A multi-parameter coordinated control system for beer brewing

Abstract

This invention relates to the technical field of beer brewing, specifically to a multi-parameter collaborative control system for beer brewing, comprising: a sequence construction module for constructing a multi-dimensional state vector sequence for the current batch based on the brewing process parameters; a confidence identification module for determining the data confidence at each moment; a path determination module for determining the matching path; a sequence repair module for generating a repaired multi-dimensional state vector sequence; and an instruction generation module for generating collaborative control instructions. By constructing a multi-dimensional state vector containing original and derived parameters, and using local statistical features to accurately identify low-confidence data automatically supplemented by the system, the system ultimately performs process stage identification and collaborative control based on the repaired state vector that retains the true micro-fluctuations and derivative features. This solves the problem of control misjudgment caused by data supplementation smoothing out derivative features, ensuring the stability of the brewing process and product quality.

Description

Technical Field

[0001] This invention relates to the technical field of beer brewing, and more particularly to a multi-parameter coordinated control system for beer brewing. Background Technology

[0002] In modern beer brewing, distributed control systems (DCS) or programmable logic controllers (PLCs) rely on continuous, real-time online sensor data streams for process stage identification and automatic control. These sensors typically include temperature sensors, pressure sensors, pH meters, and dissolved oxygen sensors, used to monitor the process status within the fermentation tank in real time. The control system analyzes the numerical changes and dynamic characteristics of this data, such as the rate of change and inflection points, to determine the current process stage, such as the primary fermentation stage, diacetyl reduction stage, or cold storage stage, and triggers corresponding control actions accordingly, such as adjusting the opening of cooling water valves, controlling tank pressure, or starting the agitator.

[0003] However, in actual industrial production environments, continuous acquisition of sensor data is often disrupted by various factors. Short-term network failures, gateway device restarts, communication buffer overflows, or momentary malfunctions of data acquisition cards can all lead to the complete loss of sensor data over a period of time. The duration of data loss can range from a few minutes to tens of minutes. To ensure the data integrity and display continuity of the control system, existing industrial real-time databases or middleware typically employ data completion strategies to automatically fill in missing segments. The most common completion methods include zero-order hold, which continuously fills in with the last valid value before the data loss, and linear interpolation, which linearly connects the valid values ​​before and after the missing segment.

[0004] While the aforementioned data completion methods maintain the visual continuity and structural integrity of the time series in form, their fundamental flaw lies in smoothing out the micro-fluctuations and derivative characteristics inherent in the real data. Even in relatively stable process stages, real sensor data exhibits natural fluctuations due to measurement noise, inherent variations in biochemical reactions, and minor adjustments in the control system; these fluctuations constitute the micro-texture of the data. More importantly, the turning points of process stages are often reflected in changes in the first derivative (rate of change) of the data. For example, a slowing rate of pH decrease indicates the end of primary fermentation, and a change in the rate of temperature increase foreshadows the start of diacetyl reduction. However, the completed data generated by zero-order hold or linear interpolation presents an unnatural, absolutely smooth or linear transition, completely eliminating the original inflection points and derivative characteristics. For control logic that relies on slope or derivative characteristics for process judgment, such an absolutely smooth or linear transition state can lead to serious control misjudgments. For example, the control system originally judges whether the main fermentation has ended by monitoring whether the rate of pH decrease approaches zero. However, during the data completion phase, the smooth pH curve will show a continuous plateau period, causing the system to mistakenly believe that fermentation has stopped, thus triggering heating or stirring actions prematurely and interfering with the normal fermentation process. Similarly, in the diacetyl reduction stage, the smooth temperature curve may mask the actual change in the heating rate, resulting in insufficient or excessive reduction time. Such misjudgments caused by data completion not only affect batch stability but may also lead to a decline in product quality and energy waste. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention proposes a multi-parameter collaborative control system for beer brewing, the specific solution of which is as follows: A multi-parameter coordinated control system for beer brewing includes: A sequence module is constructed to collect the original process parameters at each moment, generate derived process parameters based on the original process parameters, use the original process parameters and derived process parameters together as brewing process parameters, and construct a multidimensional state vector sequence for the current batch based on the brewing process parameters. The confidence identification module is used to determine the data confidence at each time step based on the local statistical characteristics of the data in the multidimensional state vector sequence. The path determination module is used to compare the multidimensional state vector sequence of the current batch with the preset standard benchmark batch sequence, and determine the matching path through a multivariate coupling constraint mechanism. The repair sequence module is used to repair the multidimensional state vector sequence of the current batch on the matching path based on the standard benchmark batch sequence and the data confidence, and generate a repaired multidimensional state vector sequence. The instruction generation module is used to generate coordinated control instructions based on the repaired multidimensional state vector sequence.

[0006] Furthermore, in the sequence construction module, the multidimensional state vector sequence is composed of multidimensional state vectors at multiple times, and the multidimensional state vector at each time is composed of multiple brewing process parameters collected at the current time.

[0007] Furthermore, the original process parameters include at least one of temperature, pressure, pH value, and dissolved oxygen; the derived process parameters include the first derivative of the pH value.

[0008] Furthermore, in the confidence identification module, based on the local statistical characteristics of the data in the multidimensional state vector sequence, the data confidence at each time step is determined as follows: Calculate the local variance at each time step; The local variance is compared with a preset threshold. If the local variance is less than the preset threshold, the data at the current moment is determined to be low-confidence data.

[0009] Furthermore, in the path determination module, a multivariate coupling constraint mechanism is used to determine the matching path, as follows: Based on the multidimensional state vector sequence of the current batch, generate the covariance matrix of the standard benchmark batch sequence; Based on the covariance matrix, calculate the weighted distance between the multidimensional state vector sequence of the current batch and the standard baseline batch sequence; Using the weighted distance as the path cost, the minimum cumulative cost path from the start point to the end point of the sequence is determined using the dynamic time warping algorithm, and this path is used as the matching path.

[0010] Furthermore, based on the covariance matrix, the weighted distance between the current batch's multidimensional state vector sequence and the standard baseline batch sequence is calculated as follows: The weighted distance is a weighted Mahalanobis distance; the covariance matrix between multiple process parameters in the standard benchmark batch sequence is obtained, and the covariance matrix is ​​used to characterize the coupling relationship between different process parameters; Construct a variable importance matrix, which is a diagonal matrix, and its diagonal elements correspond to the weight coefficients of the multiple process parameters in the matching process. The weighted Mahalanobis distance is calculated based on the inverse of the covariance matrix, the variable importance matrix, and the difference between the multidimensional state vector of the current batch and the corresponding vector in the standard benchmark batch sequence.

[0011] Furthermore, in the repair sequence module, the multidimensional state vector sequence of the current batch is repaired to generate a repaired multidimensional state vector sequence, as follows: When determining the matching path, the weight of the path cost of low-confidence data is reduced, so that the determination of the matching path is mainly determined by high-confidence data. After determining the matching path, the temporal features in the corresponding segment of the standard benchmark batch sequence are mapped onto the low-confidence data of the current batch to generate a repaired multidimensional state vector sequence.

[0012] Furthermore, the time-series characteristics include the data's changing trends and derivative characteristics.

[0013] Furthermore, in the instruction generation module, based on the repaired multidimensional state vector sequence, collaborative control instructions are generated as follows: Based on the repaired multidimensional state vector sequence, the current process stage is identified; Based on the stated process stage, the corresponding process control logic is invoked to generate and output control commands.

[0014] Compared with the prior art, the present invention can achieve at least the following beneficial effects: 1. This invention generates derived process parameters based on original process parameters, incorporating both static values ​​and dynamic change characteristics into a multi-dimensional state vector. This allows subsequent data processing to utilize both the absolute values ​​and trends of the parameters, providing richer information dimensions for accurate identification of process stages. By calculating the local variance at each moment and comparing it with a preset threshold, it can accurately identify non-naturally smooth data segments caused by zero-order preservation or linear interpolation and assign them low-confidence labels. Furthermore, by quantifying the coupling relationship between multiple parameters through the covariance matrix and using weighted distance as the matching cost, the determination of the matching path considers not only the similarity of single-variable values ​​but also whether the coupling relationship between multiple parameters conforms to the standard process. This multi-variable constraint mechanism ensures that even with similar single-variable waveforms... In some cases, incorrect matching paths can be eliminated by the differences in other coupled variables. Using a confidence-weighted dynamic time warping algorithm, high-confidence data dominates the matching path determination. After determining the matching path, the temporal characteristics of the corresponding segment in the standard benchmark batch are mapped to the low-confidence data segment. This allows the original data segment to be smoothly supplemented by the system, and its inherent micro-fluctuations and inflection point characteristics can be recovered through historical knowledge. After data repair, process stage identification and control command generation are performed based on the state vector sequence that retains the correct temporal characteristics after repair. This enables the control system to accurately perceive the current process stage and trigger corresponding control actions at the correct time, fundamentally avoiding misjudgments caused by data supplementation and ensuring the stability of the fermentation process and the consistency of product quality.

[0015] 2. This invention constructs a multidimensional state vector containing original and derived parameters, accurately identifies low-confidence data automatically supplemented by the system using local statistical features, and introduces a multivariate coupling constraint mechanism to eliminate mismatches of recurring modes during the matching process using weighted Mahalanobis distance. Furthermore, through a confidence-weighted dynamic time warping algorithm, the time-series characteristics of the standard benchmark batch are mapped to the low-confidence data segment on the matching path dominated by high-confidence data. Finally, based on the state vector that retains the true micro-fluctuations and derivative characteristics after repair, process stage identification and coordinated control are performed, thereby solving the problem of control misjudgment caused by data supplementation and smoothing of derivative characteristics, ensuring the stability of the brewing process and product quality. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a system principle block diagram of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] Please refer to Figure 1 This invention provides a multi-parameter coordinated control system for beer brewing, comprising: A sequence module is constructed to collect the original process parameters at each moment, generate derived process parameters based on the original process parameters, use the original process parameters and derived process parameters together as brewing process parameters, and construct a multidimensional state vector sequence for the current batch based on the brewing process parameters.

[0019] In an optional embodiment, the original process parameters include at least one of temperature, pressure, pH, and dissolved oxygen; the derived process parameters include the first derivative of the pH value.

[0020] It should be noted that the original process parameters refer to the physicochemical indicators directly collected by online sensors during the beer brewing process. Taking the beer fermentation stage as an example, temperature sensors monitor the temperature changes of the mash in the fermentation tank in real time, typically fluctuating within the range of 8-15℃; pressure sensors monitor the top pressure inside the tank, generally between 0.5-1.5 bar; pH sensors monitor the acidity and alkalinity of the mash, with the pH gradually decreasing from around 5.2 to around 4.2 during the primary fermentation period; and dissolved oxygen sensors monitor the oxygen content in the mash, which is rapidly depleted to near zero after yeast inoculation. These parameters are the basic data sources for process control in the DCS / PLC system.

[0021] It should be noted that derived process parameters refer to quadratic features obtained by mathematically transforming the original process parameters. These parameters are used to enhance the perception of changes in process conditions. In this scheme, the first derivative of pH is taken as an example, i.e., the rate of change of pH over time, ΔpH / Δt. This parameter can sensitively reflect the activity level of the fermentation process. For example, during the start-up phase of the main fermentation, the pH value drops rapidly, and ΔpH shows a large negative value; during the end phase of the main fermentation, the pH value tends to stabilize, and ΔpH approaches zero; during the diacetyl reduction phase, the pH value may rise slightly, and ΔpH shows a positive value. This derivative feature is of great significance for judging fermentation inflection points, while simply relying on the original pH value is insufficient to capture such dynamic changes.

[0022] In an optional embodiment, in the sequence construction module, the multidimensional state vector sequence is composed of multidimensional state vectors at multiple times, and the multidimensional state vector at each time moment is composed of multiple brewing process parameters collected at the current time moment.

[0023] It should be noted that the multidimensional state vector sequence combines multiple brewing process parameters at the same moment into a single vector and arranges them in chronological order to form a sequence. Taking time t as an example, its multidimensional state vector can be represented as follows: ,in Represents temperature. Represents pressure, Represents pH value. Represents dissolved oxygen. The first derivative represents the pH value. This multidimensional vector construction method has the following advantages: First, it preserves the correspondence between parameters at the same moment, laying the foundation for subsequent multivariate coupling analysis; second, it unifies heterogeneous sensor data into vector form, facilitating the application of sequence processing algorithms such as dynamic time warping; third, by introducing derivative features, the vector simultaneously contains static numerical values ​​and dynamic change information, more comprehensively depicting the instantaneous state of the brewing process.

[0024] It's important to note that the process of constructing the multidimensional state vector sequence needs to be matched with the data acquisition frequency. In actual brewing production, the data acquisition frequency of the DCS / PLC system is typically once per minute or higher. Taking once per minute as an example, a fermentation batch lasting 10 days will generate more than 14,000 multidimensional state vectors at various moments, forming a state vector sequence with a dimension of 5×14,000. This sequence completely records the evolution trajectory of each process parameter throughout the entire fermentation process, providing fundamental data support for subsequent data confidence assessment, state matching, and feature repair.

[0025] The confidence identification module is used to determine the data confidence at each time step based on the local statistical characteristics of the data in the multidimensional state vector sequence.

[0026] It is important to note that data confidence is a key indicator used in this solution to distinguish between real sensor data and automatically completed data. In the DCS / PLC system of the beer brewing process, when faults such as network interruption, gateway restart, or buffer overflow occur, sensor data for a period of time may be lost in its entirety. To maintain the continuity of the data stream, industrial databases or middleware typically use zero-order hold (i.e., continuously filling with the last valid value before the loss) or linear interpolation strategies to automatically complete the missing segments. While this completed data ensures the integrity of the time series, its statistical characteristics are fundamentally different from those of real sensor data: even in a relatively stable process stage, real sensor data will contain minor variations caused by factors such as measurement noise and micro-fluctuations; while the automatically completed data exhibits an unnatural, absolutely smooth state, with local variance approaching zero. Therefore, by analyzing the variance information in the local statistical characteristics, these low-confidence completed data segments can be effectively identified and marked.

[0027] In an optional embodiment, the confidence identification module determines the data confidence at each time step based on the local statistical characteristics of the data in the multidimensional state vector sequence, as follows: Calculate the local variance at each time step; It's important to note that calculating local variance requires a sliding window mechanism. A sliding window is a fixed-length time window centered (or ending) on ​​the current moment, encompassing data points from several moments before and after it. The choice of window length needs to consider both the data acquisition frequency and the inherent periodicity of process fluctuations. Taking beer fermentation as an example, the data acquisition frequency is typically once per minute, while the micro-fluctuations of parameters such as temperature and pH values ​​usually occur on the order of several minutes. Therefore, the sliding window length can be set to 5-15 minutes, which captures short-term fluctuations without smoothing out the true process changes due to an excessively large window. For each time point ti, all data points within the window are taken, and the variance of these data points is calculated; this is the local variance at that time point. This local variance value reflects the degree of fluctuation in the data around that moment.

[0028] The local variance is compared with a preset threshold. It should be noted that the preset threshold corresponds to the sensor noise floor threshold. Sensor noise floor refers to the random fluctuations that still exist in the sensor output signal when the measured physical quantity is completely constant; its amplitude is determined by the sensor's own accuracy and resolution. Taking an industrial pH sensor as an example, its measurement accuracy is typically ±0.02 pH, and the fluctuation range of the output signal under steady-state conditions is approximately 0.01 to 0.03 pH, corresponding to a variance on the order of magnitude of... Between. This noise floor threshold can be calibrated by measuring the standard buffer solution under laboratory conditions for an extended period of time, or it can be obtained statistically from historical data within a period of stable process and with complete known data. The local variance at a given moment... When the data is significantly lower than this noise floor threshold (e.g., less than 1 / 10 of the noise floor threshold), it indicates that the data fluctuation level near that moment is even lower than the sensor's inherent noise level. This is an extremely unnatural phenomenon and strongly suggests that the data segment is a false smoothed data generated by automatic padding.

[0029] If the local variance is less than the preset threshold, the data at the current moment is determined to be low-confidence data.

[0030] It's important to note that the determination of low-confidence data directly impacts subsequent data processing. When data is classified as low-confidence, it's assigned a very small confidence weight (e.g., Wt=0.01), while normally fluctuating data is assigned a high confidence weight (e.g., Wt=1). This confidence weight plays a crucial role in the subsequent Confidence-Weighted Dynamic Time Warping (FW-DTW) algorithm: for low-confidence data segments, the algorithm significantly reduces the weight of their numerical differences in path cost calculation, ensuring that the search for matching paths is primarily determined by the preceding and following high-confidence anchor data segments. This mechanism ensures that the algorithm effectively ignores falsely smoothed segments that are automatically filled in, instead relying on reliable real data to find the correct process stage mapping relationship, thus laying the foundation for subsequent feature repair.

[0031] It should be noted that this confidence assessment method based on local variance is not only applicable to identifying data that has been filled in by the system after a complete data segment was missing, but can also effectively identify other types of abnormal data. For example, when a sensor malfunctions and causes the output signal to freeze at a certain value, the local variance within that time period will also approach zero, thus it will be correctly identified as low-confidence data. Similarly, when intermittent interruptions in data communication cause some data points to be lost and filled in by interpolation, although the interpolated segment may not be perfectly smooth, its local variance is still significantly lower than the normal fluctuation level, and it can also be effectively detected. The universality of this method enables this invention to address various data quality problems and significantly improves the robustness of the system.

[0032] The path determination module is used to compare the multidimensional state vector sequence of the current batch with the preset standard benchmark batch sequence and determine the matching path through a multivariate coupling constraint mechanism.

[0033] It should be noted that the standard benchmark batch sequence is a multi-dimensional state vector sequence of one or more historically excellent batches pre-stored in this solution, serving as a reference template for state matching and feature repair of the current batch. The selection of the standard benchmark batch must meet the following conditions: first, the brewing process of this batch was successful, and the final product quality met the standards; second, the sensor data of this batch is complete, without missing or abnormal data; and third, the process trajectory of this batch is representative and can reflect typical variation patterns under normal production conditions. In practical applications, multiple historically excellent batches can be selected, and a standard batch template can be constructed through alignment and averaging methods, or a recognized benchmark batch can be directly selected as the standard benchmark. The establishment of the standard benchmark batch sequence is equivalent to establishing a standard state fingerprint database for each process stage of the brewing process, providing a reference basis for subsequent comparison and matching.

[0034] In an optional embodiment, the path determination module determines the matching path through a multivariate coupling constraint mechanism, as follows: Based on the multidimensional state vector sequence of the current batch, generate the covariance matrix of the standard benchmark batch sequence; Based on the covariance matrix, calculate the weighted distance between the multidimensional state vector sequence of the current batch and the standard baseline batch sequence; Using the weighted distance as the path cost, the minimum cumulative cost path from the start point to the end point of the sequence is determined using the dynamic time warping algorithm, and this path is used as the matching path.

[0035] It's important to note that the multivariate coupling constraint mechanism addresses the problem of reproducible process mode interference in beer fermentation. Taking high-concentration dilution brewing as an example, this process may involve multiple sugar additions, each resulting in a similar decrease-resurgence waveform in the saccharide / specific gravity curve. Traditional univariate matching algorithms rely solely on the saccharide curve for state identification, easily misinterpreting the current second sugar addition as the historical first, leading to control timing chaos. However, with multivariate coupling constraints, the situation is entirely different: although the saccharide curve waveform is similar between the two additions, the coupling relationship between auxiliary variables such as alcohol concentration (accumulated increase), pH value (decreased), and dissolved oxygen consumption rate (slowed down) and saccharide changes fundamentally. This multivariate coupling relationship constitutes a unique process fingerprint, effectively eliminating the ambiguity of univariate matching.

[0036] It should be noted that the covariance matrix is ​​used in this invention to quantify the strength and direction of the coupling relationship between different process parameters. Taking the primary fermentation stage of beer as an example, temperature and pressure usually show a positive correlation (increased temperature leads to increased tank pressure), and the covariance is positive; pH and dissolved oxygen may show a specific phase relationship, and the covariance reflects the degree of synchronous fluctuation between the two; sugar content and alcohol concentration show a near-perfect negative correlation, and the absolute value of the covariance is large. The covariance matrix Σ is an n×n symmetric matrix (n is the dimension of the multidimensional state vector), and its diagonal elements... Represents the variance of the i-th parameter, with off-diagonal elements. This represents the covariance between the i-th and j-th parameters. Essentially, this matrix is ​​a statistical description of the inherent relationships between various physical quantities during the brewing process, reflecting how these parameters should coordinate and change under normal process fluctuation patterns. When the relationship between the parameters in a data segment contradicts the coupling pattern described by the covariance matrix, that segment is judged as an anomaly or a mismatch.

[0037] It's important to note that the variable importance matrix M is another key design element of this invention, used to assign differentiated weights to different process parameters during the matching process. In actual brewing, the sensitivity of each parameter varies across different process stages. For example, during the primary fermentation start-up stage, the rate of dissolved oxygen consumption is more important than its absolute value; during the diacetyl reduction stage, subtle changes in pH are more indicative than temperature; and when distinguishing between the first and second sugar additions, the cumulative alcohol concentration is more discriminative than the sugar content itself. The variable importance matrix M is designed to reflect these differences: it is a diagonal matrix where diagonal elements m11, m22, ..., mnn correspond to the importance of parameters such as temperature, pressure, pH, dissolved oxygen, and ΔpH during the matching process. These weighting coefficients can be set based on the experience of process experts or automatically learned from historical data using machine learning methods. By introducing the variable importance matrix, the matching algorithm can focus on the more discriminative key parameters based on the characteristics of the current process stage, thereby improving the accuracy of the matching.

[0038] In an optional embodiment, based on the covariance matrix, the weighted distance between the current batch's multidimensional state vector sequence and the standard baseline batch sequence is calculated as follows: The weighted distance is a weighted Mahalanobis distance; the covariance matrix between multiple process parameters in the standard benchmark batch sequence is obtained, and the covariance matrix is ​​used to characterize the coupling relationship between different process parameters; Construct a variable importance matrix, which is a diagonal matrix, and its diagonal elements correspond to the weight coefficients of the multiple process parameters in the matching process. The weighted Mahalanobis distance is calculated based on the inverse of the covariance matrix, the variable importance matrix, and the difference between the multidimensional state vector of the current batch and the corresponding vector in the standard benchmark batch sequence.

[0039] It should be noted that the calculation process of the weighted Mahalanobis distance incorporates information from the two matrices mentioned above.

[0040] For the multidimensional state vector of the current batch at a certain moment State vector at the corresponding time in the standard benchmark batch First, calculate the difference vector between the two. This difference vector reflects the degree of deviation between the current batch and the standard batch in various parameters. Then, it is calculated using the inverse of the covariance matrix. A decorrelation transformation is performed on the difference vector to eliminate the influence of coupling relationships between parameters. Finally, a variable importance matrix M is introduced to weight the decorrelated differences. The complete calculation formula is as follows: This distance metric considers not only the numerical differences of the parameters, but also whether these differences conform to a normal coupling pattern. Even if two vectors have similar values ​​for individual parameters, if their coupling pattern does not conform to the pattern described by the covariance matrix, the calculated Mahalanobis distance will still be large, thus effectively avoiding mismatches caused by univariate similarity.

[0041] It's important to note that the dynamic time warping algorithm here addresses the nonlinear correspondence between two sequences on the time axis. In actual brewing, fermentation speeds may differ between batches: some batches ferment quickly in the early stages and slowly in the later stages; others ferment in the opposite way. If a simple point-to-point Euclidean distance is used for matching, this stretching and compression on the time axis can lead to significant matching errors. The dynamic time warping algorithm, by allowing stretching and compression on the time axis, can find the optimal nonlinear alignment path between two sequences. In this invention, the algorithm uses weighted Mahalanobis distance as the path cost, starting from the beginning of the sequence and progressively calculating the minimum cumulative cost to reach each position, ultimately obtaining the optimal path from the beginning to the end. This path is the matching path, indicating which time period in the standard baseline batch each moment (or segment) of the current batch corresponds to. It is particularly noteworthy that, because the weighted Mahalanobis distance introduces multivariate coupling constraints, the algorithm automatically avoids erroneous paths that, while similar in univariate variables, do not match in multivariate coupling when searching for the optimal path, thus ensuring the accuracy of the matching path.

[0042] It's important to note that the matching path determined through the above steps has significant dual value. Firstly, it establishes a state lock between the current batch and the standard batch, enabling the control system to accurately determine which stage of the standard process trajectory it's currently in. Even if the original data of the current batch contains low-confidence filler segments, the algorithm can find the correct mapping relationship through the preceding and following anchor points. Secondly, this matching path lays the foundation for subsequent feature repair. Once the corresponding position of each low-confidence segment in the current batch in the standard batch is determined, the temporal features of that segment in the standard batch can be transferred to the current batch, achieving effective repair of missing features.

[0043] The repair sequence module is used to repair the multidimensional state vector sequence of the current batch on the matching path based on the standard benchmark batch sequence and the data confidence level, and generate a repaired multidimensional state vector sequence.

[0044] It's important to note that low-confidence data refers to data segments deemed spuriously smoothed by the system's automatic completion. These segments are characterized by local variances far below the sensor's noise floor threshold, exhibiting an unnatural, absolutely smooth state. While this type of data maintains numerical continuity over time, its inherent micro-fluctuations and derivative characteristics have been completely smoothed out. Directly using it for process status assessment and control decisions can easily lead to misjudgments. For example, consider the temperature curve during the primary fermentation stage of beer. Real data typically exhibits slight sawtooth fluctuations due to the dynamic balance between fermentation heat release and cooling control, while the zero-order hold data automatically completed by the system is a flat straight line. If the control system relies on the slope to determine fermentation activity, the straight line segment might be misjudged as fermentation stagnation, potentially triggering incorrect heating or stirring operations. Therefore, this solution uses data confidence weights to clearly distinguish between low-confidence and high-confidence data and treats them differently in subsequent processing.

[0045] In an optional embodiment, the repair sequence module repairs the multidimensional state vector sequence of the current batch to generate a repaired multidimensional state vector sequence, as follows: When determining the matching path, the weight of the path cost of low-confidence data is reduced, so that the determination of the matching path is mainly determined by high-confidence data. After determining the matching path, the temporal features in the corresponding segment of the standard benchmark batch sequence are mapped onto the low-confidence data of the current batch to generate a repaired multidimensional state vector sequence.

[0046] It should be noted that traditional DTW algorithms typically use Euclidean or Mahalanobis distance directly when calculating the matching cost between two sequence points, treating all data points equally. When low-confidence padding exists in the sequence, these spurious smoothing points can interfere with path searching, potentially causing the algorithm to deviate from the correct path in an attempt to match these meaningless points. This invention addresses this problem by introducing confidence weights into the path cost calculation: for high-confidence data points (… ), maintaining the original weighted Mahalanobis distance as a cost; for low-confidence data points ( ,like The cost of this point is multiplied by a very small coefficient, making the matching cost of that point approach zero. This means that when searching for the path with the minimum cumulative cost, the algorithm can skip these low-confidence points almost for free, thus forcing the matching path to be mainly determined by the preceding and following high-confidence anchor points.

[0047] It's important to note that this confidence-weighted path search mechanism has a clear physical meaning. Taking the beer fermentation process as an example, suppose the current batch suffers data loss due to a network failure between time t1 and t2. The system uses zero-order hold-up to complete the data, forming a smooth straight line. Meanwhile, the data from the time periods before and after this batch are normal and have high confidence. When registering with a standard baseline batch using the FW-DTW algorithm, the algorithm finds that attempting to match the smooth segment of the current batch with a non-smooth segment of the standard baseline batch incurs a significant distance cost; however, ignoring this smooth data and directly aligning the high-confidence segments before and after with their corresponding segments in the standard baseline batch results in a lower total cost. Because the matching cost of the smooth segment is reduced by weight, the algorithm tends to choose the latter path, successfully mapping the high-confidence segment of the current batch to the correct stage in the standard baseline batch. This process is equivalent to locking the global state using reliable anchor data, while effectively bypassing unreliable data segments.

[0048] It's important to note that once the optimal matching path is determined, meaning the corresponding time period in the standard baseline batch for each low-confidence segment in the current batch is clear, feature mapping repair can proceed. The basic idea behind feature mapping repair is that while the low-confidence segments in the current batch have lost their true fluctuation information, the data for the corresponding time period in the standard baseline batch is complete and reliable, containing the micro-fluctuation characteristics and trends expected for that process stage. Therefore, the time-series features of this segment in the standard baseline batch can be transferred to the low-confidence segments of the current batch, thereby recovering reasonable data that conforms to the process rules. In practice, features such as the mean, variance, and first derivative sequence of each parameter within the corresponding segment of the standard baseline batch can be extracted. Then, using the actual measured values ​​of the start and end points of the low-confidence segments in the current batch as a benchmark, the repaired data can be generated through interpolation or waveform synthesis. For example, if the temperature of the current batch is 12.0℃ at time t1 and 11.5℃ at time t2, while the standard reference batch shows a linear decreasing trend during the corresponding time period, a repair curve can be generated that linearly decreases from 12.0℃ to 11.5℃. If the standard reference batch has slight periodic fluctuations during this time period, these fluctuations can also be superimposed on the repair curve to make it closer to the output characteristics of the real sensor.

[0049] In an optional embodiment, the time-series characteristics include the data's trend of change and derivative characteristics.

[0050] It should be noted that, in this invention, temporal characteristics specifically refer to the dynamic pattern of data changes over time, including at least two aspects: trends and derivative characteristics. Trends refer to the overall direction of parameters on the time axis, such as whether temperature rises, falls, or remains stable; derivative characteristics are the rate of change information reflected by the first and higher-order derivatives. In beer brewing, these temporal characteristics have significant process indication value. Taking pH as an example, its trend can determine the fermentation process: during the initial stage of primary fermentation, pH drops rapidly, and the first derivative is a large negative value; during the final stage of primary fermentation, pH tends to stabilize, and the first derivative approaches zero; during the diacetyl reduction stage, pH may rise slightly, and the first derivative is positive. If missing data segments are only filled in using zero-order maintenance, these derivative characteristics will be completely lost, and the control system will be unable to perceive changes in the process stages. Through feature mapping repair, the derivative characteristics are restored, and the control system can regain accurate perception of the process status based on the repaired data.

[0051] It should be noted that the multidimensional state vector sequence after feature mapping repair is no longer a simple continuation of the original sensor data, but a high-quality data product incorporating historical knowledge. The repaired sequence has the following characteristics: First, numerically, the endpoints of the low-confidence segment remain continuous with the actual measured values, ensuring overall data consistency; second, in terms of dynamic characteristics, the low-confidence segment is endowed with the same micro-fluctuations and derivative characteristics as the standard baseline batch, making its statistical properties indistinguishable from the actual sensor data; third, regarding multivariate coupling relationships, since the mapping process is performed synchronously on the entire multidimensional state vector, the coupling relationships between various parameters are also maintained. Such repaired data can be directly used for subsequent process stage identification and control decisions, with a reliability comparable to the actual data.

[0052] It should be noted that the generated, repaired multidimensional state vector sequence provides a reliable data foundation for the final coordinated regulation. Subsequently, the control system will identify the current process stage based on the repaired sequence and trigger corresponding control commands. Because the repaired sequence eliminates the spurious smoothing effect caused by the system's default completion and restores the proper micro-fluctuation characteristics, the identification of process stages will be more accurate, and the timing of control actions will be more reasonable. For example, in the pH repair data, the algorithm can correctly capture the moment when the pH decrease rate slows down, thus accurately determining the end of the main fermentation and promptly initiating the diacetyl reduction incubation program, avoiding misjudgments and delays caused by smoothed data.

[0053] The instruction generation module is used to generate coordinated control instructions based on the repaired multidimensional state vector sequence.

[0054] It's important to note that beer fermentation is typically divided into several consecutive stages, such as: yeast inoculation, early primary fermentation (active fermentation phase), mid-primary fermentation (high-foaming phase), late primary fermentation (sugar reduction phase), diacetyl reduction, and cold storage. Each stage has different control targets and strategies for parameters like temperature, pressure, and pH. For example, the early primary fermentation stage requires maintaining a suitable temperature to promote yeast growth, usually controlled at 9-11°C; the late primary fermentation stage requires a slight increase in temperature to accelerate diacetyl reduction, usually to 12-14°C; and after the diacetyl reduction phase, the temperature needs to be rapidly reduced to 0-2°C for cold storage. Accurately identifying the current stage is crucial for triggering correct control actions. Traditional identification methods rely on raw sensor data. If a segment of data is missing and automatically smoothed out by the system, derivative characteristics are lost, leading to misjudgment of the stage. For instance, the diacetyl reduction phase, which should involve temperature increases, might be misjudged as still being in the primary fermentation phase, thus missing the optimal adjustment window.

[0055] It should be noted that, compared with the original data, the repaired sequence has the following advantages: First, the micro-fluctuations and derivative features of the low-confidence segment are recovered, enabling the algorithm to accurately capture the dynamic changes of parameters; second, through multivariate coupling constraint matching, the repaired sequence is correctly aligned with the standard benchmark batch on the time axis, making the current position on the standard process trajectory clearly known; third, the repaired sequence maintains the normal coupling relationship between various parameters, avoiding misjudgments caused by univariate anomalies. Therefore, stage identification based on the repaired sequence has higher accuracy and robustness.

[0056] In an optional embodiment, the instruction generation module generates coordinated control instructions based on the repaired multidimensional state vector sequence, as follows: Based on the repaired multidimensional state vector sequence, the current process stage is identified; Based on the stated process stage, the corresponding process control logic is invoked to generate and output control commands.

[0057] It should be noted that there are several methods for implementing process stage identification. A preferred approach is to directly determine the current stage by matching the repaired sequence with the standard baseline batch sequence. Since the optimal matching path between the current batch and the standard baseline batch has been found using the FW-DTW algorithm, this path clearly indicates which time point in the standard baseline batch corresponds to the current moment. The standard baseline batch itself already labels the process stage to which each time point belongs, for example, through manual annotation or historical data clustering. Therefore, simply querying the stage label for the corresponding time point in the standard baseline batch is sufficient to determine the current process stage. Another approach is to use a threshold method or machine learning classifier: based on whether key parameters in the repaired sequence (such as pH decrease rate, sugar consumption rate, alcohol accumulation, etc.) reach a preset threshold, or by inputting the repaired sequence fragment into a pre-trained stage classification model to output the probability distribution of the current stage. Regardless of the method used, the high-quality data of the repaired sequence ensures the reliability of the identification results.

[0058] It should be noted that coordinated control commands refer to control signals issued simultaneously to multiple actuators (such as cooling water valves, heaters, pressurization valves, agitators, etc.) to achieve coordinated adjustment of multiple parameters. Beer brewing is a highly coupled multivariate process, and adjusting a single parameter often affects other parameters. For example, opening a cooling water valve to lower the temperature will cause a drop in tank pressure, which in turn affects the solubility of dissolved oxygen; if the temperature is adjusted alone without considering pressure compensation, it may cause process fluctuations. Therefore, this invention emphasizes coordinated control, that is, after identifying the current process stage, the control logic invoked is a set of pre-designed multivariate control strategies that can simultaneously adjust multiple parameters, making them change coordinatedly according to a preset trajectory. For example, when the diacetyl reduction stage is identified, the control logic may simultaneously execute: gradually increasing the cooling water temperature setpoint (heating), appropriately increasing the tank pressure to suppress foam (pressurization), and turning on the agitator to promote yeast suspension (stirring). These actions coordinately complete the stage transition.

[0059] It should be noted that process control logic is typically pre-stored in the control system in stages, with each stage corresponding to a set of control objectives and control algorithms. Control objectives can be setpoints (e.g., temperature set to 12℃), set trajectories (e.g., heating at a rate of 0.1℃ / h), or constraints (e.g., pH not lower than 4.0). Control algorithms can employ PID control, model predictive control (MPC), or rule-based control, etc. This invention does not limit the specific control algorithm; its core lies in the fact that the generation of control commands is no longer based on potentially distorted raw data, but on repaired high-confidence data, thereby ensuring that the triggering timing and adjustment range of the control actions meet the process requirements.

[0060] It should be noted that after generating and outputting control commands, the control system sends the commands to the field actuators to complete closed-loop control. For example, if the control command is to raise the fermentation tank temperature from the current 12°C to 14°C at a rate of 0.5°C / h, the DCS system will adjust the opening of the cooling water valve or the power of the heater to make the temperature change along the set trajectory. Simultaneously, the system continuously collects new sensor data, entering the next round of data confidence assessment, matching, repair, and adjustment cycle, forming a dynamic and adaptive closed-loop control process. In this way, even if data loss occurs again in the future, the system can still make correct decisions based on historical knowledge and repaired data, thereby ensuring the stability of the entire brewing process and the consistency of product quality.

[0061] It should be noted that, because the repaired multidimensional state vector sequence retains the correct temporal characteristics and multivariate coupling relationships, the accuracy of process stage identification is significantly improved, the triggering timing of control commands is precise, and misjudgments and delays caused by data smoothing are avoided. Ultimately, the beer fermentation process can operate strictly according to the preset process trajectory, batch-to-batch consistency is enhanced, product quality stability is improved, and energy waste and raw material loss caused by misoperation are reduced.

[0062] In summary, this patent systematically solves the problems of temporal feature distortion and control misjudgment caused by missing data and automatic system completion through a complete technical chain of data confidence identification, multivariate coupling constraint matching, feature mapping repair, and regulation based on repaired data, thus achieving robust control of the beer brewing process under incomplete data conditions.

[0063] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0064] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0065] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0067] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0068] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-parameter coordinated control system for beer brewing, characterized in that, include: A sequence module is constructed to collect the original process parameters at each moment, generate derived process parameters based on the original process parameters, use the original process parameters and derived process parameters together as brewing process parameters, and construct a multidimensional state vector sequence for the current batch based on the brewing process parameters. The confidence identification module is used to determine the data confidence at each time step based on the local statistical characteristics of the data in the multidimensional state vector sequence. The path determination module is used to compare the multidimensional state vector sequence of the current batch with the preset standard benchmark batch sequence, and determine the matching path through a multivariate coupling constraint mechanism. The repair sequence module is used to repair the multidimensional state vector sequence of the current batch on the matching path based on the standard benchmark batch sequence and the data confidence, and generate a repaired multidimensional state vector sequence. The instruction generation module is used to generate coordinated control instructions based on the repaired multidimensional state vector sequence.

2. The multi-parameter coordinated control system for beer brewing as described in claim 1, characterized in that: In the sequence construction module, the multidimensional state vector sequence is composed of multidimensional state vectors at multiple times, and the multidimensional state vector at each time is composed of multiple brewing process parameters collected at the current time.

3. The multi-parameter coordinated control system for beer brewing as described in claim 2, characterized in that: The original process parameters include at least one of temperature, pressure, pH value, and dissolved oxygen; the derived process parameters include the first derivative of the pH value.

4. The multi-parameter coordinated control system for beer brewing as described in claim 1, characterized in that: In the confidence identification module, the data confidence at each time step is determined based on the local statistical characteristics of the data in the multidimensional state vector sequence, as follows: Calculate the local variance at each time step; The local variance is compared with a preset threshold. If the local variance is less than the preset threshold, the data at the current moment is determined to be low-confidence data.

5. The multi-parameter coordinated control system for beer brewing as described in claim 1, characterized in that: In the path determination module, a multivariate coupling constraint mechanism is used to determine the matching path, as follows: Based on the multidimensional state vector sequence of the current batch, generate the covariance matrix of the standard benchmark batch sequence; Based on the covariance matrix, calculate the weighted distance between the multidimensional state vector sequence of the current batch and the standard baseline batch sequence; Using the weighted distance as the path cost, the minimum cumulative cost path from the start point to the end point of the sequence is determined using the dynamic time warping algorithm, and this path is used as the matching path.

6. The multi-parameter coordinated control system for beer brewing as described in claim 5, characterized in that: Based on the covariance matrix, the weighted distance between the current batch's multidimensional state vector sequence and the standard baseline batch sequence is calculated as follows: The weighted distance is a weighted Mahalanobis distance; the covariance matrix between multiple process parameters in the standard benchmark batch sequence is obtained, and the covariance matrix is ​​used to characterize the coupling relationship between different process parameters; Construct a variable importance matrix, which is a diagonal matrix, and its diagonal elements correspond to the weight coefficients of the multiple process parameters in the matching process. The weighted Mahalanobis distance is calculated based on the inverse of the covariance matrix, the variable importance matrix, and the difference between the multidimensional state vector of the current batch and the corresponding vector in the standard benchmark batch sequence.

7. The multi-parameter coordinated control system for beer brewing as described in claim 1, characterized in that: In the repair sequence module, the multidimensional state vector sequence of the current batch is repaired to generate a repaired multidimensional state vector sequence, as follows: When determining the matching path, the weight of the path cost of low-confidence data is reduced, so that the determination of the matching path is mainly determined by high-confidence data. After determining the matching path, the temporal features in the corresponding segment of the standard benchmark batch sequence are mapped onto the low-confidence data of the current batch to generate a repaired multidimensional state vector sequence.

8. The multi-parameter coordinated control system for beer brewing as described in claim 7, characterized in that: The time-series characteristics include the data's changing trends and derivative characteristics.

9. The multi-parameter coordinated control system for beer brewing as described in claim 1, characterized in that: In the instruction generation module, based on the repaired multidimensional state vector sequence, coordinated control instructions are generated as follows: Based on the repaired multidimensional state vector sequence, the current process stage is identified; Based on the stated process stage, the corresponding process control logic is invoked to generate and output control commands.

Images