Industrial control system parameter optimization method and system for microfluidic hydrogenation reaction

Abstract

The invention relates to the technical field of industrial control, and particularly discloses a microfluidic hydrogenation reaction-oriented industrial control system parameter optimization method and system, and the method comprises the steps: obtaining in-channel reaction parameters, inter-channel coupling parameters and load state parameters of a multichannel microfluidic reactor, and carrying out the preprocessing to obtain feature data; and inputting the feature data into a pre-constructed hybrid prediction model, and predicting the reaction state change of each channel and the coupling interference between the channels. According to the method, a mechanism equation + data compensation network hybrid prediction model is constructed, and a dual-target adaptive tradeoff mechanism is introduced, so that accurate prediction and control priority dynamic matching of a multichannel microfluidic reaction state and coupling interference are realized; aiming at the problem of local physical degradation, the system creatively disassembles a coupling equation in a spatial dimension and introduces heat transfer retardation and flow resistance surge penalty factors, so that the conflict that global smooth prediction covers local extreme mutation is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and in particular to a method and system for optimizing parameters of an industrial control system for microfluidic hydrogenation reactions. Background Technology

[0002] Microfluidic hydrogenation reactions have shown great application potential in modern chemical engineering and the synthesis of high-value-added fine chemicals due to their extremely high mass and heat transfer efficiency and the inherent safety of continuous flow. However, industrial-grade microfluidic systems are usually composed of a large number of micron-sized reaction channels coupled in parallel, and the reaction process is accompanied by gas-liquid-solid multiphase flow and strong exothermic effects.

[0003] As the scale of parallel channels expands, the highly nonlinear reaction dynamics within the channels and the strong heat and flow coupling interference between channels become intertwined, posing a significant challenge to the precise optimization of parameters and the maintenance of global steady state in industrial control systems.

[0004] Existing control methods typically rely on idealized macroscopic mechanism models or purely data-driven black-box models. However, in practical engineering applications, this exposes a conflict between smooth global model predictions and nonlinear abrupt changes in local physical distortions. Specifically:

[0005] In practical multi-channel hydrogenation reactions, extreme physical phenomena such as gas-locking or localized uneven scaling of the catalyst are extremely prone to occur, leading to a sharp increase in local fluid resistance or a significant decrease in heat transfer efficiency. Traditional macroscopic prediction models are often based on the assumption of continuous and uniform fluid and heat transfer. This smooth calculation of global parameters can completely mask local nonlinear extreme conditions, causing the control system to fail to detect local physical degradation in time and thus issue misaligned parameter adjustment commands, which can easily lead to reaction quenching or even localized thermal runaway. Summary of the Invention

[0006] This invention aims to at least partially address one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a method and system for optimizing parameters of an industrial control system for microfluidic hydrogenation reactions, thereby improving the overall system reaction yield.

[0007] To achieve the above objectives, a first aspect of the present invention proposes a method for optimizing parameters of an industrial control system for microfluidic hydrogenation reactions, comprising:

[0008] The reaction parameters within the channels, the coupling parameters between channels, and the loading state parameters of the multi-channel microfluidic reactor were obtained, and the characteristic data were obtained by preprocessing.

[0009] The feature data is input into a pre-built hybrid prediction model to predict the change in the reaction state of each channel and the coupling interference between channels;

[0010] Based on the changes in the reaction state and the coupling interference, the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient are calculated respectively to determine the control target priority of each channel.

[0011] Based on the control target priority and the feature data, the optimal control parameters for each channel are generated by the model predictive control algorithm and then executed to control the multi-channel microfluidic reactor.

[0012] The prediction process of the hybrid prediction model includes:

[0013] The basic reaction rate and basic mass transfer rate under uncoupled conditions are calculated using a pre-defined mechanistic equation.

[0014] The feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value of the corresponding channel and the coupling interference value between the channels.

[0015] The reaction state drift value and the coupling interference value are used to fuse and correct the basic reaction rate and the basic mass transfer rate to obtain the reaction state change.

[0016] To achieve the above objectives, a second aspect of the present invention proposes an industrial control system parameter optimization system for microfluidic hydrogenation reactions, the method comprising:

[0017] The data acquisition and preprocessing module is used to acquire the reaction parameters within the channels, the coupling parameters between channels, and the load state parameters of the multi-channel microfluidic reactor, and to preprocess them to obtain characteristic data.

[0018] The hybrid prediction module is used to input the feature data into a pre-built hybrid prediction model to predict the reaction state changes of each channel and the coupling interference between channels;

[0019] The target priority determination module is used to calculate the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient based on the reaction state change and the coupling interference, so as to determine the control target priority of each channel.

[0020] The collaborative control optimization module is used to generate the optimal control parameters for each channel based on the control target priority and the feature data through a model predictive control algorithm, and then issue the parameters for execution to control the multi-channel microfluidic reactor.

[0021] Specifically, the hybrid prediction module, when performing prediction, is used for:

[0022] The basic reaction rate and basic mass transfer rate under uncoupled conditions are calculated using a pre-defined mechanistic equation.

[0023] The feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value of the corresponding channel and the coupling interference value between the channels.

[0024] The reaction state drift value and the coupling interference value are used to fuse and correct the basic reaction rate and the basic mass transfer rate to obtain the reaction state change.

[0025] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described method for optimizing parameters of an industrial control system for microfluidic hydrogenation reactions.

[0026] The present invention relates to a parameter optimization method and system for industrial control systems for microfluidic hydrogenation reactions. By constructing a hybrid prediction model of mechanism equations and data compensation networks and introducing a dual-objective adaptive trade-off mechanism, it achieves accurate prediction and dynamic matching of control priorities for the multi-channel microfluidic reaction state and coupling interference. To address the problem of local physical degradation, the system creatively decomposes the coupling equations in the spatial dimension and introduces heat transfer hindrance and flow resistance surge penalty factors, effectively resolving the conflict between global smooth prediction and local extreme mutations, and achieving accurate perception of local hot spots and pressure build-up.

[0027] To address the issue of multi-channel interference, this invention not only effectively reduces measurement and disturbance lag using a two-stage compensation algorithm based on the rate of state change, but also innovatively reconstructs the exponential safety barrier and asymmetric hydraulic shock suppression term in the objective optimization function. This significantly alleviates the contradiction between single-channel rapid load reduction suppression and the hydraulic stability of the globally shared pipeline. The overall technical solution achieves smooth disturbance rejection and safe, flexible coordination of complex microfluidic systems under extreme conditions, improving the overall system's response yield and control robustness while ensuring the physical safety boundaries of multiple channels. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the parameter optimization method for industrial control systems oriented towards microfluidic hydrogenation reactions provided by the present invention.

[0029] Figure 2 This is a comparison chart of the original noisy measurement data and the feature data after Kalman filtering and normalization fusion processing in the parameter optimization method for industrial control systems for microfluidic hydrogenation provided by this invention.

[0030] Figure 3 This is a two-stage compensation response curve of control increment for sensor and physical conduction hysteresis in the industrial control system parameter optimization method for microfluidic hydrogenation reaction provided by the present invention.

[0031] Figure 4 This is a three-dimensional mapping surface plot of the single-channel conflict trade-off coefficient as a function of process window deviation and by-product concentration in the industrial control system parameter optimization method for microfluidic hydrogenation reaction provided by this invention.

[0032] Figure 5 This is a pseudo-color image of the evolution of the axial spatial temperature field and the accumulation of local hot spots in the industrial control system parameter optimization method for microfluidic hydrogenation reaction provided by the present invention.

[0033] Figure 6 This is a comparison curve of the influence of linear by-product penalty term and exponential barrier penalty term on the safety control of the target channel in the parameter optimization method for industrial control system of microfluidic hydrogenation reaction provided by the present invention.

[0034] Figure 7 This is a comparison diagram of the transient response of pipeline feed velocity and hydraulic pressure drop before and after introducing asymmetric impact suppression in the industrial control system parameter optimization method for microfluidic hydrogenation reaction provided by the present invention.

[0035] Figure 8 This is a schematic diagram illustrating the implementation of the parameter optimization system for industrial control systems oriented towards microfluidic hydrogenation reactions provided by the present invention.

[0036] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0037] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0038] The following description, with reference to the accompanying drawings, outlines an industrial control system parameter optimization method, system, and electronic device for microfluidic hydrogenation reactions according to embodiments of the present invention.

[0039] Example 1:

[0040] In existing technologies, microfluidic hydrogenation reactions involve complex multiphase systems with gas, liquid, and solid catalysts, exhibiting extremely high specific surface areas and mass and heat transfer rates within micrometer-scale channels. However, as the number of parallel channels increases, the nonlinear reaction kinetics within the microchannels, coupled with heat transfer and fluid distribution coupling interference between channels, become superimposed, making it difficult for traditional single-loop control strategies to maintain the global stability of the system.

[0041] Therefore, this embodiment provides a parameter optimization method for an industrial control system for microfluidic hydrogenation reactions. This method relies on an industrial control system that includes data acquisition hardware, edge computing nodes, and a central control server. By introducing a hybrid prediction model and a model predictive control algorithm, it achieves collaborative parameter optimization of a multi-channel microfluidic reactor.

[0042] For example, the parameter optimization method for an industrial control system for microfluidic hydrogenation reactions provided in this embodiment is specifically applied to an industrial control system, and the method specifically includes the following:

[0043] Step S100: The industrial control system first acquires the reaction parameters within the channels, the coupling parameters between channels, and the load state parameters of the multi-channel microfluidic reactor, and performs preprocessing to obtain characteristic data.

[0044] In this step, the intra-channel reaction parameters refer to the physical and chemical quantities characterizing the fluid state and reaction process within a single microchannel, including the channel inlet temperature, channel outlet temperature, local temperature measurement points along the channel's axial distribution, inlet and outlet fluid pressure drops, hydrogen inlet flow rate, substrate feed concentration, and spectral intensity of the outlet product. Inter-channel coupling parameters characterize the mutual influence between adjacent or nearby microchannels due to their physical proximity, including the thermal conductivity temperature difference between adjacent channels, the fluid distribution coefficient of the overall feed manifold, and the thermal resistance converted from the geometric distance between adjacent channels. Load state parameters characterize the current overall operating load and catalyst health of the microfluidic reactor, including the characteristic values ​​of catalyst activity decay within the channel and the time-varying rate of hydraulic resistance caused by microscale fouling on the tube walls.

[0045] After acquiring the raw sensor data, the industrial control system needs to perform rigorous preprocessing to obtain feature data because different sensors have different sampling frequencies and different dimensions of different physical quantities. The process of preprocessing to obtain feature data includes the following steps:

[0046] First, linear interpolation is used to align the intra-channel reaction parameters, inter-channel coupling parameters, and load state parameters in time. Since the sampling period of the spectrometer is usually longer than that of the temperature sensor, the industrial control system uses the shortest sampling period as the reference time axis. Using the real physical data from two discrete sampling moments, a linear mapping relationship is constructed to calculate the characteristic values ​​at the virtual time nodes, thereby eliminating the time-scale misalignment of multi-source heterogeneous data.

[0047] Secondly, dimensional differences are eliminated through a standardization formula. The standardization formula is expressed as:

[0048] ;

[0049] in, Indicates the first Each channel is in The value after standardization; Indicates the first Each channel is in The original parameter measurements at time; This represents the moving average of the measured values ​​of this type of parameter within the preset data window; This represents the standard deviation of the measured values ​​of this type of parameter within the preset data window. This formula maps data such as temperature, pressure, and concentration, which have different physical dimensions and numerical magnitudes, to a dimensionless standard normal distribution interval with a mean of 0 and a variance of 1, thus avoiding gradient bias problems in subsequent model training and prediction processes caused by differences in numerical magnitudes.

[0050] It is also important to note that after eliminating dimensional differences, the industrial control system employs a Kalman filter algorithm to perform multi-source data fusion on the dimensional difference-eliminating data, outputting the aforementioned feature data. In this process, the Kalman filter algorithm, based on the state-space state transition matrix and observation matrix of the microfluidic system, combined with the process noise covariance matrix and measurement noise covariance matrix, dynamically optimizes the optimal gain between prior estimates and posterior measurements. Through recursive solution, the system can effectively filter out high-frequency electromagnetic interference noise and fluid pulsation noise in industrial environments, extracting smooth and high-fidelity feature data.

[0051] like Figure 2 This diagram demonstrates the processing results of the data acquisition and preprocessing module in an industrial control system. The horizontal axis of the attached diagram represents time in seconds; the left vertical axis represents the raw measurement data in Kelvin; and the right vertical axis represents the characteristic data in dimensionless units.

[0052] The light gray oscillating waveform curve in the figure represents the native noisy temperature measurement data collected from the industrial site, which includes high-frequency electromagnetic interference and multiphase flow pulsation noise.

[0053] The dark red smooth step curve in the figure represents the feature data output after the original data has been processed by the standardization formula to eliminate dimensional differences, and the Kalman filter algorithm has been used to fuse multi-source data.

[0054] At the 50-second mark, the microchannel heats up, raising the average baseline of the native temperature data from 380 Kelvin to 385 Kelvin. During this process, the deep red curve filters out random noise with a standard deviation of 2.5 Kelvin, reflecting the specific temperature change trend and mapping it to feature data with a value of 2.0 on the right-hand ordinate.

[0055] The waveform transformation process demonstrates that the preprocessing step can filter out interference noise and extract smooth feature data, providing a basis for subsequent models.

[0056] Step S200: In a complex microfluidic pipeline network, there is a physical transmission delay in the transfer of fluid from the actuator to the reaction area, and there is also a response delay in the sensor's measurement of the chemical reaction results. The heat transfer between channels also has thermal inertia in the time dimension.

[0057] Directly inputting feature data with lag into the control algorithm will cause phase lag or even closed-loop oscillation in the control system. Therefore, before generating optimal control parameters, the industrial control system needs to identify the intra-channel measurement lag time and inter-channel disturbance lag time of each channel, and use a two-stage compensation algorithm based on the rate of change of state to pre-correct the feature data, generating corrected feature data.

[0058] For example, the specific process of using a two-level compensation algorithm based on the rate of change of state to pre-correct the feature data includes two levels: intra-channel hysteresis compensation and inter-channel interference hysteresis compensation.

[0059] To address measurement lag within the channel, the industrial control system utilizes the Taylor expansion truncation approximation principle to calculate the lag compensation characteristics within the channel. The corresponding calculation formula is as follows:

[0060] ;

[0061] in, This represents the estimated current actual state after forward compensation calculation; This indicates the current measurement value that has a time lag after being collected and preprocessed by sensors in the industrial field. This represents the first-order time rate of change of the current measured value on the time scale, characterizing the rate of change of the parameter; This represents the second-order time rate of change of the current measurement value on the time scale, characterizing the acceleration of parameter change; This represents the measurement lag time within a pre-selected or identified channel, and its value is equivalent to the sum of the fluid transit time and the sensor's inherent response time. By introducing correction terms for first-order velocity and second-order acceleration, the system can predict and reconstruct the masked true physical state in advance when fluctuations occur in the upstream feed.

[0062] Regarding the inter-channel interference hysteresis, considering that heat conduction between adjacent channels is not instantaneous but rather diffuses nonlinearly over time, the system calculates the inter-channel interference compensation control increment. The corresponding calculation formula is as follows:

[0063] ;

[0064] in, This indicates the channel after interference hysteresis compensation calculation. The actual control increment output value; This represents the basic control increment without considering the attenuation effect after disturbance hysteresis; This indicates the channel predicted by the subsequent model. For the channel Inter-channel coupling interference value; Indicates channel Heat or pressure fluctuations are transferred to the channel Required inter-channel interference hysteresis time; This represents the discrete control cycle of an industrial control system. The formula uses an exponential decay term to describe the dissipation characteristics of physical disturbances in the spatial and temporal media, preventing the control system from prematurely intervening in external disturbances that have not yet fully reached the current channel.

[0065] After processing by the above two-stage compensation algorithm, the industrial control system uses the compensated parameters as corrected feature data for subsequent steps.

[0066] like Figure 3 This demonstrates the technical features of the system's two-stage compensation algorithm when measurement and physical conduction lags exist. In the attached figure, the horizontal axis represents time in seconds; the left vertical axis represents temperature in Kelvin; and the right vertical axis represents the control increment as a percentage.

[0067] The dark gray dashed line in the figure represents the actual reaction core temperature inside the microchannel, which rises from 380 Kelvin to 386 Kelvin in the 2nd second.

[0068] The light blue uncompensated sensor measurement curve exhibits a delayed response due to the measurement lag time.

[0069] The system uses the rate of change of time to perform in-channel hysteresis compensation, generating the dark red in-channel hysteresis compensation correction curve in the figure. This curve pre-corrects the state estimate, reducing the phase deviation from the actual temperature data.

[0070] To address inter-channel interference hysteresis, the green inter-channel interference hysteresis compensation response curve in the figure does not directly output the basic control increment of 5.0 percentage points. Instead, it calculates the actual compensation control increment using exponential decay characteristics, gradually reducing it to 4.595 percentage points. This process demonstrates the control increment adjustment method performed by the system in response to external interference.

[0071] Step S300: Next, the industrial control system inputs the corrected feature data output from the compensation step into the pre-built hybrid prediction model to predict the reaction state changes of each channel and the coupling interference between channels. Microfluidic hydrogenation reactions exhibit strong exothermic and locally limited nonlinear characteristics. Pure mechanistic models are difficult to accurately describe local drift due to the difficulty of parameter calibration, while pure data network models lack physical boundary constraints and are prone to divergence. This embodiment employs a hybrid prediction model, fully combining the advantages of both.

[0072] Optionally, the prediction process of the hybrid prediction model includes the following inherent logical division:

[0073] First, the fundamental reaction rate and fundamental mass transfer rate under uncoupled conditions are calculated using pre-defined mechanistic equations. These equations include the Arrhenius equation, which describes the intrinsic kinetics of chemical reactions, and the two-film theory model, which describes gas-liquid two-phase mass transfer. Under ideal conditions without considering external disturbances, the system calculates the theoretical fundamental state values ​​based on the current input characteristics.

[0074] Next, the feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value and the coupling interference value between the corresponding channels. In this step, the data compensation network is a deep learning network architecture constructed by stacking long short-term memory network layers, fully connected layers, and attention mechanism layers. The reaction state drift value refers to the deviation between the actual state and the theoretical state caused by hidden physical degradation that is difficult to model using conventional mechanistic equations, such as slow catalyst deactivation and fine fouling on the inner wall of microchannels. The coupling interference value refers to the amount of parasitic heat loss caused by temperature gradients between adjacent channels and the fluid starvation effect caused by non-uniform distribution of flow in the main pipe.

[0075] It is also important to note that the process of fusing and correcting the basic reaction rate and the basic mass transfer rate using the reaction state drift value and the coupling interference value to obtain the reaction state change process hinges on the online reconstruction of kinetic parameters. The specific process includes:

[0076] The reaction kinetic constant is corrected based on the following formula:

[0077] ;

[0078] in, This represents the actual reaction kinetic constant after fusion correction. This represents the uncoupled dynamic constant calculated using the Arrhenius law through the aforementioned mechanistic equations, i.e., the theoretical pre-exponential factor; This represents the preset fusion weighting coefficient, used to balance the confidence of the mechanism output and the confidence of the data compensation. This represents the confidence level of the response state output by the data compensation network for a future prediction step. This formula multiplicatively applies the black-box dynamic features extracted by the data network to the mechanistic features, achieving parameter drift compensation guided by physical laws.

[0079] Subsequently, the industrial control system substitutes the corrected reaction kinetic constants into the mechanistic equation to update the basic reaction rate, and then substitutes the updated basic reaction rate and the coupled disturbance value into the coupled correlation equation, which includes heat conduction and flow distribution between channels. The coupled correlation equation, simplified from the law of conservation of energy and the Navier-Stokes momentum conservation equation, describes how the heat released from each channel is redistributed spatially through the heat transfer medium, and how the inlet fluid dynamic pressure is distributed to each capillary branch. By jointly solving these partial differential correlation equations with correction coefficients, the future reaction state changes of each channel are finally calculated accurately.

[0080] Step S400: In multi-channel parallel reactions, a single channel often faces the contradiction between improving substrate conversion and suppressing side reactions and prolonging catalyst lifetime; while at the global level, it faces the contradiction between maintaining the uniformity of products in each channel and allowing single channels to freely seek optimization.

[0081] Therefore, the industrial control system calculates the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient based on the changes in the reaction state and the coupling interference, in order to determine the priority of the control objectives for each channel.

[0082] For example, the steps for calculating the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient are as follows:

[0083] To address the conflicting process objectives within a single microfluidic reaction unit, the single-channel conflict trade-off coefficient is calculated using the following formula:

[0084] ;

[0085] in, Indicates the first Each channel is in The single-channel conflict trade-off coefficient at any given time quantifies the urgency of the current process state deviating from the optimal operating condition; The absolute window deviation of the current reaction conditions in the channel from the ideal process window is represented by a weighted norm of the temperature deviation ratio and the pressure deviation ratio. It indicates the rate of catalyst activity decay, characterizing the degree of catalyst aging; This represents the preset deviation threshold, which is the maximum process deviation limit allowed by the physical safety of the microfluidic reactor. This represents the preset decay threshold, which is the critical decay rate at which the catalyst is irreversibly destroyed. This represents the normalized value of the byproduct, which is the ratio of the current byproduct concentration to the maximum allowable concentration.

[0086] The calculation logic indicates that when a channel deviates from the normal window and the catalyst decays rapidly, the system increases the regulation priority within that channel; however, if the byproducts of that channel are approaching their limit, the weight is forcibly suppressed through a negative penalty term to avoid triggering dangerous side reaction temperatures.

[0087] like Figure 4 This diagram illustrates the three-dimensional mapping relationship between the single-channel conflict trade-off coefficient and the process window deviation and by-product concentration in an industrial control system. The attached diagram is a three-dimensional surface plot; the horizontal axis represents the window deviation (dimensionless); the vertical axis represents the normalized by-product value (dimensionless); and the horizontal axis represents the single-channel conflict trade-off coefficient (dimensionless).

[0088] The three-dimensional surface in the figure reflects the parameter calculation results under the target adaptive trade-off mechanism. When the normalized value of the by-product is at a low level such as 0.40 and the window deviation increases towards the preset deviation threshold of 0.20, the surface height increases and the corresponding single-channel conflict trade-off coefficient rises to 0.340, indicating that the internal adjustment priority calculated by the system increases with the increase of deviation.

[0089] As the vertical axis value, representing the normalized value of by-products, increases and approaches the limit value of 1, the surface height decreases and approaches 0. This graphical feature reflects the role of the negative penalty term in the computational logic, that is, reducing the weight of the tradeoff coefficient when the by-product concentration is high, thus suppressing the priority of specific channels.

[0090] To address the coordinated fluid and heat distribution problem across multiple parallel channels in the entire system, the overall multi-channel tradeoff coefficient is calculated using the following formula:

[0091] ;

[0092] in, This represents the overall tradeoff coefficient for multiple channels, which assesses the severity of overall imbalance in the equipment from a global, macroscopic perspective. This represents the maximum extreme value of the single-channel conflict trade-off coefficient extracted from all parallel channels, used to identify the bottleneck channel with the worst state in the current system; It represents the current range of reaction conversion rates among multiple channels, that is, the difference between the channel with the highest conversion rate and the channel with the lowest conversion rate, and characterizes the product quality dispersion of the system. This indicates a preset safety tolerance threshold. Exceeding this tolerance means that the mixed products will not meet industrial-grade purity standards. This represents the equilibrium weighting coefficient, which adjusts the proportion of the range term in the overall evaluation; This indicates the total number of parallel channels in the microfluidic reactor; This represents the upper limit boundary of the maximum allowable single-channel conflict coefficient historically or theoretically. This formula achieves a dual evaluation of local worst-case scenarios and global uniformity deviation, providing a quantitative basis for allocating multi-objective optimization weights in subsequent model predictive control algorithms.

[0093] Step S500: After the aforementioned complex nonlinear prediction and priority quantization, the industrial control system will generate optimal control parameters for each channel based on the control target priority and the corrected characteristic data using a model predictive control algorithm, and then issue the parameters for execution to control the multi-channel microfluidic reactor. The core concept of the model predictive control (MPC) algorithm is to solve a future-oriented multi-step finite-time open-loop optimization problem online within each control cycle, and use the first control action of the optimized sequence as the current output.

[0094] For example, the calculation process for generating the optimal control parameters for each channel using the model predictive control algorithm includes:

[0095] First, a target optimization function is constructed, which includes reaction conversion rate, by-product indicators, target yield constraints, and multi-channel mean square error. The specific form is as follows:

[0096] ;

[0097] Where J represents the target optimization value that the model predictive control algorithm needs to minimize in the prediction time domain, i.e., the comprehensive penalty cost; Indicates the first The actual predicted conversion rate of each channel at the end of the prediction time domain reflects the utilization level of raw materials; Indicates the first The byproduct yield at the end of the prediction time domain of each channel reflects the level of selective degradation of the chemical reaction. Indicates the first The expected output material quantity of each channel; This indicates the target output set by the upper-level production plan; Indicates all The arithmetic mean of the multi-channel predicted conversion rates of the actual predicted conversion rates of each multi-channel channel; This indicates the incentive weighting coefficient for improving conversion rates; This represents the weighting coefficient for the penalty of suppressing side reactions; This represents the weighting coefficient for the capacity tracking error penalty; This represents the global conversion rate consistency coordination weight coefficient. The first three terms of this objective function focus on the self-optimization of each independent subsystem, while the fourth term imposes soft constraints to force the parallel units to maintain coordination.

[0098] To achieve adaptive disturbance rejection under complex operating conditions, the system must combine the single-channel conflict trade-off coefficient with the multi-channel overall trade-off coefficient to optimize the weighting coefficient. to Parameter sensitivity is dynamically allocated, and the optimal control parameters are obtained by solving the objective optimization function. The specific allocation logic is as follows:

[0099] When a certain channel A sharp increase indicates that the channel faces internal danger, and the system nonlinearly amplifies the corresponding channel through a mapping relationship. Weight, while reducing The weighting forces the control algorithm to prioritize sacrificing the capacity of this local channel in order to strongly suppress side reactions;

[0100] when When the water level is higher than the normal safe level, it indicates a serious imbalance in the multi-channel load, or even material flow deviation. At this point, the control algorithm globally increases the collaborative weight coefficient. The value of guides the solver to find a combined solution of control surfaces that can reduce the extreme differences in conversion rates among the channels. The dynamically reconstructed objective function is then analyzed using the interior-point method or the effective set method. A rolling optimization solution with actuator saturation and amplitude limiting constraints is performed, and the final outputs are the heater duty cycle command, feed micropump speed, and hydrogen mass flow controller opening degree for each channel as the optimal control parameters.

[0101] Step S600: After the instruction is issued and executed to control the multichannel microfluidic reactor, the method further includes a rigorous closed-loop correction step to resist time-varying model mismatch.

[0102] The industrial control system continuously collects actual reaction indicators after executing the optimal control parameters, including the actual detected physical temperature, concentration, and yield; then, it calculates the norm of the prediction error between the actual reaction indicators and the reaction state changes calculated by the previous hybrid prediction model.

[0103] Optionally, when the prediction error of a single channel or the prediction error between channels exceeds their respective preset error convergence thresholds, it indicates that the pore structure of the internal catalyst bed has changed or the fluid properties have undergone a phase change, leading to distortion of the original model. In this case, the system uses a joint algorithm of recursive least squares and sequential quadratic programming to update the physical coupling coefficients (such as the heat transfer coefficient calibration value) and the weight coefficients of the data compensation network in the hybrid prediction model online.

[0104] Recursive least squares algorithm, due to its low computational cost and adjustable forgetting factor, is used to refresh mechanism parameters with slow time-varying characteristics in real time online; while sequential quadratic programming algorithm has a strong ability to solve nonlinear constraints and is used to perform low-frequency deep gradient iteration optimization on the high-dimensional neuron connection weight matrix of the neural network at the background computing node, thereby ensuring that the control system can achieve self-evolution throughout the entire life cycle of the chemical plant.

[0105] For example, the following constructs a running scenario containing two parallel microfluidic reaction channels (channel 1 and channel 2):

[0106] Set the current control cycle as ;

[0107] The current industrial conditions faced by the microfluidic system are as follows: transient hysteresis fluctuations occur in channel 1 due to micro-flow instability; channel 2 is potentially affected by the transmission of local hot spots from channel 1.

[0108] First, the industrial control system acquires the measured temperature of channel 1 at the current moment as the raw parameter measurement value. Within a preset data window containing 100 sampling points, the system calculates the sliding average temperature of channel 1. Standard deviation .

[0109] Substituting it into the dimensionless standardization formula:

[0110] ;

[0111] The calculation shows that: .

[0112] The dimensionless value of 2.0 indicates that the current temperature has deviated from the average baseline by two standard deviations and will be passed to the backend as feature data.

[0113] Secondly, due to the physical transit phenomenon of fluid through the micro-mixing zone, the measurement lag time within channel 1 is pre-specified. The sensor reports the current temperature measurement value. The control system's memory stack calculates the current first-order time rate of change using the finite difference method. Second-order time rate of change Substituting this into the channel lag compensation formula:

[0114] ;

[0115] The calculation process is as follows:

[0116] ;

[0117] ;

[0118] This calculation result alerts the optimizer that although the current sensor reading is only 385.0K, after overcoming the hysteresis shielding effect, the actual temperature of the reaction core area has risen to 386.0K, avoiding the problem of the cooling control command being issued too late due to the sensor's hysteresis response.

[0119] Simultaneously, considering the interference hysteresis between channels, assuming the system calculates the basic control increment (valve opening increase command) corresponding to channel 2 as follows: The coupling interference value generated by the high-temperature effect in channel 1 to channel 2 is predicted by the neural network model. .

[0120] Known inter-channel interference lag time Control cycle .

[0121] Substitute into the incremental formula for inter-channel interference compensation control:

[0122] ;

[0123] The calculation process is as follows:

[0124] Calculation of the exponential decay term: ;

[0125] Actual compensation coefficient: ;

[0126] ;

[0127] Since heat conduction requires physical time to dissipate, the transient strong interference from the adjacent channel at the current moment has not yet fully penetrated into this channel. Therefore, it is only necessary to slightly reduce the originally calculated valve increment from 5.0% to 4.595%, thus preventing the instability of channel 2 due to sensitive feedforward regulation.

[0128] Next, for channel 1, the mechanistic equations, based on Arrhenius's law, calculate the theoretical value of the current uncoupled dynamic constant. The data compensation network outputs values ​​characterizing the reaction state drift caused by increased fouling on the pipe wall or increased diffusion within the catalyst. At this time, the system sets the fusion weight coefficient. Substituting into the reaction kinetic constant fusion correction formula:

[0129] ;

[0130] The calculation process is as follows:

[0131] ;

[0132] ;

[0133] In summary, the ideal constant calculated purely theoretically (0.85) is precisely reduced and corrected to the true dynamic constant (0.748) after the data-driven network is introduced to extract and fuse the non-ideal physical degradation features (-0.15), thereby greatly improving the fit of the subsequent state solution of the coupled correlation equation.

[0134] Assuming that the current window deviation of channel 1 is determined by detection... Catalyst activity decay rate Preset deviation threshold Preset attenuation threshold Current normalized value of by-products .

[0135] Substitute into the formula for calculating the single-channel conflict tradeoff coefficient:

[0136] ;

[0137] The calculation process is as follows:

[0138] ;

[0139] Similarly, assuming channel 2 is in good condition, the calculation shows... At this point, the channel experiencing the most severe conflict in the system is channel 1. .

[0140] The system detected a significant difference in the prediction conversion rates between Channel 1 and Channel 2. Preset safety tolerance threshold Balance weight coefficient The historically set maximum conflict mutation anomaly constant Substituting into the multi-channel overall tradeoff formula:

[0141] ;

[0142] The calculation process is as follows:

[0143] ;

[0144] The system is based on the calculations as well as ,because The value is significantly higher than that of channel 2, indicating that the model predictive control algorithm is performing better in parsing the data containing... to When optimizing the objective function, the internal priority sensitivity mapping table is triggered, automatically reducing the capacity weight of channel 1. And increase the byproduct inhibition term .

[0145] At the same time, due to global imbalance Located in the lower-middle range of 0.353, the system decides to maintain the global consensus weights without significantly altering them. Under the premise of allowing channel 1 to autonomously reduce the feed flow rate for unidirectional self-healing adjustment, the optimal physical resource scheduling that balances local degradation and global capacity is achieved.

[0146] The above calculations demonstrate that when faced with complex nonlinearity, delay characteristics, and multi-path coupling in microfluidic responses, this industrial control system effectively resolves the contradictions between local and global factors, and between suppression and stability, by relying on mechanism data hybrid modeling, predictive compensation, and adaptive target collaborative optimization techniques, thus ensuring the high-precision and robust operation of industrial equipment.

[0147] Example 2:

[0148] Building upon Example 1, this example further focuses on the extremely complex spatial conditions of a microfluidic reactor. During the actual continuous operation of a microfluidic hydrogenation reaction, due to the complexity of the gas-liquid-solid three-phase interface and the microscopic inhomogeneity of the catalyst bed, localized micro-gas plugs or localized non-uniform scaling of the catalyst can easily occur within a single channel. This rapid deterioration of local physical properties disrupts the assumptions of continuity and uniformity in the distribution of fluid and heat within the channel.

[0149] If industrial control systems still employ predictive models that globally average channels, local hotspot accumulation or fluid pressure buildup will be completely masked, leading to severe lag in control commands and even localized physical damage. Therefore, this embodiment adaptively reduces and reconstructs the spatial dimension of the hybrid predictive model in Embodiment 1, aiming to achieve high-fidelity state prediction under extreme conditions through accurate identification of local abnormal physical phenomena and equation decomposition.

[0150] For example, the process of calculating the change in the reaction state by substituting the updated base reaction rate and the coupled interference value into the coupled correlation equation that includes inter-channel heat conduction and flow distribution includes the following steps for anomaly perception and model reconstruction in the spatial dimension:

[0151] Step 1: The industrial control system first acquires the temperature measurement data of multiple spatially distributed temperature measurement points in the reaction parameters within the channel, and calculates the variance of the spatial temperature gradient along the spatial dimension of the channel.

[0152] In this step, the multiple spatially distributed temperature measurement points refer to a miniature temperature sensor array nodes physically arranged at equal or non-equal intervals along the fluid flow direction of the microfluidic channel, i.e., the axial dimension. The data from these nodes is not merely used to calculate the average temperature of the channel, but rather to reconstruct the three-dimensional temperature field profile within the channel. The temperature measurement data refers to the transient temperature physical quantity output by each temperature measurement point at the same sampling moment.

[0153] Optionally, the system logically calculates the variance of the spatial temperature gradient according to the following formula:

[0154] First, calculate the local spatial temperature gradient between adjacent temperature measurement points:

[0155] ;

[0156] in, Indicates the first Each channel is in Time of the first The spatial temperature gradient of each temperature measurement zone; and These represent the first and second digits of the fluid flow along the positive direction of the channel. The spatially distributed temperature measurement points and the first Temperature measurement data from spatially distributed temperature measurement points; and These represent the physical coordinates of the corresponding spatially distributed temperature measurement points along the axial direction of the microfluidic channel.

[0157] Next, calculate the average spatial temperature gradient of the channel at the current moment:

[0158] ;

[0159] in, Indicates the first Each channel is in The average spatial temperature gradient at time t; This indicates the total number of spatially distributed temperature measurement points deployed within the channel.

[0160] Finally, the variance of the spatial temperature gradient along the channel spatial dimension is calculated:

[0161] ;

[0162] in, Indicates the first Each channel is in The variance of the spatial temperature gradient is calculated at any given time. This variance characteristic value quantifies the dispersion of heat distribution within the channel from a purely mathematical statistical perspective. Under normal uniform reaction conditions, since mass transfer and heat release are relatively stable, this variance value remains at a very low baseline level; once there is a violent local reaction or local heat accumulation, this variance value will jump exponentially.

[0163] Step 2: When the variance of the spatial temperature gradient is greater than the first preset threshold and the rate of change of the pressure drop in the channel is greater than the second preset threshold, the local deterioration section is located based on the distribution of the temperature measurement data.

[0164] It is important to note that a single temperature distribution anomaly is insufficient to definitively determine the occurrence of physical degradation, as even short-term fluctuations in normal feed concentration can trigger slight shifts in thermal distribution. Therefore, the industrial control system incorporates fluid dynamics parameters for cross-validation. The pressure drop rate represents the first derivative of the pressure difference between the microchannel inlet and outlet fluid over time.

[0165] Optionally, the first preset threshold refers to the critical variance eigenvalue boundary obtained through offline training and calibration using historical normal operating data to define the disruption of temperature field uniformity; the second preset threshold refers to the safe boundary value of the pressure drop change rate to define the nonlinear abnormal surge in fluid resistance. When both thresholds are breached simultaneously, it indicates that local catalyst sintering or gas blockage caused by the accumulation of gaseous products has most likely occurred inside the channel.

[0166] Once local physical degradation is confirmed, the system immediately backtracks all... Specific values, looking for deviations from the average spatial temperature gradient The most severely affected temperature measurement section. The specific process for locating locally deteriorated sections based on the distribution of the temperature measurement data is as follows: the continuous spatial physical coordinate range with the largest positive gradient anomaly and the largest negative gradient anomaly is defined as the locally deteriorated section, while the channel space outside this section continues to be judged as the normal section.

[0167] like Figure 5 This diagram illustrates the control system's process of sensing and locating spatial anomalies in microfluidic channels. The attached diagram is a two-dimensional pseudo-color image; the horizontal axis represents spatial location in millimeters; the vertical axis represents time in seconds; and the color bars represent temperature in Kelvin.

[0168] The color distribution in the figure reflects the changes in the temperature field inside the channel over time and space. Before 20 seconds, the temperature distribution in the channel is relatively uniform, maintaining a baseline value of around 380 Kelvin. After 20 seconds, brighter areas representing higher temperatures appear in a segment located 40 to 60 millimeters in space, with the local maximum temperature rising and exceeding 400 Kelvin. This change in color gradient reflects a change in the dispersion of heat distribution inside the channel.

[0169] When the calculated spatial temperature gradient variance is greater than the first preset threshold and the channel pressure drop change rate is greater than the second preset threshold, the control system, based on the distribution of temperature measurement data, defines the continuous spatial physical coordinate range with positive and negative gradient anomalies as a local deterioration section.

[0170] Step 3: Based on the spatial temperature gradient variance, calculate the heat transfer hindrance penalty factor and flow resistance surge factor corresponding to the local deterioration section.

[0171] Since the physical phases within the locally deteriorated section are drastically different from those in the normal section—for example, the disappearance of the gas-liquid interface caused by gas plugging can significantly reduce the local convective heat transfer coefficient—it is necessary to generate a specific penalty coefficient for this section.

[0172] For example, the heat transfer hindrance penalty factor is generated using an exponentially decaying mapping function, and the specific calculation formula is as follows:

[0173] ;

[0174] in, This represents the heat transfer hindrance penalty factor corresponding to the localized deterioration section; The thermodynamic penalty coefficient, which represents the system's preset sensitivity to heat transfer degradation, is used to indicate the system's pre-defined sensitivity. This refers to the aforementioned first preset threshold; This represents an exponential operation with the natural constant as the base. The physical meaning of this formula is that the more severe the degradation (characterized by the variance exceeding a threshold), the more significantly the theoretical thermal conductivity of that local region decreases, and the penalty factor gradually approaches zero. Meanwhile, the flow resistance surge factor is generated using a linear proportional mapping function, and the specific calculation formula is as follows:

[0175] ;

[0176] in, This represents the flow resistance surge factor corresponding to the localized deterioration section; This represents the system's preset dynamic penalty coefficient, characterizing its sensitivity to surges in fluid resistance; This represents the rate of change of pressure drop at the current moment; This represents the aforementioned second preset threshold. This calculation logic ensures that the more severe the local blockage, the more the resistance factor amplifies linearly.

[0177] Step 4: Decompose the coupled correlation equation into a first correlation sub-equation and a second correlation sub-equation in the spatial solution dimension.

[0178] It should be noted that in conventional model predictive control, the coupled correlation equations involving heat conduction and flow distribution between channels are a set of partial differential equations that span the entire length of the microchannel. However, in order to isolate the effects of degradation, the solver of the industrial control system, after locating the physical boundary of the locally degraded section, forcibly cuts off the original partial differential solution domain in the spatial dimension.

[0179] The first correlation equation is specifically used to describe the momentum, mass, and heat transfer relationships within the locally deteriorated section; the second correlation equation is used to further describe the transfer relationships in other normal sections outside the locally deteriorated section. This decomposition in the spatial solution dimension fundamentally avoids the mathematical diffusion of local anomalous parameters into the global normal region, that is, it avoids the false smoothing phenomenon in the numerical solution process.

[0180] Step 5: Substitute the heat transfer hindrance penalty factor, the flow resistance surge factor, the updated basic reaction rate, and the coupling interference value into the first correlation equation to calculate the first predicted sub-state of the localized deterioration segment.

[0181] It is important to note that when substituting the first correlation equation, the original equation structure undergoes a fundamental change in the coefficient matrix. The standard local convective heat transfer coefficient in the original mechanistic model is multiplied by the heat transfer hindrance penalty factor, thus being significantly weakened; the original standard flow channel friction coefficient is multiplied by the flow resistance surge factor, thus being significantly amplified. The industrial control system inputs these severely penalized physical constants, along with the updated basic reaction rate derived in Example 1 and the coupling interference value, into the solver. Using the finite difference method or the finite volume method, it independently solves for the discrete state variables of the locally deteriorated section in the future finite time domain. The solved local extreme temperature field distribution vector and the local component concentration distribution vector together constitute the first predicted sub-state.

[0182] Step 6: Use the coupled correlation equation as the second correlation sub-equation to calculate the second predicted sub-state of the normal segment, and numerically concatenate the first predicted sub-state and the second predicted sub-state on the spatial boundary to obtain the reaction state change.

[0183] Since no airlock or severe scaling occurs in the normal section, the industrial control system directly uses the original coupled correlation equation without penalty factor as the second correlation sub-equation to perform conventional partial differential evolution calculations, thereby obtaining the second predicted sub-state.

[0184] After solving for the states of the two subdomains separately, numerical stitching is necessary to restore the macroscopic physical continuity of the entire channel. During this stitching process, the continuity boundary conditions of the microfluidic channel at the disassembly boundary must be satisfied. Specifically, the system enforces the application of first-type boundary conditions (Dirichlet boundary conditions, ensuring absolute continuity of temperature and concentration scalars) and second-type boundary conditions (Neumann boundary conditions, ensuring conservation of heat flux and mass flux) at the numerical stitching interface.

[0185] By solving this multi-domain spliced ​​boundary continuity equation set, the industrial control system finally reconstructs a complete full-channel parameter prediction curve containing local steep peaks (abnormal hot spots) and abnormal pressure drop steps. This curve serves as the final change in the reaction state and is output to the subsequent target priority module and model predictive control module.

[0186] Example 3:

[0187] In actual industrial deployments of microfluidic hydrogenation reaction systems, multiple microchannels are typically arranged in parallel and share the same main feed manifold and main drain channel. Under this highly physically coupled hardware architecture, the conventional model predictive control objective optimization function constructed in Example 1 can effectively coordinate the conversion rate and capacity of each channel when the system is in a steady state.

[0188] However, when a parallel channel enters a local surge condition of rapid deterioration of side reactions due to localized catalyst bed damage or fluid deviation, the conventional linear weighted optimization algorithm, in order to minimize the penalty term of the objective function for that channel, will instantaneously output an extreme feed reduction command. This instantaneous, large-scale valve closure not only causes the fluid kinetic energy to be rapidly converted into pressure potential energy, triggering a strong hydrodynamic shock in the shared feed manifold, but also instantly disrupts the fluid pressure drop steady state of the other normal parallel channels, leading to the paralysis of the entire production system.

[0189] Therefore, in the process of solving the objective optimization function to obtain the optimal control parameters, this embodiment introduces a dynamic monitoring and objective optimization function anti-impact structure reconstruction mechanism, which aims to suppress local crises while ensuring the physical and hydraulic stability of the multi-channel shared pipeline from a mathematical perspective.

[0190] For example, based on Embodiment 1 and Embodiment 2, this embodiment further illustrates the process of solving the objective optimization function to obtain the optimal control parameters. This process specifically includes the following steps:

[0191] Step 1: In each discrete control cycle, the industrial control system continuously calculates the time rate of change of the by-product yield in each channel.

[0192] In microfluidic hydrogenation reaction systems, byproduct yield not only characterizes the reaction selectivity but also serves as a direct precursor signal to whether local thermal runaway is occurring within the channel. Conventional absolute numerical monitoring suffers from the physical limitation of alarm lag; therefore, industrial control systems introduce the first derivative in the time dimension as a predictive monitoring indicator.

[0193] Optionally, considering the unavoidable high-frequency electromagnetic noise and background fluctuations of detection instruments in the industrial field data acquisition link, directly using the difference between two adjacent single-step control cycles to calculate the derivative is highly likely to cause false triggering of the system. Therefore, the industrial control system uses a weighted difference algorithm based on a sliding time window to calculate the time change rate of by-product yield, and the corresponding calculation formula is:

[0194] ;

[0195] in, Indicates the first Each channel is in The rate of change of the by-product yield calculated at each moment; Indicates the first Each channel is in The current byproduct yield is obtained in real time after processing by online spectral detection hardware and data preprocessing module; Indicates the first Byproduct yield of each channel at historical moments within a sliding time window; This represents the inferred index sequence of historical data points within the sliding time window; this index is used to trace back historical states. This represents the total depth of the preset sliding window, indicating the span of the system's historical data tracing. This indicates the discrete control cycle step size in an industrial control system. Indicates the corresponding to the first The time decay weighting coefficient for each historical data point is set, and the closer the historical data point is to the current time, the larger the value of the weighting coefficient, exhibiting an exponential decreasing characteristic.

[0196] Using this formula, industrial control systems can accurately and smoothly extract the true acceleration trend of by-product generation rate while filtering out high-frequency jitter interference at the detection front end.

[0197] Step two: The industrial control system performs a threshold comparison on the calculated time change rate. When it is found that the time change rate of the by-product yield of the target channel is greater than the third preset threshold, it determines that the target channel has entered a local surge condition and triggers a program to reconstruct the target optimization function.

[0198] In this step, the target channel refers to a specific single channel or a set of multiple channels in a multi-channel microfluidic reactor that is currently monitored by the industrial control system as exhibiting an abnormally accelerated deterioration in the development trend of side reactions. The third preset threshold is a safety boundary derivative value that the system calibrates offline based on the design heat dissipation limit of the microfluidic channel, the catalyst heat capacity, and historical accident data.

[0199] It is also important to note that when the time-varying rate of the byproduct yield in the target channel exceeds the third preset threshold, it means that the exothermic rate of the side reaction inside the channel has exceeded the physical heat dissipation rate of the channel's microstructure, and the system is at a critical point of deteriorating positive feedback. At this point, if the conventional objective optimization function containing a linear penalty term in Example 1 is continued to be used, the algorithm will exhibit a constant marginal utility characteristic of the control strength, and will be unable to exert a strong restraining force on the state that is about to approach the safety limit. Therefore, the system must reconstruct the topology of the objective optimization function from linear to nonlinear at this moment.

[0200] Step three, the first core point of the industrial control system's reconfiguration procedure is to replace the linear byproduct penalty term for the target channel in the objective optimization function with an exponential barrier penalty term.

[0201] In the original objective optimization function of Example 1, the penalty term for by-products is expressed as a linear product, that is, the product of the optimization weight coefficient and the by-product yield. The linear penalty term has a constant partial derivative mathematically, meaning that regardless of whether the by-product yield is low or approaching its explosive limit, the mathematical "gain" gained by the optimization solver for each unit reduction in by-products is exactly the same. In the game of multi-objective optimization, this constant gain is easily masked by conversion rate or production capacity targets, causing the solver to compromise on output and allow risks to occur.

[0202] Optionally, the system replaces the original linear terms with a specific exponential mapping function, and the reconstructed exponential barrier penalty term is expressed as:

[0203] ;

[0204] in, This represents the exponential barrier penalty term reconstructed for the target channel; This represents the preset barrier weight coefficient, used to adjust the baseline magnitude of this nonlinear penalty term in the global reconstruction objective optimization function; This indicates the byproduct yield of the target channel within the current prediction domain; This represents the preset safe limit for byproduct yield, which characterizes the physical red line of byproduct concentration that the microfluidic system can tolerate at the physical level.

[0205] This exponential barrier penalty term exhibits excellent state repulsion characteristics at the mathematical level. When the by-product yield of the target channel is within a low safe range, the value of this term is relatively flat and will not excessively interfere with the system's normal optimization of production capacity and conversion rate; however, once the by-product yield of the target channel approaches the preset safe limit value for by-product yield, the denominator term will approach zero, causing the calculation result of the entire exponential term to geometrically explode towards positive infinity.

[0206] Since the ultimate goal of the solver is to find the set of control input parameters that minimizes the global objective function, this penalty of approaching infinity will forcibly deprive production capacity and conversion rate of their say in the game, forcing the solver to prioritize outputting control strategies that reduce by-product yields at all costs, thereby effectively ensuring the chemical safety of the system at the algorithm level.

[0207] like Figure 6 This diagram illustrates the difference in the impact of different penalty terms on the control objective before and after the objective optimization function is reconstructed. The horizontal axis of the attached figure represents the by-product yield (dimensionless); the vertical axis represents the penalty term output value (dimensionless).

[0208] The vertical dotted line in the figure is the red safety limit value marked in the legend, with a value of 1.0. The light blue dashed line represents the linear byproduct penalty term before reconstruction, with a straight waveform. The penalty term output value changes linearly and proportionally with the byproduct yield. The dark red solid line represents the exponential barrier penalty term curve generated by reconstruction.

[0209] When the by-product yield is in a low range, such as 0.5, the value of the dark red solid line changes relatively smoothly. When the by-product yield exceeds 0.8 and approaches the safety limit of 1.0, the denominator of the exponential barrier penalty term decreases, causing the dark red curve to exhibit a non-linear increasing shape, with its output value rising and exceeding 100. This non-linear waveform characteristic causes the optimization solver to adjust the weight allocation of the by-product reduction control strategy during the calculation.

[0210] Step four, the second core point of the industrial control system's reconfiguration procedure, is to add an asymmetric shock suppression term to the objective optimization function.

[0211] As mentioned earlier, when the solver faces an exponentially large barrier penalty term approaching infinity, it tends to output extremely forceful feed valve closing commands within a single control cycle in order to quickly escape the danger zone. In the hydraulic model of a microfluidic system sharing a manifold, drastic changes in flow velocity will generate extremely high transient water hammer pressure rises according to the principle of transient fluid flow. If this valve closing amplitude is not constrained, the measure to protect one channel will directly destroy the entire parallel network.

[0212] For example, the industrial control system superimposes a new penalty dimension into the reconstructed objective optimization function, and the specific calculation formula for the asymmetric shock suppression term is as follows:

[0213] ;

[0214] in, This represents the newly added asymmetric shock suppression term in the objective optimization function; This represents the preset impact suppression weight coefficient, used to adjust the sensitivity of hydraulic safety constraints in the objective function; This represents the control output value of the target channel, which is deduced and proposed by the solver within the control cycle. When the value is positive, it indicates an increase in the feed flow rate; when the value is negative, it indicates a decrease in the feed flow rate. This represents the maximum value function, used to output the larger of two input parameters; This represents the preset safe feed reduction threshold. This value is the maximum absolute physical value of the flow reduction allowed in a single control cycle without causing water hammer in the pipeline, calculated offline based on the material stiffness of the main manifold, the fluid compressibility coefficient, and the wave velocity equation. It is a positive number.

[0215] It is important to note that when the system needs to open the valve to increase the feed (i.e., the planned feed control increment is positive) or the valve closes slowly (i.e., the absolute value of the planned feed control increment is less than the set feed safety reduction threshold), the result of the internal parameter calculation in the formula is negative. After processing by the maximum value function, the output of the entire asymmetric impact suppression term is constant at zero. This means that the control system does not impose any additional penalty on safe adjustment actions.

[0216] However, when the solver attempts to issue a drastic valve-closing command, causing the feed control increment to become negative and its absolute value to exceed the feed safety reduction threshold, the internal calculation result turns positive. At this point, the portion exceeding the safety boundary is nonlinearly amplified through a squaring operation and applied to the global objective function. This rigid-softening quadratic penalty structure allows the solver to find an effective balance between maximizing pressure reduction byproducts and considering hydraulic penalties.

[0217] like Figure 7 This figure illustrates the system's transient response before and after adding an asymmetric shock suppression term to the objective function. The attached figure consists of two subplots, with the horizontal axis of both subplots representing time in seconds.

[0218] The vertical axis of the upper subplot represents the feed velocity, in milliliters per second; the vertical axis of the lower subplot represents the hydraulic pressure drop, in kilopascals.

[0219] The light blue dashed line represents the state without the introduction of the suppression term. In the upper subplot, the feed flow rate decreases from 50 ml / s to 10 ml / s in the second second. In the corresponding lower subplot, the hydraulic pressure drop curve shows a numerical increase, reaching a peak of 800 kPa, and then exhibits an oscillating waveform.

[0220] The dark red solid line represents the response curve after the introduction of the asymmetric impact suppression term. This mechanism imposes a constraint on the negative control increment exceeding the set threshold. Therefore, the feed velocity curve in the upper subgraph shows a gradual downward trend. In the corresponding lower subgraph, the oscillation amplitude of the hydraulic pressure drop curve decreases, and the maximum pressure value transitions smoothly and stabilizes at around 380 kPa.

[0221] This comparison demonstrates that the asymmetric impact suppression term limits the single adjustment range of the feed control command, reducing the fluctuation range of water hammer pressure rise in the pipeline network.

[0222] Step five: After the reconstruction process involving the replacement of penalty terms and the superposition of new constraints, the industrial control system obtains a complex objective optimization function that includes nonlinear barriers and asymmetric damping. At this point, the system calls the underlying optimizer to solve the reconstructed objective optimization function and generate the optimal control parameters.

[0223] Alternatively, since the reconstructed objective function has high nonconvexity and complex exponential nonlinear boundaries, traditional linear programming algorithms or simple gradient descent methods cannot guarantee the convergence of the solution and may even easily get trapped in local dead zones.

[0224] Therefore, the industrial control system is configured with a nonlinear solution engine driven by a combination of interior-point method and sequential quadratic programming in this step. Within the prediction time domain of each control step, the solution engine uses the feed control increment and heating power increment as decision variable vectors. Within a hard constraint space that satisfies the physical limits of the actuators (such as the upper and lower limits of the micro-pump speed), it iteratively searches for a sequence of decision variables that minimizes the reconstructed objective function. Subsequently, the first component of this decision variable sequence is extracted as the optimal control parameter actually issued in the current control cycle, driving the corresponding electronically controlled valve and heating actuator.

[0225] Example 4:

[0226] like Figure 8As shown, this embodiment provides an industrial control system parameter optimization system for microfluidic hydrogenation reactions. This system aims to solve the underlying technical challenges of multi-channel collaborative control in complex chemical scenarios.

[0227] Microfluidic hydrogenation reactions are widely used in fine chemical engineering due to their extremely high mass and heat transfer efficiency and intrinsic safety of continuous flow. However, with the increasing scale of parallel channels in industrial-grade microfluidic systems, the highly nonlinear gas-liquid-solid three-phase reaction kinetics within the channels are intertwined with strong heat and flow coupling interference between channels. Existing conventional control systems, relying on idealized macroscopic mechanism models, reveal a serious conflict between the smooth prediction of the global model and the abrupt nonlinear distortions of local physical changes in practical engineering, which can easily lead to local thermal runaway or gas blockage.

[0228] In addition, existing control systems often output extreme single-channel intervention commands when dealing with local sudden anomalies, thereby triggering destructive hydraulic shocks in the shared pipeline and disrupting the physical steady state of the global system.

[0229] To overcome the aforementioned shortcomings, this embodiment provides an industrial control system parameter optimization system for microfluidic hydrogenation reactions, which is constructed based on a programmable logic controller (PLC), a distributed control system (DCS), and processing algorithms deployed on edge computing nodes. The system is characterized by including: a data acquisition and preprocessing module, a hybrid prediction module, a target priority determination module, and a collaborative control optimization module.

[0230] First, the system includes a data acquisition and preprocessing module, which is used to acquire the reaction parameters within the channels, the coupling parameters between channels, and the load state parameters of the multi-channel microfluidic reactor, and to preprocess them to obtain characteristic data.

[0231] In this module, data acquisition is achieved through a high-precision sensor array deployed at the bottom layer and an industrial bus communication protocol. The reaction parameters acquired by the system within the channel include not only macroscopic inlet and outlet temperatures and pressures, but also microscopic temperature field distributions at multiple nodes along the channel axis and state quantities such as component concentrations fed back by an online spectrometer. The inter-channel coupling parameters aim to quantify the physical interference between parallel channels, mainly including the thermal conduction temperature difference gradient at the interface of adjacent channels and the real-time distribution ratio of the main feed pipe to each branch. The load state parameters are used to characterize the physical health of the system, such as the time-dependent degradation index of catalyst activity and the rate of change in flow resistance caused by scaling on the inner wall of the pipeline.

[0232] Since the aforementioned multi-source heterogeneous data have significant differences in sampling frequency and numerical magnitude, the data acquisition and preprocessing module is further configured with a preprocessing unit. By performing operations such as time series alignment, dimensionless standardization, and Kalman filtering, high-frequency electromagnetic noise and fluid pulsation interference from the industrial site are eliminated, thereby outputting high-fidelity feature data with a unified mathematical scale, laying a reliable data foundation for subsequent advanced calculations.

[0233] Secondly, the system includes a hybrid prediction module, which inputs the feature data into a pre-built hybrid prediction model to predict the reaction state changes of each channel and the coupling interference between channels.

[0234] This module is the core computational engine that resolves the conflict between global smooth prediction and local extreme mutations. Purely mechanistic models are prone to distortion under extreme conditions, while purely data-driven black-box models lack physical constraints. Therefore, in this embodiment, the hybrid prediction module specifically performs the following three dimensions of processing when executing predictions:

[0235] First, the basic reaction rate and basic mass transfer rate under uncoupled conditions are calculated using pre-defined mechanistic equations. This sub-unit incorporates chemical kinetic equations based on Arrhenius's law and mass transfer equations based on the two-film theory. Using pre-processed characteristic data, and assuming that each channel operates ideally and independently without external interference, the fundamental physical quantities supported by rigorous thermodynamic and kinetic theories are calculated.

[0236] Second, the feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value of the corresponding channel and the coupling interference value between the channels. This sub-unit relies on a deep learning network architecture (such as a long short-term memory network) to specifically model the dark physical dynamics that are difficult to describe precisely using conventional mechanistic equations. Its output reaction state drift value characterizes the nonlinear deviation between the actual state and the theoretical state caused by changes in the catalyst's microstructure or the heterogeneous distribution of the fluid; the coupling interference value quantifies the parasitic heat loss and material flow deviation effects caused by the close spacing between channels.

[0237] Third, the reaction state drift value and the coupling interference value are used to fuse and correct the basic reaction rate and the basic mass transfer rate to obtain the reaction state change. This sub-unit multiplicatively applies the hidden dynamic features extracted from the data network, in the mathematical form of a nonlinear penalty term or compensation factor, to the theoretical values ​​calculated by the mechanistic equation. Through this fusion correction architecture of "mechanism-based and data-optimized", the hybrid prediction module can effectively detect extreme anomalies such as local micro-gas plug accumulation and accurately output the reaction state changes and coupling distribution that approximate the complex working conditions of real industrial sites, significantly improving the prediction fidelity of the model in localized deterioration sections.

[0238] Furthermore, the system includes a target priority determination module, which calculates the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient based on the change in the reaction state and the coupling interference, so as to determine the control target priority of each channel.

[0239] In multi-channel microfluidic systems, each channel faces an internal trade-off between improving conversion efficiency and reducing byproducts, while the system as a whole also needs to balance the optimization of individual channels with the uniformity of output across multiple channels. This module uses a specific multi-objective evaluation function to calculate the conflict trade-off coefficient for each single channel to assess the urgency of its current process conditions deviating from the safe or optimal window; simultaneously, it calculates the overall trade-off coefficient for multiple channels to monitor the conversion rate range and load imbalance among the parallel branches of the entire system. Based on the cross-comparison of these two-dimensional coefficients, the module can dynamically assign the current control objective priority to each channel.

[0240] For example, when a channel is at risk of thermal runaway, the suppression of side reactions within that channel is given the highest priority; while when the system is in a stable state, the uniformity of material distribution across multiple channels is given a high priority. This module effectively resolves resource scheduling conflicts, enabling the system control to possess adaptive flexibility and intelligence.

[0241] Finally, the system includes a collaborative control optimization module, which generates optimal control parameters for each channel based on the control target priority and the feature data using a model predictive control algorithm, and issues the parameters for execution to control the multi-channel microfluidic reactor.

[0242] This module receives the high-precision predicted state and priority weights output by the preceding module and performs multivariate rolling optimization within a finite prediction time domain. Specifically, this module is equipped with a nonlinear optimization solver that dynamically reconstructs the target optimization function based on the determined control objective priority. Especially when dealing with local critical conditions, this module strictly limits the severity of control actions by introducing specific barrier constraints and damping constraints on the actuator's amplitude at the algorithm's underlying level. Under the premise of satisfying the aforementioned hard and soft constraints of reconstruction, the solver calculates the specific optimal control parameter sequence, such as the heating duty cycle, the speed of the micro-infusion pump, and the opening degree of the mass flow controller.

[0243] Through the instruction issuance and hardware execution of this collaborative control optimization module, the system can not only accurately and smoothly curb the abnormal surge of by-products at the single channel level, but also significantly reduce the possibility of strong hydraulic shocks caused by the explosion-proof valve closure of a single channel in the shared feed manifold due to the control of instruction mutation amplitude in the algorithm.

[0244] Overall, the parameter optimization system described in this embodiment, driven by both data and mechanisms, achieves high-precision anti-disturbance and global process stability by systematically calculating spatial dimensionality reduction and dynamic target trade-offs through systematic operations. This is achieved while ensuring the hydraulic safety boundary of the multi-channel microfluidic hardware pipeline, resulting in significant beneficial effects for industrial applications.

[0245] Example 5:

[0246] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0247] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0248] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0249] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0250] The memory 103 stores a computer program corresponding to a parameter optimization method for an industrial control system for microfluidic hydrogenation reactions according to the above embodiments of the present invention. This computer program is executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0251] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0252] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optimizing parameters of an industrial control system for microfluidic hydrogenation reactions, characterized in that, include: The reaction parameters within the channels, the coupling parameters between channels, and the loading state parameters of the multi-channel microfluidic reactor were obtained, and the characteristic data were obtained by preprocessing. The feature data is input into a pre-built hybrid prediction model to predict the change in the reaction state of each channel and the coupling interference between channels; Based on the changes in the reaction state and the coupling interference, the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient are calculated respectively to determine the control target priority of each channel. Based on the control target priority and the feature data, the optimal control parameters for each channel are generated by the model predictive control algorithm and then executed to control the multi-channel microfluidic reactor. The prediction process of the hybrid prediction model includes: The basic reaction rate and basic mass transfer rate under uncoupled conditions are calculated using a pre-defined mechanistic equation. The feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value of the corresponding channel and the coupling interference value between the channels. The reaction state drift value and the coupling interference value are used to fuse and correct the basic reaction rate and the basic mass transfer rate to obtain the reaction state change.

2. The method according to claim 1, characterized in that, The process of preprocessing to obtain feature data includes: The intra-channel reaction parameters, inter-channel coupling parameters, and load state parameters are time-aligned using linear interpolation. Through standardized formulas Eliminating dimensional differences, among which The values ​​are standardized. For the first Each channel is in Parameter measurements at time [time] and These are the mean and standard deviation of the parameters within the preset data window, respectively. The Kalman filter algorithm is used to perform multi-source data fusion on the data after eliminating dimensional differences, and the feature data is output.

3. The method according to claim 1, characterized in that, The process of fusing and correcting the basic reaction rate and the basic mass transfer rate using the reaction state drift value and the coupling interference value includes: Based on formula Corrected reaction kinetic constants, where The corrected reaction kinetic constants are... These are the uncoupled dynamic constants calculated using the aforementioned mechanistic equations. The preset fusion weight coefficients, The reaction state drift value output by the data compensation network; The modified reaction kinetic constant is substituted into the mechanistic equation to update the basic reaction rate, and the updated basic reaction rate and the coupling interference value are substituted into the coupling correlation equation that includes inter-channel heat conduction and flow distribution to calculate the reaction state change.

4. The method according to claim 1, characterized in that, The calculation of the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient includes: According to the formula Calculate the single-channel conflict tradeoff coefficient. ,in The deviation of the channel from the preset attenuation threshold window. This represents the catalyst activity decay rate. and These are the corresponding preset deviation threshold and preset attenuation threshold, respectively. Normalized value for by-products; According to the formula Calculate the overall tradeoff coefficient of the multi-channel system. ,in This represents the current range of reaction conversion rates across multiple channels. The preset safety tolerance threshold, For the balanced weighting coefficients, This represents the total number of parallel channels in the microfluidic reactor. This is the upper limit of the maximum single-channel collision coefficient allowed historically or theoretically.

5. The method according to claim 1, characterized in that, Before generating the optimal control parameters for each channel using a model predictive control algorithm based on the control target priority and the feature data, the method further includes: Identify the intra-channel measurement lag time and inter-channel interference lag time for each channel; The feature data is pre-corrected using a two-stage compensation algorithm based on the rate of change of state, resulting in corrected feature data. The step of generating optimal control parameters for each channel based on the control target priority and the feature data using a model predictive control algorithm includes: Based on the control target priority and the corrected feature data, the optimal control parameters for each channel are generated by the model predictive control algorithm.

6. The method according to claim 1, characterized in that, The generation of optimal control parameters for each channel using a model predictive control algorithm includes: Construct an objective optimization function that includes reaction conversion rate, by-product indicators, target yield constraints, and multi-channel mean square error: ； in, and The first Actual predicted conversion rate and by-product rate for each channel and The first The expected output and target output of each channel This represents the average conversion rate predicted across multiple channels. to To optimize the weighting coefficients; the single-channel conflict trade-off coefficients and the multi-channel overall trade-off coefficients are combined to optimize the weighting coefficients. to The parameter sensitivity is dynamically allocated, and the optimal control parameters are obtained by solving the objective optimization function.

7. The method according to claim 6, characterized in that, The process of solving the objective optimization function to obtain the optimal control parameters includes: Calculate the rate of change of the by-product yield in each channel over time; When the time rate of change of the byproduct yield of the target channel is found to be greater than a third preset threshold, the target optimization function is reconstructed: The linear byproduct penalty term for the target channel in the objective optimization function is added. Replace with exponential barrier penalty item ,in The preset barrier weight coefficient, This is the preset safety limit value for by-product yield; Add an asymmetric shock suppression term to the objective optimization function ,in The preset impact suppression weighting coefficient, This refers to the feed control increment of the target channel in the current control cycle. The preset threshold for the safe reduction of feed rate; The reconstructed objective optimization function is used to generate the optimal control parameters.

8. The method according to claim 1, characterized in that, After the execution of the command to control the multi-channel microfluidic reactor, the method further includes: Continuously collect actual response indicators after executing the optimal control parameters; Calculate the prediction error between the actual reaction index and the change in reaction state; When the single-channel prediction error or the inter-channel prediction error exceeds the respective preset error convergence threshold, a joint algorithm of recursive least squares and sequential quadratic programming is used to update the physical coupling coefficients in the hybrid prediction model and the weight coefficients of the data compensation network online.

9. The method according to claim 3, characterized in that, The process of substituting the updated base reaction rate and the coupling interference value into the coupling correlation equation that includes inter-channel heat conduction and flow distribution to calculate the change in reaction state includes: Acquire temperature measurement data from multiple spatially distributed temperature measurement points within the channel reaction parameters, and calculate the variance of the spatial temperature gradient along the channel spatial dimension; When the variance of the spatial temperature gradient is greater than a first preset threshold and the rate of change of the pressure drop in the channel is greater than a second preset threshold, the local deterioration section is located based on the distribution of the temperature measurement data. Based on the spatial temperature gradient variance, calculate the heat transfer hindrance penalty factor and flow resistance surge factor corresponding to the local deterioration section; The coupled correlation equation is split into a first correlation sub-equation and a second correlation sub-equation in the spatial solution dimension; Substitute the heat transfer hindrance penalty factor, the flow resistance surge factor, the updated base reaction rate, and the coupling interference value into the first correlation equation to calculate the first predicted sub-state of the local deterioration segment. The coupled correlation equation is used as the second correlation sub-equation to calculate the second predicted sub-state of the normal segment, and the first predicted sub-state and the second predicted sub-state are numerically concatenated on the spatial boundary to obtain the reaction state change.

10. A parameter optimization system for an industrial control system for microfluidic hydrogenation reactions, characterized in that, include: The data acquisition and preprocessing module is used to acquire the reaction parameters within the channels, the coupling parameters between channels, and the load state parameters of the multi-channel microfluidic reactor, and to preprocess them to obtain characteristic data. The hybrid prediction module is used to input the feature data into a pre-built hybrid prediction model to predict the reaction state changes of each channel and the coupling interference between channels; The target priority determination module is used to calculate the single-channel conflict trade-off coefficient and the multi-channel overall trade-off coefficient based on the reaction state change and the coupling interference, so as to determine the control target priority of each channel. The collaborative control optimization module is used to generate the optimal control parameters for each channel based on the control target priority and the feature data through a model predictive control algorithm, and then issue the parameters for execution to control the multi-channel microfluidic reactor. Specifically, the hybrid prediction module, when performing prediction, is used for: The basic reaction rate and basic mass transfer rate under uncoupled conditions are calculated using a pre-defined mechanistic equation. The feature data is input into a pre-trained data compensation network, which outputs the reaction state drift value of the corresponding channel and the coupling interference value between the channels. The reaction state drift value and the coupling interference value are used to fuse and correct the basic reaction rate and the basic mass transfer rate to obtain the reaction state change.

Images