Method for controlling stability of selenium content in konjak functional food

Abstract

This invention relates to the field of food processing technology and proposes a method for controlling the stability of selenium content in konjac functional foods. The method includes: continuously collecting multiple batches of historical samples using atomic spectroscopy to obtain an original time-series observation sequence; measuring the temporal fluctuation entropy and dynamically assigning weights to the original time-series observation sequence to construct an adaptive weighted observation sequence; recursively segmenting the fluctuation period based on the adaptive weighted observation sequence to obtain a steady-state operating sub-interval and its nonlinear compensation factor; identifying the drift trend of the current production batch measurement value to obtain the instantaneous drift amplitude; and correcting the instantaneous drift amplitude based on the nonlinear compensation factor of the corresponding steady-state sub-interval to obtain a stabilization control parameter. This invention effectively suppresses instrument drift and noise interference through historical data purification, fluctuation entropy measurement, and nonlinear compensation correction, achieving precise selenium content control and batch-to-batch stability.

Description

Technical Field

[0001] This invention relates to the field of food processing technology, and in particular to a method for controlling the stability of selenium content in konjac functional foods. Background Technology

[0002] Existing technologies lack adaptive correction for the temporal drift of atomic spectrometer light sources, resulting in spurious fluctuations introduced by instrument drift in the absorbance sequences of continuously acquired historical samples. They also lack methods for quantifying the disorder of local fluctuations in selenium content time-series data, making it impossible to distinguish reliable fluctuations from random noise, thus severely impacting subsequent periodic analyses. Furthermore, they struggle to automatically identify the fluctuation cycle boundaries and internal stable sub-intervals at different production stages, typically relying on manual experience for division, which is highly subjective and unable to adapt to process changes. Finally, they lack nonlinear compensation models for each stable sub-interval, and conventional linear proportional control cannot fit the marginal effects of selenium content drift, leading to low compensation accuracy.

[0003] In existing technologies, there is a lack of nonlinear correction methods that incorporate historical fluctuation patterns to address the real-time drift trend of selenium content in current production batches. Typically, only single-point deviation feedback is used, which is insufficient to handle cumulative drift across multiple batches and nonlinear response characteristics. This leads to control parameter lag or overshoot, ultimately affecting the batch-to-batch stability of selenium content in konjac functional foods. Therefore, there is an urgent need to develop a selenium content stability control method based on temporal fluctuation entropy measurement, recursive period segmentation, and nonlinear compensation factor mapping. This method aims to address the limitations of existing technologies in adaptively handling light source drift, noise interference, period segmentation, and nonlinear compensation, thereby improving the accuracy and robustness of selenium content control. Summary of the Invention

[0004] This invention provides a method for controlling the stability of selenium content in konjac functional foods to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for controlling the stability of selenium content in konjac functional foods, comprising:

[0006] S1: Based on atomic spectroscopy, multiple batches of historical samples of selenium content in selenium-enriched konjac functional food were continuously collected to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food.

[0007] S2: Perform time-series fluctuation entropy measurement on the original time-series observation sequence to obtain the fluctuation disorder of the selenium content of the selenium-enriched konjac functional food. Based on the fluctuation disorder, perform dynamic weight allocation on the original time-series observation sequence to construct an adaptive weighted observation sequence of the selenium content of the selenium-enriched konjac functional food.

[0008] S3: Based on the adaptive weighted observation sequence, the fluctuation period of selenium content in the selenium-enriched konjac functional food is recursively segmented to obtain the steady-state operating sub-interval of the fluctuation period. The selenium content in the steady-state operating sub-interval is deviated and compensated to obtain the nonlinear compensation factor of the steady-state operating sub-interval.

[0009] S4: Identify the drift trend of the selenium content measurement value of the current production batch of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the selenium content measurement value of the current production batch.

[0010] S5: Based on the nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs, the instantaneous drift amplitude is nonlinearly corrected to obtain the stabilization control parameter of the current production batch selenium content.

[0011] In a preferred embodiment, the step of continuously collecting historical samples of selenium content from multiple batches of selenium-enriched konjac functional foods based on atomic spectroscopy to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional foods includes:

[0012] Based on the temporal drift characteristics of atomic spectral excitation sources, the spectral temporal drift decoupling of multiple batches of historical samples of selenium content in selenium-enriched konjac functional food was performed to obtain the stable absorbance sequence of the multiple batches of historical samples.

[0013] The stabilized absorbance sequence is resampled at equal intervals to obtain the time-interval observation sample set of the multiple batches of historical samples;

[0014] Local density peaks are selected from the equidistant observation sample set to obtain representative selenium content measurements from the multiple batches of historical samples;

[0015] By performing sliding window residual outlier stripping on the representative selenium content measurement, the historical selenium content observation chain of the multiple batches of historical samples is obtained.

[0016] The historical selenium content observation chain was reconstructed by phase alignment of discontinuities to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food.

[0017] In a preferred embodiment, the step of measuring the temporal fluctuation entropy of the original time-series observation sequence to obtain the fluctuation disorder of the selenium content of the selenium-enriched konjac functional food, and then dynamically assigning weights to the original time-series observation sequence based on the fluctuation disorder to construct an adaptive weighted observation sequence of the selenium content of the selenium-enriched konjac functional food, includes:

[0018] The original time-series observation sequence is progressively windowed to obtain the observation subsequence of the original time-series observation sequence;

[0019] The observed subsequence is projected into the state space to obtain the projected feature string of the observed subsequence;

[0020] Self-information assignment is performed on the projected feature string to obtain the fluctuation entropy estimate of the projected feature string;

[0021] Based on the fluctuation entropy estimation, the original time series observation sequence is reconstructed and assigned values ​​inversely to obtain the adaptive weight coefficients of the original time series observation sequence;

[0022] Based on the adaptive weighting coefficients, the original time-series observation sequences are numerically multiplied and superimposed to obtain the adaptive weighted observation sequence of selenium content in the selenium-enriched konjac functional food.

[0023] In a preferred embodiment, the step of assigning self-information to the projected feature string to obtain a fluctuation entropy estimate of the projected feature string includes:

[0024] The frequency of occurrence of the projected feature string is obtained by enumerating the repetition patterns of the projected feature string.

[0025] The frequency of appearance is rounded down by performing a negative logarithmic transformation to obtain the original score of the self-information of the projected feature string;

[0026] The original self-information score is reshaped by probability normalization to obtain the information weight of the projected feature string;

[0027] The information weights are weighted, accumulated, and aggregated to obtain the fluctuation entropy estimate of the projected feature string.

[0028] In a preferred embodiment, the step of reconstructing and assigning values ​​to the original time-series observation sequence based on the fluctuation entropy estimation to obtain the adaptive weight coefficients of the original time-series observation sequence includes:

[0029] The fluctuation entropy estimate is numerically reversed to obtain the entropy reciprocal sequence of the original time series observation sequence;

[0030] The entropy reciprocal sequence is subjected to amplitude normalization stretching to obtain the weighted basis of the entropy reciprocal sequence;

[0031] Based on the weighted basis, the original time-series observation sequence is inversely weighted and embedded to obtain the initial weight coefficients of the original time-series observation sequence;

[0032] Neighborhood smoothing suppression is applied to the initial weight coefficients to obtain the adaptive weight coefficients of the original time-series observation sequence.

[0033] In a preferred embodiment, the step of recursively segmenting the fluctuation period of selenium content in the selenium-enriched konjac functional food based on the adaptive weighted observation sequence to obtain a steady-state operating sub-interval of the fluctuation period, and performing deviation compensation mapping on the selenium content within the steady-state operating sub-interval to obtain the nonlinear compensation factor of the steady-state operating sub-interval, includes:

[0034] Phase space delay reconstruction is performed on the adaptive weighted observation sequence to obtain the multidimensional embedded trajectory of the adaptive weighted observation sequence;

[0035] Extreme point orientation tracking is performed on the multidimensional embedded trajectory to obtain the inflection point marker sequence of the multidimensional embedded trajectory;

[0036] Based on the inflection point marker sequence, the adaptive weighted observation sequence is subjected to interval bisection iterative cutting to obtain the periodic segments of the adaptive weighted observation sequence;

[0037] The periodic segments are subjected to a fluctuation amplitude homogeneity test, and the periodic segments that pass the homogeneity test are marked as steady-state operating sub-intervals;

[0038] The target deviation is decoupled from the adaptive weighted observation sequence of the steady-state operating sub-interval to obtain the deviation residual vector of the steady-state operating sub-interval;

[0039] The deviation residual vector is subjected to nonlinear amplitude compression encoding to obtain the nonlinear compensation factor of the steady-state operating sub-interval.

[0040] In a preferred embodiment, the step of identifying the drift trend of the current production batch's selenium content measurement value of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the current production batch's selenium content measurement value includes:

[0041] The selenium content of the current production batch of the selenium-enriched konjac functional food is mapped by the slope of the previous and next neighborhoods to obtain the local change rate sequence of the selenium content measurement value of the current production batch.

[0042] Based on the sign consistency of the local rate of change sequence, the trend orientation of the selenium content measurement value of the current production batch is determined to obtain the drift direction label of the selenium content measurement value of the current production batch.

[0043] By performing amplitude cumulative integral on the local rate of change sequence, the total cumulative offset of the selenium content measurement value of the current production batch is obtained;

[0044] The instantaneous drift amplitude of the selenium content measurement value of the current production batch is obtained by polarity amplitude coupling between the drift direction label and the cumulative offset.

[0045] In a preferred embodiment, the formula for calculating the instantaneous drift amplitude is as follows, including:

[0046] ;

[0047] In the formula, The drift direction label, This is the total cumulative offset. This represents the number of batches that drifted in the same direction consecutively before the current moment. The preset inertia enhancement coefficient, The preset directional attenuation factor, The preset amplitude compression factor, For hyperbolic tangent operator, The instantaneous drift amplitude, It is a natural constant. This represents the absolute value, and 1 represents the base scaling factor.

[0048] In a preferred embodiment, the nonlinear compensation factor based on the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs, and the subsequent nonlinear correction of the instantaneous drift amplitude to obtain the stabilization control parameters for the current production batch selenium content, includes:

[0049] The nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs is divided into equal probability intervals to obtain the coding partition boundary of the nonlinear compensation factor;

[0050] Based on the penetration depth of the instantaneous drift amplitude within the coding partition boundary, the instantaneous drift amplitude is subjected to interval membership determination to obtain the segment label to which the instantaneous drift amplitude belongs;

[0051] Based on the attribution segment label, intra-segment interpolation mapping is performed on the instantaneous drift amplitude to obtain the nonlinear compensation estimate of the instantaneous drift amplitude;

[0052] Boundary clamping constraints are applied to the nonlinear compensation estimate to obtain the stabilization control parameters for the selenium content of the current production batch.

[0053] In a preferred embodiment, the formula for calculating the nonlinear compensation estimate is as follows:

[0054] ;

[0055] in, The instantaneous drift amplitude, The total number of intervals divided by the coding partition boundary. For the first The center position of the segment corresponding to the boundary of each coded partition For the first The compensation base value pre-stored in each coded partition, The preset attenuation coefficient, Represents the absolute value operator. The preset shape factor is greater than 1. This represents the natural exponential function. The nonlinear compensation estimate is given.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. This invention significantly improves the signal-to-noise ratio and characterization effectiveness of observational data by constructing a full-chain data purification and enhancement mechanism from the spectral acquisition source to the measurement of fluctuation characteristics. Specifically, on the one hand, by decoupling the temporal drift of the atomic spectral excitation source, isochronous resampling, and local density peak selection, spurious fluctuations and outliers introduced by instrument drift are effectively removed. Combined with discontinuity phase alignment reconstruction, a more realistic original time-series observation sequence reflecting the dynamic changes in selenium content is obtained, solving the problem of severe interference from instrument noise and drift in historical data in existing technologies. On the other hand, a temporal fluctuation entropy measurement technique is innovatively introduced. Through progressive window segmentation, state space projection, and self-information assignment, the disorder of selenium content fluctuations is quantified, and an adaptive weighting coefficient is reconstructed based on this inverse proportion. This mechanism can intelligently distinguish between reliable fluctuations caused by process adjustments and disorderly disturbances caused by random noise. It assigns low weight to high-entropy noise segments and high weight to low-entropy trend segments, thereby constructing a high-fidelity, high-sensitivity adaptive weighted observation sequence, laying a solid data foundation for subsequent accurate period division and compensation control.

[0058] 2. This invention significantly improves the accuracy and robustness of selenium content control by establishing a closed-loop control strategy that integrates historical cycle learning, real-time drift dynamic identification, and nonlinear intelligent correction. On one hand, based on adaptive weighted observation sequences, phase space delay reconstruction and recursive segmentation automatically identify and divide steady-state operation sub-intervals at different production stages, adapting to process fluctuations without human intervention. Simultaneously, nonlinear amplitude compression encoding is applied to the deviation residuals within each sub-interval, generating a dedicated nonlinear compensation factor that accurately fits the marginal effect of selenium content drift, solving the problems of low accuracy and inability to handle nonlinear responses in conventional linear control. On the other hand, in the real-time control stage, by performing neighborhood slope mapping and cumulative offset integration on the current batch's measurement values, the amplitude and direction labels of instantaneous drift are accurately captured. Combined with the inertial trend of historical consecutive batches, the instantaneous drift amplitude reflecting the cumulative effect of multiple batches is calculated. Finally, based on the nonlinear compensation factor of the steady-state sub-interval to which the current batch belongs, equal-probability interval partitioning and interpolation mapping are performed to achieve nonlinear correction of the instantaneous drift. This process effectively avoids control lag and overshoot caused by single-point feedback, and significantly enhances the batch-to-batch consistency and stability of selenium content in konjac functional foods during multi-batch continuous production. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a method for controlling the stability of selenium content in konjac functional food according to an embodiment of the present invention.

[0060] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0061] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0062] This application provides a method for controlling the stability of selenium content in konjac functional foods. The execution subject of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for controlling the stability of selenium content in konjac functional foods can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0063] Reference Figure 1 The diagram shown is a flowchart illustrating a method for controlling the stability of selenium content in konjac functional foods according to an embodiment of the present invention. In this embodiment, the method for controlling the stability of selenium content in konjac functional foods includes:

[0064] S1: Based on atomic spectroscopy, multiple batches of historical samples of selenium content in selenium-enriched konjac functional food were continuously collected to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food.

[0065] In this embodiment of the invention, the step of continuously collecting historical samples of selenium content from multiple batches of selenium-enriched konjac functional foods based on atomic spectroscopy to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional foods includes:

[0066] Based on the temporal drift characteristics of atomic spectral excitation sources, the spectral temporal drift decoupling of multiple batches of historical samples of selenium content in selenium-enriched konjac functional food was performed to obtain the stable absorbance sequence of the multiple batches of historical samples.

[0067] The stabilized absorbance sequence is resampled at equal intervals to obtain the time-interval observation sample set of the multiple batches of historical samples;

[0068] Local density peaks are selected from the equidistant observation sample set to obtain representative selenium content measurements from the multiple batches of historical samples;

[0069] By performing sliding window residual outlier stripping on the representative selenium content measurement, the historical selenium content observation chain of the multiple batches of historical samples is obtained.

[0070] The historical selenium content observation chain was reconstructed by phase alignment of discontinuities to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food.

[0071] Based on the temporal drift characteristics of atomic spectral excitation sources, spectral temporal drift decoupling was performed on multiple batches of historical samples of selenium-enriched konjac functional foods to obtain a stable absorbance sequence for these batches. Before testing each batch of konjac samples, a standard reference substance with a known selenium content was measured, and the measured absorbance values ​​of the standard reference substance at each time point were recorded and compared with the calibrated absorbance values ​​to obtain a drift curve. The deviation between the original absorbance reading of the konjac sample to be tested and the corresponding time point on the drift curve was offset to restore the absorbance value to the baseline level under the condition that the excitation source was not drifted, forming a stable absorbance sequence that only reflects the difference in selenium content of the sample itself.

[0072] The stabilized absorbance sequence is resampled at equal intervals to obtain an isochronous observation sample set of the multiple batches of historical samples. The entire stabilized absorbance sequence is resampled using the standard process cycle length between two adjacent production batches as a fixed time step. If an original detection record exists at each time step node, the absorbance value is directly retained; otherwise, the node value is filled by linear trend extension using adjacent original detection records, thereby obtaining an isochronous observation sample set with uniform time intervals.

[0073] Local density peaks are selected from the equidistant observation sample set to obtain representative selenium content measurements from the multiple batches of historical samples. A numerical window of a preset width is expanded outwards from each equidistant observation point as the center. The number of other observation points falling within this window is counted as the local neighborhood density value of that center point, and all observation points are traversed point by point. Observation points whose local neighborhood density values ​​reach a local maximum and are higher than the density levels of their surrounding neighborhoods are marked as density peak points. The absorbance values ​​corresponding to the density peak points are retained as representative selenium content measurements for that local area, while other observation points are discarded.

[0074] Outlier removal using a sliding window residual is performed on the representative selenium content measures to obtain a historical selenium content observation chain from multiple batches of historical samples. A fixed-length sliding window moves point by point along the representative selenium content measure sequence in chronological order. Each time the window moves, the median value of all measure values ​​within the window is calculated as a trend baseline, and the residual difference between each measure value and the trend baseline is calculated. Measure points whose absolute residual value exceeds a preset threshold are marked as outliers and deleted from the sequence. The resulting gaps are smoothly filled by adjacent retained measure values. After traversing all measure points, a continuous historical selenium content observation chain without outliers is obtained.

[0075] The historical selenium content observation chain is reconstructed by phase alignment of discontinuities to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food. Points in the observation chain where the time interval between adjacent observation points exceeds a preset continuity threshold are identified and marked as discontinuities. The observation chain is then divided into several continuous segments using these discontinuities as boundaries. The starting time of the earliest continuous segment throughout the entire time period is used as a unified zero-phase reference point. The time difference between the starting time of each other continuous segment and the zero-phase reference point is calculated as the phase offset of that segment. All continuous segments are rearranged on the time axis according to their respective phase offsets and then spliced ​​together to obtain the original time-series observation sequence that is phase-continuous and uninterrupted.

[0076] Beneficial Effects: Spectral temporal drift decoupling and isochronous resampling were performed on multiple batches of historical samples of selenium content in selenium-enriched konjac functional foods. This eliminated the systematic measurement error introduced by the temporal drift of the atomic spectrometer excitation source and transformed the non-uniform detection data into an isochronous observation sample set, providing a uniform and consistent time reference for fluctuation analysis. Representative selenium content measurements were retained by selecting local density peaks and removing redundant noise. Abnormal jump values ​​were removed by sliding window residual outlier stripping. Finally, phase-continuous original time-series observation sequences were obtained through discontinuity point phase alignment reconstruction. This significantly improved the signal-to-noise ratio and characterization effectiveness of the historical data, laying a reliable foundation for subsequent fluctuation feature extraction.

[0077] Building upon this foundation, the fluctuation entropy of the projected feature string is estimated through progressive window segmentation and state-space projection. Based on inverse proportional reconstruction, low weights are assigned to high-entropy noise segments and high weights to low-entropy trend segments, constructing an adaptive weighted observation sequence that effectively suppresses the interference of random noise on periodic analysis. Furthermore, steady-state operating sub-intervals are automatically identified through phase-space delay reconstruction and recursive segmentation, and nonlinear compensation factors are generated to fit the marginal effect of selenium content drift. In the real-time control phase, the instantaneous drift amplitude of the current batch measurement value is combined with the nonlinear compensation factor of the corresponding sub-interval for correction, resulting in stable control parameters. This method achieves closed-loop control from historical data purification and fluctuation pattern learning to real-time nonlinear compensation, significantly improving the batch-to-batch consistency and stability of selenium content in the continuous multi-batch production of konjac functional foods.

[0078] S2: Perform time-series fluctuation entropy measurement on the original time-series observation sequence to obtain the fluctuation disorder of the selenium content of the selenium-enriched konjac functional food. Based on the fluctuation disorder, perform dynamic weight allocation on the original time-series observation sequence to construct an adaptive weighted observation sequence of the selenium content of the selenium-enriched konjac functional food.

[0079] In this embodiment of the invention, the step of measuring the temporal fluctuation entropy of the original time-series observation sequence to obtain the fluctuation disorder of the selenium content of the selenium-enriched konjac functional food, and then dynamically assigning weights to the original time-series observation sequence based on the fluctuation disorder to construct an adaptive weighted observation sequence of the selenium content of the selenium-enriched konjac functional food, includes:

[0080] The original time-series observation sequence is progressively windowed to obtain the observation subsequence of the original time-series observation sequence;

[0081] The observed subsequence is projected into the state space to obtain the projected feature string of the observed subsequence;

[0082] Self-information assignment is performed on the projected feature string to obtain the fluctuation entropy estimate of the projected feature string;

[0083] Based on the fluctuation entropy estimation, the original time series observation sequence is reconstructed and assigned values ​​inversely to obtain the adaptive weight coefficients of the original time series observation sequence;

[0084] Based on the adaptive weighting coefficients, the original time-series observation sequences are numerically multiplied and superimposed to obtain the adaptive weighted observation sequence of selenium content in the selenium-enriched konjac functional food.

[0085] The step of assigning self-information to the projected feature string to obtain the fluctuation entropy estimate of the projected feature string includes:

[0086] The frequency of occurrence of the projected feature string is obtained by enumerating the repetition patterns of the projected feature string.

[0087] The frequency of appearance is rounded down by performing a negative logarithmic transformation to obtain the original score of the self-information of the projected feature string;

[0088] The original self-information score is reshaped by probability normalization to obtain the information weight of the projected feature string;

[0089] The information weights are weighted, accumulated, and aggregated to obtain the fluctuation entropy estimate of the projected feature string.

[0090] The step of reconstructing and assigning values ​​to the original time-series observation sequence based on the fluctuation entropy estimation to obtain the adaptive weight coefficients of the original time-series observation sequence includes:

[0091] The fluctuation entropy estimate is numerically reversed to obtain the entropy reciprocal sequence of the original time series observation sequence;

[0092] The entropy reciprocal sequence is subjected to amplitude normalization stretching to obtain the weighted basis of the entropy reciprocal sequence;

[0093] Based on the weighted basis, the original time-series observation sequence is inversely weighted and embedded to obtain the initial weight coefficients of the original time-series observation sequence;

[0094] Neighborhood smoothing suppression is applied to the initial weight coefficients to obtain the adaptive weight coefficients of the original time-series observation sequence.

[0095] The original time-series observation sequence is progressively segmented into windows. A fixed-length observation window slides forward along the time axis of the original time-series observation sequence with a fixed step size. Each slide extracts a segment of continuous selenium content observation values ​​contained within the window as an independent observation subsequence. This process continues until the sliding window covers all observation points of the original time-series observation sequence. All observation subsequences are arranged sequentially according to the window sliding order to form an observation subsequence set.

[0096] State-space projection is performed on the observation subsequence. The selenium content numerical sequence in each observation subsequence is extracted. The difference between each value and its adjacent values ​​is used as the local change feature of the value. The rising segment, falling segment and stationary segment of the selenium content value are mapped to different state symbols respectively. According to the change order of the selenium content values ​​in the observation subsequence, each value is converted into the corresponding state symbol. Thus, the numerical observation subsequence is transformed into a symbol string composed of finite state symbols. This symbol string is the projection feature string.

[0097] Repeated patterns are enumerated for the projected feature string. All possible combinations of adjacent state symbols in the projected feature string are scanned. The number of times each combination pattern appears in the entire projected feature string is counted. This number of occurrences is the appearance frequency of the combination pattern. All combination patterns and their corresponding appearance frequencies are recorded to form a frequency statistics table.

[0098] The frequency of appearance is rounded down by performing a negative logarithmic transformation. The frequency of appearance for each combination pattern in the frequency statistics table is extracted. The ratio of the frequency of appearance to the total length of the projected feature string is calculated. The ratio is then subjected to a logarithmic operation with a constant base and the negative number is taken. The result is used as the original score of the self-information of the combination pattern. The original scores of the self-information of all combination patterns are stored in the pattern score mapping table.

[0099] The original self-information scores are reshaped by probability normalization. The original self-information scores of all combined modes in the mode score mapping table are divided one by one by the sum of all original self-information scores to obtain the normalized score corresponding to each combined mode. This normalized score is the information weight of the combined mode in the projected feature string.

[0100] The information weights are weighted and aggregated. The information weight of each combination mode is multiplied by the original score of the self-information of that combination mode to obtain the weighted self-information of that combination mode. The weighted self-information of all combination modes in the projected feature string is summed. The summation result is the fluctuation entropy estimate of the projected feature string. The magnitude of the fluctuation entropy estimate reflects the degree of disorder and irregularity of the selenium content fluctuation within the observed subsequence.

[0101] A numerical reverse mapping is performed on the fluctuation entropy estimate. The fluctuation entropy estimate corresponding to each observation subsequence in the original time series observation sequence is extracted, and the reciprocal of the fluctuation entropy estimate is calculated as its reverse mapping result. The reverse mapping results of all observation subsequences are arranged in the time order of the original time series observation sequence to obtain the entropy reciprocal sequence. In the entropy reciprocal sequence, the position with a larger fluctuation entropy estimate corresponds to a smaller reciprocal value, and the position with a smaller fluctuation entropy estimate corresponds to a larger reciprocal value.

[0102] The entropy reciprocal sequence is subjected to amplitude normalization stretching. The maximum and minimum values ​​in the entropy reciprocal sequence are extracted. The minimum value is subtracted from each entropy reciprocal and then divided by the difference between the maximum and minimum values. This linearly compresses all values ​​into a fixed numerical range. Each value within this range is the weight basis of the corresponding observation subsequence. The closer the weight basis is to the upper limit of the numerical range, the stronger the fluctuation order of the observation subsequence.

[0103] The original time-series observation sequence is embedded by weight inverse ratio based on the weight basis. The weight basis of each observation subsequence is used as the basic weight coefficient of each selenium content observation value in the observation subsequence. The basic weight coefficients of all observation subsequences are assigned point by point according to their corresponding observation point positions in the original time-series observation sequence, so as to obtain the initial weight coefficient of each selenium content observation value in the original time-series observation sequence.

[0104] The initial weight coefficients are subjected to neighborhood smoothing suppression. A smooth neighborhood is formed by extending a fixed number of observation points to the front and back of each observation point with the initial weight coefficient as the center. The arithmetic mean of all initial weight coefficients in the smooth neighborhood is calculated and the arithmetic mean is used as the updated weight coefficient of the center observation point. This process is repeated point by point to traverse all observation points in the original time series observation sequence to obtain the adaptive weight coefficients after neighborhood smoothing.

[0105] The original time-series observation sequence is numerically multiplied and superimposed based on the adaptive weight coefficient. The selenium content observation value at each moment in the original time-series observation sequence is multiplied by the adaptive weight coefficient at the corresponding moment to obtain the weighted observation value at that moment. All the weighted observation values ​​at all moments are arranged in the original time order to form an adaptive weighted observation sequence of selenium content of selenium-enriched konjac functional food that corresponds one-to-one with the original time-series observation sequence.

[0106] Beneficial Effects: By progressively windowing and projecting the state space onto the original time-series observation sequence, the continuously changing selenium content numerical sequence is transformed into a projected feature string composed of finite state symbols. This allows for the symbolic representation of local variation patterns in selenium content fluctuations, facilitating subsequent statistics and quantification. Repeating patterns are enumerated and frequency of occurrence is statistically analyzed within the projected feature string. After negative logarithmic transformation, rounding, and probability normalization, the information weight of each pattern is obtained. Weighted accumulation and aggregation are then used to calculate the fluctuation entropy estimate of the projected feature string. This fluctuation entropy estimate objectively reflects the degree of disorder in selenium content fluctuations within each observation subsequence, providing a quantitative basis for distinguishing reliable process fluctuations from random noise. Based on the fluctuation entropy estimate, numerical reversal mapping and amplitude normalization stretching are performed to obtain the weight basis of each observation subsequence. Furthermore, by applying inverse weight embedding and neighborhood smoothing suppression to the original time-series observation sequence, adaptive weight coefficients for each observation point are generated, achieving automatic suppression of high-entropy noise segments and automatic enhancement of low-entropy trend segments. Finally, an adaptive weighted observation sequence was constructed by superimposing numerical products. This sequence significantly reduced the contribution ratio of random interference components while preserving the true trend of selenium content fluctuations. It provided a high signal-to-noise ratio input data basis for subsequent recursive segmentation of fluctuation cycles and identification of steady-state operation sub-intervals, effectively improving the robustness of selenium content stability control method to noise interference and its sensitivity to weak fluctuation characteristics.

[0107] S3: Based on the adaptive weighted observation sequence, the fluctuation period of selenium content in the selenium-enriched konjac functional food is recursively segmented to obtain the steady-state operating sub-interval of the fluctuation period. The selenium content in the steady-state operating sub-interval is deviated and compensated to obtain the nonlinear compensation factor of the steady-state operating sub-interval.

[0108] In this embodiment of the invention, the recursive segmentation of the fluctuation period of selenium content in the selenium-enriched konjac functional food based on the adaptive weighted observation sequence to obtain a steady-state operating sub-interval of the fluctuation period, and the deviation compensation mapping of selenium content within the steady-state operating sub-interval to obtain the nonlinear compensation factor of the steady-state operating sub-interval, includes:

[0109] Phase space delay reconstruction is performed on the adaptive weighted observation sequence to obtain the multidimensional embedded trajectory of the adaptive weighted observation sequence;

[0110] Extreme point orientation tracking is performed on the multidimensional embedded trajectory to obtain the inflection point marker sequence of the multidimensional embedded trajectory;

[0111] Based on the inflection point marker sequence, the adaptive weighted observation sequence is subjected to interval bisection iterative cutting to obtain the periodic segments of the adaptive weighted observation sequence;

[0112] The periodic segments are subjected to a fluctuation amplitude homogeneity test, and the periodic segments that pass the homogeneity test are marked as steady-state operating sub-intervals;

[0113] The target deviation is decoupled from the adaptive weighted observation sequence of the steady-state operating sub-interval to obtain the deviation residual vector of the steady-state operating sub-interval;

[0114] The deviation residual vector is subjected to nonlinear amplitude compression encoding to obtain the nonlinear compensation factor of the steady-state operating sub-interval.

[0115] Phase space delay reconstruction is performed on the adaptive weighted observation sequence. A fixed number of time delay steps are set, and each selenium content observation value in the sequence is paired with the observation value corresponding to the fixed number of steps backward and twice the fixed number of steps, respectively, to form a trajectory point in multidimensional space. All trajectory points are connected in the original time order to obtain the multidimensional embedded trajectory.

[0116] The distance change between adjacent trajectory points is calculated point by point along the multidimensional embedded trajectory. When the distance change changes from positive to negative, it is marked as a positive turning point, and when it changes from negative to positive, it is marked as a negative turning point. All turning points are arranged in chronological order to form a turning point marking sequence.

[0117] The adaptive weighted observation sequence is divided into initial segments by using adjacent markers in the inflection point marker sequence as boundaries. The average fluctuation amplitude of the observation values ​​in each initial segment is calculated. If the amplitude exceeds the preset upper limit, the segment is divided into two parts at the point of maximum fluctuation amplitude. The division is repeated until the average fluctuation amplitude of all sub-segments does not exceed the upper limit. The resulting sub-segments are the periodic segments.

[0118] For each periodic segment, the difference between its maximum and minimum values ​​is calculated as the actual fluctuation amplitude. This is then compared with the actual fluctuation amplitude of adjacent periodic segments. If the amplitude difference between adjacent segments is less than a preset homogeneity threshold, the periodic segment is marked as a steady-state operating sub-interval.

[0119] The arithmetic mean of all observations within each steady-state operating sub-interval is calculated as the selenium content baseline. The deviation is obtained by subtracting the baseline from each observation and arranged in chronological order to form the deviation residual vector.

[0120] The largest absolute value in the deviation residual vector is taken as the maximum deviation amplitude. Each deviation is divided by the maximum deviation amplitude to obtain the normalized deviation value. The normalized deviation value is nonlinearly mapped so that its output value gradually flattens as the deviation increases. All the mapped output values ​​are combined in the original order to obtain the nonlinear compensation factor for the steady-state operation sub-interval.

[0121] Beneficial Effects: By reconstructing the phase space of the adaptively weighted observation sequence using phase space delay, a multidimensional embedded trajectory is obtained, extending one-dimensional time-series data to a high-dimensional space and clearly revealing the nonlinear dynamic structure of selenium content fluctuations. Extreme point directional tracking of the multidimensional embedded trajectory yields a sequence of inflection point markers, accurately pinpointing the critical moments when the fluctuation direction changes, providing an objective boundary basis for period segmentation. Based on the inflection point marker sequence, interval bisection iterative segmentation is performed, combined with fluctuation amplitude homogeneity testing, automatically identifying and marking steady-state operating sub-intervals with smooth and consistent fluctuations. This achieves adaptive division of the stable control window during production, ensuring the accuracy and consistency of period segmentation without the need for manual threshold setting. Decoupling the observation sequence within the steady-state operating sub-intervals by target deviation decomposes the selenium content into a baseline level and instantaneous deviation, obtaining a deviation residual vector that separates steady-state mean information from fluctuation information. Nonlinear amplitude compression encoding is applied to the deviation residual vector, making the compensation intensity responsive when the deviation is small and saturating when the deviation is large. This accurately fits the marginal diminishing characteristics of selenium content drift compensation, avoiding over-adjustment caused by linear compensation. Each steady-state operating sub-interval acquires a unique nonlinear compensation factor, providing a precise mapping relationship for applying differentiated nonlinear compensation at different process stages, effectively improving the adaptability and compensation accuracy of the control method.

[0122] S4: Identify the drift trend of the selenium content measurement value of the current production batch of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the selenium content measurement value of the current production batch.

[0123] In this embodiment of the invention, the step of identifying the drift trend of the current production batch selenium content measurement value of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the current production batch selenium content measurement value includes:

[0124] The selenium content of the current production batch of the selenium-enriched konjac functional food is mapped by the slope of the previous and next neighborhoods to obtain the local change rate sequence of the selenium content measurement value of the current production batch.

[0125] Based on the sign consistency of the local rate of change sequence, the trend orientation of the selenium content measurement value of the current production batch is determined to obtain the drift direction label of the selenium content measurement value of the current production batch.

[0126] By performing amplitude cumulative integral on the local rate of change sequence, the total cumulative offset of the selenium content measurement value of the current production batch is obtained;

[0127] The instantaneous drift amplitude of the selenium content measurement value of the current production batch is obtained by polarity amplitude coupling between the drift direction label and the cumulative offset.

[0128] The formula for calculating the instantaneous drift amplitude is as follows:

[0129] ;

[0130] In the formula, The drift direction label, This is the total cumulative offset. This represents the number of batches that drifted in the same direction consecutively before the current moment. The preset inertia enhancement coefficient, The preset directional attenuation factor, The preset amplitude compression factor, For hyperbolic tangent operator, The instantaneous drift amplitude, It is a natural constant. This represents the absolute value, and 1 represents the base scaling factor.

[0131] The selenium content measurement values ​​of the current production batch of selenium-enriched konjac functional food are mapped using a neighboring slope. The selenium content measurement value of the current production batch is extracted and its position on the production time axis is determined. A neighborhood interval is formed by tracing back a fixed number of measured batches from this position and extending forward by an equal number of unmeasured batches. Within the neighborhood interval, the difference in selenium content between each batch and the next batch is divided by the time interval between the two batches to obtain the local rate of change at that batch position. The local rate of change calculated from all batch positions is arranged in chronological order to form the local rate of change sequence of the selenium content measurement values ​​of the current production batch. Positive values ​​in the local rate of change sequence indicate that the selenium content is increasing during that period, while negative values ​​indicate that the selenium content is decreasing.

[0132] Based on the sign consistency of the local rate of change sequence, the trend direction of the selenium content measurement value of the current production batch is determined. The number of positive values ​​and negative values ​​in the local rate of change sequence is counted. If the proportion of positive values ​​in the total length of the local rate of change sequence exceeds the preset sign consistency judgment ratio, the selenium content of the current production batch is determined to be in an upward drift trend. If the proportion of negative values ​​exceeds the judgment ratio, the selenium content of the current production batch is determined to be in a downward drift trend. If neither the number of positive nor negative values ​​exceeds the judgment ratio, the selenium content of the current production batch is determined to be in a state of no significant drift. An upward drift trend is assigned a drift direction label with a positive sign value, a downward drift trend is assigned a drift direction label with a negative sign value, and a state of no significant drift is assigned a drift direction label with a zero value.

[0133] The amplitude of the local rate of change sequence is accumulated and integrated. Starting from the first value of the local rate of change sequence, the sequence is traversed sequentially. Each local rate of change value is multiplied by the time interval between the two batches to obtain the selenium content change element within that time period. The selenium content change elements within all time periods in the local rate of change sequence are summed. During the summation, positive elements increase the summation value, while negative elements decrease it. The final summation result obtained after the traversal is completed is the total cumulative offset of the selenium content measurement value of the current production batch. The sign of the total cumulative offset is consistent with the dominant direction of the local rate of change, and its magnitude reflects the overall shift of the selenium content since the beginning of the neighborhood interval.

[0134] The drift direction label and the total cumulative offset are coupled by polarity amplitude. The sign value of the drift direction label is extracted and multiplied by the absolute value of the total cumulative offset to obtain the offset amplitude with directional information. At the same time, the number of batches with the same drift direction label before the current time is obtained as the continuous drift batch count. Based on the value of the continuous drift batch count and the total cumulative offset, an inertial enhancement factor and an amplitude compression factor are calculated. The offset amplitude with directional information is multiplied by the inertial enhancement factor and then by the amplitude compression factor. The result is the instantaneous drift amplitude of the selenium content measurement value of the current production batch. The instantaneous drift amplitude contains comprehensive information of drift direction, drift accumulation degree and continuous drift inertial enhancement effect.

[0135] In the calculation of instantaneous drift amplitude, the drift direction label determines the polarity of the final instantaneous drift amplitude, and the cumulative offset serves as the base offset in subsequent nonlinear transformations. The number of consecutive batches drifting in the same direction is used to construct the inertia enhancement mechanism. The more consecutive batches drifting in the same direction, the larger the value of the inertia enhancement factor, allowing the instantaneous drift amplitude to gain additional amplification when drifting in the same direction continues, thus more sensitively capturing trend-based drifts. The value of the cumulative offset is processed by a nonlinear compression function consisting of a natural constant and a preset amplitude compression coefficient. This compression function has an approximately linear response when the cumulative offset is small, but gradually saturates when the cumulative offset is large, making the instantaneous drift amplitude sensitive to changes in the early stages of drift and maintaining a stable output when the drift is significant. The hyperbolic tangent operator acts on the intermediate variable consisting of the cumulative offset, the number of consecutive drift batches, and preset parameters, smoothly compressing the intermediate variable to a finite numerical range to limit the upper limit of the inertia enhancement effect. Finally, the instantaneous drift amplitude is formed by the combined effect of the drift direction label, the inertia enhancement term, and the nonlinear compression term.

[0136] Beneficial Effects: By mapping the selenium content measurement values ​​of the current production batch to the slope of the preceding and following neighborhoods, a local rate of change sequence is obtained. The continuous trend within the neighborhood replaces the instantaneous change at a single moment, effectively suppressing the interference of random errors from single-point measurements on drift direction determination. Drift direction labels are obtained by trend-oriented discrimination based on the sign consistency of the local rate of change sequence. A statistical voting mechanism, rather than a simple positive / negative judgment, is used to determine the drift direction, improving the reliability and noise resistance of direction determination. The cumulative offset is obtained by accumulating the amplitude of the local rate of change sequence, transforming discrete rates of change into continuous cumulative offsets. This allows for quantitative expression of the offset degree and eliminates amplitude jumps caused by rate of change fluctuations. Polarity amplitude coupling between the drift direction label and the cumulative offset yields the instantaneous drift amplitude. Simultaneously, the number of batches with consecutive identical drift directions is introduced as a basis for inertia enhancement. This ensures that the instantaneous drift amplitude reflects both the current batch's drift state and incorporates the trend inertia of historical drifts, providing a more sensitive response to persistent drift trends while moderately suppressing occasional fluctuations. The instantaneous drift amplitude is calculated using a hyperbolic tangent operator for nonlinear smoothing and compression, which makes the output value maintain high sensitivity when the drift is small and tend to saturate when the drift is large. This avoids excessive amplification of the control quantity under extreme drift conditions and provides a stable drift metric input that conforms to the physical response law for subsequent nonlinear compensation correction.

[0137] S5: Based on the nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs, the instantaneous drift amplitude is nonlinearly corrected to obtain the stabilization control parameter of the current production batch selenium content.

[0138] In this embodiment of the invention, the nonlinear compensation factor based on the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs, and the nonlinear correction of the instantaneous drift amplitude to obtain the stabilization control parameters of the current production batch selenium content, includes:

[0139] The nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs is divided into equal probability intervals to obtain the coding partition boundary of the nonlinear compensation factor;

[0140] Based on the penetration depth of the instantaneous drift amplitude within the coding partition boundary, the instantaneous drift amplitude is subjected to interval membership determination to obtain the segment label to which the instantaneous drift amplitude belongs;

[0141] Based on the attribution segment label, intra-segment interpolation mapping is performed on the instantaneous drift amplitude to obtain the nonlinear compensation estimate of the instantaneous drift amplitude;

[0142] Boundary clamping constraints are applied to the nonlinear compensation estimate to obtain the stabilization control parameters for the selenium content of the current production batch.

[0143] The formula for calculating the nonlinear compensation estimate is as follows:

[0144] ;

[0145] in, The instantaneous drift amplitude, The total number of intervals divided by the coding partition boundary. For the first The center position of the segment corresponding to the boundary of each coded partition For the first The compensation base value pre-stored in each coded partition, The preset attenuation coefficient, Represents the absolute value operator. The preset shape factor is greater than 1. This represents the natural exponential function. The nonlinear compensation estimate is given.

[0146] The nonlinear compensation factor of the steady-state operating sub-interval to which the selenium content measurement value of the current production batch belongs is divided into equal probability intervals. All compensation factor values ​​in the nonlinear compensation factor coding sequence corresponding to the steady-state operating sub-interval are extracted and sorted in ascending order. The cumulative distribution of the sorted value sequence is calculated to find the division point position that can evenly divide all compensation factor values ​​into several equal parts. The value range between two adjacent division points constitutes a coding partition. The boundary values ​​of all coding partitions are arranged in order to form the coding partition boundary of the nonlinear compensation factor. The coding partition boundary divides the entire possible value range of the nonlinear compensation factor into several continuous and non-overlapping value interval segments.

[0147] The instantaneous drift amplitude is used to determine its interval membership based on the penetration depth within the coding partition boundary. The instantaneous drift amplitude value is extracted and compared with the upper and lower limits of each interval segment within the coding partition boundary to determine which coding partition the instantaneous drift amplitude value falls into. The ratio of the distance between the instantaneous drift amplitude value and the upper and lower boundaries of the coding partition is calculated as the penetration depth measure. The degree of membership of the instantaneous drift amplitude in the coding partition is determined based on the magnitude of the penetration depth. The closer the penetration depth is to the center of the interval, the higher the degree of membership. The coding partition number into which the instantaneous drift amplitude falls is recorded as the segment label of the instantaneous drift amplitude.

[0148] Intra-region interpolation mapping of instantaneous drift amplitude is performed based on the home segment label. A pre-stored compensation base value of the coding partition pointed to by the home segment label is retrieved. At the same time, the pre-stored compensation base values ​​of the preceding and following coding partitions adjacent to this coding partition are retrieved. The relative position of the instantaneous drift amplitude value in the home coding partition is used as the interpolation ratio. Linear or non-linear numerical interpolation calculations are performed between the compensation base value of the home coding partition and the compensation base values ​​of the adjacent coding partitions according to the interpolation ratio. This ensures that when the instantaneous drift amplitude is located at the center of the home coding partition, the interpolation result approaches the compensation base value of that partition. When the instantaneous drift amplitude is close to the partition boundary, the interpolation result smoothly transitions to the compensation base value of the adjacent partition. The value obtained by the interpolation calculation is the non-linear compensation estimate of the instantaneous drift amplitude.

[0149] Boundary clamping constraints are applied to the nonlinear compensation estimate. The calculated nonlinear compensation estimate is compared with the maximum and minimum values ​​of the historical compensation amount in the steady-state operation sub-interval. If the nonlinear compensation estimate is greater than the maximum historical compensation amount, it is forcibly truncated to the maximum historical compensation amount. If the nonlinear compensation estimate is less than the minimum historical compensation amount, it is forcibly truncated to the minimum historical compensation amount. If the nonlinear compensation estimate is between the maximum and minimum values, it remains unchanged. The value after boundary clamping constraint processing is the stabilization control parameter of the selenium content of the current production batch. This stabilization control parameter will be directly used to guide the adjustment operation of selenium content during the production process.

[0150] In the calculation of the nonlinear compensation estimate, the instantaneous drift amplitude is used as an independent variable. The total number of intervals divided by the coding partition boundary determines the resolution of the compensation mapping. Each coding partition corresponds to a pre-stored compensation base value, which is the statistical result of the optimal compensation amount under different deviations within the steady-state sub-interval during the historical data learning phase. The distance between the instantaneous drift amplitude and the center position of each coding partition is used to calculate a distance decay weight. The farther the coding partition is, the more significant the weight decay. The natural exponential function is used to convert the distance into a smoothly decaying weight coefficient. The preset decay coefficient controls the rate at which the weight decays with increasing distance. The shape factor is greater than one, which makes the distance measurement adopt a nonlinear measurement method greater than Euclidean distance, enhancing the weight contribution of closer coding partitions and suppressing the interference of farther coding partitions. The compensation base values ​​corresponding to all coding partitions are weighted and summed according to their distance decay weights, and then divided by the sum of all weights to obtain a compensation estimate based on spatial distance weighting. This compensation estimate realizes a smooth nonlinear mapping relationship from instantaneous drift amplitude to compensation amount.

[0151] Beneficial Effects: By dividing the nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch's selenium content measurement value belongs into equiprobable intervals, the coded partition boundaries are obtained. The continuously changing compensation factor is discretized into a finite number of coded partitions with statistical equilibrium, which not only preserves the main distribution characteristics of the compensation factor but also reduces the complexity of subsequent mapping calculations. Based on the penetration depth of the instantaneous drift amplitude within the coded partition boundaries, interval membership discrimination is performed to obtain the belonging segment label. The relationship between the instantaneous drift amplitude and each coded partition is measured using penetration depth rather than hard boundary attribution, making the membership discrimination both fuzzy and continuous, avoiding abrupt changes in compensation amount due to small numerical changes near the partition boundaries. Based on the belonging segment label, intra-regional interpolation mapping of the instantaneous drift amplitude is performed to obtain the nonlinear compensation prediction value. Weighted fusion is performed using the distance attenuation weight between the instantaneous drift amplitude and the center position of each coded partition, making the compensation prediction value smoothly transition between coded partitions and accurately fitting the nonlinear response surface of selenium content drift compensation.

[0152] By applying boundary clamping constraints to the nonlinear compensation estimate, stable control parameters are obtained. Limiting the compensation amount to a historically reasonable range effectively prevents excessive or insufficient output of compensation due to abnormal instantaneous drift amplitude, ensuring the physical feasibility and safety of the control parameters. The overall process organically combines the instantaneous drift amplitude of the current batch with the nonlinear compensation factor learned a priori from the steady-state operating sub-interval. This fully utilizes the compensation experience rules of the historical stable interval and achieves generalized compensation mapping for unseen drift amplitudes through distance-attenuation weighted interpolation. The final output stable control parameters are both targeted and robust, accurately guiding the real-time adjustment of selenium content during the production of selenium-enriched konjac functional foods, significantly improving the control accuracy and production stability of batch-to-batch selenium content consistency.

[0153] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0154] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for controlling the stability of selenium content in konjac functional foods, characterized in that, The method includes: S1: Based on atomic spectroscopy, multiple batches of historical samples of selenium content in selenium-enriched konjac functional food were continuously collected to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food. S2: Perform time-series fluctuation entropy measurement on the original time-series observation sequence to obtain the fluctuation disorder of the selenium content of the selenium-enriched konjac functional food. Based on the fluctuation disorder, perform dynamic weight allocation on the original time-series observation sequence to construct an adaptive weighted observation sequence of the selenium content of the selenium-enriched konjac functional food. S3: Based on the adaptive weighted observation sequence, the fluctuation period of selenium content in the selenium-enriched konjac functional food is recursively segmented to obtain the steady-state operating sub-interval of the fluctuation period. The selenium content in the steady-state operating sub-interval is deviated and compensated to obtain the nonlinear compensation factor of the steady-state operating sub-interval. S4: Identify the drift trend of the selenium content measurement value of the current production batch of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the selenium content measurement value of the current production batch. S5: Based on the nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs, the instantaneous drift amplitude is nonlinearly corrected to obtain the stabilization control parameter of the current production batch selenium content.

2. The method for controlling the stability of selenium content in konjac functional food as described in claim 1, characterized in that, The method, based on atomic spectroscopy, involves continuous data collection from multiple batches of historical samples of selenium content in selenium-enriched konjac functional foods, resulting in the original time-series observation sequence of selenium content in these foods. This sequence includes: Based on the temporal drift characteristics of atomic spectral excitation sources, the spectral temporal drift decoupling of multiple batches of historical samples of selenium content in selenium-enriched konjac functional food was performed to obtain the stable absorbance sequence of the multiple batches of historical samples. The stabilized absorbance sequence is resampled at equal intervals to obtain the time-interval observation sample set of the multiple batches of historical samples; Local density peaks are selected from the equidistant observation sample set to obtain representative selenium content measurements from the multiple batches of historical samples; By performing sliding window residual outlier stripping on the representative selenium content measurement, the historical selenium content observation chain of the multiple batches of historical samples is obtained. The historical selenium content observation chain was reconstructed by phase alignment of discontinuities to obtain the original time-series observation sequence of selenium content in the selenium-enriched konjac functional food.

3. The method for controlling the stability of selenium content in konjac functional food as described in claim 1, characterized in that, The process involves measuring the temporal fluctuation entropy of the original time-series observation sequence to obtain the fluctuation disorder of the selenium content in the selenium-enriched konjac functional food. Based on this fluctuation disorder, a dynamic weight allocation is performed on the original time-series observation sequence to construct an adaptive weighted observation sequence for the selenium content of the selenium-enriched konjac functional food, including: The original time-series observation sequence is progressively windowed to obtain the observation subsequence of the original time-series observation sequence; The observed subsequence is projected into the state space to obtain the projected feature string of the observed subsequence; Self-information assignment is performed on the projected feature string to obtain the fluctuation entropy estimate of the projected feature string; Based on the fluctuation entropy estimation, the original time series observation sequence is reconstructed and assigned values ​​inversely to obtain the adaptive weight coefficients of the original time series observation sequence; Based on the adaptive weighting coefficients, the original time-series observation sequences are numerically multiplied and superimposed to obtain the adaptive weighted observation sequence of selenium content in the selenium-enriched konjac functional food.

4. The method for controlling the stability of selenium content in konjac functional food as described in claim 3, characterized in that, The step of assigning self-information to the projected feature string to obtain the fluctuation entropy estimate of the projected feature string includes: The frequency of occurrence of the projected feature string is obtained by enumerating the repetition patterns of the projected feature string. The frequency of appearance is rounded down by performing a negative logarithmic transformation to obtain the original score of the self-information of the projected feature string; The original self-information score is reshaped by probability normalization to obtain the information weight of the projected feature string; The information weights are weighted, accumulated, and aggregated to obtain the fluctuation entropy estimate of the projected feature string.

5. The method for controlling the stability of selenium content in konjac functional food as described in claim 3, characterized in that, The step of reconstructing and assigning values ​​to the original time-series observation sequence based on the fluctuation entropy estimation to obtain the adaptive weight coefficients of the original time-series observation sequence includes: The fluctuation entropy estimate is numerically reversed to obtain the entropy reciprocal sequence of the original time series observation sequence; The entropy reciprocal sequence is subjected to amplitude normalization stretching to obtain the weighted basis of the entropy reciprocal sequence; Based on the weighted basis, the original time-series observation sequence is inversely weighted and embedded to obtain the initial weight coefficients of the original time-series observation sequence; Neighborhood smoothing suppression is applied to the initial weight coefficients to obtain the adaptive weight coefficients of the original time-series observation sequence.

6. The method for controlling the stability of selenium content in konjac functional food as described in claim 1, characterized in that, Based on the adaptive weighted observation sequence, the fluctuation period of selenium content in the selenium-enriched konjac functional food is recursively segmented to obtain a steady-state operating sub-interval of the fluctuation period. Deviation compensation mapping is then performed on the selenium content within the steady-state operating sub-interval to obtain the nonlinear compensation factor for the steady-state operating sub-interval, including: Phase space delay reconstruction is performed on the adaptive weighted observation sequence to obtain the multidimensional embedded trajectory of the adaptive weighted observation sequence; Extreme point orientation tracking is performed on the multidimensional embedded trajectory to obtain the inflection point marker sequence of the multidimensional embedded trajectory; Based on the inflection point marker sequence, the adaptive weighted observation sequence is subjected to interval bisection iterative cutting to obtain the periodic segments of the adaptive weighted observation sequence; The periodic segments are subjected to a fluctuation amplitude homogeneity test, and the periodic segments that pass the homogeneity test are marked as steady-state operating sub-intervals; The target deviation is decoupled from the adaptive weighted observation sequence of the steady-state operating sub-interval to obtain the deviation residual vector of the steady-state operating sub-interval; The deviation residual vector is subjected to nonlinear amplitude compression encoding to obtain the nonlinear compensation factor of the steady-state operating sub-interval.

7. The method for controlling the stability of selenium content in konjac functional food as described in claim 1, characterized in that, The step of identifying the drift trend of the selenium content measurement value of the current production batch of the selenium-enriched konjac functional food to obtain the instantaneous drift amplitude of the selenium content measurement value of the current production batch includes: The selenium content of the current production batch of the selenium-enriched konjac functional food is mapped by the slope of the previous and next neighborhoods to obtain the local change rate sequence of the selenium content measurement value of the current production batch. Based on the sign consistency of the local rate of change sequence, the trend orientation of the selenium content measurement value of the current production batch is determined to obtain the drift direction label of the selenium content measurement value of the current production batch. By performing amplitude cumulative integral on the local rate of change sequence, the total cumulative offset of the selenium content measurement value of the current production batch is obtained; The instantaneous drift amplitude of the selenium content measurement value of the current production batch is obtained by polarity amplitude coupling between the drift direction label and the cumulative offset.

8. The method for controlling the stability of selenium content in konjac functional food as described in claim 7, characterized in that, The formula for calculating the instantaneous drift amplitude is as follows: ； In the formula, The drift direction label, This is the total cumulative offset. This represents the number of batches that drifted in the same direction consecutively before the current moment. The preset inertia enhancement coefficient, The preset directional attenuation factor, The preset amplitude compression factor, For hyperbolic tangent operator, The instantaneous drift amplitude, It is a natural constant. This represents the absolute value, and 1 represents the base scaling factor.

9. The method for controlling the stability of selenium content in konjac functional food as described in claim 1, characterized in that, The nonlinear compensation factor based on the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs is used to nonlinearly correct the instantaneous drift amplitude, thereby obtaining the stabilization control parameters for the selenium content of the current production batch, including: The nonlinear compensation factor of the steady-state operating sub-interval to which the current production batch selenium content measurement value belongs is divided into equal probability intervals to obtain the coding partition boundary of the nonlinear compensation factor; Based on the penetration depth of the instantaneous drift amplitude within the coding partition boundary, the instantaneous drift amplitude is subjected to interval membership determination to obtain the segment label to which the instantaneous drift amplitude belongs; Based on the attribution segment label, intra-segment interpolation mapping is performed on the instantaneous drift amplitude to obtain the nonlinear compensation estimate of the instantaneous drift amplitude; Boundary clamping constraints are applied to the nonlinear compensation estimate to obtain the stabilization control parameters for the selenium content of the current production batch.

10. The method for controlling the stability of selenium content in konjac functional food as described in claim 9, characterized in that, The formula for calculating the nonlinear compensation estimate is as follows: ； in, The instantaneous drift amplitude, The total number of intervals divided by the coding partition boundary. For the first The center position of the segment corresponding to the boundary of each coded partition For the first The compensation base value pre-stored in each coded partition, The preset attenuation coefficient, Represents the absolute value operator. The preset shape factor is greater than 1. This represents the natural exponential function. The nonlinear compensation estimate is given.

Images