A rice wine fermentation monitoring system based on the Internet of Things

Abstract

The present invention relates to the technical field of Internet of Things, and in particular to a rice wine fermentation monitoring system based on the Internet of Things. In the present invention, a model construction module integrates material heat conduction and gas diffusion equations to form a physical field calculation basis, and embeds a metabolic environment dynamic coupling function to generate a joint calculation model; an inversion analysis module uses sensor temperature and humidity data as boundary constraints, iteratively corrects microbial metabolic rate parameters through inverse parameter optimization, and generates a global metabolic activity spatial distribution field, breaking through the limitations of traditional point-like monitoring on spatial heterogeneity perception; a deduction and prediction module initializes a model with the activity field, outputs a dual change curve of the temperature field and the activity gradient, captures a critical mutation point, and generates a targeted control signal based on the spatial coordinates and the prediction time window, thereby achieving advanced regulation of key turning points and maintaining fermentation stability and flavor quality.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to a rice wine fermentation monitoring system based on the Internet of Things. Background Art

[0002] Traditional rice wine fermentation monitoring relies mainly on a small number of temperature / humidity sensors placed on the fermentation tank wall or liquid surface, and achieves rough status monitoring through periodic data collection. This method has some drawbacks:

[0003] Sparsely distributed sensors struggle to fully perceive the differences in microbial activity distribution within the three-dimensional space of the tank. The interplay between heat transfer, gas convection, and microbial metabolism during fermentation exhibits spatial heterogeneity (e.g., high temperatures and high activity coexist in the core area while low temperatures and low activity occur at the edges). However, point-based data is limited in its ability to capture the true global state.

[0004] The periodic upload mechanism has difficulty capturing rapidly evolving critical events such as sudden temperature rises and activity cliff drops, which can lead to delayed control instructions and increase the risk of fermentation deterioration (such as rancidity or bacterial contamination) and process loss of control (such as temperature collapse).

[0005] Existing technologies have not yet effectively integrated the joint modeling capabilities of biological metabolic behavior and physical field evolution, and there is still room for improvement in the closed-loop monitoring system that inverts the spatial distribution of activity from discrete data and predicts evolutionary trends. Summary of the Invention

[0006] The purpose of the present invention is to provide a rice wine fermentation monitoring system based on the Internet of Things to solve the problems raised in the above-mentioned background technology. The specific technologies include how to invert the three-dimensional spatial distribution field of microbial metabolic activity in the fermentation tank through sparse sensor data to solve the problem of process misjudgment caused by the inability of traditional point monitoring to perceive global activity heterogeneity; and how to predict the continuous evolution trend of key state parameters based on the activity spatial distribution field to solve the regulation lag problem caused by the inability of the periodic sampling mechanism to capture the turning point in time, and this prediction needs to rely on the activity distribution field output by the previous problem to realize spatiotemporal coupling decision-making.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The rice wine fermentation monitoring system based on the Internet of Things includes a model construction module, an inversion analysis module and a deduction and prediction module, wherein:

[0009] The model construction module integrates the material heat conduction equation and the gas convection-diffusion equation to form a physical field calculation basis, and embeds the metabolic environment dynamic coupling function; converts the heat production rate into the heat source term of the heat conduction equation to drive the temperature field inversion and deduction; converts the carbon dioxide generation rate into the convection velocity correlation term of the gas convection-diffusion equation to drive the airflow field evolution simulation; this process establishes a biological-physical joint computing engine, providing a unified mathematical model basis for inversion (space) and deduction (time).

[0010] The inversion analysis module performs inverse parameter optimization, in which the boundary constraint assimilation unit matches the sensor temperature and humidity data to the spatial boundary nodes of the joint calculation model, converting them into boundary constraints of the physical field calculation basis (including the actual node temperature values ​​and humidity gradient thresholds), forcing the model solution space to conform to the real fermentation environment; this process is used to eliminate the risk of boundary distortion in traditional point monitoring.

[0011] The metabolic field iterative optimization unit in the inversion analysis module aims to minimize the temperature field output residual. It uses gradient descent combined with a regularization constraint algorithm to iteratively correct the microbial metabolic rate parameters. By running the model forward solver, calculating the residual, updating the parameters through the adjoint field method, and loading the spatial smoothing regularization term, the distribution of microbial metabolic rates in the entire tank is inverted from sparse data.

[0012] The activity field reorganization output unit in the inversion analysis module reorganizes the parameter values ​​in three dimensions based on the spatial discretization rules of the physical field calculation basis. After cleaning outliers and smoothing the edges, it outputs three-dimensional encapsulated data containing the spatial coordinate system, timestamp and activity intensity matrix, generating a metabolic activity spatial distribution field that represents the global microbial activity. This process converts sparse sensor data into a three-dimensional activity heat map, solving the problem of monitoring spatial heterogeneity of activity distribution.

[0013] The prediction module first writes the metabolic activity spatial distribution field data into the metabolic environment dynamic coupling function input port, verifies the parameter range and cuts off the over-limit value, and verifies the physical coupling state of the activity field with the temperature field and airflow field through the self-check program to ensure that the initial state of the model accurately reflects the real biological state.

[0014] Then, the transient partial differential equations are iteratively solved on the continuous time axis. The temperature field distribution and airflow field are updated in each time step. The metabolic rate correction value is reversely output through the metabolic environment dynamic coupling function, the metabolic rate parameters at the next moment are updated, and the temperature field change curve and the activity intensity gradient curve are simultaneously generated.

[0015] When the dual change curve detects a critical mutation point, the first-order difference of the sliding window is applied to the temperature field change curve to capture the slope mutation, and the second-order derivative zero value is applied to the activity intensity gradient curve to detect and locate the acceleration anomaly point; the physical field calculation base spatial coordinates are extracted to determine the abnormal area, and the predicted time window for the occurrence of the critical event is calculated based on the mutation slope. Finally, the spatial coordinates and the time window are compiled into directional control instructions, including a step-by-step stirring operation sequence, a pulse spray cooling operation sequence, a unique equipment identifier and a failure fallback strategy, to achieve closed-loop control from critical point prediction to precise execution, and solve the problem of time dimension lag.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] Activity field reconstruction technology, based on physical field calculations, converts sparse sensor data into a spatial distribution field of metabolic activity, addressing the lack of spatial heterogeneity in traditional point-based monitoring and enabling operators to intuitively identify local activity attenuation areas and hot spots.

[0018] The dual-change curve deduction mechanism based on initialization of the activity distribution field, combined with a collaborative detection algorithm based on temperature field slope mutation and activity gradient acceleration, enables advanced prediction of critical events such as fermentation stagnation and high-temperature runaway.

[0019] Directed control instructions generated by binding spatial coordinates to a predicted time window (such as step stirring in cold spot areas and pulse spraying in overheated areas) trigger targeted intervention before physical anomalies manifest, significantly reducing energy waste caused by ineffective cooling of the entire tank. This also maintains a balanced distribution of microbial activity, preventing rancidity and spoilage while improving the stability of flavor synthesis.

[0020] Ultimately, the process was upgraded from passive response to active prevention and control, providing precise and controllable technical support for high-quality rice wine brewing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the overall module of the present invention;

[0022] Figure 2 Schematic diagram of the inversion analysis module unit of the present invention.

[0023] In the figure: 100, model building module; 200, inversion analysis module; 201, boundary constraint assimilation unit; 202, metabolic field iterative optimization unit; 203, activity field reorganization output unit; 300, deduction and prediction module. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] Next, see Figure 1 The present invention provides a technical solution: a rice wine fermentation monitoring system based on the Internet of Things, including a model building module 100, an inversion analysis module 200 and a deduction and prediction module 300.

[0026] The model building module 100 first integrates two types of core equations. The first type of equations describes the heat transfer law inside the rice mash material, and characterizes the temperature diffusion process from the high-temperature area to the low-temperature area through the partial differential equation of material heat conduction; the second type of equations describes the gas movement characteristics produced by fermentation, and uses the gas convection diffusion equation to simulate the flow and mixing phenomena of gases such as carbon dioxide due to concentration differences; the two types of equations are mathematically linked to establish a unified framework, so that the temperature distribution and airflow movement state can be calculated synchronously and fed back to each other in real time, forming a physical field calculation basis.

[0027] The model building module 100 embeds a key function, namely the metabolic environment dynamic coupling function, in the physical field calculation base. This metabolic environment dynamic coupling function is based on the dynamic data of microbial metabolic activities (heat production rate, CO2 generation rate), and performs the operation of converting metabolic information into a physical field driving source; specifically, the heat production rate is converted into the heat source term of the heat conduction equation in the base, and the CO2 generation rate is converted into a key parameter (such as a source term or a convection velocity correlation term) that affects the convection-diffusion equation of airflow motion, thereby establishing a quantitative channel between biological metabolic activities and physical field changes.

[0028] The model building module 100 uses the embedded metabolic environment dynamic coupling function to establish a bidirectional coupling loop on the physical field calculation base, in which microbial metabolism drives the evolution of the physical field (temperature, airflow) through this mechanism; at the same time, the state of the physical field (such as temperature change, gas concentration distribution) is used as an input parameter feedback to regulate the metabolic rate function of the microorganism. Through this process, the three major dynamic systems of biological metabolic activities, heat transfer processes and gas movement are tightly coupled together, and finally a self-consistent joint computing model with a complete feedback loop is formed, which becomes the core engine for simulating the dynamics of the entire fermentation system.

[0029] See also Figure 2The boundary constraint assimilation unit 201 in the inversion analysis module 200 receives the temperature and humidity measurement data periodically uploaded by the sensor group deployed in the fermentation container, maps the data to the spatial boundary nodes corresponding to the joint calculation model, and converts it into boundary constraints of the physical field calculation basis, limiting the model solution space to conform to the real fermentation environment. Specifically, it includes:

[0030] First, spatial mapping calibration is performed to accurately match each data point to the corresponding spatial boundary node in the joint computing model (usually located at key locations such as the tank wall and liquid surface) based on the geometric structure of the fermentation vessel and the physical coordinates of the sensor;

[0031] After matching, the raw data is converted into boundary constraints with clear physical meaning, based on the data assimilation rules defined by the physical field computational foundation. These constraints include the measured temperature value at a specified node and the humidity gradient threshold for adjacent regions. These constraints force the solution space of the joint computational model to satisfy the physical boundary conditions of the actual fermentation environment, ensuring that subsequent inversion calculations are performed within the feasible domain. For example, when the sensor detects that the temperature in the tank top area is lower than the value predicted by the computational model, this node is automatically set as the model's strong constraint boundary, forcing the joint computational model to approximate the measured results by adjusting internal parameters rather than modifying boundary values. This process builds a bridge from discrete measured data to the continuous physical field model, providing a priori constraints that conform to thermodynamics and mass transfer laws for the inverse solution.

[0032] The metabolic field iterative optimization unit 202 in the inversion analysis module 200 initiates an inverse optimization process based on boundary constraints, with minimizing the sum of squared residuals between the theoretical temperature field output by the joint calculation model and the temperature field measured by the sensor as the sole optimization objective. The microbial metabolic rate parameter value (the input variable of the metabolic environment dynamic coupling function) of each discrete spatial unit in the joint calculation model is iteratively adjusted to drive model recalculation. When the adjustment of the microbial metabolic rate parameter value reduces the temperature residual to within a preset tolerance, the parameter field is determined to have converged. The above process specifically includes:

[0033] Under the premise that the boundary constraints are in effect, the core calculation process of inverse parameter optimization is initiated. This process uses minimizing the sum of squares of the temperature field residuals as a single-objective optimization function. Specifically, it is expressed as the residual norm of the theoretical value of the global temperature field output by the joint calculation model and the actual value measured by the sensor at the corresponding spatial location. The optimization object is defined as the microbial metabolic rate parameter of each discrete spatial unit in the model (i.e., the input variable of the metabolic environment dynamic coupling function). An iterative algorithm with gradient descent combined with regularization constraints is used. Each iteration consists of three steps:

[0034] A forward solver for running the joint computational model based on current metabolic rate parameters;

[0035] Calculate the residual between the temperature field output by the model and the measured value at the monitoring point;

[0036] The gradient vector of the metabolic rate parameter with respect to the objective function is calculated using the adjoint field method, and the parameter values ​​of each spatial unit are updated according to the gradient direction and then repeated. To avoid overfitting, a spatial smoothing regularization term is introduced to limit the sudden change of adjacent unit parameters. When the global residual change after three consecutive iterations is less than a preset tolerance threshold (such as two percent), the parameter field is judged to have converged. During the optimization process, the metabolic environment dynamic coupling function always serves as the core mathematical model connecting microbial metabolic activity and the temperature physical field, driving the model to update the output in response to parameter changes.

[0037] The activity field reorganization output unit 203 in the inversion analysis module 200 restructures the converged full-space microbial metabolic rate parameter values ​​based on the spatial discrete framework of the physical field calculation basis and outputs them as scalar field data with continuous spatial distribution, i.e., the metabolic activity spatial distribution field that characterizes the real-time microbial activity of the entire fermentation tank. The specific process includes:

[0038] First, based on the spatial discretization rules of the physical field calculation basis (such as finite element grids or isometric voxels), the microbial metabolic rate parameter values ​​of each spatial unit in the entire domain are reorganized through three-dimensional spatial interpolation.

[0039] For grid node parameter values, bilinear interpolation is used to generate continuous spatial distribution; for unstructured grids, distance-weighted interpolation is used to construct continuous field functions;

[0040] After reorganization, a scalar field matrix with spatial topological consistency is formed. Each element value in the matrix represents the relative metabolic activity intensity of the corresponding spatial position (normalized to a scale of zero to one hundred). The matrix is ​​then cleaned of outliers and smoothed to eliminate residual error interference at boundary constraint points.

[0041] The final output of the metabolic activity spatial distribution field is encapsulated in a three-dimensional data structure, including a spatial coordinate system, a timestamp, and an activity intensity matrix. The metabolic activity spatial distribution field can be directly visualized in the three-dimensional model of the fermentation tank on the monitoring interface, where warm-toned areas correspond to areas of high metabolic activity (such as the fermentation core area), and cool-toned areas correspond to areas of activity attenuation (such as the edge of the tank). It also serves as the initialization input for the subsequent deduction and prediction module 300, completing the technical closed loop from measured data to a panoramic view of biological activity.

[0042] The deduction and prediction module 300 receives the metabolic activity spatial distribution field output by the activity field reorganization output unit 203. The field data is encapsulated in the form of a three-dimensional scalar matrix, which contains the microbial metabolic rate parameter value of each spatial unit (such as 1 cm³ voxel); the deduction and prediction module 300 performs the initial state injection operation, first parsing the spatial coordinate system of the distribution field (which must strictly match the physical field calculation base grid of the joint calculation model), and writing the metabolic rate parameter value of each grid node into the metabolic environment dynamic coupling function input port of the model, overwriting the original initial value of the model; at the same time, the parameter range (0-100 % relative activity intensity), and boundary truncation is performed on over-limit values. After the injection is completed, the physical coupling state of the activity field, temperature field, and airflow field is verified through the model's built-in self-check program. For example, the high-activity area should match the positive gradient of the heat flux, and the low-activity area should be accompanied by gas retention characteristics. This process ensures that the initial conditions of the model accurately reflect the actual biological state of the current fermenter, avoiding the deduction offset of traditional prediction models caused by initial parameter mismatch. This step needs to be controlled within 5 seconds to meet real-time requirements, and finally outputs an initialized joint calculation model instance, laying a foundation for biological accuracy for transient deduction.

[0043] The deduction and prediction module 300 starts the iterative solution of the transient partial differential equations based on the initialized joint calculation model. It uses the fourth-order Runge-Kutta method to advance the calculation with a fixed time step (default 10 seconds). Four layers of nested operations are performed in each time step, specifically including:

[0044] Physical field update layer: Based on the current active parameters, the partial differential equation of heat conduction of hot materials is solved to obtain the temperature field distribution, and the gas convection and diffusion equation is solved simultaneously to update the airflow field;

[0045] Biofeedback layer: inputs the new temperature field and carbon dioxide concentration field into the metabolic environment dynamic coupling function, and outputs the metabolic rate correction value in reverse;

[0046] Parameter iteration layer: Newton-Raphson method is used to update the metabolic rate parameters of each spatial unit at the next moment;

[0047] Temperature field change curve: record the global average temperature, local hot spots (such as the highest temperature unit) and cold areas (such as the tank wall unit) temperature values;

[0048] Activity intensity gradient curve: calculate the change in metabolic rate per unit time and generate a spatially weighted gradient field;

[0049] Ultimately, a dual change curve is formed that simultaneously characterizes the evolution trends of the physical field and the biological field.

[0050] The deduction and prediction module 300 uses an intelligent mutation detection algorithm to detect the dual change curves. It applies a sliding window first-order difference (window width of 1 minute) to the temperature field change curve to capture sudden changes in the absolute value of the slope (threshold of ±3°C / min). It also applies a second-order derivative zero value detection (threshold of ±0.5% / min²) to the activity intensity gradient curve to locate acceleration anomalies. It performs a three-layer response for each detected mutation point, calculating the base space coordinates based on the physical field associated with the mutation point to determine the location of the abnormal area. It also calculates the time window for the occurrence of critical events based on the sudden change slopes of the dual change curves. It then converts the abnormal area location and this time window into a control signal for driving actuators (such as a stirring paddle and cooling tube). Specifically, this includes:

[0051] Spatial positioning and tracing: tracing back the joint computing model operation log to extract the physical field calculation base space coordinates corresponding to the mutation moment (such as the 3D grid ID of the heat source sudden increase area);

[0052] Time window estimation: Based on the extrapolation of the mutation slope exponent, the estimated time for the target physical quantity to reach the safety threshold is calculated. The formula is as follows:

[0053] T_max=t_now+(Thresh-T_now) / k, where k is the mutation slope, T_max represents the time point when the predicted physical quantity reaches the safety threshold, t_now represents the timestamp of the mutation point, Thresh represents the safety threshold (critical value of temperature or activity), and T_now is the measured physical quantity value at the current moment;

[0054] Instruction code generation: Convert the grid ID to the physical coordinates of the actuator (such as the X, Y, Z coordinates of the stirring paddle with a tolerance of ±2cm); define the execution window [t_act, t_end], where t_act = T_max - Δt_safe, t_act represents the time point when the actuator starts the operation, t_end represents the time point when the actuator stops the operation, and Δt_safe is the action time margin to avoid hysteresis and extra advance; the low-temperature activity decay zone triggers stepped stirring, and the high-temperature mutation zone triggers pulse spray cooling;

[0055] The final output structured control signal message contains the device ID, spatial coordinate set, time window, action sequence and failure fallback strategy (such as switching to emergency discharge when cooling fails).

[0056] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A rice wine fermentation monitoring system based on the Internet of Things, characterized in that: It includes a model building module (100), an inversion analysis module (200) and a deduction and prediction module (300), wherein: The model construction module (100) integrates the material heat conduction partial differential equation and the gas convection diffusion equation to form a physical field calculation base, and embeds the metabolic environment dynamic coupling function to couple the biological metabolic behavior and the physical field evolution to generate a joint calculation model; The inversion analysis module (200) obtains the temperature and humidity data of the sensor group, uses the temperature and humidity data as boundary constraints, performs inverse parameter optimization on the joint calculation model, and takes minimizing the residual between the temperature field output of the model and the actual sensor measurement value as the goal, iteratively corrects the microbial metabolic rate parameter value of each spatial unit in the model, and generates a converged metabolic activity spatial distribution field; The deduction and prediction module (300) uses a joint calculation model to predict time evolution and sets the metabolic activity spatial distribution field as the initial state of the model; it iteratively solves the transient partial differential equation group of the joint calculation model on a continuous time axis, and outputs a dual change curve including the temperature field and the activity intensity gradient in the future time window. When the dual change curve detects a critical mutation point, a control signal for driving the fermentation actuator is generated according to the spatial coordinates of the abnormal area and the predicted time window.

2. The rice wine fermentation monitoring system based on the Internet of Things according to claim 1, characterized in that, The metabolic environment dynamic coupling function converts the heat production rate into a heat source term of the heat conduction equation in the physical field calculation base, and converts the carbon dioxide generation rate into a convection velocity correlation term that affects the gas convection diffusion equation.

3. The rice wine fermentation monitoring system based on the Internet of Things according to claim 1, characterized in that: The inversion analysis module (200) includes a boundary constraint assimilation unit (201), and the boundary constraint assimilation unit (201) generates a boundary constraint process specifically including: The temperature and humidity data of the sensor group are matched to the spatial boundary nodes of the joint computing model according to the geometric structure of the fermentation container, and are converted into boundary constraints containing the actual node temperature values ​​and humidity gradient thresholds based on the data assimilation rules of the physical field calculation basis.

4. The rice wine fermentation monitoring system based on the Internet of Things according to claim 1, characterized in that: The inverse analysis module (200) includes a metabolic field iterative optimization unit (202), and the metabolic field iterative optimization unit (202) uses an iterative algorithm combining gradient descent with regularization constraints to perform inverse parameter optimization, specifically including: The forward solver of the joint computational model is run based on the current metabolic rate parameters. The residual between the temperature field output by the joint computational model and the actual value measured by the sensor is calculated. The metabolic rate parameter value is updated through the adjoint field method, and the spatial smoothing regularization term is loaded to limit parameter mutations.

5. The rice wine fermentation monitoring system based on the Internet of Things according to claim 1, characterized in that: The inversion analysis module (200) includes an activity field reorganization output unit (203), and the process of generating a metabolic activity spatial distribution field by the activity field reorganization output unit (203) specifically includes: The three-dimensional interpolation and reorganization of the microbial metabolic rate parameter values ​​are performed based on the spatial discretization rules of the physical field calculation basis; the outlier cleaning and edge smoothing processing are performed on the reorganized scalar field matrix; and the three-dimensional packaged data containing the spatial coordinate system, timestamp and activity intensity matrix are output as the spatial distribution field of metabolic activity.

6. The rice wine fermentation monitoring system based on the Internet of Things according to claim 1, characterized in that: The deduction and prediction module (300) initialization operation includes: Write the data of the metabolic activity spatial distribution field into the input port of the metabolic environment dynamic coupling function; check the parameter range and truncate the out-of-limit value; verify the physical coupling state of the activity field with the temperature field and airflow field through the self-test program.

7. The rice wine fermentation monitoring system based on Internet of Things according to claim 1, characterized in that: The generation process of the dual change curve specifically includes: The temperature field distribution of the material heat conduction partial differential equation and the airflow field of the gas convection-diffusion equation are updated in each time step; the metabolic rate correction value is reversely output through the metabolic environment dynamic coupling function; the metabolic rate parameters of each spatial unit at the next moment are updated and a double change curve is generated.

8. The rice wine fermentation monitoring system based on Internet of Things according to claim 1, characterized in that: The detection process of the critical mutation point includes: The sliding window first-order difference is applied to the temperature field change curve to capture the slope mutation; the second-order derivative zero value detection is applied to the activity intensity gradient curve to locate the acceleration abnormal point.

9. The rice wine fermentation monitoring system based on Internet of Things according to claim 1, characterized in that: The generation process of the control signal includes: The physical field associated with the mutation point is extracted to calculate the base space coordinates to determine the abnormal area; the predicted time window for the occurrence of critical events is calculated based on the mutation slope; the spatial coordinates of the abnormal area and the predicted time window are compiled into directional control instructions.

10. The rice wine fermentation monitoring system based on Internet of Things according to claim 9, characterized in that: The directional control instructions specifically include: Step-by-step stirring sequence in the low-temperature activity decay zone; Pulse spray cooling operation sequence in high temperature mutation zone; Unique device identifier, spatial coordinate set, and failure fallback strategy.

Images