A modeling method for circulating fan of dry quenching system based on mechanism and data fusion

Abstract

The present application relates to a kind of mechanism and data fusion-based dry quenching system circulating fan modeling method, including the acquisition and pre-processing of data, circulating fan physical information neural network PINN model is established: physical information neural network PINN model is established, the ordinary differential equation group of circulating fan is embedded into neural network structure, and it is as part of loss function, model training and optimization: the parameter of neural network is adjusted by back propagation algorithm, loss function is minimized, in the training process, historical data is used to fit the model, to learn the dynamic characteristics and physical law of circulating fan system, model effect verification, model practical application;Realize the efficient solution of circulating fan system, solve the optimization problem that circulating fan system can encounter in the process of operation, improve the accuracy and robustness of control, provide an effective solution for the operation management of dry quenching circulating fan system, with wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and information technology, and in particular to a modeling method for circulating fans in dry quenching systems based on mechanism and data fusion. Background Technology

[0002] Ensuring the normal operation of circulating fans is crucial. The operation of dry quenching circulating fans can be effectively optimized by introducing advanced control systems, optimizing process parameters, establishing monitoring systems, improving operating strategies, and increasing equipment utilization. "Dong Hongfei. Research on Online Vibration Monitoring System for Circulating Fans in Coking Plants [D]. Northeastern University, 2016." An online detection system based on dry quenching circulating fans was designed to ensure the safe operation of the fans, enabling predictive maintenance and reducing economic losses caused by equipment accidents during dry quenching. "Wang Xiaojun. Predictive Maintenance of Circulating Fans in Dry Quenching [J]. Fuel and Chemical Industry, 2021, 52(04):26-28+31." The principle and process of the circulating fan of a 200t / h dry quenching unit were introduced. Typical fault phenomena were classified through spectrum analysis, achieving the purpose of predictive maintenance. "Li Huafeng, Wei Wei, Ding Zhen, et al. Relay Protection Solution for Main Motor of Circulating Fan in Dry Quenching [J]. Fuel and Chemical Industry, 2014, 45(04):27-29." The application of sampling value differential protection algorithm and magnetic balance differential protection technology for non-power frequency operation of motors was considered, providing an operation optimization scheme for the relay protection problem of the main motor of large dry quenching circulating fans.

[0003] The challenges in optimizing the operation of dry quenching circulating fans lie in: 1) the dry quenching circulating fan system has multiple operating conditions and is not stable; 2) the circulating fan consumes a large amount of energy during the dry quenching process, and optimizing energy consumption is a major challenge; 3) the circulating fan needs to be adjusted quickly and precisely according to the needs of the production process. Currently, there is a lack of an effective method that can systematically solve all of the above problems simultaneously. Summary of the Invention

[0004] This invention provides a modeling method for circulating fans in dry quenching systems based on mechanism and data fusion. This method solves the ordinary differential equations of the circulating fan system by utilizing Physical Information Neural Network (PINN) technology, combined with the physical mechanism and historical data of the circulating fan, to achieve efficient solution of the circulating fan system. This addresses optimization problems that may be encountered during the operation of the circulating fan system. Furthermore, by combining physical mechanism and data fusion, the accuracy and robustness of control are improved. This provides an effective solution for the operation and management of dry quenching circulating fan systems and has broad application prospects.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A modeling method for circulating fans in a dry quenching system based on mechanism and data fusion includes the following steps:

[0007] S1. Data Acquisition and Preprocessing: Collect and prepare historical data of the circulating fan system, including input parameters, state variables, and time series data;

[0008] S2. Establishment of the Physical Information Neural Network (PINN) Model for Circulating Fans: Establish the Physical Information Neural Network (PINN) model. In this model, the ordinary differential equations of the circulating fan are embedded into the neural network structure and used as part of the loss function.

[0009] S3. Model Training and Optimization: The parameters of the neural network are adjusted through the backpropagation algorithm to minimize the loss function. During the training process, historical data is used to fit the model in order to learn the dynamic characteristics and physical laws of the circulating fan system.

[0010] S4. Model Performance Validation: Use the validation dataset to evaluate the model's generalization ability and prediction accuracy, and adjust the model's hyperparameters based on the validation results, thereby validating and optimizing the trained PINN physical information neural network model.

[0011] S5. Practical application of the model: The trained physical information neural network PINN model is applied to the real-time data of the circulating fan system to solve and optimize the circulating fan system in real time, and to optimize and adjust the operating parameters of the circulating fan online in real time based on the current state.

[0012] Furthermore, in the PINN physical information neural network model of the circulating fan, the circulating fan is set to operate in axisymmetric flow. The following nonlinear Moore-Greitzer partial differential equations are established using the deviation from the stable axisymmetric flow and the flow-pressure relationship, i.e., the Moore-Greitzer model:

[0013]

[0014] Where Ψ is the total pressure coefficient, Φ is the circulating fan flow coefficient, and ψ c These are axisymmetric characteristic lines, where ξ is dimensionless time, and H and W are the half-height and half-width of the cubic axisymmetric characteristic line. This is the inverse matrix of the micro-nozzle characteristic curve, where a represents the inertial effect through the circulating fan, m represents the duct measurement parameters of the circulating fan, B represents the Greitzer parameters, and l... c It is the effective flow length of the gas passing through the circulating fan, ψ c0 Let J be the dimensionless pressure rise constant through the circulating fan when there is no flow, and let A be the amplitude of the flow disturbance potential.2 , ψ c Choose the following cubic equation:

[0015]

[0016] The intermediate equations involved in the Moore-Greitzer equations are:

[0017]

[0018] Where, ψ p γ is the pressure ratio coefficient within the cavity, R is the average radius of the circulating fan rotor, γ is the relative opening of the throttle valve, and A is the pressure ratio coefficient within the cavity. c V is the flow channel area of ​​the circulating fan. P The volume of the cavity is expressed in cubic meters (m³). 3 U a a is the average rotational speed of the circulating fan within its speed regulation range. s It is the speed of sound, in m / s; α is the airflow angle of the circulating fan; U is the tangential velocity at the average radius of the circulating fan rotor; ρ is the inlet atmospheric density; σ is the slip coefficient; l T The length of the outlet pipe of the throttle valve is a fixed parameter of the circulating fan. In formula (1), A is used. 2 Replace J,

[0019]

[0020] in,

[0021] Taking H=1 and W=1, and based on the relationship between rotational speed and pressure rise, the performance curve ψ of the circulating fan under variable speed is obtained. cn :

[0022]

[0023] Where k is a constant blocking coefficient, obtained from the law of conservation of energy:

[0024]

[0025] Therefore, we get:

[0026] Based on the theory of electric drive:

[0027] Where n is the rotational speed of the circulating fan, n0 is the initial rotational speed of the circulating fan, λ is the coefficient of inertia of the rotating shaft, and T L Tc is the output torque of the synchronous motor, and Tc is the rotational torque of the circulating fan.

[0028] The rotational torque of a circulating fan is equal to the change in the angular momentum of the airflow, as shown in the following formula:

[0029] T c =m c RC θ =σm c RU(10)

[0030] Among them, C θ The component of the absolute velocity at the impeller inlet in the tangential direction, after simplification:

[0031]

[0032] Among them, U T The dynamic equation for the circulating fan speed is obtained from the rotor tangential speed adjustment command:

[0033]

[0034] Where, n T This is a speed adjustment command;

[0035] Mass balance analysis was performed inside the cavity and at the outlet throttle valve: the mass flow rate entering the cavity is m c The mass flow rate after passing through the throttle valve at the outlet of the circulating fan is m t The changes in gas density and pressure within the cavity are caused by the change in the opening of the outlet throttle valve. The mass balance equation within the cavity is:

[0036]

[0037] Where, ρ P This refers to the gas density inside the cavity. When the gas inside the cavity is isentropic, the change in gas density is proportional to the change in pressure, as follows:

[0038]

[0039] Among them, P P This refers to the gas pressure inside the cavity, expressed in Pa.

[0040] The momentum conservation equation at the outlet throttling valve of the circulating fan is:

[0041]

[0042] Where, Φ T F is the dimensionless mass flow rate at the outlet of the throttle valve. T For the characteristic equation of the throttle valve, l T The length of the outlet pipe of the throttle valve is given, and pipe losses are ignored.

[0043]

[0044] Simultaneously, a linear process is used to describe the throttle valve regulation process:

[0045]

[0046] Among them, T γ For the corresponding parameters, γ T The dynamic equation for the circulating fan outlet pressure, based on the throttle valve adjustment command, is as follows:

[0047]

[0048] Using ρU a A c To achieve dimensionless mass flow rate of the circulating fan, utilizing... The dimensionless transformation of the fluid pressure difference in the circulating fan yields:

[0049]

[0050] Further refinement of the Moore-Greitzer model yields:

[0051]

[0052] To verify the function of the surge valve, the above formulas were integrated to obtain:

[0053]

[0054] Where γ is the relative opening of the throttle valve, and μ is the opening of the anti-surge valve.

[0055] Furthermore, the neural network structure in the PINN model of the physical information neural network includes an input layer, a hidden layer, and an output layer. The input layer has 2 nodes and the input parameters are γ and μ. The hidden layer is configured with multiple nodes. The output layer has 2 nodes and the corresponding output parameters are Ψ and Φ.

[0056] Furthermore, the loss function in step S2 includes a data fitting term and a physical constraint term. The data fitting term measures the degree to which the model fits the historical data, and the physical constraint term is used to ensure that the solution learned by the model satisfies the physical laws of the system.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1) By utilizing historical data of the circulating fan system and the physical mechanism of the circulating fan, a neural network model, namely PINN, was constructed that comprehensively considers physical laws and data characteristics. During the training process, not only the fitting of historical data was considered, but also the physical mechanism of the circulating fan system was integrated. This model can optimize the operation of the circulating fan system in real time, solve the optimization problems that the circulating fan system may encounter during operation, and improve the accuracy and robustness of the control.

[0059] 2) By embedding the set of ordinary differential equations of the circulating fan into the neural network structure and using it as part of the loss function, it is ensured that the solution learned by the neural network satisfies the physical laws of the system.

[0060] 3) The circulating fan status obtained by solving the PINN network is used to optimize and adjust the operating parameters of the circulating fan in real time online in combination with the current status, so as to improve the efficiency and performance of the system;

[0061] 4) It improves the understanding and control of the dynamic behavior of the circulating fan system, providing an effective solution for the operation and management of the dry quenching circulating fan system, and has broad application prospects. Attached Figure Description

[0062] Figure 1 This is a flowchart of the solution structure of the present invention.

[0063] Figure 2 This is a schematic diagram of the PINN (Physical Information Neural Network) solution network structure described in this invention.

[0064] Figure 3 This invention describes the maximum error in using PINN to calculate the flow rate and pressure rise of the circulating fan at different speeds. Detailed Implementation

[0065] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0066] See Figure 1 This is a flowchart illustrating the solution structure of the present invention. The present invention provides a modeling method for circulating fans in a dry quenching system based on mechanism and data fusion, comprising the following steps:

[0067] S1. Data Acquisition and Preprocessing: By collecting and preparing historical data of the circulating fan system, including input parameters, state variables, and time series data, the reliability and integrity of the data are ensured. This includes the following steps:

[0068] S1.1 Data Acquisition and Preprocessing: Historical data is read from the real-time database of the industrial site and subjected to noise reduction and anomaly processing.

[0069] (1) Noise Treatment:

[0070] The collected data is denoised using wavelet denoising, and the collected data is represented as follows:

[0071] d i =f i +εz i , i = 1, ..., N(22)

[0072] Where, d i For the collected data, f i The data is noise-free, ε represents the noise level, and z represents the noise level. i The noise is independent and identically distributed Gaussian white noise, N is the data volume, and i is the i-th data point. Wavelet decomposition is performed on the collected data:

[0073] W0d=W0f+εW0z(23)

[0074] Where d represents the collected data, f represents noise-free data, z represents Gaussian white noise, W0 represents the coefficients, and the wavelet coefficient function is:

[0075]

[0076] in,

[0077] Further, the denoised signal is obtained:

[0078]

[0079] Among them, using Replace t N w is the signal threshold. For wavelet coefficient functions;

[0080] (2) Exception handling

[0081] When missing data is found, the following actions are taken: if the missing data is less than 1 hour, the average of the data before and after the missing data is selected as the missing data segment; if the missing data is more than 1 hour, the data segment is deleted directly.

[0082] Data preparation: Collect historical data of the circulating fan system, including the input parameters, state variables and corresponding time series data of the circulating fan.

[0083] S2. Establishment of the PINN physical information neural network model for circulating fans: see... Figure 2 A physical information neural network (PINN) model was established. In this model, the ordinary differential equations of the circulating fan were embedded into the neural network structure and used as part of the loss function.

[0084] In the PINN physical information neural network model of the circulating fan, the circulating fan is set to operate in axisymmetric flow. The following nonlinear Moore-Greitzer partial differential equations are established using the deviation from the stable axisymmetric flow and the flow-pressure relationship, i.e., the Moore-Greitzer model:

[0085]

[0086] Where Ψ is the total pressure coefficient, Φ is the circulating fan flow coefficient, and ψ c These are axisymmetric characteristic lines, where ξ is dimensionless time, and H and W are the half-height and half-width of the cubic axisymmetric characteristic line. This is the inverse matrix of the micro-nozzle characteristic curve, where a represents the inertial effect through the circulating fan, m represents the duct measurement parameters of the circulating fan, B represents the Greitzer parameters, and l... c It is the effective flow length of the gas passing through the circulating fan, ψ c0 Let J be the dimensionless pressure rise constant through the circulating fan when there is no flow, and let A be the amplitude of the flow disturbance potential. 2 , ψ c Choose the following cubic equation:

[0087]

[0088] The intermediate equations involved in the Moore-Greitzer equations are:

[0089]

[0090] Where, ψ p γ is the pressure ratio coefficient within the cavity, R is the average radius of the circulating fan rotor, γ is the relative opening of the throttle valve, and A is the pressure ratio coefficient within the cavity. c V is the flow channel area of ​​the circulating fan. P The volume of the cavity is expressed in cubic meters (m³). 3 U a a is the average rotational speed of the circulating fan within its speed regulation range. s It is the speed of sound, in m / s; α is the airflow angle of the circulating fan; U is the tangential velocity at the average radius of the circulating fan rotor; ρ is the inlet atmospheric density; σ is the slip coefficient; l T The length of the outlet pipe of the throttle valve is a fixed parameter of the circulating fan. In formula (1), A is used. 2 Replace J,

[0091]

[0092] in,

[0093] Taking H=1 and W=1, and based on the relationship between rotational speed and pressure rise, the performance curve ψ of the circulating fan under variable speed is obtained. cn :

[0094]

[0095] Where k is a constant blocking coefficient, obtained from the law of conservation of energy:

[0096]

[0097] Therefore, we get:

[0098] Based on the theory of electric drive:

[0099] Where n is the rotational speed of the circulating fan, n0 is the initial rotational speed of the circulating fan, λ is the coefficient of inertia of the rotating shaft, and T L Tc is the output torque of the synchronous motor, and Tc is the rotational torque of the circulating fan.

[0100] The rotational torque of a circulating fan is equal to the change in the angular momentum of the airflow, as shown in the following formula:

[0101] T c =m c RC θ =σm c RU(35)

[0102] Among them, C θ The component of the absolute velocity at the impeller inlet in the tangential direction, after simplification:

[0103]

[0104] Among them, U T The dynamic equation for the circulating fan speed is obtained from the rotor tangential speed adjustment command:

[0105]

[0106] Where, n T This is a speed adjustment command;

[0107] Mass balance analysis was performed inside the cavity and at the outlet throttle valve: the mass flow rate entering the cavity is m c The mass flow rate after passing through the throttle valve at the outlet of the circulating fan is m t The changes in gas density and pressure within the cavity are caused by the change in the opening of the outlet throttle valve. The mass balance equation within the cavity is:

[0108]

[0109] Where, ρ PThis refers to the gas density inside the cavity. When the gas inside the cavity is isentropic, the change in gas density is proportional to the change in pressure, as follows:

[0110]

[0111] Among them, P P This refers to the gas pressure inside the cavity, expressed in Pa.

[0112] The momentum conservation equation at the outlet throttling valve of the circulating fan is:

[0113]

[0114] Where, Φ T F is the dimensionless mass flow rate at the outlet of the throttle valve. T For the characteristic equation of the throttle valve, l T The length of the outlet pipe of the throttle valve is given, and pipe losses are ignored.

[0115]

[0116] Simultaneously, a linear process is used to describe the throttle valve regulation process:

[0117]

[0118] Among them, T γ For the corresponding parameters, γ T The dynamic equation for the circulating fan outlet pressure, based on the throttle valve adjustment command, is as follows:

[0119]

[0120] Using ρU a A c To achieve dimensionless mass flow rate of the circulating fan, utilizing... The dimensionless transformation of the fluid pressure difference in the circulating fan yields:

[0121]

[0122] Further refinement of the Moore-Greitzer model yields:

[0123]

[0124] To verify the function of the surge valve, the above formulas were integrated to obtain:

[0125]

[0126] Where γ is the relative opening of the throttle valve, and μ is the opening of the anti-surge valve;

[0127] PINN neural network structure:

[0128] Input layer: The input layer has 2 nodes, corresponding to the input parameters γ and μ;

[0129] Hidden layer: Select a hidden layer with enough nodes to capture the complexity of the system. Considering the complexity of the system of equations and possible nonlinear relationships, select a deep neural network with multiple hidden layers and nodes. Select one with two hidden layers, each with 64 nodes.

[0130] Output layer: The output layer has 2 nodes, corresponding to the output parameters Ψ and Φ. Considering that the output is a continuous value, a linear activation function is chosen as the activation function of the output layer.

[0131] Introducing physical constraints: The set of ordinary differential equations of the circulating fan is embedded into the structure of the neural network as part of the loss function to ensure that the learned solution satisfies the physical laws of the system;

[0132] Define the loss function: The loss function consists of a data fit term and a physical constraint term. The data fit term measures how well the model fits the historical data, while the physical constraint term ensures that the solution learned by the model satisfies the physical laws of the system.

[0133] S3. Model Training and Optimization: The parameters of the neural network are adjusted using the backpropagation algorithm to minimize the loss function. During training, historical data is used to fit the model in order to learn the dynamic characteristics and physical laws of the circulating fan system. This specifically includes the following processes:

[0134] (1) Randomly initialize parameters: Randomly initialize the weights and biases of the neural network to break symmetry and accelerate convergence;

[0135] (2) Backpropagation algorithm: The gradient descent optimization algorithm is used to continuously adjust the network parameters through backpropagation in order to minimize the loss function;

[0136] (3) Data-driven learning: Using historical data for model training, the neural network learns the dynamic characteristics and physical laws of the circulating fan system.

[0137] S4. Model Performance Validation: Use the validation dataset to evaluate the model's generalization ability and prediction accuracy, and adjust the model's hyperparameters based on the validation results. Adjust the hyperparameters of the neural network, such as the learning rate and the number of hidden layer nodes, to improve the model's performance, thereby validating and optimizing the trained Physical Information Neural Network (PINN) model.

[0138] S5. Practical application of the model: The trained physical information neural network PINN model is applied to the real-time data of the circulating fan system to solve and optimize the circulating fan system in real time, and to optimize and adjust the operating parameters of the circulating fan online in real time based on the current state.

[0139] Table 1 presents a comparison of the solutions obtained using the PINN (Physical Information Neural Network) model and the differential solution with the actual values. Table 2 presents the solution results of the circulating fan model at different speeds. Figure 3 The root mean square error (RMSE) and mean absolute percentage error (MAPE) were selected as evaluation indicators.

[0140] Table 1 Results of solving the circulating fan model in this invention

[0141]

[0142] Table 2. Results of solving the circulating fan model at different speeds in this invention.

[0143]

[0144]

[0145] The analysis in Table 1 shows that, compared with the traditional differential solution method, the Physical Information Neural Network (PINN) model significantly improves the computational efficiency of solving the circulating fan model, reducing the time from 0.412 seconds to 0.004 seconds. It also improves the prediction accuracy of flow rate and pressure rise, with the maximum error reduced by 0.05 and 0.032, respectively. This indicates that the Physical Information Neural Network (PINN) model has high practicality and reliability in handling complex fluid dynamics problems.

[0146] As shown in Table 2, the predictive performance of the PINN (Physical Information Neural Network) model varies at different rotational speeds. However, it maintains low RMSE and MAPE values ​​at all speeds, indicating that the model has high accuracy and reliability under different working conditions. These results demonstrate that the PINN model of the present invention has significant advantages in computational efficiency and prediction accuracy, and is suitable for performance optimization in practical engineering applications.

[0147] The above embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the above embodiments. Unless otherwise specified, the methods used in the above embodiments are conventional methods.

Claims

1. A modeling method for circulating fans in a dry quenching system based on mechanism and data fusion, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect and prepare historical data of the circulating fan system, including input parameters, state variables, and time series data; S2. Establishment of the Physical Information Neural Network (PINN) Model for Circulating Fans: Establish the Physical Information Neural Network (PINN) model. In this model, the ordinary differential equations of the circulating fan are embedded into the neural network structure and used as part of the loss function. In the PINN physical information neural network model of the circulating fan, the circulating fan is set to operate in axisymmetric flow. The following nonlinear Moore-Greitzer partial differential equations are established using the deviation from the stable axisymmetric flow and the flow-pressure relationship, i.e., the Moore-Greitzer model: （1）； Where Ψ is the total pressure coefficient, It is the flow coefficient of the circulating fan. It is an axisymmetric characteristic line. Let H be the dimensionless time, and H and W be the half-height and half-width of the cubic axisymmetric characteristic line. is the inverse matrix of the micro-nozzle characteristic curve, where a represents the inertial effect through the circulating fan, m represents the duct measurement parameters of the circulating fan, and B represents the Greitzer parameters. It is the effective flow length through which the gas passes in the circulating fan. Let J be the dimensionless pressure rise constant through the circulating fan when there is no flow, and J be the amplitude of the flow disturbance potential. , Choose the following cubic equation: （2）； The intermediate equations involved in the Moore-Greitzer equations are: （3）； in, This is the pressure ratio coefficient within the cavity. The average radius of the circulating fan rotor. This refers to the relative opening of the throttle valve. The area of ​​the circulating fan channel. The volume of the cavity is expressed in units of... , This refers to the average rotational speed of the circulating fan within its speed regulation range. It is the speed of sound, measured in m / s. Let U be the airflow angle of the circulating fan, and U be the tangential velocity at the average radius of the circulating fan rotor. The density of the inlet atmosphere. This is the slip coefficient. The length of the outlet pipe of the throttle valve is a fixed parameter of the circulating fan. In formula (1), the length of the outlet pipe is used. Replace J, （4）； in, (5); With H=1 and W=1, the performance curve of the circulating fan under variable speed is obtained based on the relationship between rotational speed and pressure rise. : （6）； Where k is a constant blocking coefficient, obtained from the law of conservation of energy: （7）； Therefore, we get: (8); Based on the theory of electric drive: (9); in, The speed of the circulating fan. This is the initial speed of the circulating fan. T is the coefficient of inertia of the rotating axis. L Tc is the output torque of the synchronous motor, and Tc is the rotational torque of the circulating fan. The rotational torque of a circulating fan is equal to the change in the angular momentum of the airflow, as shown in the following formula: （10）； in, The component of the absolute velocity at the impeller inlet in the tangential direction, after simplification: （11）； in, The dynamic equation for the circulating fan speed is obtained from the rotor tangential speed adjustment command: （12）； in, This is a speed adjustment command; Mass balance analysis was performed inside the cavity and at the outlet throttle valve: the mass flow rate entering the cavity is m c The mass flow rate after passing through the throttle valve at the outlet of the circulating fan is The changes in gas density and pressure within the cavity are caused by the change in the opening of the outlet throttle valve. The mass balance equation within the cavity is: （13）； in, This refers to the gas density inside the cavity. When the gas inside the cavity is isentropic, the change in gas density is proportional to the change in pressure, as follows: （14）； in, This refers to the gas pressure inside the cavity, expressed in Pa. The momentum conservation equation at the outlet throttling valve of the circulating fan is: （15）； in, F is the dimensionless mass flow rate at the outlet of the throttle valve. T The characteristic equation of the throttle valve is... The length of the outlet pipe of the throttle valve is given, and pipe losses are ignored. （16）； Simultaneously, a linear process is used to describe the throttle valve regulation process: （17）； in, For the corresponding parameters, The dynamic equation for the circulating fan outlet pressure, based on the throttle valve adjustment command, is as follows: （18）； use To achieve dimensionless mass flow rate of the circulating fan, utilizing... The dimensionless transformation of the fluid pressure difference in the circulating fan yields: （19）； Further refinement of the Moore-Greitzer model yields: （20）； To verify the function of the surge valve, the above formulas were integrated to obtain: （21）； in, This refers to the relative opening of the throttle valve. To prevent surge valve opening; S3. Model Training and Optimization: The parameters of the neural network are adjusted through the backpropagation algorithm to minimize the loss function. During the training process, historical data is used to fit the model in order to learn the dynamic characteristics and physical laws of the circulating fan system. S4. Model Performance Validation: Use the validation dataset to evaluate the model's generalization ability and prediction accuracy, and adjust the model's hyperparameters based on the validation results, thereby validating and optimizing the trained PINN physical information neural network model. S5. Practical application of the model: The trained physical information neural network PINN model is applied to the real-time data of the circulating fan system to solve and optimize the circulating fan system in real time, and to optimize and adjust the operating parameters of the circulating fan online in real time based on the current state.

2. The method for modeling circulating fans in a dry quenching system based on mechanism and data fusion as described in claim 1, characterized in that, The PINN (Physical Information Neural Network) model comprises an input layer, a hidden layer, and an output layer. The input layer has two nodes and the input parameters are... and The hidden layer is configured with multiple layers, each containing multiple nodes. The output layer has two nodes, and the corresponding output parameters are as follows: and .

3. The method for modeling circulating fans in a dry quenching system based on mechanism and data fusion as described in claim 1, characterized in that, The loss function in step S2 includes a data fitting term and a physical constraint term. The data fitting term measures how well the model fits the historical data, and the physical constraint term is used to ensure that the solution learned by the model satisfies the physical laws of the system.

Images