Plant extraction process parameter intelligent optimization control method

Abstract

This invention discloses an intelligent optimization and control method for plant extraction process parameters, specifically relating to the field of plant extraction process control. It addresses the problem that existing technologies, relying on fixed quality indicators, cannot accurately reflect the migration and dynamic changes of multiple components at different stages, leading to low accuracy in parameter optimization. The method involves collecting pre-process information of the raw materials to be processed and forming a process observation sequence to construct an initial state profile. Boundary retrieval is performed on the process observation sequence to form a stage state chain. Mirror denoising and quantization calculations are performed on each stage state segment to obtain the target precipitation amount of the denoised dry basis and the non-target carry-out amount of the stripped solid dry basis. The results of the comprehensive analysis of these two factors are correlated with the target quality structure to determine the dominant control direction, forming a stage parameter sequence. Finally, the stage parameter sequence and stage state chain are read to generate an extraction execution instruction set, which is sequentially sent to the extraction equipment to complete the entire batch extraction operation.

Description

Technical Field

[0001] This invention relates to the field of plant extraction process control, and more specifically, to a method for intelligent optimization and control of plant extraction process parameters. Background Technology

[0002] In plant extraction, the release of raw material components, the precipitation of impurities, and changes in the liquid phase state are interdependent. The extraction efficiency is not only affected by predetermined process parameters but also closely related to the transitions between extraction stages and the evolution of the process state. To improve the monitorability of the extraction process, existing technologies have introduced near-infrared spectroscopy, quality index analysis, and data mining methods to determine the extraction endpoint and optimize the process. For example, the existing technical literature "A Quality Control Method for Traditional Chinese Medicine Extraction Process Based on Data Mining" (Publication No.: CN113588590A) discloses a technical solution for determining the extraction endpoint and the optimal process by collecting near-infrared spectra of the extract, selecting quality indicators, establishing the relationship between process parameters and quality indicators, and setting quality indicator evaluation values.

[0003] However, the control basis of such technical solutions still relies on a few predetermined quality indicators and their comprehensive evaluation results, making it difficult to truly reflect the overall state under the combined effects of multi-component migration, impurity release, and process stage changes during plant extraction. Plant extraction is essentially a continuously changing dynamic process, with different components releasing at different rhythms. Existing solutions typically compress complex processes into fixed indicator combinations for ease of modeling and judgment, with the highest evaluation value corresponding to the optimal process. The resulting judgment only corresponds to the indicators already included in the model and cannot necessarily represent the true quality structure of the extract and its adaptability to subsequent processes. This easily leads to discrepancies between the evaluation results and the actual optimal process state, thus limiting the accuracy and applicability of intelligent optimization control of process parameters.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in existing technologies, this invention provides an intelligent optimization and control method for plant extraction process parameters. This method collects the pre-process information of the raw materials in the batch to be processed and forms a process observation sequence to construct an initial state profile. Then, it performs boundary retrieval on the process observation sequence to form a stage state chain. Subsequently, it performs mirror denoising and quantization calculations on each stage state segment to obtain the target precipitation amount of the mirror-denoised dry basis and the non-target carry-out amount of the stripped solid dry basis. These two are then correlated with the target quality structure to determine the dominant control direction and form a stage parameter sequence. Finally, based on the stage parameter sequence and stage state chain, it generates an extraction execution instruction set and sequentially sends it to the extraction equipment to complete the entire batch extraction operation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: S1: Collect the pre-state information of the batch of raw materials to be processed, collect the near-infrared spectrum of the extract, the concentration of the target component, the total solids concentration and the effective extract volume of the observation sequence according to the preset sampling interval, write the pre-state information of the raw materials into the starting position and align it with a unified timestamp, record the initial dry basis feed amount and write it into the initial state file. S2: Read the process observation sequence in the initial state file, perform boundary retrieval based on the concentration and spectral changes at adjacent sampling times, mark the stage boundary point when the change direction changes and the main feature band position switches, and divide and connect according to the boundary point to form a stage state chain; S3: Read the sequence of each segment according to the stage state chain, and after mirror expansion and center back substitution denoising, accumulate and calculate to obtain the target precipitation amount of the mirror denoised dry base and the non-target carry-out amount of the stripped solid dry base. Input the stage discrimination model to output the orientation coefficient, and write the dominant control direction into the stage parameter sequence after the target quality structure is determined. S4: Read the stage parameter sequence and stage state chain, write the first and second corresponding process parameters into the start execution bit, and write the subsequent parameters in sequence to form the extraction execution instruction set, and send them to the extraction equipment in sequence to complete the extraction operation.

[0007] Furthermore, the raw material pre-processing information includes the raw material moisture content, raw material crushing state, and near-infrared characterization information of the raw material before entering the tank. The raw material moisture content, raw material crushing state, and near-infrared characterization information of the raw material before entering the extraction process are collected and recorded.

[0008] Furthermore, the process observation sequence continuously collects near-infrared spectra, target component concentrations, total solids concentrations, and effective extract volumes of the extract at preset sampling intervals. The raw material pre-state information is written into the starting position of the process observation sequence and each sampling item is aligned with a unified timestamp. The initial dry basis feed amount is recorded and written into the initial state file.

[0009] Furthermore, the boundary search uses the changes in target component concentration, total solids concentration, and near-infrared spectrum of the extract between adjacent sampling times as the basis for determining continuity. The boundary search is performed on the process observation sequence moment by moment. When the direction of change changes and the position of the main feature band of the near-infrared spectrum of the extract switches accordingly, the stage boundary point is marked.

[0010] Furthermore, the process observation sequence is divided according to all stage boundary points to obtain multiple stage state segments that are connected end to end and have a fixed time order. Then, the stage state segments are sequentially connected according to their positions in the process observation sequence to form a stage state chain.

[0011] Furthermore, the target component concentration sequence, total solids concentration sequence, and effective extract volume sequence in each stage state segment are read according to the stage state chain. The first-end mirror expansion and the last-end mirror expansion are performed on the target component concentration sequence and the total solids concentration sequence of each stage state segment to obtain the mirror completion sequence. Then, the center substitution is performed to obtain the mirror denoised target component concentration sequence and the mirror denoised total solids concentration sequence.

[0012] Furthermore, the target component concentration sequence of the mirror-denoising method is accumulated item by item according to the sampling interval and the corresponding effective extract volume sequence, and the target precipitation amount of the mirror-denoising dry basis is calculated based on the initial dry basis feed amount. The non-target solid concentration sequence is obtained by subtracting the target component concentration sequence of the mirror-denoising method from the total solid concentration sequence of the mirror-denoising method, and then accumulated. The non-target carry-out amount of the stripped solid dry basis is calculated based on the initial dry basis feed amount.

[0013] Furthermore, the target precipitation amount of the mirror-denoised dry substrate and the non-target carry-out amount of the stripped solid dry substrate are input into the stage discrimination model. The stage discrimination model outputs the stage orientation coefficient. The stage orientation coefficient is then matched with the preset target quality structure to obtain the dominant control direction and written into the stage parameter sequence according to the order of the stage state chain.

[0014] Furthermore, the stage parameter sequence and stage state chain are read, and the process parameters corresponding to the stage state segment at the beginning of the stage state chain are written into the start execution bit. The process parameters corresponding to each stage state segment at the subsequent position of the stage state chain are written into the subsequent execution bits in sequence to form the extraction execution instruction set.

[0015] Furthermore, each execution bit in the extraction execution instruction set contains the extraction temperature, extraction duration, solvent addition amount, and stirring intensity of the corresponding stage state segment. The extraction temperature, extraction duration, solvent addition amount, and stirring intensity corresponding to each execution bit are sequentially sent to the extraction equipment to complete the extraction operation according to the execution bit sequence of the extraction execution instruction set.

[0016] The technical effects and advantages of the intelligent optimization and control method for plant extraction process parameters of this invention are as follows: Compared with existing technologies, this invention constructs an initial state archive containing raw material pre-state information and process observation sequences, achieving unified time-aligned management of raw material initial characteristics and extraction dynamic data, thus avoiding the problem of disconnect between static and dynamic data. It employs a boundary retrieval method based on the direction of concentration change and the position switching of the main spectral feature band to divide the extraction process into different stage state segments, forming a stage state chain, thereby identifying the turning points of multi-component release rhythms. After performing mirror expansion and center substitution denoising on each stage state segment, the cumulative conversion calculation of the dry basis target precipitate and non-target carry-over amounts is performed, and the results are input into a stage discrimination model to generate orientation coefficients. These coefficients are then compared with the target mass structure to determine the dominant control direction, forming a stage parameter sequence arranged in stage order. Finally, the stage parameter sequence is converted into an extraction execution instruction set and sequentially sent to the extraction equipment, allowing the process parameters to automatically adjust as the actual stage evolves. This scheme can reduce the adaptation deviation that may be caused by fixed index compression modeling, which is beneficial to improving the matching between extraction parameters and the mass structure of the extract, and improving the intelligence and operational stability of plant extraction process parameter control. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an intelligent optimization control method for plant extraction process parameters according to the present invention. Figure 2 This is a schematic diagram illustrating the initial state file construction of the present invention; Figure 3 This is a schematic diagram of the boundary retrieval and stage state chain formation of the present invention; Figure 4 This is a schematic diagram of the image denoising and quantization calculation of the present invention; Figure 5 This is a schematic diagram illustrating the stage discrimination, stage parameter sequence generation, and extraction and execution instruction set issuance of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figures 1-5 This invention provides a method for intelligent optimization and control of plant extraction process parameters, comprising: S1: Collect the pre-state information of the batch of raw materials to be processed, collect the near-infrared spectrum of the extract, the concentration of the target component, the total solids concentration and the effective extract volume of the observation sequence according to the preset sampling interval, write the pre-state information of the raw materials into the starting position and align it with a unified timestamp, record the initial dry basis feed amount and write it into the initial state file. S2: Read the process observation sequence in the initial state file, perform boundary retrieval based on the concentration and spectral changes at adjacent sampling times, mark the stage boundary point when the change direction changes and the main feature band position switches, and divide and connect according to the boundary point to form a stage state chain; S3: Read the sequence of each segment according to the stage state chain, and after mirror expansion and center back substitution denoising, accumulate and calculate to obtain the target precipitation amount of the mirror denoised dry base and the non-target carry-out amount of the stripped solid dry base. Input the stage discrimination model to output the orientation coefficient, and write the dominant control direction into the stage parameter sequence after the target quality structure is determined. S4: Read the stage parameter sequence and stage state chain, write the first and second corresponding process parameters into the start execution bit, and write the subsequent parameters in sequence to form the extraction execution instruction set, and send them to the extraction equipment in sequence to complete the extraction operation.

[0020] This invention adopts a phased dynamic optimization approach. First, an initial state file containing raw material pre-state information and process observation sequence is constructed. Then, the observation sequence is delineated and retrieved to form a phased state chain by switching between concentration changes and spectral features. Subsequently, the state segments of each phase are subjected to mirror expansion and center substitution denoising processing to calculate the dry basis target precipitation amount and non-target carry-over amount and input into the phase discrimination model to generate orientation coefficients. These coefficients are then combined with the target quality structure determination to form a phase parameter sequence. Finally, the phase parameter sequence is transformed into an extraction execution instruction set with a fixed order and sequentially sent to the extraction equipment to achieve closed-loop control of intelligent adjustment of process parameters as the extraction stage evolves.

[0021] Firstly, this addresses the problem that existing technologies cannot accurately reflect the overall extraction status due to the lack of initial raw material conditions and asynchronous dynamic process data. In step S1, the characteristics of the raw materials before the batch to be processed enters the extraction process are comprehensively recorded. After extraction begins, near-infrared spectra of the extract, target component concentrations, total solids concentrations, and effective extract volumes are continuously collected at preset sampling intervals to form a process observation sequence. By writing the raw material pre-state information into the sequence start position and aligning each sampling item with a unified timestamp, and simultaneously recording the initial dry basis feed amount and constructing an initial state profile, a complete and unified data starting point is provided for accurately tracking the dynamic evolution of plant extraction.

[0022] S101: Collection of raw material pre-state information.

[0023] Before the batch to be processed enters the extraction process, the raw material pre-state information is collected according to a predetermined procedure to record the initial physical properties and spectral characterization before entering the tank, serving as the starting point for subsequent process evolution analysis. The raw material pre-state information includes at least: the raw material moisture content, the raw material pulverization state, and the near-infrared characterization information before entering the tank. Specifically, the raw material moisture content can be determined using a standard drying and weighing method and expressed as a percentage of moisture by mass; the raw material pulverization state can be determined using sieve analysis and expressed as particle size distribution parameters; the near-infrared characterization information before entering the tank can be obtained by scanning with a near-infrared spectrometer within a preset wavelength range and expressed as absorption spectral curve data. During data collection, it is preferable to first detect the raw material moisture content, then record the pulverization state parameters, and finally perform a near-infrared scan before entering the tank, storing the obtained data uniformly. In one embodiment, when processing a batch of Salvia miltiorrhiza medicinal materials, the raw material moisture content is measured to be 12.5%, the pulverization state is 80% of the raw material passing through a 40-mesh sieve, and the near-infrared characterization information before entering the tank is a spectral absorption peak sequence within the wavelength range of 900-1700 nm. After the data collection is completed, the raw material pre-state information is temporarily stored and written into the starting position of the process observation sequence in subsequent steps to ensure that raw material batch differences are included in a unified monitoring framework.

[0024] S102: Formation of process observation sequences.

[0025] After the extraction process begins, near-infrared spectra, target component concentrations, total solids concentrations, and effective extract volumes of the extract are continuously collected at preset sampling intervals, forming a process observation sequence arranged in chronological order to capture the dynamic evolution of the extraction process. The process observation sequence can be represented as Rk = {tk, Sk, Cg,k, Cs,k, Vk}, where tk represents the timestamp of the k-th sampling moment, Sk represents the near-infrared spectrum of the extract at that moment, Cg,k represents the target component concentration, Cs,k represents the total solids concentration, and Vk represents the effective extract volume. The target component concentration can be determined by high-performance liquid chromatography or chemical titration, the total solids concentration can be determined by evaporation and weighing, and the effective extract volume can be continuously monitored by a flow meter or level sensor. The sampling interval is preferably set to a fixed time unit, such as once every 1 minute or every 5 minutes; each sampling generates a set of quaternary or pentaneous observation data, which is appended to the process observation sequence in chronological order until the extraction is completed. In one embodiment, for a batch of Salvia miltiorrhiza extract, the sampling interval is 5 minutes. The acquisition results show that the characteristic peaks of the near-infrared spectrum of the extract migrate, the concentration of the target component gradually increases from the initial 0.8 mg / mL, the total solids concentration changes synchronously, and the effective extract volume accumulates to a predetermined proportion of the designed volume. Thus, the synchronous change trajectory of multiple parameters can be completely recorded, providing raw time-series data for the segmentation of the state chain in subsequent stages.

[0026] S103: Data alignment and start position writing.

[0027] After the process observation sequence is completed, the raw material pre-state information is written into the starting position of the process observation sequence, and each sampling item is aligned with a unified timestamp. This ensures that each sampling moment corresponds to a unique near-infrared spectrum of the extract, target component concentration, total solids concentration, and effective extract volume, thereby guaranteeing the temporal continuity of the entire sequence. The preferred formula is tk=t0+k. Δt is used for timestamp alignment, where tk represents the timestamp of the k-th sampling moment, t0 represents the initial reference time corresponding to the extraction start moment, k represents the sampling sequence number, and Δt represents the preset sampling interval. For situations with acquisition delays or asynchronous output from multiple sensors, each sampling item can be mapped to the unified sampling grid closest to the current sampling moment; when the time deviation exceeds the preset tolerance, linear interpolation or nearest neighbor completion can be used to complete the alignment. During the write operation, the raw material pre-state information is first written as the first element of the sequence, and then a unified reference timestamp is configured for each sampling point in the process observation sequence. In one embodiment, when the initial reference time is set to the extraction start moment and the sampling interval is 300 seconds, the timestamp corresponding to the 5th sampling moment is t5 = t0 + 5. 300 seconds, or 1500 seconds. Through the alignment process described above, the timing deviation caused by asynchronous sampling can be eliminated, providing a unified timing benchmark for subsequent stage boundary point retrieval.

[0028] S104: Construction of the initial state file.

[0029] After completing the above alignment, record the initial dry basis feed amount of the batch to be processed, and write the raw material pre-state information, process observation sequence, and initial dry basis feed amount into the same data unit to form the initial state file of the batch to be processed, so as to achieve unified management of data structure. The initial dry basis feed amount is denoted as M0, preferably obtained by converting the raw material wet weight and moisture content, for example, M0=Mwet (1-W), where Mwet represents the wet weight of the raw material and W represents the moisture content of the raw material. The initial state profile can be stored using a structured record method, such as a database record, a structured file, or a memory object, and should include at least a raw material pre-state information field, a process observation sequence field, and an initial dry basis feed amount field. In one embodiment, for a batch of plant raw materials, the measured initial dry basis feed amount is 500 kg, which is then packaged with the pre-state information and the process observation sequence to form the initial state profile. This construction method ensures that a complete initial observation dataset for the batch to be processed is established, providing a unified and traceable data starting point for the entire intelligent optimization control process.

[0030] By collecting raw material pre-state information, forming process observation sequences, writing data alignment, and constructing initial state archives in S1, a unified data foundation containing complete pre-state and dynamic observations is formed. The subsequent step S2 can directly read the process observation sequences in the initial state archives to perform boundary retrieval, thereby achieving a natural transition from the initial archives to the stage state chain. It also enables the generation of stage parameter sequences to be based on the real multi-component release rhythm, which helps to improve the coherence and accuracy of intelligent optimization control of plant extraction process parameters.

[0031] By collecting raw material pre-state information in S1, forming process observation sequences, writing timestamps, and constructing initial state archives, a unified initial state archive containing raw material initial characteristics and synchronous observation data of multiple parameters throughout the process is finally generated for each batch to be processed. This archive achieves strict temporal continuity integration of static pre-state and dynamic sampling items.

[0032] After the initial state file is formed, the recorded process observation sequence needs to be structurally divided into stages to identify key turning points in the component release rhythm during the extraction process. Step S2 reads the process observation sequence in the initial state file, using the changes in target component concentration, total solids concentration, and the switching of the main characteristic band position of the near-infrared spectrum of the extract between adjacent sampling times as the basis for continuity determination. The process observation sequence is then divided into stages according to all stage boundary points, ultimately forming a stage state chain composed of multiple sequentially connected stage state segments with a fixed time order.

[0033] S201: Reading the initial state file.

[0034] After the initial state file is prepared, the process observation sequence within it is read as the complete input data source for subsequent boundary retrieval, ensuring that all judgment criteria originate from time-aligned multi-parameter synchronous records. During reading, the corresponding database unit or structured storage unit can be directly accessed, and all sampling records in the process observation sequence are extracted in chronological order. Each sampling moment preferably corresponds to at least one set of near-infrared spectra of the extract, target component concentration, total solids concentration, and effective extract volume data, while maintaining the original acquisition order. In one embodiment, when processing the extraction of a batch of Astragalus membranaceus raw materials, the initial state file is stored in the server-side database; after reading, a process observation sequence covering the entire extraction time is obtained, with continuous timestamps and complete fields for each sampling item. Through the above reading operation, the unique source and consistent structure of the input data can be ensured, laying the foundation for the construction of subsequent continuity judgment criteria.

[0035] S202: Construction of the criteria for determining continuity.

[0036] After the process observation sequence is read, the changes in target component concentration, total solids concentration, and the position of the main characteristic band in the near-infrared spectrum of the extract between adjacent sampling times are used as the criteria for continuity determination, and the changes are calculated time-by-time. Preferably, the target component concentration sequence is denoted as Cg,k, the total solids concentration sequence is denoted as Cs,k, and the position of the main characteristic band in the near-infrared spectrum of the extract after smoothing preprocessing is denoted as Bk, where Bk can be defined as the wavelength position corresponding to the main peak of absorption intensity or the extreme point of the first derivative within a preset wavelength range. The changes between adjacent sampling times can be expressed as ΔCg,k=Cg,k-Cg,k-1, ΔCs,k=Cs,k-Cs,k-1, and ΔBk=Bk-Bk-1, respectively. To reduce misjudgments caused by noise disturbances, concentration change thresholds εg and εs, as well as a characteristic band shift threshold ελ, can be further set. Only when |ΔCg,k|≥εg, |ΔCs,k|≥εs, and |ΔBk|≥ελ is the change at that moment considered significant. In one embodiment, for the process observation sequence of a batch of Salvia miltiorrhiza extraction, at the 15th sampling moment, the concentration difference of the target component changes from negative to positive, the total solids concentration difference reverses synchronously, and the position of the main characteristic band changes beyond a preset threshold. This establishes a quantitative judgment criterion that reflects the dynamic trend reversal of extraction. The concentration change threshold refers to the preset critical value at which the change in the concentration of the target component between two adjacent sampling moments reaches the stage switching judgment requirement. The characteristic band shift threshold refers to the preset critical value at which the shift in the center wavelength or peak position of the spectral characteristic band between two adjacent sampling moments reaches the process state change judgment requirement. The specific values ​​can be determined based on pre-experiments with the same type of plant raw materials, statistical analysis of historical process data, or model training results.

[0037] S203: Boundary retrieval and stage boundary marking.

[0038] After the continuity determination criteria are constructed, boundary searches are performed on the process observation sequence time-by-time. When the direction of change in target component concentration changes, the direction of change in total solids concentration changes, and the position of the main characteristic band of the near-infrared spectrum of the extract changes between consecutive sampling times, the current sampling time is marked as the stage boundary point. The following determination rule is preferably adopted: if ΔCg,k ΔCg,k-1<0, and |ΔCg,k|≥εg, and |ΔCg,k-1|≥εg; simultaneously ΔCs,k If ΔCs,k-1 < 0, and |ΔCs,k| ≥ εs, and |ΔCs,k-1| ≥ εs; and |Bk - Bk-1| ≥ ελ, then the k-th sampling time is marked as the stage boundary point. Here, ΔCg,k represents the difference between the target component concentration at the current sampling time and the target component concentration at the previous time, ΔCs,k represents the difference between the total solids concentration at the current sampling time and the total solids concentration at the previous time, and Bk represents the position of the main characteristic band of the near-infrared spectrum of the extract at the corresponding time. To avoid multiple adjacent sampling points triggering the boundary simultaneously, a minimum interval constraint m can be further set; when the sampling interval between two candidate boundary points is less than m, the candidate point with the higher comprehensive score of the change amplitude or the earlier time position is retained. Thus, the stage boundary point marking can balance sensitivity and stability.

[0039] S204: Stage state segmentation and stage state chain formation.

[0040] After all stage boundary points are marked, the process observation sequence is segmented according to these points, resulting in multiple consecutive stage state segments with a fixed temporal order. These segments are then sequentially connected according to their position within the process observation sequence to form a stage state chain for the batch to be processed. Specifically, data between two adjacent stage boundary points can be defined as a stage state segment; data from the start time to the first stage boundary point constitutes the first stage state segment, and data from the last stage boundary point to the end of extraction constitutes the final stage state segment. Each stage state segment maintains the continuity of its original sampling time and has no temporal overlap or gaps with preceding and following segments. In one embodiment, when the stage boundary points are located at the 28th, 65th, and 120th sampling times, four stage state segments can be obtained and connected in their original temporal order to form a complete stage state chain. Through this processing, the staged structure of the dynamic observation data is reorganized, providing ordered input for subsequent staged data reading, denoising, and discrimination.

[0041] By using S2 to perform boundary retrieval of the process observation sequence, marking of stage boundary points, segmentation of stage state fragments, and sequential connection according to their positions, a stage state chain for the batch to be processed was finally constructed. This chain clearly delineates the boundary characteristics of different stages during the plant extraction process, including the release of target components, the precipitation of impurities, and changes in the liquid phase state.

[0042] After the stage state chain is established, it is necessary to perform denoising processing and quantitative analysis on the concentration sequences within each stage state segment to extract the dominant optimization direction of the stage. Step S3 reads the target component concentration sequence, total solid concentration sequence, and effective extraction liquid volume sequence in each stage state segment one by one according to the stage state chain, performs head-end mirror extension, tail-end mirror extension, and central back substitution denoising on the target component concentration sequence and the total solid concentration sequence respectively, obtains the mirror denoised dry basis target precipitation amount and the stripped solid dry basis non-target carry-out amount through逐项 accumulation and conversion with the initial dry basis feeding amount, and then inputs them into the stage discrimination model to generate the stage selection coefficient and perform corresponding determination with the preset target quality structure, and finally forms a stage parameter sequence arranged in the order of stages.

[0043] S301: Sequentially reading of the stage state chain.

[0044] After the stage state chain is formed, the target component concentration sequence, total solid concentration sequence, and effective extraction liquid volume sequence in each stage state segment are read one by one according to this stage state chain. The reading order is consistent with the segment order in the stage state chain, that is, from the first stage state segment to the last stage state segment in sequence. For each stage state segment, the corresponding three sequences inside it are completely extracted, and the alignment relationship between each sampling moment is kept unchanged. Through the above sequential segment reading method, it can be ensured that the subsequent denoising calculation and stage discrimination are strictly based on the continuous data within a single stage.

[0045] S302: Mirror extension and central back substitution denoising.

[0046] After each sequence is read, mirror extension and central back substitution denoising are performed on the target component concentration sequence and the total solid concentration sequence of each stage state segment respectively. Taking any sequence to be processed X = {x1, x2,..., xn} as an example, it is preferably to select p sampling points at each of the head and tail ends of the sequence for mirror extension, where p is preferably 3 - 5 sampling points and satisfies p < n / 2. After mirror extension, the mirror-completed sequence X' = {xp,..., x2, x1, x1, x2,..., xn, xn, xn - 1,..., xn - p + 1} is obtained. On this basis, denoising is performed on the mirror-completed sequence by the central window back substitution method: taking the current sampling point as the center, an odd window with a length of L is selected for local smoothing calculation, L is preferably 3, 5 or 7, and the smoothing result at the center position of the window is filled back to the corresponding position of the original sequence, so as to obtain the mirror denoised sequence. Central back substitution can be implemented by moving average, weighted average or Savitzky-Golay local fitting method, and its essence is to weaken the influence of noise and mutation points at the boundary positions on the result while retaining the central trend of the sequence. The target component concentration sequence and the total solid concentration sequence are processed in the same way to obtain the mirror denoised target component concentration sequence and the mirror denoised total solid concentration sequence respectively.

[0047] S303: Calculation of the amount of target precipitate on the denoised dry basis.

[0048] After obtaining the target component concentration sequence for mirror-image denoising, this sequence is accumulated item by item with the effective extract volume sequence of the corresponding stage state segment, and the mirror-image denoising dry basis target precipitation amount corresponding to that stage state segment is calculated by combining it with the initial dry basis feed amount. Preferably, the target component concentration for mirror-image denoising is denoised as... Given that the effective extract volume increment is ΔVk and the initial dry basis feed amount is M0, the target precipitation amount of the mirror-image denoised dry basis in this stage, Eg,s, can be expressed as Eg,s=(Σ ,k ΔVk) / M0, where the summation range is all sampling times within the state segment of this stage. When the effective extract volume is recorded using cumulative volume, ΔVk = Vk - Vk-1 can be set; when the effective extract volume is obtained by converting instantaneous flow rate, interval conversion can be performed based on the time interval between adjacent sampling times Δtk = tk - tk-1 and the corresponding instantaneous flow rates Qk-1 and Qk, preferably setting ΔVk = (Qk + Qk-1) × Δtk / 2. In one embodiment, the concentration of the target component in the image-denoised sample at each sampling time is first multiplied by the increment of the effective extract volume in the corresponding sampling interval, then the product terms are accumulated, and finally converted using the initial dry basis feed amount to obtain the target precipitation amount in the dry basis of this stage. Thus, the target group analysis level under different batches and different feed scales can be unified to a comparable dry basis scale.

[0049] S304: Calculation of non-target carry-out amount of stripped solid dry basis.

[0050] After calculating the target precipitation amount, the target component concentration sequence is subtracted from the image-denoised total solids concentration sequence for each stage state segment to obtain the non-target solids concentration sequence. This non-target carry-over amount of the stripped solids dry basis is then calculated. Preferably, the image-denoised total solids concentration is denoised as... ,k, the concentration of the target component for image denoising is ,k, then the non-target solid concentration ,k= ,k- Let En,s be the amount of non-targeted solids carried out during this stage. Then En,s = (Σ ,k ΔVk) / M0. In the specific calculation, the concentration of non-target solids is first calculated at each sampling time, then multiplied by the increase in effective extract volume within the corresponding sampling interval and accumulated, finally converted to the initial dry basis feed amount. This calculation allows for the quantitative characterization of solids carried out during the extraction process, other than the target component, providing a reverse constraint index that aligns with the target precipitation amount for subsequent direction selection.

[0051] S305: Generation of stage orientation coefficients and determination of dominant control direction.

[0052] After calculating the target extraction amount of the denoised image substrate and the non-target carry-out amount of the stripped solid substrate, both are input into a pre-defined stage discrimination model to output the stage orientation coefficients for the corresponding stages. The stage discrimination model can be constructed using a multilayer perceptron neural network. Specifically, during construction, over 2000 stage samples from historical extraction batches can be collected. The target extraction amount of the denoised image substrate and the non-target carry-out amount of the stripped solid substrate are used as input features, and the stage orientation coefficients obtained from expert experience annotations or backtracking from the best historical batches are used as supervision labels. The model structure can be set to an input layer with 2 nodes, 3 hidden layers, and 128, 64, and 32 neurons respectively, and an output layer with 1 node. During training, the backpropagation algorithm and Adam optimizer can be used, with an initial learning rate of 0.001, 500 iterations, a batch size of 64, an L2 regularization coefficient of 0.01, and an early stopping mechanism. Training is terminated when the validation set loss no longer decreases for 10 consecutive epochs. To facilitate consistent input of data from different batches, the input features are preferably normalized according to the statistical range of the training set; the output layer preferably uses a hyperbolic tangent activation function or an equivalent bounded mapping method to ensure that the stage orientation coefficient αs falls within the interval [-1, 1]. After training, the model achieves a prediction accuracy of 0.93 on the independent test set. During actual inference, the input feature vector Xs=[Eg,s, En,s] of the current stage is input into the model to obtain the corresponding stage orientation coefficient αs; this coefficient is then compared with the preset target quality structure to determine the dominant control direction of the state segment of that stage, and written into the stage parameter sequence according to the stage state chain order.

[0053] For example, the process of determining the correspondence between the stage orientation coefficient and the preset target quality structure is as follows: A target quality structure table is pre-constructed, and the dominant control direction and corresponding process parameter adjustment strategy for each interval are recorded using the stage orientation coefficient interval as an index. During determination, the current stage orientation coefficient αs is directly read, its corresponding interval is located, and the corresponding record is extracted to complete the determination of the dominant control direction. Preferably, the baseline process parameters T0, t0, L0, and N0 represent the extraction temperature, extraction time, solvent addition amount, and stirring intensity of the current stage, respectively. The target quality structure table may include the following four intervals: When αs ≥ 0.65, the dominant control direction is determined to be the target release enhancement direction, and the corresponding adjustment strategy is T = T0 + (10-15) degrees Celsius, t = t0 + (15-30) minutes, L = L0 + 1.0. M0, N=N0+(20-50) rpm; when 0.25≤αs<0.65, the dominant control direction is determined to be the balanced release direction, and the corresponding adjustment strategy is to keep T unchanged, t=t0-(10-20) minutes, L maintain the original design value, and N=N0-(10-30) rpm; when -0.25≤αs<0.25, the dominant control direction is determined to be the neutral maintenance direction, and the corresponding adjustment strategy is to keep T=T0-5 degrees Celsius, t=t0-20 minutes, and L=L0-0.5 M0, N=N0-(30-50) rpm; when αs<-0.25, the dominant control direction is determined to be the impurity suppression direction, and the corresponding adjustment strategy is T=T0-(15-20) degrees Celsius, t=t0-(30-45) minutes, L=L0-1.5 M0, N = N0 - (50-80) rpm. Before execution, the adjusted parameters can be trimmed according to the equipment's allowable range and the upper and lower limits of process safety to avoid exceeding the limits. The above interval division and parameter adjustment range can be set based on the statistical results of historical best batch data to ensure that the judgment result matches the true quality structure of the extract.

[0054] In one embodiment, when processing the third-stage state segment of a batch of Salvia miltiorrhiza extract, the stage selection coefficient output by the stage discrimination model is 0.72. Since this value falls within the range of αs ≥ 0.65, the system directly extracts the "target release enhancement direction" from the target quality structure table as the dominant control direction and calls the corresponding process parameter adjustment strategy. Specifically, the extraction temperature is increased from the current set value to 92 degrees Celsius, the extraction time is extended to 135 minutes, the solvent addition is increased to 9 times the initial dry basis feed amount, and the stirring intensity is increased to 230 rpm. This set of parameters is written into the third position of the stage parameter sequence. The entire determination process can be completed within 0.02 seconds without manual recalculation, thus ensuring the real-time performance and accuracy of stage switching. In this embodiment, after the above determination and adjustment, at the end of the third stage operation, the proportion of the target group analysis is 18.7% higher than that of the unoptimized batch, the non-target carry-over amount is reduced by 21.4%, and the final quality structure of the entire batch of extract meets the requirements of the subsequent concentration process. Therefore, it can be seen that the corresponding determination mechanism can realize the direct mapping from the stage orientation coefficient to the specific process parameter, and has good real-time performance, repeatability and stability.

[0055] By reading each stage state segment in S3, performing mirror denoising, quantifying the target precipitate and non-target carry-over amounts, generating stage orientation coefficients, and determining their correspondence with the target quality structure, a stage parameter sequence for the batch to be processed was finally constructed. This sequence fully records the dominant control direction of each stage state segment and its corresponding extraction temperature, extraction time, solvent addition amount, and stirring intensity.

[0056] After the stage parameter sequence is formed, it needs to be converted into control instructions that the extraction equipment can directly execute. Step S4 reads the stage parameter sequence and stage state chain, and writes the corresponding process parameters into each execution bit of the extraction execution instruction set according to the order of the segments in the stage state chain, forming an extraction execution instruction set with a fixed sequence. Then, according to the execution bit order, the extraction temperature, extraction time, solvent addition amount and stirring intensity of each execution bit are sent to the extraction equipment, thereby realizing the staged and precise operation control of the plant extraction process.

[0057] S401: Read the stage parameter sequence and stage state chain.

[0058] After the stage parameter sequence is formed, the stage parameter sequence and stage state chain are read to obtain the full, ordered data required for instruction generation. During reading, a unified storage unit can be directly accessed, and the entire contents of both sequences can be extracted completely. Simultaneously, length consistency checks and sequence correspondence checks are preferably performed to ensure that the number of segments in the stage state chain is completely consistent with the number of parameter groups in the stage parameter sequence, and that the two correspond one-to-one in time order. Through the above reading and verification, stage parameter mismatches or execution bit offsets can be avoided.

[0059] S402: Writing and fetching execution bits to form the execution instruction set.

[0060] After reading, the process parameters corresponding to the stage state segment at the beginning of the stage state chain are written to the starting execution position, and the process parameters corresponding to each stage state segment at subsequent positions in the stage state chain are sequentially written to the subsequent execution positions, thus forming an extraction execution instruction set with a fixed sequence. The extraction execution instruction set can be defined as a set of control instructions composed of multiple execution positions arranged in the order of the stage state chain; each execution position corresponds to a stage state segment and includes at least extraction temperature, extraction duration, solvent addition amount, and stirring intensity. The writing process preferably uses a sequential appending method, and each execution position is appended with a stage number, start condition, and end condition so that the device can switch execution according to stages. In one embodiment, when the dominant control direction of the first segment of the stage state chain is enhanced release, the starting execution position can be written with an extraction temperature of 85 degrees Celsius, an extraction duration of 120 minutes, a solvent addition amount of 8 times the dry basis feed amount, and a stirring intensity of 200 rpm. Thus, a control instruction structure with clear temporal relationships and execution semantics can be established.

[0061] S403: Issuance of the extraction and execution instruction set and completion of the batch extraction and execution.

[0062] After the extraction execution instruction set is formed, the extraction temperature, extraction time, solvent addition amount, and stirring intensity corresponding to each execution position are sequentially sent to the extraction equipment according to the order of each execution position. This allows the plant extraction process to complete the entire batch extraction operation according to the execution sequence corresponding to the stage state chain. Parameters can be transmitted bit by bit through the industrial control protocol, and the equipment feedback status can be received in real time until all execution positions are processed. In one embodiment, the extraction execution instruction set for a batch of Salvia miltiorrhiza extract includes four execution bits: the first execution bit corresponds to an extraction temperature of 85 degrees Celsius, an extraction time of 120 minutes, a solvent addition amount of 8 times the initial dry basis feed amount, and a stirring intensity of 200 rpm; the second execution bit corresponds to an extraction temperature of 92 degrees Celsius, an extraction time of 90 minutes, a solvent addition amount of 6 times the initial dry basis feed amount, and a stirring intensity of 180 rpm; the third execution bit corresponds to an extraction temperature of 78 degrees Celsius, an extraction time of 150 minutes, a solvent addition amount of 7 times the initial dry basis feed amount, and a stirring intensity of 150 rpm; the fourth execution bit corresponds to an extraction temperature of 70 degrees Celsius, an extraction time of 60 minutes, a solvent addition amount of 5 times the initial dry basis feed amount, and a stirring intensity of 120 rpm. The equipment PLC controller can trigger parameter updates according to the stage state chain sequence and confirm the match between the stage switching point and the concentration and spectral characteristics based on the online monitoring results. After the entire batch extraction operation was completed, the proportion of the target group in the extract reached the optimal design requirements, and the amount of non-target components carried out remained at a low level, thus realizing the complete transformation from stage parameters to the actual operation process.

[0063] By reading the stage parameter sequence and stage state chain in S4, writing the execution bits sequentially, forming the extraction execution instruction set, and sending the parameters bit by bit to the extraction device, an extraction execution instruction set with a strict stage execution order was finally generated and the whole batch extraction operation was completed, so that the process parameters of each stage were executed accurately.

[0064] Specifically, the above are merely preferred embodiments of this application and are not intended to limit this application.

[0065] In the description of this specification, references to terms such as "an embodiment," "example," and "specific example" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for intelligent optimization and control of plant extraction process parameters, characterized in that, Including the following steps: S1: Collect the pre-state information of the batch of raw materials to be processed, collect the near-infrared spectrum of the extract, the concentration of the target component, the total solids concentration and the effective extract volume of the observation sequence according to the preset sampling interval, write the pre-state information of the raw materials into the starting position and align it with a unified timestamp, record the initial dry basis feed amount and write it into the initial state file. S2: Read the process observation sequence in the initial state file, perform boundary retrieval based on the concentration and spectral changes at adjacent sampling times, mark the stage boundary point when the change direction changes and the main feature band position switches, and divide and connect according to the boundary point to form a stage state chain; S3: Read the sequence of each segment according to the stage state chain, and after mirror expansion and center back substitution denoising, accumulate and calculate to obtain the target precipitation amount of the mirror denoised dry base and the non-target carry-out amount of the stripped solid dry base. Input the stage discrimination model to output the orientation coefficient, and write the dominant control direction into the stage parameter sequence after the target quality structure is determined. S4: Read the stage parameter sequence and stage state chain, write the first and second corresponding process parameters into the start execution bit, and write the subsequent parameters in sequence to form the extraction execution instruction set, and send them to the extraction equipment in sequence to complete the extraction operation.

2. The intelligent optimization and control method for plant extraction process parameters according to claim 1, characterized in that, Step S1 includes: The raw material pre-processing information includes the raw material moisture content, raw material crushing state, and near-infrared characterization information of the raw material before entering the tank. The raw material moisture content, raw material crushing state, and near-infrared characterization information of the raw material before entering the extraction process are collected and recorded.

3. The intelligent optimization and control method for plant extraction process parameters according to claim 2, characterized in that, Step S1 also includes: The process observation sequence continuously collects near-infrared spectra, target component concentrations, total solids concentrations, and effective extract volumes of the extract at preset sampling intervals. The raw material pre-state information is written into the starting position of the process observation sequence and each sampling item is aligned with a unified timestamp. The initial dry basis feed amount is recorded and written into the initial state file.

4. The intelligent optimization and control method for plant extraction process parameters according to claim 1, characterized in that, Step S2 includes: Boundary retrieval uses changes in target component concentration, total solids concentration, and near-infrared spectrum of extract between adjacent sampling times as the basis for determining continuity. Boundary retrieval is performed on the process observation sequence moment by moment. When the direction of change changes and the position of the main characteristic band of the near-infrared spectrum of extract switches accordingly, the stage boundary point is marked.

5. A method for intelligent optimization and control of plant extraction process parameters according to claim 4, characterized in that, Step S2 also includes: The process observation sequence is divided into multiple stage state segments that are connected end to end and have a fixed time order according to all stage boundary points. Then, the stage state segments are sequentially connected according to their positions in the process observation sequence to form a stage state chain.

6. The intelligent optimization and control method for plant extraction process parameters according to claim 1, characterized in that, Step S3 includes: According to the stage state chain, the target component concentration sequence, total solids concentration sequence and effective extract volume sequence are read from each stage state segment. The first end mirror expansion and the last end mirror expansion are performed on the target component concentration sequence and total solids concentration sequence of each stage state segment to obtain the mirror completion sequence. Then, the center substitution is performed to obtain the mirror denoised target component concentration sequence and the mirror denoised total solids concentration sequence.

7. The intelligent optimization and control method for plant extraction process parameters according to claim 6, characterized in that, Step S3 also includes: The target component concentration sequence of the mirror-denoised sample is accumulated item by item according to the sampling interval and the corresponding effective extract volume sequence. The target precipitation amount of the mirror-denoised dry basis is calculated by converting it with the initial dry basis feed amount. The non-target solid concentration sequence is obtained by subtracting the target component concentration sequence of the mirror-denoised sample from the total solid concentration sequence of the mirror-denoised sample. The non-target solid concentration sequence is then accumulated and converted with the initial dry basis feed amount to obtain the non-target carry-out amount of the stripped solid dry basis.

8. The intelligent optimization control method for plant extraction process parameters according to claim 7, characterized in that, Step S3 also includes: The target precipitation amount of the mirror-denoised dry substrate and the non-target carry-out amount of the stripped solid dry substrate are input into the stage discrimination model. The stage discrimination model outputs the stage orientation coefficient. The stage orientation coefficient is then matched with the preset target quality structure to obtain the dominant control direction and written into the stage parameter sequence according to the order of the stage state chain.

9. A method for intelligent optimization and control of plant extraction process parameters according to claim 1, characterized in that, Step S4 includes: Read the stage parameter sequence and stage state chain, write the process parameters corresponding to the stage state segment at the beginning of the stage state chain into the start execution bit, and write the process parameters corresponding to each stage state segment at the subsequent positions of the stage state chain into the subsequent execution bits to form the extraction execution instruction set.

10. A method for intelligent optimization and control of plant extraction process parameters according to claim 9, characterized in that, Step S4 also includes: Each execution bit in the extraction execution instruction set contains the extraction temperature, extraction time, solvent addition amount, and stirring intensity of the corresponding stage state segment. According to the execution bit sequence of the extraction execution instruction set, the extraction temperature, extraction time, solvent addition amount, and stirring intensity corresponding to each execution bit are sequentially sent to the extraction equipment to complete the extraction operation.

Images