Parameter optimization method for yucca saponin extraction process based on data processing

Abstract

The present application relates to the technical field of data processing, and particularly relates to a parameter optimization method for yucca saponin extraction process based on data processing, which comprises the following steps: obtaining multi-dimensional parameter data in the yucca saponin extraction process to construct all process parameter combinations, and initializing a particle swarm containing multiple particles, and each particle represents a set of process parameter combinations; outputting the predicted yield of the process parameter combination in the target iteration by using a machine learning model, and taking the particle corresponding to the process parameter combination with the highest predicted yield as the optimal particle in the target iteration; updating the particle swarm by using an improved particle swarm algorithm to output a global optimal particle, and taking the process parameter combination corresponding to the global optimal particle as the best extraction process parameter. The present application solves the problem of lack of flexible adjustment ability of the existing algorithm in different iteration stages.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a parameter optimization method for the yucca saponin extraction process based on data processing. Background Technology

[0002] Yucca saponins, as functional active ingredients derived from natural plants, are widely used in food, pharmaceuticals, feed additives, and daily chemicals. Their extraction efficiency and purity directly affect the quality and application effects of the final product. In actual production, the extraction process of yucca saponins usually involves the synergistic effects of multiple key factors, such as extraction temperature, extract concentration, material-to-liquid ratio, and extraction time. These factors often have complex nonlinear coupling relationships, resulting in the extraction process exhibiting significant multi-parameter, multi-variable, and dynamic characteristics.

[0003] Traditional process optimization methods often rely on single-factor experiments or orthogonal experimental designs, adjusting parameters one by one to find the optimal combination. However, these methods are not only time-consuming and costly, but also fail to fully reflect the interactions between parameters, often only obtaining local optima, which is insufficient to meet the high efficiency and stability requirements of industrial production. With the development of data processing technology and intelligent optimization algorithms, particle swarm optimization (PSO) has gradually become an effective means of optimizing the extraction process parameters of yucca saponins. This type of method treats different combinations of process parameters as individuals in the search space, continuously updating them during the iterative process to approach the optimal solution, thereby significantly improving optimization efficiency.

[0004] However, in practical applications, traditional particle swarm optimization (PSO) algorithms still have certain limitations in such complex process scenarios, especially since the setting of inertia weights has a crucial impact on optimization results. Inertia weights are used to control the motion inertia of particles during the search process, and their magnitude directly determines the balance between global exploration and local exploitation. In existing methods, inertia weights typically employ fixed values ​​or simple linear decreasing strategies, lacking dynamic response capabilities to the actual search state. Furthermore, since the multidimensional parameter data extracted during the process often exhibits volatility and uncertainty, fixed or empirically set inertia weights are difficult to adapt to the search requirements at different stages, resulting in a lack of flexible adjustment capabilities in existing algorithms at different iteration stages. Summary of the Invention

[0005] To address the problem of existing algorithms lacking flexible adjustment capabilities at different iteration stages as described in the background section, this invention provides the following solution.

[0006] In a first aspect, the present invention provides a parameter optimization method for the extraction process of yucca saponins based on data processing, comprising: acquiring multidimensional parameter data during the extraction process of yucca saponins to construct all combinations of process parameters, and initializing a particle swarm containing multiple particles, wherein each particle represents a combination of process parameters; using a machine learning model to output the predicted yield of the combination of process parameters under the target iteration, and taking the particle corresponding to the combination of process parameters with the highest predicted yield as the optimal particle under the target iteration; using an improved particle swarm algorithm to iteratively update the particle swarm to output the globally optimal particle, and assigning the process parameters corresponding to the globally optimal particle to the... The combination serves as the optimal extraction process parameters; wherein, the improved particle swarm optimization algorithm includes an inertia weight, which is the product of an adaptive adjustment factor and an initial value, and the adaptive adjustment factor is positively correlated with the perturbation potential energy under the target iteration; the perturbation potential energy is positively correlated with the parameter divergence, the difference between the prediction yield of the optimal particle under the target iteration and the mean prediction yield of all particles; the parameter divergence is positively correlated with the multidimensional parameter data contained in each particle, and negatively correlated with the multidimensional parameter data contained in the optimal particle under the target iteration, and the target iteration is any iteration in the iterative update of the particle swarm by the particle swarm optimization algorithm.

[0007] The aforementioned technical solution introduces an adjustment mechanism that is correlated with the population distribution and the degree of advantage of the optimal result. This allows the inertia weight to dynamically change with the search process, enhancing the overall exploration capability when the population distribution is relatively dispersed or the advantage of the optimal result is obvious, and strengthening the convergence characteristics as the population gradually becomes more consistent. This enables adaptive control of the search rhythm. Simultaneously, through comprehensive analysis of the degree of population difference and the relative level of the optimal result, the algorithm can more accurately perceive its current stage, effectively avoiding blind search, excessive oscillation, and premature entrapment in local optima. Overall, this not only improves the ability to characterize complex multi-parameter coupling relationships but also significantly enhances the stability, convergence speed, and global optimization capability of the optimization process. Ultimately, it yields more accurate and practically valuable combinations of extraction process parameters, improving the efficiency and consistency of the extraction process.

[0008] Furthermore, the first Adaptive adjustment factor in the next iteration for: , , For the first sequence The perturbation potential energy under the next iteration This represents the maximum number of iterations.

[0009] The aforementioned technical solution combines energy information reflecting changes in the search state of the swarm with the overall iterative process, enabling the adjustment mechanism to exhibit an adaptive trend of shifting from highly sensitive response to stable convergence as the search phase progresses. In the early stages of optimization, it has a stronger ability to perceive and amplify state differences between individuals, effectively enhancing the global exploration of the solution space and avoiding premature entrapment in local optima. As iterations deepen, the swarm search gradually becomes more ordered and stable, and the adjustment intensity naturally weakens, allowing the search behavior to gradually transition from large-scale probing to refined approximation. This reduces ineffective perturbations while improving the ability to focus on potentially high-quality regions. It not only achieves a balanced transition between exploration and development but also effectively improves the stability and convergence efficiency of the overall search process, significantly improving both the accuracy and reliability of the optimization results.

[0010] Furthermore, the first Perturbation potential energy in the next iteration for: , For the first Parameter divergence in the next iteration The total number of particles, For the first Predicted yield of the optimal particle in the next iteration For the first The mean of the predicted yields of all particles in the next iteration. These are the preset hyperparameters.

[0011] The above technical solution characterizes the dispersion of the population distribution in the parameter space and the advantage of the current best result relative to the overall level, so that the description of the search state not only reflects the spatial distribution characteristics, but also reflects the differences in optimization quality. On this basis, by strengthening the gap between the best result and the overall average level, the guiding role can be amplified when the best result is significantly better than the overall level, thereby promoting the population to gather towards the high-quality area at an accelerated pace. When the overall difference is small, the fluctuation effect is weakened, which helps to maintain the stability of the search process.

[0012] Furthermore, the first Parameter divergence in the next iteration for: , For the first In the nth iteration The particle in the first The values ​​of each parameter dimension For the first In the nth iteration, the optimal particle is at the... The values ​​of each parameter dimension , The first Upper and lower limits for each parameter dimension. The total number of particles, This represents the total number of parameter dimensions.

[0013] The aforementioned technical solution comprehensively characterizes the dispersion and aggregation trends of the population by accumulating differences across various dimensions, enabling the algorithm to clearly identify whether it is currently in the full exploration phase or the gradual convergence phase. Based on this, it not only helps avoid evaluation bias caused by differences in parameter scales but also enhances the ability to perceive the search state, thus providing a reliable basis for the dynamic adjustment of subsequent search strategies. This effectively enhances the stability and controllability of the optimization process, reduces blind searches and ineffective iterations, lowers the risk of getting trapped in local optima, and improves overall optimization efficiency and result accuracy.

[0014] Furthermore, it also includes denoising and standardizing the multidimensional parameter data.

[0015] Furthermore, the denoising process employs mean filtering or median filtering.

[0016] Furthermore, the standardization adopts Z-score standardization.

[0017] Furthermore, it also includes terminating the iteration early when the change in the prediction yield of the globally optimal particle is less than a set threshold in consecutive iterations.

[0018] The above technical solution introduces a termination decision mechanism based on the trend of optimal result changes, so that the optimization process no longer relies solely on the preset number of iterations, but can make adaptive decisions based on the actual improvement of the search results. When the improvement of the optimal result is consistently within a very small range in multiple consecutive iterations, it indicates that the current search has become stable and it is difficult to obtain substantial improvement by continuing iteration. At this time, timely termination of calculation can effectively avoid the waste of time and computing resources caused by invalid search, thereby significantly improving the overall optimization efficiency.

[0019] Furthermore, the machine learning model is a Gaussian process regression model.

[0020] The above technical solution uses a Gaussian process regression model to predict the extraction effect corresponding to the combination of process parameters. It can achieve high-precision modeling of complex nonlinear relationships even with a limited number of samples and fluctuating data, so that the coupling effect between parameters can be more fully characterized, thereby improving the accuracy and stability of the prediction results. At the same time, this type of model has good uncertainty expression ability and can reflect the reliability of the results during the prediction process, which helps to improve the credibility of the optimal solution selection and reduce the risk of misjudgment.

[0021] Furthermore, the multidimensional parameter data includes: extraction temperature, extract concentration, and material-to-liquid ratio.

[0022] The beneficial effects of this invention are as follows: This invention organically integrates multidimensional process data modeling, prediction and evaluation, and swarm intelligence optimization to construct a dynamic optimization mechanism oriented towards extraction results in a complex parameter space. This allows for continuous and efficient screening and updating of various process parameter combinations during iteration, significantly improving parameter optimization efficiency and reducing reliance on numerous actual experiments. A linkage adjustment mechanism based on the population distribution state and the relative advantage of optimal results is introduced, enabling the algorithm to automatically balance the overall search range and local convergence ability at different stages. It maintains sufficient exploration in the early stages and achieves stable convergence in the later stages, effectively avoiding premature entry into local optima or ineffective oscillations. Simultaneously, a termination strategy based on the trend of predicted results allows for timely termination of calculations when optimization tends to stabilize, reducing redundant iterations and improving overall operating efficiency. Overall, this invention can more accurately characterize the coupling relationships between multiple parameters, enhance the adaptability and robustness of the optimization process, significantly improve the stability, consistency, and output level of the extraction process, while reducing energy consumption and time costs, demonstrating significant engineering application value. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a parameter optimization method for a data processing-based yucca saponin extraction process according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the comparison of the predicted yield with the number of iterations before and after the improvement of the particle swarm optimization algorithm for the data processing-based yucca saponin extraction process according to an embodiment of the present invention. Detailed Implementation

[0024] Example of a parameter optimization method for yucca saponin extraction process based on data processing.

[0025] like Figure 1 The flowchart shown below illustrates the parameter optimization method for yucca saponin extraction based on data processing, according to an embodiment of the present invention, which includes the following steps: S1: Obtain multidimensional parameter data during the extraction of yucca saponins to construct all process parameter combinations, and initialize a particle swarm containing multiple particles, with each particle representing a set of process parameter combinations.

[0026] In a preferred embodiment, multidimensional parameter data from the yucca saponin extraction process is acquired to construct all combinations of process parameters. A particle swarm containing multiple particles is then initialized based on this multidimensional parameter data, where each particle represents a complete combination of process parameters for global search and optimization within the parameter space. Specifically, the multidimensional parameter data includes: extraction temperature, extract concentration, and solid-liquid ratio. Extraction temperature directly affects the dissolution rate and thermal stability of yucca saponins; extract concentration determines the polarity of the solvent system, thus affecting the dissolution efficiency of the active ingredients; and the solid-liquid ratio reflects the degree of contact and mass transfer conditions between the raw material and the solvent.

[0027] Furthermore, the process includes preprocessing the multidimensional parameter data to improve data quality and the accuracy and stability of subsequent optimization results. The preprocessing includes two key steps: denoising and standardization. In the denoising stage, random noise introduced during data acquisition due to sensor accuracy limitations, environmental disturbances, or data transmission errors is smoothed by applying either mean filtering or median filtering to the multidimensional parameter data. Specifically, mean filtering reduces the impact of random fluctuations by averaging neighborhood data, making it suitable for scenarios with relatively uniform noise distribution. Median filtering, on the other hand, effectively suppresses sudden impulse noise or outliers by replacing the current value with the median of neighborhood data. Through these denoising processes, random interference in the original data can be effectively reduced, improving the data's authenticity and continuity, thus providing a more reliable data foundation for subsequent particle swarm optimization and significantly improving the stability and convergence accuracy of the optimization results.

[0028] After denoising, the multidimensional parameter data is further standardized. The Z-score standardization method is preferred for scaling each parameter, transforming the parameter values ​​into a standard normal distribution centered on the mean and scaled by the standard deviation. This standardization effectively eliminates the influence of differences in units and ranges between parameters, ensuring that all parameters participate in particle swarm optimization on the same numerical scale. This prevents a single parameter from dominating the optimization process due to its excessively large value range, improving the algorithm's search balance and convergence efficiency in multidimensional space. Furthermore, standardization enhances the comparability of different parameters, improving the accuracy of fitness function evaluation and making the optimization results more aligned with actual process requirements.

[0029] S2: The particle swarm is iteratively updated using an improved particle swarm algorithm to output the globally optimal particle. The combination of process parameters corresponding to the globally optimal particle is then used as the optimal extraction process parameters.

[0030] like Figure 2As shown in the figure, the predicted yield of the yucca saponin extraction process based on data processing in the embodiment of the present invention is compared with the number of iterations before and after the improvement of the particle swarm algorithm.

[0031] In a preferred embodiment, the improved particle swarm optimization algorithm includes inertia weights, which are the product of an adaptive adjustment factor and an initial value. Adaptive adjustment factor in the next iteration for: , , For the first sequence The perturbation potential energy under the next iteration This represents the maximum number of iterations.

[0032] By introducing a dynamically changing adjustment mechanism during particle swarm optimization, the energy differences between different stages are mapped, making the adjustment process more sensitive to changes in the early stages and smoothing out in the later stages, thus avoiding instability caused by drastic weight fluctuations. Furthermore, the adjustment results are gradually decayed in conjunction with the iteration progress, enabling the algorithm to maintain strong global exploration capabilities in the early stages and gradually enhance local convergence capabilities in the later stages. This dual constraint mechanism allows particles to have greater inertia in the early stages of the search to fully traverse the solution space, while gradually reducing inertia as the optimization process progresses, thus encouraging particles to concentrate more in high-quality regions and accelerating convergence. Simultaneously, by combining the adjustment process with changes in the overall state of the swarm, the algorithm can adaptively adjust according to the actual search situation, rather than relying on fixed empirical parameters. This effectively improves the algorithm's adaptability to different problem scenarios, reduces premature convergence and oscillations, and significantly enhances the stability, convergence accuracy, and global optimization performance of the optimization process.

[0033] No. Perturbation potential energy in the next iteration for: , For the first Parameter divergence in the next iteration The total number of particles, For the first Predicted yield of the optimal particle in the next iteration For the first The mean of the predicted yields of all particles in the next iteration. The hyperparameters are preset. Specifically, the optimal particle is defined as follows: the particle corresponding to the process parameter combination with the highest predicted yield under the target iteration is selected from the predicted yields of the machine learning model. The machine learning model is a Gaussian process regression model.

[0034] By introducing a metric mechanism closely related to the overall search state during particle swarm optimization, this approach utilizes the distribution differences among particles to characterize the dispersion of the current search space, reflecting the swarm's exploration capability. Furthermore, it measures optimization progress by considering the advantage of the current optimal result relative to the overall level. By reinforcing these differences, the guiding role of the optimal result is effectively amplified when it significantly outperforms the overall level, while excessive fluctuations are suppressed when the overall level tends to be consistent, thus achieving an adaptive balance between exploration and convergence. Simultaneously, a prediction mechanism based on Gaussian process regression is introduced to evaluate the effectiveness of various process parameter combinations at the current stage. This ensures that the determination of the optimal individual no longer relies on actual experimental feedback but on high-confidence prediction results, thereby reducing experimental costs and time consumption while improving search efficiency and stability. Through this structural design, the premature convergence or blind search problems that easily occur in particle swarm optimization in complex multi-parameter spaces can be effectively avoided, improving global optimization capability and result reliability.

[0035] No. Parameter divergence in the next iteration for: , For the first In the nth iteration The particle in the first The values ​​of each parameter dimension For the first In the nth iteration, the optimal particle is at the... The values ​​of each parameter dimension , The first Upper and lower limits for each parameter dimension. The total number of particles, This represents the total number of parameter dimensions.

[0036] By uniformly quantifying the deviation of each individual in the population from the current optimal solution, a precise characterization of the overall distribution state is formed. The core logic lies in using the current optimal result as a reference benchmark, comparing the values ​​of each individual across all parameter dimensions with this benchmark, and normalizing the values ​​by incorporating the allowable range of variation for each parameter. This ensures fair comparison of parameters with different dimensions and value ranges on the same scale. Subsequently, the deviations across each dimension are accumulated to comprehensively reflect the overall distance relationship between individuals and the optimal solution. Finally, by summarizing the deviations of all individuals, an evaluation result representing the dispersion of the population is obtained. This approach not only accurately reflects the clustering or dispersion state of the population during the current search process but also avoids measurement distortion caused by differences in parameter scales, thus significantly improving the accuracy of describing search behavior. In practical applications, this method helps determine the stage of the algorithm. A highly concentrated population indicates gradual convergence, while a relatively dispersed distribution indicates strong exploratory capabilities. Therefore, it provides a reliable basis for dynamic adjustments to subsequent search strategies, further enhancing the stability and global optimization ability of the overall optimization process, while effectively reducing the risk of getting trapped in local optima.

[0037] Furthermore, during the iterative optimization process of the particle swarm optimization algorithm, an adaptive termination mechanism based on the trend of optimal result changes is introduced: Within multiple consecutive preset iteration cycles, the predicted yield changes of the process parameter combination corresponding to the globally optimal particle are monitored in real time. When the improvement in predicted yield between adjacent iterations consistently falls below a preset threshold, the current optimization process is determined to have converged, and subsequent iterations are terminated early. Specifically, by serializing and recording the globally optimal result output in each iteration and dynamically evaluating its change magnitude, the process of the algorithm gradually transitioning from a rapid improvement phase to a stable phase can be effectively identified, thus allowing the search to end promptly without reaching the maximum number of iterations.

[0038] This invention deeply integrates multidimensional process data processing, predictive modeling, and swarm intelligence optimization methods. It constructs a data-driven optimization mechanism within a complex process parameter extraction space, transforming the parameter search process from one reliant on experience or extensive trial-and-error experiments into a highly efficient iterative process based on prediction results. This significantly reduces experimental costs and improves optimization efficiency. Simultaneously, by denoising and standardizing the multidimensional parameter data, data quality and consistency are effectively improved, enhancing the reliability of model input and making subsequent prediction results more stable and accurate. During the optimization process, a dynamic adjustment mechanism reflecting the swarm distribution and the degree of advantage of the optimal result is introduced. The search process exhibits adaptive adjustment capabilities at different stages, maintaining sufficient exploration capacity in the early stages and achieving stable convergence in the later stages, thereby effectively avoiding getting trapped in local optima and excessive oscillations. Furthermore, by combining probabilistic modeling-based prediction methods to evaluate the process effects, the selection of the optimal solution becomes more forward-looking and accurate. An early termination mechanism reduces invalid iterations, further improving overall computational efficiency. In summary, this scheme can significantly improve search efficiency and stability while ensuring optimization accuracy, enabling refined control of key process parameters. This, in turn, improves the yield, consistency, and process controllability of yucca saponin extraction, demonstrating strong practical application value.

[0039] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0040] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A parameter optimization method for yucca saponin extraction process based on data processing, characterized in that, include: Multidimensional parameter data of yucca saponin extraction process were obtained to construct all process parameter combinations, and a particle swarm containing multiple particles was initialized, with each particle representing a set of process parameter combinations. The predicted yield of the process parameter combination under the target iteration is output by the machine learning model, and the particle corresponding to the process parameter combination with the highest predicted yield is taken as the optimal particle under the target iteration. An improved particle swarm optimization algorithm is used to iteratively update the particle swarm to output the globally optimal particle. The combination of process parameters corresponding to the globally optimal particle is then used as the optimal extraction process parameters. The improved particle swarm optimization algorithm includes an inertia weight, which is the product of an adaptive adjustment factor and an initial value. The adaptive adjustment factor is positively correlated with the perturbation potential energy under the target iteration. The perturbation potential energy is positively correlated with the parameter divergence, the difference between the prediction yield of the optimal particle under the target iteration and the mean prediction yield of all particles. The parameter divergence is positively correlated with the multidimensional parameter data contained in each particle and negatively correlated with the multidimensional parameter data contained in the optimal particle under the target iteration. The target iteration is any iteration in the particle swarm optimization algorithm's iterative update of the particle swarm.

2. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, No. Adaptive adjustment factor in the next iteration for: , , For the first sequence The perturbation potential energy under the next iteration This represents the maximum number of iterations.

3. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, No. Perturbation potential energy in the next iteration for: , For the first Parameter divergence in the next iteration The total number of particles, For the first Predicted yield of the optimal particle in the next iteration For the first The mean of the predicted yields of all particles in the next iteration. These are the preset hyperparameters.

4. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, No. Parameter divergence in the next iteration for: , For the first In the nth iteration The particle in the first The values ​​of each parameter dimension For the first In the nth iteration, the optimal particle is at the... The values ​​of each parameter dimension , The first Upper and lower limits for each parameter dimension. The total number of particles, This represents the total number of parameter dimensions.

5. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, It also includes denoising and standardizing the multidimensional parameter data.

6. The parameter optimization method for the yucca saponin extraction process based on data processing according to claim 5, characterized in that, The noise reduction employs mean filtering or median filtering.

7. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 5, characterized in that, The standardization adopts Z-score standardization.

8. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, It also includes terminating the iteration early when the change in the prediction yield of the globally optimal particle is less than a set threshold in consecutive iterations.

9. The parameter optimization method for the yucca saponin extraction process based on data processing according to claim 1, characterized in that, The machine learning model is a Gaussian process regression model.

10. The parameter optimization method for yucca saponin extraction process based on data processing according to claim 1, characterized in that, The multidimensional parameter data includes: extraction temperature, extract concentration, and material-to-liquid ratio.

Images