Dynamic matrix control method and system for exosome extraction

Abstract

The invention discloses a dynamic matrix control method and system for exosome extraction, and relates to the technical field of exosome extraction, the system and the method realize digital integration and dynamic analysis of execution elements, sensors and control logic by constructing and maintaining a uniform dynamic state matrix in real time, and the system and the method are suitable for large-scale industrial production. Flow path control is converted from a fixed mode to flexible self-adaptive regulation and control, the action of a pump valve is automatically adjusted according to parameters such as the liquid level, and the automation degree and the process stability are remarkably improved. The membrane package load is dynamically balanced through an intelligent algorithm, process parameters are finely adjusted, membrane package blockage is relieved while the extraction efficiency and the product consistency are improved, and the service life of consumables is prolonged. In the knowledge application level, the system has autonomous learning and evolution capabilities, and quantitative evaluation and iterative optimization of the historical formula are realized by converting operation data into a structured performance file and constructing a knowledge base.

Description

Technical Field

[0001] This invention relates to the field of exosome extraction technology, and more specifically, to a dynamic matrix control method and system for exosome extraction. Background Technology

[0002] Exosomes, as extracellular vesicles with multiple values ​​including intercellular communication mediators, disease diagnostic biomarkers, and drug delivery carriers, have become a research hotspot in the biomedical field. Their clinical translation places stringent demands on the efficiency, purity, reproducibility, and activity preservation of extraction technologies. However, traditional exosome extraction methods and related equipment have significant limitations in their control modes. They mostly employ fixed flow path configurations and parameter settings, lacking flexible adaptability and the ability to automatically adjust the actions of actuators according to changes in operating conditions during the extraction process. This often requires manual intervention midway, which not only reduces operational efficiency but also easily introduces contamination and operational errors, thus hindering the automation and standardization of exosome extraction.

[0003] Existing exosome extraction equipment suffers from relatively crude process control, failing to establish a synergistic regulation mechanism for multi-dimensional process parameters. During extraction, core process parameters such as transmembrane pressure, concentration ratio, and fluid composition are interconnected and dynamically change. However, traditional equipment lacks comprehensive monitoring and intelligent analysis of these parameters, making it difficult to achieve dynamic balance among them. This not only leads to significant batch-to-batch fluctuations in product concentration and purity but also easily causes problems such as uneven membrane loading and clogging, shortening the lifespan of core consumables. Furthermore, it cannot accurately match the control requirements of different process stages, affecting the stability and reliability of the extraction process.

[0004] The sources of exosome extraction samples are diverse, and the characteristics of different sample types vary significantly, requiring high adaptability of extraction processes. Traditional extraction techniques lack structured process knowledge accumulation and iterative optimization mechanisms, relying heavily on manual trial and error in process development. The transfer and adaptation of process parameters across different scenarios is challenging, making it difficult to quickly generate targeted control schemes. This results in long process development cycles, high optimization costs, and difficulty in establishing standardized extraction procedures, severely hindering the rapid implementation and promotion of exosome extraction technology in various research and clinical settings. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a dynamic matrix control method and system for exosome extraction.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A dynamic matrix control method for exosome extraction includes the following steps: Step 1: During initialization, the system loads the basic state matrix according to the process formula selected by the user. Based on the fundamental state matrix Dynamically generate real-time dynamic state matrix Furthermore, a dynamic correction factor ΔM is introduced to update the dynamic matrix in real time. Step 2: Based on the real-time updated dynamic matrix, update the real-time process health index HI(t) in real time, with the goal of keeping HI(t) in the preset optimal range, and fine-tune the parameters of the actuator in M(t); HI(t) = α*NTMP(t) + β*Nη(t) + γ*NEC(t), where: NTMP(t) = TMP(t) / TMP optimal(t) is the normalized transmembrane pressure; TMP(t) = Pin(t) - Pout(t) is the real-time calculated value, Pin(t) is the inlet pressure, Pout(t) is the outlet pressure; TMP optimal(t) is the dynamic optimal pressure threshold; Nη(t) = η(t) / η_target is the normalized concentration ratio; NEC(t) = |EC(t) - EC baseline| / EC baseline is the normalized conductivity deviation, used to indirectly monitor changes in fluid composition; EC(t) is the real-time conductivity; EC baseline is the baseline conductivity. Step 3: After each process run, the system automatically performs a performance evaluation for each process step, defining a Process Segment Performance Vector (PFVs) to quantify the overall performance of that segment. All segment PFVs, along with their corresponding initial dynamic matrix, are used for this evaluation. The sample feature vector, final product quality index, and key M(t) evolution trajectory features together constitute a process instance record, which is stored in the process knowledge base. A comprehensive fitness score F is calculated for each record in the process knowledge base, and the formula is iteratively self-optimized based on the comprehensive fitness score.

[0007] Furthermore, the real-time dynamic state matrix ;in, Let be the set of state vectors of all peristaltic pumps at time t. Assuming there are N peristaltic pumps in the system, then... The individual state vector of each pump (i=1,2,...,N) are constructed by sequentially concatenating them, and can be represented as follows: ; Let be the individual state vector of the i-th peristaltic pump; Let be the set of state vectors of all electromagnetic pinch valves at time t. Assume there are M electromagnetic pinch valves in the system. The individual state vector of each electromagnetic pinch valve (j=1,2,...,M) are constructed by sequentially concatenating them, and represented as follows: Vj(t) is the individual state vector of the j-th electromagnetic clamp valve.

[0008] Furthermore, ,in, This indicates the start / stop status of the peristaltic pump. For real-time calculation of the rotational speed of the peristaltic pump, This is the inherent flow coefficient of the peristaltic pump.

[0009] Furthermore, Vj(t)=[State j,Priority j], where State j∈{OPEN,CLOSE} represents the on / off state of the j-th solenoid pinch valve, and Priority j is the response priority of the j-th solenoid pinch valve during dynamic adjustment.

[0010] Furthermore, Let be the set of all sensor reading vectors at time t, including inlet pressure Pin(t), outlet pressure Pout(t), feed liquid weight W1(t), filtrate weight W3(t), conductivity EC(t), and each liquid level L(t).

[0011] Furthermore, the dynamic correction factor ΔM = K pump·(L target-L current(t))+K valve·Hysteresis(ΔL(t) / Δt); where L current(t) is the real-time liquid level of the critical container, L target is the desired stable liquid level or threshold; K pump and K valve are the correction coefficient matrices for pump speed and valve opening and closing, respectively; Hysteresis() is a function containing hysteresis.

[0012] Furthermore, TMP optimal(t) = TMP base * [1 + κ * (η(t) / η target)], where TMP base is the membrane base pressure value, η(t) is the real-time concentration ratio, κ is the process progress influence coefficient, and η target is the target concentration ratio.

[0013] Furthermore, PFVs = [εs, ρs, ωs, ts]; where εs is the fragment extraction efficiency, εs = (ΔW_product·C_product) / (ΔW_processing·C_initial) × 100%; ρs is the purity evaluation index, ωs = (TMP_s_end - TMP_s_start) / TMP_s_start × 100%; and ts is the fragment extraction time.

[0014] Furthermore, a dynamic matrix control system for exosome extraction includes: The initial formula loading module is used to load the basic state matrix according to the process formula selected by the user; The dynamic matrix generation module is based on the loaded... Create the real-time dynamic state matrix for this run. ; The primary real-time control module is used to make real-time, adaptive primary adjustments to M(t) based on the dynamic correction factor ΔM. The execution parameter fine-tuning module aims to maintain HI(t) within a preset optimal range by fine-tuning the execution element parameters in M(t). The data archiving module is used to perform performance analysis on each process step of this operation, generate process segment performance vectors (PFVs), and store complete process instance records in the process knowledge base. The formula iteration optimization module is used to evaluate the overall fit score F of this run. If this run is an optimization experiment for a known scenario, the system will compare its F value with the historical records and may automatically start a new round of parameter fine-tuning and running tests. Through iterative loops, it will find the local optimal formula M0optimal for this scenario and update the knowledge base.

[0015] Compared with the prior art, the present invention has the following beneficial effects: At the control architecture level, this system and method construct and maintain a unified dynamic state matrix in real time. This allows for the digital integration and dynamic analysis of the states of all system actuators, sensor readings, and control logic, achieving a fundamental shift from fixed, rigid flow path control to flexible, adaptive real-time regulation. The system can automatically adjust pump and valve actions based on key operating parameters such as liquid level, effectively avoiding the need for manual intervention in traditional equipment and significantly improving the automation level and process stability.

[0016] Secondly, at the process optimization level, an innovative process health index integrating multimodal sensor information is proposed, and a closed-loop collaborative control mechanism is constructed based on this index. This system can comprehensively consider multiple factors such as transmembrane pressure, concentration process, and solution purity, and dynamically balance membrane load and finely adjust process parameters through intelligent algorithms. This not only improves exosome extraction efficiency and product consistency, but also greatly alleviates membrane blockage problems and extends the service life of core consumables.

[0017] Finally, at the knowledge application level, the system possesses the ability to autonomously learn and evolve process knowledge. By transforming each run's data into structured performance profiles and building a knowledge base, the system can quantitatively evaluate and iteratively optimize historical formulas. More importantly, based on the sample characteristics of new tasks, the system can intelligently match or transfer existing process knowledge to quickly generate highly adaptable control schemes. This completely changes the traditional process development model that relies on experience-based trial and error, providing strong support for the standardization and rapid transfer of exosome extraction processes. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the principle of a dynamic matrix control method for exosome extraction. Figure 2 This is a flowchart illustrating the principle of iterative self-optimization of recipes based on comprehensive fitness scores. Detailed Implementation

[0019] Example 1: Refer to Figures 1-2 A dynamic matrix control method for exosome extraction includes the following steps: Step 1: During initialization, the system loads the basic state matrix according to the process formula selected by the user. (During initialization, the system sets the state of M(t) to be the same as...) Consistent, meaning the initial values ​​of P(t0) and V(t0) originate from... The target state defined in the table, S(t0), is acquired from the initial readings of the sensor, based on the fundamental state matrix. Dynamically generate real-time dynamic state matrix Furthermore, a dynamic correction factor ΔM is introduced to update the dynamic matrix in real time.

[0020] Real-time dynamic state matrix ;in, Let be the set of state vectors of all peristaltic pumps at time t. Assuming there are N peristaltic pumps in the system, then... The individual state vector of each pump (i=1,2,...,N) are constructed by sequentially concatenating them, and can be represented as follows: ; Let be the individual state vector of the i-th peristaltic pump, specifically defined as: .in, This indicates the start / stop status of the peristaltic pump. For real-time calculation of the rotational speed of the peristaltic pump, This is the inherent flow coefficient of the peristaltic pump. Let be the set of state vectors of all electromagnetic pinch valves at time t. Assume there are M electromagnetic pinch valves in the system. The individual state vector of each electromagnetic pinch valve (j=1,2,...,M) are constructed by sequentially concatenating them, and represented as follows: Vj(t) is the individual state vector of the j-th electromagnetic pinch valve, Vj(t) = [State j, Priority j], State j ∈ {OPEN, CLOSE} represents the opening and closing state of the j-th electromagnetic pinch valve, and Priority j is the response priority of the j-th electromagnetic pinch valve during dynamic adjustment. Let be the set of all sensor reading vectors at time t, including inlet pressure Pin(t), outlet pressure Pout(t), feed liquid weight W1(t), filtrate weight W3(t), conductivity EC(t), and each liquid level L(t).

[0021] The dynamic correction factor ΔM = K pump·(L target - L current(t)) + K valve·Hysteresis(ΔL(t) / Δt); where L current(t) is the real-time liquid level of the critical container (such as the raw material bottle), L target is the desired stable liquid level or threshold, K pump and K valve are correction coefficient matrices for pump speed and valve opening / closing, respectively, and Hysteresis() is a function containing hysteresis to prevent frequent actions. For example, when the raw material liquid level L current drops rapidly, ΔM will calculate an instruction to lower the raw material pump Speed ​​and may open the backup raw material passage valve in advance to stabilize the feed flow rate.

[0022] The dynamic matrix is ​​updated in real time, specifically as follows: The system monitors the liquid level Lcurrent(t) of key containers (such as raw material bottles and buffer bottles) in real time. When the liquid level change reaches a preset condition, the calculation of the dynamic correction factor ΔM is triggered; then, the matrix state is updated. The dynamic correction factor ΔM is applied to the current real-time dynamic state matrix M(t): M(t)←Update(M(t),ΔM); This update process is specifically manifested as: adjusting the Speed ​​i(t) value of the corresponding peristaltic pump and the State j(t) state of the corresponding solenoid clamp valve in M(t) according to the calculation result of ΔM. Finally, the actuators are driven. The PLC updates the matrix according to the updated M(t). and The state vector generates real-time control signals to drive the corresponding peristaltic pump and solenoid clamp valve, enabling precise fluid delivery and flexible flow path configuration.

[0023] Step 2: Based on the real-time dynamic matrix, update the real-time process health index HI(t) in real time, with the goal of maintaining HI(t) in the preset optimal range (e.g., [0.95, 1.05]), and fine-tune the parameters of the actuator in M(t).

[0024] HI(t) = α*NTMP(t) + β*Nη(t) + γ*NEC(t), where: NTMP(t) = TMP(t) / TMP optimal(t) is the normalized transmembrane pressure. TMP(t) = Pin(t) - Pout(t) is the real-time calculated value, Pin(t) is the inlet pressure, and Pout(t) is the outlet pressure; TMP optimal(t) is the dynamic optimal pressure threshold, which is calculated as: TMP optimal(t) = TMP base * [1 + κ * (η(t) / η target)). TMP base is the membrane packing pressure limit (representing the safe upper limit of transmembrane pressure allowed by the system in the initial stage of the process (when the concentration ratio is close to 0) or under the most pressure-sensitive conditions, which can be preset to 0.25 MPa), η(t) is the real-time concentration ratio, κ is the process progress influence coefficient (a dimensionless empirical coefficient used to quantify the influence of the process concentration process on the optimal pressure threshold. Its value is usually between 0.05 and 0.3, determined through process development experiments), and ηtarget is the target concentration ratio (the preset expected final concentration ratio value to be achieved in the current concentration step). This model allows the pressure threshold to adaptively adjust with the concentration process, rather than being a fixed value. Nη(t) = η(t) / η_target is the normalized concentration ratio, where η(t) is calculated in real time based on the weight of the feed liquid and the weight of the filtrate. NEC(t) = |EC(t) - EC baseline| / EC. The baseline represents the normalized conductivity deviation, used to indirectly monitor changes in fluid composition. EC(t) is the real-time conductivity, measured by an online conductivity sensor installed in a specific flow path of the system (typically the retentate line at the membrane outlet or the final product collection line). EC baseline is the baseline (or target) conductivity, a reference conductivity value set according to the target of the current process step. α, β, and γ are weighting coefficients (α + β + γ = 1), which can be dynamically configured according to different process stages (such as concentration and filtration) to highlight the control focus of the current stage. For example, in the initial concentration stage, transmembrane pressure (α) is given the highest weight to ensure the system is highly sensitive to pressure fluctuations, maintaining TMP near a low TMP base, creating conditions for the stable formation of the core filter layer (gel layer). The concentration ratio (β) has a lower weight and is used to track progress. The conductivity (γ) has a very low weight, so the change in impurity concentration is not significant. The main objective is volume reduction, so α can be 0.5, β can be 0.45, and γ can be 0.05.

[0025] To maintain HI(t) within a preset optimal range (e.g., [0.95, 1.05]), the parameters of the actuators in M(t) are fine-tuned, specifically as follows: Closed-loop regulation of pump speed: When HI(t) deviates from the target range, the system calculates the pump speed adjustment ΔSpeedi. For example, proportional regulation is used: Speedi(t+Δt)=Speedi(t)+Kp*(HI target-HI(t)). Wherein, Kp is the proportional gain coefficient (the physical meaning of Kp is: the pump speed adjustment caused by a unit health deviation (HI target-HI(t)), its value is determined through system debugging or self-tuning; too large will cause system oscillation, too small will result in slow response), and HItarget is the target health (usually 1). This adjustment directly updates Speedi(t) in M(t). Preventive dynamic pressure balancing (valve-pump coordination): To prevent membrane blockage and improve utilization, the system executes a dynamic load balancing algorithm in parallel. The relative load of each parallel membrane module branch is calculated in real time as Load k(t) = TMP k(t) / ΣTMP(t), where TMP k(t) is the transmembrane pressure of the k-th membrane module, specifically the transmembrane pressure value of the k-th independent membrane module (or branch) obtained by measurement or calculation at time t; ΣTMP(t) is the sum of the total transmembrane pressures of the system. If the Load k(t) of a certain branch is continuously higher than the average load, a coordinated adjustment command is generated: the equivalent opening of the inlet valve Vk of the high-load branch is slightly reduced (by adjusting its Priority j or using PWM control) to reduce its instantaneous flow. To maintain the stability of the total system flow, the Speed ​​i(t) of the main feed pump or the corresponding distribution pump is slightly increased in conjunction. The result of this coordinated adjustment is the synchronous update of the state j(t) or priority j(t) of the relevant valves in M(t), as well as the Speed_i(t) of the relevant pumps, to achieve a smooth redistribution of flow and make the pressure of each membrane module tend to be balanced. After making the above intelligent decisions, the system synchronizes the updated parameters (pump speed, valve status / priority) in M(t) to the physical execution layer. Based on the latest state of M(t), the PLC drives the corresponding peristaltic pump and solenoid clamp valve to achieve closed-loop control.

[0026] Step 3: After each process run, the system automatically performs a performance evaluation on each process step (such as concentration and filtration), defining a process segment performance vector (PFVs) to quantify the overall performance of that segment. All segment PFVs, along with their corresponding initial dynamic matrix, are used for this evaluation. The sample feature vector (e.g., type: serum=1, volume: 200mL, initial conductivity, etc.), the final product quality indicators (e.g., particle size distribution, protein concentration), and the key M(t) evolution trajectory features (e.g., average HI value, pressure fluctuation variance) together constitute a process instance record, which is stored in the process knowledge base. A comprehensive fitness score F is calculated for each record in the process knowledge base, and the formulation is iteratively self-optimized based on the comprehensive fitness score.

[0027] PFVs=[εs,ρs,ωs,ts]; where εs is the fragment extraction efficiency, εs=(ΔW_product·C_product) / (ΔW_treatment·C_initial)×100% (ΔW_product: the dry weight or effective weight of the target exosome product obtained in this step; C_product: the concentration of the target exosome product obtained in this step; C_initial: the initial concentration of the target exosome in the feed solution to be treated at the beginning of this step; ΔW_treatment: the theoretical total mass of exosomes contained in the feed solution (or feed liquid) treated in this step); ρs is the purity evaluation index (calculated through a pre-built purity prediction model. The model is established by: during the system process development stage, collecting time-series data of the normalized conductivity deviation NEC(t) monitored online during the operation of multiple historical process segments, and extracting its final value NEC). The system calculates the end and trend characteristics (such as the area under the curve AUC_NEC and the decay time constant τ); simultaneously, it performs offline purity analysis on the output of the corresponding fragment to obtain the calibrated purity value Purity_offline; and establishes a mapping relationship through regression analysis: ρs = a·NEC_end + b·AUC_NEC + c·τ + d (where a, b, c, and d are model coefficients). In subsequent routine operation, the system calculates the NEC_end and trend characteristics of the current fragment in real time, and substitutes them into the model to output the estimated purity evaluation index ρs of the fragment online, which is used for knowledge base construction and process optimization; ωs is the membrane bag loss rate, ωs = (TMP_s_end - TMP_s_start) / TMP_s_start × 100%; ts is the fragment consumption time. TMP_s_start is the segment start transmembrane pressure, which refers to the initial transmembrane pressure value measured and recorded immediately after the system has been running stably at the beginning of the process segment; TMP_s_end is the segment end transmembrane pressure, which refers to the instantaneous transmembrane pressure value recorded by the system when the process segment ends and meets the termination conditions.

[0028] The overall fitness score F = g·ε_total + h·ρ_total - j·ω_avg - k·t_total; where ε_total and ρ_total are the global efficiency and purity, ω_avg is the average membrane loss, t_total is the total time, and g, h, j, and k are the weighting coefficients configured by the user according to their priorities.

[0029] The system has an optimization engine. When faced with a known scenario (such as "extracting exosomes from 200 mL of serum"): if records already exist: the system will search all records for that scenario and automatically select the record with the highest F-value. As a recommended formula.

[0030] If optimization is needed: the user or system can select a baseline recipe ( The optimization process will be initiated. The system will automatically fine-tune the parameters within a safe parameter space based on algorithms such as gradient descent or Bayesian optimization. Key parameters in the process (such as pump speed target and concentration ratio threshold η target for each step) are determined, and a "design-execution-evaluation" loop is initiated. After each run, the new PFV and F value are recorded. This iterative process continues until a locally optimal formulation is found where the F value no longer increases significantly. And update the knowledge base.

[0031] Example 2: A dynamic matrix control system for exosome extraction, comprising: The initial formula loading module is used to load the basic state matrix according to the process formula selected by the user.

[0032] The dynamic matrix generation module is based on the loaded... Create the real-time dynamic state matrix for this run. .

[0033] The primary real-time control module is used to make real-time, adaptive primary adjustments to M(t) based on the dynamic correction factor ΔM.

[0034] The execution parameter fine-tuning module aims to maintain HI(t) within a preset optimal range by fine-tuning the execution element parameters in M(t).

[0035] The data archiving module is used to perform performance analysis on each process step of this operation, generate process segment performance vectors (PFVs), and store complete process instance records in the process knowledge base.

[0036] The formula iteration optimization module is used to evaluate the overall fit score F of this run. If this run is an optimization experiment for a known scenario, the system will compare its F value with the historical records and may automatically start a new round of parameter fine-tuning and running tests. Through iterative loops, it will find the local optimal formula M0optimal for this scenario and update the knowledge base.

[0037] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0038] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0039] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0040] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0041] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0042] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0043] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic matrix control method for exosome extraction, characterized by, Includes the following steps: Step 1: At initialization, the system loads the base state matrix according to the process recipe selected by the user Based on the base state matrix The real-time dynamic state matrix is dynamically generated And the dynamic correction factor ΔM is introduced, and the dynamic matrix is updated in real time; Step 2: Based on the real-time updated dynamic matrix, update the real-time process health index HI(t) in real time, with the goal of keeping HI(t) in the preset optimal range, and fine-tune the parameters of the actuator in M(t); HI(t) = α*NTMP(t) + β*Nη(t) + γ*NEC(t), where: NTMP(t) = TMP(t) / TMP optimal(t) is the normalized transmembrane pressure; TMP(t) = Pin(t) - Pout(t) is the real-time calculated value, Pin(t) is the inlet pressure, Pout(t) is the outlet pressure; TMP optimal(t) is the dynamic optimal pressure threshold; Nη(t) = η(t) / η_target is the normalized concentration ratio; NEC(t) = |EC(t) - EC baseline| / EC baseline is the normalized conductivity deviation, used to indirectly monitor changes in fluid composition; EC(t) is the real-time conductivity; EC baseline is the baseline conductivity. Step three: after each process run, the system automatically evaluates the performance of each process step, defines a process fragment performance vector PFVs to quantify the overall performance of the fragment, all PFVs of the fragments together with their corresponding initial dynamic matrix , sample feature vector, final product quality index and key M(t) evolution trajectory feature, together constitute a process instance record, stored in the process knowledge base, for each record in the process knowledge base, a comprehensive fitness score F is calculated, and the recipe is iteratively self-optimized based on the comprehensive fitness score.

2. A dynamic matrix control method for exosome extraction according to claim 1, characterized in that, Real-time dynamic state matrix ; wherein, is a state vector set of all peristaltic pumps at time t, assuming there are N peristaltic pumps in the system, is formed by concatenating the individual state vectors of each pump (i = 1, 2,..., N) in order, denoted as ; is the individual state vector of the i-th peristaltic pump; is the state vector set of all electromagnetic pinch valves at time t, and there are M electromagnetic pinch valves in the system. is the state vector set of all electromagnetic pinch valves at time t, and there are M electromagnetic pinch valves in the system. is the state vector set of all electromagnetic pinch valves at time t, and there are M electromagnetic pinch valves in the system. is the state vector set of all electromagnetic pinch valves at time t, and there are M electromagnetic pinch valves in the system.

3. A dynamic matrix control method for exosome extraction according to claim 2, wherein, wherein, represents the start-stop state of the peristaltic pump, is the real-time calculated rotational speed of the peristaltic pump, is the inherent flow coefficient of the peristaltic pump.

4. The dynamic matrix control method for exosome extraction of claim 2, wherein, Vj(t)=[State j,Priority j], where State j∈{OPEN,CLOSE} represents the on / off state of the j-th solenoid pinch valve, and Priority j is the response priority of the j-th solenoid pinch valve during dynamic adjustment.

5. The dynamic matrix control method for exosome extraction of claim 2, wherein, Let S(t) be the set of all sensor readings at time t, including the inlet pressure Pin(t), the outlet pressure Pout(t), the feedstock weight Wl(t), the filtrate weight W3(t), the conductivity EC(t), and the liquid levels L(t).

6. The dynamic matrix control method for exosome extraction of claim 1, wherein, The dynamic correction factor ΔM = K pump·(L target-L current(t))+K valve·Hysteresis(ΔL(t) / Δt); where L current(t) is the real-time liquid level of the critical container, L target is the desired stable liquid level or threshold; K pump and Kvalve are the correction coefficient matrices for pump speed and valve opening / closing, respectively; Hysteresis() is a function containing hysteresis.

7. The dynamic matrix control method for exosome extraction of claim 1, wherein, TMPoptimal(t) = TMP base * [1 + κ * (η(t) / η target)], where TMP base is the membrane base pressure value, η(t) is the real-time concentration ratio, κ is the process progress influence coefficient, and η target is the target concentration ratio.

8. The dynamic matrix control method for exosome extraction of claim 1, wherein, PFVs=[εs,ρs,ωs,ts]; where εs is the fragment extraction efficiency, εs=(ΔW_product·C_product) / (ΔW_processing·C_initial)×100%; ρs is the purity evaluation index, ωs=(TMP_s_end-TMP_s_start) / TMP_s_start×100%; ts is the fragment extraction time.

9. A dynamic matrix control system for exosome extraction, applied to the dynamic matrix control method for exosome extraction according to any one of claims 1-8, characterized in that, include: The initial formula loading module is used to load the basic state matrix according to the process formula selected by the user; a dynamic matrix generation module that creates a real-time dynamic state matrix for the current run based on the loaded ;​ The primary real-time control module is used to make real-time, adaptive primary adjustments to M(t) based on the dynamic correction factor ΔM. The execution parameter fine-tuning module aims to maintain HI(t) within a preset optimal range by fine-tuning the execution element parameters in M(t). A data encapsulation module is used to analyze the performance of each process step in the current run, generate process fragment performance vectors PFVs, and store the complete process instance record into the process knowledge base; A recipe iteration optimization module is used to evaluate the comprehensive fitness score F of the current run. If the current run is an optimization experiment for a known scenario, the system compares its F value with the historical record and may automatically start a new round of parameter fine-tuning and run test, find the local optimal recipe M0optimal for the scenario through an iterative loop, and update the knowledge base.

Images