Centrifugal belt type blood cell separation and detection system and detection method

Abstract

The invention discloses a centrifugal belt type blood cell separation detection system and detection method, and belongs to the technical field of blood cell detection.The method comprises the steps that an initial optimal control track is planned and generated according to a state baseline and physiological indexes of a blood donor; tracking an actual separation trajectory, comparing the actual separation trajectory with the initial trajectory, and quantizing to generate a real-time dynamic deviation vector; when the deviation exceeds a dynamic threshold value, performing causal attribution analysis to generate a deviation attribution weight matrix; performing iterative fine tuning on the control instruction according to an attribution result until convergence, and outputting a closed-loop steady-state control flow field to execute compensation; and finally, evaluating and collecting performance, constructing a digital twinning process resume, and feeding back a function mapping relation of an optimization system. According to the method, a cooperative control strategy integrating prediction planning, real-time tracking, causal attribution, dynamic compensation and learning optimization is adopted, and intelligent detection and control of precision, self-adaption and continuous optimization of the blood cell separation process can be achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of blood cell detection, and particularly relates to a centrifugal belt type blood cell separation detection system and a detection method. BACKGROUND

[0002] The centrifugal belt type blood cell separation detection system is an important medical and scientific research equipment, and its core function is to accurately separate whole blood into different components such as plasma, platelets, red blood cells and white blood cells through a centrifugal force field, so as to perform subsequent quantitative detection, component analysis or therapeutic collection. This technology plays a key role in the fields of clinical diagnosis, blood product preparation and cell therapy, and its separation accuracy and efficiency directly affect the reliability of the detection results and the treatment effect.

[0003] The existing blood cell separation detection system usually adopts a fixed parameter program or a simple single variable feedback control in the control strategy. For example, the speed of the centrifuge and the flow rate of the blood pump can be set according to a general model, or the position of the interface of a certain component can be monitored by an optical sensor, and the pump speed can be simply adjusted based on the deviation of the position. These methods can complete the basic separation task when facing ideal or standardized samples, and constitute the mainstream technical scheme in the current field.

[0004] However, the above-mentioned prior art scheme has obvious technical defects in actual application. Since the physiological indicators of blood of different blood donors, such as hematocrit, plasma protein concentration and blood viscosity, have significant individual differences, it is often difficult to achieve ideal separation effect by using fixed control parameters. At the same time, the simple feedback control lacks the ability to diagnose the root cause of the deviation, and cannot distinguish whether the disturbance is caused by the change of blood characteristics or the drift of equipment performance, resulting in blind and low-efficiency adjustment measures, which is easy to produce control oscillation or response lag, and then affects the purity and recovery rate of the target component. SUMMARY

[0005] To solve the above problems, the present application provides a centrifugal belt type blood cell separation detection system and a detection method, which adopts a collaborative control strategy integrating prediction planning, real-time tracking, causal attribution, dynamic compensation and learning optimization, and can realize intelligent detection and control of the precision, self-adaptation and continuous optimization of the blood cell separation process.

[0006] The above-mentioned purpose can be realized by the following scheme: A centrifugal belt type blood cell separation detection system, comprising: An individual separation strategy planning module, which generates an initial fluid dynamics state baseline by coupling a centrifugal belt type separator for self-checking, combines the physiological indicators of blood donors and the quantitative indicators of target collection components, performs multivariate coupled metrology operation, and generates an initial optimal control trajectory; A trajectory tracking and deviation analysis module performs separation according to an initial optimal control trajectory, collects a high-frequency optical image data stream of a multi-layer component interface, obtains a real-time separation trajectory, and compares the real-time separation trajectory with the initial optimal control trajectory to quantitatively generate a real-time dynamic deviation vector; A deviation traceability and causal attribution module risk-weights time points of the initial optimal control trajectory, generates a transient deviation threshold that changes with a process stage, triggers attribution analysis when an amplitude of the real-time dynamic deviation vector exceeds the transient deviation threshold, and generates a deviation attribution weight matrix; A dynamic re-planning and compensation control module reversely compensates to calculate instant compensation parameters according to the deviation attribution weight matrix, fuses the instant compensation parameters with the initial optimal control trajectory to form a modified control instruction, continuously receives the real-time dynamic deviation vector to iteratively fine-tune the modified control instruction until convergence, and outputs a closed-loop steady-state control flow field; A process evaluation and knowledge sedimentation module performs quantitative detection on collected products to generate performance evaluation indexes after target component extraction and remaining component feedback, integrates the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix, and the closed-loop steady-state control flow field, constructs and outputs a digital twin process history, weights the digital twin process history according to the performance evaluation indexes, and feeds back a result to iteratively optimize a function mapping relationship.

[0007] Preferably, the individual separation strategy planning module comprises: A multi-dimensional state vector construction unit is configured to calibrate and generate an initial fluid dynamics state baseline by performing a pipeline self-check on a coupled centrifugal belt separator, fuse the state baseline, received donor physiological indexes, and target collection component quantitative indexes, and construct an initial multi-dimensional state vector containing physical characteristics and separation task constraints; An optimal control trajectory solving unit is configured to establish a function mapping relationship of individual separation characteristics by multivariate coupled metering operation based on the initial multi-dimensional state vector, and perform constraint optimization solving to generate an initial optimal control trajectory.

[0008] Preferably, the trajectory tracking and deviation analysis module comprises: An interface image analysis and trajectory construction unit is configured to collect a high-frequency optical image data stream of a multi-layer component interface, identify boundaries of each component layer by performing edge detection and gradient analysis on the optical image data stream, and convert pixel coordinates into physical radial coordinates to construct a real-time separation trajectory; A trajectory comparison and deviation quantification unit is configured to issue an execution instruction to hardware according to the initial optimal control trajectory, synchronously receive the real-time separation trajectory, and quantitatively generate a real-time dynamic deviation vector by performing time series vector difference operation at each time point.

[0009] Preferably, the deviation traceability and causal attribution module comprises: a dynamic risk assessment and triggering unit for parsing the initial optimal control trajectory, identifying the key separation stage and weighting the risk to generate a transient deviation threshold; a real-time physical cause analysis unit for collecting real-time dynamic deviation vectors and synchronized multi-dimensional sensing data when the amplitude of the monitored real-time dynamic deviation vector exceeds the transient deviation threshold, extracting and generating a physical cause feature spectrum through cross-correlation operation and time-frequency analysis; a root cause feature regression and weight quantification unit for performing multi-element nonlinear regression quantification on the physical cause feature spectrum and the donor physiological indicators, tracing and analyzing the deviation root cause caused by individual differences, and generating a deviation attribution weight matrix.

[0010] Preferably, the multi-dimensional sensing data includes rheological state data and device execution state data, wherein: the rheological state data is used to characterize the transient changes of blood rheological properties and quantify the physical disturbance source caused by donor individual physiological differences; the device execution state data is used to monitor the real-time working state of the separation execution hardware and identify the execution deviation caused by device performance drift or pipeline dynamic impedance changes.

[0011] Preferably, the system further comprises: fusing the initial multi-dimensional state vector and the deviation attribution weight matrix, generating a mapping fidelity evaluation matrix through correlation modeling and difference calculation; according to the mapping fidelity evaluation matrix, identifying the low-fidelity area in the function mapping relationship, and performing targeted parameter optimization and correction amount calculation to generate a set of feedforward correction parameters.

[0012] Preferably, the dynamic re-planning and compensation control module comprises: a compensation parameter solving unit for quantitatively calculating the real-time dynamic deviation vector through inverse compensation based on the deviation attribution weight matrix to solve the instant compensation parameters; an instruction fusion and re-planning unit for superimposing the instant compensation parameters on the subsequent path of the initial optimal control trajectory to generate a modified control instruction of the initial version, and starting a closed-loop iterative fine-tuning process; an iterative convergence and flow field generation unit for continuously receiving the real-time dynamic deviation vector as a feedback signal during the closed-loop iterative fine-tuning process, and adjusting the modified control instruction according to the feedback signal until the amplitude of the real-time dynamic deviation vector is stabilized within the transient deviation threshold, and solidifying the converged modified control instruction as a closed-loop steady-state control flow field.

[0013] Preferably, solving the instant compensation parameters comprises: parsing the deviation attribution weight matrix, identifying and locking the dominant disturbance factor with the first contribution weight; The target hardware to be adjusted is determined according to a dominant disturbance factor, and a control adjustment amount for the target hardware is calculated in combination with the amplitude and direction of the real-time dynamic deviation vector, so as to generate instant compensation parameters.

[0014] Preferably, the process evaluation and knowledge accumulation module comprises: A final performance evaluation unit is configured to perform quantitative detection on the final collection product after the extraction and back-feeding operation is completed, and compare the quantitative detection result with the quantitative index of the target collection component to generate a performance evaluation index; A process history construction unit is configured to integrate the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix and the closed-loop steady-state control flow field, perform structure and time sequence alignment, construct and output a digital twin process history; A function mapping optimization unit is configured to perform weighted adjustment on the function mapping relationship according to the performance evaluation index, in combination with the feedforward correction parameter set and the digital twin process history.

[0015] A centrifugal belt blood cell separation and detection method based on the above system, comprising the following steps: An initial fluid dynamics state baseline is generated by performing self-checking on the coupled centrifugal belt separator, and a multivariate coupled metrology operation is performed in combination with the physiological index of the blood donor and the quantitative index of the target collection component to generate an initial optimal control trajectory; The separation is performed according to the initial optimal control trajectory, high-frequency collection of the optical image data stream of the multi-layer component interface is performed to obtain a real-time separation trajectory, and the real-time separation trajectory is compared with the initial optimal control trajectory to quantitatively generate a real-time dynamic deviation vector; The time point of the initial optimal control trajectory is risk-weighted to generate a transient deviation threshold that changes with the process stage, and the attribution analysis is triggered when the amplitude of the real-time dynamic deviation vector exceeds the transient deviation threshold to generate a deviation attribution weight matrix; The instant compensation parameters are calculated according to the deviation attribution weight matrix, and the initial optimal control trajectory is fused to form a modified control instruction, the real-time dynamic deviation vector is continuously received to iteratively fine-tune the modified control instruction until convergence, and a closed-loop steady-state control flow field is output; After the target component extraction and the remaining component back-feeding are completed, the collection product is quantitatively detected to generate a performance evaluation index, the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix and the closed-loop steady-state control flow field are integrated to construct and output a digital twin process history, the digital twin process history is weighted learned according to the performance evaluation index, and the result is fed back to iteratively optimize the function mapping relationship.

[0016] The present application has the following advantages: The present application can fully consider the physiological indicators of the blood donor and the equipment state before the separation task starts by constructing a personalized initial separation strategy combined with feedforward correction, and plan the optimal control trajectory. This makes the entire separation process have initial accuracy from the beginning, effectively avoids the early large deviation and inefficient adjustment caused by individual differences in the traditional fixed parameter method, shortens the time to reach a stable separation state, and improves the overall operation efficiency.

[0017] The present application realizes real-time diagnosis and adaptive control of the separation process. By tracking the actual separation trajectory with high precision, quantifying the deviation from the theoretical trajectory, and starting causal attribution analysis when the deviation exceeds the dynamic threshold, it can identify whether the deviation is caused by changes in the physiological characteristics of the blood donor or equipment performance drift. This root cause-based diagnostic capability enables the subsequent dynamic compensation control to be targeted, achieving control accuracy and system robustness far beyond traditional single feedback loops, ensuring separation quality under complex dynamic disturbances.

[0018] The present application establishes a complete knowledge sedimentation and self-evolution mechanism. By structuring the full process data of each separation into a digital twin process history, and combining it with the final performance evaluation index for weighted learning, it can continuously optimize its core function mapping relationship from each successful or failed experience. It is no longer a static execution tool, but an intelligent system that can continuously improve its prediction, diagnosis, and control capabilities with increasing application times, achieving long-term and continuous optimization of performance. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is the framework diagram of the system of the present application; Figure 2 is the complete architecture schematic diagram of the system of the present application; Figure 3 is the trajectory deviation dynamic heat map in the embodiment of the present application; Figure 4 is the physiological indicator and deviation degree relationship diagram in the embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0021] Embodiment 1: A centrifugal belt blood cell separation detection system adopts a collaborative control strategy integrating prediction planning, real-time tracking, causal attribution, dynamic compensation, and learning optimization, which can realize intelligent detection and control of the precision, adaptability, and continuous optimization of the blood cell separation process.

[0022] As shown in Figure 1 and Figure 2 , the system specifically includes: An individual separation strategy planning module is configured to perform self-checking by coupling a centrifugal belt separator to generate an initial fluid dynamics state baseline, and perform multivariate coupling calculation by combining physiological indicators of a blood donor and quantitative indicators of target collection components, to establish a functional mapping relationship of individual separation characteristics and perform reverse solving (constrained optimization solving) to generate an initial optimal control trajectory; A trajectory tracking and deviation analysis module is configured to perform separation according to the initial optimal control trajectory, and collect high-frequency optical image data streams of multi-layer component interfaces, to obtain real-time separation trajectories by performing real-time gradient analysis and coordinate transformation on the optical image data streams, and compare the real-time separation trajectories with the initial optimal control trajectory to quantitatively generate real-time dynamic deviation vectors; A deviation tracing and causal attribution module is configured to risk-weight time points of the initial optimal control trajectory to generate instantaneous deviation thresholds varying with process stages, and trigger attribution analysis when an amplitude of the real-time dynamic deviation vector exceeds the instantaneous deviation threshold, and perform regression quantification on results and physiological indicators of the blood donor to generate a deviation attribution weight matrix; A dynamic re-planning and compensation control module is configured to reversely compensate to calculate instant compensation parameters according to the deviation attribution weight matrix, fuse the instant compensation parameters with the initial optimal control trajectory to form a modified control instruction, continuously receive the real-time dynamic deviation vector to iteratively fine-tune the modified control instruction until convergence, and output a closed-loop steady-state control flow field; A process evaluation and knowledge sedimentation module is configured to perform quantitative detection on collected products to generate performance evaluation indicators after target component extraction and remaining component return, integrate the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix, and the closed-loop steady-state control flow field to construct and output a digital twin process history, weight-learn the digital twin process history according to the performance evaluation indicators, and feed back results to iteratively optimize the functional mapping relationship.

[0023] The collaborative control strategy integrates prediction planning, real-time tracking, causal attribution, dynamic compensation, and learning optimization, and can realize intelligent detection and control of blood cell separation process in terms of precision, self-adaptation, and continuous optimization.

[0024] The individual separation strategy planning module includes: A multi-dimensional state vector construction unit is configured to perform pipeline self-checking by coupling a centrifugal belt separator to calibrate and generate an initial fluid dynamics state baseline, and construct an initial multi-dimensional state vector containing physical characteristics and separation task constraints by fusing the state baseline, received physiological indicators of a blood donor, and quantitative indicators of target collection components. After the centrifugal belt separator and the main machine function interface complete the physical coupling, first start the pipeline self-check. By pumping a standard fluid with known physical properties, such as medical saline, into the disposable pipeline system, while changing the output power of the fluid pump group and the rotational angular velocity of the centrifuge, and monitoring the response data in the pipeline system in real time by the built-in pressure and flow sensor. By fitting analysis of the collected response data, the inherent fluid resistance, effective internal volume of the current pipeline system and the actual output performance of the pump group are calibrated, and an initial fluid dynamics state baseline is generated to represent the unique physical properties of the current hardware.

[0025] Subsequently, the physiological indicators of the blood donor input from the operation terminal, such as hematocrit, pre-collection platelet count and blood viscosity, and the target collection component quantification indicators defining the specific needs of this separation task, such as the target collection amount of platelets, the target collection volume of plasma and the allowed maximum separation time, are received.

[0026] Finally, the initial fluid dynamics state baseline, the physiological indicators of the blood donor and the target collection component quantification indicators are structured and fused, and an initial multi-dimensional state vector is constructed. Each dimension of the vector corresponds to a specific physical or biological parameter, providing a complete and individualized initial condition for subsequent calculations.

[0027] The optimal control trajectory calculation unit is used to establish a function mapping relationship of individual separation characteristics based on the initial multi-dimensional state vector through multivariate coupled metrology operation, and to perform constrained optimization solving to generate an initial optimal control trajectory.

[0028] Taking the initial multi-dimensional state vector as input, multivariate coupled metrology operation is performed to establish a function mapping relationship that can represent the nonlinear and strong coupling relationship between control parameters and separation results. This function mapping relationship describes how a set of time-varying control inputs, such as the centrifugal speed sequence and the pump group flow rate sequence, will affect the dynamic formation, migration and stability of the component interfaces during the separation process under specific initial conditions.

[0029] After establishing this relationship, the problem of generating an optimal control trajectory is converted into a constrained optimization solving problem. The goal of this problem is to find a set of optimal control input sequences so that the cost function, which integrates separation efficiency, collection purity, operation time and equipment running stability, reaches a minimum value. This solving process is constrained by a series of boundary conditions, which are derived from the target collection component quantification indicators and equipment safety specifications, such as the final collection amount must reach the target value, the pipeline pressure must not exceed the safety threshold, the total separation time must not exceed the upper limit, etc.

[0030] The numerical optimization algorithm, such as the sequential quadratic programming method, is used to solve the constrained optimization problem reversely, and a set of time-varying, theoretically optimal operation instruction sequence is calculated, which is the initial optimal control trajectory.

[0031] The trajectory tracking and deviation analysis module includes: The interface image analysis and trajectory construction unit is used for high-frequency acquisition of multi-layer component interface optical image data stream, and through edge detection and gradient analysis on the optical image data stream, each component layer boundary is identified, and pixel coordinates are converted into physical radial coordinates to construct a real-time separation trajectory. When the centrifuge is running, the optical image sensor, such as a charge-coupled device camera, configured in the separation area continuously captures images in the centrifugal belt separator at a frequency of not less than 50 frames per second, forming a continuous and dynamic optical image data stream. For each frame of image in the optical image data stream, first, the edge detection is applied to calculate the change rate of luminance or chrominance value between adjacent pixels in the image, i.e. the gradient, to identify the area with the most dramatic change.

[0032] These areas correspond to clear interfaces between blood components of different densities in the physical space, such as the boundary between the platelet-poor plasma layer and the platelet-rich plasma layer. After identifying these component layer boundaries, their pixel coordinate sequences in the image coordinate system are extracted. Then, a key coordinate transformation operation is performed to convert these pixel coordinate sequences into physical radial coordinate sequences with the centrifuge rotation axis as the origin according to the mapping relationship calibrated during device initialization. Finally, the continuously generated physical radial coordinate sequences are connected to construct one or more real-time separation trajectories that can describe the dynamic migration process of each component interface in the physical space.

[0033] The trajectory comparison and deviation quantification unit is used to issue execution instructions to hardware according to the initial optimal control trajectory, synchronously receive real-time separation trajectories, and quantize and generate real-time dynamic deviation vectors by performing time series vector difference operation at each time node.

[0034] On the one hand, according to the initial optimal control trajectory, execution instructions are issued to the centrifuge drive motor and fluid pump group and other hardware at each time node during the separation process. On the other hand, real-time separation trajectories are synchronously received. In order to quantify the deviation between the actual separation process and the theoretical plan, a time series vector difference operation is performed at each time node.

[0035] This operation will perform vector subtraction between the physical radial coordinate vector of each component interface extracted from the real-time separation trajectory at the current time, and the physical radial coordinate vector of each component interface that should theoretically be at this time from the initial optimal control trajectory. The result of the operation is the real-time dynamic deviation vector, the amplitude of which represents the deviation of the actual interface from the ideal interface position, and the direction indicates whether the deviation is towards or away from the rotation center, as shown in Figure 3 The figure shows the deviation of multiple key separation characteristics at different time nodes during a separation process, and the darker areas in the figure clearly indicate significant deviation events quantified by the real-time dynamic deviation vector.

[0036] The deviation traceability and causal attribution module includes: A dynamic risk assessment and triggering unit for analyzing the initial optimal control trajectory, identifying key separation stages and performing risk weighting to generate an instantaneous deviation threshold; The process of generating the instantaneous deviation threshold begins with the analysis of the initial optimal control trajectory. Not all stages in the separation process are equally important. For example, in the key separation stage where the platelet-rich plasma layer is about to form and needs to be accurately extracted, the control accuracy requirement is higher than that in the stable blood inlet or blood return stage. Therefore, by assigning higher risk weights to such key separation stages, a dynamic instantaneous deviation threshold that changes with the process stage can be calculated and generated. This threshold will be tightened in key stages, allowing a smaller deviation range; while in non-key stages, it will be relaxed, thus realizing an intelligent and non-uniform monitoring strategy.

[0037] A real-time physical cause analysis unit for collecting real-time dynamic deviation vectors and synchronized multi-dimensional sensor data when the amplitude of the real-time dynamic deviation vector exceeds the instantaneous deviation threshold, extracting and generating a physical cause feature spectrum through cross-correlation operation and time-frequency analysis; When the amplitude of the real-time dynamic deviation vector exceeds the instantaneous deviation threshold generated in the previous step, real-time analysis of the physical cause is triggered. This analysis process will simultaneously collect two core data streams: one is the real-time dynamic deviation vector representing the deviation phenomenon itself, and the other is the multi-dimensional sensor data from the pump group pressure, fluid viscosity, etc. Through cross-correlation operation on these two data streams, the delay relationship and correlation strength between the deviation phenomenon and each physical parameter change can be revealed. Further, using time-frequency analysis methods such as short-time Fourier transform, the multi-dimensional sensor data can be processed to capture the frequency characteristics of pressure fluctuations or viscosity mutations, etc. Finally, by integrating the calculated correlation strength, delay information and frequency characteristics, a physical cause feature spectrum that can quantitatively represent various direct physical disturbance sources is extracted and generated.

[0038] a root cause feature regression and weight quantification unit configured to perform a multivariate nonlinear regression quantification between the physical cause feature spectrum and the donor physiological indicators, trace and analyze the deviation root cause caused by individual physiological differences, and generate a deviation attribution weight matrix.

[0039] The process of generating the deviation attribution weight matrix aims to trace the root cause from direct physical phenomena to physiological reasons. This process performs a multivariate nonlinear regression quantification between the physical cause feature spectrum generated in the previous step and the donor physiological indicators initially input. For example, the regression analysis can reveal a strong correlation between the continuous high blood viscosity phenomenon embodied in the physical cause feature spectrum and the higher initial hematocrit value in the donor physiological indicators, as shown in FIG. 4. Figure 4 As shown in FIG. 4, each bubble in the figure represents a historical separation event, the horizontal coordinate is the donor hematocrit, the vertical coordinate is the maximum deviation amplitude occurring during the separation process, and the size of the bubble represents the convergence time required to correct the deviation.

[0040] Through this deep trace analysis, the deviation root cause caused by individual physiological differences can be analyzed, and the contribution of these root factors to the current deviation phenomenon is quantified, and finally a deviation attribution weight matrix is generated. This matrix clearly reveals which physiological characteristics of the donor cause the current separation process deviation to a certain extent.

[0041] The multidimensional sensing data includes rheological state data and device execution state data, wherein: The rheological state data is used to characterize the instantaneous changes in blood rheological properties and quantify the physical disturbance source caused by individual physiological differences of the donor; The purpose of generating rheological state data is to capture the changes in the physical properties of blood as a non-Newtonian fluid during the separation process in real time. For this purpose, online rheological sensors such as micro-electro-mechanical system viscometers are integrated at key nodes of the blood flow path, such as the outlet of the blood inlet pump. The sensor calculates the apparent viscosity of the blood in real time by measuring the relationship between the pressure drop and flow rate of the blood fluid in the micro channel. By continuously processing the sensor output signal, a series of rheological state data changing over time can be generated, such as real-time apparent viscosity curves. These data directly reflect the disturbance caused by the physical properties of the donor's blood itself, providing direct physical evidence for subsequent attribution analysis to distinguish between changes in blood properties and device execution problems.

[0042] The device execution state data is used to monitor the real-time working state of the separation execution hardware and identify the execution deviation caused by device performance drift or changes in pipe dynamic impedance.

[0043] The purpose of generating the equipment execution state data is to comprehensively monitor the differences between the actual responses of the hardware components and the instructions. For this purpose, corresponding monitoring sensors are deployed on the key hardware execution mechanisms. For example, the instantaneous rotational speed of the centrifuge is monitored in real time by an optical encoder integrated on the motor shaft; the instantaneous flow rate of each fluid pump group is measured by a high-precision ultrasonic flow meter; the pressure in the pipeline system is monitored by multiple pressure sensors distributed at different positions. The readings of these sensors are collected at a high frequency and compared in real time with the instruction values issued by the control system. For example, the actual flow rate measured by the flow meter is compared with the flow rate target value issued to the pump, and any significant difference may indicate a performance drift of the pump or an increase in dynamic impedance due to partial blockage of the pipeline. These equipment execution state data constitute a direct monitoring of the behavior of the hardware and can capture execution deviations caused by the equipment itself.

[0044] The system further comprises: The initial multi-dimensional state vector and the deviation attribution weight matrix are fused to generate a mapping fidelity evaluation matrix through correlation modeling and difference calculation; The process of generating the mapping fidelity evaluation matrix is a kind of prospective self-checking mechanism. First, based on the initial multi-dimensional state vector of the current separation task, a similarity matching is performed in the historical data knowledge base to retrieve and call the historical separation cases closest to the current donor physiological indicators. Then, the deviation attribution weight matrices corresponding to these retrieved historical cases are fused to generate an expected deviation attribution weight matrix for the current individual. Next, through correlation modeling, the correlation strength between each input parameter in the initial multi-dimensional state vector, such as hematocrit, and the potential disturbance factors indicated in the expected deviation attribution weight matrix is analyzed. At the same time, through difference calculation, the gap between the prediction sensitivity of the function mapping relationship to these key disturbance factors and the actual sensitivity observed in the historical data is quantified. Finally, the correlation strength and the difference are integrated to build a mapping fidelity evaluation matrix, each element of which represents the prediction credibility of the function mapping relationship in handling a certain parameter combination.

[0045] According to the mapping fidelity evaluation matrix, low-fidelity regions in the function mapping relationship are identified, and targeted parameter optimization and correction amount calculation are performed to generate a set of feedforward correction parameters.

[0046] The process of generating the set of feedforward correction parameters starts with the analysis of the mapping fidelity evaluation matrix generated in the previous step. By setting a fidelity threshold, low-fidelity regions with weak prediction ability in the function mapping relationship can be identified, which usually correspond to the response of the model to some extreme or rare physiological parameter combinations. Once these regions are identified, targeted parameter optimization and correction amount calculation are initiated.

[0047] This process does not reconstruct the entire function mapping relationship, but introduces a set of correction parameters in the low-fidelity area, and uses a local optimization algorithm to adjust these correction parameters, so that the prediction results of the function mapping relationship in these specific areas can better fit the rules revealed by the historical data. After optimization, the final calculated set of correction parameters constitutes the set of feedforward correction parameters. This parameter set will be applied as a personalized patch to the function mapping relationship when generating the initial optimal control trajectory, thereby improving the accuracy of planning before the task officially begins.

[0048] The dynamic re-planning and compensation control module includes: A compensation parameter solving unit is configured to quantitatively calculate the real-time dynamic deviation vector by inverse compensation based on the deviation attribution weight matrix, and solve the instant compensation parameters. The process of solving the instant compensation parameters is a model inversion calculation, which aims to inversely calculate the control adjustment amount required to offset the deviation according to the known deviation result and the contribution weight of each cause. This calculation can be represented by the following formula: ; Wherein, is the instant compensation parameters calculated at time point t, which is a vector whose internal elements correspond to the specific control adjustment amount required to be applied to each hardware actuator, such as centrifuge speed, blood pump flow rate, etc.; K is a diagonal gain matrix, each gain coefficient on the diagonal line is derived from and depends on the initial fluid dynamics state baseline, and the gain coefficient is determined by analyzing the calibrated hardware response characteristics in the initial fluid dynamics state baseline, which represents the difference in response sensitivity of different hardware actuators to the same control instruction change, ensuring that the application of compensation amount can match the actual physical performance of the hardware; W is the deviation attribution weight matrix, which quantifies the contribution of each root cause, such as a specific aspect of the physiological indicators of blood donors, to the current observed separation deviation; is the real-time dynamic deviation vector obtained at time point t, which accurately describes the difference between the actual separation state and the ideal state. Through the calculation of this formula, an abstract deviation phenomenon can be converted into a set of quantified instant compensation parameters that can directly act on the hardware.

[0049] An instruction fusion and re-planning unit is configured to superimpose the instant compensation parameters onto the subsequent path of the initial optimal control trajectory, generate the initial version of the modified control instructions, and start the closed-loop iterative fine-tuning process. The process of generating the initial version of the modified control instruction is the step of applying the instant compensation parameter calculated in the previous step to the future control plan. This process is not simply a numerical addition, but through a smoothing algorithm, such as spline interpolation, the instant compensation parameter is integrated into the path of the initial optimal control trajectory in the future period of time, to avoid the generation of control mutation, so as to ensure the smooth transition of the hardware execution process. After generating this initial version of the modified control instruction, it is immediately issued to the hardware executor, and a high-frequency closed-loop iterative fine-tuning cycle is simultaneously activated.

[0050] The iterative convergence and flow field generation unit is used to continuously receive the real-time dynamic deviation vector as a feedback signal during the closed-loop iterative fine-tuning process, and adjust the modified control instruction according to the feedback signal until the amplitude of the real-time dynamic deviation vector is stable within the instantaneous deviation threshold. The converged modified control instruction is solidified as a closed-loop steady-state control flow field.

[0051] During the closed-loop iterative fine-tuning process, the real-time dynamic deviation vector is continuously received as a feedback signal, and the modified control instruction is continuously fine-tuned according to the size, trend and historical accumulation of the feedback signal. This iterative adjustment process continues until the amplitude of the real-time dynamic deviation vector converges stably within the instantaneous deviation threshold corresponding to the current separation stage. This convergence state indicates that the disturbance has been successfully suppressed and the system has re-entered an acceptable stable operating interval. Once convergence is achieved, the modified control instruction sequence that has been stabilized and proven effective is solidified to form and output a closed-loop steady-state control flow field.

[0052] The instant compensation parameter includes: Analyzing the deviation attribution weight matrix to identify and lock the dominant disturbance factor with the first contribution weight; The process of identifying and locking the dominant disturbance factor aims to find the main cause of the current separation deviation from among the many potential influencing factors. This process begins with the analysis of the deviation attribution weight matrix, which is a structured data set that quantifies the contribution weight values of each potential disturbance source, such as blood viscosity and performance drift of specific pump groups, to the current observed separation deviation. By traversing the matrix, the disturbance source corresponding to the element with the highest weight value can be found and determined. This disturbance source with the highest weight value is identified as the dominant disturbance factor of the current deviation event, providing a single and focused optimization target for subsequent compensation control.

[0053] According to the dominant disturbance factor, the target hardware to be adjusted is determined, and the control adjustment amount for the target hardware is calculated based on the amplitude and direction of the real-time dynamic deviation vector, generating the instant compensation parameter.

[0054] The process of generating the instant compensation parameters is to convert the abstract attribution results into specific executable operations. After determining the dominant disturbance factor, the target hardware to be adjusted is determined according to the built-in causal knowledge base. The causal knowledge base defines the mapping relationship between different physical disturbance factors and controllable hardware actuators. For example, if the dominant disturbance factor is identified as high blood viscosity, the causal knowledge base will indicate that the target hardware to be adjusted is the centrifuge motor or the fluid pump set. After locking the target hardware, the control adjustment amount for the target hardware is calculated in combination with the amplitude and direction of the real-time dynamic deviation vector. Among them, the amplitude of the real-time dynamic deviation vector determines the adjustment strength, and its direction determines the adjustment direction, such as increasing or decreasing the speed. The final calculated set of explicit control adjustment amounts constitutes the final generated instant compensation parameters.

[0055] The process evaluation and knowledge sedimentation module includes: The final performance evaluation unit is used to quantitatively detect the final collection product after the extraction and back infusion operation is completed, and compare it with the target collection component quantitative index to generate a performance evaluation index; The process of generating the performance evaluation index starts after a complete separation task is completed. Quantitative detection of the final collection product is performed by offline analysis instruments, such as blood cell analyzers, to obtain measured values of key quality indicators such as platelet recovery rate and collection purity. These measured values are compared one by one with the target collection component quantitative index set at the beginning of the task, a series of quantitative differences are calculated, and the differences are integrated to generate one or a group of numerical values, which is the performance evaluation index for comprehensively evaluating the final performance of this separation task.

[0056] The process history construction unit is used to integrate the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix, and the closed-loop steady-state control flow field, to perform structured and time sequence alignment, construct and output the digital twin process history; The process of constructing the digital twin process history is to integrate all the core data features generated during the entire task. These data features include: the initial optimal control trajectory, the complete real-time dynamic deviation vector time sequence, all the generated deviation attribution weight matrices, and the finally formed closed-loop steady-state control flow field. By structuring and time sequence aligning these heterogeneous data, they are associated according to a unified time stamp to form a multi-dimensional time series data set. This highly structured data set completely records the entire dynamic process from planning to execution, to deviation-diagnosis-adjustment, and constitutes a digital twin process history corresponding to the physical separation process, and is stored for subsequent analysis and learning.

[0057] The function mapping optimization unit is configured to iteratively optimize the function mapping relationship according to the performance evaluation index, combined with the feedforward correction parameter set and the digital twin process history.

[0058] The process of iteratively optimizing the function mapping relationship is the core link to realize long-term learning and self-evolution. This process takes the performance evaluation index generated in the previous step as a kind of learning reward and punishment signal. For example, if the performance evaluation index shows excellent results, the successful digital twin process history of this time will obtain a higher weight in learning. This learning algorithm combines two key inputs: one is the feedforward correction parameter set containing theoretical correction suggestions, and the other is the digital twin process history containing the whole process of this practice. According to the weight given by the performance evaluation index, the function mapping relationship is iteratively optimized, so that the relationship can adopt more strategies that guide high-performance results when planning in the future, and correct the prediction patterns that have led to adverse results. The completion of each separate task provides new learning samples for the optimization of the function mapping relationship, so that its performance can continuously evolve.

[0059] In order to verify the feasibility and effectiveness in actual application, the embodiment is applied to the individualized cell therapy preparation process of a clinical research laboratory. This process requires the separation of a specific number of platelets from the peripheral blood of donors with different health conditions for subsequent cell culture and preparation, and puts forward very high requirements for the precision, stability and automation degree of the separation process. The traditional method relies on fixed operation procedures, and when faced with physiological differences between donors, often appears separation efficiency fluctuation, product quality uneven and the need for experienced technical personnel to intervene throughout the process, etc.

[0060] The centrifugal belt blood cell separation and detection system proposed by the embodiment is used for a specific donor B, whose physiological indicators are: hematocrit 42%, pre-harvest platelet count 180×10 9 / L, and a single apheresis task with a target of collecting 2.0×10 11 platelets is performed.

[0061] First, individualized separation strategy planning is performed. After physically coupling the disposable centrifugal belt separator with the host computer, pipeline self-checking is performed, the flow resistance is monitored by pumping physiological saline with a known viscosity, and the initial fluid dynamics state baseline for this task is calibrated and generated. Subsequently, the initial multi-dimensional state vector is constructed by fusing the state baseline, the physiological indicators of donor B, and the target collection component quantization indicators containing the target collection amount. Based on the vector, the function mapping relationship of individual separation characteristics is established through multivariate coupled metrology operation, and the constrained optimization solution is performed, and finally an initial optimal control trajectory customized for donor B is generated.

[0062] In the separation process execution phase, hardware commands were issued according to the generated initial optimal control trajectory, while high-frequency optical image data stream of blood layering in the separation chamber was collected. When the separation process proceeded to the 22nd minute, the real-time separation trajectory was constructed by performing edge detection and gradient analysis on the optical image data stream, and converting the identified component layer boundary pixel coordinates into physical radial coordinates. By performing time-series vector difference operation on the real-time separation trajectory and the initial optimal control trajectory, a real-time dynamic deviation vector was quantitatively generated. It was detected that the amplitude of the vector exceeded the instantaneous deviation threshold of this phase generated according to the risk weighting of the initial optimal control trajectory.

[0063] Subsequently, the traceability and causal attribution analysis of the deviation was triggered. At this time, multi-dimensional sensing data was synchronously collected, among which the rheological state data showed that the apparent viscosity of the blood had a small increase, and the device execution state data showed that the actual output flow rate of the blood pump was slightly lower than the instruction value. By performing cross-correlation operation and time-frequency analysis on the real-time dynamic deviation vector and these multi-dimensional sensing data, a physical cause feature spectrum was generated. Then, by performing multivariate nonlinear regression quantification on the physical cause feature spectrum and the physiological indicators of donor B, a deviation attribution weight matrix was finally generated. The matrix clearly pointed out that the contribution weight of the blood viscosity increase caused by the physiological differences of the donor individual was 0.7, and the contribution weight of the execution deviation caused by the performance drift of the device was 0.3.

[0064] Based on the attribution result, dynamic re-planning and compensation control was started. First, the deviation attribution weight matrix was analyzed to identify and lock the blood viscosity increase with the first contribution weight as the dominant disturbance factor. According to the dominant disturbance factor, the target hardware to be adjusted was determined as the centrifuge motor. Combined with the amplitude and direction of the real-time dynamic deviation vector, the immediate compensation parameter of increasing the centrifuge speed by 60 RPM was calculated, and it was smoothly superimposed on the subsequent path of the initial optimal control trajectory to generate the initial version of the modified control instruction. Subsequently, a closed-loop iterative fine-tuning process was started, and the speed was continuously fine-adjusted according to the feedback real-time dynamic deviation vector until the amplitude of the deviation vector was stably converged within the instantaneous deviation threshold. At this moment, the stable control instruction sequence was solidified as a closed-loop steady-state control flow field, and guided the subsequent separation process.

[0065] After the task was completed, the collected product was quantitatively detected, and the results showed that the actual collected platelets were 2.1 x 10 11One, the purity is as high as 98.9%, the total time consumption is 65 minutes, and excellent performance evaluation indexes are generated. Subsequently, all data of the initial optimal control trajectory of the task, the real-time dynamic deviation vector, the deviation attribution weight matrix and the closed-loop steady-state control flow field are integrated to construct and output a digital twin process record of the blood donor B. According to the excellent performance evaluation indexes, in combination with the theoretical feedforward correction parameter set generated by analyzing historical data before this operation and the digital twin process record generated this time, the core function mapping relationship is weighted learned and iteratively optimized.

[0066] In a six-month clinical application, 80 blood donors with different physiological characteristics were selected and randomly divided into an experimental group and a control group. The average collection time of the experimental group was shortened by about 18%, the average recovery rate of the target component was increased by about 15%, and the batch consistency of the collected products was improved by more than 60%. Most importantly, the demand for human intervention throughout the process of the experimental group was reduced by more than 90%. In addition, by tracking the data of the experimental group for six consecutive months, it was found that with the continuous accumulation of the digital twin process record and the continuous iterative optimization of the function mapping relationship, the planning accuracy of the initial optimal control trajectory was improved month by month, which was manifested as an average decrease of about 8% in the frequency of triggering dynamic compensation during the separation process every month, proving that the present application has the ability of self-learning and continuous evolution.

[0067] Embodiment 2: A centrifugal belt type blood cell separation and detection method, comprising the following steps: Performing self-checking by coupling the centrifugal belt separator to generate an initial fluid dynamics state baseline, performing multivariate coupled metrology operation in combination with the physiological indicators of the blood donor and the quantitative indicators of the target collected components to generate an initial optimal control trajectory; Performing separation according to the initial optimal control trajectory, high-frequency collection of optical image data streams of multi-layer component interfaces to obtain real-time separation trajectories, and comparison with the initial optimal control trajectory to quantitatively generate real-time dynamic deviation vectors; Risk weighting of the time points of the initial optimal control trajectory to generate instantaneous deviation thresholds varying with process stages, triggering attribution analysis when the amplitude of the real-time dynamic deviation vector exceeds the instantaneous deviation threshold to generate a deviation attribution weight matrix; According to the deviation attribution weight matrix, the real-time compensation parameters are calculated by inverse compensation, and the initial optimal control trajectory is fused to form a modified control instruction, and the real-time dynamic deviation vector is continuously received to iteratively fine-tune the modified control instruction until convergence, and a closed-loop steady-state control flow field is output; After the target component extraction and the remaining component back feeding are completed, the collected product is quantitatively detected to generate a performance evaluation index, the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix and the closed-loop steady-state control flow field are integrated, a digital twin process history is constructed and output, the digital twin process history is weighted learned according to the performance evaluation index, and a result feedback iterative optimization function mapping relationship is mapped.

[0068] The formula appearing above can translate physical quantities of different properties into unitless standard values or same-dimension superimposable parameters through the principle of dimensional consistency and mathematical standardization means (for example, normalization processing, dimensionless parameter conversion or unit unification), so as to eliminate the interference of different dimensions on the operation logic, so that the formula has mathematical operation rationality and objective law adaptability while retaining the original data distribution characteristics. The electrical connection between the above-mentioned units does not necessarily mean direct connection of the line, and indirect connection can also be used as long as the purpose of the application is achieved. The above is only an exemplary embodiment of the application, and cannot limit the scope of the application.

Claims

1. A centrifugal belt blood cell separation and detection system, comprising: Comprise: An individual separation strategy planning module, which performs self-checking by coupling a centrifugal belt separator to generate an initial fluid dynamics state baseline, combines donor physiological indicators and target collection component quantitative indicators, performs multivariate coupled metrology operations, and generates an initial optimal control trajectory; A trajectory tracking and deviation analysis module, which performs separation based on the initial optimal control trajectory, high-frequency acquisition of multi-layer component interface optical image data streams, and obtaining of real-time separation trajectories, and comparison with the initial optimal control trajectory to quantitatively generate real-time dynamic deviation vectors; A deviation tracing and causal attribution module, which risk-weights time points of the initial optimal control trajectory to generate instantaneous deviation thresholds that vary with process stages, triggers attribution analysis when the amplitude of the real-time dynamic deviation vector exceeds the instantaneous deviation threshold, and generates a deviation attribution weight matrix; A dynamic re-planning and compensation control module, which reversely compensates to calculate instant compensation parameters based on the deviation attribution weight matrix, fuses with the initial optimal control trajectory to form a modified control instruction, continuously receives the real-time dynamic deviation vector to iteratively fine-tune the modified control instruction until convergence, and outputs a closed-loop steady-state control flow field; A process evaluation and knowledge sedimentation module, which, after completing target component extraction and returning remaining components, quantitatively detects collection products to generate performance evaluation indicators, integrates the initial optimal control trajectory, real-time dynamic deviation vector, deviation attribution weight matrix, and closed-loop steady-state control flow field, constructs and outputs a digital twin process history, and learns from the digital twin process history based on the performance evaluation indicators to feed back the result to iteratively optimize the function mapping relationship.

2. The centrifugal belt blood cell separation and detection system according to claim 1, wherein, The individual separation strategy planning module comprises: A multi-dimensional state vector construction unit for calibrating and generating an initial fluid dynamics state baseline by performing pipeline self-checking through coupling a centrifugal belt separator, and fusing the state baseline, received donor physiological indicators, and target collection component quantitative indicators to construct an initial multi-dimensional state vector containing physical characteristics and separation task constraints; An optimal control trajectory solving unit for establishing a function mapping relationship of individual separation characteristics through multivariate coupled metrology operations based on the initial multi-dimensional state vector, and performing constraint optimization solving to generate an initial optimal control trajectory.

3. The centrifugal belt blood cell separation and detection system according to claim 1, wherein, The trajectory tracking and deviation analysis module comprises: An interface image analysis and trajectory construction unit for high-frequency acquisition of multi-layer component interface optical image data streams, and identifying component layer boundaries through edge detection and gradient analysis on the optical image data streams, and transforming pixel coordinates into physical radial coordinates to construct real-time separation trajectories; A trajectory comparison and deviation quantification unit for issuing execution instructions to hardware based on the initial optimal control trajectory, synchronously receiving real-time separation trajectories, and quantitatively generating real-time dynamic deviation vectors through time series vector difference operations at each time node.

4. The centrifugal belt blood cell separation and detection system according to claim 1, wherein, The deviation tracing and causal attribution module comprises: A dynamic risk assessment and triggering unit for analyzing the initial optimal control trajectory, identifying key separation stages, and risk-weighting to generate instantaneous deviation thresholds; A real-time physical cause analysis unit is configured to, when an amplitude of a real-time dynamic deviation vector exceeds a transient deviation threshold, collect the real-time dynamic deviation vector and synchronized multi-dimensional sensing data, and extract and generate a physical cause feature spectrum through cross-correlation operation and time-frequency analysis; A root cause feature regression and weight quantization unit is configured to perform multi-element nonlinear regression quantization on the physical cause feature spectrum and physiological indicators of the blood donor, trace and analyze deviation roots caused by individual differences, and generate a deviation attribution weight matrix.

5. The centrifugal belt blood cell separation and detection system according to claim 4, wherein, The multi-dimensional sensing data includes rheological state data and device execution state data, wherein: The rheological state data is used to represent transient changes in blood rheological characteristics and to quantify physical disturbance sources caused by individual physiological differences of the blood donor; The device execution state data is used to monitor real-time working states of separation execution hardware and to identify execution deviations caused by device performance drift or changes in pipe dynamic impedance.

6. The centrifugal belt blood cell separation and detection system according to claim 2, wherein, The system further includes: A fusion initial multi-dimensional state vector and the deviation attribution weight matrix are fused to generate a mapping fidelity evaluation matrix through correlation modeling and difference calculation; According to the mapping fidelity evaluation matrix, low-fidelity areas in the function mapping relationship are identified, and targeted parameter optimization and correction amount calculation are performed to generate a set of feedforward correction parameters.

7. The centrifugal belt blood cell separation and detection system of claim 1, wherein, The dynamic re-planning and compensation control module includes: A compensation parameter solving unit is configured to, based on the deviation attribution weight matrix, perform quantitative calculation on the real-time dynamic deviation vector through inverse compensation to solve instant compensation parameters; An instruction fusion and re-planning unit is configured to superimpose the instant compensation parameters on a subsequent path of the initial optimal control trajectory to generate a corrected control instruction of an initial version, and to start a closed-loop iterative fine-tuning process; An iterative convergence and flow field generation unit is configured to, in the closed-loop iterative fine-tuning process, continuously receive the real-time dynamic deviation vector as a feedback signal, and adjust the corrected control instruction according to the feedback signal until the amplitude of the real-time dynamic deviation vector is stabilized within the transient deviation threshold, and the converged corrected control instruction is solidified as a closed-loop steady-state control flow field.

8. The centrifugal belt blood cell separation and detection system according to claim 7, wherein, Solving the instant compensation parameters includes: Analyzing the deviation attribution weight matrix to identify and lock a dominant disturbance factor with the highest contribution weight; According to the dominant disturbance factor, a target hardware to be adjusted is determined, and a control adjustment amount for the target hardware is calculated based on the amplitude and direction of the real-time dynamic deviation vector to generate the instant compensation parameters.

9. The centrifugal belt blood cell separation and detection system of claim 6, wherein, The process evaluation and knowledge sedimentation module includes: A final performance evaluation unit is configured to, after completing the extraction and return operation, perform quantitative detection on the final collected product, compare it with the quantitative indicators of the target collection components, and generate a performance evaluation indicator; A process history construction unit is configured to integrate the initial optimal control trajectory, the real-time dynamic deviation vector, the deviation attribution weight matrix, and the closed-loop steady-state control flow field, perform structure and time sequence alignment, construct and output a digital twin process history; A function mapping optimization unit is configured to, according to the performance evaluation indicator, combine the feedforward correction parameter set and the digital twin process history for weighted adjustment, and iteratively optimize the function mapping relationship.

10. A centrifugal belt blood cell separation method based on the system of any one of claims 1-9, characterized by the steps of ​ Performing self-check by coupling centrifugal belt separator to generate initial fluid dynamics state baseline, combining donor physiological indicators with target collection component quantification indicators, performing multivariate coupling quantification operation to generate initial optimal control trajectory; Performing separation according to initial optimal control trajectory, high-frequency collection of multi-layer component interface optical image data stream to obtain real-time separation trajectory, and comparison with initial optimal control trajectory to generate real-time dynamic deviation vector; Risk weighting of initial optimal control trajectory time points to generate instantaneous deviation threshold that changes with process stage, triggering attribution analysis when real-time dynamic deviation vector amplitude exceeds instantaneous deviation threshold to generate deviation attribution weight matrix; According to deviation attribution weight matrix, calculating immediate compensation parameters by reverse compensation, and fusing with initial optimal control trajectory to form modified control instruction, continuously receiving real-time dynamic deviation vector to iteratively fine-tune modified control instruction until convergence, and outputting closed-loop steady-state control flow field; After completing target component extraction and returning remaining components, performing quantitative detection on collection products to generate performance evaluation indicators, integrating initial optimal control trajectory, real-time dynamic deviation vector, deviation attribution weight matrix and closed-loop steady-state control flow field to construct and output digital twin process history, and performing weighted learning on digital twin process history according to performance evaluation indicators, and feeding back the result to iterative optimization function mapping relationship.

Images