A biomass carbonization process control method and system based on artificial intelligence

Abstract

The application discloses a biomass carbonization process control method and system based on artificial intelligence, relates to the technical field of biomass carbonization, and comprises the following steps: performing real-time scanning on chemical components of biomass, storing collected data into a chemical component vector, acquiring biomass particle size distribution and porosity information, and generating a physical property vector; calculating the decomposition rate of components and recursively updating the chemical components, analyzing the reaction rate based on the decomposition rate and the mass fraction of the chemical components, calculating the reaction heat generated by the decomposition reaction of each component, collecting data of an array of temperature sensors in a furnace, and calculating the conduction heat flow by using a finite difference algorithm.The method realizes the coupled simulation of thermochemical reaction and heat transfer by calculating the decomposition rate in real time by using the Arrhenius formula and recursively updating the chemical components, and dynamically updating the temperature distribution by using the finite difference algorithm.

Description

Technical Field

[0001] This invention relates to the field of biomass carbonization technology, and in particular to a biomass carbonization process control method and system based on artificial intelligence. Background Technology

[0002] Biomass carbonization technology is a process that uses pyrolysis to decompose biomass raw materials in an oxygen-deficient environment to produce solid char, gaseous and liquid products. In this process, solid char, as a high-energy-density solid fuel or carbon-based material, has broad market prospects. The reaction process of biomass carbonization involves complex physical, chemical and heat transfer mechanisms. For example, variables such as the chemical composition, particle size characteristics, pore structure, decomposition reaction rate and gas generation behavior of biomass have a significant impact on the carbonization process. At present, traditional carbonization process control methods are mainly based on empirical parameters to adjust reaction conditions. In addition, existing monitoring methods are mostly limited to single-variable monitoring.

[0003] However, in the process of biomass carbonization, traditional carbonization process control methods based on empirical models cannot capture the dynamic changes of biomass chemical components in real time. Their limitation lies in the failure to fully utilize the correlation between chemical components, physical properties, and gas generation data, resulting in low prediction accuracy of carbonization reaction rate and gas generation rate, and limiting the control effect. On the other hand, existing methods have failed to successfully achieve multivariate feedback closed-loop control in complex environments, especially lacking effective solutions for real-time adjustment of high-dimensional variables (such as temperature field, gas concentration, and decomposition rate), resulting in the underutilization of dynamic changes in the reaction process. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based biomass carbonization process control method and system to solve the problem that traditional empirical model-based carbonization process control methods cannot capture the dynamic changes of biomass chemical components in real time. Their limitation lies in the failure to fully utilize the correlation between chemical components, physical properties, and gas generation data, resulting in low prediction accuracy of carbonization reaction rate and gas generation rate, and limited control effect. On the other hand, existing methods have failed to successfully achieve multivariate feedback closed-loop control in complex environments, especially lacking effective solutions for real-time adjustment of high-dimensional variables (such as temperature field, gas concentration, and decomposition rate), leading to the problem that the dynamic changes in the reaction process are not fully utilized.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for controlling the biomass carbonization process based on artificial intelligence, comprising:

[0008] Real-time scanning of biomass chemical components is performed, and the collected data is stored in a chemical component vector to obtain biomass particle size distribution and porosity information, and to generate a physical property vector.

[0009] The decomposition rate of the components is calculated and the chemical composition is updated recursively. The reaction rate is analyzed based on the decomposition rate and the mass fraction of the chemical components, and the exothermic reaction generated by the decomposition reaction of each component is calculated. Data from the furnace temperature sensor array is collected, and the conduction heat flow is calculated using the finite difference algorithm. The furnace temperature distribution is analyzed and updated. Gas concentration data is collected to calculate the volatile matter generation rate, and the total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method.

[0010] Based on the collected physical property vector data, the effective specific surface area is defined, the contribution of surface area to the reaction rate is analyzed, the particle size influence factor is defined, the solid carbon generation rate is analyzed in real time using a kinetic model, and a regression model is constructed using an artificial neural network (ANN) to output predicted carbon quality data.

[0011] An SVM prediction model was constructed and optimized using the artificial bee colony algorithm. The deviation between the predicted gas concentration data and the updated gas concentration data was calculated, and the deviation between the predicted carbon mass data and the actual carbon mass data was combined to form a feedback signal vector.

[0012] A deep Q-network is used to define the state space vector and action space vector, and a reward function is defined based on the feedback signal vector to predict the optimal control output.

[0013] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, the following steps are included: analyzing and updating the temperature distribution within the furnace, collecting gas concentration data to calculate the volatile matter generation rate, and accumulating the contributions of all chemical components to obtain the total gas generation rate; based on the monitored gas flow rate and gas concentration vector, updating the gas concentration distribution within the control volume using the finite volume method, including...

[0014] Based on the collected chemical component vector data, the decomposition rate is calculated using the Arrhenius formula, and the chemical components are updated using the mass fraction recursive formula.

[0015] The reaction rate of the carbonization process is calculated based on the decomposition rate and the mass fraction of the chemical components, and the heat of chemical reaction generated by the decomposition reaction of each component is calculated through the enthalpy value.

[0016] Based on the data collected from the temperature sensor array inside the carbonization furnace, the finite difference algorithm is used to calculate the conduction heat flow according to the gradient of the temperature sampling points. Combined with the heat released by the chemical reaction, the temperature distribution inside the furnace is analyzed and the temperature distribution value is updated.

[0017] The gas concentration vector is composed of the gas concentrations of carbon monoxide, carbon dioxide, methane and hydrogen. The volatile matter generation rate is calculated based on the decomposition rate of different components. The total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method (FVM).

[0018] Data is recorded based on updated temperature distribution data, gas concentration data, and gas generation rate data in the carbonization furnace.

[0019] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, the method includes: real-time analysis of the solid carbon generation rate using a kinetic model, and constructing a regression model using an artificial neural network (ANN) to output predicted carbon quality data, including:

[0020] Based on the collected physical property vector data, and considering the influence of particle specific surface area and porosity, the effective specific surface area is defined comprehensively.

[0021] The particle size influence factor is defined based on the contribution of effective specific surface area to the reaction rate.

[0022] The carbon fixation rate is determined based on the content of different components in the chemical composition vector data and the fixed proportion converted into solid carbon.

[0023] Based on carbon fixation rate, particle size influencing factors, and temperature distribution data, a kinetic model was used to describe the chemical process of biomass carbon conversion in the furnace and to analyze the solid carbon generation rate.

[0024] A regression model is constructed using an artificial neural network (ANN), including an input layer, hidden layers, and an output layer.

[0025] The input layer takes in updated temperature distribution data, gas generation rate data, particle size influence factor and carbon fixation rate, the hidden layer uses ReLU activation function to capture nonlinear dynamic characteristics, and the output layer outputs predicted carbon mass data.

[0026] Based on the calibrated training set data, the cross-entropy loss function is selected to calculate the computational loss of the model's predicted output. The Adam optimizer is used for gradient descent optimization to update the model parameters. The iteration stops when the change in computational loss is less than or equal to the historical loss threshold, and the model parameters are output to update the model.

[0027] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, wherein: the feedback signal vector is formed by combining the deviation values ​​between the predicted carbon quality data and the actual carbon quality data, including:

[0028] The SVM prediction model is constructed using a support vector machine and a kernel function. The updated temperature distribution data, gas generation rate data, and flow rate data are input, and the concentration of each gas is predicted.

[0029] The sum function of the SVM prediction model is selected as the radial basis function, the optimization objective is determined to be minimizing the model prediction error, and the computational loss of the model prediction output is calculated by the cross-entropy loss function.

[0030] The artificial bee colony algorithm is used to iteratively optimize the kernel parameters and regularization parameters. The parameter solutions are randomly assigned and used as the population. The calculated loss is used as the fitness value. A global search is performed to output parameters based on the optimization objective and update the SVM prediction model.

[0031] The deviation between the predicted gas concentration data and the updated gas concentration data is calculated, and the deviation between the predicted carbon mass data and the actual carbon mass data is combined to form a feedback signal vector, which includes gas concentration deviation and carbon mass deviation.

[0032] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, the step of defining a reward function based on the feedback signal vector and predicting the optimal control output includes:

[0033] Closed-loop regulation is based on reinforcement learning algorithm, in which temperature distribution data, gas concentration data and gas generation rate data are defined to form a state space vector, and temperature adjustment, airflow rate adjustment and reaction time adjustment are used as action space vectors.

[0034] The reward function is defined as the deviation minimization function. A deep Q-network (DQN) is used to predict the optimal value of each action space vector as the control output.

[0035] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, the step of acquiring biomass particle size distribution and porosity information and generating a physical property vector includes:

[0036] High-precision particle size analyzers and porosity detectors are used to obtain biomass particle size distribution and porosity information, and physical property vectors are generated.

[0037] As a preferred embodiment of the artificial intelligence-based biomass carbonization process control method of the present invention, wherein: the real-time scanning of biomass chemical components and the storage of collected data into a chemical component vector include,

[0038] The biomass chemical components were scanned in real time using a near-infrared spectroscopy analyzer (NIRS) to generate data on the mass content of cellulose, hemicellulose, and lignin. Each sample generated a value, which was stored in a chemical component vector.

[0039] Secondly, the present invention provides an artificial intelligence-based biomass carbonization process control system, comprising,

[0040] Real-time acquisition module: Real-time scanning of the chemical composition and physical properties of biomass;

[0041] Decomposition reaction module: Calculates the decomposition rate of chemical components and recursively updates the component mass parameters. It also analyzes the reaction rate and the exothermic reaction of the chemical reaction and supports dynamic heat flow analysis.

[0042] Temperature and gas monitoring module: Based on the temperature and gas concentration data collected by the in-furnace sensors, the temperature distribution and gas concentration distribution are updated using the finite difference method and the finite volume method, respectively;

[0043] Physical property analysis module: comprehensively defines the effective specific surface area and particle size influence factors, quantifies the contribution of physical properties to the reaction rate, and supports real-time kinetic model calculation of solid carbon formation rate;

[0044] Artificial Neural Network Regression Module: Predicts carbon mass data in real time using an ANN regression model;

[0045] Support Vector Machine Prediction Optimization Module: Based on SVM prediction of gas concentration data, the kernel parameters are optimized using the artificial bee colony algorithm, and a feedback signal vector is constructed to correct the bias of the model;

[0046] Reinforcement learning control module: It uses a deep Q-network to define the state space and action space, and predicts the optimal control output based on the feedback signal vector and reward function.

[0047] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the artificial intelligence-based biomass carbonization process control method as described in the first aspect of the present invention.

[0048] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based biomass carbonization process control method as described in the first aspect of the present invention.

[0049] The beneficial effects of this invention are as follows: By calculating the decomposition rate in real time using the Arrhenius formula and recursively updating the chemical composition, and by combining the finite difference algorithm for dynamic updating of the temperature distribution, the coupled simulation of thermochemical reaction and heat transfer is realized. By comprehensively analyzing the solid carbon formation rate through particle size influence factors and temperature distribution data, the bias that may be caused by neglecting particle size effects or multi-component coupling effects in traditional carbonization models is effectively avoided. By combining SVM, artificial bee colony algorithm and feedback signal vector, global optimization can be achieved while capturing complex nonlinear features. Based on feedback closed loop, rapid correction is achieved. The optimal value of the action space vector is predicted through deep Q-network (DQN) and used as the control output. In the process of combining state space and reward function, the limitations of traditional regulation methods in dealing with high-dimensional complex regulation scenarios are effectively solved. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of the process control method for biomass carbonization based on artificial intelligence in Example 1.

[0052] Figure 2 This is a schematic diagram of the structure of the biomass carbonization process control system based on artificial intelligence in Example 1. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0056] Example 1, referring to Figures 1 to 2This is the first embodiment of the present invention, which provides a method for controlling the biomass carbonization process based on artificial intelligence, including the following steps:

[0057] S1, real-time scanning of biomass chemical components, data collection and storage into chemical component vector, obtaining biomass particle size distribution and porosity information, and generating physical property vector;

[0058] Preferably, the chemical components of biomass are scanned in real time, and the collected data is stored in a chemical component vector, including:

[0059] The biomass chemical components were scanned in real time using a near-infrared spectroscopy (NIRS) analyzer to generate data on the mass content of cellulose, hemicellulose, and lignin. Each sample generated one value, which was stored in a chemical component vector, represented as follows:

[0060] ;

[0061] in, This represents the vector of chemical components at time t. This indicates the cellulose mass content over time t. This indicates the hemicellulose mass content over time t. The lignin content is expressed as a time interval (t).

[0062] Furthermore, information on biomass particle size distribution and porosity is obtained to generate a physical property vector, including,

[0063] High-precision particle size analyzer and porosity analyzer are used to obtain biomass particle size distribution and porosity information, and a physical property vector is generated. The physical property vector is represented as follows:

[0064] ;

[0065] in, A vector representing the physical properties at time t. This represents the average particle size over time t.

[0066] This represents the porosity percentage over time t.

[0067] S2, calculate the decomposition rate of the components and recursively update the chemical components, analyze the reaction rate based on the decomposition rate and the mass fraction of the chemical components, calculate the heat of reaction generated by the decomposition reaction of each component, collect data from the furnace temperature sensor array, calculate the conduction heat flow using the finite difference algorithm, analyze and update the furnace temperature distribution, collect gas concentration data to calculate the volatile matter generation rate, and accumulate the contributions of all chemical components to obtain the total gas generation rate. Based on the monitored gas flow rate and gas concentration vector, update the gas concentration distribution within the control volume using the finite volume method.

[0068] Preferably, the furnace temperature distribution is analyzed and updated, gas concentration data is collected to calculate the volatile matter generation rate, and the total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method, including...

[0069] Based on the collected chemical component vector data, the decomposition rate is calculated using the Arrhenius equation, and the chemical components are updated using a mass fraction recursive formula, as follows:

[0070] ;

[0071] ;

[0072] ;

[0073] in, This represents the decomposition rate at time t. The frequency factor representing component i is obtained through experimental calibration. The activation energy of component i represents the minimum energy required for a chemical reaction to occur; its specific value can be obtained from literature or experimental data. This represents the gas constant 8.314. , The reaction temperature is indicated by data collected from a real-time temperature measuring device. Indicates time At time step Update the mass fraction of the i-th chemical component. This represents the mass fraction of the i-th chemical component at time t. Indicates the decomposition rate of the component;

[0074] The reaction rate of the carbonization process is calculated based on the decomposition rate and the mass fraction of the chemical components, and the heat of chemical reaction released by the decomposition reaction of each component is calculated using the enthalpy value, expressed as:

[0075] ;

[0076] ;

[0077] in, This represents the reaction rate of component i at time t. The amount of heat released in a chemical reaction at time t is called the heat released. The decomposition enthalpy value of each component is indicated by experimental determination or literature review. Indicates the total number of components;

[0078] In a carbonization furnace, heat transfer primarily occurs through convection and conduction, resulting in a dynamic temperature field distribution. Data from an array of temperature sensors within the furnace is collected. Based on the gradient at each temperature sampling point, a finite difference algorithm is used to calculate the conductive heat flux. This, combined with the heat released by the chemical reaction, is used to analyze the temperature distribution within the furnace and update the temperature distribution values, as expressed below:

[0079] ;

[0080] ;

[0081] in, This indicates thermal conductivity, which is set according to the furnace wall material. The temperature distribution variable is represented by data collected in real time from an in-furnace temperature sensor. , as well as These represent the temperature data at time steps i+1, i, and i-1, respectively. This represents the temperature data at position i at time step n+1. Indicates time interval, Indicates the spatial node spacing. This represents the heat released by the chemical reaction at time step n. Indicates the density of biomass materials. Indicates the specific heat capacity of biomass materials;

[0082] Gas mass flow meters and concentration sensors are deployed at the gas inlet and outlet of the gas channel. The gas concentrations of carbon monoxide, carbon dioxide, methane, and hydrogen are collected to form a gas concentration vector. The volatile matter generation rate is calculated based on the decomposition rates of different components, and the total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method (FVM), as shown below:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] in, Represents the gas concentration vector. Indicates gas flow rate, Indicates the gas generation rate, The rate of gas emission loss is determined experimentally from the equipment's exhaust emissions. and Let these represent the gas concentration vectors at time steps n and n+1, respectively. This represents the volatile matter formation rate of component i at time t. This represents the gas yield when component i decomposes, i.e., the mass fraction of gas generated from the decomposition of 1 unit of component i (obtained experimentally). This represents the total gas generation rate at time t;

[0088] Data is recorded based on updated temperature distribution data, gas concentration data, and gas generation rate data in the carbonization furnace.

[0089] By calculating the decomposition rate in real time using the Arrhenius equation and recursively updating the chemical composition, combined with the dynamic updating of the temperature distribution using the finite difference algorithm, a coupled simulation of thermochemical reaction and heat transfer is achieved. The calculation of the decomposition rate directly depends on real-time temperature data (acquired by sensors), while the updating of the temperature distribution is based on the heat released by the chemical reaction (derived from the decomposition rate). This bidirectional dependence forms a positive feedback loop: temperature changes immediately affect the decomposition of chemical components (e.g., increased temperature accelerates decomposition), while the heat released by decomposition is incorporated into the heat conduction equation through the finite difference algorithm, making the temperature field simulation more accurate, avoiding the problem of temperature and reaction being disconnected in traditional open-loop control, and reducing the risk of thermal runaway.

[0090] The carbonization process reaction rate, calculated based on the decomposition rate and the mass fraction of chemical components, can not only obtain an accurate value of the reaction rate from the decomposition rate, but also mark the energy distribution of each reaction stage by combining the heat of exothermic reaction of the products.

[0091] By calculating the conduction heat flow using the finite difference algorithm and combining it with the heat released by the chemical reaction to update the furnace temperature distribution data, the shortcomings of purely chemical or physical methods in lacking global dynamics are effectively filled. Under conditions of large temperature field changes, it can simultaneously ensure the accuracy of data updates and the dynamic uniformity of the reaction area. By using concentration sensors and mass flow meters at the gas inlet and outlet, combined with the finite volume method (FVM) to update the gas concentration distribution, it can not only accurately calculate the gas concentration changes and generation rate, but also dynamically reflect the actual effect of gas distribution by combining flow rate and discharge rate.

[0092] By combining the calculation of volatile matter generation rate and the cumulative update of total gas generation rate with the dynamic adjustment of automatic feedback gas concentration vector, the generation rate and decomposition rate can still be kept consistent even when volatile matter participates in frequency fluctuations. This ensures stable gas generation efficiency and avoids the impact of excessive variable fluctuations on system performance. Compared with traditional methods, the addition of comprehensive feedback control of gas generation rate significantly improves gas utilization.

[0093] By recording and updating temperature distribution, gas concentration, and generation rate data, and linking them with the heat release and decomposition rate of the chemical reaction, a closed-loop data chain is formed, providing high-fidelity input for subsequent artificial intelligence predictions. This breaks down the barriers of traditional segmented models: the heat release of the chemical reaction affects the temperature, the temperature drives the decomposition rate, and the gas generation rate feeds back to the concentration distribution, forming a data closed loop. This integration enables the system to capture nonlinear interactions (such as the hysteretic effect of porosity on the reaction rate), reduce cumulative errors, and provide low-noise input for reinforcement learning optimization.

[0094] S3. Based on the collected physical property vector data, the effective specific surface area is defined, the contribution of surface area to the reaction rate is analyzed, the particle size influence factor is defined, the solid carbon generation rate is analyzed in real time using the kinetic model, and the regression model is constructed using the artificial neural network (ANN) to output predicted carbon quality data.

[0095] Preferably, the solid carbon formation rate is analyzed in real time using a kinetic model, and a regression model is constructed using an artificial neural network (ANN) to output predicted carbon quality data, including:

[0096] Based on the collected physical property vector data, considering the influence of particle specific surface area and porosity, the effective specific surface area is comprehensively defined and expressed as:

[0097] ;

[0098] in, This represents the effective specific surface area of ​​the particles. This indicates the proportion of the particle's external filling rate. Indicates the porosity of the particles. This represents the specific surface area of ​​the particles. Indicates the average particle size;

[0099] Based on the contribution of effective specific surface area to the reaction rate, the particle size influence factor is defined as follows:

[0100] ;

[0101] in, Indicates the particle size influencing factor. Represents the normalization constant;

[0102] For normalization constant Based on reasonable industrial assumptions, the volume and surface area formulas are used for calculation, expressed as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] in, Indicates the reference volume. Indicates the reference surface area. This represents the average reference particle size.

[0108] Indicates the reference porosity ratio. This represents the effective specific surface area of ​​the reference particle. Used to ensure At that time, the particle size influence factor was 1;

[0109] Based on the content of different components in the chemical composition vector data, and combined with the fixed proportion of conversion into solid carbon (reference values ​​can be used uniformly based on literature data or existing kinetic models), the carbon fixation rate is determined.

[0110] Based on carbon fixation rate, particle size influencing factors, and temperature distribution data, a kinetic model is used to describe the chemical process of biomass carbon conversion in the furnace, and the solid carbon formation rate is analyzed, expressed as:

[0111] ;

[0112] in, This represents the mass of carbon at time t. This represents the real-time calculated rate of solid carbon formation. Indicates update Decomposition rate, Indicates the carbon fixation rate. This represents the temperature data updated at location x and time t.

[0113] A regression model is constructed using an artificial neural network (ANN), including an input layer, hidden layers, and an output layer.

[0114] The input layer takes in updated temperature distribution data, gas generation rate data, particle size influence factor and carbon fixation rate, the hidden layer uses ReLU activation function to capture nonlinear dynamic characteristics, and the output layer outputs predicted carbon mass data.

[0115] Based on the calibrated training set data, the cross-entropy loss function is selected to calculate the computational loss of the model's predicted output. The Adam optimizer is used for gradient descent optimization to update the model parameters. The iteration stops when the change in computational loss is less than or equal to the historical loss threshold, and the model parameters are output to update the model.

[0116] By defining the effective specific surface area of ​​particles and considering factors such as specific surface area, filling rate, porosity, and particle size, and combining normalized volume, surface area, and reference parameters to calculate the particle size influence factor, the shortcomings of traditional methods that only consider specific surface area and ignore geometric and structural characteristics can be eliminated when constructing input parameters based on particle properties. This allows for a more accurate quantification of the actual impact of particle properties on reaction rate.

[0117] By introducing carbon fixation rate and combining it with chemical component content and mass conversion ratio, particle characteristics and decomposition kinetics are further combined, which solves the problem of insufficient processing capacity of traditional models for different raw materials, while ensuring the logical consistency of reaction rate and generation rate during carbonization.

[0118] By describing the chemical reaction of the in-furnace carbonization process using a kinetic model based on carbon fixation rate, particle size influencing factors, and temperature distribution data, we can not only dynamically adjust the reaction rate, but also grasp the local influence of temperature gradient on the generation rate, providing a reliable basis for the analysis of the complex coupling relationship between the in-furnace temperature field and chemical reaction.

[0119] By comprehensively analyzing the solid carbon formation rate using particle size influence factors and temperature distribution data, the biases that may be caused by neglecting particle size effects or multi-component coupling effects in traditional carbonization models are effectively avoided. At the same time, the accuracy of the carbonization reaction description is improved by introducing particle data, making the calculation of the solid carbon formation rate closer to the actual physical process.

[0120] Training a regression model using an artificial neural network (ANN) effectively utilizes the ANN's ability to fit nonlinear dynamic systems, accurately capturing dynamic characteristics under multivariate input conditions.

[0121] S4. Construct an SVM prediction model and optimize it using the artificial bee colony algorithm. Calculate the deviation between the predicted gas concentration data and the updated gas concentration data. Combine the deviation between the predicted carbon mass data and the actual carbon mass data to form a feedback signal vector.

[0122] Preferably, the deviation between the predicted carbon mass data and the actual carbon mass data is used to form a feedback signal vector, including:

[0123] The SVM prediction model is constructed using a support vector machine and a kernel function. The updated temperature distribution data, gas generation rate data, and flow rate data are input, and the concentration of each gas is predicted.

[0124] The sum function of the SVM prediction model is selected as the radial basis function, the optimization objective is determined to be minimizing the model prediction error, and the computational loss of the model prediction output is calculated by the cross-entropy loss function.

[0125] The artificial bee colony algorithm is used to iteratively optimize the kernel parameters and regularization parameters. The parameter solutions are randomly assigned and used as the population. The calculated loss is used as the fitness value. A global search is performed to output parameters based on the optimization objective and update the SVM prediction model.

[0126] The deviation between the predicted gas concentration data and the updated gas concentration data is calculated, and the deviation between the predicted carbon mass data and the actual carbon mass data is combined to form a feedback signal vector, which includes gas concentration deviation and carbon mass deviation.

[0127] By training the input temperature distribution, gas generation rate, and flow velocity data using an SVM model combined with a radial basis function kernel, and predicting gas concentration, this method leverages the nonlinear mapping properties of the kernel function to map complex multidimensional input data to a high-dimensional space. This accurately captures the nonlinear relationship between gas concentration and complex input variables. The cross-entropy loss function is used to calculate the error of the SVM model output, and an artificial bee colony algorithm is introduced to optimize the kernel parameters and regularization parameters. This achieves autonomous adaptive iterative optimization of the model. Compared with traditional hyperparameter tuning methods (such as grid search), this method can explore the parameter space more efficiently, avoid local optima traps, and obtain more accurate and reliable kernel parameters and regularization weights.

[0128] By optimizing the loss function based on fitness value using the artificial bee colony algorithm and updating the SVM prediction model with output parameters based on global search, the flexibility of the SVM kernel function and the global optimization capability of the artificial bee colony algorithm are organically combined, allowing the model to capture more stable features and ensuring that the nonlinear relationship between input variables is always maintained in the optimization space, thereby further reducing prediction bias and enhancing the generalization ability of the SVM model.

[0129] By calculating the deviation between predicted and measured gas concentration data, and the deviation between predicted and measured carbon mass, a feedback signal vector is generated, which can effectively achieve closed-loop control and enable the prediction model to dynamically self-correct based on the deviation.

[0130] By inputting the gas concentration deviation and carbon mass deviation values ​​into a joint feedback signal vector into the system, multiple error sources can be effectively integrated, breaking the limitations of local optimization that may be caused by a single deviation source. This ensures that the prediction results of gas concentration and carbon mass are simultaneously optimized in the global scope, improving the overall ability of the model to handle complex multi-objective problems.

[0131] By combining SVM, artificial bee colony algorithm and feedback signal vector, this approach can capture complex nonlinear features while achieving global optimization and make rapid corrections based on feedback loop. Compared with using nonlinear mapping or optimization algorithms alone, this approach can demonstrate extremely high prediction accuracy, adaptability and robustness in dynamic industrial scenarios.

[0132] S5 utilizes a deep Q-network to define the state space vector and action space vector, and defines a reward function based on the feedback signal vector to predict the optimal control output;

[0133] Preferably, a reward function is defined based on the feedback signal vector to predict the optimal control output, including:

[0134] Closed-loop regulation is based on reinforcement learning algorithm, in which temperature distribution data, gas concentration data and gas generation rate data are defined to form a state space vector, and temperature adjustment, airflow rate adjustment and reaction time adjustment are used as action space vectors.

[0135] Define the reward function as the minimum bias, expressed as:

[0136] ;

[0137] in, Represents the objective function value. Indicates carbon quality deviation. Indicates gas concentration deviation;

[0138] A deep Q-network (DQN) is used to predict the optimal value of each action space vector, which serves as the control output.

[0139] By defining a state space vector based on temperature distribution data, gas concentration data, and gas generation rate data, and setting an action space vector that includes temperature adjustment, airflow rate adjustment, and reaction time adjustment, the direct correlation between the system's dynamic state and the regulation variables is realized, thereby comprehensively reflecting the complex coupling relationship of the system.

[0140] By using the minimization of bias as the reward function, and combining the definitions of carbon mass bias and gas concentration bias, an optimization objective for reinforcement learning closed-loop regulation is constructed. This allows multiple regulation objectives to be centralized into a unified optimization framework, thereby avoiding the possibility of other variable biases caused by single-objective optimization.

[0141] By predicting the optimal value of the action space vector using a Deep Q-Network (DQN) and using it as the control output, this method effectively addresses the limitations of traditional regulation methods in handling high-dimensional and complex regulation scenarios by combining state space and reward functions. DQN leverages deep learning to fit the complex nonlinear relationship between states and reward functions, thereby quickly finding the global optimum under multi-action regulation. Compared to general tabular algorithms like Q-learning, DQN, combined with deep learning, can adapt to high-dimensional states and variables in complex industrial scenarios.

[0142] By leveraging the closed-loop mechanism of state, action, and reward in reinforcement learning algorithms, temperature distribution, gas generation rate, and concentration data are dynamically fed back to action adjustment, achieving collaborative optimization of real-time adjustment and prediction. Action optimization is driven by a reward mechanism that minimizes the deviation of a unified objective function, and optimal action output is achieved by combining the DQN architecture. This continuously reduces prediction deviation and achieves cumulative optimization across time steps.

[0143] By integrating carbon mass deviation and gas concentration deviation into a unified reward function, the global consistency of the output results is enhanced, thereby avoiding the situation where optimizing a single variable may lead to the deterioration of other performance indicators.

[0144] This embodiment also provides an artificial intelligence-based biomass carbonization process control system, including,

[0145] Real-time acquisition module: Real-time scanning of the chemical composition and physical properties of biomass;

[0146] Decomposition reaction module: Calculates the decomposition rate of chemical components and recursively updates the component mass parameters. It also analyzes the reaction rate and the exothermic reaction of the chemical reaction and supports dynamic heat flow analysis.

[0147] Temperature and gas monitoring module: Based on the temperature and gas concentration data collected by the in-furnace sensors, the temperature distribution and gas concentration distribution are updated using the finite difference method and the finite volume method, respectively;

[0148] Physical property analysis module: comprehensively defines the effective specific surface area and particle size influence factors, quantifies the contribution of physical properties to the reaction rate, and supports real-time kinetic model calculation of solid carbon formation rate;

[0149] Artificial Neural Network Regression Module: Predicts carbon mass data in real time using an ANN regression model;

[0150] Support Vector Machine Prediction Optimization Module: Based on SVM prediction of gas concentration data, the kernel parameters are optimized using the artificial bee colony algorithm, and a feedback signal vector is constructed to correct the bias of the model;

[0151] Reinforcement learning control module: It uses a deep Q-network to define the state space and action space, and predicts the optimal control output based on the feedback signal vector and reward function.

[0152] This embodiment also provides a computer device applicable to the case of an artificial intelligence-based biomass carbonization process control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the artificial intelligence-based biomass carbonization process control method proposed in the above embodiment.

[0153] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0154] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the artificial intelligence-based biomass carbonization process control method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0155] In summary, this invention calculates the decomposition rate in real time using the Arrhenius equation and recursively updates the chemical composition. Combined with the finite difference algorithm for dynamic updating of temperature distribution, it achieves coupled simulation of thermochemical reaction and heat transfer. By comprehensively analyzing the solid carbon formation rate through particle size influence factors and temperature distribution data, it effectively avoids the biases that may be caused by neglecting particle size effects or multi-component coupling in traditional carbonization models. Through the coupled combination of SVM, artificial bee colony algorithm, and feedback signal vector, it can capture complex nonlinear features while achieving global optimization. Based on feedback closed loop, it can quickly correct the problem and predict the optimal value of the action space vector through deep Q-network (DQN), which is then used as the control output. By combining state space and reward function, it effectively solves the limitation of traditional regulation methods in dealing with high-dimensional complex regulation scenarios.

[0156] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling the biomass carbonization process based on artificial intelligence, characterized in that, include: Real-time scanning of biomass chemical components is performed, and the collected data is stored in a chemical component vector to obtain biomass particle size distribution and porosity information, and to generate a physical property vector. The decomposition rate of the components is calculated and the chemical composition is updated recursively. The reaction rate is analyzed based on the decomposition rate and the mass fraction of the chemical components, and the exothermic reaction generated by the decomposition reaction of each component is calculated. Data from the furnace temperature sensor array is collected, and the conduction heat flow is calculated using the finite difference algorithm. The furnace temperature distribution is analyzed and updated. Gas concentration data is collected to calculate the volatile matter generation rate, and the total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method. Based on the collected physical property vector data, the effective specific surface area is defined, the contribution of surface area to the reaction rate is analyzed, the particle size influence factor is defined, the solid carbon generation rate is analyzed in real time using a kinetic model, and a regression model is constructed using an artificial neural network (ANN) to output predicted carbon quality data. An SVM prediction model was constructed and optimized using the artificial bee colony algorithm. The deviation between the predicted gas concentration data and the updated gas concentration data was calculated, and the deviation between the predicted carbon mass data and the actual carbon mass data was combined to form a feedback signal vector. A deep Q-network is used to define the state space vector and action space vector, and a reward function is defined based on the feedback signal vector to predict the optimal control output.

2. The biomass carbonization process control method based on artificial intelligence as described in claim 1, characterized in that: The process involves analyzing and updating the temperature distribution within the furnace, collecting gas concentration data to calculate the volatile matter generation rate, and accumulating the contributions of all chemical components to obtain the total gas generation rate. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method. Based on the collected chemical component vector data, the decomposition rate is calculated using the Arrhenius formula, and the chemical components are updated using the mass fraction recursive formula. The reaction rate of the carbonization process is calculated based on the decomposition rate and the mass fraction of the chemical components, and the heat of chemical reaction generated by the decomposition reaction of each component is calculated through the enthalpy value. Based on the data collected from the temperature sensor array inside the carbonization furnace, the finite difference algorithm is used to calculate the conduction heat flow according to the gradient of the temperature sampling points. Combined with the heat released by the chemical reaction, the temperature distribution inside the furnace is analyzed and the temperature distribution value is updated. The gas concentration vector is composed of the gas concentrations of carbon monoxide, carbon dioxide, methane and hydrogen. The volatile matter generation rate is calculated based on the decomposition rate of different components. The total gas generation rate is obtained by accumulating the contributions of all chemical components. Based on the monitored gas flow rate and gas concentration vector, the gas concentration distribution within the control volume is updated using the finite volume method (FVM). Data is recorded based on updated temperature distribution data, gas concentration data, and gas generation rate data in the carbonization furnace.

3. The biomass carbonization process control method based on artificial intelligence as described in claim 2, characterized in that: The method involves using a kinetic model to analyze the solid carbon formation rate in real time, and using an artificial neural network (ANN) to construct a regression model to output predicted carbon quality data, including... Based on the collected physical property vector data, and considering the influence of particle specific surface area and porosity, the effective specific surface area is defined comprehensively. The particle size influence factor is defined based on the contribution of effective specific surface area to the reaction rate. The carbon fixation rate is determined based on the content of different components in the chemical composition vector data and the fixed proportion converted into solid carbon. Based on carbon fixation rate, particle size influencing factors, and temperature distribution data, a kinetic model was used to describe the chemical process of biomass carbon conversion in the furnace and to analyze the solid carbon generation rate. A regression model is constructed using an artificial neural network (ANN), including an input layer, hidden layers, and an output layer. The input layer takes in updated temperature distribution data, gas generation rate data, particle size influence factor and carbon fixation rate, the hidden layer uses ReLU activation function to capture nonlinear dynamic characteristics, and the output layer outputs predicted carbon mass data. Based on the calibrated training set data, the cross-entropy loss function is selected to calculate the computational loss of the model's predicted output. The Adam optimizer is used for gradient descent optimization to update the model parameters. The iteration stops when the change in computational loss is less than or equal to the historical loss threshold, and the model parameters are output to update the model.

4. The biomass carbonization process control method based on artificial intelligence as described in claim 3, characterized in that: The deviation between the predicted carbon mass data and the actual carbon mass data forms a feedback signal vector, including: The SVM prediction model is constructed using a support vector machine and a kernel function. The updated temperature distribution data, gas generation rate data, and flow rate data are input, and the concentration of each gas is predicted. The sum function of the SVM prediction model is selected as the radial basis function, the optimization objective is determined to be minimizing the model prediction error, and the computational loss of the model prediction output is calculated by the cross-entropy loss function. The artificial bee colony algorithm is used to iteratively optimize the kernel parameters and regularization parameters. The parameter solutions are randomly assigned and used as the population. The calculated loss is used as the fitness value. A global search is performed to output parameters based on the optimization objective and update the SVM prediction model. The deviation between the predicted gas concentration data and the updated gas concentration data is calculated, and the deviation between the predicted carbon mass data and the actual carbon mass data is combined to form a feedback signal vector, which includes gas concentration deviation and carbon mass deviation.

5. The biomass carbonization process control method based on artificial intelligence as described in claim 4, characterized in that: The step of defining a reward function based on the feedback signal vector and predicting the optimal control output includes... Closed-loop regulation is based on reinforcement learning algorithm, in which temperature distribution data, gas concentration data and gas generation rate data are defined to form a state space vector, and temperature adjustment, airflow rate adjustment and reaction time adjustment are used as action space vectors. The reward function is defined as the deviation minimization function. A deep Q-network (DQN) is used to predict the optimal value of each action space vector as the control output.

6. The biomass carbonization process control method based on artificial intelligence as described in claim 2, characterized in that: The process of acquiring biomass particle size distribution and porosity information and generating a physical property vector includes, High-precision particle size analyzers and porosity detectors are used to obtain biomass particle size distribution and porosity information, and physical property vectors are generated.

7. The biomass carbonization process control method based on artificial intelligence as described in claim 2, characterized in that: The real-time scanning of biomass chemical components and the storage of collected data in a chemical component vector include, The biomass chemical components were scanned in real time using a near-infrared spectroscopy analyzer (NIRS) to generate data on the mass content of cellulose, hemicellulose, and lignin. Each sample generated a value, which was stored in a chemical component vector.

8. A biomass carbonization process control system based on artificial intelligence, based on the biomass carbonization process control method based on artificial intelligence as described in any one of claims 1 to 7, characterized in that: include, Real-time acquisition module: Real-time scanning of the chemical composition and physical properties of biomass; Decomposition reaction module: Calculates the decomposition rate of chemical components and recursively updates the component mass parameters. It also analyzes the reaction rate and the exothermic reaction of the chemical reaction and supports dynamic heat flow analysis. Temperature and gas monitoring module: Based on the temperature and gas concentration data collected by the in-furnace sensors, the temperature distribution and gas concentration distribution are updated using the finite difference method and the finite volume method, respectively; Physical property analysis module: comprehensively defines the effective specific surface area and particle size influence factors, quantifies the contribution of physical properties to the reaction rate, and supports real-time kinetic model calculation of solid carbon formation rate; Artificial Neural Network Regression Module: Predicts carbon mass data in real time using an ANN regression model; Support Vector Machine Prediction Optimization Module: Based on SVM prediction of gas concentration data, the kernel parameters are optimized using the artificial bee colony algorithm, and a feedback signal vector is constructed to correct the bias of the model; Reinforcement learning control module: It uses a deep Q-network to define the state space and action space, and predicts the optimal control output based on the feedback signal vector and reward function.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based biomass carbonization process control method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the biomass carbonization process control method based on artificial intelligence as described in any one of claims 1 to 7.

Images