Micro-fluidic chip control mode distribution method considering valve switching balance

Abstract

The invention discloses a micro-fluidic chip control mode distribution method considering valve switching balance. The method comprises the following steps: acquiring a flow valve switching matrix and a multiplexing combination matrix at different moments as input; according to the flow valve switching matrix and the multiplexing combination matrix, constructing a particle coding mode for control mode distribution, and meanwhile, constraining the number of control valves and the maximum number of valve switching times; initializing the position and speed of the particle swarm, and calculating the initial fitness of each particle based on the number of constraint control valves and the maximum number of valve switching times; and adopting an algorithm based on discrete particle swarm optimization, dynamically grouping particle swarms through a role-based learning strategy, and generating a control mode distribution scheme and the like. According to the method, a control mode distribution scheme with the minimum number of control valves and the minimum switching frequency of the maximum control valve can be obtained.

Description

Technical Field

[0001] This invention relates to the field of microfluidic chip control mode allocation technology, and in particular to a microfluidic chip control mode allocation method that takes into account valve switching balance. Background Technology

[0002] CFMBs consist of two types of layers: a control layer and a flow layer. The control layer contains control channels for pressure conduction, while the flow layer contains flow channels for transporting reagents and fluid samples. [1] There is a valve at the intersection of the control channel and the flow channel, which acts as a switch. Figure 13 A 3D schematic diagram of the valve and its associated channels is presented. The valve is a flexible membrane made of polydimethylsiloxane (PDMS) material. [2] To control the opening and closing of the valve, external pressure is first transmitted to the control channel, and then from the control channel to the valve, thus controlling its opening and closing. When the valve is subjected to high pressure, it will squeeze downwards to prevent fluid movement in the flow channel; when the pressure is low, the fluid in the flow channel can pass freely through the valve. Figure 13 A cross-sectional schematic diagram is presented, showing the control valve closing and impeding fluid movement.

[0003] With the rapid development of biochip manufacturing technology, tens of thousands of valves can now be integrated into a chip the size of a coin. [3] However, due to limitations in chip area and component size, allocating a separate control port for each valve is impractical. This would not only significantly increase the manufacturing complexity and cost of the chip but also increase reliance on external pneumatic controllers. [4] Meanwhile, with the increasing integration of microfluidic biochips and the growing complexity of experiments, the number of valves is constantly increasing, making traditional independent port control methods insufficient. Therefore, using multiplexer (MUX) technology to control valves has become a highly efficient solution. [5-6] .

[0004] The control logic architecture based on multiplexers was initially proposed by Thorsen et al. [7] This architecture uses binary encoding, through 2 N Each control channel is independently addressed and driven. Each flow channel has a state controlled by a combination array of binary valves. Zhu et al. further proposed a multi-channel control method to simultaneously address multiple control channels, and then utilize idle coding capacity to introduce backup channels, thereby improving the fault tolerance of the control channels. [8]Liang et al. proposed a MUX coding strategy based on combinatorial coding, utilizing Sperner's theorem to achieve the theoretically maximum coding capacity. This allows for the control of more independent control channels using fewer flow channels and valves, further improving resource utilization and the reliability of control logic. [9] Wang et al. proposed a Hamming distance-based microvalve switching sequence optimization method to improve the reliability of the control layer MUX in microfluidic biochips.

[10] Building upon this foundation, Wang et al. proposed an XOR-based pressure refresh algorithm to address the pressure degradation problem in MUX.

[11] Huang et al. proposed a reinforcement learning-based method for synthesizing control logic in a fully programmable valve array biochip, further optimizing the control logic architecture to improve chip efficiency and reliability.

[12] Furthermore, valves and channels on the MUX may encounter manufacturing defects, including valve defects, blockages, and leaks. [13-15] To address this, Liu et al. proposed an automated fault detection method for MUX, which can quickly and accurately detect various faults occurring in MUX.

[16] . References [1]X. Huang, Y. Pan, GL Zhang, B. Li, W. Guo, T.-Y. Ho, and U.Schlicht-mann "Pathdriver+: Enhanced path-driven architecture design for flow-based microfluidic biochips," IEEE Transactions on Computer-AidedDesign ofIntegrated Circuits and Systems, vol. 41, no. 7, pp. 2185–2198, 2021. [2]Unger MA, Chou HP, Thorsen T, et al. Monolithic microfabricatedvalves and pumps by multilayersoftlithography[J]. Science, 2000, 288(5463): 113−116 [3]AraciIE,QuakeSR. Microfluidicverylargescaleintegration (mVLSI)with integrated micromechanical valves[J]. Lab on a Chip, 2012, 12(16): 2803−2806 [4]S. Liang, M. Li, T.-M. Tseng, U. Schlichtmann, and T.-Y. Ho,“Comux:Combinatorial-coding-based high-performance microfluidic controlmultiplexer design,” in The 41st IEEE / ACM International Conference onComputer-Aided Design (ICCAD), 2022. [5]T.-M. Tseng, B. Li, M. Li, T.-Y. Ho, and U. Schlichtmann,“Reliability-aware synthesis with dynamic device mapping and fluid routingfor flow-based microfluidic biochips,” IEEE Transactions on Computer-AidedDesign of Integrated Circuits and Systems, vol. 35, no. 12, pp. 1981–1994,2016. [6]M. Shayan, S. Bhattacharjee, Y.-A. Song, K. Chakrabarty, and R.Karri,“Toward secure microfluidic fully programmable valve array biochips,”IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 27,no. 12, pp. 2755–2766, 2019. [7]ThorsenT,MaerklSJ,QuakeSR. Microfluidiclarge-scale integration[J].Science, 2002, 298(5593): 580−584 [8]Zhu Ying, Huang Xing, Li Bing, et al. Multicontrol: Advancedcontrol-logic syn-thesis for flow-basedmicrofluidicbiochips[J]. IEEETransactionsonComputer-AidedDesignofIntegratedCircuitsand Systems, 2019, 39(10): 2489−2502 [9]Liang Siyuan, Li Mengchu, Tseng T M, et al. CoMUX: Combinatorial-coding-basedhigh-performancemicrofluidiccontrolmultiplexer design[C / OL] / / Procofthe41stACMIntConfonComputer-Aided Design. New York: ACM, 2022 [2024-06-19].

[10] Wang Qin, Zuo Shiliang, Yao Hailong, et al. Hamming-distance-based valveswitchingoptimizationforcontrol-layermultiplexinginflow-basedmicrofluidicbiochips[C] / / Procofthe22ndAsiaandSouth PacificDesignAutomationConf.Piscataway,NJ:IEEE,2017: 524−529

[11] Wang Qin, Xu Yue, Zuo Shiliang, et al. Pressure-aware controllayer optimizationforflow-basedmicrofluidicbiochips[J]. IEEE TransactionsonBiomedicalCircuitsandSystems, 2017, 11(6): 1488−1499

[12] HuangXing,CaiHuayang,GuoWenzhong,etal. Control-logic synthesisoffullyprogrammablevalvearrayusingreinforcement learning[J].IEEETransactionsonComputer-AidedDesignof Integrated Circuits and Systems,2024, 43(1): 277−290

[13] K. Hu, F. Yu, T.-Y. Ho, and K. Chakrabarty, “Testing of flow-based microfluidic biochips: Fault modeling, test generation, andexperimental demonstration,” IEEE Transactions on Computer-Aided Design ofIntegrated Circuits and Systems, vol. 33, no. 10, pp. 1463–1475, 2014.

[14] K. Hu, T.-Y. Ho, and K. Chakrabarty, “Test generation and design-for-testability for flow-based mvlsi microfluidic biochips,” in Proceedingsof IEEE VLSI Test Symposium, pp. 1–6, IEEE, 2014.

[15] G. Liu, Y. Zhu, W. Guo, and X. Huang, “Fault-tolerance-oriented physical design for fully programmable valve array biochips,” in Proceedings of ACM / IEEE Design Automation Conference, pp. 1–6, IEEE, 2023.

[16] Liu G, Zeng Y, Zhu Y, et al. Towards automated testing of multiplexers in fully programmable valve array biochips[C] / / 2024 29th Asiaand South Pacific Design Automation Conference (ASP-DAC). IEEE, 2024: 570-575. Summary of the Invention In view of this, the purpose of this invention is to propose a microfluidic chip control mode allocation method that takes into account the balance of valve switching, so as to obtain a control mode allocation scheme with the minimum number of control valves and the minimum number of maximum control valve switching times.

[0005] According to one aspect of the present invention, a microfluidic chip control mode allocation method considering valve switching balance is provided, the method comprising: The process involves obtaining the flow valve switching matrix and multiplexing combination matrix at different times as input; constructing a particle encoding method for control mode allocation based on the flow valve switching matrix and multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times; initializing the particle swarm position and velocity, and calculating the initial fitness of each particle; employing an algorithm based on discrete particle swarm optimization, dynamically grouping the particle swarm through a role-based learning strategy to generate a control mode allocation scheme; the role-based learning strategy includes: the globally optimal particle integrating the experience of the optimal particles in each subgroup for updating; the optimal particle in a subgroup learning from the experience of the optimal particles in the upper subgroup; and ordinary particles learning from the experience of the current optimal particle in their subgroup; updating particle positions through crossover and mutation operations, and outputting a globally optimal control mode allocation scheme that satisfies the constraints.

[0006] In the aforementioned technical solution, the microfluidic chip control mode allocation method has significant advantages. First, it constructs a particle encoding method that simultaneously constrains the number of control valves and the maximum number of valve switching operations by acquiring the flow valve switching matrix and multiplexing combination matrix at different times as input, thus optimizing the constraints on control mode allocation from the source. Second, during particle swarm optimization, a role-based learning strategy is used to dynamically group the particle swarm. The globally optimal particle integrates the experience of the optimal particles in each subgroup for updating, the optimal particle in a subgroup learns from the experience of the optimal particle in the upper subgroup, and ordinary particles learn from the experience of the current optimal particle in their subgroup. This multi-level, multi-directional learning mechanism effectively improves the global search capability and solution accuracy of the particle swarm, avoiding getting trapped in local optima. Finally, crossover and mutation operations are used to update particle positions, further enhancing the algorithm's global search capability. Ultimately, the algorithm outputs a globally optimal control mode allocation scheme that satisfies the constraints, effectively improving the reliability and control efficiency of the microfluidic chip. In summary, this method not only optimizes the constraints of control mode allocation, but also improves the algorithm's global search capability and solution accuracy through role-based learning strategies and crossover mutation operations. Ultimately, it outputs a globally optimal control mode allocation scheme that satisfies the constraints, significantly improving the reliability and control efficiency of microfluidic chips.

[0007] In some embodiments, a particle coding method for control mode allocation is constructed based on the flow valve switching matrix and the multiplexing combination matrix, while simultaneously constraining the number of control valves and the maximum number of valve switching operations, including: Define the particle encoding format, particle The location code is:

[0008] in, Set the value to 0 , indicating that the number is The control mode is assigned to the first Multiplexer combinations; Indicates the number of control channels. L Indicates the maximum number of multiplexing combinations; Construction constraints: .

[0009] In the above technical solution, the particle encoding method assigned by this control mode has a clear structure and constraints. First, by defining the particle encoding format, the particle position is encoded as... Each of them Set the value to , and 0 It is clearly stated that the number is The control mode is assigned to the first Multiplexing combinations. This coding method clearly defines the mapping relationship between control modes and multiplexing combinations, providing a foundation for subsequent optimization processes. Secondly, constraints are constructed, that is, for any... and ,when At that time, it was required This approach ensures that each control mode is uniquely assigned within the multiplexing combination, avoiding duplicate allocations and effectively constraining the number of control valves and the maximum number of valve switching operations, thereby improving the rationality and reliability of control mode allocation. By defining a clear particle coding format and strict constraints, a scientific and reasonable coding mechanism is provided for the allocation of control modes in microfluidic chips. This ensures both the uniqueness of control mode allocation and effectively constrains the number of control valves and the maximum number of valve switching operations, laying a solid foundation for the implementation of subsequent optimization algorithms and contributing to improved control efficiency and reliability of microfluidic chips.

[0010] In some embodiments, initializing the particle swarm position and velocity, and calculating the initial fitness of each particle, includes: randomly generating... Given each particle position, initialize the particle velocity vector and calculate the initial fitness of each particle:

[0011] in, and This represents two weighting factors. Indicates the number of control valves. This indicates the maximum number of times the control valve can be switched.

[0012] In the above technical solution, this method lays a solid foundation for the particle swarm optimization algorithm in the initialization phase. First, by randomly generating... n The algorithm first identifies the position of each particle and initializes its velocity vector, providing diverse initial solutions for subsequent optimization and enhancing the algorithm's exploratory capabilities. Secondly, it calculates the initial fitness of each particle and employs a fitness function. This method explicitly quantifies the optimization objective as a fitness value. The fitness function design not only considers minimizing the number of control valves but also balances the number of valve switching operations, enabling the algorithm to balance these two key factors simultaneously during optimization. Through reasonable particle initialization and fitness function design, this method provides diverse initial solutions and clear optimization directions for the particle swarm optimization algorithm, allowing it to efficiently explore the global optimum in subsequent iterations while maintaining a balance between the number of control valves and the number of switching operations. This significantly improves the optimization effect and reliability of microfluidic chip control mode allocation.

[0013] In some embodiments, an algorithm based on discrete particle swarm optimization is used to dynamically group the particle swarm through a role-based learning strategy, including: obtaining the historical maximum and minimum values ​​of particle fitness values; determining the subgroup boundary based on the historical maximum and minimum values; assigning particles to the corresponding subgroups according to the subgroup boundary; assigning different roles to particles, including globally optimal particles, subgroup optimal particles, and ordinary particles, thereby guiding different levels of learning strategies; dynamically adjusting the subgroup size to maintain population diversity; and verifying whether the subgroup division satisfies the upper and lower bound constraints of fitness.

[0014] In the above technical solution, this method achieves efficient dynamic grouping and optimization of particle swarms through an algorithm based on discrete particle swarm optimization and a role-based learning strategy. First, by obtaining the historical maximum and minimum values ​​of particle fitness, the sub-swarm boundaries are determined, and particles are assigned to corresponding sub-swarms accordingly. This dynamic grouping method based on fitness levels effectively maintains population diversity and provides a foundation for different levels of learning strategies. Second, different roles are assigned to particles, including globally optimal particles, sub-swarm optimal particles, and ordinary particles, guiding different levels of learning strategies: globally optimal particles integrate the experience of optimal particles from each sub-swarm for updates; sub-swarm optimal particles learn from the experience of optimal particles in upper-level sub-swarms; and ordinary particles learn from the experience of the current sub-swarm's optimal particle within their own sub-swarm. This hierarchical learning mechanism not only enhances information interaction between particles but also improves the global search capability and solution accuracy of the swarm. Finally, the sub-swarm size is dynamically adjusted to maintain population diversity, and the sub-swarm partitioning is verified to satisfy the upper and lower bound constraints of fitness, further ensuring the stability and effectiveness of the algorithm. By employing dynamic grouping and role-based learning strategies, the global search capability and solution accuracy of the particle swarm optimization algorithm are significantly improved, while maintaining population diversity and effectively avoiding local optima problems, thus providing an efficient and reliable optimization scheme for the allocation of control modes in microfluidic chips.

[0015] In some embodiments, particle positions are updated through crossover and mutation operations, and the particle updates are as follows:

[0016] in, Indicates inertia weight,

[0017] ; and It is the acceleration factor. For inertial components, Perceived through individual experience. For group experience perception. In the above technical solution, this method significantly improves the performance of the particle swarm optimization algorithm by updating particle positions through crossover and mutation operations. First, the particle update formula integrates three parts: inertial component, individual experience perception, and swarm experience perception. The inertial weight is dynamically adjusted according to the particle's fitness, ensuring a balance between global and local searches. Second, through crossover (individual experience perception) and mutation (inertial component), particles can retain valuable information and introduce new solution spaces, preventing the algorithm from getting trapped in local optima. Finally, swarm experience perception further guides particles to evolve towards the global optimum, enhancing the algorithm's global search capability. Overall, this method of updating particle positions through crossover and mutation operations not only enhances particle diversity but also improves the algorithm's global search capability and solution accuracy, providing an efficient and reliable optimization method for microfluidic chip control mode allocation.

[0018] In some embodiments, the inertial component update is achieved through a mutation operation, and the update formula is as follows:

[0019] in, It is a mutated particle. It is a random number between [0,1). It is a set of control mode numbers that have never been assigned; for particles in the t-th iteration , Through exchange and The control mode represented is used to update the particles; Then use Replace any of the control modes The control mode it represents; This allows you to choose any control mode from all available control modes to replace the previous one. If the control mode represented by the selected control mode is duplicated after replacement, the operation will be cancelled.

[0020] In the above technical solution, this method updates the inertial component through mutation operations, effectively enhancing the exploration capability of the particle swarm optimization algorithm. Specifically, the mutation operation employs three different strategies based on the relationship between random numbers and inertial weights: performing a swap operation to update particles by exchanging two control modes; performing a rep operation to replace the current control mode with any control mode; and performing a mut operation in other cases to randomly select a control mode from all control modes for replacement. If a duplicate control mode occurs after replacement, the operation is canceled to ensure the feasibility of the solution. This mutation operation design not only introduces randomness and increases particle diversity but also avoids local optima problems through different replacement strategies, while ensuring the feasibility constraint of the solution. The inertial component update achieved through mutation operations significantly improves the global search capability and solution diversity of the particle swarm optimization algorithm, providing a more efficient and reliable optimization mechanism for control mode allocation in microfluidic chips.

[0021] In some embodiments, the perception of individual experience is achieved through a crossover operation, and the update formula is expressed as follows:

[0022] in, These are the particles after the crossover. It is a random number in [0,1). Representing the t The best historical position of an individual particle in the next iteration. This indicates that two particles are performing an interleaved operation. The individual learning factor is defined as follows:

[0023] in, The individual learning factor is the optimal particle in the swarm. The individual learning factor is the optimal particle of the subpopulation. For ordinary particles, the individual learning factor. It represents the first t The optimal particle of the population in the next iteration. Representing the t In the nth iteration j The optimal particle of a subpopulation Representing the t During the nth iteration i The subpopulation to which each particle belongs.

[0024] In the aforementioned technical solution, this method enables each particle to select a different learning factor based on its role, thereby performing targeted cross-operations during the individual experience perception stage. When the random number is less than the individual learning factor, the particle cross-operates with its own historical best position, thus introducing a new solution space while preserving valuable information. Otherwise, the particle position remains unchanged. This mechanism not only enhances the local search capability of the particles but also introduces diversity through cross-operations, preventing the algorithm from getting trapped in local optima. Through this role-based individual experience perception mechanism, this method can better balance global and local search capabilities during the optimization process, significantly improving the performance of the particle swarm optimization algorithm and providing a more efficient and reliable optimization method for microfluidic chip control mode allocation.

[0025] In some embodiments, the group experience perception is achieved through role differentiation and cross-referencing, with the update formulas for different roles as follows:

[0026] The calculation formula is:

[0027] The calculation formula is:

[0028] and, Satisfy the following formula:

[0029] in, The group learning factor is the optimal particle in the group. The group learning factor is the optimal particle of the subgroup. For the population learning factor of ordinary particles, Represents the particles after the intersection. It is a random number in [0,1). This indicates an iterative crossover operation, used to guide cross-learning between the globally optimal particle and the optimal particles of each subgroup. The number of subpopulations; A random number in [0,1). Indicates the first j The proportion of the fitness of the best particle in a subpopulation to the total fitness of the best particles in all non-global optimal subpopulations.

[0030] In the above technical solution, this method achieves group experience perception through role-differentiated crossover, significantly improving the global search capability of the particle swarm optimization algorithm. Different crossover strategies are adopted for particles with different roles, enhancing the collaborative optimization capability among particles. Simultaneously, by dynamically adjusting the crossover operation, it ensures that the optimal particle in each subgroup can effectively learn through crossover based on its fitness ratio, further improving algorithm performance. First, this method designs differentiated crossover strategies for particles with different roles. The globally optimal particle is updated by integrating the experience of the optimal particles from all subgroups; the optimal particle in a subgroup learns from the experience of neighboring, better subgroups; ordinary particles learn from the optimal particle in the current subgroup. Second, by dynamically adjusting the crossover operation, it ensures that the optimal particle in each subgroup can learn through crossover based on its fitness ratio, further improving the global search capability. Specifically, Re ( j The formula dynamically adjusts according to the fitness ratio of particles to ensure the effectiveness of the crossover operation. Finally, in this way, the algorithm not only retains the excellent information of particles but also introduces new solution spaces, increasing particle diversity and avoiding the algorithm getting trapped in local optima. The group experience perception mechanism achieved through role-differentiated crossover significantly improves the global search capability and solution diversity of the particle swarm optimization algorithm. This method adopts differentiated crossover strategies based on particles with different roles and ensures that the optimal particles in each subgroup can effectively learn through crossover according to their fitness ratio by dynamically adjusting the crossover operation. This not only enhances the collaborative optimization capability among particles but also effectively avoids local optima problems, providing a more efficient and reliable optimization method for the allocation of control modes in microfluidic chips.

[0031] According to another aspect of the present invention, a microfluidic chip control mode allocation system considering valve switching balance is provided. Based on the above method, it includes: an acquisition module for acquiring flow valve switching matrices and multiplexing combination matrices at different times; an encoding module for constructing a particle encoding method for control mode allocation based on the flow valve switching matrices and multiplexing combination matrices, while constraining the number of control valves and the maximum number of valve switching times; initializing the particle swarm position and velocity, and calculating the initial fitness of each particle; a strategy module for dynamically grouping the particle swarm using a discrete particle swarm optimization-based algorithm and a role-based learning strategy to generate a control mode allocation scheme; the role-based learning strategy includes: the globally optimal particle integrating the experience of the optimal particles in each subgroup for updating; the optimal particle in a subgroup learning from the experience of the optimal particles in the upper subgroup; and ordinary particles learning the experience of the current optimal particle in the subgroup; and an update module for updating the particle positions through crossover and mutation operations, and outputting a globally optimal control mode allocation scheme that satisfies the constraints.

[0032] In order to better utilize the above method, this application proposes a microfluidic chip control mode allocation system that considers valve switching balance. Each module corresponds to a step of the above method, and its specific principle has been described above and will not be repeated here. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating an embodiment of the method of the present invention; Figure 2 This is a flowchart of the DPSO-RL algorithm according to an embodiment of the method of the present invention; Figure 3 This is a schematic diagram of the DPSO-RL learning network according to an embodiment of the method of the present invention; Figure 4 This is an example diagram of a multiplexing matrix according to an embodiment of the method of the present invention; Figure 5 This is the valve logic simplification and logic forest construction process of Scheme A in an embodiment of the method of the present invention; Figure 6 This is the result of constructing a valve logic forest for allocation schemes B and C in one embodiment of the method of the present invention; Figure 7 This is the logical forest wiring diagram corresponding to allocation scheme A in one embodiment of the method of the present invention; Figure 8 This is the logical forest wiring diagram corresponding to allocation scheme C in one embodiment of the method of the present invention; Figure 9 This is a schematic diagram of the variation operation of an embodiment of the method of the present invention; Figure 10 This is a schematic diagram of the cross operation of an embodiment of the method of the present invention; Figure 11 This is a schematic diagram of the test results of the DPSO-RL algorithm according to an embodiment of the method of the present invention; Figure 12 This is a schematic diagram of the structure of an embodiment of the system of the present invention; Figure 13 The diagram shows the structure of existing CFMBs; the left diagram is a 3D schematic of the valve and its associated passage, and the right diagram is a cross-sectional schematic of the valve closing and impeding fluid movement. Detailed Implementation

[0035] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] This invention provides a microfluidic chip control mode allocation method that considers valve switching balance. Current control mode allocation algorithms primarily focus on minimizing the number of control valves in the MUX (Micro-UX) and do not adequately address the switching balance during the control process. Some control valves may be frequently switched during biochemical assays, while others are switched relatively infrequently, leading to uneven distribution of switching load. Frequent valve switching can cause elastic degradation, potentially resulting in valve failure and affecting the normal operation of biochemical assays, or even causing structural damage to the chip in severe cases. Furthermore, the imbalance in valve switching frequency can lead to localized overload, creating a bottleneck in system reliability. Therefore, while ensuring the normal execution of biochemical assays, it is necessary not only to optimize the control mode to reduce the number of valves but also to balance the valve switching frequency to maximize the reliability of the biochip.

[0037] Example 1 Please see Figure 1 A microfluidic chip control mode allocation method considering valve switching balance, the algorithm flow is as follows: Figure 2 As shown, firstly, the position and velocity of the particle swarm are initialized using the flow valve switching matrix and multiplexing combination matrix at different times as input, and the algorithm parameters are set. Then, the fitness value of each particle is calculated, and the particle swarm is dynamically grouped according to the fitness level, with the size of each subgroup adjusted accordingly. Next, different roles are assigned to the particles, including globally optimal particles, subgroup optimal particles, and ordinary particles, to guide different levels of learning strategies: globally optimal particles integrate the experience of the best particles in each subgroup for updates; subgroup optimal particles learn from the experience of the best particles in the upper subgroups; and ordinary particles learn from the experience of the best particle in the current subgroup within their own subgroup. In each iteration, the overall learning network of the particles is as follows: Figure 3 As shown, the particle positions are updated according to the above learning strategy, continuously exploring the optimal control mode configuration until the maximum number of iterations is met, at which point a globally optimal control mode allocation scheme is output. Specifically, the method includes: S1. Obtain the flow valve switching matrix and multiplexing combination matrix at different times as input; In this embodiment, the following input matrix is ​​obtained: flow valve switching matrix ( T × VMatrix: Record T At this time step V The state of each valve (0 = open, 1 = closed). Multiplexing combination matrix ( C × V Matrix): Definition C One control channel and V The connection relationship of each valve (0 / 1 indicates whether it is connected).

[0038] S2. Construct a particle coding scheme for control mode allocation based on the flow valve switching matrix and the multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times; initialize the particle swarm position and velocity, and calculate the initial fitness of each particle; in this embodiment, S2-1, constructing a particle coding scheme for control mode allocation based on the flow valve switching matrix and the multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times, includes: S2-11, Define the particle encoding format, particle The location code is:

[0039] in, Set the value to 0 , indicating that the number is The control mode is assigned to the first Multiplexer combinations; Indicates the number of control channels. L Indicates the maximum number of multiplexing combinations; S2-12, Construction Constraints:

[0040] For example, in the control mode assignment problem, suppose a multiplexer contains N If the combination of control channels used for synchronous switching has been determined, i.e., multiplexing combination, then initialization is required. Each control port. Furthermore, these ports can generate... This paper presents the given multiplexing combinations in a matrix format to more intuitively describe their control relationships. (The text then repeats the first instance, which seems to be an error and can be omitted.) Figure 4 Taking a multiplexing matrix as an example, this article uses a portion of a multiplexing matrix from a biochemical experimental application. In this matrix, each row represents a multiplexing combination, each column represents a control channel, and "0" and "1" represent the closed and open states of the valves, respectively. Five flow valves ( - The system connects to the core input via five control channels and employs four multiplexing combinations to achieve multi-channel control. Therefore, three pairs of complementary control ports need to be initialized, resulting in a total of six control ports. The control modes generated by these control ports and their numbering are as follows: 0: 1: 2: 3:

[0041] 4: 5: 6: 7:

[0042] Figure 4 The particle code corresponding to the control mode allocation scheme C in the diagram is: cp(0)=0, cp(1)=4, cp(2)=5, cp(3)=7. Furthermore, the control mode corresponding to each combination is listed next to each row of the matrix, with three control mode allocation schemes: A, B, and C, to visually represent the control logic. For example, in allocation scheme A, the control mode... It has been assigned to the multiplexer combination 01100. If the control mode... If activated, then with the flow valve and The status of the connected control channels will change simultaneously, and the flow valves... and They will be opened simultaneously under the pressure drive of the control channel.

[0043] To evaluate the valve quantity allocation schemes for each control mode, this paper uses literature...

[14] The allocation scheme is transformed into a logic tree, and then the resource usage of each allocation scheme is obtained through logical simplification and logic forest construction. For example, Figure 5This illustrates the process corresponding to allocation scheme A, where the root node and leaf nodes represent the core input and flow valves, respectively, the internal nodes represent control valves, and each edge represents a control path connecting the control valves. The final logical forest constructed by allocation scheme A contains 18 control valves and 25 control paths. Conversely, as... Figure 6 As shown, the logic forests of allocation schemes B and C contain only 11 control valves and 17 control paths. The wiring configurations of the logic forests constructed by allocation schemes A and C are as follows: Figure 7 and Figure 8 As shown, there are 6 control ports. , , , , and Connected to a pressure source, this controls the opening and closing of five flow valves. Figure 7 and Figure 8 As can be seen, assigning different control modes to different multiplexing combinations leads to significant differences in the number of control valves required in the control logic. Reducing the number of control valves not only helps to shorten the control path length, but also effectively reduces the complexity of the control logic design, thereby reducing the chip manufacturing cost.

[0044] In control logic design, the pressure values ​​of the control ports are usually complementary; only one port can have high pressure at a time. For example... Figure 7 As shown, and As a pair of complementary control ports, allocation scheme A is a multiplexed combination of 01001, and the control mode is as follows: That is, state 100. The control mode assigned to combination 01000 is... That is, state 111. Assume the valve state needs to change from... Figure 7 (a) The "01001" at the time shown switches to as follows Figure 7 (b) At time “01000”, the control valve of the multiplexer needs to complete the state transition from 100 to 111. , , and Each corresponding control valve requires a switching operation, therefore the multiplexer needs to perform a total of 4 switching operations. Based on the above switching situation, when using... Figure 8 When using allocation scheme C as shown, the control mode for this scheme, which is a multiplexed combination 01001 allocation, is... At this point, you only need to switch The corresponding control valve state can achieve the same function. It can be seen that scheme C only requires the multiplexer to perform one switching operation. Compared to scheme A, scheme C reduces the number of control valve switching operations by three while achieving the same control function, significantly reducing resource usage and improving control efficiency. Furthermore, the above also demonstrates how to calculate the number of control valve switching operations between two adjacent moments. To more fully illustrate this process, the flow valve switching action matrix SY at different moments is used as an example, as shown in the following formula:

[0045] Each row in the matrix represents the state of the flow valve at a certain moment. The top-to-bottom arrangement of the matrix reflects the change of the flow valve state over time. By analyzing the state differences between adjacent rows in the matrix, the number of control valve switching events occurring between each moment can be counted, thereby quantifying the switching balance of the control mode allocation scheme. Table 1 shows the number of valve switching events corresponding to different allocation schemes, where TA represents the number of control valve switching events for allocation scheme A, TB represents the number of control valve switching events for allocation scheme B, and TC represents the number of control valve switching events for allocation scheme C.

[0046] Table 1 shows the number of control valve switching times for different allocation schemes after completing the SY switch.

[0047] As can be seen from the results of the logic forest construction above, allocation schemes B and C both use fewer resources than scheme A. Furthermore, as shown in Table 1, the maximum number of control valve switching times for both allocation schemes A and B is 7, but the minimum number of switching times is 3 and 0 respectively, indicating a significant difference in the number of switching times among the control valves. This is because existing control mode allocation algorithms only optimize the number of control valves and do not consider the balance between the number of switching times, resulting in some control valves bearing excessive switching overhead. In contrast, allocation scheme C not only consumes fewer resources, but also has a maximum of only 3 control valve switching times and a minimum of only 1. It reduces the number of control valve switching times while balancing the switching times among the various valves. Moreover, since each control valve is a potential point of failure, reducing the number of control valves also means reducing the number of potential failure points, thereby improving reliability while reducing costs. Simultaneously, balancing the number of valve switching times prevents some valves from accelerating elastic degradation due to overuse, further extending the valve's operational durability and enhancing the overall reliability of the chip. Based on the above analysis, the problem model for optimizing the CFMBs control mode allocation for valve switching balance is as follows: Input: Flow valve switching action matrix at different times, multiplexing matrix. Output: Control mode allocation scheme considering valve switching balance. Objective: Minimize the number of control valves and minimize the maximum number of valve switching operations.

[0048] In this embodiment, S2-2, initializing the particle swarm position and velocity, and calculating the initial fitness of each particle, includes: S2-21, randomly generating... Given each particle position, initialize the particle velocity vector and calculate the initial fitness of each particle:

[0049] in, and This represents two weighting factors. Indicates the number of control valves. This indicates the maximum number of times the control valve can be switched. Since the goal of this invention is to minimize the maximum number of times the control valve can be switched and the number of resources required in the multiplexer, particles with higher fitness represent control mode allocation schemes of higher quality.

[0050] S3. An algorithm based on discrete particle swarm optimization is adopted, and the particle swarm is dynamically grouped through a role-based learning strategy to generate a control mode allocation scheme. The role-based learning strategy includes: the global optimal particle integrates the experience of the optimal particles of each subgroup and updates it; the optimal particle of a subgroup learns the experience of the optimal particle of the upper subgroup; and ordinary particles learn the experience of the current optimal particle of the subgroup within the subgroup. In this embodiment, S3, an algorithm based on discrete particle swarm optimization is used to dynamically group the particle swarm through a role-based learning strategy, including: S31, obtaining the historical maximum and minimum values ​​of particle fitness, determining the subgroup boundary based on the historical maximum and minimum values, and assigning particles to the corresponding subgroups according to the subgroup boundary; S32, assigning different roles to particles, including globally optimal particles, subgroup optimal particles, and ordinary particles, thereby guiding different levels of learning strategies; S33, dynamically adjusting the subgroup size to maintain population diversity, and verifying whether the subgroup division satisfies the upper and lower bound constraints of fitness.

[0051] For example, during the particle swarm optimization update process, the swarm is divided into 6 subgroups based on the fitness values ​​of the particles, rather than a single overall population. This grouping method better maintains population diversity, and the size of each subgroup is dynamically adjusted according to the fitness level, effectively enhancing the ability to explore high-quality control mode allocation schemes and avoiding getting trapped in local optima. The formula for subgroup division is as follows:

[0052] in, and These represent the historical maximum and minimum values ​​of the fitness function, respectively. This represents the boundary of each subgroup. Indicates the first t In the nth iteration j Individual population. Indicates the first t In the nth iteration i The position of each particle. Represents particles The fitness value. This is used to determine the upper and lower bounds of the fitness of each subpopulation, so as to avoid the extreme imbalance of the size of each subpopulation affecting the performance of the algorithm.

[0053] S4. Update particle positions through crossover and mutation operations, and output the globally optimal control mode allocation scheme that satisfies the constraints. For example, the particle update formula is as follows:

[0054] in, It is the inertial weight, representing the probability of a particle undergoing a mutation operation. and It is the acceleration factor, representing the probability of particles performing crossover operations. For inertial components, Perceived through individual experience. For group experience perception. In the PSO algorithm, inertia weight is a crucial parameter that directly affects the adjustment ability of particles. To achieve a balance between global search and local exploitation, this invention employs an adaptive inertia weight, as shown in the following formula:

[0055] This formula maps all inertia coefficients to the interval [0.4, 0.8], which is a common range for PSO inertia weights. For particles with higher fitness, smaller inertia weights are assigned, which helps them perform more refined local searches near the current high-quality solution. Particles with lower fitness or located in the middle interval are given larger inertia weights, thus encouraging them to conduct broader global exploration.

[0056] In this embodiment, the inertial component of the particle is updated through a mutation operation. For example, the update formula is expressed as follows:

[0057] in, It is a mutated particle. It is a random number between [0,1). It is a set of control mode numbers that have never been assigned. For particles in the t-th iteration... , Through exchange and The control mode represented is used to update the particles; Then use Replace any of the control modes The control mode it represents; It means selecting any control mode from all control modes (including already assigned control modes) to replace it. If the control mode represented by the given information is repeated after replacement, which violates the feasibility constraint, the operation will be cancelled. Figure 9 Displaying particles Three ways to update the position through mutation during the iteration process. Figure 9 (a) Use the swap(0,3) operation to swap the first and fourth assigned control modes. Figure 9 (b) Using the rep(1) operation, in the set of control modes that have never been assigned. Take out the control mode numbered 6 and replace the control mode originally assigned to the second multiplexing combination. Figure 9(c) The mut(2) operation is used to attempt to replace the current control mode of the third multiplexing combination with the control mode numbered 4. However, after the replacement, the control mode numbers assigned to the second and third multiplexing combinations are the same, resulting in a duplicate. Therefore, the replacement operation is canceled.

[0058] In this embodiment, the individual experience perception of particles is achieved through cross-operation. For example, the update formula is expressed as follows:

[0059] in These are the particles after the crossover. It is a random number in [0,1). Representing the t The best historical position of an individual particle in the next iteration. This indicates that two particles are performing an interleaved operation. The individual learning factor is defined as follows:

[0060] in, The individual learning factor is the optimal particle in the swarm. The individual learning factor is the optimal particle of the subpopulation. For ordinary particles, the individual learning factor. Representing the t The optimal particle of the population in the next iteration. Representing the t In the nth iteration j The optimal particle of a subpopulation Representing the t During the nth iteration i The subpopulation to which each particle belongs. Figure 10 This demonstrates the cross-learning process of particles during individual experiential perception. At this moment, the particle's current position... Its individual historical best position Perform a crossover operation to generate new positions. For positions where the values ​​of the two encoded positions are the same, directly retain that value in the new particle; for positions with different values, start from... and A random value is selected from the corresponding values ​​and assigned to the new particle. If, after the crossover operation, a duplicate control mode number appears in the new particle, violating the feasibility constraint, the crossover result is canceled, and the crossover operation is re-executed until the no-duplicate constraint is met. This crossover operation helps particles retain some valuable information during the evolution process.

[0061] In this embodiment, group experience perception is achieved through role-differentiated cross-learning. For example, when particles perform group experience perception, the globally optimal particle learns from the optimal particles of each subgroup through cross-learning. The optimal particle of a subgroup (not the globally optimal) learns from the optimal particles of neighboring subgroups with higher-level skills to obtain a higher-quality search direction. Ordinary particles learn from the optimal particles of their respective subgroups. The update formulas for different roles are as follows:

[0062] in, The group learning factor is the optimal particle in the group. The group learning factor is the optimal particle of the subgroup. For the population learning factor of ordinary particles, Represents the particles after the intersection. It is a random number in [0,1). This represents an iterative crossover operation, used to guide cross-learning between the globally optimal particle and the optimal particles of each subgroup. By definition, The calculation formula is:

[0063] in For the number of subpopulations, A random number in [0,1). Indicates the first j The proportion of the fitness of the best particle (not globally optimal) in a subpopulation to the total fitness of the best particles in all non-globally optimal subpopulations, according to the definition. The calculation formula is:

[0064] and, Satisfy the following formula:

[0065] Furthermore, the crossover operation used in the swarm experience perception stage is the same as that in the individual experience perception stage. This operation allows particles to continuously absorb better genetic information from the swarm, constantly optimizing their own state during the iteration process, thereby improving global search capabilities and converging to the global optimum.

[0066] The experiments of this invention use literature

[16] The proposed algorithm was tested using six benchmarks: RA30, CPA, mRNA, R0, R1, and R2. These included colorimetric protein assays (CPA) performed on an RA30 chip, in vitro diagnostic (IVD) assays performed on a CPA chip, IVD detection applied to an mRNA chip, and three synthetic biology assays: R0, R1, and R2.

[16] Detailed information regarding the above benchmarks is shown in Table 1, where... and These represent the number of control modes and the number of control channels used in the control logic, respectively. The number of valve switching states at different times. In addition, the parameters used by the proposed algorithm are also listed in Table 2, where Ms is the maximum number of logic trees that can be continuously merged when constructing the logic forest.

[0067] Table 2. Benchmark Details and Algorithm Parameters

[0068] To verify the effectiveness of the role-based learning strategy, the DPSO algorithm without the role-based learning strategy is compared with the DPSO-RL algorithm with the role-based learning strategy. The DPSO algorithm without the role-based learning strategy learns from both the individual historical best and the group historical best for all particles. The comparison results are shown in Tables 3 and 4. The DPSO-RL algorithm improves the optimal number of valves, worst number of valves, average number of valves, and standard deviation by 15.96%, 20.08%, 18.37%, and 20.54%, respectively. It improves the optimal maximum number of valve switching times, worst maximum number of valve switching times, average maximum number of valve switching times, and standard deviation by 5.97%, 10.49%, 9.28%, and 14.03%, respectively.

[0069] Table 3 Comparison of valve count between DPSO and DPSO-RL

[0070] Table 4 Comparison of the maximum number of valve switching operations between DPSO and DPSO-RL

[0071] In DPSO, particles, when learning from collective experience, always tend to learn from the swarm's best particle. This singular learning object leads to homogenization of the swarm particles, making them prone to getting trapped in local optima early in the search and difficult to escape once trapped. In contrast, DPSO-RL employs a role-based learning strategy, enhancing information interaction and diversity among particles through role division: the swarm's best particle integrates the experience of multiple sub-swarm's best particles for updates, improving the overall globality of the search; sub-swarm's best particles learn from the best particles of higher-level sub-swarms, thus borrowing search directions from better subgroups; and ordinary particles learn from the best particle of their current subgroup, enhancing collaborative optimization capabilities within the subgroup. This hierarchical, multi-directional learning mechanism effectively guides particles towards high-quality solutions while preserving the swarm's diversity and exploratory capabilities, thereby improving the algorithm's convergence accuracy and stability.

[0072] After introducing a role-based learning strategy, the stability and solution quality of the algorithm are significantly improved. To verify the effectiveness of the proposed algorithm, this section compares it with that in the literature.

[16] The algorithm was compared with the improved DPSO algorithm. As shown in Table 5, the algorithm of the present invention improved the efficiency by 18.87%, 19.00%, 15.99%, and 12.44% in terms of optimal valve number, worst valve number, average valve number, and standard deviation, respectively.

[0073] Table 5 DPSO-RL and Literature

[16] Comparison of the number of valves in the algorithm

[0074] Furthermore, as shown in Table 6, optimization results of 6.80%, 14.74%, 12.57%, and 22.77% were achieved in terms of optimal maximum valve switching count, worst maximum valve switching count, average maximum valve switching count, and standard deviation, respectively. Experimental results demonstrate that the algorithm of this invention possesses high stability and can obtain a higher-quality control mode allocation scheme that considers valve switching balance.

[0075] Table 6 DPSO-RL and Literature

[16] Comparison of maximum valve switching counts in the algorithm

[0076] To further verify the superiority of the DPSO-RL algorithm proposed in this invention in the control mode allocation problem considering valve switching equalization, it is compared with that in the literature.

[16] This paper compares the proposed DDQN control mode method based on deep reinforcement learning. As shown in Table 7, the algorithm of this invention achieves optimizations of 9.88% and 11.6% in the number of valves and the maximum number of valve switching, respectively. The results show that the algorithm of this invention outperforms the reinforcement learning method under the DDQN architecture in both key optimization indicators, further highlighting its superiority in the control mode allocation problem considering valve switching balance.

[0077] Table 7 DPSO-RL and Literature

[16] Algorithm Comparison

[0078] Furthermore, comparing the algorithm DPSO-RL of this invention with that in the literature...

[16] The algorithm's convergence speed and accuracy on six benchmark tests. Figure 11 As shown, the DPSO-RL algorithm proposed in this invention achieves higher accuracy solutions in all benchmark tests, significantly outperforming the solutions in the literature.

[16] The method described in this paper, DPSO-RL, exhibits faster convergence speed in the early stages and maintains high optimization potential in the later stages, ultimately converging to a better global solution. This invention effectively maintains the diversity of the particle swarm, preventing it from prematurely falling into local optima. This allows particles to continuously explore and escape local traps during the search process, making it easier for them to converge towards the global optimum. In contrast, the literature...

[16] Due to a lack of sufficient diversity maintenance mechanisms, the convergence curve of the algorithm tends to plateau in the mid-to-late stages, which can easily lead to premature convergence and eventually get stuck in local optima, making it difficult to obtain high-precision solutions.

[0079] Based on the above embodiments, it can be seen that the present invention has the following advantages: 1. For the first time, a control mode allocation optimization model that simultaneously constrains the number of control valves and the maximum number of control valve switching times was constructed, which is of great significance for reducing the design complexity of control logic and improving the reliability of biochips.

[0080] 2. A control mode allocation algorithm for continuous microfluidic biochips considering valve switching balance is proposed, called the DPSO-RL (Discrete Particle Swarm Optimization with Human Social Learning Adaptation) algorithm. This algorithm introduces a role-based learning strategy, dividing particles into multiple subpopulations based on fitness, and introducing the roles of global optimal particle (Gbest), subpopulation optimal particle (Sbest), and ordinary particle (Pbest), so that a hierarchical learning mechanism is formed within the population, improving the exploration quality of control mode allocation schemes. 3. A discrete learning mechanism based on crossover and mutation is designed to achieve effective discretization of the continuous particle swarm optimization algorithm in the control mode allocation problem. This mechanism introduces crossover and mutation operations under feasibility constraints, which not only improves the global search capability of the algorithm, but also effectively improves the accuracy of the solution and ensures the quality of the obtained control mode allocation scheme. 4. Six test cases were used to evaluate the proposed control mode allocation algorithm. Experimental results show that the proposed algorithm can obtain a control mode allocation scheme with the minimum number of control valves and the minimum maximum number of control valve switching times. Example 2 Please see Figure 12 A microfluidic chip control mode allocation system considering valve switching balance, based on the above method, includes: The acquisition module is used to acquire the flow valve switching matrix and multiplexing combination matrix at different times; The encoding module is used to construct the particle encoding method for control mode allocation based on the flow valve switching matrix and the multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times; initialize the particle swarm position and velocity, and calculate the initial fitness of each particle; The strategy module is used to dynamically group the particle swarm using a discrete particle swarm optimization algorithm and a role-based learning strategy to generate a control mode allocation scheme. The role-based learning strategy includes: the global optimal particle integrates the experience of the optimal particles in each subgroup and updates it; the optimal particle in a subgroup learns the experience of the optimal particle in the upper subgroup; and ordinary particles learn the experience of the current optimal particle in their subgroup. The update module is used to update particle positions through crossover and mutation operations, and output a globally optimal control mode allocation scheme that satisfies the constraints.

[0081] In order to better utilize the above method, this application proposes a microfluidic chip control mode allocation system that considers valve switching balance. Each module corresponds to a step of the above method, and its specific principle has been described above and will not be repeated here.

[0082] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A microfluidic chip control mode allocation method considering valve switching balance, characterized in that, The method includes: The flow valve switching matrix and multiplexing combination matrix at different times are obtained as input; The particle coding method for control mode allocation is constructed based on the flow valve switching matrix and the multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times; the particle swarm position and velocity are initialized, and the initial fitness of each particle is calculated based on the constrained number of control valves and the maximum number of valve switching times; An algorithm based on discrete particle swarm optimization is adopted, and a role-based learning strategy is used to dynamically group the particle swarm to generate a control mode allocation scheme. The role-based learning strategy includes: the global optimal particle integrates the experience of the optimal particles in each subgroup and updates it; the optimal particle in a subgroup learns the experience of the optimal particle in the upper subgroup; and ordinary particles learn the experience of the current optimal particle in their subgroup. The particle positions are updated through crossover and mutation operations, and the globally optimal control mode allocation scheme that satisfies the constraints is output.

2. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 1, characterized in that, A particle coding method for control mode allocation is constructed based on the flow valve switching matrix and the multiplexing combination matrix, while simultaneously constraining the number of control valves and the maximum number of valve switching operations, including: Define the particle encoding format, particle The location code is: in, Set the value to 0 , indicating that the number is The control mode is assigned to the first Multiplexer combinations; Indicates the number of control channels. L Indicates the maximum number of multiplexing combinations; Construction constraints: .

3. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 1, characterized in that, Initialize the particle swarm's position and velocity, and calculate the initial fitness of each particle based on the constraint control valve number and maximum valve switching count, including: Randomly generated Given each particle position, initialize the particle velocity vector and calculate the initial fitness of each particle: in, and This represents two weighting factors. Indicates the number of control valves. This indicates the maximum number of times the control valve can be switched.

4. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 1, characterized in that, An algorithm based on discrete particle swarm optimization is adopted, which dynamically groups the particle swarm through a role-based learning strategy, including: Obtain the historical maximum and minimum values ​​of particle fitness, determine the subgroup boundary based on the historical maximum and minimum values, and assign particles to the corresponding subgroups according to the subgroup boundary; Different roles are assigned to particles, including global optimal particles, subgroup optimal particles, and ordinary particles, to guide different levels of learning strategies. The subpopulation size is dynamically adjusted to maintain population diversity, and the subpopulation partitioning is verified to satisfy the fitness upper and lower bound constraints.

5. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 1, characterized in that, Updating particle positions through crossover and mutation operations includes: The particle update formula is as follows: in, Indicates inertia weight, and It is the acceleration factor. For inertial components, Perceived through individual experience. For the perception of group experience. and These represent the historical maximum and minimum values ​​of the fitness function, respectively. Represents particles The fitness value.

6. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 5, characterized in that, The inertial component update is achieved through a mutation operation, and the update formula is as follows: in, It is a mutated particle. It is a random number between [0,1). It is a set of control mode numbers that have never been assigned; for particles in the t-th iteration , Through exchange and The control mode represented is used to update the particles; Then use Replace any of the control modes The control mode it represents; This allows you to choose any control mode from all available control modes to replace the previous one. If the control mode represented by the selected control mode is duplicated after replacement, the operation will be cancelled.

7. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 5, characterized in that, The individual experience perception is achieved through cross operations, and the update formula is as follows: in, These are the particles after the crossover. It is a random number in [0,1). Representing the t The best historical position of an individual particle in the next iteration. This indicates that two particles are performing an interleaved operation. The individual learning factor is defined as follows: in, The individual learning factor is the optimal particle in the swarm. The individual learning factor is the optimal particle of the subpopulation. For ordinary particles, the individual learning factor. It represents the first t The optimal particle of the population in the next iteration. Representing the t In the nth iteration j The optimal particle of a subpopulation Representing the t During the nth iteration i The subpopulation to which each particle belongs.

8. The microfluidic chip control mode allocation method considering valve switching balance as described in claim 5, wherein the group experience perception is achieved through role differentiation and cross-validation, and the update formulas for different roles are as follows: The calculation formula is: The calculation formula is: and, Satisfy the following formula: in, The group learning factor is the optimal particle in the group. The group learning factor is the optimal particle of the subgroup. For the population learning factor of ordinary particles, Represents the particles after the intersection. It is a random number in [0,1). This indicates an iterative crossover operation, used to guide cross-learning between the globally optimal particle and the optimal particles of each subgroup. The number of subpopulations; A random number in [0,1). Indicates the first j The proportion of the fitness of the best particle in a subpopulation to the total fitness of the best particles in all non-global optimal subpopulations.

9. A microfluidic chip control mode allocation system considering valve switching balance, characterized in that, Based on the method according to any one of claims 1-8, it includes: The acquisition module is used to acquire the flow valve switching matrix and multiplexing combination matrix at different times; The encoding module is used to construct the particle encoding method for control mode allocation based on the flow valve switching matrix and the multiplexing combination matrix, while constraining the number of control valves and the maximum number of valve switching times; initialize the particle swarm position and velocity, and calculate the initial fitness of each particle; The strategy module is used to dynamically group the particle swarm using a discrete particle swarm optimization algorithm and a role-based learning strategy to generate a control mode allocation scheme. The role-based learning strategy includes: the global optimal particle integrates the experience of the optimal particles in each subgroup and updates it; the optimal particle in a subgroup learns the experience of the optimal particle in the upper subgroup; and ordinary particles learn the experience of the current optimal particle in their subgroup. The update module is used to update particle positions through crossover and mutation operations, and output a globally optimal control mode allocation scheme that satisfies the constraints.

Images