A closed-loop control method for temperature deviation in pesticide production
By constructing exothermic characteristic values using multi-point temperature and flow sensors in pesticide production, and combining them with volumetric Kalman filtering to identify reaction stages and weighted fusion models, the problem of accumulated temperature estimation bias in multi-step series reactions of pesticides is solved, achieving continuity and accuracy in temperature regulation.
Patent Information
- Application Number
- CN202611515026.9
- 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
In existing technologies for multi-step cascade reactions of pesticides, when using a single model for temperature control, the temperature state cannot be accurately estimated. This leads to a mismatch between the state transition model and the actual system during the switching of reaction stages, resulting in accumulated temperature deviations and inaccurate adjustment commands.
By setting up multiple temperature sensors and flow meters in the reactor, exothermic characteristic values are constructed. The heat transfer contribution is removed by using the heat transfer ratio coefficient. Combined with volumetric Kalman filtering, the reaction stage is identified and the state transition model is matched. Weighted fusion is performed during the model switching transition period to ensure the continuity and accuracy of temperature estimation.
Under nonlinear exothermic changes and measurement disturbances, the system maintains the continuity of temperature estimation, reduces the accumulation of temperature deviations and overshoot, lowers the risk of thermal runaway and product quality fluctuations, and ensures the accuracy and stability of temperature regulation.
Smart Images

Figure CN123114142A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of temperature deviation adjustment technology, and in particular to a closed-loop adjustment method for temperature deviation in pesticide production. Background Technology
[0002] The synthesis reactions in pesticide production have strict requirements for temperature control, as the reaction temperature directly affects the conversion rate and purity of the product. In actual production, pesticide synthesis typically involves multiple cascaded reactions, with raw materials undergoing sequential reaction steps such as condensation, cyclization, oxidation, or esterification to ultimately generate the target product. Each reaction step has its own target temperature range and exothermic characteristics. Temperature deviations in previous reactions will change the composition ratio and concentration distribution of intermediate products, causing the initial material state of subsequent reactions to deviate from the process setpoint. This deviation is gradually propagated and amplified along the reaction path, and by the end of the reaction, the deviation between the actual temperature and the process target may be much greater than the initial deviation. Therefore, when adjusting temperature deviations in multi-step cascaded pesticide reactions, it is necessary to accurately estimate the temperature state at each sampling moment in order to generate timely adjustment commands that match the current operating conditions.
[0003] In the prior art, Chinese patent application CN122346209A discloses an intelligent PID control method for reactor temperature, including constructing a reactor temperature model and using a first-order inertial plus delay to describe the dynamic characteristics of the reactor temperature; constructing a Kalman filter PID control model based on an improved RBF neural network; normalizing the deviation between the target temperature and the real-time temperature of the reactor; inputting the normalized temperature deviation value into the Kalman filter PID control model, and outputting a temperature control signal to the regulating valve to adjust the reactor temperature. This method uses an improved RBF neural network algorithm to intelligently tune the PID control parameters and introduces Kalman filtering, enabling the temperature system to achieve rapid response even under nonlinear, large hysteresis, and disturbance environments. However, this method uses a single model to describe the dynamic characteristics of the reactor temperature and does not consider the significant differences in the exothermic rates of different stages in the multi-step series reaction of pesticides. When the reaction stage switches, a mismatch occurs between the filter state transition model and the actual system, and the deviation between the predicted result and the actual temperature continues to expand over several consecutive sampling periods, making it impossible to obtain reliable state estimation information for temperature deviation adjustment.
[0004] Volumetric Kalman filtering is a nonlinear state estimation algorithm that approximates the probability distribution of a system by generating deterministic sampling points in the state space, performing state prediction and observation correction at each sampling time. This algorithm uses a single process model to describe the dynamic behavior of the system when establishing the state transition equations. However, the exothermic rate, activation energy, and material heat capacity differ significantly at different stages of a pesticide multi-step reaction. When the reaction switches from one stage to the next, the thermodynamic behavior of the system changes abruptly in a short time, resulting in a mismatch between the state transition relationship established in the previous stage and the actual characteristics of the new stage. After the stage switch, the filter continues to perform state prediction based on the invalidated model, and the deviation between the prediction result and the actual temperature continues to widen over several consecutive sampling periods, making it impossible to obtain reliable state estimation information for temperature deviation adjustment. Summary of the Invention
[0005] To address the problems of existing reactor temperature control methods that use a single model and suffer from mismatch between the state transition model and the actual system during the switching of multi-step series reaction stages in pesticide production, leading to accumulated temperature estimation deviations and inaccurate control commands, this invention provides a closed-loop control method for pesticide production temperature deviations.
[0006] This invention provides a closed-loop control method for temperature deviation in pesticide production, employing the following technical solution: S1. Sampling and preprocessing are performed using temperature sensors installed at different vertical positions inside the reactor and flow meters on the inlet pipes of the cold and hot media to obtain the time series of the reactor's overall temperature, cold medium flow rate, and hot medium flow rate; S2. The temperature change rate is determined based on the overall reactor temperature, and the net cooling flow rate is determined based on the cold and hot medium flow rates. The heat transfer contribution is removed using the heat transfer ratio coefficient, and the exothermic characteristic value is calculated; S3. Based on the deviation of the exothermic characteristic value from the recent baseline level and the recent fluctuation amplitude, a stage switching detection quantity is constructed. After determining the stage switching, the quantity is retrieved from the model library. Select the state transition model whose nominal exothermic characteristic value is closest to the current exothermic characteristic value; S4. During the model switching transition period, use the state transition model before switching and the selected state transition model to perform volumetric Kalman state prediction, respectively. Use the transition coefficient to fuse the state prediction value and the prediction error covariance matrix, and combine it with the current comprehensive temperature value in the reactor for observation and update to obtain the temperature estimate; S5. Determine the temperature deviation based on the temperature estimate and the target temperature of the current reaction stage, and adjust the opening of the cold medium valve and the hot medium valve accordingly, and repeat S1 to S5 in a cycle for each sampling period.
[0007] To address the problems of existing pesticide multi-step reactions controlled by a single model based on measured temperature, which struggles to distinguish between exothermic reactions and heat transfer in the medium, suffers from estimation lag during abrupt stage changes, and is prone to valve over-adjustment, this invention integrates multi-point temperatures and hot and cold medium flow rates to construct exothermic characteristics that eliminate the influence of heat transfer. Based on this, reaction stages are identified and matched with state transition models. During the switching period, the prediction results of the old and new models are smoothly integrated, and the hot and cold valves are adjusted in a closed loop. This allows for continuous temperature estimation in production processes where nonlinear exothermic changes and measurement disturbances coexist, reducing overshoot and oscillations near the target temperature, and mitigating the risks of thermal runaway and product quality fluctuations.
[0008] Preferably, the method for obtaining the time series of the overall temperature value, cold medium flow rate, and hot medium flow rate inside the reactor includes: acquiring continuous sampling values of each temperature sensor, cold medium flow rate, and hot medium flow rate, which are equally spaced along the vertical direction of the reactor's inner wall, according to a preset sampling period; performing sliding window median filtering on the continuous sampling values of each temperature sensor, and taking the arithmetic mean of the median filtering results at the same sampling time to obtain the overall temperature value inside the reactor at each sampling time; performing sliding window arithmetic mean processing on the continuous sampling values of cold medium flow rate and hot medium flow rate, respectively, to obtain the cold medium flow rate and hot medium flow rate at each sampling time; and arranging the overall temperature value, cold medium flow rate, and hot medium flow rate inside the reactor according to the sampling time to obtain the time series of the overall temperature value, cold medium flow rate, and hot medium flow rate inside the reactor.
[0009] Preferably, the method for calculating the heat transfer ratio coefficient includes: during the preheating stage before the reactants are added, obtaining the comprehensive temperature value, cold medium flow rate, and hot medium flow rate inside the vessel according to a preset sampling period; determining the temperature change rate based on the comprehensive temperature value inside the vessel at adjacent sampling times, and determining the net cooling flow rate based on the cold medium flow rate and hot medium flow rate at the same sampling time; performing least squares fitting with the net cooling flow rate as the independent variable and the temperature change rate as the dependent variable, and determining the fitted ratio parameter as the heat transfer ratio coefficient.
[0010] Calibrate the ratio between net cooling flow rate and temperature change rate during the preheating stage before reactants are added. This can reduce the interference of reaction exothermics on parameter identification and make the heat transfer ratio coefficient adaptable to the actual reactor, media pipeline and production environment, providing a reliable parameter basis for accurately separating the temperature changes caused by hot and cold media.
[0011] Preferably, the method for calculating the exothermic characteristic value includes: calculating the ratio of the difference between the comprehensive temperature values inside the vessel at adjacent sampling times to the sampling period to obtain the temperature change rate; calculating the difference between the flow rate of the cold medium and the flow rate of the hot medium at the same sampling time to obtain the net cooling flow rate; calculating the product of the heat transfer ratio coefficient and the net cooling flow rate to obtain the heat transfer cooling rate; superimposing the heat transfer cooling rate onto the temperature change rate to obtain the reaction exothermic contribution value; and taking the arithmetic mean of the reaction exothermic contribution values within the sliding window to obtain the exothermic characteristic value.
[0012] The heat exchange cooling rate corresponding to the net cooling flow rate is compensated from the actual temperature change rate, so that the obtained features more centrally reflect the heat exothermic contribution of the reaction itself. The short-term sampling fluctuations are weakened by the moving average, thus avoiding the increase in cooling amount from masking the enhancement of heat exothermic effect, and providing a stable basis for the identification of multi-step reaction stages.
[0013] Preferably, constructing the stage switching detection quantity based on the offset of the exothermic feature value relative to the recent baseline level and the recent fluctuation range includes: obtaining a preset number of exothermic feature values before the current sampling time; calculating the arithmetic mean of the preset number of exothermic feature values to obtain the recent baseline level; calculating the standard deviation of the preset number of exothermic feature values to obtain the recent fluctuation range; calculating the absolute value of the difference between the current exothermic feature value and the recent baseline level, and calculating the ratio of the absolute value to the recent fluctuation range to obtain the stage switching detection quantity.
[0014] The current exothermic offset is normalized using the recent mean and standard deviation, so that the test results can be adaptively adjusted according to the baseline level and fluctuation degree of different batches and different reaction stages. Therefore, it can highlight the exothermic changes that exceed the normal fluctuation range and reduce missed detections or misjudgments caused by differences in operating conditions.
[0015] Preferably, the process of selecting the state transition model from the model library that is closest to the nominal exothermic characteristic value level and the current exothermic characteristic value includes: comparing the stage switching detection quantity with the switching judgment threshold; determining that a stage switching has occurred when the stage switching detection quantity reaches or exceeds the switching judgment threshold for a consecutive preset number of sampling times; calculating the absolute difference between the current exothermic characteristic value and the nominal exothermic characteristic value level corresponding to each state transition model in the model library; and determining the state transition model with the smallest corresponding absolute difference as the state transition model whose nominal exothermic characteristic value level is closest to the current exothermic characteristic value.
[0016] Preferably, the method for obtaining the model library includes: executing a complete multi-step reaction process according to the process formula before formal production, and obtaining the exothermic characteristic value sequence corresponding to each reaction stage; dividing the exothermic characteristic value sequence according to the start and end times of each reaction stage, and removing the transition segments at both ends of each reaction stage to obtain the stable operating range of each reaction stage; calculating the arithmetic mean of the exothermic characteristic values within each stable operating range to obtain the nominal exothermic characteristic value level corresponding to each reaction stage; establishing multiple state transition models with the same model structure, and setting the nominal exothermic characteristic value level corresponding to each reaction stage as the exothermic rate parameter of the corresponding state transition model; and storing each state transition model and its corresponding nominal exothermic characteristic value level together to obtain the model library.
[0017] The nominal exothermic level is extracted from the stable operating range of the complete process flow, and the interference of transition data at both ends of the stage is eliminated, so that the model library can characterize the typical thermal behavior of each reaction stage. The use of a unified model structure and the association with the corresponding exothermic parameters also facilitates the comparison, calling and subsequent formula maintenance between models.
[0018] Preferably, the method for calculating the transition coefficient includes: recording the model switching trigger time when the switching occurs, and setting the transition coefficient corresponding to the model switching trigger time as the initial value; determining the increment step size of the transition coefficient according to the preset model switching transition period; increasing the transition coefficient periodically according to the increment step size from the model switching trigger time until the model switching transition period ends and the transition coefficient reaches the termination value, thereby obtaining the transition coefficient. The transition coefficient is used to gradually reduce the fusion weight of the state transition model before the switch and gradually increase the fusion weight of the selected state transition model.
[0019] The fusion weights of the old and new models are adjusted periodically during the preset transition period, so that the state prediction can continuously migrate with the change of reaction stage, avoiding sudden changes in model parameters that cause temperature estimation jumps and valve command jumps. It is especially suitable for the stage transition conditions where the heat release rate changes rapidly but the reaction state still has continuity.
[0020] Preferably, the method for calculating the temperature estimate includes: generating volumetric sampling points based on the state estimate and error covariance matrix of the previous sampling time; inputting the volumetric sampling points into the state transition model before switching and the selected state transition model for propagation and weighted summation to obtain the state prediction values and prediction error covariance matrices of the two state transition models respectively; weightedly fusing the state prediction values of the two state transition models according to the transition coefficient, and weightedly fusing the prediction error covariance matrices of the two state transition models according to the transition coefficient; using the fused state prediction values and prediction error covariance matrices as the prediction input for observation updates, and combining them with the comprehensive temperature value inside the vessel at the current sampling time for observation updates; after the model switching transition period ends, only the selected state transition model is used for state prediction and observation updates to obtain the temperature estimate.
[0021] Preferably, the method for obtaining the opening degree of the cold medium valve and the hot medium valve includes: calculating the difference between the estimated temperature value and the target temperature of the current reaction stage to obtain the temperature deviation; when the temperature deviation is positive, increasing the current opening degree of the cold medium valve according to the product of the temperature deviation and the cold medium regulation gain, and decreasing the current opening degree of the hot medium valve according to the product of the temperature deviation and the hot medium regulation gain; when the temperature deviation is negative, decreasing the current opening degree of the cold medium valve according to the product of the absolute value of the temperature deviation and the cold medium regulation gain, and increasing the current opening degree of the hot medium valve according to the product of the absolute value of the temperature deviation and the hot medium regulation gain; and limiting the opening degree of the adjusted valves to obtain the opening degree of the cold medium valve and the hot medium valve.
[0022] The present invention has the following technical effects: By extracting the contribution of the heat exchange system from temperature data to identify the changing characteristics of the reaction's exothermic response, the exothermic levels of each reaction stage become numerically distinguishable. This allows for the detection of signs of exothermic characteristic migration even in the early stages of stage transition, providing timely stage transition signals for the filter. Upon detecting a stage transition, the filter automatically selects the model with the closest thermodynamic characteristics to the current stage by comparing the distance between the current exothermic level and the nominal levels of each candidate state transition model. This ensures that the state transition equation remains consistent with the actual system at each stage, eliminating the problem of continuous mismatch between a single model and the actual system after a stage transition. During model replacement, the prediction results of the two models before and after the transition are progressively weighted and fused using a transition coefficient. The filter's state estimation does not exhibit jumps or oscillations at the model transition point. Even during the most critical period of temperature control during stage transitions, it can still output continuous and reliable temperature state estimation information. This ensures that the control commands generated based on this estimation information match the actual needs of the current operating conditions before and after each stage transition, reducing the problem of gradual accumulation and amplification of temperature deviations along the reaction path in multi-step series reactions of pesticides. Attached Figure Description
[0023] Figure 1 This is a flowchart of a closed-loop adjustment method for pesticide production temperature deviation according to the present invention.
[0024] Figure 2 This is a schematic diagram comparing the estimated temperature state with the actual temperature before and after the phase switch. Detailed Implementation
[0025] 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, not all, of the embodiments of the present invention. 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.
[0026] This invention discloses a closed-loop regulation method for pesticide production temperature deviation, referring to... Figure 1 This includes the following steps: S1: Data Acquisition and Preprocessing.
[0027] Three sets of platinum resistance temperature sensors are installed at equal intervals along the vertical direction on the inner wall of the reactor, located at the top, middle, and bottom of the reactor body, respectively, with a sampling period of 2 seconds. One turbine flow meter is installed on each of the cold and hot medium inlet pipes of the reactor jacket, with the same sampling period as the temperature sensors. At each sampling moment, three temperature values and two flow rate values are recorded, forming a 5-dimensional sampling vector.
[0028] The collected raw data is preprocessed according to the following procedure: For continuous sampling values from each group of temperature sensors, the median value is taken within a sliding window of 5 sampling periods and used to replace the current raw temperature reading, eliminating occasional spike interference. The arithmetic mean of the median filtering results from the three groups of temperature sensors at the same time is taken to obtain the comprehensive temperature value inside the vessel at that time. subscript Indicates the first At each sampling time point, the raw readings of the cold and hot medium flow rates are averaged over a sliding window of equal length to obtain the cold medium flow rate at that time point. and heat medium flow rate The combined temperature, cold medium flow rate, and hot medium flow rate at each sampling time are combined into a 3D preprocessing vector, which is then arranged in chronological order to form a time series for use in subsequent steps.
[0029] S2: Construct exothermic characteristic values.
[0030] The multi-step cascade reaction of pesticides involves different types of chemical transformation processes at each stage, and the enthalpy change and exothermic rate of each stage differ significantly. When the reaction transitions from one stage to the next, the heat release per unit time of the system shifts from one level to another over several sampling periods. This change in exothermic level is reflected in the temperature acquisition data as follows: the rate of temperature change fluctuates around a relatively stable baseline within each stage, but deviates from the original baseline at the stage transition point, transitioning to the level corresponding to the new stage.
[0031] However, the temperature change in the reactor is not solely determined by the exothermic reaction. At each sampling moment, the control system continuously performs heat exchange operations through the distribution of hot and cold media flow rates, and this heat exchange process also contributes to the temperature change. The temperature acquisition values combine the effects of both exothermic reaction and heat exchange; directly using the rate of temperature change cannot accurately reflect the exothermic level of the current stage because fluctuations in heat exchange with control actions mask the true change in the exothermic level. For volumetric Kalman filtering, the applicability of the state transition model depends on whether the exothermic characteristics of the current stage are consistent with the model description. When the exothermic level changes due to stage switching, the deviation between the model prediction and the actual temperature gradually accumulates over the sampling period. Therefore, it is necessary to separate the contribution of the exothermic reaction itself from the temperature and control data, so that each reaction stage is characterized at different numerical levels, providing input for subsequent identification of stage switching.
[0032] Calculate the temperature change rate at each sampling time, which is the difference between the combined internal temperature values of two adjacent sampling times divided by the sampling period. Calculate the net cooling flow rate at each sampling time, which is the cold medium flow rate minus the hot medium flow rate; a positive value indicates that cooling is dominant, and a negative value indicates that heating is dominant.
[0033] During the preheating stage before reactant addition, no chemical reaction occurs within the reactor, and temperature changes are entirely determined by the heat exchange process. Temperature change rate and net cooling flow rate data for 200 consecutive sampling periods during the preheating stage were collected, and the heat transfer ratio coefficient was calibrated using the least squares method. Its physical meaning is the rate of temperature change corresponding to a unit net cooling flow rate, and its value is determined by the heat exchange area of the reactor, the jacket volume, and the total heat capacity of the material. 200 sampling cycles correspond to a calibration duration of 400 seconds, covering the main range of temperature rise from the initial value to the process start temperature during the preheating stage, ensuring that the net cooling flow rate in the calibration data has a sufficient range of variation.
[0034] At each sampling time, the heat transfer ratio coefficient is used. The contribution of the heat exchange process to the rate of temperature change is separated to obtain the contribution value of the exothermic reaction. :
[0035] in For the first The rate of temperature change at each sampling time This is the heat transfer proportionality coefficient. For the first Net cooling flow rate at each sampling time. This is the rate at which the temperature decreases during the heat exchange process, plus the actual rate of temperature change. The heat transfer effect is then offset, and the remaining part is the contribution of the exothermic reaction to the temperature change.
[0036] For continuous The heat release contribution values at each sampling time are averaged over a sliding window to obtain the heat release characteristic value. :
[0037] in The length of the sliding window. For the first The contribution of heat release from the reaction at each sampling time. Sliding window length. Fifteen sampling periods were taken, each corresponding to 30 seconds. The transition process between adjacent stages in the pesticide synthesis reaction is usually completed within 20 to 40 seconds. The 30-second window length covers the main time span of a transition process, eliminating single-point random fluctuations while preserving the trend of changes in the exothermic level.
[0038] exothermic characteristic value This reflects the first The average contribution of exothermic reaction to the temperature change inside the vessel within the first 30 seconds of each sampling time. Within the same reaction stage, the exothermic rate is relatively stable, and the exothermic characteristic value... The enthalpy change and reaction rate fluctuate around the characteristic level of this stage, with a limited range of fluctuation. When the reaction enters the next stage, the enthalpy change and reaction rate of the new stage are different from those of the previous stage, and the exothermic characteristic value is different. This then migrates to the numerical level corresponding to the new stage. Each stage has its own exothermic characteristic value. The horizontal range allows for numerical differentiation between different stages. The heat transfer ratio coefficient has been used in the calculation process. After eliminating the temperature contribution of the heat exchange system, the exothermic characteristic value The change in temperature only reflects the alteration of the exothermic characteristics of the reaction system itself and is unaffected by adjustment operations. In multi-step series reactions, the temperature deviation generated by the preceding steps is transmitted to subsequent steps by changing the composition of intermediate products, causing the actual exothermic level of the subsequent steps to deviate from the nominal value of that stage, thus affecting the exothermic characteristic value. It can reflect this deviation in real time, providing numerical basis for subsequent steps to determine the stage switching.
[0039] S3: Establish a phase switching determination and multi-model matching mechanism.
[0040] In multi-step cascade reactions of pesticides, the transition between stages is not instantaneous but occurs gradually over a transition period. During this transition, the reaction of the previous stage has not yet completely terminated, while the reaction of the next stage has already begun. The coexistence of these two reaction processes leaves the exothermic characteristics of the system in an intermediate state as it transitions from the old to the new level. For volumetric Kalman filtering, the mismatch between the state transition model and the actual system deepens over time after the stage transition, and the residual between the temperature value predicted by the model and the temperature value observed by the sensor gradually deviates from the normal range. If the current stage is determined solely by the absolute value of the exothermic characteristic, the determination result is prone to delay or misjudgment when the exothermic characteristic value has not yet fully transitioned to the new level during the transition period.
[0041] Therefore, a mechanism is needed to detect signs of heat transfer shifts early in the transition period, and to select a model matching the current stage from a set of pre-established state transition models after the stage switch, replacing the model currently in use in the filter. The sensitivity of the mechanism needs to strike a balance between two factors: excessive sensitivity will misinterpret normal heat transfer fluctuations within a stage as stage switching, while excessive insensitivity will prevent the detection of a switch in the early transition period.
[0042] Calculate the exothermic characteristic value at each sampling time. The degree of deviation relative to recent baseline levels. (Based on data prior to the current moment.) The arithmetic mean of the exothermic characteristic values at each sampling time is used as the recent baseline level. Use 50 sampling periods, corresponding to 100 seconds, to cover a sufficient duration within a stable phase to establish a reliable baseline. Calculate before the current time step. The standard deviation of the exothermic characteristic value at each sampling time is used to characterize the normal fluctuation range of the exothermic characteristic value within the current stage.
[0043] Switching detection volume during the construction phase :
[0044] in For the first The exothermic characteristic value at each sampling time, For the first Before each sampling time The arithmetic mean of the exothermic characteristic values at each sampling time. For the first Before each sampling time Standard deviation of the exothermic characteristic values at each sampling time. Phase switching detection quantity. This indicates the degree to which the current exothermic characteristic value deviates from the recent baseline level, and is normalized using the normal fluctuation range as a scale.
[0045] Set switching threshold The value is 3.5. This represents the detection quantity during stage switching. continuous The switching threshold is reached at any sampling time. If the number of consecutive determinations is higher, a switch to the reaction phase has been determined. A value of 3 corresponds to 6 seconds. The threshold must be reached continuously, not in a single instance, to eliminate occasional shifts in the exothermic characteristic value caused by momentary sensor interference or uneven stirring.
[0046] Switch judgment threshold The basis for choosing 3.5 is that within the normal reaction phase, the fluctuation of the exothermic characteristic value around the baseline level approximately follows a normal distribution. The one-sided exceedance probability corresponding to 3.5 times the standard deviation is less than 0.02%. Normal fluctuations within the phase will almost never trigger this threshold, while the migration of the exothermic level caused by phase switching is usually more than 4 times the standard deviation and can be reliably detected.
[0047] When a phase transition is triggered, model matching is performed from a pre-established model library. The model library contains... There are 10 state transition models, each corresponding to a typical stage in the multi-step reaction of pesticides. The nominal level of the exothermic characteristic value of each stage was recorded when the model was established. ,in Number the model. Calculate the current exothermic eigenvalue. The model with the smallest distance from the nominal level of each model is selected as the matching model for the current stage.
[0048] in For the matching model number, For the first The exothermic characteristic value at each sampling time, For the first The nominal exothermic eigenvalue level of each model. The nominal exothermic eigenvalue level at each stage. The data is obtained as follows: Before formal production, a complete multi-step reaction process is executed according to the process formula. During the reaction, temperature data is collected and exothermic characteristic values are calculated according to steps S1 and S2. Intervals are divided according to the start and end times of each stage in the process specification. Within the stable operating interval of each stage, the arithmetic mean of the exothermic characteristic value sequence is taken as the nominal exothermic characteristic value level for that stage. The stable operating range is the interval from 30 seconds after the start of the stage to 30 seconds before the end of the stage. The transition segments at both ends of the stage are removed to ensure that the exothermic characteristic values involved in the calculation are at the steady-state level of the stage.
[0049] The state transition model for each stage uses the nominal exothermic characteristic value level of the corresponding stage. As a parameter for the heat release rate, it is related to the heat transfer ratio coefficient calibrated in step S2. Together they constitute the state transition equation for this stage. The models have the same structure, only the values of the heat release rate parameter are different, corresponding to the nominal heat release characteristic value level of each stage.
[0050] After matching is complete, the number will be... The state transition model is passed to subsequent steps to replace the model currently used in the capacitive Kalman filter.
[0051] Phase switching detection volume Using recent baseline levels and recent fluctuation ranges as references, the degree to which the current exothermic characteristic value deviates from the steady state is measured. This relativistic detection method allows the judgment criteria to automatically adapt to different exothermic levels and fluctuation ranges at each stage, eliminating the need to set absolute thresholds for each stage. In multi-step series reactions, when a temperature deviation in the preceding step causes the actual exothermic level of subsequent steps to deviate from the nominal value, the stage switching detection quantity... The baseline and fluctuation range are dynamically updated based on recent actual observation data, thus preventing failure due to baseline drift caused by accumulated biases. The model matching step is based on exothermic characteristic values. The numerical distance between the model and each candidate model is used to select the most suitable model, ensuring that the volumetric Kalman filter can use the state transition equation that is consistent with the current thermodynamic properties in a timely manner after the stage switch.
[0052] S4: Improved state prediction and transition correction for volumetric Kalman filtering.
[0053] When the switching occurs in step S3 and a matching model is selected, the capacitive Kalman filter needs to replace the new state transition model in the filtering recursion process. At the moment of replacement, the dynamic characteristics of the system described by the new model differ from those of the old model, while the state estimates and error covariance matrix accumulated by the filter are based on the old model. If the new model is directly loaded and the recursion continues immediately according to the standard capacitive Kalman filter procedure, there will be incompatibility between the state estimates accumulated under the old model and the prediction logic of the new model. The state prediction results will jump at the switching moment, and the magnitude of the jump depends on the degree of difference between the old and new models.
[0054] This abrupt change is transmitted to the observation update step through the Kalman gain, causing the filter to oscillate within several sampling periods after the switch, resulting in a significant decrease in estimation accuracy during the transition period. In the temperature deviation regulation of multi-step pesticide reactions, the decrease in estimation accuracy during the transition period directly leads to the regulation command deviating from the actual demand, and the stage switching moment is precisely the period when the exothermic characteristics change most drastically and the temperature control demand is most urgent. Therefore, it is necessary to correct the state prediction results during the transition period after the model switch, so that the filter can smoothly transition from the estimation state of the old model to the estimation state of the new model.
[0055] At the model switching trigger moment, record the current filter state estimate and error covariance matrix. Set the transition period length. Ten sampling periods were taken, each corresponding to 20 seconds. The transition process of switching pesticide synthesis reaction stages was completed within 20 to 40 seconds. The length of the transition period was taken as the duration of the initial stage of the transition process, so that the correction effect covered the initial period with the greatest model difference.
[0056] Calculate the transition coefficient at each sampling time during the transition period. :
[0057] in At the current sampling time, The trigger time for model switching The transition period length. Transition coefficient. It increases linearly from 0 to 1.
[0058] At each sampling time, the current state is predicted using both the model before and after the switch.
[0059] In this scheme, the state vector of the volumetric Kalman filter is the overall temperature value inside the vessel, and the state transition equation describes the temperature evolution relationship between two adjacent sampling times. The state transition equations corresponding to each model are:
[0060] in For the first Predicted temperature value at each sampling time. For the first Temperature state estimate at each sampling time, For the first The heat release rate parameters for each stage of the model. The heat transfer ratio coefficient calibrated in step S2, For the first Net cooling flow rate at each sampling time. The sampling period is This is for process noise. Heat release rate parameter. This refers to the nominal exothermic characteristic value level of this stage. The two values are the same.
[0061] Using the state transition equation of the model before the switch, and following the sampling point generation, propagation, and weighted summation process of standard capacitive Kalman filtering, the predicted values of the model before the switch are obtained. Using the state transition equation of the switched model, the predicted values of the switched model are obtained following the same procedure. In the subscript, 0 represents the model before the switch, and 1 represents the model after the switch. The two predicted values are weighted and fused using a transition coefficient to obtain the corrected predicted value. :
[0062] in These are the state predictions of the model before the switch. These are the predicted state values of the model after the switch. This is the transition coefficient.
[0063] The same weighted fusion is applied to the prediction error covariance matrix:
[0064] in This is the prediction error covariance matrix of the model before the switch. This is the prediction error covariance matrix of the model after the switch. This is the covariance matrix of the fused prediction error.
[0065] Fusion corrected predictions and prediction error covariance matrix The prediction results from the standard volumetric Kalman filter are then used in subsequent observation update steps. These updates follow the standard volumetric Kalman filter procedure, using the current combined temperature value inside the vessel at the sampling time. As an observation, the predicted value is corrected into a posterior state estimate and a posterior error covariance matrix by using Kalman gain.
[0066] After the transition period ends, the filter runs entirely using the switched model, no longer performs weighted fusion, and resumes the recursive process of standard capacitive Kalman filtering.
[0067] Transition coefficient The linearly increasing design allows the filter to maintain estimation continuity primarily by relying on the predictive inertia of the pre-switch model during the initial model switching phase, gradually transferring control to the post-switch model as the transition progresses. This gradual transition avoids state prediction jumps caused by sudden model replacement. During the transition period, the predictions from both the pre- and post-switch models participate in state estimation simultaneously, and the filter output smoothly transitions between them. The synchronous fusion of the error covariance matrix ensures the continuity of Kalman gain calculation during the transition period. During the stage switching phase of the pesticide multi-step reaction, the thermodynamic properties of the reaction system are transitioning from one stage to the next. The linear change of the transition coefficient is consistent with the gradual transition of the reaction system in terms of time rhythm, synchronizing the filter's model switching process with the actual evolution of the physical system. This ensures the reliability of temperature state estimation during the transition period and provides continuous and stable state estimation input for temperature deviation adjustment in step S5.
[0068] like Figure 2 As shown, the improved method and the standard method track the actual temperature before and after a phase switch: the estimated value of the improved method transitions smoothly during the switch period, while the estimated value of the standard method shows obvious jumps and oscillations.
[0069] S5: Temperature deviation adjustment based on improved filtering results.
[0070] The improved voluminous Kalman filter, after step S4, outputs a current temperature state estimate at each sampling time. This estimate maintains consistency with the actual temperature state before and after the reaction phase switching. At each sampling time, the temperature component is extracted from the posterior state estimate from the improved voluminous Kalman filter and used as the current temperature estimate. Read the target temperature corresponding to the current reaction stage from the process parameter table. Calculate temperature deviation :
[0071] in For the first Temperature estimate at each sampling time, This is the target temperature for the current reaction stage. Temperature deviation. A positive value indicates that the current temperature is higher than the target value, while a negative value indicates that it is lower than the target value.
[0072] Based on temperature deviation The sign and amplitude determine the adjustment direction and adjustment force. When the temperature deviation... When the value is positive, increase the opening of the cold medium valve and decrease the opening of the hot medium valve to enhance cooling and reduce the temperature inside the reactor. When the temperature deviation... When the value is negative, the opening of the cold medium valve is reduced and the opening of the hot medium valve is increased to enhance heating and raise the temperature inside the reactor. Cold medium valve opening adjustment amount. and the adjustment amount of the opening of the heat medium valve According to temperature deviation Proportional generation:
[0073]
[0074] in Gain adjustment for cold medium To adjust the gain for the heat medium, For the first Temperature deviation at each sampling time. Cold medium adjustment gain. and heat medium adjustment gain Based on the heat exchange capacity and valve characteristics of the reactor, the values were determined through a step response experiment during the system commissioning phase. Specifically, a substitute liquid with a heat capacity similar to the production material was injected into the reactor. After maintaining a stable temperature inside the reactor, the opening of the cold medium valve was increased by 10% from its current value. The overall temperature inside the reactor was recorded from the start of the response until it stabilized again. The reciprocal of the ratio of the final temperature change to the valve opening change was taken as the cold medium regulation gain. A step test was performed on the hot medium valve using the same method, and the reciprocal of the ratio of the final temperature change to the valve opening change was taken as the hot medium regulation gain. In this embodiment, with a reactor volume of 3 cubic meters and a jacket heat exchange area of 8 square meters, the cold medium adjustment gain is... The calibration value is 5.2% / ℃, and the heat transfer medium adjustment gain is... The calibration value is 4.8% / ℃.
[0075] The adjustment amount is added to the current valve opening to obtain a new valve opening command, which is then sent to the actuator.
[0076] After the valve opening command is executed, the flow rates of the hot and cold media change accordingly, and the temperature inside the reactor responds in subsequent sampling cycles. At the next sampling moment, the temperature sensor in step S1 collects new temperature data, which is preprocessed and then fed into step S2 to calculate the exothermic characteristic value. Step S3 performs stage switching detection on the exothermic characteristic value. Step S4 uses the improved volumetric Kalman filter to update the state estimate using the new observation data. Step S5 again calculates the temperature deviation based on the updated estimate and generates an adjustment command.
[0077] The processes S1 to S5 described above are executed sequentially once per sampling cycle, forming a continuously operating control loop. During the period when no switching occurs in the reaction phase, the phase switching detection quantity in step S3... If the temperature is below the threshold, the volumetric Kalman filter uses the state transition model corresponding to the current stage for normal recursion, and the temperature deviation adjustment operates according to steady-state conditions. When the reaction enters the next stage, the exothermic characteristic value of step S2... The transition to the new stage begins. Step S3 detects the stage switch and matches the new model. Step S4 ensures a smooth transition of the filter through weighted fusion during the transition period. Step S5 continues to generate adjustment commands based on the reliable temperature state estimates maintained during the transition period. Throughout the entire period from the occurrence of the stage switch to the completion of the filter's model transition, there is no interruption of estimation information or inaccuracy of adjustment commands in the temperature deviation adjustment.
[0078] Once all stages of the multi-step series reaction are completed, the reactor temperature drops to a safe range, and the control loop stops operating. Throughout the reaction process, the improved volumetric Kalman filter provides temperature estimation information consistent with the current reaction state for temperature deviation regulation within each stage and during stage transitions through stage switching determination, model matching, and transition correction. This ensures that the control commands continuously match the actual temperature control requirements of each stage, reducing the problem of temperature deviation accumulating and amplifying along the reaction path in multi-step reactions.
[0079] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A closed-loop control method for temperature deviation in pesticide production, characterized in that, Including the following steps: S1. By sampling and preprocessing temperature sensors installed at different vertical positions inside the reactor and flow meters on the inlet pipes of cold and hot media, the time series of the comprehensive temperature value inside the reactor, the flow rate of cold media and the flow rate of hot media are obtained; S2. The temperature change rate is determined based on the comprehensive temperature value inside the reactor, the net cooling flow rate is determined based on the flow rates of cold media and hot media, the heat transfer contribution is removed using the heat transfer ratio coefficient and the heat release characteristic value is calculated. S3. Construct a phase switching detection quantity based on the offset of the exothermic characteristic value relative to the recent baseline level and the recent fluctuation range. After determining the phase switching, select the state transition model from the model library that is closest to the nominal exothermic characteristic value level and the current exothermic characteristic value. S4. During the model switching transition period, the volumetric Kalman state prediction is performed using both the state transition model before switching and the selected state transition model. The state prediction value and the prediction error covariance matrix are fused using the transition coefficient, and the temperature estimate is obtained by observation and updating in conjunction with the current comprehensive temperature value inside the reactor. S5. The temperature deviation is determined based on the temperature estimate and the target temperature of the current reaction stage. The opening of the cold medium valve and the hot medium valve is adjusted accordingly, and S1 to S5 are executed cyclically for each sampling cycle.
2. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The method for obtaining the time series of the combined temperature value, cold medium flow rate, and hot medium flow rate inside the reactor includes: acquiring continuous sampling values of each temperature sensor, cold medium flow rate, and hot medium flow rate, which are equally spaced along the vertical direction of the reactor's inner wall, according to a preset sampling period; performing sliding window median filtering on the continuous sampling values of each temperature sensor, and taking the arithmetic mean of the median filtering results at the same sampling time to obtain the combined temperature value inside the reactor at each sampling time; performing sliding window arithmetic mean processing on the continuous sampling values of cold medium flow rate and hot medium flow rate, respectively, to obtain the cold medium flow rate and hot medium flow rate at each sampling time; and arranging the combined temperature value, cold medium flow rate, and hot medium flow rate inside the reactor according to the sampling time to obtain the time series of the combined temperature value, cold medium flow rate, and hot medium flow rate inside the reactor.
3. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The calculation method for the heat transfer ratio coefficient includes: during the preheating stage before reactant input, obtaining the comprehensive temperature value, cold medium flow rate, and hot medium flow rate inside the vessel according to a preset sampling period; determining the temperature change rate based on the comprehensive temperature value inside the vessel at adjacent sampling times, and determining the net cooling flow rate based on the cold medium flow rate and hot medium flow rate at the same sampling time; performing least squares fitting with the net cooling flow rate as the independent variable and the temperature change rate as the dependent variable, and determining the fitted ratio parameter as the heat transfer ratio coefficient.
4. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The calculation method for the exothermic characteristic value includes: calculating the ratio of the difference between the comprehensive temperature values inside the vessel at adjacent sampling times to the sampling period to obtain the temperature change rate; calculating the difference between the flow rate of the cold medium and the flow rate of the hot medium at the same sampling time to obtain the net cooling flow rate; calculating the product of the heat transfer ratio coefficient and the net cooling flow rate to obtain the heat transfer cooling rate; superimposing the heat transfer cooling rate onto the temperature change rate to obtain the reaction exothermic contribution value; and taking the arithmetic mean of the reaction exothermic contribution values within the sliding window to obtain the exothermic characteristic value.
5. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The phase switching detection quantity is constructed based on the offset of the exothermic feature value relative to the recent baseline level and the recent fluctuation range. This includes: obtaining a preset number of exothermic feature values before the current sampling time; calculating the arithmetic mean of the preset number of exothermic feature values to obtain the recent baseline level; calculating the standard deviation of the preset number of exothermic feature values to obtain the recent fluctuation range; calculating the absolute value of the difference between the current exothermic feature value and the recent baseline level, and calculating the ratio of the absolute value to the recent fluctuation range to obtain the phase switching detection quantity.
6. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, Selecting the state transition model from the model library whose nominal exothermic characteristic value level is closest to the current exothermic characteristic value includes: comparing the stage switching detection quantity with the switching judgment threshold; determining that a stage switching has occurred when the stage switching detection quantity reaches or exceeds the switching judgment threshold for a consecutive preset number of sampling times; calculating the absolute difference between the current exothermic characteristic value and the nominal exothermic characteristic value level corresponding to each state transition model in the model library; and determining the state transition model with the smallest corresponding absolute difference as the state transition model whose nominal exothermic characteristic value level is closest to the current exothermic characteristic value.
7. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The method for acquiring the model library includes: executing a complete multi-step reaction process according to the process formula before formal production to obtain the exothermic characteristic value sequence corresponding to each reaction stage; dividing the exothermic characteristic value sequence according to the start and end times of each reaction stage, and removing the transition segments at both ends of each reaction stage to obtain the stable operating range of each reaction stage; calculating the arithmetic mean of the exothermic characteristic values within each stable operating range to obtain the nominal exothermic characteristic value level corresponding to each reaction stage; establishing multiple state transition models with the same model structure, and setting the nominal exothermic characteristic value level corresponding to each reaction stage as the exothermic rate parameter of the corresponding state transition model; and storing each state transition model and its corresponding nominal exothermic characteristic value level together to obtain the model library.
8. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The calculation method of the transition coefficient includes: recording the model switching trigger time when the switching occurs and setting the transition coefficient corresponding to the model switching trigger time as the initial value; determining the increment step size of the transition coefficient according to the preset model switching transition period; increasing the transition coefficient in each sampling period according to the increment step size from the model switching trigger time until the model switching transition period ends and the transition coefficient reaches the termination value, thus obtaining the transition coefficient. The transition coefficient is used to gradually reduce the fusion weight of the state transition model before the switch and gradually increase the fusion weight of the selected state transition model.
9. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The method for calculating the temperature estimate includes: generating volumetric sampling points based on the state estimate and error covariance matrix of the previous sampling time; inputting the volumetric sampling points into the state transition model before switching and the selected state transition model for propagation and weighted summation to obtain the state prediction values and prediction error covariance matrices of the two state transition models respectively; weighted fusion of the state prediction values of the two state transition models according to the transition coefficient, and weighted fusion of the prediction error covariance matrices of the two state transition models according to the transition coefficient; using the fused state prediction values and prediction error covariance matrices as the prediction input for observation updates, and combining them with the comprehensive temperature value inside the vessel at the current sampling time for observation updates; after the model switching transition period ends, only the selected state transition model is used for state prediction and observation updates to obtain the temperature estimate.
10. The closed-loop regulation method for pesticide production temperature deviation according to claim 1, characterized in that, The method for obtaining the opening degree of cold medium valves and hot medium valves includes: calculating the difference between the estimated temperature and the target temperature of the current reaction stage to obtain the temperature deviation; when the temperature deviation is positive, increasing the current opening degree of the cold medium valve according to the product of the temperature deviation and the cold medium regulation gain, and decreasing the current opening degree of the hot medium valve according to the product of the temperature deviation and the hot medium regulation gain; when the temperature deviation is negative, decreasing the current opening degree of the cold medium valve according to the product of the absolute value of the temperature deviation and the cold medium regulation gain, and increasing the current opening degree of the hot medium valve according to the product of the absolute value of the temperature deviation and the hot medium regulation gain; and limiting the opening degree of the adjusted valves to obtain the opening degree of the cold medium valves and hot medium valves.
Citation Information
Patent Citations
Intelligent PID (Proportion Integration Differentiation) control method for temperature of reaction kettle
CN122346209A
