Methods, apparatus, and electronic equipment for assisting in process optimization

Abstract

This application relates to a method, apparatus, and electronic device for assisting in process optimization. The method includes: obtaining a condition parameter space and an optimization objective; sampling the condition parameter space to obtain multiple sampled condition points; predicting the multiple sampled condition points using a preset model to obtain an estimate of the optimization objective for each sampled condition point; processing the estimated optimization objective for each sampled condition point using a preset acquisition function to determine recommended condition points from the multiple sampled condition points; and outputting the recommended condition points. By automatically generating recommended process condition parameter values ​​through Bayesian optimization, the process optimization process can be guided, shortening the optimization cycle, obtaining better process conditions with fewer experiments, saving material costs during optimization, and reducing reliance on the prior knowledge of technical personnel.

Description

Technical Field

[0001] This application relates to the field of chemical production technology, and in particular to a method, apparatus and electronic equipment for assisting in process optimization. Background Technology

[0002] Chemical production typically requires process optimization before commencement. Optimization goals include increasing yield, purity, and reducing costs. The focus of optimization is on process conditions, including the type and amount of raw materials, reaction time, temperature, and catalyst type. A typical process optimization involves initial synthesis process development, followed by multiple rounds of condition optimization in both laboratory and production environments with varying degrees of detail, ultimately yielding process conditions that meet the optimization objectives. This optimization process often requires extensive trial and error, consuming significant resources, manpower, and time. Furthermore, condition exploration relies heavily on the knowledge and experience of technical personnel, incurring substantial time investment in data retrieval and analysis, resulting in a high learning curve.

[0003] Design of experiments (DoE) can systematically complete single-objective or multi-objective optimization, but this method requires exhaustively exploring the combinations of conditions within a limited search space to give a final recommendation. The overall optimization process has a long cycle and high material and labor costs. Summary of the Invention

[0004] To address or partially address the problems existing in related technologies, this application provides a method, apparatus, and electronic device for assisting in process optimization. By automatically generating recommended process condition parameter values ​​through Bayesian optimization to guide the process optimization process, the optimization cycle can be shortened, better process conditions can be obtained with fewer experiments, material costs consumed during the optimization process can be saved, and reliance on the prior knowledge of technical personnel can be reduced.

[0005] The first aspect of this application provides a method for assisting in process optimization. The method includes: obtaining a condition parameter space and an optimization objective, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters; sampling the condition parameter space to obtain multiple sampled condition points, wherein a set of values ​​for the process condition parameters represents one sampled condition point; predicting the multiple sampled condition points using a preset model to obtain an estimate of the optimization objective for each sampled condition point; processing the estimate of the optimization objective for each sampled condition point using a preset acquisition function to determine a recommended condition point from the multiple sampled condition points; and outputting the recommended condition point.

[0006] A second aspect of this application provides an apparatus for assisting in process optimization. The apparatus includes: an acquisition module configured to acquire a condition parameter space and an optimization objective, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters; a sampling module configured to sample the condition parameter space to obtain multiple sampling condition points, wherein a set of values ​​for the process condition parameters represents one sampling condition point; a prediction module configured to predict the multiple sampling condition points using a preset model to obtain an estimate of the optimization objective for each sampling condition point; a recommendation module configured to process the estimate of the optimization objective for each sampling condition using a preset acquisition function to determine recommended condition points from the multiple sampling condition points; and an output module configured to output the recommended condition points.

[0007] A third aspect of this application provides an electronic device, including a processor and a memory, wherein executable code is stored in the memory, and when the executable code is executed by the processor, the processor performs the method described above.

[0008] A fourth aspect of this application also provides a computer-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the above-described method.

[0009] The fifth aspect of this application also provides a computer program product including executable code that, when executed by a processor, implements the above-described method.

[0010] The method for assisting in process optimization provided in this application automatically generates recommended process condition parameter values ​​through Bayesian optimization, guiding the process optimization process. This can shorten the process optimization cycle, obtain better process conditions with fewer experiments, save material costs consumed in the process optimization process, and reduce reliance on the prior knowledge of technical personnel.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0012] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0013] Figure 1 The illustration shows a schematic diagram of an application scenario of the method according to an embodiment of this application;

[0014] Figure 2 A flowchart illustrating a method for assisting in process optimization according to an embodiment of this application is shown schematically;

[0015] Figure 3 A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically;

[0016] Figure 4 A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically;

[0017] Figure 5 A yield distribution diagram according to an embodiment of this application is illustrated schematically;

[0018] Figure 6 A schematic diagram illustrating the optimization process according to an embodiment of this application is shown.

[0019] Figure 7A A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically;

[0020] Figure 7B A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically;

[0021] Figure 8 This schematic diagram illustrates a structural block diagram of an apparatus for assisting in process optimization according to an embodiment of this application;

[0022] Figure 9 A schematic block diagram of an electronic device implementing an embodiment of this application is shown. Detailed Implementation

[0023] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0024] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The terms "comprising," "including," etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0025] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0026] It should be understood that although the terms "first," "second," "third," etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0027] Bayesian optimization is an active optimization algorithm that, based on the fitting results of machine learning, selects the sample points with the greatest potential to become the global optimum for exploration, and locates and converges to the global optimum with as few samplings as possible.

[0028] This application improves the process optimization process using Bayesian optimization. Based on historical experimental conditions and results obtained during the optimization process, it dynamically recommends the exploration conditions to be selected for subsequent optimization steps. Through algorithm iteration, it continuously obtains feedback on results and dynamically adjusts the optimization direction, quickly converging to the globally optimal conditions and shortening the overall cycle of process optimization.

[0029] The following will describe in detail, with reference to the accompanying drawings, a method and apparatus for assisting in process optimization according to an embodiment of this application.

[0030] Figure 1 This illustration schematically depicts an exemplary system architecture applicable to methods for assisting in process optimization according to embodiments of this application. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not mean that the embodiments of this application cannot be used in other devices, systems, environments or scenarios.

[0031] See Figure 1 The system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0032] Users can use terminal devices 101, 102, and 103 to interact with other terminal devices and server 105 via network 104 to receive or send information. Terminal devices 101, 102, and 103 can be equipped with various communication client applications, such as process design applications, material design applications, web browser applications, database applications, search applications, instant messaging tools, email clients, social media platform software, and so on.

[0033] Terminal devices 101, 102, and 103 include, but are not limited to, smart desktop computers, tablet computers, laptop computers, and other electronic devices that can support functions such as internet access, modeling, analysis and calculation, and design.

[0034] Server 105 can receive conditional parameter space and optimization objectives, and perform functions such as model prediction and preset evaluation function calculation. It can also send recommendation results to terminal devices 101, 102, and 103. For example, server 105 can be a backend management server, a server cluster, or a cloud server.

[0035] It should be noted that the number of terminal devices, networks, and servers is merely illustrative. Depending on the implementation requirements, any number of terminal devices, networks, and cloud components can be included.

[0036] Figure 2 A flowchart illustrating a method for assisting in process optimization according to an embodiment of this application is shown schematically.

[0037] like Figure 2 As shown, the method includes operations S210-S250.

[0038] In operation S210, the condition parameter space and optimization objective are obtained, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters.

[0039] In operation S220, the condition parameter space is sampled to obtain multiple sampling condition points, wherein a set of values ​​of the process condition parameters represents a sampling condition point.

[0040] In operation S230, multiple sampling condition points are predicted using a preset model to obtain an estimate of the optimization target for each sampling condition point.

[0041] In operation S240, the optimization target of each sampling condition is estimated by a preset acquisition function, and the recommended condition point is determined from multiple sampling condition points.

[0042] When operating S250, output the recommended condition points.

[0043] According to embodiments of this application, users can define a condition parameter space based on actual needs. The condition parameter space includes multiple sets of values ​​for process condition parameters, such as all possible values. For example, for a process synthesizing compound C from compound A and compound B, the process condition parameters may include the ratio of compounds A and B, reaction temperature, reaction time, catalyst type, etc. As an example, a condition parameter space can be defined as shown in Table 1.

[0044] Process condition parameters Range of values The ratio of compounds A and B [0.20,5.0] reaction temperature [60.0, 100] (degrees Celsius) reaction time [40.0, 100] (minutes) Catalyst types {D,E,F}

[0045] Among them, a set of process condition parameters includes the values ​​of the proportion of compound A and compound B, reaction temperature, reaction time and catalyst type. At least one parameter in any two sets of process condition parameters is different.

[0046] According to embodiments of this application, users can also define optimization objectives, which may include, for example, yield, purity, waste liquid rate, and production cost. The selection of optimization objectives can be set according to actual needs; users can choose one optimization objective or multiple optimization objectives simultaneously. For example, in a process optimization process, three optimization objectives—yield, purity, and energy consumption—can be selected. It is understood that optimization objectives are not limited to the above-mentioned types; any numerical optimization objective in the actual production process is supported and is not limited here.

[0047] It is important to note that the optimization objective here does not include the expected value of the optimization objective. However, as an optional implementation, the expected value of the optimization objective can be defined and used as an automatic termination condition for determining the optimization. For example, if the optimization objective is yield, the expected value can be set to 70%. In this way, when the yield reaches 70% or higher, the optimization process will end directly.

[0048] According to embodiments of this application, multiple sampling condition points in the condition parameter space can be predicted by a preset model to obtain an estimate of the optimization target for each sampling condition point.

[0049] According to embodiments of this application, the preset model may include a machine learning model, such as an extreme tree model, a random forest model, or a Gaussian process, with the extreme tree model being preferred in this application.

[0050] According to embodiments of this application, sampling condition points are combinations of process condition parameters obtained by sampling in a condition parameter space. Taking the condition parameter space shown in Table 1 as an example, a sampling condition point can be represented as {1.0, 80.0, 60.0, D}, which means that the ratio of compounds A and B is 1:1, the reaction temperature is 80.0 degrees Celsius, the reaction time is 60.0 minutes, and catalyst D is used. In this embodiment, a large number of sampling condition points can be selected, for example, 5000 sampling condition points can be selected in each round, thereby allowing estimation of a large number of process condition combinations. The total number of samples can be set. Random uniform sampling or other sampling strategies can be selected.

[0051] According to embodiments of this application, the estimate can be represented as the mean and standard deviation, or, as an equivalent solution, variance can be used instead of standard deviation to represent uncertainty. A preset model can process each sampling condition point to obtain an estimate of the optimization objective for that sampling condition point. For example, for a given sampling condition point, the preset model can estimate the mean of the optimization objective (such as yield) for that sampling condition point as μ and the standard deviation as σ based on existing empirical data (i.e., model parameters).

[0052] According to embodiments of this application, the estimation of the optimization objective for each sampling condition can be processed by a preset acquisition function, and a recommended condition point can be determined from multiple sampling condition points.

[0053] In Bayesian optimization, the acquisition function reflects the potential benefit of adding new sample points at a given location to the optimization process; that is, locations with high uncertainty or high estimated values ​​are more likely to be recommended by the acquisition function. According to embodiments of this application, the acquisition function may include at least one of the following: Expected Improvement (EI), log EI, Probability of Improvement (PI), Lower Confidence Bound (LCB), Upper Confidence Bound (UCB), or Thompson sampling. This application preferably uses EI.

[0054] The specific model and sampling function chosen in Bayesian optimization are determined by the characteristics of the process optimization scenario, such as: diverse numerical types of conditions; non-monotonic and non-smooth functional relationships between conditions and results, and the potential for sharp drops; highly sparse optimal points within the condition space; high cost of obtaining single-point results for each experiment (usually 1-2 days); and limited iteration cycles (generally no more than a few dozen). Based on these characteristics, the inventors found that the extreme tree ensemble learning model (e.g., decision tree) combined with the sampling function EI performs best, rather than the Gaussian process suitable for continuous numerical values. Furthermore, Thompson sampling, which has higher randomness and exploration costs, was not chosen as the sampling function.

[0055] It is understandable that the preset acquisition function can also be an evaluation algorithm set by the user, and this is not the only limitation.

[0056] According to an embodiment of this application, the estimation of the optimization objective for each sampling condition point is processed by a preset acquisition function, and recommended condition points are determined from multiple sampling condition points. This includes: evaluating the estimation of the optimization objective for each sampling condition point using the preset acquisition function to obtain an evaluation result for each sampling condition point; and determining the sampling condition points whose evaluation results satisfy a first preset condition as recommended condition points.

[0057] For example, when EI is selected, the EI value of each sampling condition point can be calculated, and all sampling condition points can be sorted according to their EI values. The top N sampling condition points are then selected as recommended condition points, and these recommended condition points are output to guide technicians in conducting further experiments. Here, N is a positive integer, for example, N=3. That is, the first preset condition is the three sampling condition points whose EI values ​​are ranked from highest to lowest. The output method may include output through a device such as a display or speaker. It is understood that the first preset condition can also be the sampling condition points whose EI values ​​rank in the top M%, where M is a positive number, for example, M can be 10%, 20%, 25%, or other values. The first preset condition can also be all sampling condition points whose EI values ​​are greater than a preset value; this is not a unique limitation.

[0058] The following explanation uses EI as an example. In this embodiment, the optimization objective is the yield.

[0059] The formula for calculating EI is:

[0060]

[0061]

[0062] Where x refers to the sampling condition point. f(x) + ) refers to the current best yield value, x +This refers to the process conditions for obtaining this optimal value. Ф(z) refers to the values ​​of f(x) from -∞ on a standard normal distribution with mean μ(x) and standard deviation σ(x). + The integral of ) . σ in z is σ(x).

[0063] According to the embodiments of this application, the EI value of each sampling point can be calculated based on the mean μ and standard deviation σ obtained from the preset model, and all the obtained EI values ​​are sorted from high to low, and the process conditions with the highest sampling ranking are recommended to the technicians.

[0064] According to an embodiment of this application, after obtaining recommended condition points sorted by EI value from high to low, a technician can design experiments based on the process condition parameter values ​​of the recommended condition points to obtain the target values ​​of the corresponding optimization objectives, such as the yield, purity, and waste liquid rate under the given process condition parameter values. If the target value of the optimization objective is satisfactory, the optimization process is complete; if the target value is not satisfactory, the above steps can be repeated until the result is satisfactory.

[0065] This application embodiment automatically generates recommended process condition parameter values ​​through Bayesian optimization, guiding the process optimization process. This can shorten the process optimization cycle, obtain better process conditions with fewer experiments, save material costs consumed in the process optimization process, and reduce reliance on the prior knowledge of technical personnel.

[0066] Figure 3 A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically.

[0067] like Figure 3 As shown, the method is in Figure 2 Based on the illustrated embodiment, operations S310 and S320 may also be included. Specifically, operations S310 and S320 may be located between operations S210 and S220.

[0068] In operation S310, initial training samples are obtained, which include pre-acquired historical condition points and the target values ​​of the optimization objectives corresponding to those historical condition points.

[0069] In operation S320, the initial model is trained using initial training samples to obtain the preset model.

[0070] According to embodiments of this application, training samples are combinations of process condition parameters labeled with target values ​​for the optimization objective, including the parameter values ​​of the process condition parameters and the target value of the optimization objective under those parameter values. Taking the optimization objective as yield as an example, training samples may include combinations of process condition parameter values ​​x1, corresponding to a yield of y1; combinations of process condition parameter values ​​x2, corresponding to a yield of y2; and combinations of process condition parameter values ​​x3, corresponding to a yield of y3. For ease of explanation, x1, x2, and x3 are referred to as parameter points, the form of which is described above in the description of sampling condition points and will not be repeated here. y1, y2, and y3 are the labels corresponding to parameter points x1, x2, and x3. A set of training samples may include one or more parameter points and labels corresponding to those parameter points(s). The initial model may be an initial extreme tree model.

[0071] According to an embodiment of this application, a preset model can be trained using training samples to obtain an updated preset model. After performing operation S240, operation S310 can be returned to perform the next round of calculations.

[0072] Before the first prediction, the model can be initially trained using training samples to adjust the model parameters. Ideally, there should be at least three historical condition points. These historical condition points can be parameter points from experiments already conducted, parameter points selected and obtained by technicians based on their experience, or parameter points randomly generated in the condition parameter space and obtained through experiments.

[0073] In other embodiments, incremental training samples can be obtained, which include recommended condition points and the target values ​​of the optimization objectives corresponding to the recommended condition points. According to embodiments of this disclosure, initial training samples may not be input. The first prediction is performed using random model parameters. After obtaining recommended condition points, technicians can design experiments based on the values ​​of the process condition parameters of the recommended condition points to obtain the target values ​​of the corresponding optimization objectives, thereby constructing training samples to train the model, and then performing the next round of calculations. The training samples can be newly obtained recommended condition points and their target values, or a combination of recommended condition points and their target values ​​with the training samples from the previous round. Furthermore, the training samples may also include condition points and their target values ​​obtained by technicians through experimental verification or literature collection; these condition points are different from the recommended condition points mentioned above.

[0074] According to embodiments of this disclosure, recommended condition points are generated during each iteration. Technicians can design experiments based on these recommended condition points to obtain target values ​​for the optimization objective under more combinations of process condition parameter values. These results can be used to further train a pre-defined model. Technicians can conduct experiments according to the recommended condition points, or they can choose other condition points to conduct experiments, as long as the experimental conditions are within the condition parameter space and the target value of the optimization objective can be obtained.

[0075] According to embodiments of this application, by training and updating a preset model during the optimization process, the accuracy of estimating the optimization target of the predicted sampling condition points can be continuously improved, the process optimization cycle can be further shortened, better process conditions can be obtained with fewer experiments, and material costs consumed during the process optimization process can be saved, while reducing the reliance on the prior knowledge of technical personnel.

[0076] According to embodiments of this application, before training an initial model with initial training samples, the method may further include a verification operation, the content of which may include one or more of the following.

[0077] (1) Verify whether the condition parameter space is set correctly, which may include, but is not limited to, verifying whether the upper and lower boundaries and optional parameters of the condition parameter space are accurately defined. Different parameter types can be set for different process parameters, corresponding to different verification methods.

[0078] For example, for numerical process parameters, upper and lower boundaries and / or step sizes can be defined. For instance, for a floating-point process parameter with a value range of [0, 10], during verification, the lower boundary 0 can be checked to see if it is less than or equal to the upper boundary 10. Similarly, for a process parameter with a value range of [0, 10] and a step size of 2, in addition to checking if the lower boundary 0 is less than or equal to the upper boundary 10, the verification can also check if the difference between the upper boundary 10 and the lower boundary 0 is divisible by 2.

[0079] For optional process parameters, such as {a,b,c}, you can check if the list is empty.

[0080] The parameter data type settings in the condition parameter space of this application embodiment are fully adapted to the types of conditions that may occur in process optimization. For example, the reagent type is set as a discrete value, the reagent dosage or temperature is set as a continuous value, and the stirring speed level is set as a sequential sequence.

[0081] (2) Verify whether the historical parameter values ​​of the process condition parameters of the initial training sample are within the condition parameter space. If the historical parameter values ​​of the initial training sample are not within the condition parameter space, the parameter point can be deleted.

[0082] (3) Verify whether the optimized parameter configuration is suitable for the requirements of the initial model. The optimized parameter configuration may include, but is not limited to, one or more of the following: the selection of the preset acquisition function, the number of sampling condition points, the number of recommended condition points, and the random number seed.

[0083] According to embodiments of this application, the condition parameter space can also be adjusted after each iteration. That is, the method further includes adjusting the condition parameter space in response to receiving a space adjustment instruction.

[0084] For example, if satisfactory results cannot be obtained using the existing condition parameter space, or if new condition values ​​are available (e.g., a new solvent is purchased), the condition parameter space can be expanded. Conversely, if there is a specific local region to explore, or some existing condition values ​​are no longer applicable (e.g., a certain catalyst is unavailable), the condition parameter space can be narrowed. Alternatively, a new solvent can replace the existing solvent.

[0085] The embodiments of this application support the real-time adjustment (enlargement, reduction, replacement, etc.) of the condition parameter space to meet the needs of technicians in process optimization scenarios. This invention can quickly converge to the globally optimal process conditions within a variable condition parameter space, shortening the overall process optimization cycle and thus saving time and material costs.

[0086]

Example

[0087] [Testing Method]

[0088] The inventors used data from the paper "Nanoscale High-Throughput Chemistry for the Synthesis of Complex Molecules" (authors Alexander Buitrago Santanilla et al.) published in Science Journal, Volume 347, Issue 6217, for their experiments.

[0089] This paper focuses on optimizing the process conditions for the coupling reaction of 3-bromopyridine. It tested the process conditions that yielded the highest yield among 16 × 16 × 6 = 1536 combinations of 16 nucleophilic substrates, 6 organic bases, and 16 Pd catalysts. The substrate amount was 0.1 μmol, the solvent was DMSO, the molar ratio of catalyst to substrate was 0.1, the molar ratio of base to substrate was 2, and the temperature was room temperature. All experiments were performed in 1536-well plates. After 22 hours of reaction, ultra-high performance liquid chromatography (UPLC) was used for chromatographic analysis. The ratio of the product LC peak area to the internal standard LC peak area (Pd / IS, i.e., product / internal standard) was used to measure the relative yield of different reactions.

[0090] In this embodiment, experimental data from the paper are used for simulation to test the optimization effect of the implementation method of this application.

[0091] Figure 4 A flowchart illustrating a method for assisting in process optimization according to an embodiment of this application is shown schematically.

[0092] like Figure 4 As shown, the method includes operations S401-S410.

[0093] In operation S401, input the conditional parameter space and optimization objective, input the starting parameter points, and input the optimization parameter configuration. The optimization parameter configuration can include selecting which initial model to use, which preset evaluation function to choose, the number of sampling points, the number of recommendations per round, and the random number seed, etc.

[0094] In this embodiment, the condition parameter space consists of 1536 combinations of process condition parameters, comprising 16 nucleophilic substrates, 6 organic bases, and 16 Pd catalysts; the optimization objective is yield; 7 combinations of process condition parameters are randomly generated as starting parameter points; the initial model is an extreme tree model; the preset evaluation function is EI; the number of sampling points is 5000 (in this embodiment, this can ensure all parameter points in the condition parameter space are predicted in each round; it can also be set to less than 1536); 3 recommended condition points are recommended in each round; the random number seed is 0; and the hyperparameters of the extreme tree model all use default values. Furthermore, since this embodiment uses known experimental data for testing, a maximum number of iterations can be set, for example, 10, so that the system runs for at most 10 rounds before ending the optimization.

[0095] In operation S402, it is determined whether the optimization objective has been achieved as expected. Similarly, since this embodiment uses known experimental data for testing, it can automatically determine whether the optimization objective has been achieved as expected during each iteration, thereby deciding whether to end the optimization process. However, since the first iteration will inevitably fail to achieve the expected result, this operation can be skipped. In real-world scenarios, technicians can proactively determine whether the optimization objective has been achieved as expected and whether to execute the next round of optimization.

[0096] In operation S403, verify whether the parameter values ​​of the process condition parameters of the training samples are all within the condition parameter space.

[0097] The seven initial parameter points mentioned above can be used as initial training samples.

[0098] In operation S404, an initial model is trained based on training samples to update the model parameters and obtain the preset model.

[0099] Specifically, the above 7 starting parameter points can be used as initial training samples to train the initial extreme tree model, and the trained preset model can be used for subsequent predictions.

[0100] In operation S405, sampling is performed in the condition parameter space to obtain the sampling condition points.

[0101] In operation S406, the sampling condition points are predicted using a preset model to obtain the mean and standard deviation of the optimization target for each sampling condition point.

[0102] In operation S407, the preset acquisition function EI is used to calculate the EI value of each sampling condition point, and a batch of sampling condition points with larger EI values ​​are used as recommended condition points.

[0103] In operation S408, the technical personnel obtain the target values ​​of the corresponding optimization objectives by conducting experiments on the recommended condition points. Since this embodiment uses known experimental data for testing, the corresponding yield results from the paper can be directly retrieved.

[0104] In operation S409, the condition parameter space is adjusted according to the strategy. This operation is optional; in this embodiment, the condition parameter space is not adjusted.

[0105] In operation S410, the training samples are updated. Specifically, the initial training samples consisting of the 7 starting parameter points can be updated using the recommended condition points.

[0106] In operation S408, if the target value of the optimization objective corresponding to all recommended condition points does not meet the expected optimization result, the condition parameter space can be adjusted and the training samples updated for the next iteration. The above operation is repeated until there is a condition point among the recommended condition points whose target value meets the expected optimization result, then the iteration can end.

[0107] [Test Results]

[0108] The actual yield distribution provided in the paper is as follows: Figure 5 As shown, the horizontal axis represents Pd / IS, and the vertical axis represents the number of distributions. Statistically, there are 934 data points with Pd / IS = 0, 1507 data points with Pd / IS < 20, and only 29 data points with Pd / IS > 20. The three highest Pd / IS values ​​are 39.81, 28.96, and 27.54, with the highest yield condition combination being approximately one-quarter higher than the second-highest yield condition.

[0109] In this embodiment, the performance of the seven initial parameter points was poor, with Pd / IS values ​​of 4.91, 3.0, 0.8, 0, 0, 0, and 0, respectively. After the first round of prediction, sampling, and recommendation, the Pd / IS values ​​of the three recommended condition points were 15.48, 18.65, and 11.71, respectively. After four rounds of iteration, the optimal combination with a yield of 39.81 was found, and the performance of each iteration is as follows: Figure 6 The line connecting the points represents the highest Pd / Is value found in the current round. It can be seen that in the first iteration, the algorithm can recommend a combination of production conditions with Pd / IS = 18.65 based on the data performance, and by the 4th round (i.e., a total of 12 experiments), the optimal production combination has been found.

[0110] The method for assisting in process optimization provided in this application shortens the overall process optimization cycle and obtains a better combination of process conditions with fewer iterations. Furthermore, this method helps save material costs in the process of finding the optimal combination of conditions, reduces the reliance on prior knowledge for technicians in finding the optimal combination of conditions, and can provide suggestions to technicians through a preset model to guide their next steps.

[0111] Figure 7A A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically.

[0112] like Figure 7A As shown, the method is in Figure 2 Based on the illustrated embodiment, operations S710 to S730 are also included.

[0113] In operation S710, obtain the target estimate of the optimization objective corresponding to the recommended condition point.

[0114] In operation S720, recommended condition points whose target estimated value satisfies the second preset condition are selected as candidate condition points.

[0115] When operating the S730, candidate condition points are output.

[0116] According to an embodiment of this application, operation S250 can output recommended condition points to a subsequent processing unit to execute the above operations S710 to S730, and finally output candidate condition points to the user.

[0117] According to embodiments of this application, the target estimate of the optimization objective can be determined using the estimate of the optimization objective obtained in operation S230. For example, if the estimate of the optimization objective for a certain recommended condition point is the mean μ, the standard deviation σ can be disregarded, and this mean can be used as the target estimate.

[0118] According to embodiments of this application, candidate condition points can be, for example, several recommended condition points with higher target estimates. For instance, if there are 5 recommended condition points, the target estimates of the optimization objectives of the multiple recommended condition points can be sorted, and the top 3 recommended condition points with higher target estimates can be selected as candidate condition points and output to the user.

[0119] According to the technical solution of the embodiments of this application, the recommended condition points can be further screened by optimizing the target estimate of the target in order to obtain more valuable candidate condition points.

[0120] Figure 7B A flowchart illustrating a method for assisting in process optimization according to another embodiment of this application is shown schematically. Figure 7B As shown, the method is in Figure 2 Based on this, it can also include operations of S740 to S760.

[0121] When operating S740, obtain the target value of the optimization objective corresponding to the recommended condition point.

[0122] In operation S750, it checks whether the target value of the recommended condition point has reached the expected value. If it has not reached the expected value, operation S760 is executed.

[0123] When operating the S760, the preset model is updated using the recommended condition points and their target values, and then the next iteration is performed.

[0124] According to embodiments of this application, the target value is a value obtained by experimenters after conducting actual experiments based on recommended condition points, while the target estimate mentioned above is a value estimated by the system; the two are different concepts. If the target value does not reach the expected value, the recommended condition points and target value generated in this round can be used to update the training samples, and the preset model can be retrained using the new training samples to obtain an updated preset model, thereby continuously optimizing the recommended condition points.

[0125] Still refer to Figure 7B The method may also include operation S770. Operation S770 occurs after operation S750 and before operation S760.

[0126] During operation S770, a space adjustment strategy is obtained, and the condition parameter space is adjusted according to the space adjustment strategy.

[0127] At this point, it is possible to re-verify whether the adjusted conditional parameter space settings are correct, and / or whether the recommended conditional points in the updated training samples are still within the adjusted conditional parameter space.

[0128] According to embodiments of this application, a spatial adjustment strategy can be obtained when the target values ​​of all recommended condition points or the target values ​​of all candidate condition points fail to meet the expected values. For example, the spatial adjustment strategy can be pre-stored in memory, allowing it to be read directly from memory when needed. Alternatively, a prompt can be issued to the user when the target values ​​of all recommended condition points fail to meet the expected values, enabling the user to set a spatial adjustment strategy accordingly. This spatial adjustment strategy typically expands the condition parameter space, allowing selection within a larger space to potentially yield better condition points.

[0129] Another aspect of this application provides an apparatus for assisting in process optimization.

[0130] Figure 8 The diagram illustrates a structural block diagram of an apparatus for assisting in process optimization according to an embodiment of this application.

[0131] like Figure 8 As shown, the device 800 for assisting in process optimization includes: an acquisition module 810, a sampling module 820, a prediction module 830, a recommendation module 840, and an output module 850.

[0132] The acquisition module 810 is configured to acquire a condition parameter space and an optimization objective, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters.

[0133] The sampling module 820 is configured to sample the condition parameter space to obtain multiple sampling condition points, wherein a set of values ​​of the process condition parameters represents a sampling condition point.

[0134] The prediction module 830 is configured to predict multiple sampling condition points using a preset model to obtain an estimate of the optimization target for each sampling condition point.

[0135] The recommendation module 840 is configured to process the estimation of the optimization objective for each sampling condition through a preset acquisition function, and determine the recommended condition point from multiple sampling condition points.

[0136] Output module 850 is configured to output recommended condition points.

[0137] In some embodiments, the device further includes a training module configured to obtain initial training samples and train an initial model using the initial training samples to obtain a preset model. The initial training samples include pre-acquired historical condition points and their corresponding target values ​​for optimization objectives; the historical condition points include historical parameter values ​​of the process condition parameters.

[0138] In some embodiments, the device further includes a verification module configured to perform at least one of the following:

[0139] Verify that the condition parameter space is set correctly; and / or

[0140] Verify whether the historical parameter values ​​of the process conditions of the initial training samples are within the condition parameter space; and / or

[0141] Verify whether the optimized parameter configuration is suitable for the requirements of the initial model. The optimized parameter configuration includes one or more of the following: the selection of the preset acquisition function, the number of sampling condition points, the number of recommended condition points, and the random number seed.

[0142] In some embodiments, the recommendation module is configured to evaluate the estimation of the optimization target for each sampling condition point using a preset acquisition function to obtain an evaluation result for each sampling condition point; and to determine the sampling condition points whose evaluation results satisfy a first preset condition as recommendation condition points.

[0143] In some embodiments, the estimation of the optimization objective includes the mean and standard deviation, and the preset acquisition function is the EI function.

[0144] In some embodiments, the device further includes a candidate condition point recommendation module, configured to obtain the target estimate of the optimization target corresponding to the recommended condition point; select the recommended condition point whose target estimate satisfies a second preset condition as a candidate condition point; and output the candidate condition point.

[0145] In some embodiments, the device further includes a detection module configured to acquire the target value of the optimization target corresponding to the recommended condition point; detect whether the target value of the recommended condition point has reached the expected value; and if the target values ​​of the recommended condition points have not reached the expected value, update the preset model using the recommended condition points and their target values.

[0146] In some embodiments, the device further includes an adjustment module configured to acquire a spatial adjustment strategy and adjust the condition parameter space according to the spatial adjustment strategy.

[0147] Regarding the apparatus 800 for assisting in process optimization in the above embodiments, the specific manner in which each module and unit performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated further here.

[0148] Another aspect of this application provides an electronic device.

[0149] Figure 9 A block diagram schematically illustrates an electronic device implementing an embodiment of this application.

[0150] See Figure 9 The electronic device 900 includes a memory 910 and a processor 920.

[0151] The processor 920 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0152] Memory 910 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. ROM may store static data or instructions required by the processor 920 or other modules of the computer. Permanent storage devices may be read-write storage devices. Permanent storage devices may be non-volatile storage devices that retain stored instructions and data even when the computer is powered off. In some embodiments, permanent storage devices use mass storage devices (e.g., magnetic or optical disks, flash memory) as permanent storage devices. In other embodiments, permanent storage devices may be removable storage devices (e.g., floppy disks, optical drives). System memory may be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory. System memory may store some or all of the instructions and data required by the processor during operation. Furthermore, memory 910 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (e.g., DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 910 may include a removable storage device that is readable and / or writable, such as a laser disc (CD), a read-only digital multifunction optical disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-high-density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves or transient electronic signals transmitted wirelessly or via wired connections.

[0153] The memory 910 stores executable code, which, when processed by the processor 920, can cause the processor 920 to execute part or all of the methods described above.

[0154] Furthermore, the method according to this application can also be implemented as a computer program or computer program product, which includes computer program code instructions for performing some or all of the steps in the method described above.

[0155] Alternatively, this application may be implemented as a computer-readable storage medium (or a non-transitory machine-readable storage medium or a machine-readable storage medium) storing executable code (or computer program or computer instruction code) thereon, which, when executed by a processor of an electronic device (or server, etc.), causes the processor to perform part or all of the steps of the methods described above according to this application.

[0156] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for assisting in optimizing a process, characterized by, The method includes: Obtain a condition parameter space and an optimization objective, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters; the process condition parameters include at least one of the following: feed type, dosage, feed ratio, reaction temperature, reaction time, and catalyst type; and the optimization objective includes at least one of the following: yield, purity, energy consumption, waste liquid rate, and production cost. The condition parameter space is sampled to obtain multiple sampling condition points, wherein a set of values ​​of the process condition parameters represents a sampling condition point; By predicting multiple sampling condition points using a preset model, an estimate of the optimization target for each sampling condition point is obtained; The optimization target for each sampling condition point is estimated by processing it through a preset acquisition function, and a recommended condition point is determined from the plurality of sampling condition points; and Output the recommended condition points; Obtain the target value of the optimization objective corresponding to the recommended condition point. The target value of the optimization objective is obtained through actual experiments based on the process condition parameter values ​​of the corresponding recommended condition point. Detect whether the target value of the recommended condition point has reached the expected value; If the target values ​​of the recommended condition points do not reach the expected values, obtain a spatial adjustment strategy; The condition parameter space is adjusted according to the space adjustment strategy, and the preset model is updated using the recommended condition points and the target values ​​of the recommended condition points in the adjusted condition parameter space. If there is a recommended condition point where the target value reaches the expected value, the operation ends.

2. The method of claim 1, wherein, Before predicting multiple sampling condition points using a preset model, the method further includes: Obtain initial training samples, which include pre-acquired historical condition points and the target values ​​of the optimization objectives corresponding to the historical condition points. The historical condition points include historical parameter values ​​of the process condition parameters. The initial model is trained using the initial training samples to obtain the preset model.

3. The method according to claim 2, characterized in that, Before training the initial model using the initial training samples, the method further includes at least one of the following operations: Verify that the settings of the condition parameter space are correct; Verify whether the historical parameter values ​​of the process condition parameters of the initial training sample are within the condition parameter space; Verify whether the optimized parameter configuration is suitable for the requirements of the initial model. The optimized parameter configuration includes one or more of the following: selection of preset acquisition function, number of sampling condition points, number of recommended condition points, and random number seed.

4. The method according to claim 1, characterized in that, The step of processing the estimation of the optimization target for each sampling condition point through a preset acquisition function, and determining the recommended condition point from the plurality of sampling condition points, includes: The estimation of the optimization target for each sampling condition point is evaluated using a preset acquisition function to obtain the evaluation result for each sampling condition point; The sampling condition points whose evaluation results meet the first preset conditions are determined as recommended condition points.

5. The method according to claim 4, characterized in that, The step of predicting multiple sampling condition points using a preset model to obtain an estimate of the optimization objective for each sampling condition point includes: The mean and standard deviation of the optimization objective for each sampling condition point are obtained by predicting multiple sampling condition points using a preset model. The step of evaluating the estimation of the optimization objective for each sampling condition point using a preset acquisition function to obtain the evaluation result for each sampling condition point includes: The EI value for each sampling condition point is obtained by calculating the mean and standard deviation of the optimization target using a preset acquisition function EI.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Obtain the target estimate of the optimization objective corresponding to the recommended condition point; Recommended condition points whose target estimated values ​​satisfy the second preset condition are selected as candidate condition points; Output the candidate condition points.

7. An apparatus for assisting in process optimization, characterized in that, The device includes: The acquisition module is configured to acquire a condition parameter space and an optimization objective, wherein the condition parameter space includes multiple sets of values ​​for process condition parameters; the process condition parameters include at least one of the following: feed type, dosage, feed ratio, reaction temperature, reaction time, and catalyst type; and the optimization objective includes at least one of the following: yield, purity, energy consumption, waste liquid rate, and production cost. The sampling module is configured to sample the condition parameter space to obtain multiple sampling condition points, wherein a set of values ​​of the process condition parameters represents a sampling condition point; The prediction module is configured to predict multiple sampling condition points using a preset model to obtain an estimate of the optimization target for each sampling condition point; The recommendation module is configured to process the estimation of the optimization target for each of the sampling condition points using a preset acquisition function, and determine recommendation condition points from the plurality of sampling condition points; and The output module is configured to output the recommended condition points; The detection module is configured to obtain the target value of the optimization target corresponding to the recommended condition point, wherein the target value of the optimization target is obtained through actual experiments based on the process condition parameter values ​​of the corresponding recommended condition point; detect whether the target value of the recommended condition point has reached the expected value; if the target values ​​of the recommended condition points have not reached the expected value, update the preset model using the recommended condition points and their target values; if there is a recommended condition point whose target value has reached the expected value, then the operation ends. The adjustment module is configured to acquire a spatial adjustment strategy and adjust the condition parameter space according to the spatial adjustment strategy. When the detection module performs the step of updating the preset model using the recommended condition points and the target values ​​of the recommended condition points, it is configured to update the preset model using the recommended condition points and the target values ​​of the recommended condition points in the adjusted condition parameter space.

8. An electronic device, characterized in that, include: processor; as well as A memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores executable code that, when executed by a processor of an electronic device, causes the processor to perform the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes executable code, which, when executed by a processor, implements the method according to any one of claims 1-6.

Images