Large-scale fluid extraction separation intelligent control method based on mixing tank structure

Abstract

The invention provides a large-scale fluid extraction and separation intelligent control method based on a mixing tank structure, and relates to the technical field of intelligent control, and the method comprises the steps: collecting parameter information in a mixing tank through double-cycle sampling, dividing flow field partitions based on three-dimensional numerical characteristics, collecting concentration gradient information of a target object, and building a digital twinborn model; executing a partition self-adaptive control strategy to dynamically adjust stirring parameters, controlling extraction agent supplement, adjusting the opening degree of a partition plate, and monitoring and updating the control strategy in real time. According to the method, precise zone control is realized, the extraction efficiency is improved, the energy consumption is reduced, and the method adapts to complex working condition changes.

Description

TECHNICAL FIELD

[0001] The present application relates to intelligent control technology, and in particular to a large-scale fluid extraction separation intelligent control method based on a mixed tank structure. BACKGROUND

[0002] Fluid extraction separation technology is a commonly used material separation and purification method in chemical, metallurgical, pharmaceutical and other industries. In traditional mixed tank fluid extraction separation control, single parameter sampling and global uniform stirring are generally used. The operator usually sets the stirring speed and extractant addition amount according to experience, and lacks accurate grasp of the internal flow field distribution and local mass transfer efficiency of the mixed tank. The prior art has the following defects and deficiencies in the intelligent control of large-scale fluid extraction separation: Existing extraction separation usually uses single cycle parameter sampling method, which cannot simultaneously consider rapid response capability and steady state observation accuracy, and it is difficult to realize accurate control.

[0003] The prior art usually regards the mixed tank as a single flow field area for overall control, ignoring the non-uniformity of the flow field inside the large mixed tank. Due to the complexity of the fluid dynamics characteristics inside the mixed tank, it is difficult for a single stirring device to form a uniform flow field distribution in the entire mixed tank, resulting in significant differences in extraction efficiency in different areas of the tank, which limits the improvement of overall extraction efficiency.

[0004] The existing control method uses pre-set fixed parameters for operation, and cannot automatically adjust the control strategy according to the concentration gradient changes in the actual extraction process. When the raw material composition, environmental temperature and other conditions change, the stirring parameters and extractant addition strategy cannot be optimized in time, resulting in energy waste and extraction efficiency decline. SUMMARY

[0005] The embodiment of the present application provides a large-scale fluid extraction separation intelligent control method based on a mixed tank structure, which can solve the problems in the prior art.

[0006] In a first aspect, the embodiment of the present application provides a large-scale fluid extraction separation intelligent control method based on a mixed tank structure, comprising: Collecting real-time parameter information of the mixed tank of the extractant and the material mixture, and performing double-cycle sampling on the real-time parameter information according to a fast response cycle and a steady state observation cycle; Based on the double-cycle sampling data, calculate the three-dimensional numerical characteristics of the flow field distribution in the mixed tank, divide the mixed tank into multiple flow field partitions according to the three-dimensional numerical characteristics, and set independent stirring devices in each flow field partition; Real-time acquisition of target material concentration gradient information of each flow field partition, establishment of a digital twin model based on the target material concentration gradient information, and output of fluid flow characteristics and mass transfer efficiency prediction values of each partition by the digital twin model; According to the output of the digital twin model, a partition adaptive control strategy is executed to dynamically adjust the stirring speed and stirring direction of each partition, automatically control the extraction agent replenishment amount and replenishment time of each partition, and adjust the baffle opening degree according to the concentration gradient of adjacent partitions; The extraction efficiency of each partition is monitored in real time, and when an abnormal extraction efficiency is detected, the partition adaptive control strategy is updated based on the latest collected target concentration gradient information until the preset extraction efficiency is reached.

[0007] The step of collecting real-time parameter information of the mixed liquid of the extraction agent and the material in the mixing tank and performing double-period sampling on the real-time parameter information according to a fast response period and a steady-state observation period comprises: The real-time parameter information includes the pH value, temperature, concentration and target content of the mixed liquid; Based on the real-time parameter information, a parameter response time constant is calculated, and the pH value and temperature are divided into a fast response parameter group, and the concentration and target content are divided into a steady-state parameter group; The parameter change rate and parameter acceleration of the fast response parameter group are obtained, and a basic sampling frequency is calculated; the process disturbance intensity and disturbance change rate of the mixed liquid are detected, and the basic sampling frequency is adjusted to obtain a disturbance response frequency; the parameter weight corresponding to each parameter in the fast response parameter group is calculated, and the parameter weight and the corresponding disturbance response frequency are weighted to obtain a fast response sampling frequency; The parameter fluctuation variance of the steady-state parameter group is calculated, the ratio of the parameter fluctuation variance to the preset allowable fluctuation range is substituted into a logarithmic function to obtain a steady-state observation period correction coefficient; based on the correction coefficient, a preset reference period is adjusted to obtain an adaptive observation period, and the maximum value of the adaptive observation period corresponding to each parameter is selected as a steady-state sampling period; When the weighted sum of the parameter change rate and the parameter acceleration exceeds a first preset threshold, an exponential smoothing function is used to realize the transition sampling of the fast response sampling frequency to the steady-state sampling period.

[0008] Based on the double-period sampling data, the three-dimensional numerical characteristics of the flow field distribution in the mixing tank are calculated, and the step of dividing the mixing tank into multiple flow field partitions according to the three-dimensional numerical characteristics comprises: According to the fast response period data, the parameter space gradient characteristics of the spatial coordinate points of the mixing tank along the three-dimensional direction are calculated, based on the parameter space gradient characteristics, the local turbulent intensity is calculated, and based on the local turbulent intensity, the shear stress distribution of the spatial coordinate points is calculated; According to the steady-state observation period data, the parameter steady-state mean value distribution and the parameter fluctuation intensity distribution of the spatial coordinate points in the three-dimensional space are calculated, and based on the parameter steady-state mean value distribution and the parameter fluctuation intensity distribution, the mixing degree index is calculated; fuse the shear stress distribution and the mixing degree index to construct a multi-dimensional flow field feature vector, and calculate a multi-dimensional flow field feature vector similarity of adjacent spatial coordinate points; calculate a ratio of a variance to a mean of the multi-dimensional flow field feature vector, substitute the ratio into a logarithmic function to obtain a correction factor, and adjust a preset initial partition threshold value according to the correction factor to obtain an adaptive partition threshold value; divide the mixing tank into a plurality of flow field partitions based on the multi-dimensional flow field feature vector similarity and the adaptive partition threshold value, calculate a boundary gradient of the flow field partition, determine a partition boundary migration rate according to the boundary gradient, adjust a boundary position of the flow field partition based on the partition boundary migration rate, and perform smoothing processing on the adjusted partition boundary to obtain a final mixing tank flow field partition result.

[0009] establish a digital twin model based on the target substance concentration gradient information, and a step of outputting a fluid flow characteristic and a mass transfer efficiency prediction value of each partition by the digital twin model includes: collect target substance concentration data, calculate a concentration gradient of the target substance in the mixing tank by spatial difference, extract a gradient strength and a gradient direction angle of the concentration gradient, and construct a concentration gradient feature vector; input the concentration gradient feature vector into a residual neural network for feature extraction, perform feature mapping on an output result of the residual neural network according to a fluid flow law and a mass transfer law, and construct a digital twin model; based on the digital twin model, calculate a local shear stress and a vorticity distribution in the mixing tank as the fluid flow characteristic, and calculate a mass transfer coefficient to obtain the mass transfer efficiency prediction value; integrate learning of the fluid flow characteristic and the mass transfer efficiency prediction value by using a sliding time window, calculate a root mean square error of a prediction result, and dynamically update the digital twin model according to the root mean square error.

[0010] input the concentration gradient feature vector into a residual neural network for feature extraction, perform feature mapping on an output result of the residual neural network according to a fluid flow law and a mass transfer law, and construct a digital twin model; input the concentration gradient feature vector into a residual neural network, and output a deep feature representation; perform physical law mapping on the deep feature representation, respectively calculate a fluid momentum conservation deviation value and a fluid mass transfer diffusion deviation value, divide the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value by their historical maximum values to obtain normalized deviation values, and take a complementary value of the normalized deviation value as a physical mapping weight; constructing a physical rule weight matrix according to a ratio relationship between the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value, the physical rule weight matrix being used to represent a coupling relationship between different physical variables; encoding the deep feature representation based on the physical mapping weight to obtain a physical knowledge attention weight, and superimposing a product of the physical knowledge attention weight, the physical rule weight matrix and the deep feature representation to the deep feature representation to obtain a physical enhanced feature; inputting the physical enhanced feature into a mapping network to obtain a prediction result of a velocity field, a pressure field and a concentration field; calculating a model training loss value according to a fitting error between the prediction result and real data, and the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value in a sliding time window, the model training loss value including a root mean square error term; updating parameters of the residual neural network, the physical rule weight matrix and the mapping network based on the model training loss value, and dynamically optimizing the digital twin model.

[0011] According to the output of the digital twin model, a partition adaptive control strategy is executed to dynamically adjust the stirring speed and stirring direction of each partition, automatically control the extractant replenishment amount and replenishment time of each partition, and adjust the baffle opening degree according to the concentration gradient of adjacent partitions, which includes the following steps: According to the velocity field, pressure field and concentration field output by the digital twin model, the local Reynolds number and local Schmidt number of each flow field partition are calculated, the optimal stirring speed of each flow field partition is calculated based on the local Reynolds number and local Schmidt number, and the stirring direction angle is determined according to the concentration gradient of each flow field partition; The local mass transfer efficiency is calculated based on the local mass transfer coefficient and specific surface area of each flow field partition, the extractant replenishment amount is calculated according to the local mass transfer efficiency, and the extractant replenishment amount is related to the flow field partition volume, molecular diffusion coefficient and mixing time; the extractant replenishment time interval is determined based on the difference between the target concentration and the actual concentration and the concentration change rate; The concentration difference between adjacent flow field partitions is calculated, the diffusion flux is calculated according to the concentration difference and the molecular diffusion coefficient, the baffle opening degree adjustment coefficient is obtained by substituting the ratio of the diffusion flux to the reference diffusion flux into an exponential function, and the baffle opening degree between adjacent flow field partitions is determined based on the baffle opening degree adjustment coefficient; The mass transfer efficiency of each flow field partition is monitored, and when the mass transfer efficiency is lower than the target value, the control parameters of the corresponding flow field partition are adjusted to realize dynamic optimization of the stirring parameters, extractant replenishment and baffle opening degree of each flow field partition.

[0012] The local mass transfer efficiency is calculated based on the local mass transfer coefficient and specific surface area of each flow field partition, and the extractant replenishment amount is calculated according to the local mass transfer efficiency, which includes the following steps: The current concentration and the equilibrium concentration of each flow field partition are obtained, the relative difference between the current concentration and the equilibrium concentration is calculated, the characteristic size of the fluid micro-cluster is calculated based on the velocity field output by the digital twin model, the ratio of the characteristic size to the molecular diffusion coefficient is determined as the local mass transfer unit number, and the local mass transfer coefficient is calculated according to the local mass transfer unit number; the shear stress distribution of each flow field partition is collected, the droplet deformation degree is calculated based on the shear stress distribution, and the interface update coefficient is calculated according to the droplet deformation degree; the ratio of the interface update coefficient to the geometric size of the flow field partition is determined as the dynamic specific surface area; The local mass transfer efficiency is calculated according to the relative difference, the local mass transfer coefficient and the dynamic specific surface area; the mass transfer resistance coefficient is calculated based on the local mass transfer efficiency, and the mass transfer resistance coefficient exponentially decays with the increase of the local mass transfer efficiency; The product of the mass transfer resistance coefficient and a preset reference flow rate is determined as the extractant supplement amount, wherein the preset reference flow rate is equal to the volume of the flow field partition divided by the mixing time required to reach a preset mixing uniformity, and is corrected according to the molecular diffusion coefficient.

[0013] In a second aspect of the embodiments of the present application, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0014] In a third aspect of the embodiments of the present application, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0015] The present application realizes precise adaptive adjustment of the extraction process through the mixing tank partition control and the digital twin model, and improves the efficiency and stability of large-scale fluid extraction separation. The double-period sampling mechanism ensures that both the fluid dynamic changes can be quickly responded to and the long-term trend of the extraction process can be accurately grasped, solving the problem that the traditional single sampling method cannot balance the response speed and control stability.

[0016] The digital twin model of the present application can accurately predict the fluid characteristics and mass transfer efficiency of each partition, providing a reliable basis for adaptive control of the partition, changing the control strategy from experience-driven to data-driven, and greatly improving the control precision and adaptability. Through dynamic adjustment of the stirring parameters and extractant supplement scheme of each partition, the optimal allocation of resources is realized, and the energy consumption and extractant consumption are reduced.

[0017] The dynamic adjustment mechanism of the baffle opening effectively balances the concentration difference between adjacent partitions, eliminates the "dead zone" and "short circuit flow" phenomenon in the traditional mixing tank, and significantly improves the uniformity of mass transfer. The closed-loop mechanism of real-time monitoring and control strategy updating ensures that the extraction process can respond to various disturbances and abnormal conditions, improves the robustness and intelligent level, and is suitable for large-scale fluid extraction and separation applications in complex industrial environments. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of an embodiment of the application for intelligent control method of large-scale fluid extraction and separation based on mixing tank structure is shown in Figure 2 An extraction process partition adaptive control strategy execution flowchart based on a digital twin model. DETAILED DESCRIPTION

[0019] To make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0020] The technical scheme of the application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0021] Figure 1 A flowchart of an embodiment of the application for intelligent control method of large-scale fluid extraction and separation based on mixing tank structure is shown in Figure 1 The method comprises: Collecting real-time parameter information of the extractant and material mixture in the mixing tank, and performing double-period sampling on the real-time parameter information according to a fast response period and a steady-state observation period; Based on the double-period sampling data, calculating the three-dimensional numerical characteristics of the flow field distribution in the mixing tank, dividing the mixing tank into multiple flow field partitions according to the three-dimensional numerical characteristics, and setting independent stirring devices in each flow field partition; Real-time collection of target substance concentration gradient information of each flow field partition, establishment of a digital twin model based on the target substance concentration gradient information, and output of flow characteristics and mass transfer efficiency prediction values of each partition by the digital twin model; According to the output of the digital twin model, a partition adaptive control strategy is executed to dynamically adjust the stirring speed and stirring direction of each partition, automatically control the extraction agent replenishment amount and replenishment time of each partition, and adjust the baffle opening degree according to the concentration gradient of adjacent partitions. The extraction efficiency of each partition is monitored in real time, and when an abnormal extraction efficiency is detected, the partition adaptive control strategy is updated based on the latest collected target substance concentration gradient information until the preset extraction efficiency is reached.

[0022] In an optional embodiment, the step of collecting real-time parameter information of the mixed liquid of the extraction agent and the material in the mixing tank and performing double-period sampling on the real-time parameter information according to a fast response period and a steady-state observation period comprises: The real-time parameter information includes pH value, temperature, concentration and target substance content of the mixed liquid; Based on the real-time parameter information, a parameter response time constant is calculated, and the pH value and temperature are divided into a fast response parameter group, and the concentration and target substance content are divided into a steady-state parameter group; The parameter change rate and parameter acceleration of the fast response parameter group are obtained, and a basic sampling frequency is calculated; the process disturbance intensity and disturbance change rate of the mixed liquid are detected, and the basic sampling frequency is adjusted to obtain a disturbance response frequency; the parameter weight corresponding to each parameter in the fast response parameter group is calculated, and the parameter weight and the corresponding disturbance response frequency are weighted to obtain a fast response sampling frequency; The parameter fluctuation variance of the steady-state parameter group is calculated, the ratio of the parameter fluctuation variance to the preset allowable fluctuation range is substituted into a logarithmic function to obtain a correction coefficient of the steady-state observation period; the adaptive observation period is obtained by adjusting the preset reference period based on the correction coefficient, and the maximum value of the adaptive observation period corresponding to each parameter is selected as the steady-state sampling period; When the weighted sum of the parameter change rate and the parameter acceleration exceeds a first preset threshold, an exponential smoothing function is used to realize the transition sampling of the fast response sampling frequency to the steady-state sampling period.

[0023] For example, the real-time parameter information includes four key indicators, i.e., pH value, temperature, concentration and target content of the mixed solution. The response time constant of each parameter is calculated, and the response characteristics are determined by monitoring and analyzing the change rate of each parameter. The pH value and temperature usually show a rapid change characteristic, and the response time constant is small, for example, the response time constant of the pH value is about 5 seconds, and the response time constant of the temperature is about 12 seconds. The concentration and target content show a slow response, and the response time constant is large, for example, the response time constant of the concentration is about 120 seconds, and the response time constant of the target content is about 180 seconds. Based on the difference in response characteristics, the pH value and temperature are divided into a fast response parameter group, and the concentration and target content are divided into a steady-state parameter group.

[0024] For the fast response parameter group, the parameter change rate and parameter acceleration are calculated by continuous sampling. For example, when the pH value changes from 6.5 to 7.0 in 10 seconds, the change rate is 0.05 / s; if the change rate is 0.04 / s in the first 5 seconds and 0.06 / s in the last 5 seconds, the acceleration is 0.004 / s 2 . Based on these dynamic change characteristics, the basic sampling frequency is calculated, and the basic sampling frequencies of the pH value and the temperature are set to 0.2 Hz and 0.1 Hz, respectively.

[0025] Considering the influence of process disturbance factors on the sampling frequency, the disturbance intensity and disturbance change rate of the mixed solution are detected by setting multiple sensors, for example, when the pH value fluctuation caused by feeding is detected to be 0.8 units and the change rate is 0.2 units / s, the basic sampling frequency is increased by 50% to obtain the disturbance response frequency. For each parameter in the fast response parameter group, the influence coefficient of each parameter on the extraction efficiency is determined by sensitivity analysis, i.e., the influence degree of a single parameter change on the extraction efficiency under the condition that other parameters remain unchanged. The obtained influence coefficient is normalized to obtain the weight value of each parameter, for example, the weight of the pH value is 0.7, and the weight of the temperature is 0.3. The weights and the corresponding disturbance response frequencies are weighted and calculated to obtain the fast response sampling frequency, for example, the sampling frequency of the pH value is 0.35 Hz, and the sampling frequency of the temperature is 0.15 Hz.

[0026] For the steady-state parameter group, the parameter fluctuation variance is calculated to represent its stability. For example, when the average value of the concentration in 20 consecutive samplings is 5.2 mol / L, and the standard deviation is 0.15 mol / L, the parameter fluctuation variance is 0.0225. The preset allowable fluctuation range is set, such as the allowable fluctuation range of the concentration is 0.05 mol / L, and the allowable fluctuation range of the target content is 0.02 g / L. The ratio of the parameter fluctuation variance to the preset allowable fluctuation range is substituted into the logarithmic function, for example, the ratio of the concentration is 9, and the correction coefficient calculated by the logarithmic function is 0.95. Based on the correction coefficient, the preset reference period (such as 300 seconds) is adjusted to obtain the adaptive observation period of the concentration as 285 seconds. In the same way, the adaptive observation period of the target content is calculated as 320 seconds. The maximum value of the adaptive observation period corresponding to each parameter is selected as the steady-state sampling period, which is 320 seconds.

[0027] When the weighted sum of the parameter change rate and the parameter acceleration exceeds the preset threshold, it indicates that it is from dynamic transition to steady state or vice versa. For example, the first preset threshold is set as 0.08, when the change rate of the pH value is 0.04 / second, the acceleration is 0.02 / second 2 , and the weighted sum is 0.04*0.7+0.02*0.3=0.034<0.08, the original sampling mode is maintained; when the weighted sum reaches 0.09>0.08, the exponential smoothing function is used to realize the transition of the fast response sampling frequency to the steady-state sampling period, and the sampling period P t at t second can be expressed as the weighted combination of the sampling period P t-1 at the previous moment and the target sampling period T, that is, P t =α*P t-1 +(1-α)*T, wherein α is a smoothing factor, and the value range is 0 to 1, for example, it is set as 0.8. In this way, the fast sampling frequency (such as 0.35 Hz, the period is about 2.86 seconds) can be smoothly transitioned to the steady-state sampling period (such as 320 seconds).

[0028] The double-period sampling mechanism is adopted in the application, the sampling density is increased when the parameter changes sharply, and the sampling frequency is reduced when the parameter is stable, so that the sampling resources are reasonably allocated. The exponential smoothing function is used to realize the natural transition of the frequency to the period, and the information loss caused by the discontinuous sampling is avoided. The sampling method can quickly respond to the disturbance, and can save the calculation resources under the steady-state working condition, so that the adaptability and stability of the monitoring are improved.

[0029] In an optional embodiment, based on the double-period sampling data, the three-dimensional numerical characteristics of the flow field distribution in the mixing tank are calculated, and the mixing tank is divided into a plurality of flow field partitions according to the three-dimensional numerical characteristics. According to the parameter space gradient characteristics of the spatial coordinate points of the mixing tank in three-dimensional directions calculated from the fast response period data, a local turbulent intensity is calculated based on the parameter space gradient characteristics, and a shear stress distribution of the spatial coordinate points is calculated based on the local turbulent intensity; According to the parameter steady-state mean value distribution and the parameter fluctuation intensity distribution of the spatial coordinate points in three-dimensional space calculated from the steady-state observation period data, a mixing degree index is calculated based on the parameter steady-state mean value distribution and the parameter fluctuation intensity distribution; A multi-dimensional flow field feature vector is constructed by fusing the shear stress distribution and the mixing degree index, and a multi-dimensional flow field feature vector similarity of adjacent spatial coordinate points is calculated; A ratio of a variance to a mean of the multi-dimensional flow field feature vector is calculated, a correction factor is obtained by substituting the ratio into a logarithmic function, and a preset initial partition threshold value is adjusted to obtain an adaptive partition threshold value according to the correction factor; Based on the multi-dimensional flow field feature vector similarity and the adaptive partition threshold value, the mixing tank is divided into a plurality of flow field partitions, a boundary gradient of the flow field partition is calculated, and a partition boundary migration rate is determined according to the boundary gradient; based on the partition boundary migration rate, a boundary position of the flow field partition is adjusted, and the adjusted partition boundary is smoothed to obtain a final mixing tank flow field partition result.

[0030] For example, when processing the fast response period data, the sensor grid collection points arranged in the mixing tank are analyzed. These sensors are arranged at a 10-centimeter interval to form a three-dimensional grid. For a spatial coordinate point, the parameter difference between the point and its adjacent point is divided by the distance between the two points. For example, for the coordinate point (10 cm, 15 cm, 20 cm), when calculating the gradient of the pH value in the x direction, the difference between the pH value 6.8 of the point and the pH value 7.3 of the adjacent point (20 cm, 15 cm, 20 cm) is divided by the distance 10 cm to obtain the gradient 0.05 units / cm. Similarly, the y and z direction gradients are calculated as 0.03 units / cm and 0.08 units / cm, respectively.

[0031] The divergence value of the gradient vector is calculated, that is, the square root of the sum of the squares of the gradients in the x, y, and z directions, multiplied by the adjustment coefficient of the ratio of the fluid density to the viscosity. For the aforementioned coordinate point, the turbulent intensity is calculated as: turbulent intensity = (0.05 2 + 0.03 2 + 0.08 2 ) 0.5 × 5 = 0.12, where 5 is the adjustment coefficient. The shear stress calculation method is that the turbulent intensity is multiplied by the fluid dynamic viscosity and then multiplied by the flow velocity gradient. For example, the fluid dynamic viscosity of the point is 0.8 mPa·s, and the flow velocity gradient is 2.6 / s, so the shear stress = 0.12 × 0.8 × 2.6 = 0.25 Pa.

[0032] When processing the steady-state observation period data, the steady-state mean distribution of each point parameter is calculated by taking the arithmetic mean of 20 consecutive steady-state sampling data. For example, the average of 20 pH value sampling results of the aforementioned coordinate point is 6.8. The parameter fluctuation intensity distribution is calculated by calculating the standard deviation of 20 sampling data. For example, the pH value standard deviation of this point is 0.15, indicating that the fluctuation intensity is 0.15. The mixing degree index is calculated by calculating the ratio of the fluctuation intensity to the preset allowable fluctuation range, and then subtracting the smaller value between the ratio and 1 from 1 to obtain the fluctuation mixing degree; the deviation percentage of the steady-state mean of each point parameter from the target mean is calculated, and then 1 is subtracted from the deviation percentage to obtain the mean mixing degree; the final mixing degree index is the weighted average of the fluctuation mixing degree and the mean mixing degree. For example, the preset allowable fluctuation range of the pH value is 0.6, the fluctuation intensity is 0.15, the fluctuation mixing degree = 1-min(0.15 / 0.6, 1) = 0.75; the target mean of the pH value is 7.0, the actual mean is 6.8, the deviation percentage is 2.9%, the mean mixing degree = 1-0.029 = 0.971; taking the weights as 0.6 and 0.4 respectively, the final mixing degree index = 0.75x0.6 + 0.971x0.4 = 0.838.

[0033] When constructing the multi-dimensional flow field feature vector, a direct combination method is used. For each space point, its shear stress value, mixing degree index and turbulent intensity are taken to form a vector. For example, the feature vector of point (10cm, 15cm, 20cm) is [0.25, 0.838, 0.12]. The method for calculating the similarity of the feature vectors of adjacent space points is: taking the square sum of the difference of the corresponding elements of the two vectors, calculating the square root, and then subtracting the ratio of the value and the sum of the vector lengths from 1. For example, the vectors of point (10cm, 15cm, 20cm) and point (11cm, 15cm, 20cm) are [0.25, 0.838, 0.12] and [0.26, 0.825, 0.13] respectively, and the similarity is calculated as 0.91.

[0034] When calculating the ratio of the variance to the mean of the multi-dimensional flow field feature vector, the mean and the variance of each dimension in the entire mixing tank are calculated first. For example, the mean of the shear stress dimension is 0.3 Pa, the variance is 0.09, and the ratio is 0.3. The correction factor is calculated using the formula: correction factor = 1 + 0.5xln(ratio + 0.1), which is 1 + 0.5xln(0.3 + 0.1) = 1.2. The preset initial partition threshold is 0.85, multiplied by the correction factor to get the adaptive partition threshold 0.85x1.2 = 1.02. Since the maximum similarity is 1, the threshold is adjusted to 0.95.

[0035] From the center point of the bottom of the mixing tank, check the similarity of the feature vectors of adjacent points. If the similarity is greater than the threshold value 0.95, the adjacent points are divided into the same partition; if it is less than the threshold value, a new partition is created. This process is recursively performed until all points are divided. For example, 500 sampling points are divided into 4 partitions, containing 210, 150, 90 and 50 points respectively.

[0036] The similarity of the feature vectors of adjacent points on both sides of the boundary is subtracted and divided by the distance between the two points. For example, the similarity between the two points on the boundary of partition 1 and partition 2 decreases from 0.93 to 0.78, and the distance is 1 cm, then the boundary gradient is 0.15 / cm. The calculation method of the migration rate of the partition boundary is: when the boundary gradient is greater than 0.1 / cm, the migration rate is 1 mm / hour; when the boundary gradient is between 0.05 / cm and 0.1 / cm, the migration rate is 0.5 mm / hour; when the boundary gradient is less than 0.05 / cm, the migration rate is 0.2 mm / hour. According to the migration rate, the boundary position is adjusted once an hour, and the boundary moves in the direction of the gradient.

[0037] For each point on the boundary, take the weighted average of its own position and the positions of the adjacent two boundary points, with weights of 0.5, 0.3 and 0.2 respectively. For example, the weighted average of the boundary points (25cm, 30cm, 35cm) and the adjacent two points (24cm, 30.5cm, 35.2cm), (26.2cm, 29.3cm, 34.8cm) is: 25×0.5+24×0.3+26.2×0.2=25.04cm, 30×0.5+30.5×0.3+29.3×0.2=29.96cm, 35×0.5+35.2×0.3+34.8×0.2=35.04cm, i.e. the new position is (25.04cm, 29.96cm, 35.04cm). Through three iterations of smoothing processing, the final mixing tank flow field partitioning result is obtained.

[0038] The present application realizes fine adaptive partitioning of the flow field of the mixing tank, overcoming the limitations of traditional equidistant partitioning and empirical partitioning. By fusing dynamic shear stress and static mixing degree indicators, the flow field characteristics are fully captured; the adaptive partitioning threshold is introduced, making the partitioning adapt to different working conditions; the boundary migration technology is used, making the partitioning dynamically adjust with the change of the flow field. This intelligent partitioning method based on fluid physical characteristics provides a foundation for subsequent implementation of precise partitioning control, significantly improving the efficiency and stability of the extraction and separation process.

[0039] In an optional implementation, a digital twin model is established based on the target substance concentration gradient information, and the step of outputting the fluid flow characteristics and mass transfer efficiency prediction value of each partition by the digital twin model includes: Collect concentration data of target substances, calculate concentration gradient of target substances in the mixing tank by spatial difference, extract gradient strength and gradient direction angle of the concentration gradient, and construct a concentration gradient feature vector; Input the concentration gradient feature vector into a residual neural network for feature extraction, perform feature mapping on the output result of the residual neural network according to fluid flow law and mass transfer law, and construct a digital twin model; Based on the digital twin model, calculate the local shear stress and vorticity distribution in the mixing tank as fluid flow characteristics, and calculate the mass transfer coefficient to obtain a mass transfer efficiency prediction value; Integrate learning of the fluid flow characteristics and the mass transfer efficiency prediction value by using a sliding time window, calculate the root mean square error of the prediction result, and dynamically update the digital twin model according to the root mean square error.

[0040] For example, in the construction process of the digital twin model, the target substance concentration data needs to be collected first. A plurality of conductivity sensors can be used to measure synchronously at different positions of the mixing tank, the collection frequency is set to 50 Hz, a total of 60 seconds are collected, and 3000 time series data are obtained for each measurement point. These sensors are uniformly distributed along the radial and axial directions, for example, 5 measurement points are arranged in the radial direction and 8 measurement points are arranged in the axial direction, forming a total of 40 spatially distributed measurement points. After collecting the concentration data, the concentration difference between adjacent two measurement points is calculated to obtain the concentration gradient components in x, y and z directions. For example, when the distance between the adjacent two points is 0.05 meters, if the concentrations of the two points are measured to be 0.5 mol / L and 0.6 mol / L respectively, the concentration gradient in this direction is 2 mol / (L·m).

[0041] The gradient strength and the gradient direction angle are extracted from the calculated concentration gradient. The gradient strength is represented by the square root of the sum of the squares of the gradient components in x, y and z directions. For the above example, if the gradients in the three directions are 2, 1.5 and 1 mol / (L·m) respectively, the gradient strength is 2.69 mol / (L·m). The gradient direction angle is obtained by calculating the angle between the gradient vector and the coordinate axis, for example, the angle with the x-axis is 36.9 degrees, the angle with the y-axis is 56.3 degrees, and the angle with the z-axis is 68.2 degrees. The angle calculation uses the vector dot product formula, for example, the cosine value of the x-axis angle is equal to the x-direction gradient component divided by the gradient strength, that is, cos(θx)=Gx / |G|=2 / 2.69=0.74, corresponding to an angle of 36.9 degrees.

[0042] The gradient intensity and the three direction angles are combined to form a four-dimensional feature vector as the concentration gradient feature vector. For 40 measurement points, feature vectors are constructed respectively to form a complete spatial concentration gradient feature description. These feature vectors change over time, reflecting the dynamic characteristics of the mixing process. To capture the changes in the time dimension, the feature vectors of 10 consecutive time points are taken for each measurement point to form a space-time feature matrix with a dimension of 10x4, and there are 40 such feature matrices.

[0043] The concentration gradient feature vector is input into the residual neural network for feature extraction. The network contains 5 residual blocks, each containing two convolution layers and a skip connection. The first layer of convolution uses 32 3x3 convolution kernels with a step size of 1, and the second layer uses 64 3x3 convolution kernels with a step size of 2. Each layer of convolution is followed by batch normalization and a ReLU activation function. The network input is a feature matrix with a shape of 40x4, and the output is a 256-dimensional feature vector after residual block processing. The network training uses the Adam optimizer with an initial learning rate of 0.001, a batch size of 64, and is trained for 300 rounds. The loss function uses the root mean square error, and an L2 regularization term is introduced to prevent overfitting, with a regularization coefficient of 0.0001.

[0044] The output of the residual neural network is mapped according to the fluid flow law and mass transfer law. In the mapping process, two fully connected layers are designed, the first layer contains 128 neurons, and the second layer contains 64 neurons, both using ReLU activation function. The last layer is the output layer, containing 40 neurons, corresponding to the predicted values of the 40 measurement points. Physical constraints such as mass conservation and momentum conservation are introduced during mapping to ensure that the predicted results conform to physical laws. For example, by adding a physical constraint term to the loss function, the predicted flow field satisfies the non-dispersion condition, i.e. the mass conservation of fluid inflow and outflow. The specific implementation method is to calculate the divergence of the predicted velocity field, and the square sum of the divergence is taken as part of the loss function, with a weight coefficient of 0.1. At the same time, the momentum conservation constraint is added, the residual of the predicted velocity field is calculated by substituting it into the Navier-Stokes equation, and the square sum of the residual is taken as another part of the loss function, with a weight coefficient of 0.2.

[0045] The digital twin model constructed by the above steps can calculate the local shear stress and vorticity distribution in the mixing tank. The shear stress calculation is based on the velocity gradient tensor, which is derived from the spatial derivative of the velocity field predicted by the network. For a three-dimensional velocity field V(u, v, w), the velocity gradient tensor contains 9 components: ; where V represents the velocity field, u represents the x-direction velocity component, v represents the y-direction velocity component, w represents the z-direction velocity component, represents the partial derivative of u with respect to x, i.e. the x-direction velocity gradient in the x-direction, and the rest are similar. The shear stress tensor τ is calculated by the formula where μ is the dynamic viscosity of the fluid, is the velocity gradient tensor, is the transpose of the velocity gradient tensor. For example, when the x-direction velocity difference between two adjacent grid points is 0.5 m / s with a spacing of 0.01 m, is 50 / s, if the fluid viscosity is 0.001 Pa·s, then τ xy = 0.05 Pa. Here, τ xy represents the shear stress component in the xy plane. The complete shear stress includes normal and tangential components, which can be obtained by eigenvalue decomposition to get the principal shear stress and its direction.

[0046] The vorticity distribution reflects the rotational characteristics of the fluid, which is calculated by the curl of the velocity field. The three-dimensional vorticity vector ω = (ωx, ωy, ωz) is calculated by the formula , which is specifically . Here, ω represents the vorticity vector, ωx represents the x-direction vorticity component, ωy represents the y-direction vorticity component, and ωz represents the z-direction vorticity component, represents the curl operator. The vorticity magnitude |ω| = (ωx 2 + ωy 2 + ωz 2 ) 0.5 represents the rotational intensity, where |ω| represents the vorticity magnitude. For the area near the rotating agitator, the measured vorticity can reach 30 / s, indicating that the fluid rotation in this area is intense. The vorticity distribution can be visualized by isosurfaces or streamlines, which helps to identify the position and intensity of the vortex core in the mixing tank.

[0047] The digital twin model can also calculate the mass transfer coefficient to obtain the mass transfer efficiency prediction value. The mass transfer coefficient k is calculated based on the double film theory, considering the effects of concentration gradient, fluid velocity, and turbulence intensity. The calculation formula is k = D(1 + αRe β · Sc γ ) / δ, where D is the molecular diffusion coefficient, δ is the boundary layer thickness, Re is the local Reynolds number, Sc is the Schmidt number, and α, β, γ are fitting coefficients, which are 0.0096, 0.8, and 0.33, respectively. The local Reynolds number is calculated by Re = ρvL / μ, where ρ is the fluid density, v is the local flow velocity, L is the characteristic length, and μ is the dynamic viscosity. The Schmidt number is calculated by Sc = μ / (ρD). In practical applications, the boundary layer thickness δ varies with the local flow velocity and turbulence intensity, which can be calculated by the empirical formula δ = 0.01 / (1 + 0.07Re 0.78) where 0.01 is a dimensionless reference length in meters, and 0.07 and 0.78 are experimental fitting coefficients. In the high concentration gradient region (such as 2.5 mol / (L·m)) and high flow rate region (such as 0.8 m / s), the mass transfer coefficient can reach 0.0015 m / s, while in the low concentration gradient region (such as 0.5 mol / (L·m)) and low flow rate region (such as 0.1 m / s), the mass transfer coefficient is about 0.0003 m / s.

[0048] The mass transfer efficiency is calculated by the formula η = k·A·ΔC / V, where k is the mass transfer coefficient, A is the mass transfer interface area, ΔC is the concentration difference, and V is the volume of the mixing tank. The mass transfer efficiency represents the ratio of the mass transfer amount per unit time to the total mass transfer amount, reflecting the mass transfer performance of the mixing tank.

[0049] The predicted values of fluid flow characteristics and mass transfer efficiency are integrated by setting the window size to 10 seconds and the sliding step to 2 seconds, and the predicted results in each window are weighted and averaged. The integrated learning integrates the prediction results of multiple models, and the weights are dynamically allocated based on the performance of each model on the validation set. The better the performance, the higher the weight. For example, in a certain time window, the weights of the three base models are 0.5, 0.3 and 0.2 respectively. The predicted values are compared with the measured values, such as the predicted shear stress value of 0.048 Pa and the measured value of 0.05 Pa, and the predicted mass transfer coefficient value of 0.00147 m / s and the measured value of 0.0015 m / s. The root mean square error is calculated to be 0.0025, which is lower than the preset threshold of 0.005, indicating that the model accuracy meets the requirements. If the error exceeds the threshold, the model is updated by adjusting the network parameters, such as increasing the learning rate from 0.001 to 0.002 and increasing the training rounds from 100 to 150.

[0050] The present application constructs a digital twin model based on concentration gradient features, combines residual neural networks with fluid flow rules and mass transfer rules, and can accurately predict local shear stress, vorticity distribution and mass transfer coefficient in the mixing tank. The sliding time window is used for integrated learning, and the model parameters are dynamically updated according to the root mean square error, so that the digital twin model has strong adaptability and prediction accuracy.

[0051] In an alternative embodiment, the concentration gradient feature vector is input into a residual neural network for feature extraction, and the output of the residual neural network is mapped according to the fluid flow rules and mass transfer rules to construct the digital twin model. The concentration gradient feature vector is input into a residual neural network, and the deep feature representation is output. mapping the deep feature representation to physical laws, respectively calculating a fluid momentum conservation deviation value and a fluid mass transfer diffusion deviation value, dividing the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value by their historical maximum values to obtain normalized deviation values, and taking the complementary values of the normalized deviation values as physical mapping weights; constructing a physical rule weight matrix according to a ratio relationship between the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value, the physical rule weight matrix being used to represent a coupling relationship between different physical variables; encoding the deep feature representation based on the physical mapping weights to obtain a physical knowledge attention weight, and superimposing a product of the physical knowledge attention weight, the physical rule weight matrix and the deep feature representation to the deep feature representation to obtain a physical enhanced feature; inputting the physical enhanced feature into a mapping network to obtain prediction results of a velocity field, a pressure field and a concentration field; In a sliding time window, a model training loss value is calculated according to fitting errors between the prediction results and real data, and the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value, the model training loss value including a root mean square error term; parameters of the residual neural network, the physical rule weight matrix and the mapping network are updated based on the model training loss value, and a digital twin model is dynamically optimized.

[0052] For example, the concentration gradient feature vector enters the residual neural network through the input layer, the network includes 5 residual blocks, each residual block includes two convolution layers and a jump connection. The first layer of convolution uses 32 3x3 convolution kernels with a step of 1, and the second layer uses 64 3x3 convolution kernels with a step of 2. Each layer of convolution is followed by batch normalization and a ReLU activation function. The network input is a feature matrix with a shape of 40x4, and after processing by the residual block, a 256-dimensional feature vector is output, which is the deep feature representation.

[0053] When the deep feature representation is mapped to physical laws, the momentum conservation deviation value is calculated by substituting the predicted velocity field into the momentum conservation equation. Specifically, for an incompressible fluid, the momentum conservation equation is where p is the fluid density, v is the velocity vector, p is the pressure, t is the time, μ is the dynamic viscosity, F is the body force, The Laplacian of velocity represents the viscous diffusion term. When calculating the bias value, the predicted velocity field and pressure field are substituted into the equation on both sides to calculate the absolute value of the difference. For example, at the coordinate point (10, 15, 20), the predicted velocity field is (0.5, 0.3, 0.2) m / s, the pressure gradient is (0.02, 0.01, 0.03) MPa / m, the fluid density is 1000 kg / m3, and the dynamic viscosity is 0.001 Pa·s. Substituting into the left side of the equation gives a value of 12.5 N / m2, and substituting into the right side gives a value of 12.2 N / m2, with an absolute difference of 0.3 N / m2, and a normalized bias value of 0.03. The mass transfer diffusion bias value is calculated by substituting the predicted concentration field into the mass transfer diffusion equation. The mass transfer diffusion equation is where c is the concentration, t is the time, v is the fluid velocity vector, and D is the diffusion coefficient, The Laplacian of concentration describes the change in concentration caused by molecular diffusion. At the same coordinate point, the predicted concentration is 0.6 mol / L, the concentration gradient is (0.05, 0.03, 0.04) mol / (L·m), the velocity field is (0.5, 0.3, 0.2) m / s, and the diffusion coefficient is 10 -9 square meters per second. Substituting into the left side of the equation gives a value of 0.053 mol / (L·s), and substituting into the right side gives a value of 0.051 mol / (L·s), with an absolute difference of 0.002 mol / (L·s), and a normalized bias value of 0.02.

[0054] The fluid momentum conservation bias value 0.03 is divided by the historical maximum bias value 0.1 to obtain the normalized bias value 0.3, and the fluid mass transfer diffusion bias value 0.02 is divided by the historical maximum bias value 0.08 to obtain the normalized bias value 0.25. The complementary values are calculated to obtain the momentum conservation physical mapping weight of 0.7 and the mass transfer diffusion physical mapping weight of 0.75, respectively.

[0055] According to the ratio relationship 1.5 between the fluid momentum conservation bias value 0.03 and the fluid mass transfer diffusion bias value 0.02, a physical rule weight matrix is constructed. The matrix has a dimension of 256x256, the diagonal elements are set to 1, and the non-diagonal elements are determined according to the coupling strength between physical variables. For example, the value of the element in the first row and the second column is 0.8, indicating that the coupling strength between the first feature and the second feature is 0.8. This physical rule weight matrix represents the coupling relationship between different physical variables, and is used to guide the model to learn the feature representation that conforms to the physical law.

[0056] The deep feature representation is encoded based on the physical mapping weights, and the self-attention mechanism is used to multiply the deep feature representation with the momentum conservation physical mapping weight 0.7 and the mass transfer diffusion physical mapping weight 0.75 to obtain the encoded result. For example, if the first dimension value of the deep feature representation is 0.6, multiplying it by 0.7 gives 0.42, indicating that the encoding value of this dimension in momentum conservation is 0.42.

[0057] The encoded result is multiplied by the physical rule weight matrix, and then fused with the original deep feature representation. For example, the value of a certain dimension of the deep feature representation is 0.6, which is enhanced to 0.78 after encoding and physical rule matrix combination, indicating that the feature is strengthened in the physical sense.

[0058] The physical enhanced feature is input into the mapping network to obtain the prediction results of the velocity field, pressure field and concentration field. The mapping network contains three parallel fully connected layer branches, which are used to predict the velocity field, pressure field and concentration field respectively. The velocity field branch contains two hidden layers with 128 and 64 neurons respectively; the pressure field branch contains two hidden layers with 96 and 48 neurons respectively; the concentration field branch contains two hidden layers with 64 and 32 neurons respectively. All hidden layers use ReLU activation function, and the output layer does not use activation function. For example, after the physical enhanced feature is input into the mapping network, the predicted velocity at the coordinate point (10, 15, 20) is (0.48, 0.32, 0.21) m / s, the predicted pressure is 0.15 MPa, and the predicted concentration is 0.58 mol / L. These prediction results are then used to calculate the fluid flow characteristics and mass transfer efficiency.

[0059] When calculating the model training loss value in the sliding time window, the window size is set to 10 seconds and the sliding step is set to 2 seconds. The fitting error is calculated by the root mean square error, that is, the square root of the average of the sum of the squares of the difference between the predicted value and the measured value. Specifically for each physical field, the root mean square error of the predicted value and the measured value of the velocity field is 0.04 m / s, and the calculation method is to take the square of the difference between the predicted value and the measured value of the three velocity components at all time points and all space points in the window, sum them up, divide by the total number of samples, and then take the square root. The root mean square error of the pressure field is 0.01 MPa, and the root mean square error of the concentration field is 0.03 mol / L, and the calculation method is similar. The total loss value is the weighted sum of the root mean square error term, the momentum conservation deviation term and the mass transfer diffusion deviation term, and the weights are 0.5, 0.3 and 0.2 respectively. For the above example, the total loss value is 0.04 multiplied by 0.5 plus 0.03 multiplied by 0.3 plus 0.02 multiplied by 0.2, which is equal to 0.033. The momentum conservation deviation term and the mass transfer diffusion deviation term are the average values of the deviation values calculated at all time and space points in the window.

[0060] Based on the model training loss value, the parameter is updated using the back propagation algorithm and Adam optimizer, the learning rate is set to 0.001, and the batch size is 64. For the identified high error area, increase the sampling density and adjust the local learning rate. For example, when the error in the area near the stirrer is large, the learning rate of this area is increased to 0.002, and the number of sampling points is increased from 5 to 8. At the same time, the physical rule weight matrix is dynamically adjusted, and the weight value of the poor coupling relationship is reduced, such as reducing a certain coupling weight from 0.8 to 0.6. Through multiple rounds of iterative training, the model loss value is reduced from the initial 0.15 to less than 0.03, reaching the preset precision requirement. The updated model is verified on new data, such as predicting the speed of a test point as (0.51, 0.29, 0.19) m / s, and the actual value is (0.50, 0.30, 0.20) m / s, with an error of less than 5%, meeting the engineering application requirements. For the concentration field, the predicted value is 0.59 mol / L, and the actual value is 0.60 mol / L, with an error of 1.7%, also meeting the accuracy requirements.

[0061] The present application realizes the physical constraint enhancement of the digital twin model by explicitly integrating the physical law knowledge into the deep learning model. Compared with the pure data-driven method, the method can still maintain good prediction performance in the sparse data area, and the model has stronger generalization ability. At the same time, the introduction of the physical rule weight matrix and the physical knowledge attention mechanism enables the model to better capture the complex coupling relationship between physical variables, thereby improving the applicability and accuracy of the model in complex fluids.

[0062] In an optional implementation, according to the output of the digital twin model, a partition adaptive control strategy is executed to dynamically adjust the stirring speed and stirring direction of each partition, automatically control the extractant replenishment amount and replenishment time of each partition, and adjust the baffle opening degree according to the concentration gradient of adjacent partitions. According to the velocity field, pressure field and concentration field output by the digital twin model, the local Reynolds number and local Schmidt number of each flow field partition are calculated, the optimal stirring speed of each flow field partition is calculated based on the local Reynolds number and local Schmidt number, and the stirring direction angle is determined according to the concentration gradient of each flow field partition. The local mass transfer efficiency is calculated based on the local mass transfer coefficient and specific surface area of each flow field partition, the extractant replenishment amount is calculated according to the local mass transfer efficiency, and the extractant replenishment amount is related to the flow field partition volume, molecular diffusion coefficient and mixing time; the extractant replenishment time interval is determined based on the difference between the target concentration and the actual concentration and the concentration change rate; The concentration difference between adjacent flow field partitions is calculated, the diffusion flux is calculated according to the concentration difference and the molecular diffusion coefficient, the ratio of the diffusion flux to the reference diffusion flux is substituted into the exponential function to obtain the baffle opening degree adjustment coefficient, and the baffle opening degree between adjacent flow field partitions is determined based on the baffle opening degree adjustment coefficient. The mass transfer efficiency of each flow field partition is monitored, and when the mass transfer efficiency is lower than the target value, the control parameters of the corresponding flow field partition are adjusted to realize dynamic optimization of the stirring parameters, extractant supplement and baffle opening of each flow field partition.

[0063] In combination Figure 2 The extraction process partition adaptive control strategy execution flowchart based on the digital twin model is described. For example, after the digital twin model outputs the velocity field, pressure field and concentration field of each partition, the local Reynolds number is calculated. The characteristic length is the diameter of the stirring paddle, the characteristic velocity is the local average flow velocity, and the fluid density and dynamic viscosity are queried from the database according to the temperature and concentration. For example, in the first partition, the diameter of the stirring paddle is 0.2 meters, the local average flow velocity is 0.5 meters per second, the fluid density is 1050 kilograms per cubic meter, and the dynamic viscosity is 0.002 Pa·s. The calculated local Reynolds number is 52500. When calculating the local Schmidt number, the values of dynamic viscosity, fluid density and molecular diffusion coefficient are taken, such as dynamic viscosity 0.002 Pa·s, fluid density 1050 kg / m3, and molecular diffusion coefficient 10 -9 square meters per second, the calculated local Schmidt number is 1905.

[0064] The empirical relationship is used. When the local Reynolds number is in the range of 10000 to 20000, the optimal stirring speed is proportional to the 0.7 power of the Reynolds number; when the local Reynolds number is in the range of 20000 to 100000, the optimal stirring speed is proportional to the 0.5 power of the Reynolds number; and when the local Reynolds number is greater than 100000, the optimal stirring speed is proportional to the 0.3 power of the Reynolds number. For the above-mentioned first partition, the local Reynolds number is 52500, which falls within the second range, and the calculated optimal stirring speed is 2.3 revolutions per second. At the same time, the stirring speed is adjusted according to the Schmidt number. When the Schmidt number is greater than 1000, the stirring speed needs to be adjusted by 10% to 20% to enhance the turbulent mixing effect. For the above-mentioned first partition, the Schmidt number is 1905, and the stirring speed is adjusted by 15%, and the final adjustment is 2.65 revolutions per second.

[0065] The concentration gradient direction is the direction from high concentration to low concentration, and the stirring direction angle is set to be perpendicular to the concentration gradient direction to maximize the stirring effect. For example, the concentration gradient direction of the first partition is (0.7, 0.5, 0.5), and after normalization, it is (0.7071, 0.5051, 0.5051), and the angle with the x-axis is calculated to be 45 degrees, the angle with the y-axis is 60 degrees, and the angle with the z-axis is 60 degrees. Take the angle with the horizontal plane as 45 degrees and the angle with the vertical direction as 60 degrees as the stirring direction angle.

[0066] The local mass transfer coefficient is related to the molecular diffusion coefficient, Reynolds number and Schmidt number. For example, in the first partition, the molecular diffusion coefficient is 10 -9Given a flow rate of m² / s, a Reynolds number of 52500, and a Schmidt number of 1905, substituting these values ​​into the mass transfer coefficient calculation formula: k = D(1 + α·Re β ·Sc γ ) / δ. Where D is the molecular diffusion coefficient, representing the diffusion ability of a substance in a static fluid; α, β, and γ are empirical fitting coefficients, with values ​​of 0.0096, 0.8, and 0.33 respectively, used to correct for mass transfer enhancement effects under turbulent conditions; Re is the Reynolds number, characterizing the degree of fluid turbulence; Sc is the Schmidt number, characterizing the relative intensity of momentum transfer and mass transfer; δ is the boundary layer thickness, calculated using the formula δ=0.01 / (1+0.07·Re). 0.78 The estimated value is approximately 2.91 × 10⁻⁶. -5 The uncorrected mass transfer coefficient is calculated to be 4.46 × 10⁻⁶ meters, where 0.01 is the reference length (meters). -5 At a speed of m / s, after a correction factor of 0.72 to account for the influence of the stirrer geometry, the final local mass transfer coefficient is 3.2 × 10⁻⁶ m / s. -5 The specific surface area is related to the stirring intensity and the gas-liquid interface morphology. When the stirring speed is 2.65 rpm, the specific surface area is 200 m² / m³, representing the interface area per unit volume. The local mass transfer efficiency is the product of the local mass transfer coefficient and the specific surface area. The calculated local mass transfer efficiency for the first zone is 3.2 × 10⁻⁶ m² / m³. -5 Meters per second × 200 square meters per cubic meter = 0.0064 per second, which represents the ratio of transmitted mass to total transmitted mass per unit time.

[0067] The amount of extractant added is related to the volume of the flow field zone, the molecular diffusion coefficient, and the mixing time. Taking the first zone as an example, the zone volume is 0.5 cubic meters, and the molecular diffusion coefficient is 10. -9 With a mass transfer rate of m² / s, a mixing time of 120 seconds, and a local mass transfer efficiency of 0.0064 sec, the calculated extractant replenishment amount is 0.023 kg. For a target concentration of 0.8 mol / L, an actual concentration of 0.6 mol / L, a concentration difference of 0.2 mol / L, and a concentration change rate of 0.001 mol / (L·s), the extractant replenishment interval is 200 seconds.

[0068] Calculate the concentration difference between adjacent flow field zones. For example, if the concentrations in the first and second zones are 0.6 mol / L and 0.7 mol / L respectively, the concentration difference is 0.1 mol / L. Calculate the diffusion flux based on the concentration difference and the molecular diffusion coefficient (10). -9 With a density of square meters per second and a distance of 0.05 meters between the two zones, the calculated diffusion flux is 2 × 10⁻⁶. -9 Moles per square meter per second. The reference diffusion flux is taken as 10. -8Mol / (m2·s), the ratio of the diffusion flux to the reference diffusion flux is 0.2. The ratio is substituted into the exponential function to obtain the baffle opening adjustment coefficient of 0.8. The initial opening of the baffle is 50%, which is adjusted to 40% according to the adjustment coefficient. When the concentration difference increases, the baffle opening decreases to prevent rapid mixing; when the concentration difference decreases, the baffle opening increases to promote uniform mixing.

[0069] Taking the first partition as an example, the target mass transfer efficiency is 0.007 per second, and the actual monitored mass transfer efficiency is 0.0064 per second, which is lower than the target value. The stirring speed is automatically increased by 10% to 2.92 revolutions per second, the extractant supplement is increased by 5% to 0.024 kilograms, and the baffle opening is adjusted from 40% to 45%. After adjustment, the mass transfer efficiency rises to 0.0072 per second, reaching the target value. Parameter adjustment is performed every 30 seconds to continuously optimize the control parameters of each partition.

[0070] During the extraction process, local dead zones or short circuit flow phenomena may occur. When a local dead zone is detected in the third partition, the stirring speed in this region is automatically increased by 20%, and the stirring direction angle is changed by 30 degrees, effectively eliminating the dead zone. When a short circuit flow is detected between the fourth and fifth partitions, the baffle opening between the two regions is reduced to 20%, and the stirring direction angle is adjusted to make the fluid flow more uniform. Through this adaptive control strategy, various abnormal working conditions can be handled, maintaining the stability and efficiency of the extraction process.

[0071] For monitoring and adjusting the extraction efficiency, the actual extraction efficiency of each partition is calculated at a set time (e.g. 3 minutes). When the actual extraction efficiency is less than 90% of the pre-set target value (e.g. the target efficiency is 0.007 per second, and the actual efficiency is less than 0.0063 per second), or the efficiency fluctuation between two adjacent measurements exceeds 15%, it is determined as an efficiency anomaly. At this time, the concentration gradient information is re-collected, and the adaptive control strategy described above is used to optimize and adjust the stirring parameters, extractant supplement scheme and baffle opening until the actual extraction efficiency rises to more than 95% of the target value.

[0072] The present application realizes precise regulation and control of each flow field partition in the mixing tank. Compared with traditional control methods, this technology can dynamically optimize stirring parameters, extractant supplement and baffle opening according to real-time flow field state, effectively solving problems such as uneven mixing, local dead zones and short circuit flow. The adaptive control strategy significantly improves the extraction efficiency, reduces energy consumption and extractant consumption, and enhances the stability and controllability of the process, providing an efficient and intelligent control scheme for complex fluid mixing processes.

[0073] In an alternative embodiment, the local mass transfer efficiency is calculated based on the local mass transfer coefficient and specific surface area of each flow field partition, and the step of calculating the extractant supplement amount based on the local mass transfer efficiency comprises: The current concentration and the equilibrium concentration of each flow field partition are obtained, the relative difference between the current concentration and the equilibrium concentration is calculated, the characteristic size of the fluid micro-cluster is calculated based on the velocity field output by the digital twin model, the ratio of the characteristic size to the molecular diffusion coefficient is determined as the local mass transfer unit number, and the local mass transfer coefficient is calculated according to the local mass transfer unit number; the shear stress distribution of each flow field partition is collected, the droplet deformation degree is calculated based on the shear stress distribution, and the interface update coefficient is calculated according to the droplet deformation degree; the ratio of the interface update coefficient to the geometric size of the flow field partition is determined as the dynamic specific surface area; The local mass transfer efficiency is calculated according to the relative difference, the local mass transfer coefficient and the dynamic specific surface area; the mass transfer resistance coefficient is calculated based on the local mass transfer efficiency, and the mass transfer resistance coefficient exponentially decays with the increase of the local mass transfer efficiency; The product of the mass transfer resistance coefficient and the preset reference flow rate is determined as the extractant supplement amount, wherein the preset reference flow rate is equal to the volume of the flow field partition divided by the mixing time required to reach the preset mixing uniformity, and is corrected according to the molecular diffusion coefficient.

[0074] For example, taking the first partition as an example, the current concentration is 0.45 mol / L collected by the distributed concentration sensor, the equilibrium concentration is 0.72 mol / L under the corresponding working condition according to the thermodynamic database query, and the relative difference is 0.375, i.e. (0.72-0.45) / 0.72. The relative difference represents the degree of deviation from the equilibrium state, and the larger the value, the stronger the mass transfer driving force. For the second partition, the current concentration is 0.53 mol / L, and the equilibrium concentration is also 0.72 mol / L, and the relative difference is 0.264. The concentration of each partition is monitored in real time by a high-precision conductivity sensor with a sampling frequency of 10 Hz to ensure the accuracy and timeliness of the concentration data.

[0075] The characteristic size of the fluid micro-cluster is related to the local turbulent intensity, which can be calculated by the eigenvalue of the velocity gradient tensor. The average value of the velocity field gradient of the first partition is 12.5 / s, and the corresponding turbulent energy dissipation rate is 0.156 W / kg. The characteristic size of the fluid micro-cluster is calculated based on the turbulent energy dissipation rate, using the Kolmogorov microscale theory, and the characteristic size is equal to the fourth root of the ratio of the dynamic viscosity to the fluid density multiplied by the negative fourth power of the turbulent energy dissipation rate. The dynamic viscosity of the first partition is 0.002 Pa·s, and the density is 1050 kg / m3. The characteristic size of the fluid micro-cluster is calculated to be 0.00024 m. The ratio of the characteristic size to the molecular diffusion coefficient is determined as the local mass transfer unit number, and the molecular diffusion coefficient of the first partition is 10 -9square meters per second, and the local mass transfer unit number was calculated to be 240,000. According to the local mass transfer unit number, the local mass transfer coefficient was calculated, and using the penetration theory model, when the mass transfer unit number is greater than 10,000, the local mass transfer coefficient is proportional to the -0.5 power of the mass transfer unit number, and the local mass transfer coefficient of the first partition was calculated to be 6.5 x 10 -6 square meters per second. For the second partition, the average value of the velocity field gradient was 9.8 per second, the fluid micro-cluster characteristic size was calculated to be 0.00028 meters, the local mass transfer unit number was calculated to be 280,000, and the local mass transfer coefficient was calculated to be 6.0 x 10 -6 square meters per second.

[0076] The shear stress distribution of each flow field partition was collected, the average shear stress of the first partition was 0.025 Pa, the maximum shear stress was 0.045 Pa, and the minimum shear stress was 0.012 Pa. The droplet deformation degree was defined as the ratio of the shear stress to the interfacial tension multiplied by the droplet radius. For the first partition, the interfacial tension between the liquid phase and the dispersed phase was 0.032 Newton per meter, the initial radius of the dispersed phase droplet was 0.0001 meters, and the average droplet deformation degree was calculated to be 0.078. According to the droplet deformation degree, the interfacial renewal coefficient was calculated, and when the droplet deformation degree was less than 0.1, the interfacial renewal coefficient was approximately equal to 1 plus twice the deformation degree, and the interfacial renewal coefficient of the first partition was calculated to be 1.156. The ratio of the interfacial renewal coefficient to the geometric size of the flow field partition was determined as the dynamic specific surface area, the characteristic geometric size (mixing tank diameter) of the first partition was 0.4 meters, and the dynamic specific surface area was calculated to be 2.89 per meter. For the second partition, the average shear stress was 0.020 Pa, the average droplet deformation degree was calculated to be 0.063, the interfacial renewal coefficient was 1.126, and the dynamic specific surface area was 2.82 per meter.

[0077] The local mass transfer efficiency was defined as the ratio of the mass transfer amount per unit time to the total mass transfer amount, and the calculation formula was the local mass transfer coefficient multiplied by the dynamic specific surface area and then multiplied by the relative difference. The local mass transfer coefficient of the first partition was 6.5 x 10 -6 square meters per second, the dynamic specific surface area was 2.89 per meter, the relative difference was 0.375, and the local mass transfer efficiency was calculated to be 7.03 x 10 -6 per second. The local mass transfer coefficient of the second partition was 6.0 x 10 -6 square meters per second, the dynamic specific surface area was 2.82 per meter, the relative difference was 0.264, and the local mass transfer efficiency was calculated to be 4.47 x 10 -6 per second. The higher the local mass transfer efficiency, the faster the mass transfer process in the partition, and the higher the extraction efficiency.

[0078] The mass transfer resistance coefficient represents the resistance of the mass transfer process, which exponentially decays with the increase of the local mass transfer efficiency. The calculation formula of the mass transfer resistance coefficient is in the form of the negative exponent of the exponential function, with the base number being the base e of the natural logarithm and the exponent being the ratio of the negative mass transfer efficiency to the reference mass transfer efficiency. The reference mass transfer efficiency is set to 10 -5 The first partition mass transfer efficiency is 7.03 x 10 -6 The mass transfer resistance coefficient is calculated to be 1.93. The second partition mass transfer efficiency is 4.47 x 10 -6 The mass transfer resistance coefficient is calculated to be 2.65. The larger the mass transfer resistance coefficient, the more serious the resistance of the mass transfer process, and more extractant needs to be supplemented to accelerate the mass transfer process.

[0079] The product of the mass transfer resistance coefficient and the preset reference flow rate is determined as the extractant supplement amount. The preset reference flow rate is equal to the volume of the flow field partition divided by the mixing time required to reach the preset mixing uniformity, and is corrected according to the molecular diffusion coefficient. The preset mixing uniformity is defined as the ratio of the highest concentration to the lowest concentration in the partition being less than 1.05. The volume of the first partition is 0.5 cubic meters, the mixing time required to reach the preset mixing uniformity is 180 seconds, and the molecular diffusion coefficient is 10 -9 The reference molecular diffusion coefficient is 10 -8 The correction coefficient is 3.16 (the square root of the ratio of the reference molecular diffusion coefficient to the actual molecular diffusion coefficient). The preset reference flow rate of the first partition is calculated to be 0.0088 cubic meters per second, the mass transfer resistance coefficient is 1.93, and the final extractant supplement amount is 0.017 cubic meters per second, i.e. 17 liters per second. The volume of the second partition is 0.6 cubic meters, the mixing time is 210 seconds, and the preset reference flow rate is calculated to be 0.009 cubic meters per second, the mass transfer resistance coefficient is 2.65, and the final extractant supplement amount is 0.024 cubic meters per second, i.e. 24 liters per second.

[0080] In practical applications, the extractant supplement amount also needs to consider equipment limitations and economic factors. The maximum supplement amount is set to 0.03 cubic meters per second and the minimum supplement amount is set to 0.005 cubic meters per second. When the calculated value exceeds the range, the boundary value is taken. At the same time, the extractant supplement amount is dynamically adjusted according to the partition concentration, and when the concentration approaches the equilibrium value, the supplement amount gradually decreases. For example, when the first partition concentration reaches 0.65 moles per liter, the relative difference is reduced to 0.097, and the extractant supplement amount is recalculated to be 0.007 cubic meters per second. The mass transfer efficiency and extractant supplement amount of each partition are re-evaluated every 30 seconds to ensure the efficiency and economy of the extraction process.

[0081] The application realizes intelligent control of the supplement amount of the extractant by accurately calculating the local mass transfer characteristics of each flow field partition, can dynamically adjust the extractant usage according to the real-time mass transfer efficiency, and avoids the problems of waste of extractant and low mass transfer efficiency. The intelligent supplement strategy not only improves the material utilization rate and mass transfer efficiency, but also reduces the energy consumption and operation cost, and enhances the stability and controllability of the process.

[0082] In a second aspect, the application provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to perform the method described above.

[0083] In a third aspect, the application provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, implement the method described above.

[0084] The application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for executing various aspects of the application.

[0085] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A method for intelligent control of large scale fluid extraction separation based on a mixing tank structure, characterized in that, The method comprises the steps of: Collecting real-time parameter information of the mixed liquid of the extractant and the material in the mixing tank, and performing double-period sampling on the real-time parameter information according to a fast response period and a steady-state observation period; Based on the double-period sampling data, the three-dimensional numerical characteristics of the flow field distribution in the mixing tank are calculated, the mixing tank is divided into multiple flow field partitions according to the three-dimensional numerical characteristics, and independent stirring devices are arranged in each flow field partition; Real-time acquisition of the target substance concentration gradient information of each flow field partition, establishment of a digital twin model based on the target substance concentration gradient information, and output of the fluid flow characteristics and mass transfer efficiency prediction value of each partition by the digital twin model; According to the output of the digital twin model, a partition adaptive control strategy is executed to dynamically adjust the stirring speed and stirring direction of each partition, automatically control the extractant supplement amount and supplement time of each partition, and adjust the baffle opening degree according to the concentration gradient of the adjacent partition; Real-time monitoring of the extraction efficiency of each partition, updating the partition adaptive control strategy based on the latest collected target substance concentration gradient information when an abnormal extraction efficiency is detected, and reaching the preset extraction efficiency.

2. The method of claim 1, wherein, The step of collecting real-time parameter information of the mixed liquid of the extractant and the material in the mixing tank and performing double-period sampling on the real-time parameter information according to a fast response period and a steady-state observation period comprises: Based on the real-time parameter information, the parameter response time constant is calculated, and the parameters are divided into a fast response parameter group and a steady-state parameter group according to the parameter response time constant; The parameter change rate and parameter acceleration of the fast response parameter group are obtained, and the base sampling frequency is calculated; the process disturbance intensity and disturbance change rate of the mixed liquid are detected, and the base sampling frequency is adjusted to obtain the disturbance response frequency; the parameter weight corresponding to each parameter in the fast response parameter group is calculated, and the fast response sampling frequency is obtained by weighting the parameter weight and the corresponding disturbance response frequency; The parameter fluctuation of the steady-state parameter group is calculated, the ratio of the parameter fluctuation to the preset allowable fluctuation range is substituted into the logarithmic function to obtain the modification coefficient of the steady-state observation period, and the adaptive observation period is obtained by adjusting the preset reference period based on the modification coefficient; the maximum value of the adaptive observation period corresponding to each parameter is selected as the steady-state sampling period; When the weighted sum of the parameter change rate and the parameter acceleration exceeds a first preset threshold, an exponential smoothing function is used to realize the transition sampling of the fast response sampling frequency to the steady-state sampling period.

3. The method of claim 1, wherein, The step of calculating the three-dimensional numerical characteristics of the flow field distribution in the mixing tank based on the double-period sampling data and dividing the mixing tank into multiple flow field partitions according to the three-dimensional numerical characteristics comprises: According to the fast response period data, the parameter space gradient characteristics of the spatial coordinate points in the mixing tank along the three-dimensional direction are calculated, the local turbulent intensity is calculated based on the parameter space gradient characteristics, and the shear stress distribution of the spatial coordinate points is calculated based on the local turbulent intensity; According to the steady-state observation period data, the parameter steady-state mean distribution and the parameter fluctuation intensity distribution of the spatial coordinate points in the three-dimensional space are calculated, and the mixing degree index is calculated based on the parameter steady-state mean distribution and the parameter fluctuation intensity distribution; The shear stress distribution and the mixing degree index are fused to construct a multi-dimensional flow field feature vector; an adaptive partition threshold is calculated based on the multi-dimensional flow field feature vector, and the mixing tank is divided into a plurality of flow field partitions; a boundary gradient of the flow field partition is calculated, and a partition boundary migration rate is determined according to the boundary gradient; the boundary position of the flow field partition is adjusted based on the partition boundary migration rate, and a final mixing tank flow field partition result is obtained.

4. The method of claim 1, wherein, The step of establishing a digital twin model based on the target concentration gradient information includes: Collecting target concentration data, calculating the concentration gradient of the target in the mixing tank, and constructing a concentration gradient feature vector based on the concentration gradient; The concentration gradient feature vector is input into a residual neural network for feature extraction, and the output results of the residual neural network are mapped according to the fluid flow law and the mass transfer law to construct a digital twin model; Based on the digital twin model, the local shear stress and vorticity distribution in the mixing tank are calculated as the fluid flow characteristics, and the mass transfer coefficient is calculated to obtain the mass transfer efficiency prediction value; The root mean square error of the prediction result is dynamically updated based on the root mean square error of the prediction result.

5. The method of claim 4, wherein, The step of inputting the concentration gradient feature vector into a residual neural network for feature extraction, and mapping the output results of the residual neural network according to the fluid flow law and the mass transfer law to construct a digital twin model includes: The concentration gradient feature vector is input into a residual neural network, and a deep feature representation is output; The deep feature representation is mapped according to the physical law, the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value are calculated respectively, and the physical mapping weight is constructed based on the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value; A physical rule weight matrix is constructed to represent the coupling relationship between different physical variables; the deep feature representation is encoded based on the physical mapping weight, and the physical enhanced feature is obtained by combining the encoding result with the physical rule weight matrix; The physical enhanced feature is input into a mapping network to obtain the prediction results of the velocity field, the pressure field and the concentration field; In the sliding time window, the model training loss value is calculated according to the fitting error between the prediction result and the real data, and the fluid momentum conservation deviation value and the fluid mass transfer diffusion deviation value; the parameters of the residual neural network, the physical rule weight matrix and the mapping network are updated based on the model training loss value, and the digital twin model is dynamically optimized.

6. The method of claim 1, wherein, According to the output of the digital twin model, a partition adaptive control strategy is executed, the stirring speed and stirring direction of each partition are dynamically adjusted, the extraction agent replenishment amount and replenishment time of each partition are automatically controlled, and the step of adjusting the baffle opening degree according to the concentration gradient of the adjacent partition includes: According to the velocity field, pressure field and concentration field output by the digital twin model, the local Reynolds number and the local Schmidt number of each flow field partition are calculated, the optimal stirring speed of each flow field partition is calculated based on the local Reynolds number and the local Schmidt number, and the stirring direction angle is determined according to the concentration gradient of each flow field partition; The local mass transfer efficiency is calculated based on the local mass transfer coefficient and the specific surface area of each flow field partition, the extraction agent supplement amount is calculated according to the local mass transfer efficiency, and the extraction agent supplement amount is related to the flow field partition volume, the molecular diffusion coefficient and the mixing time; the extraction agent supplement time interval is determined based on the difference between the target concentration and the actual concentration and the concentration change rate; The concentration difference between adjacent flow field partitions is calculated, the baffle opening adjustment coefficient is calculated according to the concentration difference, and the baffle opening between adjacent flow field partitions is determined based on the baffle opening adjustment coefficient; The mass transfer efficiency of each flow field partition is monitored, and when the mass transfer efficiency is lower than the target value, the control parameters of the corresponding flow field partition are adjusted to realize the dynamic optimization of the stirring parameters, the extraction agent supplement and the baffle opening of each flow field partition.

7. The method of claim 6, wherein, The step of calculating the local mass transfer efficiency based on the local mass transfer coefficient and the specific surface area of each flow field partition includes: The relative difference between the current concentration and the equilibrium concentration of each flow field partition is calculated; the characteristic size of the fluid micro-cluster is calculated based on the velocity field output by the digital twin model, the ratio of the characteristic size to the molecular diffusion coefficient is determined as the local mass transfer unit number, and the local mass transfer coefficient is calculated according to the local mass transfer unit number; the droplet deformation degree is calculated based on the shear stress distribution of each flow field partition, and the interface update coefficient is calculated according to the droplet deformation degree; the ratio of the interface update coefficient to the geometric size of the flow field partition is determined as the dynamic specific surface area; The local mass transfer efficiency is calculated according to the relative difference, the local mass transfer coefficient and the dynamic specific surface area; the mass transfer resistance coefficient is calculated based on the local mass transfer efficiency; The product of the mass transfer resistance coefficient and a preset reference flow rate is determined as the extraction agent supplement amount, wherein the preset reference flow rate is equal to the flow field partition volume divided by the mixing time required to reach the preset mixing uniformity, and is corrected according to the molecular diffusion coefficient.

8. An electronic device, comprising: It includes: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1 to 7.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 7. The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 7.

Images