Lukastin reaction parameter adaptive control method based on closed-loop feedback

Abstract

The application belongs to the technical field of pharmaceutical and chemical industry control, and particularly relates to a luspaterin reaction parameter adaptive control method based on closed-loop feedback, which comprises the following steps: synchronously collecting the heat flow value, the real-time temperature in the reactor and the temperature difference between the inlet and outlet of the jacket cooling water of the reactor, smoothing the heat flow value and calculating the first-order difference to obtain the heat flow rate change amount, establishing the baseline distribution by counting the mean value and the standard deviation; constructing the upper limit of the heat flow rate abnormality discrimination, triggering the advance compensation mode when the limit is exceeded, calculating the cooling water flow forward compensation amount and the feeding rate attenuation factor; performing online identification of the heat capacity based on the measured data to obtain the actual effective heat capacity and update the forward feed model parameters; constructing the heat overrun integral amount and comparing it with the impurity overrun risk threshold value, and warning and reducing the upper limit of the feeding rate when approaching. The application moves forward the intervention time of control, effectively suppresses the heat accumulation overrun, and reduces the batch degradation risk.

Description

Technical Field

[0001] This invention belongs to the field of pharmaceutical and chemical control technology, specifically relating to an adaptive control method for ruxolitinib reaction parameters based on closed-loop feedback. Background Technology

[0002] Ruxolitinib, as a JAK1 / JAK2 kinase inhibitor active pharmaceutical ingredient (API), involves a multi-step organic reaction in its synthesis. The thermodynamic behavior of the coupling reaction stage is particularly complex, with the exothermic reaction rate fluctuating significantly with changes in substrate concentration, catalyst activity, and solvent polarity. The pharmaceutical industry has extremely stringent purity requirements for this type of API; any uncontrolled reaction leading to side reactions directly affects batch yield. Therefore, precise control of reaction process parameters is a core engineering challenge for ensuring product quality.

[0003] Patent application CN114690636A discloses a predictive temperature control method for a reaction calorimeter system based on online correction of mechanistic parameters. This patent establishes an equivalent mechanistic parameter model of the reaction calorimeter system based on heat transfer analysis for predictive output calculation. It synchronously updates the initial values ​​of the model parameters using a general parameter calibration process for reaction calorimeters. During the model-based predictive temperature control process, it adjusts the model parameters online by incorporating logic such as real-time parameter relationship comparison. This method significantly improves the stability of temperature control, effectively suppresses overshoot, and increases the convergence speed of isothermal control. With online correction of model parameters, the adaptability to a wide range of operating conditions of the reaction system is improved, reducing the impact of severe mismatch in the predictive model, and a reference trend of mechanistic parameter changes can be provided after the experiment. Combined with the online parameter calibration process for reaction calorimeter systems, the initial model parameters can be automatically updated without the need for complex or time-consuming system parameter identification.

[0004] Existing reaction control schemes typically employ PID feedback controllers to regulate single parameters such as temperature or pH. Some improved schemes introduce multi-parameter coordinated control frameworks, determining the optimal parameter combination through offline modeling and executing control actions according to a preset trajectory during the reaction execution phase. Furthermore, some schemes attempt to incorporate online sensor data to perform batch-to-batch corrections of controller parameters to compensate for deviations caused by batch variations in raw materials.

[0005] However, existing online adaptive solutions suffer from fundamental control timing defects. Current methods rely solely on thermocouple-measured temperature as the only feedback signal, reflecting the overall temperature of the reactor rather than the rate of the exothermic reaction. When the ruxolitinib coupling reaction reaches its exothermic peak, the abrupt change in the exothermic rate occurs approximately 60 to 120 seconds before a measurable temperature deviation appears. This physical response lag causes the control system to intervene only after sensing the temperature deviation. With the execution delay itself, the cooling power is only applied to the reactor after the exothermic peak has passed its halfway point. If the catalyst activity in the batch is high, the mismatch window between the exothermic rate and the system's adjustment bandwidth will further widen, leading to excessive local heat accumulation and triggering dehalogenation side reactions. This results in decreased selectivity of the target product, with the peak area percentage of major impurities exceeding the quality standard limit, causing batch degradation or scrapping. The fundamental flaw of existing solutions lies in the unavoidable physical timing misalignment between the control trigger signal and the actual progress of the exothermic process, causing compensation actions to consistently lag behind the actual evolution of the exothermic peak. Summary of the Invention

[0006] This invention provides an adaptive control method for ruxolitinib reaction parameters based on closed-loop feedback, in order to solve the technical problem that the physical response time lag between the single temperature feedback signal and the actual exothermic process leads to delayed feedforward compensation and ineffective suppression of excessive heat accumulation.

[0007] This invention provides an adaptive control method for ruxolitinib reaction parameters based on closed-loop feedback, comprising the following steps: S1, synchronously acquiring the heat flux value of the reactor, the real-time temperature inside the reactor, and the temperature difference between the inlet and outlet of the jacket cooling water; smoothing the heat flux value and calculating the first-order difference to obtain the change in heat flux rate; and statistically analyzing the mean and standard deviation of the change in heat flux rate within the initial time window of the reaction start-up to establish a baseline distribution of heat flux rate; S2, constructing an upper bound for heat flux rate anomaly detection based on the mean and standard deviation; when the change in heat flux rate exceeds the upper bound for heat flux rate anomaly detection, triggering a pre-compensation mode, and based on the change in heat flux rate, the upper bound for heat flux rate anomaly detection, and the temperature difference between the inlet and outlet of the jacket cooling water... Calculate the cooling water flow feedforward compensation amount and simultaneously calculate the feeding rate attenuation factor; S3, after completing the initial compensation adjustment, based on the cooling water flow feedforward compensation amount, the temperature difference between the jacket cooling water inlet and outlet, the duration of the compensation control action, and the measured change in the real-time temperature inside the vessel, perform batch-level online heat capacity identification to obtain the actual effective heat capacity, and update the feedforward model parameters using the actual effective heat capacity; S4, based on the deviation of the real-time temperature inside the vessel from the target temperature set value, perform time integration to construct the heat over-limit integral amount, compare the heat over-limit integral amount with the preset impurity exceedance risk threshold, and trigger an early warning when the heat over-limit integral amount approaches the impurity exceedance risk threshold, and reduce the upper limit of the feeding rate to control the heat release intensity.

[0008] By replacing the temperature result with the change in heat flow rate as the control trigger signal, the timing of control intervention is moved forward to before the physical occurrence of the exothermic peak, thus eliminating the response time lag problem of single temperature feedback; by identifying the batch heat capacity correction model parameters online, systematic deviations are eliminated; and by constructing the thermal over-limit integral quantity, the entire process quality status monitoring within the batch is realized.

[0009] Furthermore, the heat flux value is smoothed and a first-order difference is calculated to obtain the change in heat flux rate, including: The original heat flux value is smoothed by moving average using a fixed window to cancel high-frequency disturbance components and obtain a smooth heat flux signal. The first-order difference is calculated based on the sampling interval of the smoothed heat flux signal, and the change in heat flux rate is finally obtained.

[0010] The original heat flux value is smoothed by using a fixed window sliding mean to eliminate high-frequency signal noise caused by stirring and equipment vibration. Then, the first-order difference of the smoothed heat flux is obtained according to the sampling interval to obtain the change in heat flux rate. After filtering out interference, a stable and accurate early characteristic signal reflecting the increase in heat release power per unit time is obtained. This avoids noise from triggering abnormal compensation logic, ensuring the reliability of the data source for subsequent calculation of the upper limit of heat release anomaly judgment and the solution of the advance compensation amount. This allows the system to accurately capture the early weak signal of a slight acceleration in the heat release rate and ensure the recognition sensitivity of the pre-control.

[0011] Furthermore, an upper bound for judging heat flux rate anomalies is constructed based on the mean and standard deviation, including: Obtain the statistical coverage coefficient under a normal distribution; Multiplying the standard deviation by the statistical coverage factor yields the fluctuation coverage range; The mean value is added to the fluctuation range to obtain the upper bound for judging heat flux rate anomalies.

[0012] Furthermore, based on the change in heat flux rate, the upper limit of the heat flux rate anomaly detection, and the temperature difference between the inlet and outlet of the jacket cooling water, the cooling water flow feedforward compensation is calculated, including: Calculate the difference between the change in heat flux rate and the upper limit for judging heat flux rate anomalies; The specific heat capacity of the cooling water is obtained, and the specific heat capacity is multiplied by the temperature difference between the inlet and outlet of the jacket cooling water to obtain the dynamic heat transfer denominator. Divide the difference by the denominator of the dynamic heat exchange to obtain the cooling water flow feedforward compensation.

[0013] By replacing static rated values ​​with dynamic measured values, the calculation of feedforward compensation can track the actual heat exchange state of the cooling system in real time, avoiding the estimation deviation of compensation caused by differences in water supply temperature.

[0014] Furthermore, the feed rate attenuation factor is calculated simultaneously, including: Calculate the percentage of heat flux rate change that exceeds the upper limit for heat flux rate anomaly detection; Obtain the attenuation sensitivity coefficient calibrated based on the historical batch over-limit amplitude dataset; The attenuation amount is calculated based on the over-limit amplitude ratio and the attenuation sensitivity coefficient, and combined with the preset cutoff lower limit, the feeding rate attenuation factor is finally obtained.

[0015] Furthermore, based on the cooling water flow feedforward compensation, the temperature difference between the jacket cooling water inlet and outlet, the duration of the compensation control action, and the measured changes in the real-time temperature inside the vessel, batch-level online heat capacity identification is performed to obtain the actual effective heat capacity, including: The effective heat removal amount is obtained by multiplying the cooling water flow feedforward compensation amount, the specific heat capacity of the cooling water, the temperature difference between the inlet and outlet of the jacket cooling water, and the duration of the compensation control. The effective heat removal amount is divided by the measured change in real-time temperature inside the vessel to obtain the actual effective heat capacity.

[0016] By using real control event data to identify heat capacity parameters online, the calculation basis for feedforward compensation is switched to the actual thermodynamic characteristics of the material in this batch, which improves the consistency between the compensation estimate and the actual required cooling power.

[0017] Furthermore, the feedforward model parameters are updated using the actual effective heat capacity, including: Obtain the confidence range determined based on the effective heat capacity deviation range of historical batches; Determine whether the actual effective heat capacity falls within the confidence range; If the value falls within the confidence range, the nominal heat capacity in the feedforward model is replaced with the actual effective heat capacity, thus completing the update of the feedforward model parameters.

[0018] Furthermore, based on the deviation of the real-time temperature inside the vessel from the target temperature setpoint, a thermal over-limit integral is constructed by integrating over time, including: Obtain the allowable deviation bandwidth determined based on multi-point temperature measurements; Filter out the moments when the real-time temperature inside the vessel exceeds the sum of the target temperature setpoint and the allowable deviation bandwidth; The difference between the real-time temperature inside the vessel at the time of exceeding the limit and the target temperature setting is integrated over time to obtain the thermal over-limit integral.

[0019] By transforming the quality status perception that originally relied on offline detection into a real-time process quantity that can be monitored throughout the entire batch, operators can identify and intervene in quality deviation trends before the end of the batch, reducing the risk of batch degradation.

[0020] Furthermore, the process for setting the preset impurity exceedance risk threshold includes: Statistically analyze the maximum thermal exceedance integral of qualified batches in history, and calculate the historical mean and corresponding standard deviation; The initial threshold for impurity exceedance risk is obtained by adding the historical mean to the corresponding standard deviation. After a new batch is completed, the new data is incorporated into the sample set, and the historical mean and corresponding standard deviation are recalculated, ultimately achieving a rolling update of the impurity exceedance risk threshold.

[0021] Furthermore, lowering the upper limit of the feeding rate to control the exothermic intensity includes: When the thermal excess integral approaches the risk threshold for impurity exceedance, a system warning signal is triggered. The upper limit of the feeding rate is forcibly reduced to a preset ratio of the rated value; The reduced upper limit of the feeding rate is maintained until the reaction ends, thus achieving control over the exothermic intensity of the remaining reaction period.

[0022] The beneficial effects are as follows: This invention uses the first-order change in heat flux rate instead of the traditional temperature result as the control trigger signal, shifting the control intervention time forward to before the physical occurrence of the exothermic peak, thus eliminating the response time lag problem of single temperature feedback. Feedforward compensation can be triggered in the early stages of abnormal acceleration in the exothermic rate, ensuring that the actual action of cooling power occurs earlier than the moment the temperature deviation appears, effectively compressing the window for heat accumulation exceeding limits.

[0023] By automatically statistically analyzing the baseline distribution of heat flux rate during the early induction period of a batch, and objectively deriving the judgment threshold based on the measured data of this batch, the trigger sensitivity of the control system is directly bound to the active state of the catalyst in the current batch, avoiding the sensitivity drift problem caused by the solidification of empirical parameters across batches. At the same time, the rolling update design of the impurity exceedance risk threshold ensures that the quality judgment boundary continuously converges to the actual process limit as production data accumulates, and has the ability to self-calibrate during production operation. Attached Figure Description

[0024] Figure 1 This is a flowchart of an adaptive control method for ruxolitinib response parameters based on closed-loop feedback.

[0025] Figure 2 This is a line graph comparing the peak temperature trends of each reaction stage in this invention using two different methods. Detailed Implementation

[0026] 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.

[0027] An embodiment of the adaptive control method for ruxolitinib response parameters based on closed-loop feedback provided by this invention: like Figure 1 As shown, the adaptive control method for ruxolitinib response parameters based on closed-loop feedback includes the following steps: S1, synchronously acquires multiple sensor signals and establishes a baseline distribution of heat flux rate.

[0028] In an optional embodiment, after the reactor feeding stage is completed, the system synchronously initiates parallel acquisition of three sensor signals: the instantaneous heat flow value output by the heat flow meter. Real-time temperature inside the vessel and the temperature difference between the inlet and outlet of the jacket cooling water The heat flow meter directly measures the heat transferred from the reaction system to the jacket per unit time, making it the core sensor reflecting the amount of heat release rather than the result. Because the heat flow meter's sensing element directly exchanges heat with the reaction system, its physical response delay is much smaller than the thermocouple's sensing delay of the overall temperature of the vessel; there is a time difference of approximately 60 to 120 seconds between the two. This difference is the physical basis for this method to achieve forward control intervention. The three signals are synchronously written to the data buffer at a 1Hz sampling frequency for real-time reading by subsequent processing modules.

[0029] Furthermore, the original heat flux value is smoothed using a moving average within a fixed window to cancel out high-frequency disturbance components and obtain a smoothed heat flux signal. Specifically, the original heat flux signal... Smoothing was performed using a fixed window with a length of 30 seconds, corresponding to 30 sampling points at a sampling frequency of 1 Hz. This smoothing method eliminated the high-frequency disturbance component introduced by the mechanical vibration of the agitator, resulting in a smoothed heat flow signal. Based on this, the first-order difference of the smoothed heat flux signal is calculated according to the sampling interval to obtain the change in heat flux rate. :

[0030] in The sampling interval is 1 second. A positive value indicates that heat release is increasing and is a leading indicator for predicting the arrival of the heat release peak. Its physical meaning is the increase in heat release power per unit time, with the unit being W / s. Since a larger value indicates a greater increase in heat release power per unit time, the more urgent the arrival of the heat release peak is; the smaller the value, or even close to 0, the more stable the heat release intensity is, and the reaction is still in the induction phase or plateau phase.

[0031] Next, the system statistically analyzes the mean and standard deviation of heat flux rate changes within the initial time window of reaction start-up to establish a baseline distribution of heat flux rate. Specifically, this is done within the first 10% of the time window after reaction start-up. mean and standard deviation This establishes a baseline distribution of heat flux rates for this batch. This time window corresponds to the reaction induction period, during which the substrate concentration has not yet accumulated to the critical level required for the exothermic peak, and the heat flux signal is stable. These two statistics are automatically calculated entirely from early measured data of this batch, without the need for manual experience-based settings. This ensures that the baseline data is directly linked to the current batch's catalyst activity state, eliminating the risk of sensitivity drift caused by cross-batch parameter transfer from the source. It should be noted that there are many other methods in the art for denoising heat flux signals, including median filtering and exponentially weighted moving averages. Those skilled in the art can choose different smoothing methods based on the actual sampling frequency and perturbation frequency characteristics.

[0032] By using a heat flow meter as the core sensor and extracting the first-order change in heat flow rate, the system obtains an earlier indicator of the heat release process on the time axis than the temperature result, laying the signal foundation for triggering subsequent advance compensation actions and advancing the timing of the control system's detection of heat release anomalies from after the temperature deviation occurs to the heat release rate acceleration period.

[0033] S2, construct the upper bound of the judgment to trigger the advance compensation mode.

[0034] In an optional embodiment, the statistical coverage coefficient under a normal distribution is obtained. The standard deviation is multiplied by the statistical coverage coefficient to obtain the fluctuation coverage range. The mean is added to the fluctuation coverage range to finally obtain the upper bound for heat flux rate anomaly detection. This is based on the already statistically obtained batch baseline parameters. and Based on this, an upper bound for judging heat flux rate anomalies in this batch is constructed. :

[0035] During the normal exothermic induction period, It follows an approximately normal distribution. The coverage coefficient is the statistical value of the normal distribution. Taking a value of 3 corresponds to a 99.7% coverage range of normal fluctuations, that is, when Exceed When the probability of this event occurring under normal operating conditions is less than 0.3%, it can be objectively determined that the heat release rate has entered an abnormally accelerated range. The 3σ statistical coverage property, derived from the normal distribution, is an existing conclusion in statistics and is used as a statistical coverage coefficient; while and All data are derived from actual measurements in this batch; therefore, the upper limit for judging heat flux rate anomalies is set. It is fully adaptively bound to the activity status of the current batch of materials, without the intervention of cross-batch empirical constants.

[0036] When detected When the system detects that the exothermic peak precursor signal has been triggered, it immediately enters the advance compensation mode and issues two parallel control outputs: cooling water flow feedforward compensation amount. and the feeding rate attenuation factor .

[0037] Furthermore, the difference between the change in heat flow rate and the upper limit of the heat flow rate anomaly detection is calculated to obtain the specific heat capacity of the cooling water. The specific heat capacity is then multiplied by the temperature difference between the inlet and outlet of the jacket cooling water to obtain the dynamic heat transfer denominator. The difference is then divided by the dynamic heat transfer denominator to finally obtain the cooling water flow feedforward compensation amount. The calculations are based on the heat balance equation. Existing feedforward control uses the rated temperature difference... As a fixed parameter in the denominator, its original calculation formula is:

[0038] Because this formula does not consider the real-time changes in the actual heat exchange efficiency of cooling water with variations in supply water temperature and flow rate, it is problematic when there are significant seasonal differences in supply water temperature between winter and summer. The calculation will result in a consistently overestimation or underestimation of the compensation amount. This invention replaces the fixed rated temperature difference in the denominator with a value measured in real-time by temperature difference sensors at the jacket inlet and outlet. The improved formula for calculating the feedforward compensation amount is obtained:

[0039] in, For the real-time temperature difference between the inlet and outlet of the jacket cooling water, The specific heat capacity of the cooling water is used; by replacing the static rated value with the dynamic measured value, the calculation of the feedforward compensation quantity tracks the actual heat exchange state of the cooling system in real time.

[0040] Next, the proportion of heat flux rate change exceeding the upper limit for heat flux rate anomaly detection is calculated. An attenuation sensitivity coefficient, calibrated based on historical batch attenuation datasets, is obtained. The attenuation is calculated based on the attenuation proportion and the attenuation sensitivity coefficient. Combined with a preset lower cutoff limit, the feed rate attenuation factor is finally obtained. Feed rate attenuation factor The formula for calculation is:

[0041] Where the attenuation sensitivity coefficient Based on the feedforward compensation events triggered in historical batches The over-limit amplitude dataset is used as a sample. The upper quartile value of this sample set is taken as the calibration benchmark, so that... It reaches the lower limit precisely at the typical maximum exceedance. This calibration benchmark is updated rollingly with new batch data to ensure that the sensitivity coefficient remains adaptively linked to the degree of exceedance that has occurred historically in actual production. Set truncation lower bound This lower bound is objectively determined by process constraints: when the feeding rate is less than 40% of the rated rate, the substrate concentration in the reactor is insufficient to maintain the reaction driving force, and the total batch duration will exceed the upper limit of the reaction time window constraint imposed by the quality standard. If the calculated value is lower than 0.4, take 0.4 as the actual execution value and keep it constant, then wait. Falling back to The following will then be restored to 1 using the same linear relationship. It should be noted that... The cutoff lower bound is not fixed at 0.4. Those skilled in the art can adjust the lower bound accordingly based on the substrate concentration lower limit and batch time constraints of the specific reaction system, combined with experimental data from the process development stage.

[0042] By simultaneously incorporating the batch adaptive threshold and the real-time measured cooling and heat transfer status into the feedforward compensation calculation, the control system can complete the joint adjustment of cooling power and feeding rate before the exothermic peak causes a measurable deviation in the kettle temperature. The timing and magnitude of the compensation action are objectively determined by the measured process quantity of this batch, effectively eliminating the time lag caused by single temperature feedback.

[0043] S3 identifies the actual effective heat capacity online and updates the model.

[0044] In an optional embodiment, the aforementioned improved feedforward formula is based on the vessel heat capacity. As a model parameter, the nominal value of this parameter comes from the equipment's factory calibration data. Because the density and specific heat capacity of the reactants vary with the source of raw materials in different batches under actual operating conditions, there is a certain systematic deviation between the nominal heat capacity and the actual heat capacity. If this deviation is not corrected, the compensation amount will be consistently too high or too low.

[0045] After each batch of reaction reaches its exothermic peak and completes its initial compensation adjustment, the system uses measured data from this adjustment event to perform online batch-level heat capacity identification. The effective heat removal is obtained by multiplying the cooling water flow rate feedforward compensation, the specific heat capacity of the cooling water, the temperature difference between the jacket cooling water inlet and outlet, and the duration of the compensation control action. This effective heat removal is then divided by the measured change in real-time temperature inside the reactor to obtain the actual effective heat capacity. During the feedforward compensation intervention period, the system knows the actual execution... Input quantity and Measured change Based on the inverse solution of the same heat balance equation, the actual effective heat capacity of this batch is obtained. :

[0046] in To compensate for the duration of the control effect, This represents the measured change in temperature inside the vessel during this period. A negative value indicates a decrease in temperature, signifying that the cooling compensation has achieved an effective heat removal effect.

[0047] Furthermore, a confidence range determined based on the effective heat capacity deviation range of historical batches is obtained. It is then determined whether the actual effective heat capacity falls within this confidence range. If it does, the nominal heat capacity in the feedforward model is replaced with the actual effective heat capacity, thus completing the update of the feedforward model parameters. The validity of the identification results is verified through testing. The confidence level is confirmed by checking whether the value falls within the confidence range of the nominal heat capacity. The boundary of the confidence range is determined during the process development phase by statistically analyzing measured density and specific heat capacity data from multiple batches of raw materials from different sources. The upper limit of the effective heat capacity deviation range of historical batches is taken as the confidence boundary. This boundary is objectively determined by actual material characteristic data and does not require manual experience. If the identification result exceeds this confidence range, the sensor is considered abnormal. The system maintains the original nominal value and triggers an alarm, without performing the current correction, to avoid amplifying control errors due to incorrect heat capacity estimation caused by sensor malfunction. This is achieved through identification. Then, replace it in the feedforward formula that has already completed its first trigger. This process updates the model parameters for this batch. This identification process is performed only once per batch, taking data from the first exothermic peak event. The corrected parameters remain effective for the remainder of the batch, ensuring that subsequent control compensation calculations are based on the actual thermodynamic characteristics of this batch, rather than relying on equipment nominal values ​​that may deviate from the current material specifications.

[0048] It should be noted that the heat balance equation upon which heat capacity identification relies holds true during the compensation control period only if there are no significant external heat input disturbances during that time. If there is frequent switching between heating and cooling jackets in the batch process, those skilled in the art need to constrain the selection of the identification time window according to the actual jacket operation mode to ensure that the identification conditions meet the applicable prerequisites of the heat balance equation.

[0049] By performing a one-time online identification of heat capacity parameters using real control event data within a batch, the calculation benchmark for feedforward compensation is switched from the nominal value of the equipment to the actual thermodynamic characteristics of the material in this batch. This eliminates the systematic deviation introduced by changes in the source of raw materials, and significantly improves the consistency between the compensation estimate and the actual required cooling power during the subsequent exothermic peak stage.

[0050] S4, construct the thermal excess integral to control the heat release intensity.

[0051] In one optional embodiment, the dehalogenation side reaction occurs after the heat accumulation exceeding the limit event, and direct offline HPLC detection of impurity content has an analytical cycle of several hours, making it impossible to provide real-time feedback within the batch. Therefore, the system establishes an online impurity proxy index system that uses real-time measurable process quantities to replace offline analytical quantities, enabling real-time estimation of batch quality status.

[0052] The proxy index is constructed based on the following physical derivation: the dehalogenation side reaction is a competitive side reaction, and its occurrence rate is positively correlated with the duration for which the local temperature inside the reactor exceeds the upper limit of the target reaction window. The allowable deviation bandwidth, determined based on multi-point temperature measurements, is obtained. The moments when the real-time temperature inside the reactor exceeds the sum of the target temperature setpoint and the allowable deviation bandwidth are selected. The difference between the real-time temperature inside the reactor and the target temperature setpoint corresponding to the moment of exceedance is integrated over time to obtain the thermal exceedance integral. The thermal exceedance integral is defined as follows. For all items within the batch that meet the requirements The time integral of the temperature deviation corresponding to the moment:

[0053] in Set the target temperature for this reaction stage. The allowable deviation bandwidth is expressed in °C·min. The value was determined by multi-point temperature measurements during the process development phase: A multi-point thermocouple array was arranged at different locations inside the reactor to simultaneously collect temperature data from the axial temperature measurement point and local hot spots near the wall. The average of the differences between these two values ​​across multiple typical exothermic peak events was taken as the temperature reading. A larger value indicates a greater temperature gradient between the central temperature measuring point and the near-wall hot spot within the vessel, making it easier for the central temperature measuring point to underestimate the degree of local heat accumulation. Conversely, a smaller value indicates a more uniform temperature distribution within the vessel, making the central temperature measuring point more representative. This value is objectively determined by the hardware layout and material flow characteristics, and does not require manual evaluation based on experience.

[0054] Next, the maximum integral value of heat exceeding the limit for historical qualified batches is statistically analyzed, and the historical mean and corresponding standard deviation are calculated. The historical mean and corresponding standard deviation are then added together to obtain the initial impurity exceedance risk threshold. After a new batch is completed, the new data is added to the sample set, and the historical mean and corresponding standard deviation are recalculated, ultimately achieving a rolling update of the impurity exceedance risk threshold. Based on batch offline HPLC data, the historical qualified samples are statistically analyzed. Distribution, select qualified batches The historical mean of the maximum value plus one standard deviation is used as the risk threshold for impurity exceedance. Initial settings:

[0055] in For historical qualified batches The mean of the maximum values, To correspond to the standard deviation, both were calculated from historical measured data. For each new batch with offline HPLC results, this batch... The data is assigned to the corresponding qualified or downgraded sample set, and the above formula is recalculated. and Scroll refresh The entire process involves only addition, subtraction, and division operations, and no fitting model needs to be introduced.

[0056] When the heat excess component approaches the impurity exceedance risk threshold, a system warning signal is triggered, forcibly lowering the upper limit of the feeding rate to a preset ratio of the rated value. This lowered feeding rate is maintained until the reaction ends, ultimately controlling the exothermic intensity during the remaining reaction period. Specifically, during batch execution, the system performs rolling calculations in 10-minute cycles. and with Comparison. When At that time, it was confirmed that the control compensation of feedforward compensation and heat capacity correction had effectively suppressed thermal over-limit events, and the batch quality risk was under control. near When the system triggers an early warning signal to send a warning to the operator, it will forcibly reduce the upper limit of the feeding rate to a preset ratio of the rated value. For example, the reduction ratio can be 60% of the rated value. The upper limit of the feeding rate after the reduction will be maintained until the reaction ends, so as to reduce the exothermic intensity of the remaining reaction period and prevent the impurity proxy index from accumulating and exceeding the risk boundary.

[0057] After the batch ends, the system will process the data for this batch. curve, The identification values, timestamps of each control event, and compensation amounts are fully archived and correlated with subsequent offline HPLC detection results to continuously enrich the proxy indicator sample library. The settings gradually converge towards a more precise process boundary as batches accumulate. It should be noted that those skilled in the art can adjust the settings based on the actual production schedule and offline analysis cycle. The scrolling refresh frequency should be adjusted, and the minimum number of valid batches in the sample library should also be determined based on the degree of batch-to-batch variation in the specific process.

[0058] Thus, by constructing an online impurity proxy index system with thermal excess integral as the core, the system transforms the quality status perception that originally relied on offline HPLC into a real-time process quantity that can be monitored throughout the entire batch. This enables operators to identify and intervene in quality deviation trends before the end of the batch, effectively reducing the risk of batch downgrading and scrapping. At the same time, it achieves self-calibration of the quality judgment boundary as batch data accumulates.

[0059] like Figure 2 As shown, it can be observed that both broken lines reach their highest point around 90 minutes and then fall back. The peak temperature of the method of the present invention is lower than that of the existing method in each stage, indicating that the present invention has a more effective suppression effect on the peak temperature throughout the reaction process.

[0060] 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. An adaptive control method for ruxolitinib response parameters based on closed-loop feedback, characterized in that, Includes the following steps: S1. Simultaneously collect the heat flux value of the reactor, the real-time temperature inside the reactor, and the temperature difference between the inlet and outlet of the jacket cooling water. Smooth the heat flux value and calculate the first-order difference to obtain the change in heat flux rate. Within the initial time window of the reaction start-up, statistically analyze the mean and standard deviation of the change in heat flux rate to establish a baseline distribution of heat flux rate. S2. Construct an upper bound for heat flux rate anomaly detection based on the mean and standard deviation. When the change in heat flux rate exceeds this upper bound, trigger the advance compensation mode. Based on the change in heat flux rate, the upper bound for heat flux rate anomaly detection, and the temperature difference between the inlet and outlet of the jacket cooling water, calculate the feedforward compensation amount of the cooling water flow rate. Simultaneously calculate the additional... S3, After the initial compensation adjustment is completed, the actual effective heat capacity is obtained by online identification of batch-level heat capacity based on the feedforward compensation amount of cooling water flow, the temperature difference between the inlet and outlet of the jacket cooling water, the duration of the compensation control action, and the measured change in the real-time temperature inside the vessel. The feedforward model parameters are updated using the actual effective heat capacity. S4, The heat over-limit integral is constructed by time integration based on the deviation of the real-time temperature inside the vessel from the target temperature set value. The heat over-limit integral is compared with the preset impurity over-limit risk threshold. When the heat over-limit integral approaches the impurity over-limit risk threshold, an early warning is triggered, and the upper limit of the feeding rate is reduced to control the heat release intensity.

2. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, The heat flux value is smoothed and a first-order difference is calculated to obtain the change in heat flux rate, including: The original heat flux value is smoothed by moving average using a fixed window to cancel high-frequency disturbance components and obtain a smooth heat flux signal. The first-order difference is calculated based on the sampling interval of the smoothed heat flux signal, and the change in heat flux rate is finally obtained.

3. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, An upper bound for judging heat flux rate anomalies is constructed based on the mean and standard deviation, including: Obtain the statistical coverage coefficient under a normal distribution; Multiplying the standard deviation by the statistical coverage factor yields the fluctuation coverage range; The mean value is added to the fluctuation range to obtain the upper bound for judging heat flux rate anomalies.

4. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Based on the change in heat flux rate, the upper limit of heat flux rate anomaly detection, and the temperature difference between the inlet and outlet of the jacket cooling water, the cooling water flow feedforward compensation is calculated, including: Calculate the difference between the change in heat flux rate and the upper limit for judging heat flux rate anomalies; The specific heat capacity of the cooling water is obtained, and the specific heat capacity is multiplied by the temperature difference between the inlet and outlet of the jacket cooling water to obtain the dynamic heat transfer denominator. Divide the difference by the denominator of the dynamic heat exchange to obtain the cooling water flow feedforward compensation.

5. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Simultaneous calculation of the feed rate attenuation factor, including: Calculate the percentage of heat flux rate change that exceeds the upper limit for heat flux rate anomaly detection; Obtain the attenuation sensitivity coefficient calibrated based on the historical batch over-limit amplitude dataset; The attenuation amount is calculated based on the over-limit amplitude ratio and the attenuation sensitivity coefficient, and combined with the preset cutoff lower limit, the feeding rate attenuation factor is finally obtained.

6. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Based on the cooling water flow feedforward compensation, the temperature difference between the jacket cooling water inlet and outlet, the duration of the compensation control action, and the measured changes in the real-time temperature inside the vessel, the actual effective heat capacity is obtained through batch-level online heat capacity identification, including: The effective heat removal amount is obtained by multiplying the cooling water flow feedforward compensation amount, the specific heat capacity of the cooling water, the temperature difference between the inlet and outlet of the jacket cooling water, and the duration of the compensation control. The effective heat removal amount is divided by the measured change in real-time temperature inside the vessel to obtain the actual effective heat capacity.

7. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Update the feedforward model parameters using the actual effective heat capacity, including: Obtain the confidence range determined based on the effective heat capacity deviation range of historical batches; Determine whether the actual effective heat capacity falls within the confidence range; If the value falls within the confidence range, the nominal heat capacity in the feedforward model is replaced with the actual effective heat capacity, thus completing the update of the feedforward model parameters.

8. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Based on the deviation of the real-time temperature inside the vessel from the target temperature setpoint, a thermal over-limit integral is constructed by integrating over time, including: Obtain the allowable deviation bandwidth determined based on multi-point temperature measurements; Filter out the moments when the real-time temperature inside the vessel exceeds the sum of the target temperature setpoint and the allowable deviation bandwidth; The difference between the real-time temperature inside the vessel at the time of exceeding the limit and the target temperature setting is integrated over time to obtain the thermal over-limit integral.

9. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, The preset threshold for the risk of impurities exceeding the standard is set through the following process: Statistically analyze the maximum thermal exceedance integral of qualified batches in history, and calculate the historical mean and corresponding standard deviation; The initial threshold for impurity exceedance risk is obtained by adding the historical mean to the corresponding standard deviation. After a new batch is completed, the new data is incorporated into the sample set, and the historical mean and corresponding standard deviation are recalculated, ultimately achieving a rolling update of the impurity exceedance risk threshold.

10. The adaptive control method for ruxolitinib response parameters based on closed-loop feedback according to claim 1, characterized in that, Lowering the upper limit of the feeding rate to control the intensity of heat release includes: When the thermal excess integral approaches the risk threshold for impurity exceedance, a system warning signal is triggered. The upper limit of the feeding rate is forcibly reduced to a preset ratio of the rated value; The reduced upper limit of the feeding rate is maintained until the reaction ends, thus achieving control over the exothermic intensity of the remaining reaction period.

Images