Digital twin method and system for coal-biomass fluidized bed staged combustion

Abstract

This invention discloses a digital twin method and system for staged combustion in a coal-biomass fluidized bed boiler. The method establishes a comprehensive database including 3D simulation data and operational data from the coal-biomass fluidized bed boiler. A neural network is used to model and couple the gas and particulate phases separately to obtain a coal-biomass combustion model, which is used to predict the temperature, pressure, and velocity fields of the coal-biomass fluidized bed boiler. Finally, the coal-biomass combustion model is integrated with the boiler's PLC system to achieve data transmission, resulting in a digital twin model of coal-biomass combustion, enabling real-time control and dynamic optimization. This invention achieves online real-time prediction of boiler performance, and through continuous accumulation and dynamic training of operational data, it enables long-term model updates and continuous improvement in the level of intelligence.

Description

Technical Field

[0001] This invention relates to boiler combustion and digital twin technology, particularly a digital twin method and system for staged combustion in a coal-biomass fluidized bed. Background Technology

[0002] The co-combustion of coal and biomass can reduce dependence on fossil fuels and effectively utilize the characteristics of biomass to reduce emissions of greenhouse gases such as carbon dioxide (CO2). Unlike traditional staged combustion of a single fuel, staged combustion strategies in coal-biomass combustion need to be optimized for the different combustion characteristics of the two fuels. By dividing the combustion process into multiple stages and controlling the oxygen supply, temperature, and fuel distribution in each stage, staged combustion technology can better adapt to the different volatility and combustion rates of coal and biomass during combustion.

[0003] Traditional numerical simulation methods, such as CFD (Computational Fluid Dynamics) technology, can accurately describe the multiphase flow and chemical reactions in the staged combustion process of coal-biomass in fluidized bed boilers. However, these simulation methods are computationally intensive and struggle to meet the real-time response requirements of industrial sites, thereby reducing fuel utilization efficiency. They also easily lead to slagging and corrosion ash accumulation, and pollutants such as nitrogen oxides (NOx), sulfur dioxide (SO2), and particulate matter cannot be effectively emitted. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a digital twin method and system for staged combustion in coal-biomass fluidized bed boilers, which can realize online real-time prediction of coal-biomass fluidized bed boilers.

[0005] Technical solution: The digital twin method for staged combustion in a coal-biomass fluidized bed according to the present invention includes the following steps:

[0006] Define the target parameters that digital twins need to predict, and determine the input parameters and evaluation criteria;

[0007] A comprehensive database for staged combustion in coal-biomass fluidized beds was constructed using three-dimensional simulation data from computational fluid dynamics and operational data from coal-biomass fluidized bed boilers.

[0008] Based on the aforementioned comprehensive database, neural networks are used to model the gas phase and particulate phase respectively. By coupling the neural network models of the gas phase and particulate phase through the drag function and the heat transfer function between the gas and particulate phases, a coal-biomass combustion model is obtained, which is used to predict the temperature field, pressure field and velocity field of coal-biomass fluidized bed boilers.

[0009] The coal-biomass combustion model is integrated with the boiler PLC system of the coal-biomass fluidized bed to achieve data transmission, thereby obtaining a digital twin model of coal-biomass combustion.

[0010] Furthermore, the process of defining the target parameters that digital twins need to predict, determining the range of input parameters, and establishing evaluation criteria includes:

[0011] The target parameters that need to be predicted include boiler temperature field distribution, pressure field distribution, velocity field distribution, pollutant emission concentration, combustion efficiency, and thermal efficiency.

[0012] The input parameters include the structural parameters of the coal-biomass fluidized bed boiler, as well as the physical and chemical properties of the coal and biomass particles;

[0013] The evaluation criteria include the allowable deviations in combustion efficiency, pollutant emission concentration, and thermal efficiency.

[0014] Furthermore, in the comprehensive database, the 3D simulation data and real-time running data are standardized to remove outliers and perform time series alignment.

[0015] Furthermore, before integrating the coal-biomass combustion model with the coal-biomass fluidized bed boiler PLC system to achieve data transmission, the following steps are also included:

[0016] The sampled parameters are input into the coal-biomass combustion model to obtain the probability distribution of the prediction results. The relative contribution of the input parameters to the prediction results is calculated through sensitivity analysis. Based on the relative contribution, the parameters of the coal-biomass combustion model are tuned to obtain an optimized coal-biomass combustion model.

[0017] Furthermore, the probability distribution of the predicted results obtained by inputting the sampled parameters into the coal-biomass combustion model, and the calculation of the relative contribution of the input parameters to the predicted results through sensitivity analysis, includes:

[0018] Monte Carlo sampling is used to construct the uncertainty distribution of key input parameters, obtain the sampled data, and input the sampled parameters into the coal-biomass combustion model to obtain the probability distribution of the prediction results.

[0019] The Sobol method is used to perform sensitivity analysis on the input parameters and calculate the relative contribution of the input parameters to the prediction results.

[0020] Furthermore, the coupling of the neural network models of the gas and particulate phases via the drag function and the heat transfer function between the gas and particulate phases includes:

[0021]

[0022] S u =∑ p F d ;

[0023] S q =hA(T g -T p );

[0024] Among them, F d For drag force, C d A is the drag coefficient. p Let ∑ be the projected area of ​​the particle. p To sum over all particles, ρ g For gas density, u g Let u be the gas velocity. p S represents the particle velocity. u S is the momentum source term affected by the drag force. q The energy source term for heat transfer between the gas and solid phases is given by: h, the convective heat transfer coefficient, A, and T. g T represents the gas temperature. p The particle temperature;

[0025] Iterative calculation of traction force F d Then calculate S u , will S u In the input gas phase neural network model, S is iteratively calculated. q , will S q It is used to constrain the gas phase neural network model and the particle phase neural network model.

[0026] Furthermore, it also includes embedding the continuity equation, energy conservation equation, and momentum conservation equation into the neural network structure of the gas phase and particles for constraint, and optimizing the neural network model in the form of a loss function;

[0027] Loss function of gas phase neural network model as follows:

[0028]

[0029] In the formula, For continuous loss functions, Let be the momentum loss function. This is the energy loss function for the gas-phase neural network model. Here, w1, w2, w3, and w4 are weights, representing the loss function based on experimental data.

[0030] In the formula, ρ g For gas density, For gas in x i x j Predicted velocity components in direction, To predict gas pressure, μ t S is the turbulent viscosity. uC is the momentum source term affected by the drag force. p,g For the specific heat capacity of the gas, Let t be the predicted temperature of the gas, and t be time. Let be the velocity vector of the gas. For the temperature gradient of the gas, k g S is the thermal conductivity of the gas. q The energy source for heat transfer between the gas and solid phases is represented by N, where N is the total number of actual data points, and α1, α2, and α3 are weights.

[0031] Loss function of granular neural network model as follows:

[0032]

[0033] In the formula, Let be the energy loss function of the granular phase neural network model. The loss function is based on experimental data, with w5 and w6 as weights;

[0034]

[0035] In the formula, m p For particle mass, C p,p Let be the specific heat capacity of the particles. Let T be the predicted temperature of the particle, h be the convective heat transfer coefficient, A be the surface area of ​​the particle, σ be the Stefan-Boltzmann constant, ∈ be the emissivity of the particle, and T be the surface area of ​​the particle. s For ambient temperature, The rate of heat release in a chemical reaction. α4 and α5 are the position coordinates of the particle, and α4 and α5 are the weights.

[0036] The particle's position coordinates are updated using the following formula:

[0037]

[0038] In the formula, u p Here, k represents the particle velocity vector, m represents the current time, m* represents the intermediate calculation time, m+1 represents the next time step for position update, Δt is the iteration time step, and Δx is the velocity vector of the predicted particles. k a represents the particle position difference predicted by the particle phase neural network model. ext It includes the effects of gravity and drag.

[0039] The digital twin system for staged combustion of coal-biomass fluidized bed as described in this invention includes:

[0040] The parameter setting unit is used to specify the target parameters that the digital twin needs to predict, and to determine the input parameters and evaluation criteria.

[0041] The database establishment unit is used to construct a comprehensive database of staged combustion in coal-biomass fluidized beds by utilizing three-dimensional simulation data from computational fluid dynamics and operating data from coal-biomass fluidized bed boilers.

[0042] The coal-biomass combustion model building unit is used to model the gas phase and particulate phase respectively based on the comprehensive database using neural networks. By coupling the neural network models of the gas phase and particulate phase through the drag function and the heat transfer function between the gas and particulate phases, a coal-biomass combustion model is obtained, which is used to predict the temperature field, pressure field and velocity field of the coal-biomass fluidized bed boiler.

[0043] The coal-biomass combustion digital twin model establishment unit is used to integrate the coal-biomass combustion model with the boiler PLC system of the coal-biomass fluidized bed to achieve data transmission and obtain the coal-biomass combustion digital twin model.

[0044] The electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the digital twin method for staged combustion in a coal-biomass fluidized bed.

[0045] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the digital twin method for staged combustion in a coal-biomass fluidized bed.

[0046] Beneficial effects: Compared with the prior art, the advantages of the present invention are: (1) The present invention establishes a comprehensive database through multi-source data to ensure the accuracy and stability of the digital twin model under different operating conditions and adapt to the dynamic changes in boiler operating conditions; (2) The digital twin model of the present invention achieves long-term updates and continuous improvement of the intelligence level through continuous accumulation and dynamic training of operating data; (3) The present invention realizes online real-time prediction of boiler and combined performance testing of operating conditions, which facilitates operators to obtain timely feedback on boiler combustion status for optimization and subsequent energy saving and emission reduction schemes; (4) The present invention supports fault prediction. (5) This invention enables real-time and accurate prediction of speed, pressure, temperature and other data in coal-biomass fluidized bed boilers, dynamically feedback combustion status, and adjusts fuel flow and air supply according to load demand and emission targets to optimize combustion efficiency, avoid CO accumulation, reduce NOx, SO2 and particulate matter generation and reduce slagging and ash corrosion; (6) This invention adopts multi-scale multi-physics field full-chain verification to improve the accuracy and reliability of model prediction and meet the high-precision prediction requirements under complex working conditions. Attached Figure Description

[0047] Figure 1 This is a flowchart of the digital twin method for staged combustion in a coal-biomass fluidized bed according to the present invention;

[0048] Figure 2 This is a schematic diagram illustrating the physical mechanism and coupling principle of the coal-biomass combustion model in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram illustrating the establishment and application of a digital twin model for coal-biomass combustion in an embodiment of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0051] like Figure 1 As shown, the digital twin method for staged combustion in a coal-biomass fluidized bed includes the following steps.

[0052] (1) Identify the target parameters that need to be predicted and optimized for digital twins, and determine the range of input parameters and evaluation criteria.

[0053] The target parameters for prediction and optimization include boiler temperature field distribution, pressure field distribution, velocity field distribution, pollutant emission concentration, combustion efficiency, and thermal efficiency.

[0054] The input parameters include the structural dimensions of the coal-biomass fluidized bed boiler and the physical and chemical properties of the coal and biomass particles. The structural dimensions of the coal-biomass fluidized bed boiler include the boiler geometry, bed height, and heat exchange area. The physical and chemical properties of the coal and biomass particles include the density, particle size distribution, calorific value, particle shape factor, moisture content, elemental composition, volatile matter content, ash content, and fixed carbon content of the coal and biomass.

[0055] Establish evaluation criteria, with performance indicators including combustion efficiency, pollutant emission concentration, and thermal efficiency, and set an allowable deviation of ±15% for the target parameters.

[0056] Different operating condition parameters were designed, including fuel flow rate, fuel ratio, total air flow rate, primary air flow rate, secondary air flow rate, fluidization velocity, and environmental boundary conditions, including inlet temperature and pressure. Test schemes were also designed for different operating loads of 30%, 50%, and 100% to verify the system's stability and adaptability.

[0057] (2) Based on computational fluid dynamics (CFD) simulation and boiler operation data, a comprehensive database for staged combustion of coal-biomass fluidized bed was constructed.

[0058] A three-dimensional simulation model of the boiler is constructed using CFD simulation technology. The geometric model is set based on the boiler structural parameters, and the boundary conditions are set in combination with the characteristics of coal-biomass fuel and operating conditions. The simulation simulates the gas-solid two-phase flow, staged combustion reaction, heat transfer and pollutant generation process, and outputs the spatial coordinates of each calculation node and the corresponding temperature, pressure, velocity, density and component concentration.

[0059] Collect actual boiler operating data, and obtain actual boiler operating information through DCS and sensors, including boiler temperature, boiler pressure, fuel flow rate, fuel ratio, air supply temperature and pressure, air stage ratio and flue gas emission concentration.

[0060] By integrating simulation data with operational data, a comprehensive database is constructed. The data is standardized, outlier removal is performed, and time series alignment is carried out using the unified data storage format HDF5, thus completing the integration of multi-source data.

[0061] (3) Construct a coal-biomass combustion model using a neural network constrained by physical mechanisms.

[0062] like Figure 2 As shown, based on a comprehensive database, boiler structural parameters, fuel characteristics, and simulation results are extracted. A mechanistic-constrained neural network is used to model both the gas and particulate phases for multiphase flow characteristics. The differences in chemical composition, density, particle size, and volatile matter content between coal and biomass fuels, as well as the interactions between the gas and solid phases (including drag and heat transfer), are considered to couple the gas and particles, resulting in a coal-biomass combustion model. The inputs to this model are boiler structural parameters, fuel properties, initial conditions, and operating conditions. It can predict the distribution of boiler temperature, pressure, and velocity fields. Simultaneously, physical mechanisms are incorporated, with continuity equations, energy conservation equations, and momentum conservation equations embedded into the neural network structure for constraint. The coal-biomass combustion model is then optimized using a loss function.

[0063] Loss function of gas phase neural network model as follows:

[0064]

[0065] In the formula, For continuous loss functions, Let be the momentum loss function. This is the energy loss function for the gas-phase neural network model. Here, w1, w2, w3, and w4 are weights, representing the loss function based on experimental data.

[0066]

[0067] In the formula, ρ g For gas density, For gas in x i x j Predicted velocity components in direction, To predict gas pressure, μ t S is the turbulent viscosity. u C is the momentum source term affected by the drag force. p,g For the specific heat capacity of the gas, Let t be the predicted temperature of the gas, and t be time. Let be the velocity vector of the gas. For the temperature gradient of the gas, k g S is the thermal conductivity of the gas. q For heat transfer energy source terms between the gas phase and the solid phase, N is the total number of actual data points, and a1, a2, and a3 are weights.

[0068] Loss function of granular neural network model as follows:

[0069]

[0070] In the formula, Let be the energy loss function of the granular phase neural network model. The loss function is based on experimental data, and w5 and w6 are weights.

[0071]

[0072] In the formula, m p For particle mass, C p,p Let be the specific heat capacity of the particles. Let T be the predicted temperature of the particle, h be the convective heat transfer coefficient, A be the surface area of ​​the particle, σ be the Stefan-Boltzmann constant, ∈ be the emissivity of the particle, and T be the surface area of ​​the particle. s For ambient temperature, The rate of heat release in a chemical reaction. Let α4 and a5 be the position coordinates of the particle, and α4 and a5 be the weights.

[0073] The particle's position coordinates are updated using the following formula:

[0074]

[0075] In the formula, u p Here, k represents the particle velocity vector, m represents the current time, m* represents the intermediate calculation time, m+1 represents the next time step for position update, Δt is the iteration time step, and Δx is the velocity vector of the predicted particles. k a represents the particle position difference predicted by the particle phase neural network model. ext It includes the effects of gravity and drag.

[0076] The gas-solid two-phase neural network model is coupled by drag force and heat transfer:

[0077]

[0078] S u =∑ p F d ;

[0079] S q =hA(T g -T p );

[0080] Among them, F d For drag force, C d A is the drag coefficient. p Σ is the projected area of ​​the particle. p To sum over all particles, ρ g For gas density, u g Let u be the gas velocity. p S represents the particle velocity. u S is the momentum source term affected by the drag force. q The energy source term for heat transfer between the gas and solid phases is given by: h, the convective heat transfer coefficient, A, and T. g For gas temperature T p This represents the particle temperature.

[0081] Iterative calculation of traction force F d Then calculate the momentum source term S u The heat transfer source term S is iteratively calculated by inserting it into the momentum loss function of the gas-phase neural network model. q The energy loss functions of the gas-phase neural network and the particle-phase neural network are respectively used to constrain the model. Meanwhile, F... d a used for calculating particle position updates ext This process involves iteratively updating the particle position information.

[0082] (4) Model optimization and verification are carried out based on uncertainty optimization methods and consistency analysis.

[0083] The uncertainty distribution of input parameters is defined based on prior knowledge. Using a comprehensive database, Monte Carlo sampling is employed to input the samples into the coal-biomass combustion model, obtaining the probability distribution of the output results and initially quantifying the prediction uncertainty of the target parameters. Sensitivity analysis is performed using the Sobol method to calculate its relative contribution and interactive influence on the uncertainty of the target parameter prediction results, identifying the main sources and propagation paths affecting the output uncertainty. Based on the above uncertainty quantification and sensitivity analysis, uncertainty optimization methods are introduced to fine-tune the model's hyperparameters. A multi-scale validation process is constructed by integrating the coal-biomass combustion model, CFD simulation, experimental setup, and industrial boiler testing. During the validation process, uncertainty quantification methods are used for step-by-step validation and optimization, accumulating uncertainty to achieve full-chain validation. Cross-validation is then used for consistency analysis to evaluate the model's prediction accuracy and feasibility for practical industrial application, optimizing the model's adaptability.

[0084] (5) Under the framework of advanced process control (APC), the optimized coal-biomass combustion model is integrated into the boiler PLC system coupled with the existing DCS to establish a digital twin model of coal-biomass combustion and to perform real-time control and optimization of the boiler.

[0085] like Figure 3 As shown, the system connects to the boiler's DCS system via the OPC UA protocol to obtain real-time operating data, and connects to the PLC system via the Modbus protocol to realize the real-time execution of boiler control commands. A high-efficiency data transmission channel is built using Kafka to transmit the real-time collected data to a comprehensive SQL-based database for storage. A RESTful API is provided for the coal-biomass combustion digital twin model to call in real time to complete the generation of prediction and control commands.

[0086] The user interface (GUI) is developed based on the Qt framework, displaying real-time temperature, pressure, flow rate, pollutant emissions, combustion performance indicators, and operating condition inputs, supporting 2D / 3D visualization. A manual input module is provided for testing specific operating conditions or simulating optimization schemes. An anomaly alarm function is integrated, triggering alarm events based on real-time data monitoring and generating fault diagnosis reports through logical inference. WebSocket technology is used to achieve real-time data synchronization between the interface and the model, ensuring immediate feedback and visualization of control suggestions. Control commands are executed via a PLC, and historical data is stored for subsequent performance analysis and fault backtracking.

[0087] The digital twin system for staged combustion of coal-biomass fluidized bed as described in this invention includes:

[0088] The parameter setting unit is used to specify the target parameters that the digital twin needs to predict, and to determine the input parameters and evaluation criteria.

[0089] The database establishment unit is used to construct a comprehensive database of staged combustion in coal-biomass fluidized beds by utilizing three-dimensional simulation data from computational fluid dynamics and operating data from coal-biomass fluidized bed boilers.

[0090] The coal-biomass combustion model building unit is used to model the gas phase and particulate phase respectively based on the comprehensive database using neural networks. By coupling the neural network models of the gas phase and particulate phase through the drag function and the heat transfer function between the gas and particulate phases, a coal-biomass combustion model is obtained, which is used to predict the temperature field, pressure field and velocity field of the coal-biomass fluidized bed boiler.

[0091] The coal-biomass combustion digital twin model establishment unit is used to integrate the coal-biomass combustion model with the boiler PLC system of the coal-biomass fluidized bed to achieve data transmission and obtain the coal-biomass combustion digital twin model.

[0092] The electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the digital twin method for staged combustion in a coal-biomass fluidized bed.

[0093] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the digital twin method for staged combustion in a coal-biomass fluidized bed.

[0094] The computer-readable storage medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory or any other medium that can be used to store program code in the form of instructions or data structures and is accessible by a computer.

[0095] The processor is used to execute a computer program stored in memory to implement the various steps in the methods described in the above embodiments.

Claims

1. A digital twin method for staged combustion in a coal-biomass fluidized bed, characterized in that, Includes the following steps: Define the target parameters that digital twins need to predict, and determine the input parameters and evaluation criteria; A comprehensive database for staged combustion in coal-biomass fluidized beds was constructed using three-dimensional simulation data from computational fluid dynamics and operational data from coal-biomass fluidized bed boilers. Based on the comprehensive database, neural networks are used to model the gas phase and particulate phase respectively. By coupling the neural network models of the gas phase and particulate phase through the drag function and the heat transfer function between the gas and particulate phases, a coal-biomass combustion model is obtained, which is used to predict the temperature field, pressure field and velocity field of coal-biomass fluidized bed boiler. The coal-biomass combustion model is integrated with the DCS system of the coal-biomass fluidized bed boiler to obtain real-time operating data for the coal-biomass combustion model to call in real time, and to complete the generation of prediction and control instructions. The coal-biomass combustion model is integrated with the PLC system of the coal-biomass fluidized bed boiler to realize the data transmission and real-time execution of control instructions, thus obtaining a digital twin model of coal-biomass combustion. Before integrating the coal-biomass combustion model with the coal-biomass fluidized bed boiler PLC system to achieve data transmission, the following steps are also included: The sampled parameters are input into the coal-biomass combustion model to obtain the probability distribution of the prediction results. The relative contribution of the input parameters to the prediction results is calculated through sensitivity analysis. Based on the relative contribution, the parameters of the coal-biomass combustion model are tuned to obtain an optimized coal-biomass combustion model.

2. The digital twin method for staged combustion in a coal-biomass fluidized bed according to claim 1, characterized in that, The definition of the target parameters that digital twins need to predict, the determination of the input parameter range, and the evaluation criteria include: The target parameters that need to be predicted include boiler temperature field distribution, pressure field distribution, velocity field distribution, pollutant emission concentration, combustion efficiency, and thermal efficiency. The input parameters include the structural parameters of the coal-biomass fluidized bed boiler, as well as the physical and chemical properties of the coal and biomass particles; The evaluation criteria include the allowable deviations in combustion efficiency, pollutant emission concentration, and thermal efficiency.

3. The digital twin method for staged combustion in a coal-biomass fluidized bed according to claim 1, characterized in that, In the comprehensive database, the 3D simulation data and real-time running data are standardized, outliers are removed, and time series alignment is performed.

4. The digital twin method for staged combustion in a coal-biomass fluidized bed according to claim 1, characterized in that, The probability distribution of the predicted results obtained by inputting the sampled parameters into the coal-biomass combustion model, and the calculation of the relative contribution of the input parameters to the predicted results through sensitivity analysis, includes: Monte Carlo sampling is used to construct the uncertainty distribution of key input parameters, obtain the sampled data, and input the sampled parameters into the coal-biomass combustion model to obtain the probability distribution of the prediction results. The Sobol method is used to perform sensitivity analysis on the input parameters and calculate the relative contribution of the input parameters to the prediction results.

5. The digital twin method for staged combustion in a coal-biomass fluidized bed according to claim 1, characterized in that, The coupling of the neural network models of the gas and particulate phases through the drag function and the heat transfer function between the gas and particulate phases includes: S u =∑ p F d ; S q =hA(T g -T p ); Among them, F d For drag force, C d A is the drag coefficient. p Let ∑ be the projected area of ​​the particle. p To sum over all particles, ρ g For gas density, u g Let u be the gas velocity. p S represents the particle velocity. u S is the momentum source term affected by the drag force. q The energy source term for heat transfer between the gas and solid phases is given by: h, the convective heat transfer coefficient, A, and T. g T represents the gas temperature. p The particle temperature; Iterative calculation of traction force F d Then calculate S u , will S u In the input gas phase neural network model, S is iteratively calculated. q , will S q It is used to constrain the gas phase neural network model and the particle phase neural network model.

6. The digital twin method for staged combustion in a coal-biomass fluidized bed according to claim 1, characterized in that, It also includes embedding the continuity equation, energy conservation equation, and momentum conservation equation into the neural network structure of the gas phase and particles for constraint, and optimizing the neural network model in the form of a loss function; Loss function of gas phase neural network model as follows: In the formula, For continuous loss functions, Let be the momentum loss function. This is the energy loss function for the gas-phase neural network model. Here, w1, w2, w3, and w4 are weights, representing the loss function based on experimental data. In the formula, ρ g For gas density, For gas in x i x j Predicted velocity components in direction, To predict gas pressure, μ t S is the turbulent viscosity. u C is the momentum source term affected by the drag force. p，g For the specific heat capacity of the gas, Let t be the predicted temperature of the gas, and t be time. Let be the velocity vector of the gas. For the temperature gradient of the gas, k g S is the thermal conductivity of the gas. q The energy source for heat transfer between the gas and solid phases is represented by N, where N is the total number of actual data points, and α1, α2, and α3 are weights. Loss function of granular neural network model as follows: In the formula, Let be the energy loss function of the granular phase neural network model. The loss function is based on experimental data, with w5 and w6 as weights; In the formula, m p For particle mass, C p，p Let be the specific heat capacity of the particles. Let T be the predicted temperature of the particle, h be the convective heat transfer coefficient, A be the surface area of ​​the particle, σ be the Stefan-Boltzmann constant, ∈ be the emissivity of the particle, and T be the surface area of ​​the particle. s For ambient temperature, The rate of heat release in a chemical reaction. α4 and α5 are the position coordinates of the particle, and α4 and α5 are the weights. The particle's position coordinates are updated using the following formula: In the formula, u p Here, k represents the particle velocity vector, m represents the current time, m* represents the intermediate calculation time, m+1 represents the next time step for position update, Δt is the iteration time step, and Δx is the velocity vector of the predicted particles. k a represents the particle position difference predicted by the particle phase neural network model. ext It includes the effects of gravity and drag.

7. A digital twin system for staged combustion in a coal-biomass fluidized bed, characterized in that, include: The parameter setting unit is used to specify the target parameters that the digital twin needs to predict, and to determine the input parameters and evaluation criteria. The database establishment unit is used to construct a comprehensive database of staged combustion in coal-biomass fluidized beds by utilizing three-dimensional simulation data from computational fluid dynamics and operating data from coal-biomass fluidized bed boilers. The coal-biomass combustion model building unit is used to model the gas phase and particulate phase respectively using neural networks based on the comprehensive database. By coupling the neural network models of the gas phase and particulate phase through the drag function and the heat transfer function between the gas and particulate phases, a coal-biomass combustion model is obtained, which is used to predict the temperature field, pressure field and velocity field of the coal-biomass fluidized bed boiler. The coal-biomass combustion digital twin model establishment unit is used to integrate the coal-biomass combustion model with the boiler DCS system of the coal-biomass fluidized bed, acquire real-time operating data for the coal-biomass combustion model to call in real time, complete the generation of prediction and control instructions, integrate the coal-biomass combustion model with the boiler PLC system of the coal-biomass fluidized bed to realize the data transmission and real-time execution of control instructions, and obtain the coal-biomass combustion digital twin model; Before integrating the coal-biomass combustion model with the coal-biomass fluidized bed boiler PLC system to achieve data transmission, the following steps are also included: The sampled parameters are input into the coal-biomass combustion model to obtain the probability distribution of the prediction results. The relative contribution of the input parameters to the prediction results is calculated through sensitivity analysis. Based on the relative contribution, the parameters of the coal-biomass combustion model are tuned to obtain an optimized coal-biomass combustion model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the digital twin method for staged combustion of coal-biomass fluidized bed according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the digital twin method for staged combustion of coal-biomass fluidized bed according to any one of claims 1-6.

Images