Physical-data joint driven industrial composting process temperature prediction method and system

Abstract

The present application belongs to the technical field of organic solid waste resource processing, and particularly relates to a physical-data jointly driven industrial composting process temperature prediction method and system, comprising the following steps: S1: physical modeling and parameter acquisition; S2: composting data acquisition; S3: building and verifying a composting physical model; S4: building a training data set; and S5: building and training a joint prediction model. The present application realizes deep integration of physical mechanism and deep learning method by embedding residuals of physical control equations based on energy conservation and microbial dynamics as constraint terms into a loss function of a data driven model.

Description

Technical Field

[0001] This invention belongs to the field of organic solid waste resource utilization technology, specifically relating to a physical-data driven method and system for predicting temperature in industrial composting processes. Background Technology

[0002] Industrial composting is a key method for utilizing microbial metabolism to transform agricultural waste and organic solid waste into resources. The temperature of the compost pile is a crucial parameter reflecting the composting process and its operational status, directly affecting microbial activity, material degradation efficiency, and the final compost quality. Because the compost pile typically exhibits a porous media structure composed of solid particles, gaseous pores, and liquid water, and possesses large-scale and highly non-uniform characteristics, the spatiotemporal distribution of the internal temperature field exhibits significant nonlinearity and hysteresis, making it difficult to comprehensively and accurately characterize it in real-time using a limited number of physical measurement points.

[0003] Existing technologies for predicting composting temperature mainly fall into two categories: The first category is physical modeling methods based on heat transfer and reaction kinetics. These methods use governing equations for numerical simulation, but in engineering applications, they heavily rely on accurate input of physical properties and boundary conditions, thus limiting their prediction accuracy and adaptability. The second category is data-driven methods based on machine learning. The generalization ability and stability of these methods are highly dependent on the quality and distribution of historical data. Since actual operating conditions exceed the coverage of the training data, the prediction results may exhibit significant deviations.

[0004] To address the aforementioned issues, there is an urgent need for a temperature prediction method that combines high-precision fitting capabilities with physical consistency, enabling accurate prediction of the temperature field during composting. Summary of the Invention

[0005] The purpose of this invention is to provide a physical-data-driven method and system for predicting temperature in industrial composting processes, addressing the shortcomings of existing industrial and commercial integrated energy storage cabinet fire protection technologies in terms of rapid early suppression, accurate graded response, deep and continuous cooling, and system redundancy design.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a physical-data-driven method for predicting the temperature of an industrial composting process, comprising the following steps: S1: Physical modeling and parameter acquisition: The industrial compost pile is modeled as a porous media continuum composed of solid biomass, gaseous pores, and liquid water, and the geometric dimensions, environmental boundary conditions, and thermal properties of the raw materials are acquired; S2: Composting data acquisition: Time-series temperature data and corresponding operating parameters of the composting process are acquired using a sensor array arranged inside the compost pile; S3: Construction and verification of the compost pile physical model: Based on the law of conservation of energy and the microbial growth kinetics equation, physical control equations are established to describe the heat production, conduction, gas convection, and latent heat of water phase change during the composting process, thus forming the temperature prediction method for the composting process. The evolutionary physical model is used, and the measured data collected in step S2 is used for numerical solution and verification to obtain the theoretical temperature field distribution characteristics under physical constraints; S4: Construct a training dataset: Based on the physical model verified in step S3, numerical simulation is performed by changing the initial conditions and boundary conditions to generate a training dataset containing simulated temperature data and corresponding operating condition parameters, and the dataset is normalized; S5: Construct and train a joint prediction model: A data-driven prediction model is constructed using a long short-term memory network, and the training dataset generated in step S4 is used as input for training; at the same time, the residual of the physical control equation described in step S3 is used as a physical constraint loss term and embedded in the total loss function of the data-driven prediction model, and the final composting temperature prediction model is obtained through iterative optimization.

[0007] Preferably, the thermophysical parameters mentioned in step S1 include at least one of porosity, density, effective thermal conductivity, effective specific heat capacity, and adsorption isotherm.

[0008] Preferably, the microbial growth kinetic equation in step S3 adopts the Logistic growth model to describe the change of microbial biomass over time, and the rate of microbial metabolic heat production is calculated based on this change.

[0009] Preferably, in step S3, when constructing the physical model of the compost pile, the compost pile is considered as an isotropic porous medium, and its macroscopic thermophysical parameters are calculated using an equivalent method based on the weighted average of the volume or mass fractions of the solid, liquid, and gas phase components. Preferably, the normalization process in step S4 is performed using the following formula:

[0010] in, These are the normalized data values; The original data value; The maximum statistical value of this feature variable in the training set; This is the statistical minimum value of this feature variable in the training set.

[0011] Preferably, the total loss function described in step S5 Loss due to data fitting With physical constraint loss The weighted summation is expressed as follows:

[0012] in, This is the total loss function of the model; This represents the data fitting error; Errors in the physical equations; The weights for physical losses.

[0013] Preferably, the weights of the physical loss are dynamically adjusted based on an adaptive balancing strategy.

[0014] This invention also discloses a physical-data jointly driven temperature prediction system for industrial composting processes, used to implement the method, comprising: The data acquisition module is used to collect time-series temperature and humidity data in real time through a sensor array arranged in the compost pile, and simultaneously record ambient air temperature, ambient relative humidity and ventilation status parameters. The physical modeling module is used to construct and solve the physical control model of the composting process based on the geometric dimensions of the compost and the thermal properties of the raw materials, using the law of conservation of energy and the microbial growth kinetic equations, and to obtain the temperature field distribution characteristics under physical constraints. The dataset construction module is used to generate a training dataset containing simulated temperature data and operating condition parameters by changing the initial conditions and boundary conditions based on the physical control model, and to normalize the data. The joint prediction module is used to construct a long short-term memory network model, which is trained using the training dataset. The residuals of the equations of the physical control model are incorporated into the model loss function as physical constraints. The temperature prediction results of the composting process are output through iterative optimization. The central control terminal is communicatively connected to the data acquisition module, physical modeling and processing module, dataset construction module, and joint prediction module, and is used to coordinate the operation of each module, execute calculation instructions, and output prediction results.

[0015] Preferably, the data acquisition module includes multi-parameter sensor probes, signal transmitters, and data acquisition cards distributed at different depths inside the stack.

[0016] Preferably, the central control terminal includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the program to implement the steps of the method as described in any one of claims 1 to 7.

[0017] The beneficial effects of this invention are as follows: By embedding the residuals of the physical control equations based on energy conservation and microbial dynamics as constraints into the loss function of a data-driven model, this invention achieves a deep integration of physical mechanisms and deep learning methods. This method effectively overcomes the shortcomings of traditional physical models, such as reliance on precise parameters and poor adaptability, as well as the weak generalization ability and insufficient physical consistency of purely data-driven models. Under limited monitoring data conditions, it significantly improves the prediction accuracy and spatiotemporal generalization ability of nonlinear temperature field changes in the composting process, while ensuring that the prediction results conform to the basic laws of thermodynamics. This provides reliable technical support for the intelligent control, energy efficiency optimization, and stable operation of industrial composting processes. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the present invention; Figure 2 This is a diagram of the predictive model structure; Figure 3 This is a system block diagram of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that the following embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0020] like Figure 1 As shown, a physics-data jointly driven method for predicting temperature in industrial composting processes includes the following steps: Step S1: Physical Modeling and Parameter Acquisition. In this step, the industrial compost pile is considered as a porous media continuum composed of solid biomass particles, gaseous pores, and liquid water. The geometric dimensions and environmental boundary conditions of the pile are acquired. The thermophysical properties of the compost raw materials are acquired, including at least the following parameters: Structural parameters: porosity =0.35, density ρ=650kg / m³ 3 .

[0021] Thermodynamic parameters: effective thermal conductivity and effective specific heat capacity. Effective thermal conductivity can be calculated using a parallel or series model based on a weighted average of the thermal conductivity of the solid, liquid, and gas phases. The calculation formula is as follows:

[0022]

[0023] in, For effective thermal conductivity; For the first Volume fraction of the components; For the first The intrinsic thermal conductivity of the components; This is the equivalent specific heat capacity; For the first Mass fraction of the components; For the first The intrinsic specific heat capacity of each component.

[0024] In this embodiment, the solid-state thermal conductivity is 0.25 W / (mK), and the specific heat capacity is 2100 J / (kg·K); the initial water content is 60%, and water has a thermal conductivity of 0.6 W / (mK) and a specific heat capacity of 4180 J / (kg·K); air has a thermal conductivity of 0.026 W / (mK) and is extremely lightweight, so its mass is negligible. Substituting these values ​​into the formula yields λ. eff =0.3081W / (mK), c p,eff =3348J / (kg.K).

[0025] Adsorption characteristics: Adsorption isotherms describe the relationship between the water activity and equilibrium moisture content of a material, directly affecting the calculation of the latent heat of vaporization. In this embodiment, the GAB equation is used for fitting:

[0026] in To balance the adsorption amount, This refers to water activity.

[0027] Step S2: Composting Data Acquisition. Temperature and humidity data are measured using temperature and humidity sensors to collect time-series temperature data and corresponding operating parameters of the composting process.

[0028] Time-series temperature data acquisition: A multidimensional time series consisting of temperature data collected at different spatial locations within the compost pile at preset sampling time intervals, in the following form:

[0029] Where j is the temperature sensor number, This represents the temperature value collected by sensor j at time i.

[0030] Operating condition parameters are collected, including ambient air temperature, ambient relative humidity, ventilation status parameters, initial moisture content of the stack, temperature, and stack turning time information. The operating condition parameters and the time series temperature data correspond one-to-one in the time dimension.

[0031] Step S3: Constructing the reactor physical model. This step aims to establish the physical governing equations describing the spatiotemporal evolution of reactor temperature. Based on the law of conservation of energy and the Logistic equation, the following partial differential equations are constructed:

[0032] in, The rate of change of the internal thermal energy of the reactor system; This is the enthalpy flux term, representing the heat enthalpy carried in or out during the flow of matter; This is the heat conduction term based on Fourier's law; This refers to the thermal convection term caused by gas flow. Darcy velocity; This is the latent heat of phase change caused by the phase change of water. For latent heat of vaporization, The mass transfer coefficient is . The driving force is the water vapor partial pressure difference; The biopyrogen term is calculated using a microbial metabolic heat production model.

[0033] The microbial metabolic heat production model was constructed using the Logistic equation, and the calculation formula is as follows:

[0034]

[0035] in, The cumulative heat production per unit volume at time t; the theoretical maximum heat production potential per unit volume of stockpile. =1.8*10 7 J / kg; heat production rate constant =0.25d -1 Integral constant =45; The instantaneous biological heat production rate.

[0036] The above equations are numerically solved using the finite element method. The time-series temperature data collected in step S2 is used as reference data and compared with the temperature results calculated by the physical model in step S3. This allows us to obtain the theoretical temperature field distribution characteristics under physical constraints.

[0037] Step S4: Constructing the training dataset. To address the scarcity and insufficient coverage of real-world operating data, this step generates high-quality training data through numerical simulation based on the physical model validated in Step S3. Multi-condition simulation: Based on the physical model of the composting process verified in step S3, the physical model is run by changing parameters such as the initial temperature of the compost pile, environmental boundary conditions, ventilation strategy and initial moisture content, and obtaining the corresponding evolution data of compost temperature over time.

[0038] Data Combining and Normalization: The simulated temperature data is combined with the corresponding operating parameters, and the temperature data and operating parameters are then normalized. The normalization formula is as follows:

[0039] in, These are the normalized data values; The original data value; The maximum statistical value of this feature variable in the training set; This is the statistical minimum value of this feature variable in the training set.

[0040] For temperature data, set a statistical maximum value. =75℃, minimum value =30℃, substitute into the formula Normalize.

[0041] Step S5: Construct a joint prediction model. Based on the high-precision dataset obtained in Step S4, a data-driven prediction model is constructed using a Long Short-Term Memory (LSTM) network. Figure 2 The model comprises an input layer that receives historical temperature and operating condition sequences, a hidden layer that extracts temporal features, a fully connected layer that fuses features, and an output layer that predicts future temperatures. The input layer receives operating condition parameters, fermentation time, and the previous time-series internal temperature data of the pile. The output layer outputs the next time-series internal temperature data of the pile. A physical constraint loss term is constructed using the control equations described in step S3 and embedded into the total loss function of the data-driven prediction model. The total loss function is a weighted average of the data fitting error and the physical equation residuals, calculated as follows:

[0042] in, This is the total loss function of the model; This represents the data fitting error; Errors in the physical equations; The weights for physical losses.

[0043] The input layer has a dimension of 30 (24 temperature and humidity features + 6 operating condition features); the hidden layer has 64 neurons with a Dropout rate of 0.2; the fully connected layer has 32 neurons with the ReLU activation function; and the output layer has a dimension of 12 (predicting the temperature of the next 12 points).

[0044] Data fitting error: measures the mean squared error between the model's predicted values ​​and the dataset label values ​​in step S4.

[0045] Physical equations: Error measures the degree to which the prediction results violate the physical governing equations in step S3. The partial derivatives of the predicted temperature field with respect to time and space are calculated using automatic differentiation techniques and substituted into the partial differential equations to calculate the residuals.

[0046] Adaptive weights: Dynamically adjusted using an adaptive balancing strategy. The value. In the early stages of training. Set it to 0.01 to allow the model to learn the data distribution first; as iterations proceed, according to and The gradient magnitude ratio is dynamically adjusted between [0.01, 1] to make the model converge to a solution space that satisfies energy conservation. Iterative training is then performed.

[0047] In this embodiment, at the 100th iteration, the data fitting error is 0.045, the physical residual is 0.12, and the weight of the physical loss is 0.1. According to the formula... =0.045+0.1*0.12=0.057.

[0048] By minimizing The model parameters were trained for 2000 epochs using the Adam optimizer and a learning rate of 0.001. Converging to 10 -4 The magnitude of this difference ensures that the model's predictions still follow the law of conservation of energy at points in time and space where no monitoring data is available.

[0049] Based on the methods described in the above embodiments, the present invention also provides a physical-data jointly driven temperature prediction system for industrial composting processes. In this embodiment, the selected industrial compost pile geometry is a trapezoidal long stack. The bottom width of the pile is 3m, the top width is 1.5m, the pile height is 1.8m, and the pile length is 20m. Three cross-sections are selected at 5m, 10m, and 15m along the length of the pile. Sensors are arranged at the center bottom point of each cross-section at heights of 0.3m, 0.6m, 0.9m, and 1.2m above the ground, for a total of 12 points. Figure 3 As shown, the system mainly includes: Data acquisition module: used to collect time-series temperature and humidity data in real time through a sensor array arranged in the compost pile, and simultaneously record ambient air temperature, ambient relative humidity and ventilation status parameters. The data acquisition module includes several multi-parameter sensor probes, signal transmitters, and data acquisition cards distributed at different depths within the stack.

[0050] PT100 industrial-grade resistance temperature sensors and capacitive soil moisture sensors are installed at pre-set monitoring points inside the compost pile. The sensors are encased in corrosion-resistant stainless steel sheaths to withstand the high temperature and humidity of the composting environment. A meteorological monitoring unit is installed outside the compost pile to collect real-time data on ambient temperature, humidity, and wind speed.

[0051] The analog signals acquired by the sensor are converted into 4-20mA standard signals by a signal transmitter and then connected to a multi-channel data acquisition card.

[0052] The data acquisition card uploads the preprocessed time series data to the computing application layer in real time via a wired local area network.

[0053] The meteorological monitoring unit integrates a weather transmitter that measures wind speed, wind direction, and air temperature and humidity. It is installed on a column in the composting workshop at a height of 2 meters.

[0054] Physical modeling processing module: Integrated into the COMSOL simulation software of the industrial control computer, it is responsible for the physical model construction, solution and parameter correction of the S3 step; Dataset construction module: A Python script running on an industrial control computer calls the COMSOL Java API to automatically perform multi-condition simulations and complete the extraction, combination and normalization of data; Joint Prediction Module: An LSTM model program developed on an industrial control computer using the PyTorch framework, responsible for loading the trained joint prediction model. During online operation, it receives real-time data streams and continuously predicts the reactor temperature field for the next few hours. Central control terminal: An industrial computer (ICC) serves as the core of the system. Its processor executes integrated control software, which coordinates the workflow of each module: receiving and storing collected data, triggering periodic updates to the physical model, managing dataset construction tasks, executing training and online prediction of the joint prediction model, and displaying the prediction results in a visual graphical interface. Simultaneously, based on predicted temperature trends (such as predicted local overheating or insufficient temperature rise), it can automatically generate or prompt operators to implement control measures such as ventilation and turning.

[0055] The trained joint prediction model was deployed to the composting project. In actual operation, the system used real-time collected data as input, predicting the evolution of the internal temperature field of the compost pile every 30 minutes for the next 6 hours. The prediction results accurately reflected the dynamics of the pile's heating, high-temperature maintenance, and cooling stages. Particularly in predicting temperature changes after ventilation or turning, the accuracy was improved by approximately 15% compared to a pure LSTM model, and the prediction curves better conformed to physical laws (such as energy conservation). Based on the predicted "temperature bottlenecks" or "overheated areas," operators adjusted ventilation or scheduled turning in advance, effectively stabilizing the composting process and improving processing efficiency.

[0056] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A physical-data jointly driven method for predicting temperature in industrial composting processes, characterized in that, The process includes the following steps: S1: Physical modeling and parameter acquisition: The industrial compost pile is modeled as a porous media continuum composed of solid biomass, gaseous pores, and liquid water. The geometric dimensions, environmental boundary conditions, and thermal properties of the raw materials are acquired. S2: Compost data acquisition: Time-series temperature data and corresponding operating parameters of the composting process are acquired using a sensor array deployed inside the compost pile. S3: Construction and verification of the compost pile physical model: Based on the law of conservation of energy and the microbial growth kinetic equation, physical control equations describing the heat generation, conduction, gas convection, and latent heat effects of water phase change during the composting process are established. A physical model of temperature evolution during the composting process is formed, and the measured data acquired in step S2 is used for numerical solution and verification to obtain the theoretical temperature field distribution characteristics under physical constraints. S4: Construction of training dataset: Based on the physical model verified in step S3, numerical simulations are performed by changing the initial conditions and boundary conditions to generate a training dataset containing simulated temperature data and corresponding operating parameters. The dataset is then normalized. S5: Construct and train a joint prediction model: A data-driven prediction model is constructed using a long short-term memory network, and the training dataset generated in step S4 is used as input for training; at the same time, the residual of the physical control equation described in step S3 is used as a physical constraint loss term and embedded into the total loss function of the data-driven prediction model, and the final composting temperature prediction model is obtained through iterative optimization.

2. The method according to claim 1, characterized in that, The thermophysical parameters mentioned in step S1 include at least one of porosity, density, effective thermal conductivity, effective specific heat capacity, and adsorption isotherm.

3. The method according to claim 1, characterized in that, The microbial growth kinetic equation described in step S3 uses the Logistic growth model to describe the change in microbial biomass over time, and calculates the rate of microbial metabolic heat production based on this change.

4. The method according to claim 1, characterized in that, In step S3, when constructing the physical model of the compost pile, the compost pile is regarded as an isotropic porous medium, and its macroscopic thermophysical parameters are calculated using the weighted average method based on the volume fraction or mass fraction of the solid, liquid, and gas components.

5. The method according to claim 1, characterized in that, The normalization process described in step S4 is performed using the following formula: ; in, These are the normalized data values; The original data value; The maximum statistical value of this feature variable in the training set; This is the statistical minimum value of this feature variable in the training set.

6. The method according to claim 1, characterized in that, The total loss function described in step S5 Loss due to data fitting With physical constraint loss The weighted summation is expressed as follows: ; in, This is the total loss function of the model; This represents the data fitting error; Errors in the physical equations; The weights for physical losses.

7. The method according to claim 6, characterized in that, The weights of the physical loss are dynamically adjusted based on an adaptive balancing strategy.

8. A physical-data jointly driven temperature prediction system for industrial composting processes, used to implement the method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect time-series temperature and humidity data in real time through a sensor array arranged in the compost pile, and simultaneously record ambient air temperature, ambient relative humidity and ventilation status parameters. The physical modeling module is used to construct and solve the physical control model of the composting process based on the geometric dimensions of the compost and the thermal properties of the raw materials, using the law of conservation of energy and the microbial growth kinetic equations, and to obtain the temperature field distribution characteristics under physical constraints. The dataset construction module is used to generate a training dataset containing simulated temperature data and operating condition parameters by changing the initial conditions and boundary conditions based on the physical control model, and to normalize the data. The joint prediction module is used to construct a long short-term memory network model, which is trained using the training dataset. The residuals of the equations of the physical control model are incorporated into the model loss function as physical constraints. The temperature prediction results of the composting process are output through iterative optimization. The central control terminal is communicatively connected to the data acquisition module, physical modeling and processing module, dataset construction module, and joint prediction module, and is used to coordinate the operation of each module, execute calculation instructions, and output prediction results.

9. The system according to claim 8, characterized in that, The data acquisition module includes multi-parameter sensor probes, signal transmitters, and data acquisition cards distributed at different depths inside the stack.

10. The system according to claim 8, characterized in that, The central control terminal includes a memory, a processor, and a computer program stored in the memory. When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 7.

Images