration pulse (fluctuation amplitude up to Even under these conditions, it can still maintain the stability and continuity of the criterion fusion output, significantly improving the robustness of the comprehensive judgment of saturated state; S4.4: First, obtain the dynamically adjusted concentration curvature criterion weight coefficient, pressure difference growth slope criterion weight coefficient, and remaining capacity ratio criterion weight coefficient; then calculate the sum of the three weight coefficients; next, divide each weight coefficient by the sum to perform normalization processing; ensure that the sum of the normalized weight coefficients equals one; finally, form an adaptive weight set for operating conditions, providing decision fusion parameters that can be directly called for the multi-scale phase transition threshold criterion model; The dynamically adjusted concentration curvature criterion weight coefficient, pressure difference growth slope criterion weight coefficient, and remaining capacity ratio criterion weight coefficient are used as the input parameter set. The weight summation calculation method (parameter: the values ​​of the three types of weight coefficients) is adopted to realize the normalization precondition of the fusion weight set. Furthermore, by using a weight normalization method (parameters: total value, individual weight value), the proportion of each fusion weight is corrected, and a set of normalized coefficients is obtained; Furthermore, a numerical consistency verification method (parameter: normalized coefficient set) is adopted to verify that the sum of the normalized weight coefficients equals one, and a consistency verification signal is generated. Furthermore, based on the consistency verification signal and the normalized coefficient set, an adaptive weight set for the working condition is generated through a weight set construction method (parameter: three types of normalized coefficients); By normalizing the results of the dynamic adjustment of the fusion weight coefficients in the previous step, the decision fusion parameters can be directly called, thereby improving the stability and robustness of the multi-scale phase transition threshold criterion model under different working conditions. For example, in a polyolefin waste gas treatment system under abnormal operating conditions, the weighting coefficient for the concentration curvature criterion is set to 0.4, the weighting coefficient for the pressure difference growth slope criterion is set to 0.35, and the weighting coefficient for the remaining capacity ratio criterion is set to 0.25. The weighted summation calculation method is used, and the formula is as follows: in, This is the sum of the weighted coefficients. The total is calculated. Using the weight normalization method, the formula is as follows: , , .

[0016] The normalized coefficient set is The consistency check signal confirms that its sum equals After the adaptive weight set for this operating condition is input into the multi-scale phase change criterion model, the model still maintains high diagnostic transparency and judgment stability under the conditions of both sudden airflow changes and concentration fluctuations. The triggering delay of the comprehensive phase change confirmation event is significantly reduced, the system switching command is executed stably, and the risk of pollutant penetration is effectively controlled.

[0017] like Figure 4 As shown, step S5 involves inputting the initial phase change judgment criteria based on concentration response, the phase change judgment criteria based on pressure difference growth, and the remaining capacity ratio characteristics into a multi-scale phase change threshold judgment model. This is combined with an adaptive weight set based on operating conditions to perform a phase change flag activation judgment, generating a comprehensive phase change confirmation event. Specifically, this includes: S5.1: First, obtain the historical penetration sample dataset as the basis for model calibration; second, calculate the dynamic threshold of outlet concentration curvature based on the dataset to generate concentration response phase change threshold parameters; then, perform statistical analysis on the transition characteristics of bed pressure difference growth rate to determine the pressure difference growth phase change threshold parameters; next, calibrate the remaining capacity ratio threshold parameters based on the cumulative adsorption load integral slope shift characteristics; finally, integrate the three types of threshold parameters to construct a multi-scale phase change threshold criterion model, and output the model parameter set for phase change judgment benchmark. The model calibration basis is constructed based on the historical penetration sample dataset. The statistical feature extraction method (parameters: sample size, sampling time interval, air volume distribution) is used to quantify the original features of the outlet concentration curvature, bed pressure difference growth rate and cumulative adsorption load slope shift under different operating conditions. Furthermore, by using a piecewise regression analysis method (parameters: air volume interval division, curvature threshold calculation window length), the dynamic threshold of outlet concentration curvature is calculated, and the curvature threshold parameter set of the phase change trigger sensitive section is obtained. Furthermore, by using the sliding standard deviation analysis method (parameters: slope analysis window length, standard deviation calculation step size), the statistical analysis of the transition characteristics of the bed pressure difference growth rate is realized, the pressure difference growth phase transition threshold parameter is determined, and the local abrupt change law between pressure difference change and adsorption load is extracted. Furthermore, the capacity ratio threshold of the cumulative adsorption load integral slope shift characteristic is calibrated by using the Langmuir-Freundlich isothermal adsorption model inversion method (parameters: temperature, pressure, pollutant concentration), and the remaining capacity ratio judgment threshold parameter is output. By integrating the outlet concentration curvature threshold parameter, the pressure difference growth phase change threshold parameter, and the remaining capacity ratio threshold parameter, a multi-scale threshold fusion model construction method (parameters: threshold fusion weight, fusion strategy type) is adopted to construct a multi-scale phase change threshold criterion model and output the model parameter set for subsequent phase change judgment benchmark. By constructing a multi-scale model parameter set, the statistical analysis results of the previous stage are transformed into physically interpretable criterion data, thereby realizing a multi-dimensional physical phase change judgment benchmark for the operating status of the activated carbon adsorption bed. For example, in the operation of a polyolefin exhaust gas treatment system, the historical penetration sample dataset consists of 1000 records, with a sampling time interval of 5 seconds. The airflow distribution range is divided into low airflow (0.8-1.2 m³ / s). 3 / min), medium air volume (1.2-1.8m) 3 / min), high air volume (1.8-2.2 m³ / min), 3 / min). A piecewise regression method was used to calculate the threshold for the outlet concentration curvature, with a window length of 30 seconds. The curvature threshold obtained in the medium air volume range was: The sliding standard deviation analysis window was set to 60 seconds, and the step size was set to 10 seconds. The phase transition threshold for pressure differential growth was obtained as follows. kPa / (mol). Under the conditions of 25℃ temperature and 1 atm pressure, the Langmuir-Freundlich model inversion yielded a residual capacity ratio threshold range of 0.15±0.03. A multi-scale model was constructed by weighting and fusing the three types of threshold parameters (curvature threshold weight 0.4, pressure difference threshold weight 0.35, and capacity ratio threshold weight 0.25). The output parameter set significantly improves the accuracy and interpretability of phase transition determination in online system monitoring and maintains stable output under various operating conditions. S5.2: First, it receives the initial judgment criteria for phase change based on concentration response, the judgment criteria for phase change based on pressure difference growth, and the remaining capacity ratio as input objects; second, it performs weighted fusion processing on the three types of features based on the dynamic weight coefficients in the adaptive weight set of operating conditions; then, it calculates the weighted comprehensive phase change index value; next, it compares the index value with the preset threshold in the multi-scale phase change threshold judgment model; finally, it generates a phase change flag activation status identifier. Using the initial judgment criteria for phase transition in concentration response, the judgment criteria for phase transition in pressure difference growth, and the characteristics of remaining capacity ratio as standardized input objects, a feature vector construction method (parameters: three feature values ​​and a set of dynamic weight coefficients) is adopted to achieve a unified representation of multi-source physical indicators. Furthermore, a weighted fusion algorithm (parameters: three types of weight coefficients in the adaptive weight set) is used to sum the weighted products of each feature value and obtain the weighted comprehensive phase transition index value. The calculation formula is: in, , , These are the dynamic weighting coefficients for concentration curvature, pressure gradient, and residual capacity ratio, respectively. , , These are the corresponding eigenvalues; Furthermore, a normalization algorithm (parameter: normalization range [0,1]) is used to achieve interval mapping of the comprehensive phase transition index, thereby eliminating differences in different feature dimensions and generating a normalized comprehensive index. ; Furthermore, a threshold comparison method is employed (parameter: a preset threshold in the multi-scale phase transition threshold criterion model) to achieve... With threshold Real-time comparison, and generate a Boolean value for the comparison result. ; By using a discrimination algorithm (parameter: Boolean value R for comparison result), the comparison result of the previous step is transformed into a phase transition flag activation state identifier, thereby clarifying the triggering conditions of the phase transition criterion; For example, in a polyolefin waste gas treatment system, the initial judgment criterion for concentration response phase change... The value is 0.78, which is the criterion for phase transition due to pressure difference growth. The remaining capacity ratio is 0.65. The value is 0.22; adaptive weight set configuration for operating conditions. =0.4, =0.35, =0.25. Apply the weighted comprehensive phase transition index formula to calculate: The result is I = 0.564. After normalization... =0.564 (because all features have been normalized). The multi-scale phase transition threshold criterion model presets a threshold T=0.55. When performing threshold comparison, the condition I'=0.564>T is met, R is true, and the phase transition flag activation state is finally generated as "1". In this embodiment, through the execution of the above algorithm chain, the system effectively integrates multi-source physical features, achieves high-precision phase transition trigger judgment, and maintains robustness under airflow disturbance conditions.

[0018] S5.3: First, obtain the phase transition flag activation status identifier as the judgment input; second, verify whether the identifier meets the activation condition of the multi-scale phase transition threshold criterion model; then, combine the confidence adjustment mechanism of the working condition adaptive weight set; next, confirm the validity of the phase transition event; finally, generate a comprehensive phase transition confirmation event structure containing the running stage identifier, confidence index and trigger reason code. Obtain the phase transition flag activation status identifier output by S5.2, and use the conditional verification algorithm (parameter: multi-scale threshold parameter set) to realize threshold compliance verification based on the criterion model; Furthermore, through the logical matching method (parameters: phase transition flag activation state identifier, activation condition set of multi-scale phase transition criterion model), the one-to-one correspondence between the judgment identifier and the model condition is realized, and the condition matching result data is obtained; Furthermore, the confidence level of the matching results data is corrected by using a weighted confidence adjustment method (parameter: working condition adaptive weight set), and a corrected confidence vector is generated. Furthermore, through an event validity confirmation algorithm (parameters: corrected confidence vector, condition matching result data), the joint determination of the authenticity of phase transition events is realized, and an event validity flag is generated; By using a structured event coding method (parameters: event validity flag, operational phase classification result, confidence vector), the result of the previous step is transformed into a comprehensive phase transition confirmation event structure containing operational phase identifier, confidence index, and trigger cause coding, thereby achieving physically interpretable saturation state determination output; For example, in a polyolefin exhaust gas treatment system, the outlet concentration curvature threshold is set as an adaptive function of air volume. When the air volume Q = 1.2 m³ / s... 3 The curvature threshold is 0.08 per minute, and the pressure differential growth slope threshold is 0.015 Pa·kg. -1 The remaining capacity ratio threshold is 0.14. The phase transition flag output by S5.2 is in an active state with an initial confidence level of 0.89. The conditional verification algorithm determines that the flag meets the phase transition judgment conditions using a multi-scale threshold parameter set; the logical matching result shows that all conditions are successfully matched; the adaptive weight set for the current disturbance intensity... Maintaining the initial weights at 0.25, the confidence level calculated by the weighted confidence adjustment method is 0.91. The event validity confirmation algorithm generates a validity flag as true after the verification duration meets the set threshold. The structured event coding method sets the running phase identifier to "transition", and the triggering reason coding includes "curvature_spike" and "slope_jump". The final output is a comprehensive phase transition confirmation event structure: the phase identifier is transition, the confidence index is 0.91, the triggering reason coding list has two items, and the system switching time window is predicted to be [87min, 102min]. This embodiment ensures both judgment accuracy and interpretability under low abnormal disturbance conditions, and achieves real-time and accurate identification of saturation state.

[0019] Step S6: When the disturbance intensity factor exceeds a preset threshold, a 15-second short-time hysteresis loop verification mechanism is activated to verify the continued validity of the phase transition flag within the hysteresis time window, and a verification confirmation signal is output. Specifically, this includes: S6.1: Perform threshold comparison processing on the real-time acquired disturbance intensity factor to determine whether it exceeds the preset threshold, and generate a disturbance intensity over-limit signal as the trigger condition for the short-time hysteresis loop verification mechanism. For the real-time acquired disturbance intensity factor quantification value, a numerical threshold comparison method (parameter: preset disturbance threshold = 0.3) is used to determine the triggering conditions of sudden airflow change or concentration pulse abnormality. Furthermore, by using the absolute value normalization method (parameters: absolute value of air volume change rate, absolute value of intake air concentration change rate), the input disturbance intensity factor is standardized and characterized, and normalized disturbance intensity factor data is obtained. Furthermore, a difference comparison algorithm (parameters: normalized disturbance intensity factor, preset disturbance threshold) is used to determine whether the disturbance intensity exceeds the preset threshold and generate Boolean comparison result data; Furthermore, by using a state identifier generation method (parameter: Boolean comparison result data), the encoding of the disturbance intensity exceeding the limit state is realized, and the disturbance intensity exceeding the limit signal is generated; By using the disturbance intensity anomaly detection and processing method, the comparison result of the previous step is converted into a disturbance intensity over-limit signal that can be called by the short-time hysteresis verification mechanism, thereby realizing the precise triggering of the hysteresis mechanism; For example, at the polyolefin waste gas treatment site, the absolute value of the instantaneous air volume change rate was 0.12, the absolute value of the inlet air concentration change rate was 0.25, and the average air volume was 3200 m³ / s. 3 The average intake air concentration is 150 ppm per hour. After normalizing the absolute values ​​of the above data, the normalized airflow change rate is 0.15 and the normalized concentration change rate is 0.27. These are then linearly weighted and summed with a weight of 0.5 to calculate the disturbance intensity factor. as follows: in, These are the weighting coefficients. and These represent the normalized rate of change in air volume and the rate of change in intake air concentration, respectively. The calculation results are as follows: When this value is compared with a preset perturbation threshold, the formula is as follows: The comparison result is logically false, not exceeding the threshold, and the disturbance intensity exceeding the limit signal remains inactive. Under another operating condition, the normalized air volume change rate is 0.35, and the normalized concentration change rate is 0.45. Calculate the η value with equal weights: The calculation result is Compare with the threshold: If the logic is true, the disturbance intensity exceeding the limit signal is activated, triggering a 15-second short-time hysteresis loop verification mechanism. This embodiment shows that this step can significantly improve the sensitivity and accuracy of disturbance detection under abnormal operating conditions, providing a reliable triggering condition for the robustness verification of saturation state judgment under subsequent abnormal operating conditions. S6.2: Based on the disturbance intensity exceeding the limit signal, perform the initialization operation of the short time delay loop verification mechanism, set the start and end time points of the 15-second delay time window to obtain the verification time range; S6.3: Perform time continuity analysis on the phase transition flag sequence within the initialized lag time window to identify the continuous period of phase transition flag activation; calculate the duration of the activation period; compare the calculated duration with the preset minimum duration threshold; generate a preliminary verification state based on the comparison result; integrate the preliminary verification states to form a continuous validity verification result, and output the verification result. The phase transition flag sequence is received as the analysis input within a 15-second lag time window after initialization. A continuity analysis algorithm based on time series scanning (parameters: sampling period Δt, sequence length N) is used to extract the activation state index in the phase transition marker sequence; Furthermore, by using the state segmentation method (parameter: activation flag value is 1, inactivation flag value is 0), the boundary positioning of the continuous period of phase change flag activation is realized, and the start time index and end time index of each activation period are obtained; Furthermore, the continuous activation period is quantified through a duration calculation formula, as follows: in, For duration, To activate the end time, To activate the start time; Furthermore, a duration threshold comparison algorithm is used (parameter: threshold). This allows for the comparison of the activation period duration with a preset minimum duration threshold, and the generation of a binary preliminary verification status to determine whether the continuous validity is satisfied. Furthermore, by using a state integration algorithm (parameter: binary state sequence), the fusion of multiple activation verification states is achieved, generating a single continuous validity verification result; By combining continuity analysis with duration calculation, the continuity stability of the phase transition marker sequence after the hysteresis window is transformed into a continuous validity verification result, thereby achieving a quantitative assessment of the robustness of phase transition events under anomalous perturbations. For example, in the operation of a polyolefin waste gas treatment system, the hysteresis window length is set to 15 seconds, the sampling period Δt is 1 second, and the phase transition flag sequence within the window is [1,1,1,0,1,1,1,1,0,0,1,1,1,1,1]. A continuity analysis algorithm is used to scan the sequence, extracting continuous segments with a flag value of 1, locating three activation stages: the 1-3 second segment, the 5-8 second segment, and the 11-15 second segment. The duration of the first segment is calculated using a formula. = seconds, duration of the second segment = seconds, duration of the third segment = Seconds. The duration is compared to a preset minimum duration threshold. The comparison of seconds shows that the first and third segments meet the conditions, while the second segment also meets the conditions but has a shorter duration. After integrating the verification states, the output of the continuous validity verification result is "activated and valid". This result indicates that the phase transition flag remains stably activated within the hysteresis window, which is sufficient to support the subsequent switching decision logic. S6.4: Based on the continuous validity verification results, generate a verification confirmation signal to indicate whether the phase transition flag remains valid within the hysteresis time window, and output the final verification status for use by the upper-level module.

[0020] Step S7: Generate a structured saturation state label based on the verification confirmation signal and the phase transition flag activation state, including the operating stage identifier, confidence index, trigger reason code, and switching time window parameters. Specifically, this includes: S7.1: Perform data acquisition and processing on the verification confirmation signal and the phase change flag activation status to extract the decision input source of the current adsorption bed operating status; S7.2: Execute the operation stage classification algorithm based on the phase change flag activation state to determine the physical stage identifier of the adsorption bed and output the operation stage identifier; S7.3: Perform confidence fusion calculation based on the operation stage identifier, phase transition flag activation status and operating condition adaptive weight set to generate a confidence index that quantifies the reliability of the status judgment; Based on the operation phase identifier, phase transition flag activation state, and operating condition adaptive weight set as input objects for fusion calculation, a confidence-weighted fusion algorithm (parameters: operation phase identifier, phase transition flag activation state, weight coefficient set) is adopted to achieve unified quantitative processing of the three types of state features. Furthermore, by using the runtime phase identifier mapping coefficient library (parameter: the credibility benchmark value corresponding to each physical phase), the runtime phase identifier is quantified, and the phase credibility score matrix is ​​obtained. Furthermore, the stability of the phase transition state is quantified by using the phase transition flag activation state convolution integral method (parameters: flag continuity duration, flag activation frequency), and a phase transition stability score sequence is generated. Furthermore, by using the working condition adaptive weight set normalization processing algorithm (parameters: various weight coefficients), the constraint that the sum of the weight coefficients is 1 is achieved, and a normalized weight vector is generated to ensure the accuracy of the fusion calculation ratio; Furthermore, a weighted summation fusion formula is used to calculate the confidence levels of the three types of features. calculate: in, These are the weighting coefficients for the operational phase. This represents the stage credibility score. The phase transition indicator weighting coefficient, The phase transition stability score, The remaining capacity is the weighting factor. The remaining capacity is compared to the credibility score; By using a confidence-weighted fusion algorithm, the feature score results from the previous step are transformed into a single confidence index, thereby achieving a quantitative reliability evaluation of the saturation state judgment results. For example, in a polyolefin waste gas treatment system, if the operating phase is marked as "penetrating the critical zone," the phase change flag activation duration is 18 seconds, and the flag activation frequency is 0.85, the baseline value for "penetrating the critical zone" in the operating phase mapping coefficient library is set to 0.92, the phase change stability score calculated by convolution integral is 0.88, and the remaining capacity ratio confidence score is 0.80 under the current cumulative adsorption load. The normalized weight vectors of the adaptive weight set for the operating condition are as follows: stage weight coefficients. =0.40, phase transition indicator weighting coefficient =0.35, remaining capacity ratio weighting factor =0.25. Substituting into the weighted summation formula: The calculated fusion confidence index was 0.883. This result indicates that the saturation state judgment under the current operating conditions has significantly improved reliability, and can provide accurate quantitative decision-making basis for the switching control module; S7.4: Perform trigger cause coding mapping based on the phase transition flag activation state to generate a trigger cause coding sequence that records key phase transition events; S7.5: Execute a switching time window prediction algorithm using the operation phase identifier and confidence index to generate safe time window parameters for adsorption bed switching.

[0021] Step S8: Input the structured saturation state tag into the switching control unit. When the operating stage in the tag is identified as the penetration critical zone and the confidence index exceeds the safety threshold, trigger the adsorption bed switching command and update the status of the desorption queue. Specifically, this includes: S8.1: Perform data parsing processing on the structured saturation state labels to extract the operation stage identifier and confidence index as input conditions for stage determination and confidence verification, generate a parsed stage parameter set, and ensure that subsequent judgments are based on standardized label data; S8.2: Based on the analytical stage parameter set generated in S8.1, perform string matching operation between the running stage identifier and the breakthrough critical zone identifier to generate a stage matching signal, which is used to confirm whether the adsorption bed has entered a high-risk breakthrough state. S8.3: Perform floating-point comparison processing on the confidence index extracted in S8.1 and the preset safety threshold to generate a confidence level compliance signal, so as to quantify the reliability level of the state judgment result; S8.4: Input the stage matching signal generated in S8.2 and the confidence level compliance signal generated in S8.3 into the logic AND gate circuit, perform Boolean logic AND operation processing, generate the switching trigger signal, and comprehensively determine whether the adsorption bed switching conditions are fully met; The stage matching signal generated in S8.2 and the confidence level achievement signal generated in S8.3 are used as Boolean logic input objects. A dual-input logic AND operation unit (parameter: input channel 1 is the stage matching signal, input channel 2 is the confidence level achievement signal) is used to realize the synchronization effect determination of the two conditions. Furthermore, by using a logic state mapping method (parameter: high level indicates that the condition is met, low level indicates that the condition is not met), the input signal matrix is ​​converted into a binary decision vector so as to perform logical AND operation processing and obtain a Boolean synthesized signal; Furthermore, the standard Boolean AND operation function (parameter: The logic core performs a bit-by-bit synchronous logical product operation between the stage matching signal and the confidence level achievement signal, and generates a logical AND output value. Where A is the stage matching signal and B is the confidence level achievement signal; Furthermore, the result mapping mechanism (parameter: when the output value is high, the switching trigger state is set to active) is used to convert the logic and output into a switching trigger signal, ensuring that the switching execution preparation state is only entered when both the stage matching and confidence level are met. By using logical AND operation processing, the stage matching and confidence level achievement results of the previous step are transformed into a unique switching trigger signal, thereby achieving the judgment effect that the switching conditions of the adsorption bed are fully met. For example, in a polyolefin waste gas treatment operation scenario, the stage matching signal in the analytical stage parameter set is set to a high level (value 1), indicating that the current operation stage is in the penetration critical zone; the confidence index is 0.92, the preset safety threshold is 0.85, and the confidence compliance signal generated after floating-point comparison processing is a high level (value 1). The two high-level signals are input to the logic AND operation unit, and a Boolean AND function kernel is used to calculate... The system outputs a value of 1. This output value is converted into a switching trigger signal through a result mapping mechanism, and the switching trigger state is set to active. In this scenario, after the switching trigger signal is generated, the system confirms that the adsorption bed switching conditions are fully met and transmits the signal to the switching control unit to execute the adsorption bed switching command. This ensures that the switching action can be reliably started under high-risk penetration conditions, eliminating the risk of pollutant emissions exceeding standards, while maintaining the optimal level of adsorption capacity utilization. S8.5: Based on the switching trigger signal generated in S8.4, execute the generation and sending of adsorption bed switching instructions, and synchronously update the status of the queue to be desorbed, so as to realize seamless scheduling of adsorption bed resources and closed-loop guarantee of system operation safety.

[0022] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0023] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent saturation state judgment and switching method for activated carbon adsorption beds oriented to polyolefin off-gas treatment, characterized in that, The method comprises the following steps: S1: Real-time collection of volatile organic compound concentrations at the inlet and outlet of the adsorption bed, bed pressure difference, cumulative running time, instantaneous air flow rate, and inlet gas concentration to form a basic time series dataset; S2: Calculation of an outlet concentration curvature feature sequence based on the basic time series dataset, extraction of curvature values by second-order difference of a sliding window, and calibration of a dynamic curvature threshold value according to historical breakthrough samples to generate a concentration response phase change preliminary judgment basis; S3: Calculation of a pressure difference growth rate slope feature based on cumulative adsorption load integration, modeling of the bed pressure difference increment in relation to cumulative adsorption, identification of a slope transition point, and generation of a pressure difference growth phase change criterion; S4: Construction of a disturbance intensity factor according to the instantaneous air flow rate change rate and the inlet gas concentration fluctuation amplitude, dynamic adjustment of the fusion weight coefficients of the concentration curvature criterion, the pressure difference growth slope criterion, and the residual capacity ratio criterion to form a working condition adaptive weight set; S5: Input of the concentration response phase change preliminary judgment basis, the pressure difference growth phase change criterion, and the residual capacity ratio feature into a multi-scale phase change threshold criterion model, combined with the working condition adaptive weight set, to perform a phase change flag activation judgment and generate a comprehensive phase change confirmation event; S6: When the disturbance intensity factor exceeds a preset threshold, a short time lag loop verification mechanism is started to verify the sustained effectiveness of the phase change flag within a lag time window, and a verification confirmation signal is output; S7: Generation of a structured saturation state label based on the verification confirmation signal and the phase change flag activation state.

2. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, The step S7 further comprises: S8: Input of the structured saturation state label into a switching control unit, triggering of an adsorption bed switching instruction when the running phase identification in the label is a breakthrough critical region and the confidence index exceeds a safety threshold, and updating of a desorption queue state.

3. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, The real-time collection of volatile organic compound concentrations at the inlet and outlet of the adsorption bed, bed pressure difference, cumulative running time, instantaneous air flow rate, and inlet gas concentration is specifically as follows: Synchronous collection of raw analog signals output by adsorption bed inlet and outlet volatile organic compound concentration sensors, bed pressure difference sensors, instantaneous air flow rate sensors, cumulative running time counters, and inlet gas concentration sensors, wherein the synchronous collection is performed through buffer zone construction and FIFO structure for parallel buffering and phase-locked collection of the raw analog signals, the phase-locked accuracy is ±1 sampling point, a channel calibration coefficient insertion algorithm is called to perform amplitude correction, and an initial analog value matrix corresponding to physical quantities is generated.

4. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, The step S3 specifically comprises: Instantaneous adsorption load calculation and processing of the instantaneous air flow rate, inlet gas concentration, and outlet concentration in the basic time series data, acquisition of the product value of the instantaneous air flow rate and the difference between the inlet gas concentration and the outlet concentration as the instantaneous adsorption load, time cumulative summation operation on the instantaneous adsorption load to generate a cumulative adsorption load sequence, smoothing processing of the cumulative adsorption load sequence, calculation of the unit adsorption load increment based on the smoothed cumulative adsorption load sequence, and final output of a dynamic cumulative adsorption load feature sequence; Differential processing is performed on the bed pressure difference time series data in the basic time series data to extract the bed pressure difference value between the current time and the previous time. Then, the bed pressure difference value between the current time and the previous time is calculated as the pressure difference increment per unit time. Median filtering is performed on the pressure difference increment sequence per unit time. Based on the filtered pressure difference increment sequence, a time-normalized pressure difference change rate is constructed, and finally, a stable pressure difference increment feature sequence is output. The dynamic cumulative adsorption load feature sequence and the stable pressure difference increment feature sequence are correlated and modeled. The stable pressure difference increment feature sequence and the dynamic cumulative adsorption load feature sequence are aligned to form a two-dimensional correlation dataset. Local linear fitting is performed on the two-dimensional correlation dataset to calculate the slope of change. Sliding standard deviation analysis is applied to the slope of change to identify local fluctuation characteristics. The slope transition threshold is dynamically updated based on the local fluctuation characteristics, and finally the pressure difference growth rate slope feature sequence is output. The slope transition point identification process is performed on the slope feature sequence of the differential pressure growth rate. Candidate points in the slope feature sequence that exceed the dynamically updated slope transition threshold are detected. The validity of the candidate points within the preset time window is verified. The slope transition intensity quantization value is calculated for the continuously valid candidate points. Based on the quantization calculation result, it is determined whether the slope transition intensity meets the preset transition condition. Finally, the slope transition point identification signal is output. The slope transition point identifier signal is processed to generate a pressure difference growth phase transition criterion. The transition duration is verified by combining the slope transition point identifier signal with the cumulative running time to see if it exceeds a preset time threshold. Then, the transition duration verification result is fused with the slope transition intensity quantification value. A binary phase transition state flag is generated based on the fusion result. The confidence weighting calculation is performed on the binary phase transition state flag, and finally, a structured pressure difference growth phase transition criterion is output.

5. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 4, characterized in that, The structured differential pressure growth phase transition criterion includes the phase transition activation state, confidence index, and duration parameter.

6. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, Step S4 specifically includes: The raw time-series data of instantaneous air volume and intake air concentration are processed by sliding window differentiation to calculate the absolute value of the instantaneous air volume change rate and the absolute value of the intake air concentration fluctuation amplitude. Based on the average air volume and average intake air concentration of historical operating data, a normalization operation is performed to generate the quantified value of the disturbance intensity factor. Based on the comparison between the quantized value of the disturbance intensity factor and the preset disturbance threshold, an abnormal operating condition status identifier is generated when the quantized value of the disturbance intensity factor exceeds the preset disturbance threshold. Based on the abnormal operating condition status identifier, the initial fusion weight coefficients of the concentration curvature criterion, the differential pressure growth slope criterion, and the remaining capacity ratio criterion are dynamically adjusted. Normalization is performed on the weight coefficients of the dynamically adjusted concentration curvature criterion, the pressure difference growth slope criterion, and the remaining capacity ratio criterion to form an adaptive weight set for operating conditions.

7. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 6, characterized in that, The initial fusion weighting coefficients for the dynamically adjusted concentration curvature criterion, pressure difference growth slope criterion, and remaining capacity ratio criterion are as follows: the weighting coefficient of the concentration curvature criterion is adjusted from the initial value to 0.4, the weighting coefficient of the pressure difference growth slope criterion is adjusted to 0.35, and the weighting coefficient of the remaining capacity ratio criterion is adjusted to 0.

25.

8. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, Step S5 specifically includes: A historical penetration sample dataset is obtained. Based on the historical penetration sample dataset, the dynamic threshold of outlet concentration curvature is calculated, and a concentration response phase change threshold parameter is generated. The transition characteristics of the bed pressure difference growth rate are statistically analyzed to determine the pressure difference growth phase change threshold parameter. The remaining capacity ratio threshold parameter is calibrated according to the cumulative adsorption load integral slope shift characteristics. The concentration response phase change threshold parameter, the pressure difference growth phase change threshold parameter, and the remaining capacity ratio threshold parameter are integrated to construct a multi-scale phase change threshold criterion model. The concentration response phase change initial judgment criteria, the pressure difference growth phase change criterion, and the remaining capacity ratio feature are input into the multi-scale phase change threshold criterion model. Based on the dynamic weight coefficients in the operating condition adaptive weight set, the concentration response phase change initial judgment criteria, the pressure difference growth phase change criterion, and the remaining capacity ratio feature are subjected to weighted fusion processing to calculate the weighted comprehensive phase change index value. The weighted comprehensive phase change index value is compared with the preset threshold in the multi-scale phase change threshold criterion model to generate a phase change flag activation status identifier. Verify whether the phase transition flag activation status indicator meets the activation conditions of the multi-scale phase transition threshold criterion model, and combine the confidence adjustment mechanism of the adaptive weight set to confirm the validity of the phase transition event, and finally generate a comprehensive phase transition confirmation event structure.

9. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 8, characterized in that, The integrated phase change confirmation event structure includes an operational phase identifier, a confidence index, and a trigger reason code.

10. The polyolefin off-gas treatment-oriented activated carbon adsorption bed saturation state intelligent judgment and switching method according to claim 1, characterized in that, The structured saturation state label includes a running phase identifier, a confidence index, a trigger reason code, and a switching time window parameter.

Images