Virtual metering method for steam flow of brewing plant based on mechanism and data fusion

Abstract

The invention discloses a wine brewing plant steam flow virtual metering method based on mechanism and data fusion, and the method comprises the steps: an edge calculation device builds a pipeline metering mechanism model from an inlet of a wine brewing plant to each steam use end, obtains the steam use amount of each steam use end through calculation, and carries out the accumulation summation to obtain a wine brewing plant steam flow mechanism value; the edge computing device forms a virtual measurement data sample of the steam flow of the brewing plant; performing multi-dimensional feature extraction and model training on the virtual measurement data sample of the steam flow of the brewing plant by adopting a deep learning network, establishing a steam flow prediction model of the brewing plant, and obtaining a steam flow prediction value of the brewing plant; and performing fusion calculation on the steam flow mechanism value and the steam flow predicted value of the brewing plant to obtain a virtual metering value of the steam flow of the brewing plant. Accurate flow metering under the scene that the steam flow of a winery fluctuates rapidly is achieved, and the strict limitation of a traditional flow meter on straight pipe sections before and after measurement is broken through.

Description

Technical Field

[0001] The present invention belongs to the technical field of brewery steam systems, and in particular relates to a brewery steam flow virtual metering method based on mechanism and data fusion. Background Art

[0002] Steam, as the core energy source in winemaking, is also the primary source of carbon emissions and the largest consumer of water resources in wineries. Refined management of the winery's steam system, and the gradual reduction of steam energy costs in the future, to achieve energy conservation and carbon reduction are important long-term tasks for wineries. During the winemaking process, grain must first be moistened to achieve the appropriate moisture content and initial starch preparation before it is sent to the retort for core operations such as steaming and distilling. Therefore, it is necessary to rationally utilize steam for moistening to ensure that the raw materials are suitable for subsequent processes, and to use steam for distillation, to assist in grain gelatinization, and to sterilize the grain, to promote the smooth progress of the winemaking process and ensure the output and quality of the wine.

[0003] However, traditional steam flow measurement in breweries relies on vortex and orifice flowmeters. These suffer from issues like unstable metering accuracy and short service life, significantly hindering the sustainable and refined management of steam systems. For example, the unique nature of brewing requires brewers to dynamically adjust steam usage based on the microbial fermentation within the retort. Consequently, during normal brewing operations, steam flow is frequently adjusted within a preset range. When the actual steam flow is significantly less than the rated flow, the meter's strategy error rapidly increases. Furthermore, the brewing process is not continuous 24 hours a day. Frequent temperature fluctuations within the pipes cause the steam flowmeter to fluctuate from ambient to high temperatures, accelerating the thermal expansion and contraction of key components, leading to increasingly poor measurement accuracy. Therefore, avoiding the metering errors of traditional steam flowmeters and improving the accuracy, reliability, and adaptability of steam flow measurement in breweries are pressing challenges.

[0004] Based on the above technical problems, it is necessary to design a new virtual measurement method of steam flow in brewery based on mechanism and data fusion. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a virtual metering method for steam flow in a brewery based on mechanism and data fusion, which can integrate the mechanism model and the data-driven steam flow prediction model, comprehensively consider the relevant steam flow mechanism knowledge and collected multi-dimensional data in the brewing production process, capture and interpret the complex relationship and characteristics of the steam system in the brewing production process, overcome the limitations of a single mechanism model or a data-driven steam flow prediction model, and ensure the physical rationality of the metering results. It can also improve the adaptability under complex steam system working conditions, improve the accuracy and reliability of the virtual metering of steam flow in the brewery, and also provide data support for the subsequent refined control of steam consumption and energy saving and carbon reduction in the brewery; in addition, it avoids the metering deviation of traditional steam flow meters and also provides another steam flow metering method for breweries that cannot install steam flow sensors.

[0006] In order to solve the above technical problems, the technical solution of the present invention is:

[0007] The present invention provides a virtual metering method for steam flow in a brewery based on mechanism and data fusion, which includes:

[0008] S1. Based on the existing equipment components in the brewery, temperature sensors, pressure sensors, and edge computing devices were installed at the steam pipe entrance of the brewery. Pressure sensors were also installed in front of each steam consumption end in the brewery. The data measurement values ​​of each type of sensor were obtained through the edge computing device.

[0009] S2. The edge computing device uses the acquired data measurements, combined with the physical property parameters of the steam pipeline, the steam pipeline transmission resistance, and the relationship between steam flow and pressure changes, to establish a pipeline metering mechanism model from the brewery entrance to each steam-using end. It calculates the steam usage at each steam-using end and accumulates and sums the steam usage to obtain the brewery steam flow mechanism value.

[0010] S3. The edge computing device generates a virtual metering data sample of the brewery's steam flow based on the acquired data measurements, combined with the brewery's brewing production schedule, steam pipeline structure data, external environment data, structural data of each steam-using end, and process data.

[0011] S4. After performing multi-dimensional feature extraction and model training on the brewery steam flow virtual metering data samples using a deep learning network, a brewery steam flow prediction model is established to obtain a brewery steam flow prediction value;

[0012] S5. The brewery steam flow mechanism value and the steam flow prediction value are integrated and calculated to obtain a virtual measurement value of the brewery steam flow.

[0013] Furthermore, in said S1, each steam-using end in said brewing plant includes a steam end for a grain-moistening water tank and a steam end for a wine retort, and there are multiple grain-moistening water tanks and wine retorts.

[0014] Furthermore, the S2 includes:

[0015] The edge computing device obtains the temperature and pressure data measurement values ​​at the entrance of the steam pipeline of the brewery, as well as the pressure data measurement values ​​in front of each grain-wetting water tank and the pressure data measurement values ​​in front of each wine retort. Combined with the diameter, length, friction coefficient of the steam pipeline, and the calculation principle of the along-the-line resistance and local resistance during the steam transportation process of the steam pipeline, a pipeline metering mechanism model from the entrance of the brewery to each steam-using end is established. After solving and calculating the steam flow rate in front of each grain-wetting water tank and each wine retort in the pipeline, the steam consumption of each steam-using end is calculated through the steam flow rate, pipe diameter, and steam physical properties. The steam flow mechanism value of the brewery is obtained by cumulative summation; the steam physical properties include: steam density, pressure, and temperature.

[0016] Furthermore, the S3 includes:

[0017] The edge computing device obtains the temperature and pressure data measurement values ​​at the steam pipe entrance of the brewery, as well as the pressure data measurement values ​​in front of each grain-watering tank and the pressure data measurement values ​​in front of each wine retort;

[0018] Obtain the brewery's brewing production schedule, including grain tempering, grain steaming, and distillation schedules;

[0019] Obtain steam pipeline structural data, including: pipe diameter and wall thickness, pipeline length and direction, pipeline connection and branch structure data, and pipeline insulation layer parameters;

[0020] Obtain external environmental data, including: factory temperature, humidity and atmospheric pressure;

[0021] Obtain structural data for each steam-using end, including: shape, size, and wall thickness of the grain-moistening water tank, as well as the upper and lower diameters, height, volume parameters, taper, and insulation thickness of the wine retort;

[0022] Obtain process data at each steam-using end, including: grain pile temperature, grain tempering amount, tempering time, tempering stirring time and frequency, tempering water tank temperature, as well as mash temperature, retort surface temperature, steam temperature, pressure difference at different parts of the retort, and liquor parameters;

[0023] Based on the data measurement values ​​of the brewery, production schedule, steam pipeline structure data, external environment data, structure data of each steam-using end and process data, a virtual metering data sample of the brewery steam flow is formed.

[0024] Further, the S4 includes:

[0025] Construct various modules of a deep learning network for brewery steam flow prediction, including: data dimensionality reduction module, spatial feature extraction module, temporal feature extraction module, attention mechanism module, feature integration module, and steam flow prediction module;

[0026] Through the data dimensionality reduction module, the adaptive sparse kernel principal component analysis method is used to reduce the dimensionality of the virtual metering data samples to obtain key data features that affect the steam flow;

[0027] Through the spatial feature extraction module, a graph attention convolutional network is used to extract spatial features of key data features;

[0028] Through the temporal feature extraction module, a bidirectional long short-term memory network is used to extract the basic temporal features of key data features;

[0029] Through the attention mechanism module, a multi-head self-attention mechanism is used to obtain global temporal features by considering time dependence and similarity;

[0030] The feature integration module integrates the global temporal features and spatial features and outputs fusion features;

[0031] Through the steam flow prediction module, a deep belief network is used to train and learn the fusion features to output the predicted value of the steam flow of the brewery.

[0032] Further, the S5 includes:

[0033] A hybrid coordination unit is set up to learn the contribution value of the brewery steam flow mechanism value and the steam flow prediction value to the steam flow virtual metering prediction, adjust the weight factor of the steam flow mechanism value and the prediction value, and output the hybrid brewery steam flow virtual metering value;

[0034] The deviation loss between the virtual measurement value of the steam flow in the brewery and the actual steam consumption is used as the objective function to fine-tune the steam flow prediction model and update the weight factor to obtain the final virtual measurement value of the steam flow in the brewery.

[0035] Furthermore, the S5 further includes:

[0036] The pipeline metering mechanism model from the brewery entrance to each steam-using end is integrated into the steam flow prediction model of the brewery, so that the steam flow prediction model includes the physical mechanism of steam flow. At the same time, based on the quantitative index function of the prediction model and the physical constraint residual function as the penalty term, a total loss function of the steam flow prediction model integrating the physical mechanism is constructed, and the parameters of the steam flow prediction model are trained and optimized by minimizing the total loss function. Finally, the optimized steam flow prediction model is used to perform virtual metering of the steam flow of the brewery.

[0037] Furthermore, the quantitative index function includes the mean square error index function MSE, the root mean square error index function RMSE, and the mean absolute error MAE; the physical constraint residual function is obtained by transforming the mechanism model using the physical constraint fusion method, and the physical constraint fusion method includes L2 regularization method, MAE regularization, physical constraint embedding method and weak supervision constraint method.

[0038] Furthermore, the total loss function is expressed as: Loss = α1L d +α2L m , α1 and α2 are the training weight coefficients of physical constraints and prediction models respectively. The influence of physical constraints on the prediction model can be adjusted by changing the size of the weight coefficients. d is the loss part in the steam flow prediction model training process, L m is the loss part of physical constraints.

[0039] Furthermore, after the brewery steam flow prediction model is established, the method further includes: using the model interpretability theory to perform global and local interpretations on the brewery steam flow prediction model and the prediction results.

[0040] The beneficial effects of the present invention are:

[0041] (1) The present invention is based on the original equipment components of the brewery, and installs temperature sensors, pressure sensors and edge computing devices at the steam pipe inlet of the brewery, and respectively installs pressure sensors in front of each steam-using end in the brewery, and obtains the data measurement values ​​of each type of sensor through the edge computing device; it can realize multi-dimensional data acquisition, covering the steam inlet and each steam-using end, and use the edge computing device to process the data on-site, reducing the cloud transmission delay, and using the edge computing device to perform subsequent mechanism model and steam flow prediction model establishment and virtual metering calculation, thereby improving the real-time performance and accuracy of data processing and model establishment;

[0042] (2) The present invention establishes a pipeline metering mechanism model from the entrance of the brewery to each steam-using end through edge computing equipment, calculates the steam consumption of each steam-using end, and accumulates and sums to obtain the steam flow mechanism value of the brewery; the edge computing equipment forms a virtual metering data sample of the steam flow of the brewery based on the acquired data measurement value, combined with the brewing production schedule of the brewery, steam pipeline structure data, external environment data, structure data of each steam-using end and process data, and uses a deep learning network to perform multi-dimensional feature extraction and model training on the virtual metering data sample of the steam flow of the brewery, and then establishes a steam flow prediction model for the brewery; and the steam flow mechanism value and the steam flow prediction value of the brewery are integrated and calculated through the edge computing equipment to obtain the steam flow mechanism value of the brewery. Virtual flow measurement value; it can integrate the mechanism model and the data-driven steam flow prediction model, comprehensively consider the relevant steam flow mechanism knowledge and the collected multi-dimensional data in the brewing production process, capture and explain the complex relationship and characteristics of the steam system in the brewing production process, overcome the limitations of a single mechanism model or a data-driven steam flow prediction model, and ensure the physical rationality of the measurement results while improving the adaptability under complex steam system working conditions. It improves the accuracy and reliability of the virtual measurement of steam flow in the brewery, and also provides data support for the subsequent refined control of steam consumption and energy conservation and carbon reduction in the brewery. In addition, it avoids the measurement deviation of traditional steam flow meters and provides another steam flow measurement method for breweries that cannot install steam flow sensors.

[0043] (3) Compared with traditional orifice plate or vortex steam flowmeters, the virtual metering method of steam flow in the present invention has the following advantages: the metering accuracy does not change with the flow rate, and can maintain good accuracy even at extremely small flow rates; multiple parameter inputs and data cross-comparison avoid metering deviations caused by a single data source; frequent day and night temperature changes in the pipe have little effect on metering accuracy, and the measurement life is longer; there is no need to ensure strict straight pipe section requirements when installing the sensor, and installation in old factories is less difficult.

[0044] Other features and advantages will be described in the following description, and in part will become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is a flow chart of a method for virtual measurement of steam flow in a brewery based on mechanism and data fusion according to the present invention;

[0048] Figure 2 The figure is a schematic diagram of the virtual metering principle of steam flow in a brewery based on mechanism and data fusion according to the present invention. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] like Figure 1 、 Figure 2 As shown, this embodiment provides a virtual metering method for steam flow in a brewery based on mechanism and data fusion, which includes:

[0051] S1. Based on the existing equipment components in the brewery, temperature sensors, pressure sensors, and edge computing devices were installed at the steam pipe entrance of the brewery. Pressure sensors were also installed in front of each steam consumption end in the brewery. The data measurement values ​​of each type of sensor were obtained through the edge computing device.

[0052] S2. The edge computing device uses the acquired data measurements, combined with the physical property parameters of the steam pipeline, the steam pipeline transmission resistance, and the relationship between steam flow and pressure changes, to establish a pipeline metering mechanism model from the brewery entrance to each steam-using end. It calculates the steam usage at each steam-using end and accumulates and sums the steam usage to obtain the brewery steam flow mechanism value.

[0053] S3. The edge computing device generates a virtual metering data sample of the brewery's steam flow based on the acquired data measurements, combined with the brewery's brewing production schedule, steam pipeline structure data, external environment data, structural data of each steam-using end, and process data.

[0054] S4. After performing multi-dimensional feature extraction and model training on the brewery steam flow virtual metering data samples using a deep learning network, a brewery steam flow prediction model is established to obtain a brewery steam flow prediction value;

[0055] S5. The brewery steam flow mechanism value and the steam flow prediction value are integrated and calculated to obtain a virtual measurement value of the brewery steam flow.

[0056] In this embodiment, in S1, the steam-using ends in the brewery include steam ends for the grain-moistening water tank and steam ends for the wine retort, and there are multiple grain-moistening water tanks and wine retorts.

[0057] In actual applications, the grain moistening stage: steam is used to heat and moisten the brewing raw materials (such as sorghum), allowing them to absorb a certain amount of water, which facilitates better gelatinization of starch during subsequent steaming and cooking. It can also remove impurities and odors from the raw materials, improve the purity of the raw materials, and lay a good foundation for subsequent fermentation and distillation. For example, sorghum absorbs water and softens, which is conducive to thorough cooking;

[0058] The wine retort process includes:

[0059] 1) Distillation: Steam heats the fermented mash in the steamer, vaporizing the alcohol and aroma components in the mash to form wine vapor, which is then condensed into liquid wine, separating the wine from the mash. It also concentrates the alcohol and increases the alcohol content, extracts aroma substances, removes some impurities, and ensures the quality of the wine.

[0060] 2) Steaming and gelatinization of grains (if the steamer also has a steaming function): If new grain and mash are fed into the steamer together, the steam can cook the new grain and gelatinize the starch in the grain, making it easier to ferment in the cellar. In some processes, the steamer can simultaneously complete the process of picking wine and gelatinizing the new grain, solving two problems in one process, improving efficiency and saving costs.

[0061] 3) Sterilization: The high temperature of steam can sterilize the mash and equipment in the steamer, reducing the impact of bacteria, ensuring a sanitary and stable winemaking process, and promoting fermentation and improving wine quality;

[0062] When moistening the grain, if the parameters such as steam flow and moistening time are not well controlled, the grain will absorb water unevenly and gelatinize insufficiently, and after entering the retort, problems such as "insufficient steaming, less wine, and more impurities" will appear. On the contrary, if the grain is moistened properly, the steam distribution in the retort will be more even, and the distillation efficiency will be higher. In addition, according to the wine output from the retort (alcohol content, aroma, impurities), the winemaker will reversely optimize the grain moistening process. For example, if the wine is found to have a raw taste, the water temperature for moistening the grain may be adjusted, the moistening time may be extended, and the grain pretreatment may be more thorough. In summary, steam is used to moisten the grain to make the raw materials more suitable for subsequent processes. Steam is used in the retort mainly to distill wine, assist in grain gelatinization and sterilization, promote the smooth brewing process, and ensure the output and quality of the wine.

[0063] In this embodiment, S2 includes:

[0064] The edge computing device obtains the temperature and pressure data measurement values ​​at the entrance of the steam pipeline of the brewery, as well as the pressure data measurement values ​​in front of each grain-wetting water tank and the pressure data measurement values ​​in front of each wine retort. Combined with the diameter, length, friction coefficient of the steam pipeline, and the calculation principle of the along-the-line resistance and local resistance during the steam transportation process of the steam pipeline, a pipeline metering mechanism model from the entrance of the brewery to each steam-using end is established. After solving and calculating the steam flow rate in front of each grain-wetting water tank and each wine retort in the pipeline, the steam consumption of each steam-using end is calculated through the steam flow rate, pipe diameter, and steam physical properties. The steam flow mechanism value of the brewery is obtained by cumulative summation; the steam physical properties include: steam density, pressure, and temperature.

[0065] It should be noted that traditional steam flow measurement often uses vortex and orifice plate measurement methods. This type of measurement method creates pressure changes or flow pattern changes in the steam fluid by installing a resistance member (orifice plate or vortex sounder) in the steam pipe, thereby calculating the flow velocity of the steam fluid and then calculating the mass flow rate of the steam through volume and physical properties. However, traditional steam flow measurement has the following characteristics: (1) Before entering the flow meter for measurement, it is necessary to ensure a sufficient length of straight pipe section to ensure that the fluid in the pipe is in a stable laminar state to avoid unpredictable turbulent flow patterns causing deviations in the pressure drop calculation formula and affecting the generation of standard vortex; (2) The size of the small hole on the orifice plate and the size of the vortex sounder of the vortex flowmeter are designed according to the maximum rated flow in the pipeline. When the actual steam flow in the pipeline is much less than the rated flow, the measurement error will increase rapidly; (3) When the working fluid temperature of the flow meter changes frequently, it causes thermal expansion of the orifice plate material, resulting in the aperture deviation from the factory size increasing with the increase of service time, thereby making the measurement accuracy worse and worse.

[0066] During the winemaking process, steam usage must be dynamically adjusted based on the microbial fermentation within the retort. Furthermore, the winemaking process is not continuous throughout the entire 24-hour period. Rapid temperature fluctuations within the pipeline cause the flowmeter to frequently fluctuate between ambient and elevated temperatures, accelerating the thermal expansion and contraction of the flowmeter's key components. In summary, steam flowmeters can exhibit metering deviations, further leading to errors in steam usage management. Therefore, a virtual steam flow metering physical mechanism model can be established based on existing temperature and pressure measurement data within the winery. This model can calculate steam flow at each steam-consuming end, addressing the shortcomings of traditional meters and providing an alternative steam flow measurement method for wineries where steam flow sensors are not suitable.

[0067] The resistance of steam pipeline transportation is divided into along-the-line resistance and local resistance. The along-the-line resistance is the pressure drop after the steam passes through the straight pipe section, while the local resistance refers to the pressure drop of the steam after passing through the resistance elements such as tees, valves, and metering instruments. The along-the-line resistance can be calculated according to the following formula:

[0068]

[0069] ΔP is the resistance along the way; λ is the friction coefficient; L is the length of the steam pipe; D is the inner diameter of the steam pipe; ρ is the steam density; v is the steam flow rate;

[0070] The local resistance can be calculated according to the following formula:

[0071]

[0072] ΔP′ is the local resistance; ζ is the local resistance coefficient.

[0073] According to the steam system of the steam pipe diameter, length, friction coefficient, steam inlet physical parameters and various resistance parts along the way, the relationship between the steam flow rate in the steam pipe and the inlet and outlet pressure difference can be obtained through resistance calculation, and then a pipeline metering mechanism model from the inlet of the brewery to each steam-using end is established, that is, the sum of the pressure differences of multiple along-the-way resistances and local resistances is constructed, and a joint solution is performed with one data operating point per second to obtain the steam flow rate in front of each wine steamer and each grain-wetting pipe in the steam pipe. The steam consumption of each steam-using end is further calculated through the flow rate, inner diameter of the pipe and steam physical parameters, and the total steam flow mechanism value of the brewery is obtained by cumulative summation; among them, the steam mass conservation must be ensured in the calculation process, that is, the steam mass flow rate at the inlet of the brewery Equal to the steam consumption of all wine retorts in the factory and grain heating steam flow The sum of:

[0074] In this embodiment, S3 includes:

[0075] The edge computing device obtains the temperature and pressure data measurement values ​​at the steam pipe entrance of the brewery, as well as the pressure data measurement values ​​in front of each grain-watering tank and the pressure data measurement values ​​in front of each wine retort;

[0076] Obtain the brewery's brewing production schedule, including grain tempering, grain steaming, and distillation schedules;

[0077] Obtain steam pipeline structural data, including: pipe diameter and wall thickness, pipeline length and direction, pipeline connection and branch structure data, and pipeline insulation layer parameters;

[0078] Obtain external environmental data, including: factory temperature, humidity and atmospheric pressure;

[0079] Obtain structural data for each steam-using end, including: shape, size, and wall thickness of the grain-moistening water tank, as well as the upper and lower diameters, height, volume parameters, taper, and insulation thickness of the wine retort;

[0080] Obtain process data at each steam-using end, including: grain pile temperature, grain tempering amount, tempering time, tempering stirring time and frequency, tempering water tank temperature, as well as mash temperature, retort surface temperature, steam temperature, pressure difference at different parts of the retort, and liquor parameters;

[0081] Based on the data measurement values ​​of the brewery, production schedule, steam pipeline structure data, external environment data, structure data of each steam-using end and process data, a virtual metering data sample of the brewery steam flow is formed.

[0082] In actual application, the original wine steamer and grain-conditioning equipment will be equipped with corresponding process data monitoring devices. For example, by monitoring the temperature of the mash (the surface / internal temperature of the mash in different layers and areas), the distillation process and the uniformity of steam penetration can be reflected; by monitoring the steam temperature (steam at the bottom of the steamer, steam temperature through the air pipe), the steam quality and distillation efficiency can be correlated. Too high / too low will affect the flavor and yield of the wine; by monitoring the pressure difference at different parts of the wine steamer, it can assist in judging the looseness of the mash and whether it is blocked. If the differential pressure is too large, it may be that the mash is clumping and the steam circulation is not smooth. The wine parameters include the output flow, alcohol content and flavor components; by inserting multiple temperature probes into the grain-moistening water tank, the temperature changes at the center and surface of the grain pile during the moistening process are monitored to prevent the grain from becoming moldy due to too long a moistening time, or uneven temperature affecting the starch conversion during subsequent steaming; by monitoring the amount of moistening grain to match the production batch requirements, the moistening efficiency is controlled to avoid insufficient water penetration due to excessive amount; by monitoring the length of the moistening time, if the time is too short, the grain will not absorb enough water, and if it is too long, microorganisms may grow, affecting the quality of the raw materials, and it needs to be adjusted in combination with the water temperature and grain variety; by monitoring the stirring time and frequency of the moistening grain, the grain can be prevented from settling and agglomerating.

[0083] It's important to note that the structural data of each steam-consuming end is correlated with steam usage. For example, the volume and wall thickness of the grain-conditioning water tank affect grain heat absorption and heat loss, directly determining total steam consumption. The taper, height, and insulation thickness of the wine retort influence steam distribution uniformity and heat exchange efficiency. Production schedules also strongly influence steam usage. For example, longer grain conditioning time increases the window for continuous steam supply. If the conditioning time is extended from 12 hours to 18 hours, and the grain pile temperature is maintained above process requirements, steam must continuously compensate for heat losses, and usage increases linearly with time. Steaming is a steam-intensive process, with steaming duration positively correlated with steam usage and synergistically influenced by steam pressure and temperature. Distillation requires a stable steam flow to extract alcohol; longer distillation times increase cumulative steam consumption. Steam pipeline structural data determines steam transmission efficiency and heat loss rate. The external environment affects the physical properties of steam (density, condensation rate) and pipe heat dissipation, indirectly altering steam demand.

[0084] In this embodiment, the S4 includes:

[0085] Construct various modules of a deep learning network for brewery steam flow prediction, including: data dimensionality reduction module, spatial feature extraction module, temporal feature extraction module, attention mechanism module, feature integration module, and steam flow prediction module;

[0086] Through the data dimensionality reduction module, the adaptive sparse kernel principal component analysis method is used to reduce the dimensionality of the virtual metering data samples to obtain key data features that affect the steam flow;

[0087] Through the spatial feature extraction module, a graph attention convolutional network is used to extract spatial features of key data features;

[0088] Through the temporal feature extraction module, a bidirectional long short-term memory network is used to extract the basic temporal features of key data features;

[0089] Through the attention mechanism module, a multi-head self-attention mechanism is used to obtain global temporal features by considering time dependence and similarity;

[0090] The feature integration module integrates the global temporal features and spatial features and outputs fusion features;

[0091] Through the steam flow prediction module, a deep belief network is used to train and learn the fusion features to output the predicted value of the steam flow of the brewery.

[0092] In practical applications, the adaptive sparse kernel principal component analysis method applies different weights to the penalty terms in the original sparse kernel principal component analysis process. This allows for different levels of penalty based on feature importance, retaining principal components strongly correlated with steam flow and eliminating redundant information. The spatial feature extraction module, utilizing a graph attention convolutional network, abstracts the brewery's steam system into a graph structure: nodes represent physical entities such as pipeline nodes, steamers, and grain-watering tanks, while edges represent pipeline connections and heat conduction paths. An attention mechanism calculates the weights of influence of different nodes on steam flow and extracts spatial topological features. The temporal feature extraction module, utilizing a bidirectional long short-term memory network, leverages the gating mechanism of the LSTM to capture long-term dependencies in steam flow. The bidirectional structure simultaneously learns past and future temporal information. The attention mechanism module utilizes a multi-head self-attention mechanism to parallelize dependencies across different time steps, capturing similarities between non-adjacent time points using a self-attention weight matrix. The feature integration module integrates spatial and global temporal features through concatenation and weighting. The steam flow prediction module uses a deep belief network to learn the nonlinear mapping relationship between fusion features and steam flow through unsupervised pre-training + supervised fine-tuning, which is suitable for processing complex coupling relationships in brewing data.

[0093] In this embodiment, S5 includes:

[0094] A hybrid coordination unit is set up to learn the contribution value of the brewery steam flow mechanism value and the steam flow prediction value to the steam flow virtual metering prediction, adjust the weight factor of the steam flow mechanism value and the prediction value, and output the hybrid brewery steam flow virtual metering value;

[0095] The deviation loss between the virtual measurement value of the steam flow in the brewery and the actual steam consumption is used as the objective function to fine-tune the steam flow prediction model and update the weight factor to obtain the final virtual measurement value of the steam flow in the brewery.

[0096] It should be noted that the hybrid coordination unit dynamically balances the outputs of the two models using weighting factors, achieving a complementary effect where data-driven capture of dynamic fluctuations and the mechanism model ensures physical rationality. Adaptive learning algorithms (such as gradient descent and reinforcement learning) are employed, using the deviation between the mechanism value, the predicted value, and the actual steam usage as feedback to dynamically adjust the weighting factors of the mechanism value and the predicted value.

[0097] The deviation loss as an objective function is expressed as: N is the number of samples, X i is the virtual measurement value of steam flow, S i is the actual steam consumption, λ is the regularization coefficient, Ω(θ,α) is the regularization term, θ is the learning algorithm model parameter, and α is the weight factor.

[0098] In this embodiment, the S5 further includes:

[0099] The pipeline metering mechanism model from the brewery entrance to each steam-using end is integrated into the steam flow prediction model of the brewery, so that the steam flow prediction model includes the physical mechanism of steam flow. At the same time, based on the quantitative index function of the prediction model and the physical constraint residual function as the penalty term, a total loss function of the steam flow prediction model integrating the physical mechanism is constructed, and the parameters of the steam flow prediction model are trained and optimized by minimizing the total loss function. Finally, the optimized steam flow prediction model is used to perform virtual metering of the steam flow of the brewery.

[0100] In this embodiment, the quantitative index function includes the mean square error index function MSE, the root mean square error index function RMSE, and the mean absolute error MAE; the physical constraint residual function is obtained by transforming the mechanism model using the physical constraint fusion method, and the physical constraint fusion method includes L2 regularization method, MAE regularization, physical constraint embedding method and weak supervision constraint method.

[0101] In this embodiment, the total loss function is expressed as: Loss = α1L d +α2L m , α1 and α2 are the training weight coefficients of physical constraints and prediction models respectively. The influence of physical constraints on the prediction model can be adjusted by changing the size of the weight coefficients. d is the loss part in the steam flow prediction model training process, L m is the loss part of physical constraints.

[0102] In this embodiment, after the brewery steam flow prediction model is established, the method further includes: using the model interpretability theory to perform global and local interpretations on the brewery steam flow prediction model and the prediction results.

[0103] It should be noted that model interpretability refers to revealing the decision logic of the prediction model through technical means, making the prediction results transparent and understandable. In brewing steam flow prediction, interpretability is particularly important, as shown in the following aspects:

[0104] Industrial safety needs: It is necessary to clarify whether the model prediction is consistent with the production mechanism to avoid production accidents caused by black box decision-making;

[0105] Optimization iteration basis: locate model defects through interpretation of results and guide parameter adjustment;

[0106] Mechanism fusion verification: verify the consistency between the data-driven model and the brewing process mechanism.

[0107] The interpretable method is post-hoc interpretability, which provides external explanations for complex models. Common techniques include SHAP, LIME, and feature importance analysis.

[0108] Global explanation: reveals the overall decision logic of the model, focusing on the model's decision-making pattern in the entire input space, and answers the question of how the model uses all features to make predictions. The specific implementation in brewing steam flow prediction is as follows:

[0109] 1) Feature Importance Analysis

[0110] By calculating the contribution of each feature to the prediction results, we identify the key factors affecting steam flow. Using SHAP technology based on game theory, we calculate the Shapley value of each feature to measure its marginal contribution to the prediction results. For example, we identify the weight of the impact of features such as "steaming time," "pipeline diameter," and "plant temperature" on steam flow to verify whether they are consistent with the brewing process mechanism.

[0111] 2) Decision rule extraction

[0112] Extract human-understandable rules from complex models (such as "If the steaming time is greater than 3 hours and the factory temperature is greater than 70 degrees, then increase the steam consumption by 20%").

[0113] Local explanation: Explain the cause of a single prediction result and answer why the model made the prediction for a specific sample. Specific implementations include:

[0114] 1) Sample-level feature attribution

[0115] Analyze the direction and magnitude of each feature's contribution to the predicted value in a single sample: Generate a SHAP value for each feature for each sample. A positive number indicates that the feature drives the predicted value upward, while a negative number indicates a downward trend. For example, when the steam flow forecast at a certain moment is abnormally high, the SHAP value reveals that "extended steaming phase" and "aging of pipeline insulation" are the primary contributing factors, assisting in troubleshooting production anomalies.

[0116] 2) Counterfactual Explanation

[0117] This function answers how the prediction results would change if a certain feature value changed, providing suggestions for optimizing decisions. For example, if the predicted steam usage exceeds a threshold, counterfactual explanations can be used to calculate the flow reduction effects of ending steaming one hour earlier or replacing pipe insulation, assisting in developing energy-saving strategies.

[0118] The comprehensive value of model interpretability in brewing steam flow prediction is as follows:

[0119] Mechanism verification: Through global interpretation, confirm whether the model has learned the steam consumption patterns of each stage of "grain conditioning-steaming-distillation" to avoid conflicts with process common sense;

[0120] Abnormal diagnosis: Use local interpretation to locate the cause of sudden changes in steam flow (such as leaking pipe connections or increased heat dissipation due to a sudden drop in ambient temperature);

[0121] Model optimization: If the global interpretation finds that the "pipeline direction" feature has low importance, it may indicate that the model does not fully capture spatial features, and the parameters of the graph attention network need to be adjusted;

[0122] Human-machine collaboration: Provide understandable prediction basis, enhance trust in the model, and support hybrid coordination of human decision-making and model prediction (such as weight adjustment of hybrid coordination units).

[0123] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the systems, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0124] In addition, the functional modules in each embodiment of the present invention can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part. If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk, or an optical disk.

[0125] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A virtual metering method for steam flow in brewery based on mechanism and data fusion, characterized in that: It includes: S1. Based on the existing equipment components in the brewery, temperature sensors, pressure sensors, and edge computing devices were installed at the steam pipe entrance of the brewery. Pressure sensors were also installed in front of each steam consumption end in the brewery. The data measurement values ​​of each type of sensor were obtained through the edge computing device. S2. The edge computing device uses the acquired data measurements, combined with the physical property parameters of the steam pipeline, the steam pipeline transmission resistance, and the relationship between steam flow and pressure changes, to establish a pipeline metering mechanism model from the brewery entrance to each steam-using end. It calculates the steam usage at each steam-using end and accumulates and sums the steam usage to obtain the brewery steam flow mechanism value. S3. The edge computing device generates a virtual metering data sample of the brewery's steam flow based on the acquired data measurements, combined with the brewery's brewing production schedule, steam pipeline structure data, external environment data, structural data of each steam-using end, and process data. S4. After performing multi-dimensional feature extraction and model training on the brewery steam flow virtual metering data samples using a deep learning network, a brewery steam flow prediction model is established to obtain a brewery steam flow prediction value; S5. The brewery steam flow mechanism value and the steam flow prediction value are integrated and calculated to obtain a virtual measurement value of the brewery steam flow.

2. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: In said S1, each steam-using end in said brewing plant includes a steam end for a grain-moistening water tank and a steam end for a wine retort, and there are multiple grain-moistening water tanks and wine retorts.

3. The method for virtual metering of steam flow in a brewery according to claims 1 and 2, characterized in that: The S2 includes: The edge computing device obtains the temperature and pressure data measurement values ​​at the entrance of the steam pipeline of the brewery, as well as the pressure data measurement values ​​in front of each grain-wetting water tank and the pressure data measurement values ​​in front of each wine retort. Combined with the diameter, length, friction coefficient of the steam pipeline, and the calculation principle of the along-the-line resistance and local resistance during the steam transportation process of the steam pipeline, a pipeline metering mechanism model from the entrance of the brewery to each steam-using end is established. After solving and calculating the steam flow rate in front of each grain-wetting water tank and each wine retort in the pipeline, the steam consumption of each steam-using end is calculated through the steam flow rate, pipe diameter, and steam physical properties. The steam flow mechanism value of the brewery is obtained by cumulative summation; the steam physical properties include: steam density, pressure, and temperature.

4. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: The S3 includes: The edge computing device obtains the temperature and pressure data measurement values ​​at the steam pipe entrance of the brewery, as well as the pressure data measurement values ​​in front of each grain-watering tank and the pressure data measurement values ​​in front of each wine retort; Obtain the brewery's brewing production schedule, including grain tempering, grain steaming, and distillation schedules; Obtain steam pipeline structural data, including: pipe diameter and wall thickness, pipeline length and direction, pipeline connection and branch structure data, and pipeline insulation layer parameters; Obtain external environmental data, including: factory temperature, humidity and atmospheric pressure; Obtain structural data for each steam-using end, including: shape, size, and wall thickness of the grain-moistening water tank, as well as the upper and lower diameters, height, volume parameters, taper, and insulation thickness of the wine retort; Obtain process data at each steam-using end, including: grain pile temperature, grain tempering amount, tempering time, tempering stirring time and frequency, tempering water tank temperature, as well as mash temperature, retort surface temperature, steam temperature, pressure difference at different parts of the retort, and liquor parameters; Based on the data measurement values ​​of the brewery, production schedule, steam pipeline structure data, external environment data, structure data of each steam-using end and process data, a virtual metering data sample of the brewery steam flow is formed.

5. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: The S4 includes: Construct various modules of a deep learning network for brewery steam flow prediction, including: data dimensionality reduction module, spatial feature extraction module, temporal feature extraction module, attention mechanism module, feature integration module, and steam flow prediction module; Through the data dimensionality reduction module, the adaptive sparse kernel principal component analysis method is used to reduce the dimensionality of the virtual metering data samples to obtain key data features that affect the steam flow; Through the spatial feature extraction module, a graph attention convolutional network is used to extract spatial features of key data features; Through the temporal feature extraction module, a bidirectional long short-term memory network is used to extract the basic temporal features of key data features; Through the attention mechanism module, a multi-head self-attention mechanism is used to obtain global temporal features by considering time dependence and similarity; The feature integration module integrates the global temporal features and spatial features and outputs fusion features; Through the steam flow prediction module, a deep belief network is used to train and learn the fusion features to output the predicted value of the steam flow of the brewery.

6. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: The S5 includes: A hybrid coordination unit is set up to learn the contribution value of the brewery steam flow mechanism value and the steam flow prediction value to the steam flow virtual metering prediction, adjust the weight factor of the steam flow mechanism value and the prediction value, and output the hybrid brewery steam flow virtual metering value; The deviation loss between the virtual measurement value of the steam flow in the brewery and the actual steam consumption is used as the objective function to fine-tune the steam flow prediction model and update the weight factor to obtain the final virtual measurement value of the steam flow in the brewery.

7. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: The S5 further includes: The pipeline metering mechanism model from the brewery entrance to each steam-using end is integrated into the steam flow prediction model of the brewery, so that the steam flow prediction model includes the physical mechanism of steam flow. At the same time, based on the quantitative index function of the prediction model and the physical constraint residual function as the penalty term, a total loss function of the steam flow prediction model integrating the physical mechanism is constructed, and the parameters of the steam flow prediction model are trained and optimized by minimizing the total loss function. Finally, the optimized steam flow prediction model is used to perform virtual metering of the steam flow of the brewery.

8. The method for virtual metering of steam flow in a brewery according to claim 7, characterized in that: The quantitative index functions include the mean square error index function MSE, the root mean square error index function RMSE, and the mean absolute error MAE; the physical constraint residual function is obtained by transforming the mechanism model using the physical constraint fusion method, and the physical constraint fusion method includes L2 regularization method, MAE regularization, physical constraint embedding method and weak supervision constraint method.

9. The method for virtual metering of steam flow in a brewery according to claim 7, characterized in that: The total loss function is expressed as: Loss = α1L d +α2L m , α1 and α2 are the training weight coefficients of physical constraints and prediction models respectively. The influence of physical constraints on the prediction model can be adjusted by changing the size of the weight coefficients. d is the loss part in the steam flow prediction model training process, L m is the loss part of physical constraints.

10. The method for virtual metering of steam flow in a brewery according to claim 1, characterized in that: After the brewery steam flow prediction model is established, the method further includes: using the model interpretability theory to perform global and local interpretations on the brewery steam flow prediction model and the prediction results.

Images