Chromatographic sample injection control method and system based on dynamic programming

Abstract

The invention provides a chromatographic sample injection control method and system based on dynamic programming, and the method comprises the steps: collecting sample attributes, and constructing a characteristic data set of samples; according to the low-dimensional feature representation set and the most suitable equipment operation parameters of the current partition, designing a comprehensive optimization objective function to perform dynamic planning in the partition; performing global resource scheduling according to the optimal detection sequence of each partition in combination with the most suitable equipment operation parameters of the current partition, and generating a global resource plan; the real-time operation state and environment parameters of the equipment are collected, and the equipment state deviation and the environment state deviation are calculated; and performing sample detection batch by batch according to the optimized global resource plan, collecting feedback data of equipment, and integrating deviation analysis of all batches to optimize a performance evaluation result. The invention provides a comprehensive and efficient solution for chromatographic sample introduction optimization. The method not only has significant advantages in large-scale sample detection tasks, but also can adapt to diversified equipment and environment requirements.

Description

Technical Field

[0001] The invention belongs to the field of chromatographic injection control, and in particular relates to a chromatographic injection control method and system based on dynamic programming. Background Art

[0002] In the field of modern chromatography, the order of sample injection directly affects the detection efficiency and the accuracy of the results. Traditional chromatography methods rely on manual experience or simple rules, such as injection according to the order or category of sample submission. However, with the improvement of the performance of chromatography equipment and the increase in sample complexity, these traditional methods have exposed a series of limitations.

[0003] First, the chemical and physical properties of samples vary, and different samples have different requirements for equipment parameters (such as column temperature, flow rate, etc.). This diversity makes it impossible for simple sorting methods to meet optimization needs. For example, continuous injection of samples with high residence time and samples with low residence time may lead to degradation of equipment performance or cross-contamination. Secondly, there are certain resource limitations (such as maximum load, cleaning time) during the operation of the equipment. Traditional sorting methods fail to effectively consider these constraints, resulting in low resource utilization and even increased equipment loss. In addition, with the increase in analysis needs, large-scale sample detection tasks are becoming more and more common, which further exacerbates the complexity of sorting. Although static optimization methods can improve efficiency to a certain extent, they usually rely on fixed initial conditions and are difficult to dynamically respond to real-time changing environments (such as temperature and pressure fluctuations) or equipment conditions (such as operating failures or overloads).

[0004] Existing optimization techniques, such as sorting methods based on simple heuristic algorithms or optimization methods based on a single dynamic programming model, also have shortcomings in application. On the one hand, heuristic algorithms are usually designed for specific scenarios and it is difficult to take into account diverse sample characteristics and equipment parameters; on the other hand, although the single dynamic programming method has a high theoretical optimality, when faced with large-scale sample injection problems, the computational complexity increases significantly due to the explosion of the state space, making it difficult to achieve real-time optimization. At the same time, the lack of adaptive capabilities to the operating status of the equipment and the external environment makes these methods often perform poorly in practical applications.

[0005] In summary, the current chromatography injection optimization technology has prominent problems in the following aspects:

[0006] Insufficient adaptability of sample diversity: Failure to fully utilize sample characteristic information leads to suboptimal or even failed sorting results.

[0007] Low efficiency in equipment resource utilization: Equipment constraints and batch scheduling strategies are not effectively considered, resulting in serious waste of equipment resources.

[0008] Lack of dynamic adaptive capabilities: Failure to respond to device status and environmental changes in real time, resulting in poor robustness of optimization results.

[0009] To address the above issues, a new technical solution is urgently needed that can comprehensively consider sample characteristics, equipment constraints, and real-time environmental changes to systematically optimize chromatographic injection sequencing, thereby improving detection efficiency and equipment utilization, and ensuring the accuracy and robustness of test results. Summary of the invention

[0010] The purpose of the present invention is to provide a chromatographic injection control method and system based on dynamic programming, which effectively solves the problems of insufficient adaptability of sample diversity, low efficiency of equipment resource utilization, and lack of dynamic adaptive ability, and provides a comprehensive and efficient solution for chromatographic injection optimization. This method not only has significant advantages in large-scale sample detection tasks, but also can adapt to diverse equipment and environmental requirements.

[0011] In order to achieve the above object, a first aspect of the present invention provides a chromatographic injection control method based on dynamic programming, the method comprising:

[0012] S1. Collect sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes sample attributes, and then use the following standardization process to non-dimensionalize the attributes of each sample according to different dimensions of the sample characteristic value, map the dimensionless characteristic data to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then perform partition calculation based on a dynamic clustering algorithm according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, allocate the most suitable equipment operating parameters for the current partition; wherein the sample attributes include retention time, separation degree and detection sensitivity;

[0013] S2. Design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition; wherein the equipment operation parameters include the optimal column temperature of the partition equipment, the maximum load of the partition equipment, and the optimal flow rate of the partition equipment;

[0014] S3. Perform global resource scheduling based on the optimal detection sequence of each partition and the most suitable equipment operation parameters for the current partition, and generate a global resource plan, including a list and sequence of samples for each batch; wherein the global resource plan satisfies the constraints of the maximum load of the partition equipment and the minimization of the batch switching cost;

[0015] S4. Collect the real-time operating status and environmental parameters of the equipment, calculate the equipment state deviation and the environmental state deviation, respectively measure the degree of deviation between the current operating state and the state required by the scheduling plan, and if the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, trigger dynamic optimization and generate an optimized global resource plan; wherein, the dynamic optimization specifically includes:

[0016] Design a new target optimization function for global resource planning based on the current equipment state deviation and environment state deviation;

[0017] Recalculate the batch switching cost based on the real-time operating status of the equipment;

[0018] Determine the batch adjustment window, including the current batch and the next h batches;

[0019] Using a rolling optimization algorithm, the batch order is reordered within the window to minimize the objective optimization function of the new global resource plan;

[0020] Output the optimized sub-plan;

[0021] Replace the optimized sub-plan into the global resource plan to form an optimized global resource plan: If the batch covered by the window has been completed, monitor the equipment status of the subsequent batches; otherwise, dynamic optimization continues to be executed in a rolling manner;

[0022] S5. Perform sample testing batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, analyze the optimization performance evaluation results based on the deviation of all batches, and adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results.

[0023] Furthermore, the partition calculation based on the dynamic clustering algorithm is performed according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, specifically:

[0024] Randomly select k samples from the low-dimensional feature data set Z as the initial cluster centers {c1, c2, ..., c k};

[0025] Assign samples to the nearest cluster center:

[0026]

[0027] Among them, Cluster(i) represents the cluster to which sample i belongs, c j is the center of cluster j, z i It is a low-dimensional feature representation;

[0028] For each cluster, recompute its center to be the mean of the samples assigned to that cluster:

[0029]

[0030] Among them, N j is the number of samples in cluster j, Cluster(j) is the cluster to which sample j belongs;

[0031] Repeat the process of sample allocation and center updating until the center position no longer changes;

[0032] Finally, k partitions are formed, and the sample characteristics in each partition are highly similar.

[0033] Furthermore, the comprehensive optimization objective function F j , expressed as:

[0034]

[0035] Among them, T(i k ,i k+1 ) is the detection time cost, which measures the sample i k with i k+1 The detection time between k ,i k+1 ) is the cross contamination cost, which measures the sample i k with i k+1 The risk of cross contamination between k ,i k+1 ) is the uncertainty regularization term, for the detection sample i k with i k+1 The uncertainty penalty for the order between them; α and λ are the weight coefficients of cross contamination and uncertainty regularization terms, respectively, which are used to balance the influence of different optimization objectives;

[0036] The detection time cost is determined by the sample residence time difference and the device flow rate. Joint decision, the formula is:

[0037]

[0038] in, and For sample i k and i k+1 The residence time; is the equipment flow rate, which represents the adjustment factor of the detection speed;

[0039] The cross-contamination cost is determined by the sample separation r i and sensitivity e i The difference calculation of , while adding the innovation weight term to punish extreme pollution, is expressed as:

[0040]

[0041] in, For sample i k The separation degree indicates the separation effect of the sample from other samples in the equipment. A high separation degree indicates that there is less mutual interference between samples. For sample i k+1 The separation degree indicates the degree of separation between the next sample and other samples; For sample i k The sensitivity of the sample indicates that the sample has the highest responsiveness and sensitivity in the equipment detection, and the sample has the strongest detection signal; For sample i k+1 The sensitivity of β is a dynamic adjustment weight, which is generated based on the overall separation distribution of samples in the partition:

[0042] β=1+exp(-σ r )

[0043] Among them, σ r is the standard deviation of the separation of the partitioned samples, used to adjust the impact of extreme contamination;

[0044] The uncertainty regularization term introduces constraints on sample sequence fluctuations to reduce large changes in the detection sequence:

[0045]

[0046] Among them, γ is the regularization strength factor, which controls the stability of the detection sequence; is the absolute difference in sample retention time. A larger difference will increase the regularization penalty.

[0047] Furthermore, the minimization of the comprehensive cost of each pair of samples is expressed as:

[0048] Construct the state transfer formula for dynamic programming:

[0049]

[0050] Where f(i,S) represents the optimal detection cost starting from sample i and the remaining sample set S; j represents sample j, T(i,j) represents the detection time cost between samples i and j, P(i,j) represents the cross-contamination cost between samples i and j, α represents the weight coefficient of the cross-contamination cost, λ represents the weight coefficient of the uncertainty regularization term, and f(j,S\{j}) represents the optimal cost starting from sample j and removing j;

[0051] Initial state Indicates that all samples have been tested; among them, Indicates that there is no remaining sample;

[0052] Use a dynamic programming table to record the optimal path for each step, and backtrack from the final state to generate the optimal detection sequence o j ={i1,i2,…,i m}, and get the optimal detection sequence {o1,o2,…,o j ,o k}.

[0053] Furthermore, the global resource scheduling is performed according to the optimal detection sequence of each partition combined with the most suitable device operation parameters for the current partition, specifically including:

[0054] According to the maximum load of the partition device, the optimal detection sequence of the partition is o j Assign to batches in sequence

[0055] When it is found that the maximum load of the current partition device has been reached, it is necessary to switch the current generation of detection samples to the device with the minimum batch switching cost to control the operation load balance of the device;

[0056] Use a recursive optimization algorithm to solve the optimal batch order:

[0057] Initial state Indicates that all batch scheduling is completed;

[0058] The transfer formula is:

[0059]

[0060] in, is the kth batch in batch scheduling;

[0061] Generate a global scheduling plan by backtracking

[0062] Furthermore, the batch switching cost is composed of the temperature adjustment time between batches and the cross-contamination cost;

[0063] Design and optimize the objective function G to minimize the switching cost between all batches while controlling the load balance of the equipment, which can be expressed as:

[0064]

[0065] Among them, λ is the weight coefficient of load balancing; m represents the total number of batches, represents the batch switching cost, represents the batch switching cost matrix, represents the jth batch, represents the j+1th batch; the second item Improve equipment operation efficiency by controlling batch load deviation; Load avg Indicates the average load of the device. represents the load of the jth batch.

[0066] Furthermore, the target optimization function G′ of the new global resource plan is expressed as:

[0067]

[0068] in, The adjusted batch switching cost is recalculated based on the real-time status; and η are weight coefficients, which control the influence of equipment deviation and environmental adaptability respectively, Δr is the equipment state deviation, and Δe is the environmental state deviation;

[0069] The adjusted batch switching cost It is expressed as:

[0070]

[0071] in, Indicates the current temperature T real With target temperature Adjustment time; The updated cross-contamination cost, combined with the real-time sensitivity bias correction:

[0072]

[0073] Among them, Δe k =|e k -e real |Indicates the real-time deviation of sample sensitivity; indicates; r i represents the separation degree of sample i, r k represents the separation degree of sample k, e i represents the sensitivity of sample i, e k represents the sensitivity of sample k, ω represents the weight coefficient of cross-contamination cost, which is used to adjust the impact of sensitivity difference on contamination, Δe k =|e k -e real | represents the real-time deviation of sample sensitivity, e real Indicates real-time sample sensitivity.

[0074] Furthermore, the deviation analysis optimization performance evaluation results of all batches are integrated, and the target optimization function and scheduling strategy are adjusted according to the optimization performance evaluation results, specifically including:

[0075] Comprehensively consider the deviations of all batches and define the global optimization performance index Gperf , reflecting the overall adaptability and scheduling efficiency of plan execution:

[0076]

[0077] Among them, the first represents the average deviation of all batches; Δ j Represents the comprehensive deviation index; the second item Indicates the scheduling efficiency by the idle time T between batches j idle It is measured by the ratio of the execution time to the total execution time; ω is the efficiency weight, which is dynamically adjusted to meet the specific device requirements;

[0078] The comprehensive deviation index Δ j , expressed as:

[0079] Δ j =α T ·|T j exec -T j plan |+α P ·|P j exec -P j plan |+α E ·Δe

[0080] Among them, T j plan and P j plan are the target detection time and pollution level corresponding to the scheduling plan; Δe is the degree of deviation of environmental parameters, which is used to evaluate the impact of environmental changes on the execution results; the weight α T ,α P ,α E To balance the contribution of time, pollution and environment, P j exec represents the actual contamination level of batch j, P j plan represents the target contamination level of batch j in the scheduling plan;

[0081] Adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results:

[0082] If the global optimization performance index G perf If the target range is exceeded, the objective function weight α of the scheduling plan is dynamically adjusted T ,α P ,α E , balance various optimization objectives;

[0083] Update batch switching cost matrix Especially the highest environmental fluctuations and the influence of equipment aging.

[0084] In another aspect of the present invention, a chromatographic injection control system based on dynamic programming is provided, the system comprising:

[0085] A sample attribute acquisition unit is used to acquire sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes the sample attribute, and then the attribute of each sample is dimensionally processed by the following standardization process according to different dimensions of the sample characteristic value, and the dimensionally processed characteristic data is mapped to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then partition calculation based on a dynamic clustering algorithm is performed according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, the most suitable equipment operation parameters for the current partition are allocated; wherein the sample attributes include retention time, separation degree and detection sensitivity;

[0086] A detection sequence generation unit is used to design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition, so as to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition; wherein the equipment operation parameters include the optimal column temperature of the partition equipment, the maximum load of the partition equipment and the optimal flow rate of the partition equipment;

[0087] A global detection sequence generation unit is used to perform global resource scheduling according to the optimal detection sequence of each partition and the most suitable equipment operation parameters for the current partition, and generate a global resource plan, including a sample list and sequence for each batch; wherein the global resource plan satisfies the constraints of the maximum load of the partition equipment and the constraints of minimizing the batch switching cost;

[0088] The global detection sequence optimization unit is used to collect the real-time operating status and environmental parameters of the equipment, calculate the equipment state deviation and the environmental state deviation, and respectively measure the degree of deviation between the current operating state and the state required by the scheduling plan. If the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, dynamic optimization is triggered to generate an optimized global resource plan; wherein, the dynamic optimization specifically includes:

[0089] Design a new target optimization function for global resource planning based on the current equipment state deviation and environment state deviation;

[0090] Recalculate the batch switching cost based on the real-time operating status of the equipment;

[0091] Determine the batch adjustment window, including the current batch and the next h batches;

[0092] Using a rolling optimization algorithm, the batch order is reordered within the window to minimize the objective optimization function of the new global resource plan;

[0093] Output the optimized sub-plan;

[0094] Replace the optimized sub-plan into the global resource plan to form an optimized global resource plan: If the batch covered by the window has been completed, monitor the equipment status of the subsequent batches; otherwise, dynamic optimization continues to be executed in a rolling manner;

[0095] The sampling optimization unit is used to perform sample testing batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, analyze the optimization performance evaluation results based on the deviation of all batches, and adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results.

[0096] The beneficial technical effects of the present invention are at least as follows:

[0097] (1) The present invention innovatively introduces a sample partitioning method, which dynamically divides the partitions according to the chemical properties (such as retention time and separation) and physical properties (such as detection sensitivity) of the sample. By managing samples by partition, the sorting complexity problem caused by sample diversity is effectively solved, ensuring the efficiency and accuracy of optimization within the partition. A multidimensional dynamic programming model is used to optimize the sample injection order in each partition, and local optimization is achieved in combination with equipment parameters (such as flow rate and column temperature).

[0098] (2) To address the problem of low equipment resource utilization, the present invention designs an intelligent diversion algorithm to divide samples into batches between partitions and dynamically schedule the batch order to balance the equipment load and maximize resource utilization efficiency.

[0099] Batch optimization not only reduces equipment idle time, but also reduces the overall complexity of injection sequencing, and realizes the coordinated work of local optimization within partitions and global optimization between partitions.

[0100] (3) To address the problem of lack of dynamic adjustment capability, the present invention adds a real-time feedback module to monitor the device status (such as load, temperature, flow rate) and environmental changes (such as temperature fluctuations) in real time.

[0101] Dynamically adjust partition and batch optimization strategies during operation to improve the robustness and flexibility of the system and ensure that the optimization plan always adapts to the actual operation situation.

[0102] (4) The present invention effectively solves the problems of insufficient adaptability to sample diversity, low efficiency of equipment resource utilization, and lack of dynamic adaptive capabilities, and provides a comprehensive and efficient solution for chromatographic injection optimization. This method not only has significant advantages in large-scale sample detection tasks, but also can adapt to diverse equipment and environmental requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0104] Figure 1 The present invention is a flow chart of a chromatographic injection control method based on dynamic programming according to an embodiment of the present invention.

[0105] Figure 2 The present invention is a framework diagram of a chromatographic injection control system based on dynamic programming. DETAILED DESCRIPTION

[0106] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0107] like Figure 1 As shown, the chromatographic injection control method based on dynamic programming provided by an embodiment of the present invention comprises the following steps S1-S5:

[0108] S1. Collect sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes sample attributes, and then use the following standardization processing to non-dimensionalize the attributes of each sample according to the different dimensions of the sample characteristic values, map the dimensionless characteristic data to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then perform partition calculation based on a dynamic clustering algorithm according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, allocate the most suitable equipment operating parameters for the current partition; wherein the sample attributes include retention time, separation degree and detection sensitivity.

[0109] Specifically, the characteristic data set D of the received samples is {d1, d2, ..., d n}, where each sample d i Contains three attributes:

[0110] t i : Retention time (RetentionTime), in seconds;

[0111] r i : Separation Factor, dimensionless value;

[0112] e i : Detection sensitivity (Sensitivity), a dimensionless value ranging from [0,1].

[0113] Furthermore, according to the different dimensions of the sample characteristic values, the following standardized formula is used to perform dimensionless processing on the properties of each sample:

[0114]

[0115] Among them, μ t ,μ r ,μ e are the means of retention time, separation and detection sensitivity respectively; σ t ,σ r ,σ e are the standard deviations of the corresponding attributes. After standardization, the characteristic data of all samples form a set D′={d1′,d2′,…d i ′,d n ′}, the attribute of each sample is normalized to a dimensionless value.

[0116] Furthermore, the standardized characteristic data D′ is mapped to a low-dimensional feature space (usually two-dimensional or three-dimensional) to express the sample characteristics more compactly. The mapping formula is:

[0117] z i =W·d i ′ (2)

[0118] Where W is the weight matrix, and the initial value is calculated by singular value decomposition (SVD) to ensure that the main information between samples is retained; i is a low-dimensional feature representation of sample i (such as a two-dimensional or three-dimensional vector).

[0119] The low-dimensional feature data set after mapping is Z = {z1,z2,…,z n}, to facilitate subsequent partitioning operations.

[0120] Furthermore, the dynamic clustering algorithm is used to partition the low-dimensional characteristics Z of the sample to obtain k partitions Each partition contains samples with similar characteristics. The clustering process includes the following steps:

[0121] Initialization: Randomly select k samples from Z as the initial cluster centers {c1, c2, ..., c k};

[0122] Assign samples: Assign samples to the nearest cluster center:

[0123]

[0124] Cluster(i) represents the cluster to which sample i belongs, c j is the center of cluster j;

[0125] Furthermore, update the cluster center: for each cluster, recalculate its center as the average value of the samples assigned to the cluster:

[0126]

[0127] Where N j is the number of samples in cluster j.

[0128] Iteration: Repeat the process of sample allocation and center update until the center position no longer changes.

[0129] Finally, k partitions are formed, and the sample characteristics in each partition are highly similar.

[0130] Furthermore, for each partition According to the characteristic distribution of samples in the partition, the equipment operation parameters p suitable for the partition are allocated j ,include:

[0131] T j opt : The optimal column temperature of partition j is calculated by weighting the mean of the retention time distribution of samples in the partition;

[0132] The maximum load of partition j, associated with the total number of samples in the partition;

[0133] The optimal flow rate of partition j is determined according to the sensitivity distribution of the samples in the partition.

[0134] Understandably, output partition results Each partition contains a list of samples with similar characteristics; at the same time, the corresponding equipment operating parameters {p1, p2, …, p k}, these results will serve as direct input for subsequent dynamic programming optimization.

[0135] S2. Design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition.

[0136] Specifically, obtain the data and device parameters of each partition from the result of step 1:

[0137] The sample set in partition j, each sample d i =[t i ,r i ,ei ] includes the residence time t i , separation r i and sensitivity e i ;

[0138] Equipment parameters for partition j, including optimal column temperature T j opt , Maximum load and optimal flow rate

[0139] Goal: Optimize the order of sample testing for each partition j ={i1,i2,…,i m}, to minimize the combined cost of detection time and cross-contamination, while considering the uncertainty of detection sequence and equipment constraints.

[0140] Furthermore, the comprehensive optimization objective function F is constructed j , the specific formula is:

[0141]

[0142] in,

[0143] T(i k ,i k+1 ): Detection time cost, measuring sample i k with i k+1 The detection time between

[0144] P(i k ,i k+1 ): Cross-contamination cost, measuring the risk of cross-contamination between samples;

[0145] R(i k ,i k+1 ): A newly introduced uncertainty regularization term, which penalizes the uncertainty of the detection order;

[0146] α and λ: are the weight coefficients of cross contamination and uncertainty regularization terms, respectively, used to balance the impact of different optimization objectives.

[0147] Furthermore, the detection time cost T(i k ,i k+1 ) is determined by the sample retention time difference and the equipment flow rate Joint decision, the formula is:

[0148]

[0149] in, and For sample i k and ik+1 The residence time; is the equipment flow rate, which represents the adjustment factor of the detection speed.

[0150] Furthermore, the cross-contamination cost P(i k ,i k+1 ) by the sample separation r i and sensitivity e i The difference between , while adding an innovative weight term to penalize extreme pollution:

[0151]

[0152] Among them, β is a dynamically adjusted weight, which is generated based on the overall separation distribution of samples in the partition:

[0153] β=1+exp(-σ r ) (8)

[0154] where σ r is the standard deviation of the separation of the partitioned samples, used to adjust for the effects of extreme contamination.

[0155] Furthermore, the uncertainty regularization term R(i k ,i k+1 ) introduces constraints on sample sequence fluctuations to reduce large changes in the detection sequence:

[0156]

[0157] Among them, γ: regularization strength factor, controlling the stability of the detection sequence; The absolute difference in sample retention times. Larger differences will increase the regularization penalty.

[0158] Furthermore, the state transfer formula of dynamic programming is constructed:

[0159]

[0160] Among them, f(i,S): represents the optimal detection cost in the remaining sample set S starting from sample i;

[0161] Initial state Indicates that all samples have been tested.

[0162] Furthermore, a dynamic programming table is used to record the optimal path for each step, and the optimal detection sequence is generated by backtracking from the final state. j ={i1,i2,…,i m}.

[0163] Finally, the optimal detection sequence {o1,o2,…o j ,o k}; The comprehensive cost (including time, pollution, and uncertainty) of each pair of samples is used for visualization and subsequent scheduling analysis.

[0164] S3. Perform global resource scheduling based on the optimal detection sequence of each partition and the most suitable equipment operating parameters for the current partition, and generate a global resource plan, including a list and sequence of samples for each batch.

[0165] Specifically, obtain the following input data from step 2:

[0166] {o1,o2,…,o k}: The optimal detection sequence within the partition, each o j ={i1,i2,…,i m} represents the detection order of partition j;

[0167] {p1,p2,…,p k}: Device parameters corresponding to the partition, each represents the column temperature, maximum load and flow rate of partition j.

[0168] Objective: Generate a global scheduling plan S across partitions to minimize the batch switching cost while ensuring efficient use of device resources.

[0169] Furthermore, according to the maximum load of the equipment The partition sample {o j}Assign to batches in order

[0170]

[0171] Initialization: Samples are added to the current batch in the optimal order within the partition

[0172] Checking Batch Load Whether it exceeds

[0173]

[0174] where w i is the load weight of sample i, with the initial value being the sensitivity e i ; Start a new batch when the load limit is exceeded

[0175]

[0176] Furthermore, an innovative design is made: a load balancing term η is added when calculating the load to penalize large load differences between partitions, making the batch division more balanced:

[0177]

[0178] Load avg is the average load of all current batches.

[0179] Furthermore, the batch switching cost Mainly composed of temperature adjustment time between batches and cross-contamination costs:

[0180] Temperature adjustment cost: column temperature T corresponding to the batch j opt and The difference is related to

[0181] Cross-contamination cost: The separation degree of batch samples is i and sensitivity e i The difference determines the

[0182] Combining these two parts, the switching cost can be expressed as:

[0183]

[0184] The cost of cross contamination It has been defined in step 2 and will not be repeated here.

[0185] Furthermore, the optimization goal is to minimize the switching cost between all batches while controlling the load balance of the equipment:

[0186]

[0187] λ is the weight coefficient of load balancing;

[0188] The second item further improves equipment operating efficiency by controlling batch load deviations.

[0189] Furthermore, a recursive optimization algorithm is used to solve the optimal batch order:

[0190] Initial state Indicates that all batch scheduling is completed;

[0191] The transfer formula is:

[0192]

[0193] Generate a global scheduling plan by backtracking

[0194] As can be understood, the global scheduling plan S is finally output, including the sample list and order of each batch;

[0195] Batch switching cost matrix for further visualization and result analysis.

[0196] S4. Collect the real-time operating status and environmental parameters of the equipment, calculate the equipment state deviation and environmental state deviation, and measure the degree of deviation between the current operating state and the state required by the scheduling plan. If the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, dynamic optimization is triggered to generate an optimized global resource plan.

[0197] Specifically, the global scheduling plan is obtained from step S3 and the batch switching cost matrix At the same time, the following status is collected in real time:

[0198] r=[T real ,F real ,P real ]: Real-time operating status of the equipment, including column temperature T real , flow rate F real and pressure P real ;

[0199] e=[θ temp ,θ humidity ]: Environmental parameters, including ambient temperature θ temp and humidityθ humidity .

[0200] Objective: Based on real-time collected equipment and environment data, dynamically adjust the scheduling plan S to minimize deviations and resource waste in actual execution.

[0201] Furthermore, the equipment state deviation Δr and the environment state deviation Δe are defined to measure the degree of deviation between the current operating state and the state required by the scheduling plan:

[0202] Δr=|rp j |, Δe=|ee ref | (16)

[0203] in, Batch scheduling Equipment operating parameters;

[0204] e ref : Nominal working environment parameters of the equipment.

[0205] When any of Δr or Δe exceeds the preset threshold τ r or τ e , triggering dynamic optimization.

[0206] Furthermore, a new optimization objective function is defined for the adjusted scheduling plan S′, which comprehensively considers the switching cost, equipment state deviation and environmental adaptability:

[0207]

[0208] in, The adjusted batch switching cost is recalculated based on the real-time status; and η: weight coefficients, which control the influence of equipment deviation and environmental adaptability respectively.

[0209] Furthermore, the batch switching cost The impact of real-time status on temperature adjustment and cross contamination is considered:

[0210]

[0211] in,

[0212] Indicates the adjustment time between the current temperature and the target temperature;

[0213] The updated cross-contamination cost, combined with the real-time sensitivity bias correction:

[0214]

[0215] Δe k =|e k -e real |Indicates the real-time deviation of sample sensitivity.

[0216] Further, determine the adjustment window Contains the current batch and the following h batches;

[0217] Using the rolling optimization algorithm, the batch order is reordered within the window to minimize the adjusted objective function G′;

[0218] Output optimized sub-plan

[0219] Understandably, the adjusted sub-plan Replace it into the global scheduling plan S to form a new scheduling plan S′;

[0220] If the batch covered by the window has been completed, continue to monitor the equipment status of subsequent batches;

[0221] Otherwise, dynamic optimization continues to execute in a rolling manner.

[0222] Finally, the updated global scheduling plan S′ includes the order adjustment of all batches;

[0223] The deviation and adjustment times of the equipment status during the optimization process are recorded for subsequent performance analysis.

[0224] S5. Perform sample testing batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, analyze the optimization performance evaluation results based on the deviation of all batches, and adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results.

[0225] Specifically, the updated global scheduling plan is obtained from step S4 and the adjusted batch switching cost matrix The real-time operating parameters of the experimental equipment r = [T real ,F real ,P real ] and the environmental state e=[θ temp ,θ humidity ] is also used as input for result recording and optimization evaluation.

[0226] Furthermore, sample testing is performed batch by batch according to the scheduling plan S′:

[0227] In each batch Before starting, according to the required equipment parameters Adjust the device status r.

[0228] If Δr=|rp j |Exceeding the set threshold τ r , trigger recalibration of the device status and record the calibration time T calib And dynamically adjust the testing schedule.

[0229] Furthermore, sample testing is completed in batches, and the following indicators are recorded:

[0230] Total batch detection time T j exec ;

[0231] The contamination level within the batch is P j exec , measured by the dynamic match between sample characteristics and sensitivity.

[0232] Furthermore, after each batch is executed, the deviation between the actual result and the scheduling plan target is calculated, and a comprehensive deviation index is defined:

[0233] Δ j =α T ·|T j exec -T j plan |+α P ·|P j exec -P j plan |+α E·Δe(20)

[0234] Among them, T j plan and P j plan are the target detection time and pollution level corresponding to the scheduling plan; Δe is the degree of deviation of environmental parameters, which is used to evaluate the impact of environmental changes on the execution results; the weight α T ,α P ,α E A contribution to balance time, pollution and the environment.

[0235] Furthermore, the deviations of all batches are integrated to define the global optimization performance index G perf , reflecting the overall adaptability and scheduling efficiency of plan execution:

[0236]

[0237] Among them, the first term represents the average deviation of all batches;

[0238] The second term represents the scheduling efficiency, which is calculated by the idle time T between batches. j idle Measured as a ratio to the total execution time;

[0239] ω is the efficiency weight, which is dynamically adjusted to suit specific equipment requirements.

[0240] Furthermore, according to the optimization performance evaluation results, the optimization model and scheduling strategy are adjusted:

[0241] If G perf If the target range is exceeded, the objective function weight α of the scheduling plan is dynamically adjusted T ,α P ,α E , balance various optimization objectives;

[0242] Update batch switching cost matrix Especially considering higher environmental fluctuations and the effects of equipment aging.

[0243] Finally, the output is:

[0244] Execution records: actual testing time, contamination level, equipment status and environmental parameters for each batch;

[0245] Optimization evaluation: global optimization performance index G perf and key deviation indicator Δ j ;

[0246] Improvement suggestions: including objective function adjustment and plan update strategy, providing reference for the next round of scheduling optimization.

[0247] like Figure 2 As shown, in another embodiment of the present invention, a chromatographic injection control system based on dynamic programming is provided, and the system comprises:

[0248] The sample attribute acquisition unit 301 is used to acquire sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes the sample attribute, and then the attribute of each sample is dimensionally processed by the following standardization process according to the different dimensions of the sample characteristic value, and the dimensionally processed characteristic data is mapped to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then a partition calculation based on a dynamic clustering algorithm is performed according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, the most suitable equipment operation parameters for the current partition are allocated; wherein the sample attributes include retention time, separation degree and detection sensitivity;

[0249] The detection sequence generation unit 302 is used to design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition, so as to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition; wherein the equipment operation parameters include the optimal column temperature of the partition equipment, the maximum load of the partition equipment and the optimal flow rate of the partition equipment;

[0250] The global detection sequence generation unit 303 is used to perform global resource scheduling according to the optimal detection sequence of each partition and the most suitable device operation parameters for the current partition, and generate a global resource plan, including a sample list and sequence for each batch; wherein the global resource plan satisfies the constraints of the maximum load of the partition device and the constraints of minimizing the batch switching cost;

[0251] The global detection sequence optimization unit 304 is used to collect the real-time operation status and environmental parameters of the equipment, calculate the equipment state deviation and the environmental state deviation, and respectively measure the degree of deviation between the current operation state and the state required by the scheduling plan. If the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, dynamic optimization is triggered to generate an optimized global resource plan; wherein, the dynamic optimization specifically includes:

[0252] Design a new target optimization function for global resource planning based on the current equipment state deviation and environment state deviation;

[0253] Recalculate the batch switching cost based on the real-time operating status of the equipment;

[0254] Determine the batch adjustment window, including the current batch and the next h batches;

[0255] Using a rolling optimization algorithm, the batch order is reordered within the window to minimize the objective optimization function of the new global resource plan;

[0256] Output the optimized sub-plan;

[0257] Replace the optimized sub-plan into the global resource plan to form an optimized global resource plan: If the batch covered by the window has been completed, monitor the equipment status of the subsequent batches; otherwise, dynamic optimization continues to be executed in a rolling manner;

[0258] The sampling optimization unit 305 is used to perform sample detection batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, comprehensively analyze the deviation of all batches to optimize the performance evaluation results, and adjust the target optimization function and scheduling strategy according to the optimized performance evaluation results.

[0259] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0260] In addition, for technical details that are not described in detail in this embodiment, reference can be made to the parameter operation method provided in any embodiment of the present invention, and will not be repeated here.

[0261] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0262] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0263] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0264] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A chromatographic injection control method based on dynamic programming, characterized in that: The method comprises: S1. Collect sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes sample attributes, and then use the following standardization process to non-dimensionalize the attributes of each sample according to different dimensions of the sample characteristic value, map the dimensionless characteristic data to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then perform partition calculation based on a dynamic clustering algorithm according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, allocate the most suitable equipment operating parameters for the current partition; wherein the sample attributes include retention time, separation degree and detection sensitivity; S2. Design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition; wherein the equipment operation parameters include the optimal column temperature of the partition equipment, the maximum load of the partition equipment, and the optimal flow rate of the partition equipment; S3. Perform global resource scheduling based on the optimal detection sequence of each partition and the most suitable equipment operation parameters for the current partition, and generate a global resource plan, including a list and sequence of samples for each batch; wherein the global resource plan satisfies the constraints of the maximum load of the partition equipment and the minimization of the batch switching cost; S4. Collect the real-time operating status and environmental parameters of the equipment, calculate the equipment state deviation and the environmental state deviation, respectively measure the degree of deviation between the current operating state and the state required by the scheduling plan, and if the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, trigger dynamic optimization and generate an optimized global resource plan; wherein, the dynamic optimization specifically includes: Design a new target optimization function for global resource planning based on the current equipment state deviation and environment state deviation; Recalculate the batch switching cost based on the real-time operating status of the equipment; Determine the batch adjustment window, including the current batch and the next h batches; Using a rolling optimization algorithm, the batch order is reordered within the window to minimize the objective optimization function of the new global resource plan; Output the optimized sub-plan; Replace the optimized sub-plan into the global resource plan to form an optimized global resource plan: If the batch covered by the window has been completed, monitor the equipment status of the subsequent batches; otherwise, dynamic optimization continues to be executed in a rolling manner; S5. Perform sample testing batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, analyze the optimization performance evaluation results based on the deviation of all batches, and adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results.

2. The chromatographic injection control method based on dynamic programming according to claim 1, characterized in that: The partition calculation based on the dynamic clustering algorithm is performed according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, specifically: Randomly select k samples from the low-dimensional feature data set Z as the initial cluster centers {c1, c2, ..., c k }; Assign samples to the nearest cluster center: Among them, Cluster(i) represents the cluster to which sample i belongs, c j is the center of cluster j, z i It is a low-dimensional feature representation; For each cluster, recompute its center to be the mean of the samples assigned to that cluster: Among them, N j is the number of samples in cluster j, Cluster(j) is the cluster to which sample j belongs; Repeat the process of sample allocation and center updating until the center position no longer changes; Finally, k partitions are formed, and the sample characteristics in each partition are highly similar.

3. The chromatographic injection control method based on dynamic programming according to claim 1, characterized in that: The comprehensive optimization objective function F j , expressed as: Among them, T(i k ,i k+1 ) is the detection time cost, which measures the sample i k with i k+1 The detection time between k ,i k+1 ) is the cross contamination cost, which measures the sample i k with i k+1 The risk of cross contamination between k ,i k+1 ) is the uncertainty regularization term, for the detection sample i k with i k+1 The uncertainty penalty for the order between them; α and λ are the weight coefficients of cross contamination and uncertainty regularization terms, respectively, which are used to balance the influence of different optimization objectives; The detection time cost is determined by the sample residence time difference and the device flow rate. Joint decision, the formula is: in, and For sample i k and i k+1 The residence time; is the equipment flow rate, which represents the adjustment factor of the detection speed; The cross-contamination cost is determined by the sample separation r i and sensitivity e i The difference calculation of , while adding the innovation weight term to punish extreme pollution, is expressed as: in, For sample i k The separation degree indicates the separation effect of the sample from other samples in the equipment. A high separation degree indicates that there is less mutual interference between samples. For sample i k+1 The separation degree indicates the degree of separation between the next sample and other samples; For sample i k The sensitivity of the sample indicates that the sample has the highest responsiveness and sensitivity in the equipment detection, and the sample has the strongest detection signal; For sample i k+1 The sensitivity of β is a dynamic adjustment weight, which is generated based on the overall separation distribution of samples in the partition: β=1+exp(-σ r ) Among them, σ r is the standard deviation of the separation of the partitioned samples, used to adjust the impact of extreme contamination; The uncertainty regularization term introduces constraints on sample sequence fluctuations to reduce large changes in the detection sequence: Among them, γ is the regularization strength factor, which controls the stability of the detection sequence; is the absolute difference in sample retention time. A larger difference will increase the regularization penalty.

4. The chromatographic injection control method based on dynamic programming according to claim 3, characterized in that: The comprehensive cost of minimizing each pair of samples is expressed as: Construct the state transfer formula for dynamic programming: Where f(i,S) represents the optimal detection cost starting from sample i and the remaining sample set S; j represents sample j, T(i,j) represents the detection time cost between samples i and j, P(i,j) represents the cross-contamination cost between samples i and j, α represents the weight coefficient of the cross-contamination cost, λ represents the weight coefficient of the uncertainty regularization term, and f(j,S\{j}) represents the optimal cost starting from sample j and removing j; Among them, the initial state Indicates that all samples have been tested; among them, Indicates that there is no remaining sample; Use a dynamic programming table to record the optimal path for each step, and backtrack from the final state to generate the optimal detection sequence o j ={i1,i2,…,i m }, and obtain the optimal detection sequence {o1,o2,…,o j ,o k }.

5. The chromatographic injection control method based on dynamic programming according to claim 4, characterized in that: The global resource scheduling is performed according to the optimal detection sequence of each partition combined with the most suitable device operation parameters for the current partition, specifically including: According to the maximum load of the partition device, the optimal detection sequence of the partition is o j Assign to batches in sequence When it is found that the maximum load of the current partition device has been reached, it is necessary to switch the current sample to be tested to the device with the minimum batch switching cost to control the operation load balance of the device; Use a recursive optimization algorithm to solve the optimal batch order: Initial state Indicates that all batch scheduling is completed; The transfer formula is: in, is the kth batch in batch scheduling; Generate a global scheduling plan by backtracking 6. The chromatographic injection control method based on dynamic programming according to claim 1, characterized in that: The batch switching cost is composed of the temperature adjustment time between batches and the cross-contamination cost; Design and optimize the objective function G to minimize the switching cost between all batches while controlling the load balance of the equipment, which can be expressed as: Among them, λ is the weight coefficient of load balancing; m represents the total number of batches, represents the batch switching cost, represents the batch switching cost matrix, represents the jth batch, represents the j+1th batch; the second term λ· Improve equipment operation efficiency by controlling batch load deviation; Load avg Indicates the average load of the device. represents the load of the jth batch.

7. The chromatographic injection control method based on dynamic programming according to claim 6, characterized in that: The objective optimization function G′ of the new global resource plan is expressed as: in, is the adjusted batch switching cost, which is recalculated based on the real-time status; θ and η are weight coefficients, which control the influence of equipment deviation and environmental adaptability respectively; Δr is the equipment state deviation, and Δe is the environmental state deviation; The adjusted batch switching cost It is expressed as: in, Indicates the current temperature T real With target temperature Adjustment time; The updated cross-contamination cost, combined with the real-time sensitivity bias correction: Among them, Δe k =|e k -e real |Indicates the real-time deviation of sample sensitivity; indicates; r i represents the separation degree of sample i, r k represents the separation degree of sample k, e i represents the sensitivity of sample i, e k represents the sensitivity of sample k, ω represents the weight coefficient of cross-contamination cost, which is used to adjust the impact of sensitivity difference on contamination, Δe k =|e k -e real | represents the real-time deviation of sample sensitivity, e real Indicates real-time sample sensitivity.

8. The chromatographic injection control method based on dynamic programming according to claim 1, characterized in that: The deviation analysis of all batches is integrated to optimize the performance evaluation results, and the target optimization function and scheduling strategy are adjusted according to the optimized performance evaluation results, specifically including: Comprehensively consider the deviations of all batches and define the global optimization performance index G perf , reflecting the overall adaptability and scheduling efficiency of plan execution: Among them, the first represents the average deviation of all batches; Δ j Represents the comprehensive deviation index; the second item Indicates scheduling efficiency, through the idle time between batches It is measured by the ratio of the execution time to the total execution time; ω is the efficiency weight, which is dynamically adjusted to meet the specific device requirements; The comprehensive deviation index Δ j , expressed as: Among them, T j plan and P j plan are the target detection time and pollution level corresponding to the scheduling plan; Δe is the degree of deviation of environmental parameters, which is used to evaluate the impact of environmental changes on the execution results; the weight α T ,α P ,α E To balance the contribution of time, pollution and environment, P j exec represents the actual contamination level of batch j, P j plan represents the target contamination level of batch j in the scheduling plan; Adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results: If the global optimization performance index G perf If the target range is exceeded, the objective function weight α of the scheduling plan is dynamically adjusted T ,α P ,α E , balance various optimization objectives; Update batch switching cost matrix Especially the highest environmental fluctuations and the influence of equipment aging.

9. A chromatographic injection control system based on dynamic programming, characterized in that: The system comprises: A sample attribute acquisition unit is used to acquire sample attributes and construct a characteristic data set of the sample, wherein each sample feature in the characteristic data set includes the sample attribute, and then the attribute of each sample is dimensionally processed by the following standardization process according to different dimensions of the sample characteristic value, and the dimensionally processed characteristic data is mapped to a low-dimensional feature space to obtain a low-dimensional feature representation of the current sample, and then partition calculation based on a dynamic clustering algorithm is performed according to the low-dimensional feature representation to obtain k partitions, and the sample characteristics in each partition are highly similar, and then for each partition, according to the characteristic distribution of the samples in the partition, the most suitable equipment operation parameters for the current partition are allocated; wherein the sample attributes include retention time, separation degree and detection sensitivity; A detection sequence generation unit is used to design a comprehensive optimization objective function based on the low-dimensional feature representation set and the most suitable equipment operation parameters for the current partition to perform dynamic programming within the partition, so as to minimize the comprehensive cost of each pair of samples and generate the optimal detection sequence for each partition; wherein the equipment operation parameters include the optimal column temperature of the partition equipment, the maximum load of the partition equipment and the optimal flow rate of the partition equipment; A global detection sequence generation unit is used to perform global resource scheduling according to the optimal detection sequence of each partition and the most suitable equipment operation parameters for the current partition, and generate a global resource plan, including a sample list and sequence for each batch; wherein the global resource plan satisfies the constraints of the maximum load of the partition equipment and the constraints of minimizing the batch switching cost; The global detection sequence optimization unit is used to collect the real-time operating status and environmental parameters of the equipment, calculate the equipment state deviation and the environmental state deviation, and respectively measure the degree of deviation between the current operating state and the state required by the scheduling plan. If the current equipment state deviation or environmental state deviation exceeds the respective preset thresholds, dynamic optimization is triggered to generate an optimized global resource plan; wherein, the dynamic optimization specifically includes: Design a new target optimization function for global resource planning based on the current equipment state deviation and environment state deviation; Recalculate the batch switching cost based on the real-time operating status of the equipment; Determine the batch adjustment window, including the current batch and the next h batches; Using a rolling optimization algorithm, the batch order is reordered within the window to minimize the objective optimization function of the new global resource plan; Output the optimized sub-plan; Replace the optimized sub-plan into the global resource plan to form an optimized global resource plan: If the batch covered by the window has been completed, monitor the equipment status of the subsequent batches; otherwise, dynamic optimization continues to be executed in a rolling manner; The sampling optimization unit is used to perform sample testing batch by batch according to the optimized global resource plan, collect feedback data from the equipment, calculate the deviation between the actual result and the scheduling plan target, analyze the optimization performance evaluation results based on the deviation of all batches, and adjust the target optimization function and scheduling strategy according to the optimization performance evaluation results.

Images