A method and system for intelligent optimization control of petroleum coke calcination
By identifying volatile matter fluctuation patterns in real time and compensating in segments, and combining the feedforward and feedback fusion of confidence assessment, the problem of temperature lag deviation in the tank calciner was solved, and the stability of the calcined coke quality was achieved.
Patent Information
- Application Number
- CN202611516131.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-28
- Publication Date
- 2026-11-17
AI Technical Summary
The temperature of the calcination zone in the tank calciner is lagging and deviates due to the fluctuation of volatile matter in the petroleum coke entering the furnace, which affects the unstable quality of the calcined coke. The existing feedback adjustment strategy is slow to respond and cannot correct it in time.
By acquiring near-infrared spectral signals and flue gas composition signals from the top of the tank, the volatile matter is estimated in real time and fluctuation patterns are identified. The calcination zone is divided into segments, and the feedforward compensation is calculated using the segmented heat balance equation. Combined with confidence assessment, the fusion weight of feedforward and feedback is dynamically adjusted to achieve active temperature regulation.
It achieves active regulation of calcination temperature, dynamic coordination of feedforward compensation and feedback control, solves the problem of feedback regulation lag, and ensures stable quality of calcined coke.
Smart Images

Figure CN123113389A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of petroleum coke calcination control technology, and more specifically, to a method and system for intelligent optimization control of petroleum coke calcination. Background Technology
[0002] Calcination of petroleum coke is a key process in the production of prebaked anodes. It involves high-temperature heat treatment to remove volatiles and moisture from the delayed-release petroleum coke, achieving the true density and conductivity required for downstream anode preparation. This process widely utilizes tank-type calcining furnaces, which are characterized by not relying on external fuels, but rather maintaining the calcination temperature solely through the combustion of the petroleum coke's own volatiles. This results in advantages such as low burn-off rate and stable product quality.
[0003] In actual production, the volatile matter content of the petroleum coke fed into the furnace is not constant. Affected by fluctuations in the upstream delayed coking process and changes in raw material sources, the volatile matter content fluctuates continuously. Since volatile matter is the only heat source for maintaining thermal balance in the furnace, its fluctuations directly change the heat release within the furnace, causing the calcination zone temperature to deviate from the target range, thereby affecting key quality indicators such as the true density and powder resistivity of the calcined coke.
[0004] Currently, for the control of calcination temperature, existing technologies generally adopt a feedback regulation strategy. Temperature sensors detect the temperature of the calcination zone, and when the detected value deviates from the target range, the controller calculates and outputs an adjustment signal based on the deviation, changing the discharge speed to bring the temperature back to the target value. To improve the regulation quality, some solutions introduce a cascade control structure, with the flue temperature as the main controlled variable and air volume or negative pressure as auxiliary controlled variables to form inner and outer loop regulation. Other solutions use predictive control algorithms, which predict the temperature change trend by establishing a process response model and apply adjustment actions in advance.
[0005] However, feedback control can only take effect after a temperature deviation occurs. The tank-type calciner has a large heat capacity and a significant temperature response lag. Furthermore, the petroleum coke's turnover cycle within the furnace can last for several hours. When fluctuations in volatile matter cause temperature deviations, adjusting the discharge rate only becomes effective after the entire feed column has been replaced, meaning the control time lags far behind the occurrence of the disturbance.
[0006] Therefore, there is an urgent need for a calcination temperature control method that utilizes feedforward information on the volatile matter content of the raw materials entering the furnace and combines segmented compensation with dynamic coordination. Summary of the Invention
[0007] One objective of this invention is to provide an intelligent optimization control method for petroleum coke calcination. This method senses the changing trend of volatile matter content as the raw material enters the furnace, and proactively adjusts the discharge speed and air volume ratio before the temperature deviates, ensuring that the heat supply inside the furnace always matches the fluctuations in the raw material. Simultaneously, it dynamically coordinates the relationship between feedforward actions and feedback correction based on the reliability of the volatile matter detection, thereby solving the problems mentioned in the background art, namely: The temperature of the calcination zone in the tank-type calciner deviates due to the fluctuation of volatile matter in the petroleum coke entering the furnace, resulting in unstable quality of the calcined coke.
[0008] To achieve the above objective, the method includes the following steps: S1. Obtain the near-infrared spectral signal of the petroleum coke currently fed into the furnace and the component signal of the flue gas at the top of the tank, and analyze them to obtain the estimated value of volatile matter in the spectral channel and the estimated value of volatile matter in the process channel, respectively. The volatile matter estimate of the spectral channel is used as the real-time volatile matter estimate. The estimation residual is determined based on the volatile matter estimate of the process channel. The fluctuation characteristics of the real-time volatile matter estimate are extracted and the fluctuation pattern of the current volatile matter is identified. S2. Calculate the disturbance equivalent heat flux based on the difference between the real-time estimated value of volatile matter and the preset baseline volatile matter; The calcination zone is divided into multiple sections along the material moving downwards, and a heat release ratio coefficient is set for each section. The disturbance equivalent heat flow is distributed to each section according to the heat release ratio coefficient of each section to obtain the segmented disturbance heat flow of each section. The current temperature and preset target temperature of each section of the calcination zone are obtained. The corresponding compensation strategy is matched according to the fluctuation mode. The segmented disturbance heat flow and the deviation between the current temperature and the target temperature are used as inputs. The required compensation heat flow of each section is solved in reverse through the segmented heat balance equation. The compensation heat flow of each section is mapped to the feedforward compensation amount of discharge speed, negative pressure in tank and air distribution through the pre-calibrated sensitivity matrix. S3. Calculate the feedforward confidence level based on the estimated residual, calculate the coordination factor based on the feedforward confidence level, the current temperature deviation, and the equivalent heat flow of the disturbance, solve the feedback control quantity using the heat balance equation as the prediction model, and weight and fuse the feedforward compensation quantity and the feedback control quantity according to the coordination factor to output the comprehensive control command.
[0009] In the above technical solution, the control of calcination temperature does not wait until a temperature deviation occurs before adjustment begins. Instead, the starting point of regulation is moved forward to the raw material feeding stage. By using spectral methods to obtain the volatile matter content and its variation pattern in advance, compensation calculations are triggered before heat disturbances are conducted to the calcination zone, thereby eliminating the lag problem of feedback regulation at its source. Due to the large thermal inertia and long material renewal cycle of the pot furnace, if only temperature feedback is relied upon for correction, it will be too late to adjust after the temperature deviates from the target. Therefore, this solution uses feedforward compensation as the main control method. The compensation amount of discharge speed, negative pressure, and air volume is calculated based on the fluctuation amplitude and pattern of volatile matter. At the same time, the calcination zone is divided into multiple sections to establish a thermal balance relationship for each section, so that the application position of the compensation heat matches the actual heat release position. Based on this, since the spectral soft measurement itself has estimation errors, the reliability of feedforward compensation will decrease accordingly. If feedforward is still applied with fixed weights at this time, new disturbances may be introduced due to inaccurate compensation information. Therefore, this scheme further introduces confidence assessment based on the estimated residuals and dynamically adjusts the fusion weights of feedforward and feedback. This allows feedforward to dominate the adjustment when it is reliable and automatically give way to feedback correction when it is unreliable, thereby ensuring that feedforward compensation always plays a role under the premise of reliability and avoiding the risk of open-loop feedforward failing due to inaccurate information.
[0010] Based on this, the fluctuation characteristics include first-order rate of change, second-order rate of change, coefficient of variation, and dominant frequency component; The method for identifying the fluctuation characteristics is as follows: when the absolute value of the first-order rate of change exceeds the first threshold and the duration is less than the first duration threshold, it is determined to be a step-type mode. When the absolute value of the first-order rate of change is between the second threshold and the first threshold, and the first-order rate of change continues to exceed the second duration threshold with the same sign, it is determined to be a gradual change mode. When the power spectral density of the main frequency component exceeds the third threshold and the coefficient of variation is lower than the step-type determination threshold, it is determined to be a periodic oscillation mode. When the absolute value of the first-order rate of change does not exceed the second threshold and the coefficient of variation is lower than the fourth threshold, it is determined to be a steady-state mode.
[0011] In another technical solution, the segmented heat balance equation is to write the heat balance relationship for each segment separately. The heat balance relationship of each segment includes the actual heat release term of volatiles, the compensation heat flow term, the heat dissipation loss term, and the reaction endothermic term. The heat dissipation loss term is determined by the product of the heat dissipation coefficient and the temperature difference, and the reaction endothermic term is determined by the product of the reaction enthalpy change and the reaction rate. The temperature of each segment is made to approach the target temperature of each segment at a desired convergence rate, and the compensation heat flow required for each segment is solved in reverse.
[0012] In this technical solution, the design logic of fluctuation pattern recognition lies in the fact that different types of fluctuations have significantly different impact characteristics on the temperature of the calcining zone. Step-type fluctuations have large amplitudes and strong suddenness, requiring rapid response in a short time. Gradual fluctuations have smooth amplitude changes but long durations, making them more suitable for proactive adjustment based on the changing trend. The repetitive nature of periodic oscillations means that overcompensation will introduce unnecessary actuator actions, and it is more reasonable to compensate only the main frequency component through filtering. If the fluctuation patterns are not distinguished and the same compensation strategy is used for all fluctuations, it will inevitably lead to situations where the compensation effect is ideal in some operating conditions but poor or even aggravated in others, making it impossible to guarantee control quality across the entire operating range. The corresponding design logic of the segmented heat balance equation is that volatiles are not released in a concentrated manner at a fixed location in the calcining zone, but are gradually precipitated as the material layer moves downward. The timing and amount of heat release vary in different sections. Only by dividing the calcining zone into multiple sections along the material flow direction and establishing heat balance relationships for each section can the spatial application location of the feedforward compensation heat correspond to the actual heat release location. If segmentation is not performed and the entire calcination zone is treated as a whole to establish the heat balance equation, the compensation heat will be evenly distributed throughout the region, which does not match the localized concentrated characteristics of the actual heat release. The misalignment between the compensation point and the disturbance point will cause the feedforward control to fail to achieve the expected cancellation effect.
[0013] The second objective of this invention is to provide an intelligent optimization control system for petroleum coke calcination, including a data acquisition and synchronization module, a volatile matter sensing and pattern recognition module, a segmented feedforward compensation calculation module, a coordinated optimization control module, a quality feedback self-learning module, and a human-computer interaction and parameter tuning module. The data acquisition and synchronization module is used to synchronously acquire near-infrared spectral signals, flue gas composition signals from the top of the tank, and discharge speed signals. After alignment, rejection, and filtering, it outputs a synchronous data set. The volatile matter sensing and pattern recognition module is used to obtain the estimated value of volatile matter in the spectral channel by analyzing the near-infrared spectral signal, and to obtain the estimated value of volatile matter in the process channel by analyzing the flue gas component signal at the top of the tank combined with the discharge speed signal after time inversion alignment. Based on the current estimated value of the spectral channel, the module outputs the real-time estimated value of volatile matter, and calculates the estimation residual based on the deviation between the estimated value of the spectral channel before the transmission delay and the output value of the process channel. At the same time, the module extracts fluctuation features to identify the fluctuation pattern of the current volatile matter. The segmented feedforward compensation calculation module is used to calculate the disturbance equivalent heat flux based on the difference between the real-time estimated value of volatile matter and the benchmark volatile matter. After distributing the disturbance equivalent heat flux according to the heat release ratio coefficient of each segment, it combines the compensation strategy of fluctuation mode matching and temperature deviation, solves the compensation heat flux of each segment through the segmented heat balance equation, and maps it to the compensation amount of discharge speed, negative pressure in tank and air distribution through the sensitivity matrix. The coordinated optimization control module is used to determine the feedforward confidence level based on the estimated residual, calculate the coordination factor based on the feedforward confidence level, the current temperature deviation and the disturbance equivalent heat flux, and then weight and fuse the compensation quantity and the feedback control quantity according to the coordination factor to output a comprehensive control command. The quality feedback self-learning module is used to correct the soft measurement model parameters and optimize the objective function weight coefficients based on the deviation between the online estimated value of the true density of calcined coke and the laboratory measured value. The human-computer interaction and parameter tuning module is used to set and adjust control parameters and display key operating variables in real time.
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces near-infrared spectroscopy sensing into calcination temperature control, enabling volatile matter information to be acquired during the furnace feeding stage. Volatile matter fluctuations are converted into feedforward compensation quantities for discharge speed, negative pressure inside the furnace, and air distribution. The compensation action is completed before heat disturbances are conducted to the calcination zone, fundamentally solving the inherent defect of feedback control lagging behind raw material fluctuations due to the large inertia of the furnace body, and realizing the transformation of calcination temperature from passive correction to active suppression.
[0015] 2. This invention divides the calcination zone into multiple sections along the material's downward movement direction and establishes heat balance equations for each section, ensuring that the application location of the feedforward compensation heat corresponds to the actual volatile matter release location. This avoids the problem of misalignment between the compensation application point and the disturbance application point when the exothermic process is treated as a lumped parameter. Simultaneously, it matches differentiated compensation strategies according to the volatile matter fluctuation pattern, aligning the temporal characteristics of the feedforward action with the disturbance characteristics, thus achieving matching of the feedforward compensation with the disturbance distribution in both spatial and temporal dimensions.
[0016] 3. This invention introduces a confidence assessment based on the accuracy of volatile matter estimation, enabling the fusion weights of feedforward compensation and feedback control to be dynamically adjusted according to the reliability of the current information. When the estimation is accurate, the feedforward quickly suppresses disturbances, and when the estimation is inaccurate, the feedback automatically gives way to correction. This ensures that the feedforward compensation remains effective across the entire operating range, leveraging the advantages of rapid feedforward response while using feedback as a safety net. This solves the problem of decreased control quality in conventional feedforward-feedback control when petroleum coke volatile matter information is inaccurate. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the overall process of the intelligent optimization control method for petroleum coke calcination of the present invention; Figure 2 This is a flowchart of step S1 of the present invention; Figure 3 This is a flowchart of step S2 of the present invention; Figure 4 This is an overall structural diagram of the intelligent optimization control system for petroleum coke calcination of the present invention. Detailed Implementation
[0018] 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.
[0019] Currently, the fluctuating volatile matter content of petroleum coke entering the furnace of tank calciners leads to temperature lag and deviation in the calcination zone, resulting in unstable quality of the calcined coke. One objective of this invention is to provide an intelligent optimization control method for petroleum coke calcination. See [link to relevant documentation]. Figure 1 As shown, it includes the following steps: S1. Obtain the near-infrared spectral signal of the petroleum coke currently fed into the furnace and the component signal of the flue gas at the top of the tank, and analyze them to obtain the estimated value of volatile matter in the spectral channel and the estimated value of volatile matter in the process channel, respectively. The volatile matter estimate of the spectral channel is used as the real-time volatile matter estimate. The estimation residual is determined based on the volatile matter estimate of the process channel. The fluctuation characteristics of the real-time volatile matter estimate are extracted and the fluctuation pattern of the current volatile matter is identified. S2. Calculate the disturbance equivalent heat flux based on the difference between the real-time estimated value of volatile matter and the preset baseline volatile matter; The calcination zone is divided into multiple sections along the material moving downwards, and a heat release ratio coefficient is set for each section. The disturbance equivalent heat flow is distributed to each section according to the heat release ratio coefficient of each section to obtain the segmented disturbance heat flow of each section. The current temperature and preset target temperature of each section of the calcination zone are obtained. The corresponding compensation strategy is matched according to the fluctuation mode. The segmented disturbance heat flow and the deviation between the current temperature and the target temperature are used as inputs. The required compensation heat flow of each section is solved in reverse through the segmented heat balance equation. The compensation heat flow of each section is mapped to the feedforward compensation amount of discharge speed, negative pressure in tank and air distribution through the pre-calibrated sensitivity matrix. S3. Calculate the feedforward confidence level based on the estimated residual, calculate the coordination factor based on the feedforward confidence level, the current temperature deviation, and the equivalent heat flow of the disturbance, solve the feedback control quantity using the heat balance equation as the prediction model, and weight and fuse the feedforward compensation quantity and the feedback control quantity according to the coordination factor to output the comprehensive control command.
[0020] This method takes the real-time sensing of volatile matter entering the furnace as its starting point. It identifies the fluctuation pattern of volatile matter and calculates the feedforward compensation amount of discharge rate, negative pressure in the tank and air distribution accordingly. Then, it uses confidence assessment to drive the dynamic coordination of feedforward and feedback. At the same time, it uses the online estimation results of the quality of calcined coke to perform closed-loop correction of model parameters, so that temperature control can respond to changes in raw materials in advance and improve the situation of large inertial system responding lagging to volatile matter fluctuations.
[0021] like Figure 2 As shown, step S1 of the present invention involves obtaining a real-time estimated value of the volatile matter content and its fluctuation characteristics of the current material entering the furnace, providing a data basis for subsequent feedforward compensation. Specifically, it includes the following steps: S101: In this step, multiple types of sensors are deployed at the furnace front silo and the top flue gas duct of the tank calciner to simultaneously collect information on the material entering the furnace and the furnace operating status.
[0022] Specifically, a near-infrared spectrometer is installed at the feed inlet of the furnace hopper in the tank calciner. This near-infrared spectrometer uses a diffuse reflection fiber optic probe, with the probe window positioned close to the material flow path to continuously collect the reflectance spectral signal from the surface of the petroleum coke particles. The near-infrared spectrometer outputs near-infrared spectral signals at 10-second sampling intervals, with a spectral range of 1000 nm to 2500 nm, covering the characteristic absorption bands of hydrocarbon functional groups in petroleum coke.
[0023] Meanwhile, an online flue gas component analyzer is installed on the flue gas exhaust pipe at the top of the calcining furnace. This online flue gas component analyzer has built-in electrochemical and infrared sensors to detect the volume concentration of carbon monoxide, carbon dioxide and oxygen in the flue gas in real time. The sampling interval is also set to 10 seconds to ensure time synchronization with the spectral signal, and finally outputs the flue gas component concentration signal.
[0024] Furthermore, a K-type thermocouple temperature sensor is installed on the exhaust duct at the top of the tank. Its measuring end extends into the duct to collect the real-time value of the exhaust gas temperature at the top of the tank and output the temperature signal at the top of the tank.
[0025] The discharge end of the tank calciner is equipped with a discharge motor, which is driven by a frequency converter. The discharge speed is controlled by adjusting the motor's operating frequency. The current operating frequency of the discharge motor is directly read by the control system through the frequency converter's communication interface and serves as an indirect representation of the discharge speed, i.e., the discharge speed signal.
[0026] Based on this, the near-infrared spectral signal, flue gas component concentration signal, tank top temperature signal and discharge speed signal are transmitted to different channels of the data acquisition card through shielded cables. The data acquisition card performs analog-to-digital conversion on each analog signal and then transmits the digital signals to the control system.
[0027] After receiving the data, the control system aligns all signals according to the timestamp, so that the data group at each moment contains near-infrared spectral data, flue gas composition data, tank top temperature data, and discharge speed data at the same time point.
[0028] Furthermore, outlier removal is performed on the aligned data sets. The removal criteria are as follows: values that exceed the range of each sensor are marked as invalid, and values that deviate from the median by more than three standard deviations in three consecutive data sets are marked as invalid.
[0029] After outlier removal, a moving average filter is applied to the valid data. The width of the moving window is set to 5 minutes, and the arithmetic mean of all valid data within the window is used as the filtered output value at that moment.
[0030] After the above alignment, elimination, and filtering processes, the synchronized data group is obtained. The data set is continuously output in 10-second cycles and includes four types of data: the filtered near-infrared spectral signal, the filtered flue gas component concentration, the filtered tank top temperature, and the filtered discharge speed at the current moment. These data serve as inputs for the subsequent S102 step.
[0031] S102: Obtain synchronization data group According to the synchronized data group A dual-channel parallel estimation and verification / correction strategy is employed to calculate the volatile matter content of the feedstock. This method independently estimates the volatile matter content of the current feedstock through both the spectral and process channels. Based on the current estimate from the spectral channel, the process channel verifies and corrects it.
[0032] Specifically, the spectral channel uses the aforementioned near-infrared spectral signal as input and selects the absorbance of the characteristic band corresponding to the hydrocarbon functional groups in the volatiles as the independent variable to establish a partial least squares regression model. Before formal deployment, this model is calibrated using no fewer than 50 samples with known volatile matter test values. The calibration samples cover the range of volatile matter variations that may actually be encountered in the factory, and the test values are obtained according to the national standard GB / T8726, the ignition method. During model operation, the measured spectral characteristics at the current moment are input, and after regression calculation, the estimated volatile matter values of the spectral channel are output. .
[0033] The process channel uses flue gas component concentration signals, tank top temperature signals, and discharge rate signals as inputs. Because the petroleum coke entering the furnace undergoes several hours of column descent and calcination before its volatiles are released and rise with the flue gas to the tank top, the flue gas component concentration detected at the current moment actually corresponds to... Material batches that entered the furnace before the specified time. Transfer delay. It is determined by the ratio of the material column height to the current discharge speed, that is: ; In the formula, H is the height of the material column; v(t) represents the current discharge speed.
[0034] When performing calculations, the process channel first calculates the transmission delay based on the current discharge speed signal. The current flue gas component concentration and tank top temperature are compared with The estimated spectral channel values or laboratory values from previous times are time-aligned, and then the estimated volatile matter content of the historical batch of materials fed into the furnace is derived by combining a simplified mechanistic model with data-driven correction. .
[0035] The simplified mechanistic model consists of three parts: calculation of total carbon oxides, estimation of volatile matter release, and back calculation of volatile matter content. The first part is the calculation of the total amount of carbon oxides. Specifically, it involves calculating the volume fraction of carbon monoxide detected by the flue gas composition analyzer. and carbon dioxide volume fraction Adding them together, we get the total volume fraction of carbon oxides in the flue gas: ; The second part is the estimation of volatile matter release. Specifically, the amount of carbon moles emitted with the flue gas per unit time is determined by the total integral of carbon oxides, the flue gas volumetric flow rate, and the molar volume: ; In the formula, The flue gas volumetric flow rate is estimated from the discharge velocity and furnace negative pressure through the heat balance relationship. V m This represents the molar volume of the gas.
[0036] Since the carbon oxides in the flue gas come from both volatile matter combustion and coke burn-off, the portion from coke burn-off must be deducted: ; In the formula, The molar amount of carbon lost due to coke burn-off is estimated from the coke burn-off rate and discharge rate.
[0037] The third part involves the reverse calculation of volatile matter content. Specifically, the molar amount of carbon produced by the combustion of volatile matter is converted into the mass of volatile matter, and then divided by the mass of material fed into the furnace per unit time to obtain the mass fraction of volatile matter. ; In the formula, M c is the molar mass of carbon; This refers to the mass fraction of carbon in the volatile matter. The mass of material fed into the furnace per unit time.
[0038] Due to the flue gas volume flow rate in the mechanistic model Volatile carbon mass fraction and the molar amount of carbon lost from coke burning Since precise data is difficult to obtain, a data-driven correction method is used to adjust the output of the mechanistic model. Specifically, the concentration of flue gas components, discharge rate, tank top temperature, and corresponding volatile matter content from historical production records are used as training samples. The least squares method is used to fit and obtain correction coefficients k1 and k2. The corrected model output is as follows: ; In the formula, This is the output of the calculation for the mechanism.
[0039] This channel outputs the time-reversed and aligned volatile content estimate, which corresponds to... The batch of materials that were put into the furnace before the specified time.
[0040] Furthermore, since the spectral channel directly measures the material currently fed into the furnace and can provide a real-time estimate of the volatile matter at the current moment, while the process channel outputs a historical batch estimate after time inversion alignment, the two channels are inconsistent in terms of time reference. Therefore, the estimated values of the two channels are not directly weighted and averaged. Instead, the current estimated value of the spectral channel is used as the basis for verification and correction by the process channel.
[0041] Specifically, the V output from the process channel is compared with the estimated value of the spectral channel at the corresponding historical moment. By comparison, the estimated residuals of the two channels are calculated: ; The estimated residual reflects the spectral channels in The accuracy of the estimation before the time is important; the smaller the residual, the higher the reliability of the estimated values for the spectral channels; the larger the residual, the more likely there is a systematic bias in the spectral channels.
[0042] After the above processing, the real-time estimated value of volatile matter at the current moment is obtained. and estimated residuals This serves as the input for the subsequent S103 step.
[0043] S103: Real-time estimate of volatile matter based on step S102 The system extracts fluctuation characteristics and classifies patterns from the estimated volatile matter values for the current moment and the preceding 30 minutes. Since this estimate is based on the real-time output of the spectral channels, its time reference is consistent with the current material entering the furnace, and it can be directly used to characterize the fluctuation state of the volatile matter entering the furnace at the current moment.
[0044] Specifically, taking the current time t as the endpoint, all estimated values within the past 30 minutes are extracted to form a sliding window sequence: ; Wherein, window width T w=30 minutes, Δt=10 seconds is the sampling interval. This window slides forward cycle by cycle as the current moment progresses, always keeping the data from the most recent 30 minutes.
[0045] Furthermore, time-domain and frequency-domain features are calculated for the sequence data within the window: ① The first-order rate of change dV / dt is obtained by dividing the change in the estimated value between adjacent time points within the first-order difference calculation window by the sampling interval. This rate of change is used to characterize the speed of volatile matter fluctuation. ② By further differentiating the first-order rate of change, we obtain the second-order rate of change. It is used to characterize wave acceleration.
[0046] ③ Calculate the standard deviation of all estimated values within the window. with the mean The ratio of the two values is used to obtain the coefficient of variation. , used to characterize the degree of dispersion of fluctuations relative to the mean; ④ Perform a Fast Fourier Transform on the window sequence to convert the time-domain signal to the frequency domain, and extract the frequency component with the largest amplitude as the dominant frequency. It is used to determine whether a dominant cycle exists in the fluctuation.
[0047] Based on this, and using the four characteristic quantities mentioned above—first-order rate of change, second-order rate of change, coefficient of variation, and dominant frequency component—the current volatile matter fluctuations are divided into four modes through threshold criteria.
[0048] When the absolute value of the first-order rate of change The rate of change exceeds the first threshold δ1 and the duration of this high rate of change is less than [a certain value]. When this occurs, it is determined to be a step-type fluctuation, which corresponds to rapid changes caused by a sudden switch in raw material sources; when Between the second threshold δ2 and the first threshold δ1 and Maintaining the same number for more than When the frequency is determined to be a gradual fluctuation, this pattern corresponds to a continuous unidirectional change caused by the slow drift of the upstream delayed coking process; when the main frequency When the power spectral density exceeds the third threshold δ3 and the coefficient of variation CV is lower than the step-type judgment threshold, it is judged as a periodic oscillation. This mode corresponds to periodic fluctuations caused by regular changes in the incoming material ratio; when When the fluctuation does not exceed the second threshold δ2 and the coefficient of variation CV is lower than the fourth threshold δ4, it is determined to be in a steady state, indicating that the current fluctuation range is within the allowable range and there is no need to trigger feedforward compensation.
[0049] It should be noted that the above threshold values are determined based on the thermal response characteristics of the pot furnace and on-site operating experience, and are adjusted by the operators according to the actual furnace conditions.
[0050] After the pattern recognition described above, the fluctuation pattern category M(t) and fluctuation feature vector F(t) at the current moment are output to step S2, where These correspond to four modes: step type, gradual change type, periodic oscillation type, and steady state type, respectively. It includes four numerical quantities: first-order rate of change, second-order rate of change, coefficient of variation, and dominant frequency component. The fluctuation mode category M(t) and fluctuation eigenvector F(t) serve as the characteristic basis for subsequent feedforward compensation strategy selection and compensation amount calculation.
[0051] like Figure 3 As shown, step S2 of this invention is volatile matter disturbance quantification and segmented feedforward heat compensation. Its purpose is to calculate the feedforward compensation amount that can offset the influence of volatile matter disturbance in advance based on the real-time estimated value of volatile matter, fluctuation mode category and fluctuation feature vector obtained in S1, so as to provide feedforward input for the coordinated control in the subsequent step S3. Specifically, it includes the following steps: S201: This step first establishes a description of the release distribution of volatiles in different sections of the calcination zone. Petroleum coke volatiles are not released in a concentrated manner at a fixed location in the calcination zone, but rather gradually precipitate in different temperature ranges as the material layer moves downward.
[0052] The preheating zone has a temperature of approximately 400℃ to 700℃, during which light volatiles are mainly released; the low-temperature volatiles zone has a temperature of approximately 700℃ to 850℃ and the high-temperature volatiles zone has a temperature of approximately 850℃ to 1050℃, corresponding to the release of pyrolysis products with different molecular structures; the homogenization zone has a temperature of approximately 1050℃ to 1250℃, during which the remaining volatiles are finally released.
[0053] Accordingly, the calcination zone is divided into four sections along the downward movement direction of the material layer, and each section is assigned a corresponding heat release ratio coefficient, forming a heat release weight vector: ; Among them, w i Let be the heat release ratio of the volatile matter in the i-th segment, satisfying .
[0054] The coefficients for each segment were calibrated by petroleum coke thermogravimetric analysis experiments and stored in the control system database.
[0055] Based on this, the equivalent heat flux caused by the current volatile matter content relative to the baseline value is calculated to characterize the impact of volatile matter fluctuations on the furnace thermal balance: Let the current discharge rate be v(t), and the real-time estimated volatile matter content be... The baseline volatile matter content The baseline volatile matter value is taken from the furnace design conditions, and the average calorific value of the volatile matter is... The actual volatile heat flow and reference heat flow They are respectively: ; ; The difference between the two is the equivalent heat flux of the disturbance at the current moment: ; when A positive value indicates that the volatile content is too high and the heat release in the furnace is excessive; a negative value indicates that the volatile content is too low and the heat release is insufficient.
[0056] Furthermore, the equivalent heat flux of the disturbance is distributed to each segment according to a segmented heat release weight vector. Simultaneously, a release time distribution function is introduced. The transport delay and diffusion distribution of volatiles from their entry into the furnace to their release in the i-th section are characterized by this function, which is determined by the material bed descent velocity and the inter-section distance. Its peak position corresponds to the average transport delay of volatiles reaching the i-th section. Then the segmented perturbation heat flux of the i-th segment is: ; In the formula, The integral over the time axis is normalized to 1, that is... This indicates that all the heat from the disturbance will eventually be released.
[0057] Based on the above calculations, the output disturbance equivalent heat flux is obtained. And a four-segment segmented perturbation heat flux sequence: .
[0058] S202: Based on the fluctuation mode category M(t) output in step S1, match the corresponding feedforward compensation strategy. Since different fluctuation modes have different impact characteristics and propagation paths on the calcination zone temperature, matching differentiated compensation strategies can align the time and frequency domain characteristics of the feedforward action with the disturbance characteristics, avoiding the failure of a single strategy under certain operating conditions or the introduction of unnecessary adjustment actions.
[0059] Specifically, when the fluctuation mode is in steady state, it is determined that the current volatile fluctuation amplitude is within the tolerance range. Feedforward compensation will not produce substantial improvement and may even introduce unnecessary actuator actions. Therefore, feedforward is not activated, and operation is maintained only by feedback control in the subsequent step S3.
[0060] When the fluctuation pattern is step-type, a two-stage compensation strategy is adopted. Step disturbances have large amplitudes and strong impacts, requiring a distinction between rapid response and steady-state elimination. The rapid response stage acts on the negative pressure inside the tank and the air volume distribution; the response time of these actuators is on the order of seconds, enabling them to quickly absorb the disturbance impact. The steady-state correction stage acts on the discharge velocity; the response time of this actuator is on the order of minutes to hours, used to gradually eliminate accumulated thermal deviations. The transient compensation amount of the two-stage strategy is determined by multiplying the disturbance amplitude by the air volume-temperature channel gain, while the steady-state correction amount is determined by integrating the disturbance heat flux and then multiplying it by the discharge velocity channel gain.
[0061] When the fluctuation pattern is gradual, a predictive feedforward strategy is adopted. The trend of gradual fluctuation is extrapolable; the trend of volatile matter change over several control cycles is predicted based on the current first-order rate of change, and compensation is applied in advance based on the predicted values. The prediction step size is determined by the feedstock update cycle.
[0062] When the fluctuation pattern is periodic oscillation, a filter feedforward strategy is adopted. Bandpass filtering is applied to the disturbed heat flux to retain the effective fluctuation components near the dominant frequency, while high-frequency noise exceeding the actuator's response capability is not compensated to avoid frequent actuator operation.
[0063] After the above strategy matching, the strategy flag corresponding to the current moment is output for subsequent S203 calls.
[0064] S203: Based on the segmented perturbation heat flow sequence obtained in step S201 and the strategy flag determined in step S202, run the segmented dynamic heat compensation algorithm to calculate the feedforward compensation heat flow required for each segment and map it as the compensation amount of the operation variable.
[0065] Specifically, the heat balance equations for each section of the calcination zone are first written.
[0066] For the i-th segment, the thermal equilibrium relationship is: ; In the formula, C i Let i be the heat capacity of the i-th element; Let be the current temperature of the i-th segment; The actual heat release of volatiles in the i-th segment includes the baseline and the perturbation components. The feedforward compensation heat flux to be determined; Let be the heat loss of the i-th segment; This represents the endothermic reaction of the i-th calcination stage.
[0067] Heat loss is calculated as the product of the heat dissipation coefficient and the temperature difference, i.e. ,in Let be the heat dissipation coefficient of the i-th segment. The ambient temperature; The endothermic reaction is calculated as the product of the enthalpy change and the reaction rate, i.e. ,in For the enthalpy change of the calcination reaction, denoted as , where is the reaction rate at temperature i.
[0068] Furthermore, the temperature of each segment is made to converge towards the target value at a desired rate, i.e., the rate of change of the temperature of each segment is set as follows: ; In the formula, The target temperature for the i-th segment; The desired convergence time constant is one-third of the material column update cycle.
[0069] Substituting the above equation into the heat balance equation, we can solve for the required feedforward compensation heat flux for the i-th segment: ; The physical meaning of this formula is: when a certain temperature is lower than the target value, additional compensating heat needs to be injected to increase the heating rate; conversely, the compensating heat needs to be reduced and maintained by the material's own heat.
[0070] Furthermore, the compensated heat flux calculated in the four segments needs to be mapped to three executable operational variables: discharge velocity, tank negative pressure, and air distribution volume. The sensitivity matrix S is pre-calibrated through a step response experiment; this sensitivity matrix is a 4×3 matrix, where the element in the i-th row and j-th column is S. ij This represents the gain of the effect of a unit change in the j-th operand on the temperature of the i-th segment.
[0071] It should be noted that the calibration method is as follows: small step excitations are applied to the discharge velocity, tank negative pressure, and air distribution volume respectively, and the steady-state temperature response changes at each stage are recorded. The ratio of the response change to the excitation amplitude is used as the corresponding matrix element. The relationship between the piecewise compensated heat flux vector and the compensated amount of the manipulated variable is as follows: ; in, This is the feedforward compensation amount for the discharge speed; This is the feedforward compensation amount for negative pressure inside the tank; This is the feedforward compensation amount for the air volume distribution.
[0072] Based on this, the above mapping relationship is solved using the regularized least squares method: ; in, This is a regularization coefficient used to prevent the compensation amount of the solved operation variables from exceeding the limit due to ill-conditioning of the matrix; I is a 3×3 identity matrix.
[0073] After solving, apply engineering constraints to the compensation amount: ; ; Where v min and v max These are the upper and lower limits of the discharge speed, respectively; Δv max This represents the maximum permissible rate of change for a single adjustment.
[0074] The amplitude constraints and rate of change constraints for the negative pressure inside the tank and the air distribution volume are set in the same way.
[0075] Simultaneously, the expected temperature trajectory is generated by the forward integration of the piecewise heat balance equation, and each segment in the prediction time domain N p Expected temperature sequence within The expression is: ; in, .
[0076] This predicted temperature trajectory reflects the expected temperature change path of each segment under feedforward compensation. Together they form the multivariable expected temperature trajectory vector This serves as a reference for calculating the feedforward compensation amount and determining the coordination factor in subsequent step S3.
[0077] Furthermore, the output is a piecewise feedforward compensation sequence. and expected temperature trajectory Proceed to step S3.
[0078] Step S3 of the present invention involves dynamically adjusting the weights of the feedforward and feedback based on the reliability of the feedforward compensation amount, and then weighting and fusing the two to output a comprehensive control command. Specifically, this includes the following steps: S301: The reliability of feedforward compensation depends on the accuracy of volatile matter estimation. If the volatile matter estimation has a large deviation, the feedforward compensation amount calculated based on the estimation value may worsen the furnace condition. Therefore, before introducing the feedforward amount into the control, its confidence level needs to be quantitatively assessed.
[0079] Specifically, the feedforward confidence level is determined based on the estimated residuals. The feedforward confidence level is obtained by weighted summation of four indicators: estimated residuals, sensor health status, cumulative error in volatile matter prediction, and stability of fluctuation patterns. Among these, the estimated residuals... This reflects the transmission delay of the spectral channel. The accuracy of previous estimates is important; the smaller the residual, the more reliable the estimated values for the spectral channels, and the higher the feedforward confidence level.
[0080] It should be noted that sensor health status indicators Outputted by the self-test modules of each sensor, this indicator decreases when a sensor malfunctions or the signal exceeds the effective range; cumulative error in volatile matter prediction. Record the cumulative deviation between soft-measured values and subsequent laboratory values of volatile matter over a period of time to reflect whether there is systematic drift in the soft-measured model. Fluctuation pattern stability index. This characterizes whether the fluctuation pattern changes frequently within the recent window. If the pattern switches frequently, it indicates that the reliability of the pattern recognition result is reduced.
[0081] The above four indicators are weighted and combined to obtain the feedforward confidence level c(t) at the current time, which is calculated by the following formula: ; In the formula, For each weight coefficient, satisfying The values are set by the on-site operators based on the actual impact of each factor on the quality control. and These are the normalized upper limit thresholds for the estimated residuals and cumulative errors, respectively. When the actual value exceeds these thresholds, the contribution of the corresponding term drops to zero. The value of the feedforward confidence level c(t) ranges from 0 to 1. The closer c(t) is to 1, the more reliable the current feedforward information is. The closer c(t) is to 0, the lower the reliability of the feedforward information is, and the feedforward weight should be reduced.
[0082] After the above calculation, the feedforward confidence level c(t) at the current time is output for subsequent step S302 to call.
[0083] S302: Based on the feedforward confidence level c(t) obtained in step S301, combined with the current temperature deviation... The disturbance equivalent heat flux output in step S201 Dynamically calculate the feedforward-feedback coordination factor .
[0084] It should be noted that the design logic of the coordination factor is as follows: the higher the feedforward confidence level, the greater the amplitude of the disturbance equivalent heat flux. The larger the current temperature deviation, the smaller the current temperature deviation should be, and the greater the weight of the feedforward control should be. Conversely, if the current temperature deviation is already quite significant, it indicates that the feedforward compensation may not have accurately offset the disturbance or there may be unmodeled factors. In this case, the feedforward weight should be reduced and the feedback correction force should be increased.
[0085] Coordinating Factor The calculation formula is: ; in It is an sigmoid function, and its expression is: This is used to smoothly map the input to the 0 to 1 range; and These are the scale factors for the amplitude of the perturbation equivalent heat flux and the temperature deviation, respectively, used to normalize the actual engineering values to the effective range of the S-shaped function.
[0086] ,when When the value approaches 1, the feedforward behavior is dominant; when it approaches 0, the feedback behavior is dominant.
[0087] Based on the above calculations, the coordination factor at the current moment is output. This is for use in subsequent S303 steps.
[0088] S303: Feedforward compensation sequence based on the output of step S203 Expected temperature trajectory and the coordination factor output in step S302 The system performs rolling time-domain optimization to solve for the feedback control quantity, and then outputs a comprehensive control command by weighting and fusing the feedforward compensation quantity and the feedback control quantity according to the coordination factor.
[0089] First, a rolling optimization problem for the feedback control quantity is constructed: the piecewise heat balance equation established in step S203 is used as the internal prediction model, with the target temperature of each section as the internal prediction model. To track the target, in the prediction time domain N p and control time domain N c Minimize the following objective function: ; The first item is the temperature tracking item. The predicted temperature value is obtained by forward integration based on the piecewise heat balance equation. The first term is the target temperature for each section; the second term is the control smoothing term. The first term represents the change in the feedback control quantity between adjacent time points; this term is used to suppress frequent actions of the actuator. The third term is a quality penalty term. ,in This section is designed to predict the quality indicators of calcined coke based on the current control sequence. It incorporates product quality into the control optimization objectives, ensuring that control actions always serve the final quality goals.
[0090] It should be noted that α, β, and γ are weighting coefficients for each item, and their values are determined based on the control priority of the pot furnace: α corresponds to the weight of temperature tracking, reflecting the degree of impact on the quality of calcined coke when the calcination temperature deviates from the target value, and its value is usually between 0.5 and 2.0. The more sensitive the temperature is to quality, the larger the value should be; β corresponds to the weight of the smoothness of the actuator's operation, used to suppress frequent fluctuations in discharge speed, negative pressure, and air distribution, and its value is usually between 0.1 and 0.5; γ corresponds to the weight of the quality penalty item, which is dynamically adjusted by the quality deviation feedback in step S4. When the soft measurement quality deviates from the target value, γ increases to strengthen quality constraints, and its base value is set between 0.2 and 1.0. The specific values of each weight are adjusted by the on-site operators according to the actual furnace conditions.
[0091] The constraints of this rolling optimization problem include: magnitude constraints of the manipulated variables. , , and rate of change constraints , , The quality index is treated as a soft constraint; when the predicted quality exceeds the allowable range, γ is increased to bring it back to the feasible region. This problem is solved using a quadratic programming method, taking the first term of the control sequence as the feedback control quantity at the current time step. .
[0092] Furthermore, the feedforward compensation and feedback control quantities are weighted and fused according to a coordination factor to obtain the comprehensive control command: ; in This is the feedforward compensation amount. This is the feedback control variable. Final output. As a comprehensive control command for the current moment, including adjustments to the discharge speed, internal negative pressure, and air distribution volume, it is issued to the corresponding actuators. Simultaneously, the historical record of this control command is available for subsequent step S4 to recall.
[0093] Step S4 of this invention is the prediction of calcined coke quality and model self-learning correction. Its purpose is to use the large-delay feedback information of calcined coke quality detection to correct the parameters of the soft measurement model in S1 and the control model in S3 online, so that the control system maintains its performance without degradation during long-term operation. Specifically, it includes the following steps: S401: Based on the historical records of integrated control commands output by S303 and the process data synchronously collected by the control system, an online soft measurement model of the true density of calcined coke is established to estimate the quality level of the current calcined coke produced before the laboratory test results are transmitted back.
[0094] Specifically, based on the historical temperature sequence of each segment of the calcination zone. Historical discharge speed History of volatile matter entering the furnace and the residence time of materials in the furnace As an input feature. Dwell time The ratio of the current discharge rate to the height of the material column determines the average time the material takes from entering the furnace to being discharged. The output of the soft-sensor model is an estimate of the true density of the calcined coke. .
[0095] It should be noted that this soft sensing model is built using the support vector regression method, and the radial basis function kernel is selected. Before formal deployment, the model is trained with no fewer than 100 sets of historical production data samples. The training samples include corresponding process data and laboratory-measured true density values, which are obtained according to industry standard methods. During model operation, the temperature, discharge rate, volatile matter content in the furnace, and residence time collected at each stage are input into the model. After regression calculation, the model outputs an online estimate of the true density of the currently produced calcined coke.
[0096] Based on the above calculations, output the estimated true density of the calcined coke at the current moment. .
[0097] S402: Based on the online estimate of the true density of calcined coke obtained in S401, the quality deviation feedback and model correction logic is activated after the laboratory sends back the actual test values of samples produced in the same time period. Due to the inherent delay in laboratory analysis, the actual test values are usually obtained with a lag of 2 to 4 hours. Therefore, it is necessary to align the lag time before comparison.
[0098] Let the current time be t, and the detection value returned by the laboratory be... ,in This detection value corresponds to the laboratory analysis delay time. Samples produced at any given time. With step S401 Soft measurement estimate of output at time step Compare and calculate the quality deviation: ; If the absolute value of Δρ continuously exceeds a set threshold, the model correction logic is triggered. Backtracking then occurs. Estimated volatile matter before and after time point The source of the quality deviation was determined by combining the subsequent retest of the volatile matter content of the batch of materials fed into the furnace.
[0099] like The deviation between the measured values and the post-tested values exceeded the set range, indicating that the quality deviation stemmed from the estimation error of the S102 soft measurement model. Therefore, the model parameters for the spectral and process channels in S102 were corrected. Specifically, the post-tested values, along with the corresponding process data, were added to the calibration dataset as new samples. The partial least squares regression model for the spectral channel and the correction coefficients for the process channel were then refitted, allowing the soft measurement model to gradually approximate the actual volatile matter content. Simultaneously, this deviation was recorded in the cumulative error. This is for use in the subsequent confidence assessment step S301.
[0100] like The results are basically consistent with the retested values, but the quality still shows a deviation, indicating that the quality deviation stems from an unsatisfactory control effect. Therefore, the weight coefficient γ of the rolling optimization objective function in step S303 is corrected. When Δρ is positive, it indicates that the true density is too high, and there is a risk of over-calcination. In this case, γ is increased to strengthen the quality constraint, making the controller pay more attention to the quality index in subsequent optimizations. When Δρ is negative, it indicates that the true density is too low, and there is a risk of under-calcination. In this case, γ is also increased, and the expected convergence time constant in S203 is appropriately reduced. This improves the response speed of temperature control, allowing the furnace temperature to return to the target value more quickly.
[0101] After the above attribution analysis and parameter correction, the updated spectral channel model parameters and process channel model parameters are fed back to step S102, and the updated weight coefficient of the quality penalty term is fed back to step S303, thus completing the closed-loop correction of the quality deviation on the upstream model parameters.
[0102] S403: Based on the quality deviation feedback and model correction completed in step S402, the overall operating performance of the control system is evaluated periodically. Based on this evaluation, the confidence weight coefficient in step S301 and the scaling factor in step S302 are adaptively adjusted. The performance evaluation cycle is set to be executed once every 24 hours, running independently in the background without interfering with the real-time control process from S1 to S3.
[0103] The calculation method for the comprehensive performance index PI is as follows: ; In the formula, This represents the variance of the temperature fluctuation in the calcination zone over the past 24 hours, reflecting the stability of the temperature control. E represents the energy consumption per unit of product within the same time period, reflecting the economic efficiency of operation; The pass rate of calcined coke reflects the effectiveness of quality control.
[0104] For each weighting factor, its value is determined based on the factory's actual management priorities regarding temperature stability, energy consumption, and quality. When the factory prioritizes product quality, w... Q The value is relatively large, usually set between 0.5 and 0.7; when energy conservation and consumption reduction are the main objectives, w E The value is relatively large, usually set between 0.4 and 0.6; temperature stability is a fundamental indicator for ensuring production safety. T The weights are typically no less than 0.2. The specific values for each weight are determined by the operator within the above reference range based on the actual furnace conditions and management objectives, and remain fixed within the same evaluation cycle.
[0105] When the overall performance index PI exceeds the set threshold, the parameter self-learning logic is triggered to adjust the following three types of parameters.
[0106] The first type consists of the weighting coefficients of the confidence assessment function. By comparing the correlation between each input factor and the feedforward compensation bias within a past evaluation period, the weight of significantly correlated factors is increased, while the weight of insignificantly correlated factors is decreased, making the confidence assessment more closely reflect current actual operating conditions. The input factors include four items: estimation residual, sensor health, cumulative error, and mode stability.
[0107] The second category is the scaling factor in the calculation of the coordination factor. and The values of the coordination factor are re-determined based on the statistical distribution characteristics of recent disturbance amplitudes and temperature deviations, so that the coordination factor maintains a reasonable distribution range under different operating conditions and avoids extreme values that could lead to complete failure of feedforward or feedback.
[0108] The third category involves the control weights α, β, and γ in the rolling optimization objective function. The need to adjust these weights is determined based on the statistical characteristics of recent quality deviations. If the quality deviation consistently leans to one side, the weight γ of the quality penalty term is increased; if the actuator operates too frequently, the weight β of the control smoothing term is appropriately increased; if temperature tracking deviation is the main cause of quality fluctuations, the weight α of the temperature tracking term is increased.
[0109] After the above performance evaluation and parameter self-learning, the updated confidence weight coefficients, scaling factors and control weights are fed back to S301 and S303 respectively, so that the control system can maintain optimal control performance in long-term operation.
[0110] The second objective of this invention is to provide an intelligent optimization control system for petroleum coke calcination, see [link to relevant documentation]. Figure 4 As shown, it includes a data acquisition and synchronization module, a volatile matter sensing and pattern recognition module, a segmented feedforward compensation calculation module, a coordinated optimization control module, a quality feedback self-learning module, and a human-computer interaction and parameter tuning module.
[0111] The data acquisition and synchronization module performs the functions described in step S101 above. This module communicates with the near-infrared spectrometer, the online flue gas component analyzer, the thermocouple temperature sensor on the tank top, and the frequency converter of the discharge motor. It synchronously acquires near-infrared spectral signals, carbon monoxide concentration, carbon dioxide concentration, oxygen concentration in the flue gas, the flue gas temperature on the tank top, and the operating frequency of the discharge motor at a 10-second sampling cycle. The data is then aggregated by the data acquisition card and transmitted to the control system. The module includes a data alignment unit to synchronize multi-source signals according to timestamps; an outlier removal unit to mark and remove values that exceed the measurement range or deviate from the median by more than three standard deviations; and a sliding filter unit to perform sliding mean filtering on valid data with a 5-minute window width, outputting a synchronized data set.
[0112] The volatile matter sensing and pattern recognition module is used to perform the functions of steps S102 and S103 mentioned above. This module internally has a dual-channel fusion soft measurement unit, including a spectral channel and a process channel. The spectral channel takes the aforementioned near-infrared spectral signal as input and outputs an estimated value of volatile matter through a partial least squares regression model. This estimated value directly reflects the volatile matter level of the material entering the furnace at the current moment. The process channel takes the flue gas component concentration signal, the tank top temperature signal, and the discharge rate signal as input. It first calculates the transmission delay based on the discharge rate signal, then performs time inversion alignment of the current flue gas component concentration and tank top temperature with the material information before the transmission delay, and finally outputs an estimated value of volatile matter in the process channel through a combination of a simplified mechanistic model and data-driven correction. This estimated value corresponds to the batch of material entering the furnace before the transmission delay. Because the spectral channel and the process channel are not aligned on the same time base, the module outputs the current estimate of the volatile matter as the real-time estimate. Simultaneously, it compares the output value of the process channel with the estimate of the spectral channel before the transmission delay, using the deviation to determine the estimation residual. This residual is then used as one of the input parameters for subsequent confidence assessment. The module also includes a fluctuation pattern recognition unit that extracts four features from the volatile matter sequence—first-order rate of change, second-order rate of change, coefficient of variation, and dominant frequency component—with a 30-minute window. Using threshold criteria, the fluctuations are classified into four modes: step-type, gradual-type, periodic oscillation-type, and steady-state, outputting the fluctuation pattern category and fluctuation feature vector.
[0113] The segmented feedforward compensation calculation module performs the functions described in steps S201 to S203 above. This module contains a disturbance equivalent heat flux calculation unit, storing the average calorific value of volatiles, the baseline volatile content, and the segmented heat release weight vector. It calculates the disturbance equivalent heat flux by multiplying the difference between the current estimated volatile content and the baseline value by the discharge rate and the average calorific value of volatiles. This flux is then distributed to each segment according to the segmented heat release weight vector and the release time distribution function, outputting a segmented disturbance heat flux sequence. The module also includes a strategy selection and segmented compensation calculation unit. It matches the corresponding compensation strategy based on the fluctuation mode category and runs a segmented dynamic heat compensation algorithm. It writes the heat balance equation for each segment, solves for the required feedforward compensation heat flux for each segment at the desired convergence rate, and maps the four segments of compensation heat flux to the compensation amounts of three operational variables: discharge rate, tank negative pressure, and air distribution volume, while applying amplitude and rate-of-change constraints. Finally, it outputs the segmented feedforward compensation sequence and the expected temperature trajectory.
[0114] The coordinated optimization control module performs the functions described in steps S301 to S303. This module includes a confidence assessment unit that calculates the feedforward confidence based on four weighted indicators: volatile matter estimation residual, sensor health status, volatile matter prediction cumulative error, and fluctuation mode stability. It also includes a coordination factor calculation unit that takes the feedforward confidence, current temperature deviation, and disturbance equivalent heat flux amplitude as inputs, smooths these three values to the 0-1 interval using a sigmoid function, multiplies them, and outputs the coordination factor. Finally, the module includes a rolling time-domain optimization solution unit that uses the piecewise heat balance equation as its internal prediction model and the target temperature as its tracking target. It constructs an objective function containing temperature tracking, control smoothing, and quality penalty terms. Under constraints on the magnitude and rate of change of the manipulated variables, it uses a quadratic programming method to solve for the optimal feedback control sequence. Then, it weights and fuses the feedforward compensation and feedback control quantities according to the coordination factor, outputting a comprehensive control command that includes adjustments to the discharge speed, tank negative pressure, and air distribution volume.
[0115] The quality feedback self-learning module performs the functions described in steps S401 to S403 above. This module includes a quality soft sensing unit that takes the historical temperature of each section of the calcination zone, the historical discharge rate, the historical volatile matter content in the furnace, and the material residence time as input. It estimates the true density of the calcined coke online using a support vector regression model and outputs the true density estimate. The module also includes a deviation attribution and correction unit. When laboratory measured values are returned, they are compared with the corresponding soft sensing estimates to calculate the quality deviation. Based on the source of the quality deviation, the soft sensing model parameters or the weighting coefficients of the objective function are adjusted accordingly. Finally, the module includes a performance evaluation and self-learning unit that calculates comprehensive performance indicators, including temperature fluctuation variance, unit product energy consumption, and quality pass rate, over a 24-hour period. When performance indicators exceed thresholds, the confidence weighting coefficient, coordination factor scaling factor, and control weights are automatically adjusted.
[0116] The human-machine interaction and parameter tuning module is used to realize information exchange between the operator and the control system. This module has a parameter configuration interface, allowing operators to set and adjust adjustable parameters such as the baseline volatile matter content, components of the segmented exothermic weight vector, expected convergence time constant, regularization coefficient, amplitude and rate of change constraints for each actuator, confidence weight coefficient, coordination factor scaling factor, control weight of the optimization objective function, and weight of the comprehensive performance index. The module also has a data monitoring interface, which displays in real time key operating variables such as the current estimated volatile matter value, fluctuation mode type, feedforward confidence, coordination factor, temperature of each segment, and comprehensive control commands, and records and stores historical data for operator query and analysis.
[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for intelligent optimization control of petroleum coke calcination, characterized in that, Includes the following steps: S1. Obtain the near-infrared spectral signal of the petroleum coke currently fed into the furnace and the component signal of the flue gas at the top of the tank, and analyze them to obtain the estimated value of volatile matter in the spectral channel and the estimated value of volatile matter in the process channel, respectively. The volatile matter estimate of the spectral channel is used as the real-time volatile matter estimate. The estimation residual is determined based on the volatile matter estimate of the process channel. The fluctuation characteristics of the real-time volatile matter estimate are extracted and the fluctuation pattern of the current volatile matter is identified. S2. Calculate the disturbance equivalent heat flux based on the difference between the real-time estimated value of volatile matter and the preset baseline volatile matter; The calcination zone is divided into multiple sections along the material moving downwards, and a heat release ratio coefficient is set for each section. The disturbance equivalent heat flow is distributed to each section according to the heat release ratio coefficient of each section to obtain the segmented disturbance heat flow of each section. The current temperature and preset target temperature of each section of the calcination zone are obtained. The corresponding compensation strategy is matched according to the fluctuation mode. The segmented disturbance heat flow and the deviation between the current temperature and the target temperature are used as inputs. The required compensation heat flow of each section is solved in reverse through the segmented heat balance equation. The compensation heat flow of each section is mapped to the feedforward compensation amount of discharge speed, negative pressure in tank and air distribution through the pre-calibrated sensitivity matrix. S3. Calculate the feedforward confidence level based on the estimated residual, calculate the coordination factor based on the feedforward confidence level, the current temperature deviation, and the equivalent heat flow of the disturbance, solve the feedback control quantity using the heat balance equation as the prediction model, and weight and fuse the feedforward compensation quantity and the feedback control quantity according to the coordination factor to output the comprehensive control command.
2. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The flue gas composition signal at the top of the tank includes carbon monoxide concentration, carbon dioxide concentration, and oxygen concentration. The estimated value of volatile matter in the spectral channel is obtained analytically through a partial least squares regression model, which uses the absorbance of the characteristic band corresponding to the hydrocarbon functional groups in the volatile matter as the independent variable. The analytical method for the estimated value of volatile matter in the process channel is as follows: the transmission delay is calculated based on the discharge rate signal, and the flue gas composition signal at the top of the tank is time-reversed and aligned with the material information before the transmission delay. The result is obtained by combining a simplified mechanism model with data-driven correction. The simplified mechanism model includes three parts: calculation of total carbon oxides, estimation of volatile matter release, and back-calculation of volatile matter content.
3. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The fluctuation characteristics include first-order rate of change, second-order rate of change, coefficient of variation, and dominant frequency component. The method for identifying the fluctuation characteristics is as follows: when the absolute value of the first-order rate of change exceeds the first threshold and the duration is less than the first duration threshold, it is determined to be a step-type mode. When the absolute value of the first-order rate of change is between the second threshold and the first threshold, and the first-order rate of change continues to exceed the second duration threshold with the same sign, it is determined to be a gradual change mode. When the power spectral density of the main frequency component exceeds the third threshold and the coefficient of variation is lower than the step-type determination threshold, it is determined to be a periodic oscillation mode. When the absolute value of the first-order rate of change does not exceed the second threshold and the coefficient of variation is lower than the fourth threshold, it is determined to be a steady-state mode.
4. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The discharge speed signal is obtained by reading the current operating frequency of the discharge motor through the control system. The current volatile heat flux is calculated based on the discharge speed signal and the real-time estimated value of volatile matter. The reference heat flux is calculated based on the discharge speed signal and the reference volatile matter. The difference between the current volatile heat flux and the reference heat flux is used as the disturbance equivalent heat flux. The disturbance equivalent heat flux is used to characterize the change in furnace heat power caused by the deviation of the current volatile matter entering the furnace relative to the reference value.
5. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The segmented heat balance equation is to write the heat balance relationship for each segment separately. The heat balance relationship of each segment includes the actual heat release term of volatiles, the compensating heat flow term, the heat dissipation loss term, and the reaction endothermic term. The heat dissipation loss term is determined by the product of the heat dissipation coefficient and the temperature difference, and the reaction endothermic term is determined by the product of the reaction enthalpy change and the reaction rate. The temperature of each segment is allowed to approach the target temperature of each segment at a desired convergence rate, and the compensating heat flow required for each segment is solved in reverse.
6. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The rows of the sensitivity matrix correspond to each section of the calcination zone, and the columns correspond to three operational variables: discharge velocity, tank negative pressure, and air distribution volume. Each element in the sensitivity matrix is calibrated through a step response experiment. Specifically, a small step excitation is applied to the discharge velocity, tank negative pressure, and air distribution volume, and the steady-state response change of each section's temperature after the step excitation is recorded. The ratio of the steady-state response change to the step excitation amplitude is used as the sensitivity coefficient of the operational variable for that section. A mapping relationship is established between the compensation heat flux of each section and the compensation amount of the three operational variables through the sensitivity matrix. The mapping relationship is solved using the regularized least squares method to obtain the compensation amount of each operational variable. After solving, amplitude constraints and rate of change constraints are applied to the compensation amount.
7. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The estimated residual is determined by the deviation between the estimated volatile matter value of the spectral channel before the transmission delay and the estimated volatile matter value output by the process channel after time inversion alignment. The transmission delay is determined by the ratio of the column height to the current discharge speed.
8. The intelligent optimization control method for petroleum coke calcination according to claim 1, characterized in that: The feedback control quantity is solved using a rolling time-domain optimization method. The piecewise thermal balance equation is used as the prediction model, the target temperature is used as the tracking target, and the optimal control sequence is solved under the constraints of the magnitude and rate of change of the manipulated variables. The first term of the sequence is taken as the feedback control quantity at the current moment.
9. The intelligent optimization control method for petroleum coke calcination according to claim 8, characterized in that: The objective function of the rolling time-domain optimization includes a temperature tracking term, a control smoothing term, and a quality penalty term. The temperature tracking term is the sum of squares of the differences between the predicted temperature and the target temperature. The control smoothing term is the sum of squares of the changes in the control quantity at adjacent time points. The quality penalty term is the square of the difference between the predicted quality of the calcined coke and the quality target.
10. A petroleum coke calcination intelligent optimization control system, used to execute the petroleum coke calcination intelligent optimization control method as described in any one of claims 1-9, characterized in that, It includes a data acquisition and synchronization module, a volatile matter sensing and pattern recognition module, a segmented feedforward compensation calculation module, a coordinated optimization control module, a quality feedback self-learning module, and a human-computer interaction and parameter tuning module; The data acquisition and synchronization module is used to synchronously acquire near-infrared spectral signals, flue gas composition signals from the top of the tank, and discharge speed signals. After alignment, rejection, and filtering, it outputs a synchronous data set. The volatile matter sensing and pattern recognition module is used to obtain the estimated value of volatile matter in the spectral channel by analyzing the near-infrared spectral signal, and to obtain the estimated value of volatile matter in the process channel by analyzing the flue gas component signal at the top of the tank combined with the discharge speed signal after time inversion alignment. Based on the current estimated value of the spectral channel, the module outputs the real-time estimated value of volatile matter, and calculates the estimation residual based on the deviation between the estimated value of the spectral channel before the transmission delay and the output value of the process channel. At the same time, the module extracts fluctuation features to identify the fluctuation pattern of the current volatile matter. The segmented feedforward compensation calculation module is used to calculate the disturbance equivalent heat flux based on the difference between the real-time estimated value of volatile matter and the benchmark volatile matter. After distributing the disturbance equivalent heat flux according to the heat release ratio coefficient of each segment, it combines the compensation strategy of fluctuation mode matching and temperature deviation, solves the compensation heat flux of each segment through the segmented heat balance equation, and maps it to the compensation amount of discharge speed, negative pressure in tank and air distribution through the sensitivity matrix. The coordinated optimization control module is used to determine the feedforward confidence level based on the estimated residual, calculate the coordination factor based on the feedforward confidence level, the current temperature deviation and the disturbance equivalent heat flux, and then weight and fuse the compensation quantity and the feedback control quantity according to the coordination factor to output a comprehensive control command. The quality feedback self-learning module is used to correct the soft measurement model parameters and optimize the objective function weight coefficients based on the deviation between the online estimated value of the true density of calcined coke and the laboratory measured value. The human-computer interaction and parameter tuning module is used to set and adjust control parameters and display key operating variables in real time.
