Carbon black reaction furnace combustion zone temperature field-tail gas carbon oxygen ratio cooperative optimization control system

Abstract

The present application relates to the technical field of industrial process control, and specifically discloses a carbon black reaction furnace combustion zone temperature field-tail gas carbon oxygen ratio collaborative optimization control system, which synchronously collects optical measurement signals and multiple auxiliary process signals of the combustion zone; performs online quality evaluation on the optical signals and fuses the auxiliary signals to generate comprehensive diagnostic characteristic information; utilizes a pre-trained event diagnosis model to judge the pollution state of the optical window based on the characteristic information; according to the pollution state diagnosis result, the control mode is adaptively switched, the multivariable collaborative optimization control based on the complete temperature field and the carbon oxygen ratio is executed under the normal state, and the degraded optimization control based on the key point temperature and the carbon oxygen ratio is switched to under the pollution state, so as to stabilize the combustion chemical equilibrium as the core target; and after the pollution is eliminated, a gradual recovery strategy is used to smoothly transit back to the advanced optimization mode; the present application realizes the automatic intelligent management of the whole process from pollution occurrence to recovery.

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, specifically to a coordinated optimization control system for the temperature field of the combustion zone of a carbon black reactor and the carbon-oxygen ratio of the exhaust gas. Background Technology

[0002] In the complex industrial process of carbon black production, multivariate collaborative optimization control of the temperature field distribution in the combustion zone of the reactor and the carbon-oxygen ratio in the exhaust gas is a crucial step in ensuring product quality, improving energy efficiency, and achieving cleaner production. From the perspective of automatic control systems, this process presents control challenges due to multivariate coupling, strong nonlinearity, large inertia, and time-varying operating conditions, making it a typical difficult-to-control process. Currently, the industry commonly employs advanced optical sensing technologies such as tunable diode laser absorption spectroscopy to online invert the two-dimensional temperature field in the combustion zone and combine it with the carbon-oxygen ratio signal obtained from online exhaust gas analyzers to construct a dual-loop or multivariate closed-loop optimization control system. Such systems can achieve a certain degree of optimized control under ideal operating conditions, but their overall control performance is highly dependent on the accuracy and reliability of the optical measurement signals.

[0003] The existing technology has the following shortcomings:

[0004] When the critical optical measurement signals used to reconstruct the temperature field of the combustion zone are distorted due to window contamination, the existing collaborative optimization control system cannot effectively identify this fault state, nor does it have the ability to operate safely and self-recover in the event of partial sensor failure, which leads to the failure of the entire optimization process or catastrophic misoperation. Summary of the Invention

[0005] The purpose of this invention is to provide a coordinated optimization control system for the temperature field of the combustion zone and the carbon-oxygen ratio of the exhaust gas in a carbon black reactor, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] The temperature field in the combustion zone of the carbon black reactor and the carbon-oxygen ratio in the exhaust gas are jointly optimized and controlled, including:

[0008] The multi-source signal synchronous acquisition module synchronously acquires optical measurement signals from the combustion zone of the carbon black reactor as well as multiple auxiliary process signals.

[0009] The signal quality assessment and feature fusion module processes optical measurement signals online to calculate signal quality indices, and then fuses these indices with auxiliary process signals to generate comprehensive diagnostic feature information.

[0010] The optical window contamination status diagnosis module inputs comprehensive diagnostic feature information into a pre-trained event diagnosis model and outputs the contamination status judgment result corresponding to the current optical measurement window. The contamination status judgment result includes normal status, instantaneous contamination status, or continuous contamination status.

[0011] The control mode decision module uses the pollution status judgment result as the basis for mode selection; if the judgment result is a normal state, the first control mode is activated; if the judgment result is an instantaneous pollution state or a continuous pollution state, the second control mode is activated.

[0012] The adaptive collaborative control execution and recovery module performs multivariate collaborative optimization control based on the combustion zone temperature field distribution reconstructed from optical measurement signals and the real-time exhaust gas carbon-oxygen ratio measurement when executing the first control mode. When executing the second control mode, it switches to degraded optimization control based on at least one temperature measurement value and the real-time exhaust gas carbon-oxygen ratio measurement value, with stable combustion chemical balance as the core objective. It receives updated information from the signal quality assessment and feature fusion module and the optical window pollution state diagnosis module and performs cyclic diagnosis. When the pollution state judgment result continuously output by the optical window pollution state diagnosis module returns to the normal state, the control execution process returns from the second control mode to the first control mode.

[0013] As a further aspect of the present invention: the generation of comprehensive diagnostic feature information specifically includes:

[0014] Multi-dimensional quality assessment in the time and frequency domains is performed on continuously acquired optical measurement signals, and multiple independent quality sub-indicators are extracted, including signal baseline drift rate, effective absorption peak amplitude fluctuation variance, and high-frequency band noise energy ratio.

[0015] The synchronously acquired auxiliary process signals are preprocessed, including signal alignment and normalization based on process event timestamps, and dynamic change trend features of the auxiliary process signals are extracted.

[0016] Multiple independent quality sub-indices and dynamic trend characteristics are organized into fixed time windows, and the instantaneous correlation coefficients between the quality sub-indices of optical measurement signals and the changes in specific auxiliary process signals are calculated to jointly construct comprehensive diagnostic feature information.

[0017] As a further aspect of the present invention: the construction process of the event diagnosis model is as follows:

[0018] Extract multiple historical operating condition segments from historical production data, including known cleanliness and known contamination states of the optical measurement window;

[0019] For each historical working condition segment, the corresponding historical comprehensive diagnostic feature information is calculated offline according to the method of generating comprehensive diagnostic feature information, and the historical comprehensive diagnostic feature information is associated with the known state labels of the corresponding segment to form training samples;

[0020] Using training samples, a feedforward logic network based on attention weighting is trained and validated in a supervised manner, enabling the feedforward logic network to learn the mapping relationship between the comprehensive diagnostic feature information and different pollution states.

[0021] As a further aspect of the present invention: the criteria for determining the pollution state are as follows:

[0022] The criteria for judging a normal state are: multiple independent quality sub-indices of the optical measurement signal are stable within their respective preset confidence ranges, and the instantaneous correlation coefficients with the auxiliary process signal conform to the expected physical correlation.

[0023] The criteria for judging the instantaneous contamination state are: the signal baseline drift rate and the proportion of noise energy in a specific high-frequency band simultaneously undergo a step deterioration within a short time window after receiving the raw material switching event signal, and then tend to stabilize without continuing to deteriorate;

[0024] The criteria for determining a persistent pollution state are: the variance of the effective absorption peak amplitude fluctuation continuously decreases in one direction and falls below the minimum threshold, and the duration of the corresponding state exceeds the preset time window.

[0025] As a further aspect of the present invention: the execution of multivariate collaborative optimization control specifically includes:

[0026] Based on the reconstructed combustion zone temperature field distribution, several first characteristic parameters characterizing the location and distribution uniformity of the high-temperature core area are extracted; the real-time exhaust gas carbon-oxygen ratio measurement value is filtered, and the changing trend is extracted as the second characteristic parameter.

[0027] Multiple first characteristic parameters and second characteristic parameters, along with the current operating variable setpoints, are input into a dynamic prediction unit. The dynamic prediction unit, through its built-in process response relationship, synchronously predicts the evolution trajectory of the first characteristic parameters and second characteristic parameters over multiple future control cycles.

[0028] Based on the preset control targets regarding temperature field stability and carbon-oxygen ratio target range, the evolution trajectory is evaluated, and a set of operational variable adjustment amounts that make the predicted trajectory optimally satisfy the control targets are calculated in a rolling manner.

[0029] The adjusted values ​​of the manipulated variables are converted into specific actuator control commands and issued to complete one control loop.

[0030] As a further aspect of the present invention: the core objective of stabilizing combustion chemical equilibrium specifically includes:

[0031] Calculate the integral of the temperature deviation of the temperature measurement value from the safe operating benchmark; calculate the instantaneous deviation of the real-time exhaust gas carbon-oxygen ratio measurement value from the ideal stoichiometric ratio;

[0032] The integral of the temperature deviation and the instantaneous offset are weighted and fused to generate a composite control index that characterizes the overall deviation of the combustion state.

[0033] Based on the magnitude and direction of change of the composite control index, the ratio of raw material supply to combustion air flow is adjusted first to make the composite control index change in the direction of decreasing.

[0034] During the adjustment of the mixing ratio, the temperature measurement value is monitored to ensure that the temperature measurement value is always within the preset safe operating range.

[0035] As a further aspect of the present invention: the cyclic diagnosis specifically includes:

[0036] It receives real-time information on the latest signal quality assessment and comprehensive diagnostic features, as well as the results of pollution status judgment;

[0037] Trend analysis is performed on multiple pollution status judgment results received consecutively to calculate a stability metric value that characterizes the direction and degree of fluctuation of status changes.

[0038] Based on the magnitude of the stability metric, adjust the time interval for the next trigger event diagnostic model to perform a diagnosis;

[0039] When the adjusted time interval is reached, the latest comprehensive diagnostic feature information is input into the event diagnosis model to trigger a new round of diagnosis and output an updated pollution status judgment result.

[0040] As a further aspect of the present invention: the control execution process recovers from the second control mode to the first control mode, specifically including:

[0041] After obtaining the pollution status judgment results of the normal state a preset number of times, the temperature measurement value is compared with the calculated value of the temperature field at the corresponding position, which is initially reconstructed from the optical measurement signal, and the consistency deviation is calculated.

[0042] Based on the magnitude of the consistency deviation, a confidence coefficient ranging from 0 to 1 is set, and the temperature field distribution parameters reconstructed from the optical measurement signal are gradually and in stages reintroduced into the control objective function using the confidence coefficient as a weight.

[0043] During the gradual introduction process, the fluctuation range of the temperature measurement value and the real-time exhaust gas carbon-oxygen ratio measurement value is monitored in real time. If the fluctuation range does not exceed the preset recovery process tolerance, the confidence coefficient is increased until it reaches 1, thus completing the full recovery to the first control mode.

[0044] The beneficial effects of this invention are:

[0045] (1) This invention, by introducing a multi-source signal fusion and intelligent diagnostic mechanism, endows the system with self-sensing, self-diagnostic, and adaptive fault-tolerant capabilities for pollution events. Specifically, the system can identify the quality degradation of optical signals in real time and accurately distinguish the type of pollution (instantaneous or continuous), thereby automatically triggering a safety degradation of the control strategy. In degradation mode, the system switches to control based on reliable redundant temperature points and the carbon-oxygen ratio of the exhaust gas, with stable combustion chemical equilibrium as the core, thus avoiding major production risks such as furnace temperature runaway, product scrapping, and even unplanned shutdowns caused by relying on "ghost temperature field" data. This solves the long-standing problem of "measurement vulnerability" in this field, enabling advanced process control to be stably applied to actual industrial environments with variable raw materials and complex operating conditions.

[0046] (2) This invention not only achieves safe operation under contaminated conditions, but also automates the entire process management from abnormal to normal through a closed-loop "diagnosis-control-recovery" logic. During degraded control, the system continuously performs cyclical diagnosis, actively monitoring the recovery of the optical window status, rather than passively waiting for manual intervention. Once the contamination is confirmed to be cleared, the system automatically recovers to the full-function collaborative optimization mode in a gradual and smooth manner through a dynamic trust coefficient calculated based on consistency deviation. This process eliminates the need for frequent judgment and manual switching by operators, minimizing the time the system operates in degraded mode, thereby reducing hidden economic losses such as increased raw material consumption and product quality fluctuations caused by degraded control performance. At the same time, it reduces the reliance on frequent preventive cleaning and maintenance of the optical window, optimizing maintenance activities into state-based predictive maintenance, which helps extend the equipment operating cycle and comprehensively improve production efficiency and economic benefits. Attached Figure Description

[0047] The invention will now be further described with reference to the accompanying drawings.

[0048] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 As shown, this invention is a coordinated optimization control system for the temperature field of the combustion zone and the carbon-oxygen ratio of the exhaust gas in a carbon black reactor, comprising:

[0051] The multi-source signal synchronous acquisition module synchronously acquires optical measurement signals from the combustion zone of the carbon black reactor as well as multiple auxiliary process signals.

[0052] The signal quality assessment and feature fusion module processes optical measurement signals online to calculate signal quality indices, and then fuses these indices with auxiliary process signals to generate comprehensive diagnostic feature information.

[0053] The optical window contamination status diagnosis module inputs comprehensive diagnostic feature information into a pre-trained event diagnosis model and outputs the contamination status judgment result corresponding to the current optical measurement window. The contamination status judgment result includes normal status, instantaneous contamination status, or continuous contamination status.

[0054] The control mode decision module uses the pollution status judgment result as the basis for mode selection; if the judgment result is a normal state, the first control mode is activated; if the judgment result is an instantaneous pollution state or a continuous pollution state, the second control mode is activated.

[0055] The adaptive collaborative control execution and recovery module performs multivariate collaborative optimization control based on the combustion zone temperature field distribution reconstructed from optical measurement signals and the real-time exhaust gas carbon-oxygen ratio measurement when executing the first control mode. When executing the second control mode, it switches to degraded optimization control based on at least one temperature measurement value and the real-time exhaust gas carbon-oxygen ratio measurement value, with stable combustion chemical balance as the core objective. It receives updated information from the signal quality assessment and feature fusion module and the optical window pollution state diagnosis module and performs cyclic diagnosis. When the pollution state judgment result continuously output by the optical window pollution state diagnosis module returns to the normal state, the control execution process returns from the second control mode to the first control mode.

[0056] In the multi-source signal synchronous acquisition module, optical measurement signals from the combustion zone are acquired. These signals are obtained through optical windows installed at specific locations within the reactor, specifically using tunable diode laser absorption spectroscopy. Laser emitters and receivers are arranged along multiple preset measurement paths within the combustion zone. A laser beam of a preset wavelength emitted by the laser passes through the combustion zone and is received by detectors at corresponding locations. The laser performs wavelength scanning, and the detectors convert the received light intensity signals into electrical signals, forming a continuously varying raw absorption spectrum sequence over time. This sequence constitutes the optical measurement signal.

[0057] Multiple auxiliary process signals are acquired simultaneously. These auxiliary process signals include, but are not limited to: the instantaneous flow rate of feedstock oil measured by a mass flow meter, the reactor throat pressure measured by a pressure transmitter, the temperature signal of at least one reference point upstream of the combustion zone measured by a thermocouple, and the binary status signal of feedstock switching commands directly obtained from the control unit. The acquisition frequency of all auxiliary process signals is synchronized with the scanning cycle of the optical measurement signals or strictly aligned using timestamps to ensure that all data correspond to the process state at the same moment.

[0058] In the signal quality assessment and feature fusion module, online quality assessment of the optical measurement signal is performed to extract multiple independent quality sub-indicators. For each preset laser measurement path, the absorption spectrum sequence acquired in the most recent 50 consecutive scan cycles is analyzed. The signal baseline drift rate is obtained by calculating the average signal value in the non-absorption region within each scan cycle, then calculating the absolute value of the difference between the average values ​​of adjacent cycles in the 50 consecutive cycle average value sequence, and finally taking the arithmetic mean of these absolute values. The unit is the change in light intensity per cycle. The effective absorption peak amplitude fluctuation variance is calculated for the light intensity value at the center wavelength of the absorption peak, calculating the variance of the light intensity value at that position within 50 consecutive cycles. Specifically, the arithmetic mean of these 50 light intensity values ​​is first calculated, then the square of the difference between each light intensity value and the mean is calculated, and finally these squared values ​​are summed and divided by the total number of cycles. The calculation of the high-frequency noise energy ratio involves performing a fast Fourier transform on the original time-domain light intensity signal within one scan cycle, calculating the sum of the squares of the amplitudes of all frequency components in the frequency band with frequencies higher than 10 kHz in the signal spectrum, and then dividing by the sum of the squares of the amplitudes of all frequency components in the entire spectrum.

[0059] The auxiliary process signals are preprocessed and their dynamic trend characteristics are extracted. Based on a unified time stamp, the feedstock flow signal, reference point temperature signal, and optical measurement signal are aligned. After alignment, each auxiliary process signal is normalized. Specifically, within a 10-second sliding time window, the current signal value is subtracted from the minimum value within that window, and then divided by the difference between the maximum and minimum values ​​within the window. The result is mapped to the interval between 0 and 1. The dynamic trend characteristics are obtained by calculating the approximate value of the first derivative of the normalized signal. Specifically, the difference between the signal value at the current sampling time and the signal value at the previous sampling time is calculated and then divided by the time interval between the two sampling points.

[0060] The aforementioned indicators and features are integrated into comprehensive diagnostic feature information. Using a fixed time window of 5 seconds, the baseline drift rate, amplitude fluctuation variance, and noise energy ratio of all laser paths calculated within this window, along with the dynamic trend characteristics of each auxiliary process signal, are collected and organized. Furthermore, the instantaneous correlation coefficient between the baseline drift rate of the optical measurement signal and the trend characteristics of the feed oil flow rate within this time window is calculated. The calculation method involves first calculating the arithmetic mean of each of the two sequences, then summing the products of the deviations of the corresponding values ​​of the two sequences from their mean values, and finally dividing each product by the product of the standard deviation and the sequence length of each sequence. All collected quality sub-indicators, dynamic trend characteristics, and instantaneous correlation coefficients together constitute a one-dimensional array, which is the comprehensive diagnostic feature information.

[0061] In the optical window contamination status diagnosis module, the event diagnosis model is constructed through the following steps. First, multiple historical operating condition segments recording different states of the optical measurement window are extracted from the data platform storing historical production process data. These segments are obtained by querying operation event logs and maintenance records, specifically including: segments where the window has been running stably for at least 30 minutes after cleaning and maintenance, marked as "known clean state"; segments 2 minutes before and after a raw material oil switching operation, marked as "known instantaneous contamination state"; and segments where the window's performance has continuously declined due to long-term operation until maintenance (usually several hours), marked as "known continuous contamination state". Second, for each extracted historical operating condition segment, the corresponding historical comprehensive diagnostic feature information is calculated offline using the stored original optical signal and auxiliary process signal, following the aforementioned method for generating comprehensive diagnostic feature information. Then, each set of calculated historical comprehensive diagnostic feature information is manually associated with the state label (i.e., "normal", "instantaneous contamination", or "continuous contamination") corresponding to the segment's source, forming a complete training sample. Finally, the set consisting of all training samples is used to perform supervised training on an attention-weighted feedforward logic network. The network has a three-layer structure: the input layer has the same number of neurons as the dimension of the comprehensive diagnostic feature information; a hidden layer with 64 neurons activated using a linear rectified function; and an output layer with 3 neurons, each corresponding to one of the three contamination states. Its attention weighting mechanism is as follows: a weight calculation layer is introduced after the hidden layer. This layer calculates a weight between 0 and 1 for each of the 64 values ​​output from the hidden layer. The calculation method involves first mapping each value to a logistic function, then normalizing it so that the sum of all weights is 1. The output value of the hidden layer is then multiplied by the corresponding weight before being passed to the output layer. The training process uses cross-entropy as the loss function and adaptive moment estimation (IME) for optimization. Training stops when the accuracy on the validation set no longer improves after 10 consecutive training epochs, thus obtaining a model that can map comprehensive diagnostic feature information to contamination state probabilities.

[0062] It should be noted that the selection of historical operating condition segments follows the principles of ensuring the scale, diversity, and representativeness of the training samples:

[0063] Quantitative principle: To ensure the model can learn sufficient patterns, at least 50 historical segments should be extracted for each known state (clean, transient contamination, continuous contamination). This covers different raw material batches, different production loads, and different operating stages of the equipment, ensuring the model's generalization ability.

[0064] Duration basis: The duration of each segment is designed to capture the full dynamic characteristics of the state.

[0065] "Known Cleanliness Status" segment: Data is extracted from the window after it has been running stably for at least 30 minutes following cleaning, to cover normal process fluctuations.

[0066] "Known Instantaneous Contamination Status" segment: Typically, data from 2-5 minutes before and after a raw material switching event is extracted to fully capture the transient process from the occurrence of contamination to its initial dissipation.

[0067] "Known Persistent Contamination Status" segment: This segment captures the entire process from the initial appearance of contamination characteristics to the continuous decline in performance before maintenance, typically lasting several hours, to cover the slow deterioration trend.

[0068] The specific criteria and threshold definitions for judging the contamination status are as follows. The criteria for judging the normal state are: in the comprehensive diagnostic feature information generated in the latest 5-second time window, the baseline drift rate of all laser paths is less than the upper limit of their preset confidence range (e.g., 0.5% per cycle), the variance of the amplitude fluctuation of all effective absorption peaks is greater than the lower limit of their preset confidence range (e.g., 60% of the initial calibration value), the proportion of high-frequency noise energy is less than its preset upper limit (e.g., 5%), and the instantaneous correlation coefficient between the baseline drift rate of the optical signal and the trend of feedstock flow rate change is greater than 0.7. The criteria for judging the instantaneous contamination status are: within 60 seconds after the control unit issues the feedstock switching command, the comprehensive diagnostic feature information shows that more than half of the laser paths have a baseline drift rate exceeding three times the upper limit of the normal range, and their high-frequency noise energy proportion exceeds twice the upper limit of the normal range; and in the following 120 seconds, the values ​​of these indicators no longer show a continuous increasing trend, but stabilize at a new level. The criteria for determining persistent contamination are: within six consecutive time windows (i.e., 30 seconds), over 80% of the laser paths in the comprehensive diagnostic feature information have an effective absorption peak amplitude fluctuation variance that remains below its minimum threshold (e.g., 30% of the initial calibration value), and this trend shows no signs of recovery over time. During inference, the event diagnosis model calculates the degree to which the input comprehensive diagnostic feature information conforms to the above three categories of rules, and outputs the category with the highest degree of conformity as the contamination status judgment result for the current optical measurement window.

[0069] It should be noted that the specific values ​​of the key thresholds involved in the above judgment criteria (such as the upper limit of baseline drift rate, the minimum threshold of effective absorption peak amplitude fluctuation variance, and the upper limit of high-frequency band noise energy ratio) are determined comprehensively through the following methods and can be adapted according to actual applications:

[0070] Initial calibration benchmark: Under ideal conditions where the optical measurement system is installed and debugged and the window is clean, long-term (e.g., more than 24 hours) data collection is carried out, and the initial mean and standard deviation of each quality sub-index are statistically calculated as the benchmark of the "preset confidence range".

[0071] Long-term statistical patterns: Collect long-term historical operating data and analyze the changes in the statistical distribution of each quality sub-indicator before and after recognized normal production periods, known minor pollution events, and serious pollution alarms, so as to determine the statistical boundaries that distinguish different states.

[0072] Process safety boundary deduction: Combining process knowledge, we analyze the allowable temperature field reconstruction error range to ensure reaction safety and product quality, and then deduce the minimum requirements that the optical signal quality must maintain, thereby setting key minimum thresholds (such as the minimum threshold for effective absorption peak amplitude fluctuation variance).

[0073] Empirical adaptation: The threshold determined by the above method can be empirically fine-tuned within ±20% based on the specific characteristics of the raw materials (such as the fact that high aromatic oils are prone to producing smoke and dust), the furnace structure of the reactor, and the specific installation position of the optical window, in order to achieve the best diagnostic sensitivity and reliability.

[0074] In the control mode decision module, the pollution status judgment result is used as the basis for mode selection; if the judgment result is a normal state, the first control mode is activated; if the judgment result is an instantaneous pollution state or a continuous pollution state, the second control mode is activated.

[0075] In the adaptive cooperative control execution and recovery module, firstly, based on the two-dimensional temperature field distribution reconstructed from optical measurement signals, key first feature parameters are extracted. These include the location parameters of the high-temperature core region. The uniformity parameter is obtained by calculating the weighted average of the vertical coordinates of all pixels in the temperature field whose temperature values ​​are higher than a preset threshold. The weights are the differences between the temperature value of each point and the threshold. The calculation process is as follows: The entire combustion zone is divided into five equal regions longitudinally. The average temperature of each region is calculated, and then the standard deviation of these five average temperature values ​​is calculated. Finally, the uniformity is characterized by dividing 1 by the quotient of this standard deviation and the sum of a reference temperature; the larger the value, the more uniform the uniformity. Secondly, a first-order low-pass digital filter is applied to the real-time exhaust gas carbon-oxygen ratio measurement to eliminate high-frequency noise. The filtered value is denoted as... Then, the slope of its linear fit within the most recent 10-second time window is calculated, and this slope is the second characteristic parameter characterizing its changing trend. Subsequently, , and These three characteristic parameters, along with the current fuel oil flow rate setpoint Combustion air flow setpoint and raw material oil flow rate set value These inputs are fed into the dynamic prediction unit. This unit contains a simplified process model describing the dynamic relationship between the manipulated variables and feature parameters, represented as a series of difference equations. Using the aforementioned inputs as initial conditions, the model recursively predicts the next eight control cycles, with each cycle lasting one second. , and The evolutionary trajectory. Then, based on the preset control objective, namely... It needs to stabilize within the target range. Maximize The predicted trajectory needs to be evaluated to maintain it within the target range. A constrained quadratic programming method is used to iteratively solve for a set of operational variable adjustment sequences over the next eight periods. ,in, This indicates the amount of fuel oil flow adjustment. This indicates the adjustment amount of combustion air flow. This represents the adjustment amount of the feedstock oil flow rate, minimizing the overall deviation between the predicted trajectory and the target. Finally, only the adjustment amount of the first cycle in this sequence is converted into a specific valve opening degree or frequency converter command and sent to the corresponding actuator, thus completing one control loop.

[0076] The specific implementation of the downgraded optimization control is as follows. When this mode is enabled, the control core switches to stable combustion chemical equilibrium. First, the key point temperature measurements are calculated. Deviating from its safe operating benchmark Temperature deviation integral The calculation method involves calculating the temperature deviation for each control cycle (e.g., 1 second) starting from the start of degraded mode. The summation is performed, but multiplied by a forgetting factor less than 1 (e.g., 0.95) before summation to emphasize the effect of recent bias. Simultaneously, the real-time exhaust carbon-to-oxygen ratio measurement is calculated. Deviating from its ideal stoichiometry instantaneous offset ,Right now: ;in, The first step is to represent the measured carbon-oxygen ratio of the exhaust gas. Then, these two quantities are weighted and fused to generate a composite control index. The calculation formula is as follows: ;in, and Assuming preset weighting coefficients, temperature stability and chemical equilibrium are assigned different levels of importance, while satisfying the following: . indicates taking the absolute value. Then, according to The magnitude of the value and its direction of change compared to the previous period will be used to prioritize adjusting the supply of crude oil. Combustion airflow The proportions. For example, if Increase and mainly composed of The contribution is then fine-tuned to increase airflow. The adjustment strategy uses proportional control, with the adjustment amount proportional to the [missing information - likely a percentage or unit]. The current value. During this process, continuous monitoring... Once it approaches the preset safe operating range boundary, the optimization adjustment of the blend ratio is suspended, and a small fuel oil compensation adjustment is introduced to... Pull back to the safe zone to ensure process safety.

[0077] The specific implementation of the cyclical diagnosis is as follows. During the second control mode operation, diagnostic updates are continuously performed. First, the latest comprehensive diagnostic feature information from the signal quality assessment and feature fusion process, as well as the latest judgment results from the pollution status diagnosis process, are received in real time. Second, trend analysis is performed on multiple (e.g., the most recent 10) pollution status judgment results received consecutively. A stability metric is calculated. Count the number of times "normal state" appeared in these 10 results. And calculate the number of state transitions between consecutive results. (For example, changing from "instantaneous pollution" to "continuous pollution" or vice versa). Stability metric. The calculation process can be expressed as: ;in and For experience weights. A higher value indicates a more stable state. Next, based on... The size dynamically adjusts the time interval for the next diagnostic trigger. If the base interval is set to 5 seconds, the adjustment rule is as follows: If If it is below 0.3, then Seconds (frequent diagnosis); if If it is between 0.3 and 0.7, then seconds; if If it is higher than 0.7, then Seconds (reduce diagnostic frequency). Finally, upon reaching the adjusted time interval. At that time, the latest comprehensive diagnostic feature information is automatically input into the event diagnosis model, triggering a new round of diagnosis, thereby obtaining an updated pollution status judgment result.

[0078] The specific implementation of the control mode recovery transition is as follows. When the recovery conditions are met, a smooth transition from the second control mode to the first control mode is executed. First, the recovery conditions are confirmed: the pollution status judgment result continuously output by the optical window pollution status diagnosis module must be "normal state", and the number of consecutive judgments must reach a preset value (e.g., 5 times). This number of judgments is based on a trade-off between control stability and system response speed: too few judgments (e.g., 1-2 times) may lead to accidental triggering of recovery due to random fluctuations, causing control oscillations; too many judgments (e.g., more than 10 times) will unnecessarily prolong the system's running time in degraded mode. Five consecutive judgments can provide a faster recovery response while ensuring stable state recovery. After the conditions are met, at least one currently available key point temperature measurement value is used. The calculated temperature value is obtained by interpolating the temperature field initially reconstructed from the latest optical measurement signals at the corresponding spatial location. Compare and calculate their consistency deviation. The calculation expression is: Secondly, based on this consistency deviation... The size is set to a trust coefficient ranging from 0 to 1. The calculation expression is: ;

[0079] in, This is a sensitivity adjustment coefficient (e.g., a value of 5). This represents the natural exponential function. This formula ensures that when... When larger Approaching 0 (distrustful reconstruction of the temperature field), when Approaching 0 Approaching 1 (complete trust). Then, with... As weights, in the objective function of the multivariate collaborative optimization control used in the first control mode, the temperature field distribution parameters reconstructed from the optical measurement signals (such as...) are gradually and in stages reintroduced. , The optimization terms are as follows. In the initial stage, the weights of these terms are close to zero, and control mainly depends on the carbon-oxygen ratio and the critical point temperature; as... As the temperature field optimization term gradually increases, its weight also increases proportionally. During this gradual introduction process, the temperature measurements at key points are monitored in real time. Compared with real-time exhaust carbon-oxygen ratio measurement value The fluctuation range (calculated in this embodiment as its standard deviation over 10 seconds). If the fluctuation range does not exceed the preset recovery process tolerance (set in this embodiment as: temperature standard deviation < 5K, carbon-oxygen ratio standard deviation < 0.02), it indicates that the recovery process is stable, and the confidence coefficient can continue to be increased slowly. The set value or by calculating a new Update .when When the value finally reaches or stabilizes at 1, and the multivariate collaborative optimization controller has fully taken over and is running stably, it signifies the completion of the full recovery to the first control mode, and the system re-enters the advanced collaborative optimization state based on the complete temperature field and carbon-oxygen ratio.

[0080] It should be noted that the tolerance parameters in this embodiment are set based on the allowable fluctuation range of key parameters for maintaining stable operation of the carbon black reactor combustion process. For example, a temperature standard deviation of <5K means that during the recovery process, temperature fluctuations at key points within the furnace are limited to a small range allowed by the process operating procedures, ensuring that the recovery action will not impact production stability. The limitation on the carbon-oxygen ratio standard deviation similarly ensures that the combustion chemical equilibrium will not shift drastically during the recovery period. These tolerance values ​​are derived from process design parameters and long-term operational experience.

[0081] The working principle of this invention is as follows: First, the optical measurement signal of the combustion zone is acquired through a multi-source signal synchronous acquisition module, and auxiliary process signals such as feedstock flow rate, furnace pressure, and key point temperature are acquired simultaneously. The signal quality assessment and feature fusion module performs real-time quality assessment of the optical signal, extracting quality sub-indicators such as baseline drift rate, absorption peak fluctuation variance, and high-frequency noise ratio, and fuses them with the trend characteristics of the preprocessed auxiliary process signals to generate comprehensive diagnostic feature information. The optical window contamination state diagnosis module inputs this feature information into a pre-trained event diagnosis model based on an attention weighting mechanism, outputting a state judgment result of "normal," "instantaneous contamination," or "continuous contamination." The control mode decision module makes a decision based on this: if the state is normal, the first control mode is activated; if the state is contaminated, the second control mode is activated. The adaptive collaborative control execution and recovery module executes corresponding control: in the first control mode, multivariate model prediction and collaborative optimization are performed based on the reconstructed temperature field distribution and exhaust gas carbon-oxygen ratio; in the second control mode, it switches to degradation optimization based on key point temperature and exhaust gas carbon-oxygen ratio, with the core objective of stabilizing combustion chemical equilibrium. During this process, the system continuously performs cyclical diagnostics to monitor changes in the window state. When the contamination state continuously returns to normal, it dynamically introduces a confidence coefficient by calculating the consistency deviation between the temperature measurement value and the reconstructed value. This allows the control to be safely restored from the second mode to the first mode in a gradual and smooth manner, thereby achieving highly robust intelligent optimization control of carbon black production under complex working conditions.

[0082] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A coordinated optimization control system for temperature field in the combustion zone of a carbon black reactor and carbon-oxygen ratio in the exhaust gas, characterized in that, include: The multi-source signal synchronous acquisition module acquires optical measurement signals and auxiliary process signals from the combustion zone of the carbon black reactor. The auxiliary process signals include feed oil flow rate, furnace pressure, and reference point temperature. The signal quality assessment and feature fusion module processes the optical measurement signal to calculate the signal quality index, fuses the signal quality index with the auxiliary process signal, calculates the real-time correlation coefficient, and generates comprehensive diagnostic feature information. The generation of comprehensive diagnostic feature information specifically includes: Multi-dimensional quality assessment in the time and frequency domains is performed on continuously acquired optical measurement signals, and multiple independent quality sub-indicators are extracted, including signal baseline drift rate, effective absorption peak amplitude fluctuation variance, and high-frequency band noise energy ratio. The synchronously acquired auxiliary process signals are preprocessed, including signal alignment and normalization based on process event timestamps, and dynamic change trend features of the auxiliary process signals are extracted. Multiple independent quality sub-indices and dynamic trend characteristics are organized into fixed time windows, and the instantaneous correlation coefficient between the quality sub-indices of the optical measurement signal and the changes in the auxiliary process signal is calculated to jointly construct comprehensive diagnostic feature information. The optical window contamination status diagnosis module inputs comprehensive diagnostic feature information into the event diagnosis model and outputs the contamination status judgment result corresponding to the current optical measurement window, including: Normal state: Multiple independent quality sub-indices of the optical measurement signal are stable within the preset confidence range, and their correlation coefficients with the instantaneous correlation coefficients conform to the expected physical correlation; Transient contamination state: Within a short time window following a raw material switching event, the signal baseline drift rate and the proportion of high-frequency noise energy simultaneously undergo a step-like degradation, and then tend to stabilize; Continuous pollution state: The effective absorption peak amplitude fluctuation variance continuously decreases in one direction and falls below the minimum threshold, and the duration exceeds the preset time window; The control mode decision module is used to switch control modes based on the pollution status judgment result. Specifically, if the judgment result is a normal state, the first control mode is activated; if the judgment result is an instantaneous pollution state or a continuous pollution state, the second control mode is activated. The adaptive collaborative control execution and recovery module executes the first control mode under normal conditions, performing multivariate collaborative optimization control based on the temperature field distribution reconstructed from the optical signal and the real-time carbon-oxygen ratio of the exhaust gas. Under polluted conditions, it switches to a downgraded optimization control based on at least one temperature measurement value and the carbon-oxygen ratio of the exhaust gas, with stable combustion chemical balance as the core objective. After the pollution is eliminated, it calculates the consistency deviation between the temperature measurement value and the optically reconstructed temperature field at the corresponding position, sets a confidence coefficient of 0 to 1, and gradually reintroduces the temperature field distribution parameters with the confidence coefficient as the weight to achieve a gradual and smooth recovery from the second control mode to the first control mode.

2. The carbon black reactor combustion zone temperature field-tail gas carbon-oxygen ratio co-optimization control system according to claim 1, characterized in that, The process of constructing the event diagnosis model is as follows: Extract multiple historical operating condition segments from historical production data, including known cleanliness and known contamination states of the optical measurement window; For each historical working condition segment, the corresponding historical comprehensive diagnostic feature information is calculated offline according to the method of generating comprehensive diagnostic feature information, and the historical comprehensive diagnostic feature information is associated with the known state labels of the corresponding segment to form training samples; Using training samples, a feedforward logic network based on attention weighting is trained and validated in a supervised manner, enabling the feedforward logic network to learn the mapping relationship between the comprehensive diagnostic feature information and different pollution states. The attention weighting is as follows: the feedforward logic network introduces a weight calculation layer after the hidden layer. The weight calculation layer calculates a weight between 0 and 1 for each of the 64 values ​​output by the hidden layer. The calculation method is to first input each value into a logic function to map it to the range of 0 to 1, and then use a normalization operation to make the sum of all weights equal to 1. Then the output value of the hidden layer is multiplied by the corresponding weight and then passed to the output layer.

3. The temperature field-tail gas carbon-oxygen ratio co-optimization control system for the combustion zone of the carbon black reactor according to claim 1, characterized in that, The aforementioned multivariate collaborative optimization control specifically includes: Based on the reconstructed combustion zone temperature field distribution, several first characteristic parameters characterizing the location and distribution uniformity of the high-temperature core area are extracted; the real-time exhaust gas carbon-oxygen ratio measurement value is filtered, and the changing trend is extracted as the second characteristic parameter. Multiple first characteristic parameters and second characteristic parameters, along with the current operating variable setpoints, are input into a dynamic prediction unit. The dynamic prediction unit, through its built-in process response relationship, synchronously predicts the evolution trajectory of the first characteristic parameters and second characteristic parameters over multiple future control cycles. Based on the preset control targets regarding temperature field stability and carbon-oxygen ratio target range, the evolution trajectory is evaluated, and a set of operational variable adjustment amounts that make the predicted trajectory optimally satisfy the control targets are calculated in a rolling manner. The adjusted values ​​of the manipulated variables are converted into specific actuator control commands and issued to complete one control loop.

4. The carbon black reactor combustion zone temperature field-tail gas carbon-oxygen ratio co-optimization control system according to claim 1, characterized in that, The core objective of stabilizing combustion chemical equilibrium specifically includes: Calculate the integral of the temperature deviation of the temperature measurement value from the safe operating benchmark; calculate the instantaneous deviation of the real-time exhaust gas carbon-oxygen ratio measurement value from the ideal stoichiometric ratio; The integral of the temperature deviation and the instantaneous offset are weighted and fused to generate a composite control index that characterizes the overall deviation of the combustion state. Based on the magnitude and direction of change of the composite control index, the ratio of raw material supply to combustion air flow is adjusted first to make the composite control index change in the direction of decreasing. During the adjustment of the mixing ratio, the temperature measurement value is monitored to ensure that the temperature measurement value is always within the preset safe operating range.

5. The temperature field-tail gas carbon-oxygen ratio co-optimization control system for the combustion zone of the carbon black reactor according to claim 1, characterized in that, The event diagnosis model described above is used for cyclical diagnosis, specifically including: It receives real-time information on the latest signal quality assessment and comprehensive diagnostic features, as well as the results of pollution status judgment; Trend analysis is performed on multiple pollution status judgment results received consecutively to calculate a stability metric value that characterizes the direction and degree of fluctuation of status changes. Based on the magnitude of the stability metric, adjust the time interval for the next trigger event diagnostic model to perform a diagnosis; When the adjusted time interval is reached, the latest comprehensive diagnostic feature information is input into the event diagnosis model to trigger a new round of diagnosis and output an updated pollution status judgment result.

6. The temperature field-tail gas carbon-oxygen ratio co-optimization control system for the combustion zone of the carbon black reactor according to claim 1, characterized in that, The gradual and smooth recovery from the second control mode to the first control mode specifically includes: After obtaining the pollution status judgment results of the normal state a preset number of times, the temperature measurement value is compared with the calculated value of the temperature field reconstructed by the optical measurement signal at the corresponding position, and the consistency deviation is calculated. Based on the magnitude of the consistency deviation, a confidence coefficient ranging from 0 to 1 is set, and the temperature field distribution parameters reconstructed from the optical measurement signal are gradually and in stages reintroduced into the control objective function using the confidence coefficient as a weight. During the gradual introduction process, the fluctuation range of the temperature measurement value and the real-time exhaust carbon-oxygen ratio measurement value is monitored in real time. If the fluctuation range does not exceed the preset recovery process tolerance, the confidence coefficient is increased until it reaches 1, thus completing the full recovery to the first control mode.

Images