CN-121999919-A - Method for optimizing anthraquinone extraction process parameters of semen cassiae based on machine learning

CN121999919ACN 121999919 ACN121999919 ACN 121999919ACN-121999919-A

Abstract

The invention discloses a method for optimizing anthraquinone extraction process parameters of semen cassiae based on machine learning, in particular to the field of medicinal material extraction processes, which is used for solving the problem that the same group of extraction process parameters are difficult to stably multiplex due to fluctuation of semen cassiae raw materials in different batches; the method comprises the steps of taking a dry radical free anthraquinone content index as a unique batch characterization parameter, searching a batch similar sample set to generate an initial parameter set, extracting a short-time probe under the initial parameter set, collecting a primary stage sample liquid, calculating a dry radical release and a hetero-peak ratio, comprehensively analyzing to obtain a net release coefficient, generating an optimizing mode mark according to the net release coefficient to limit a process window and an operable rule strength, then training a regression prediction model in the limited window to output a target parameter set, determining an executable target parameter set, and realizing stability and reusability of process parameter optimization.

Inventors

ZHANG YONGLI
An Xuanyu
KOU YUFENG
HAO MENGEN

Assignees

三原利华生物技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. The method for optimizing the anthraquinone extraction process parameters of the cassia seeds based on machine learning is characterized by comprising the following steps: S1, completing free anthraquinone measurement and water-containing state conversion based on the same check sample entering a factory, and generating dry radical free anthraquinone content indexes according to the existing conversion caliber; S2, screening a batch similar sample set from a historical process sample record by using a dry radical free anthraquinone content index, extracting an initial parameter set corresponding to a response surface reference extraction condition from the batch similar sample set, and setting process conditions except for solvent concentration, extraction temperature and extraction time as fixed conditions; S3, extracting and collecting a stage sample liquid once by a short-time probe under an initial parameter set, converting to obtain a dry basis release and impurity peak ratio, comprehensively analyzing to obtain a net release coefficient, and generating an optimizing mode mark according to the net release coefficient to limit a subsequent process window and operable rule intensity; S4, evaluating the effectiveness of the batch similar sample set, removing abnormal records, training a regression prediction model, outputting a target parameter set by using a dry basis free anthraquinone content index and a batch response deviation index in an optimizing mode mark limiting window, and determining an executable target parameter set; s5, performing whole batch extraction by using an executable target parameter set, obtaining batch anthraquinone yield indexes, and simultaneously writing all acquired parameters and indexes into a process sample record library for subsequent batch retrieval and calling.
2. The machine learning based method for optimizing anthraquinone extraction process parameters of semen cassiae according to claim 1, wherein step S1 comprises: And (3) measuring the content of free anthraquinone wet radicals by adopting a high performance liquid chromatography based on the same check sample of the semen cassiae raw material entering the factory, measuring the water content, generating a dry radical free anthraquinone content index through dry matter proportion conversion, and establishing the dry radical free anthraquinone content index as a batch characterization parameter of the current batch.
3. The machine learning based method for optimizing anthraquinone extraction process parameters of semen cassiae according to claim 2, wherein step S2 comprises: and calculating absolute deviation by using the dry radical free anthraquinone content index and the dry radical free anthraquinone content index of each historical batch in the historical process sample records, and collecting all the historical batch records with the absolute deviation smaller than or equal to the similarity threshold as a batch similarity sample set.
4. The machine learning based method for optimizing process parameters for anthraquinone extraction of cassia seed according to claim 3, wherein step S2 further comprises: If the record directly adopting the response surface reference extraction condition exists, extracting an initial parameter set corresponding to the response surface reference extraction condition from a batch of similar sample sets, otherwise, calculating an average value of the solvent concentration, the extraction temperature and the extraction time to generate the initial parameter set, and setting the process conditions except the solvent concentration, the extraction temperature and the extraction time as fixed values corresponding to the response surface reference extraction condition.
5. The machine learning based method for optimizing anthraquinone extraction process parameters of semen cassiae as set forth in claim 4, wherein step S3 includes: short-time probe extraction is carried out by adopting an initial parameter set and combining with fixed process conditions, a stage sample liquid is collected at a preset stage, the caliber is converted according to the concentration of a detection method, the volume of the extracting liquid is combined with the dilution multiple to generate a stage leaching amount, and then the stage leaching amount is converted with the dry radical free anthraquinone content index multiplied by the dry radical theoretical total amount generated by the mass of the fed dry radical to obtain a dry radical release component.
6. The machine learning based method for optimizing process parameters of anthraquinone extraction of cassia seed of claim 5, wherein step S3 further comprises: And (3) completing baseline correction and peak collection on the phase sample liquid chromatogram, determining a target peak set integral to obtain a target peak total area, integrating effective peaks outside the target peak set to obtain a non-target peak total area, and converting the ratio of the non-target peak total area to the target peak total area to obtain a hetero-peak ratio.
7. The machine learning based method for optimizing process parameters of anthraquinone extraction of cassia seed of claim 6, wherein step S3 further comprises: inputting the dry basis release and the peak-to-peak ratio into a comprehensive analysis model trained by a batch of similar sample sets to generate a net release coefficient, and generating an optimizing mode mark according to comparison of the net release coefficient and a preset judging threshold, wherein the optimizing mode mark limits the subsequent optimizing process window and the constraint strength of an operation rule.
8. The machine learning based method for optimizing process parameters for anthraquinone extraction of cassia seed of claim 7, wherein step S4 comprises: And carrying out validity evaluation on the batch similar sample set, removing inconsistent or abnormal historical records through comparing absolute deviation of dry basis release components, absolute deviation of impurity peak ratio and lower limit threshold of batch anthraquinone yield index, reserving sample records to form a purified sample set, and training a regression prediction model based on the purified sample set.
9. The machine learning based method for optimizing process parameters for anthraquinone extraction of cassia seed of claim 8, wherein step S4 further comprises: And fixing the dry radical free anthraquinone content index and the batch response deviation index in a process window defined by the optimizing mode mark, generating a preliminary target parameter set by using regression prediction model iteration test parameter combination, and then obtaining an executable target parameter set by projection adjustment according to the equipment reachable range and the process operational rule.
10. The machine learning based method for optimizing process parameters for anthraquinone extraction of cassia seed of claim 9, wherein step S5 further comprises: And after the extraction is finished, collecting an end point extracting solution sample, generating end point dissolution total amount by combining the extracting solution volume and dilution multiple according to concentration conversion caliber, and then carrying out proportional conversion with dry basis theoretical total amount to obtain a batch anthraquinone yield index, wherein the dry basis free anthraquinone content index, the dry basis release, the peak-to-peak ratio, the net release coefficient, the executable target parameter set and the batch anthraquinone yield index are used as a complete record to be written into a process sample record library.

Description

Method for optimizing anthraquinone extraction process parameters of semen cassiae based on machine learning Technical Field The invention relates to the field of medicinal material extraction processes, in particular to a method for optimizing anthraquinone extraction process parameters of semen cassiae based on machine learning. Background In the process development and industrial production scenario of anthraquinone extraction of semen cassiae, a production unit usually needs to establish a set of executable parameter schemes around key conditions such as solvent system, temperature, extraction time and the like, and hopes that the schemes can be stably reused in subsequent batches. In the prior art, a method for optimizing relevant process conditions by ultrasonic-assisted extraction and response surface test by taking semen cassiae as a raw material has appeared, an idea of obtaining the optimal extraction process by optimization is given in a specification, and reliability of the extraction conditions is emphasized, and the application number is 201310200050. Meanwhile, a process optimization method for the traditional Chinese medicine production process also appears, a prediction evaluation model is built by sampling and evaluating production data and further introducing a machine learning nuclear model, and optimal data points are obtained by iterative optimization in an interested data space, and then a production optimization decision model is formed by training and is used for process control, and the application number is 202211273092.1. However, the above-mentioned optimum extraction process obtained by experimental design is often established on the basis of relatively fixed raw material state and experimental condition, when continuous batch production is entered, natural fluctuation of semen cassiae raw material in the aspects of water-containing state, crushing and screening state, storage and processing difference and the like can change the solvent permeation and mass transfer behavior, so that the same set of temperature, time and solvent condition have effect deviation among different batches, the fluctuation of anthraquinone related index and the processing quality are unstable, on-site trial-and-error type parameter adjustment has to be repeated, and a parameter decision basis which can be reused for a long time and can be interpreted is difficult to form. While the process optimization thought for traditional Chinese medicine production data mining provides searching a feasible region in a complex data space and performing iterative optimization, the process optimization thought is mainly based on exploration and optimal point mining of a historical data space, and if an effective constraint and identification mechanism aiming at the batch difference of raw materials and the action mechanism of process conditions is lacked, the situation that the batch change is seemingly optimal in data and difficult to reproduce is easy to appear, and finally the most critical requirement in the process of extracting the cassia seed anthraquinone cannot be met, namely, stable executable and reusable process parameter optimization results can still be output under natural fluctuation of raw materials and working conditions. In order to solve the above problems, a technical solution is now provided. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, the embodiment of the present invention provides a method for optimizing process parameters of extraction of cassia seed anthraquinone based on machine learning, which uses dry radical free anthraquinone content index as unique batch characterization parameter, searches batch similar sample set to generate initial parameter set, performs short-time probe extraction under the initial parameter set and collects primary stage sample liquid, calculates dry radical release and hetero peak ratio and comprehensively analyzes to obtain net release coefficient, generates optimizing mode mark according to the net release coefficient to limit process window and operable rule intensity, then trains regression prediction model in the limiting window to output target parameter set and determines executable target parameter set, and realizes stability and reusability of process parameter optimization to solve the problems presented in the above background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: S1, completing free anthraquinone measurement and water-containing state conversion based on the same check sample entering a factory, and generating dry radical free anthraquinone content indexes according to the existing conversion caliber; S2, screening a batch similar sample set from a historical process sample record by using a dry radical free anthraquinone content index, extracting an initial parameter set corresponding to a response surface reference extr