Air cooling tower temperature field and in-tower air flow field prediction method based on physical information neural network

Abstract

The invention relates to the field of thermal power generating unit cold end optimization, in particular to an air cooling tower temperature field and in-tower air flow field prediction method based on a physical information neural network. Comprising the steps that S1, environment temperature, environment wind speed and circulating water flow, in-tower temperature, in-tower wind speed and in-tower pressure data of an air cooling tower of the indirect air cooling unit are collected and preprocessed, and a data set is constructed; s2, based on the control equations of the heat conduction problem and the air flow problem, using initial conditions and boundary conditions to respectively establish loss functions, thereby constructing a physical information neural network; s3, training the physical information neural network of the air cooling tower of the indirect air cooling unit through a minimum loss function by using the training set obtained after preprocessing; s4, inputting the test set into the trained physical information neural network, and comparing the difference between a prediction result and a real situation; and S5, predicting a temperature field and an in-tower air flow field of the air cooling tower of the multi-working-condition indirect air cooling unit under different environmental conditions.

Description

Technical Field

[0001] The invention relates to the field of cold end optimization of thermal power units, and in particular to a method for predicting a temperature field of an air cooling tower and an air flow field in the tower based on a physical information neural network. Background Art

[0002] Thermal power generation companies have a major mission and must implement a series of energy-saving improvement strategies.

[0003] In the field of thermal power generation, indirect air cooling technology is widely used in the three northern regions (North China, Northwest China, and Northeast China) with rich coal and short of water due to its many advantages such as water saving, good antifreeze performance, and flexible and convenient operation. As the core heat dissipation equipment at the cold end of the indirect air-cooled unit, the heat exchange performance of the air-cooling tower directly affects the economic efficiency of the unit. At present, most of the indirect air-cooled coal-fired power units in service are involved in deep peak-shaving tasks, and the unit parameters change frequently. In this state, the prediction of the temperature field of the air-cooling tower and the air flow field in the tower of the indirect air-cooled unit helps to accurately understand the temperature in the tower and the air flow in the tower, avoid the impact of abnormal temperature and air vortex in local areas on the overall heat dissipation performance of the air-cooling tower, promptly discover unreliable factors in the tower, adjust the equipment, and ensure the safe, efficient and economical operation of the power plant. However, the prediction of the temperature field and air flow field in the air-cooling tower of the indirect air-cooled unit has long been a technical problem that needs to be solved urgently.

[0004] Due to the large volume of the air-cooling tower, for economic considerations, the number of flow sensors, temperature sensors, anemometers and differential pressure meters installed is generally limited. The internal conditions of the air-cooling tower are complex, and the existing measurement points cannot fully and intuitively reflect the overall heat exchange and air flow conditions of the air-cooling tower, resulting in the inability to identify and locate local high temperatures, overcooling and air vortices in the tower. At present, the presentation of the heat exchange and air flow conditions of the air-cooling tower mainly relies on the data-driven FLUENT model. This method has a large data demand and high data quality requirements, and it is difficult to perform real-time simulation according to the current working conditions. At the same time, it has a serious dependence on the quality of the grid. Unreasonable grid division or poor quality will cause the simulation results to be distorted and produce large errors. That is, at this stage, no matter what method is used, it is difficult to predict the temperature field of the air-cooling tower and the air flow field in the tower of the indirect air-cooling unit if only a small amount of limited data is used. Therefore, it is particularly important to explore a numerical method that can accurately predict the temperature field of the air-cooling tower and the air flow field in the tower with only a small number of measurement points.

[0005] As a neural network that incorporates physical information constraints during the training process, the physical information neural network can show strong generalization ability when there are fewer data samples compared to the traditional neural network. Therefore, the present invention proposes an innovative method, namely, a method for predicting the temperature field and air flow field in the air cooling tower of an indirect air cooling unit based on the physical information neural network. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention proposes a prediction method for the temperature field of the air-cooling tower and the air flow field in the tower of an indirect air-cooling unit based on a physical information neural network. The model is constructed by utilizing the measurement data of a flow sensor, a temperature sensor, an anemometer and a differential pressure meter, combining neural network technology, heat transfer control equations and air flow field control equations. By minimizing the loss function of the physical information neural network, the temperature field of the air-cooling tower and the air flow field in the tower of the indirect air-cooling unit can be quickly and accurately predicted.

[0007] The present invention adopts the following technical scheme: a method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network, comprising: S1: According to the flow sensors, internal and external temperature sensors, internal and external anemometers and differential pressure meters arranged on the air cooling tower of the indirect air cooling unit, the ambient temperature, ambient wind speed, circulating water flow rate of the air cooling tower of the indirect air cooling unit, the temperature inside the tower, the wind speed inside the tower and the pressure inside the tower are collected and preprocessed to construct a data set; S2: Based on the control equations of heat conduction and air flow problems, using initial conditions and boundary conditions, loss functions are established respectively, thereby constructing a physical information neural network; S3: Use the training set obtained after preprocessing to train the physical information neural network of the air cooling tower of the indirect air cooling unit by minimizing the loss function, and use the Adam optimization function to update the training parameters in the back propagation process; S4: Input the test set into the trained physical information neural network, compare the difference between the predicted results and the actual situation, evaluate the generalization ability of the physical information neural network prediction model, and evaluate the performance of the prediction model; S5: Based on the real-time data collected by the flow sensors, internal and external temperature sensors, internal and external anemometers and differential pressure meters arranged on the air cooling tower of the indirect air cooling unit, the temperature field and the air flow field in the air cooling tower of the indirect air cooling unit under different environmental conditions are predicted.

[0008] In some embodiments, step S1 includes: S11: According to the flow sensor, internal and external temperature sensors, internal and external anemometers and differential pressure gauge arranged on the air cooling tower of the indirect air cooling unit, the ambient temperature, ambient wind speed, the circulating water flow rate flowing into the air cooling tower and the temperature, wind speed and pressure measurement values ​​corresponding to the time and space coordinates of the measuring element in the tower are obtained at that moment; S12: preprocessing the data obtained in S11, including removing outliers, data cleaning and data normalization; S13: Arrange the preprocessed data into a measurement data set and divide it into a training set and a test set.

[0009] In some embodiments, in step S11, the measuring point of the flow sensor is arranged at the middle of the edge of each sector-shaped working area in the air cooling tower, and the number of measuring points is recommended to be arranged one for each sector-shaped working area; The measuring points of the external temperature sensor and the external anemometer are arranged in the east, west, south and north directions around the air cooling tower; one measuring point is arranged in each direction; The measuring points of the internal temperature sensor, internal anemometer and differential pressure meter are arranged in the middle and the middle of the edge of each sector-shaped working area in the air cooling tower. The number of measuring points is two temperature sensors, two anemometers and two differential pressure meters for one sector-shaped working area, with a total of four layers from top to bottom.

[0010] In some embodiments, in step S12, The ambient temperature and ambient wind speed are the average values ​​of the four sensors in the east, west, south and north. The circulating water flow is the average value of the eight sector sensors. Data normalization includes: The characteristic quantities are introduced to perform dimensionless processing on the physical quantities of ambient temperature, ambient wind speed, circulating water flow, tower temperature, tower wind speed and tower pressure. The dimensionless physical data are mapped to the interval [0,1] using the normalization method of minimum and maximum values. The formula is as follows: in, is the normalized physical quantity data; x It is the physical data after dimensionless processing; and are the maximum and minimum values ​​of the physical data set after dimensionless processing.

[0011] In some embodiments, the specific process of step S2 is as follows: S21: Build a fully connected neural network and set the number of layers, initial conditions and hyperparameters; S22: Determine the heat conduction equation, air flow field equation, initial conditions and boundary conditions of heat conduction, and initial conditions and boundary conditions of air flow field satisfied by the air cooling tower of the indirect air cooling unit; S23: The residual formed by the heat conduction equation, the initial conditions of heat conduction, and the boundary conditions of heat conduction are used as the loss function of the temperature field. The residual formed by the air flow field control equation, the initial conditions of the air flow field, and the boundary conditions of the air flow field are used as the loss function of the air flow field. Both are embedded in the physical information neural network as part of the loss function.

[0012] In some embodiments, in step S21, The fully connected neural network is represented as: in, is the input layer of the neural network; t is the time information; x,y,z Represents spatial coordinates; d is the circulating water flow rate; is the ambient temperature; is the ambient wind speed; The neural network i layer; and Represent the weight and bias of the network respectively; The neural network i -1st floor; is the activation function; H is the number of hidden layers of the neural network; is the output layer; and denote the output layer weight and bias respectively, For the H-1 layer; T Indicates temperature information; U express x Axis wind speed information; V for y Axis wind speed information; W for z Axis wind speed information; P For pressure information.

[0013] In some embodiments, in step S22, The heat conduction equation is: in, t For time, x, y, z represents the three-dimensional space coordinates of the sampling point, is the temperature corresponding to a certain time and space coordinate of the air cooling tower of the indirect air cooling unit, , and are the thermal conductivity, density and specific heat capacity of the material, respectively. Indicates the heat source of the air cooling tower of the indirect air cooling unit; The initial conditions for heat conduction are: To select the temperature at the start of sampling, is the heat source at the beginning of sampling; The boundary conditions for heat conduction are: Where: is the temperature on the specified boundary; The air flow field equation is: Where Re is the dimensionless Reynolds number, , is the air density, is the air viscosity, l is the reference length, v is the reference speed; u for x Axial wind speed; v for y Axial wind speed; w for z Axial wind speed; p For pressure; The initial conditions satisfied by the air flow field are: is the sampling start time x Axial wind speed; is the sampling start time y Axial wind speed; is the sampling start time z Axial wind speed, is the pressure at the beginning of sampling; The boundary conditions satisfied by the air flow field are: In the formula, for t The sampling time (x, y, z) coordinate position x Axial boundary wind speed; for t The sampling time (x, y, z) coordinate position y Axial boundary wind speed; for t The sampling time (x, y, z) coordinate position z Axial boundary wind speed; for t The boundary pressure at the sampling time (x, y, z) coordinate position; , , Measured by the anemometer at the boundary, Measured by a differential pressure gauge at the boundary.

[0014] In some embodiments, in step S23: The loss function of the temperature field is: in, As shown below: In the formula, represents the temperature field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; Respectively j The measured temperature and calculated temperature of each measuring point; and Respectively represent j The initial temperature and boundary temperature of each data point; is the residual of the diffusion equation for the heat conduction problem; The loss function of the air flow field is: in As shown below: In the formula, represents the air flow field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; 、 、 、 、 、 、 、 Respectively j of measuring points x Axial wind speed, y Axial wind speed, z Measured and predicted values ​​of axial wind speed and pressure; 、 、 、 、 、 、 、 Respectively represent j data points x Axial wind speed, y Axial wind speed, z Initial and boundary values ​​of axial wind speed and pressure; They are the four residuals of the Navier-Stokes equations; The total loss function is: Where L is the total loss function of the physical information neural network, is the weight of the temperature field loss function, is the weight of the air flow loss function in the tower.

[0015] In some embodiments, step S4 includes: S41: Based on the neural network trained in step S3, the corresponding data of the test set is input, and the difference between the predicted results and the actual measured data is compared. 2 Indicators evaluate the generalization ability and prediction performance of physical information neural network prediction models; If the performance of the neural network evaluation result is good, it is used to predict the temperature field of the air cooling tower and the air flow field in the tower; otherwise, return to step S3 to repeatedly train the physical information neural network until the indicators are met; S42: At this time, the physical information neural network only needs to use a small number of sensors and other measuring elements to complete the inversion prediction and real-time error correction of the temperature field and the air flow field in the tower, and adjust the layout of sensors and other measuring elements to ensure accuracy and efficiency while saving costs.

[0016] In some embodiments, step S5 includes: based on the physical information neural network prediction model obtained in steps S3 and S4, inputting the ambient temperature, ambient wind speed, circulating water flow rate, and real-time data of wind speed and pressure in all directions in the tower measured by the measuring device, using a three-dimensional vector synthesis method to obtain the vector sum of wind speeds in all directions, that is, the wind speed in the tower, and combining the wind pressure and temperature in the tower to accurately construct the temperature field of the air-cooling tower and the air flow field in the tower under different environmental conditions, so as to reflect the operation status inside the air-cooling tower in real time.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention is based on a physical information neural network, combines the measured data of a flow sensor, a temperature sensor, an anemometer and a differential pressure gauge, and the heat conduction and air flow equations of an air cooling tower of an indirect air cooling unit, and skillfully uses its physical characteristics to accurately invert the temperature field of an air cooling tower of an indirect air cooling unit under different environmental conditions and the distribution of the air flow field in the tower by using limited measuring point data through a physical information neural network. Compared with the traditional calculation method, the present invention significantly reduces the number of flow sensors, temperature sensors, anemometers and differential pressure gauges, and reduces the material cost. At the same time, even when there are fewer measuring points, the distribution of the temperature field of an air cooling tower of an indirect air cooling unit under different environmental conditions and the air flow field in the tower can be effectively revealed. When analyzing the problem, the present invention avoids the problems of high computational complexity and low efficiency of the traditional method. When the neural network makes a prediction, due to the incorporation of the physical information of heat conduction and air flow, the prior knowledge of the problem is fully utilized, so that the observation data is more fully utilized. This not only improves the operating efficiency of the neural network, but also enhances the accuracy of the calculation results, and effectively prevents the occurrence of overfitting. The present invention provides a new solution for the problem of predicting the temperature field of an air-cooling tower and the air flow field in an indirect air-cooling unit with multiple operating conditions under different environmental conditions, significantly improves the calculation efficiency, and enhances the accuracy of the inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is an overall flow chart of a method for predicting the temperature field of an air cooling tower and the air flow field in the tower of an indirect air cooling unit based on a physical information neural network of the present invention; Figure 2It is a physical information neural network framework diagram of the present invention; Figure 3 This is the air cooling tower sector distribution diagram; Figure 4 It is the sector layout diagram of air cooling tower temperature sensor, anemometer and differential pressure gauge; Figure 5 This is the longitudinal distribution diagram of the air cooling tower temperature sensor, anemometer and differential pressure gauge. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present invention more clear and explicit, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0020] like Figure 1 It is an overall flow chart of the method embodiment of the present invention, comprising the following steps: S1: According to the flow sensors, internal and external temperature sensors, internal and external anemometers and differential pressure meters arranged on the air-cooling tower of the indirect air-cooling unit, the ambient temperature, ambient wind speed, circulating water flow rate of the air-cooling tower of the indirect air-cooling unit, tower temperature, tower wind speed and tower pressure data are collected and preprocessed to construct a data set.

[0021] Specifically, step S1 includes: S11: Based on the flow sensor, internal and external temperature sensors, internal and external anemometers and differential pressure meter arranged on the air cooling tower of the indirect air cooling unit, the ambient temperature, ambient wind speed, circulating water flow rate flowing into the air cooling tower and the temperature, wind speed and pressure measurement values ​​corresponding to the time and space coordinates of the measuring elements in the tower are obtained.

[0022] Furthermore, Figure 3 This is the air cooling tower sector distribution diagram. Figure 4 This is the sector layout diagram of the air cooling tower temperature sensor, anemometer and differential pressure gauge. Figure 5 This is the longitudinal distribution diagram of the air cooling tower temperature sensor, anemometer and differential pressure gauge.

[0023] Furthermore, the measuring point of the flow sensor is arranged in the middle of the edge of each sector-shaped working area in the air-cooling tower. The recommended number of measuring points is one for one sector-shaped working area, and the arrangement height is 22 meters from the ground; the measuring points of the external temperature sensor and the external anemometer are arranged around the air-cooling tower in the east, west, south and north; the recommended number of measuring points is one in one direction, and the arrangement height is 10 meters from the ground; the measuring points of the internal temperature sensor, internal anemometer and differential pressure meter are arranged in the middle and middle of the edge of each sector-shaped working area in the air-cooling tower. The recommended number of measuring points is two temperature sensors, two anemometers and two differential pressure meters for one sector-shaped working area, with a total of four floors, with the first floor 10 meters from the ground, the second floor 50 meters from the ground, the third floor 100 meters from the ground, and the fourth floor 148 meters from the ground.

[0024] The fan-shaped working area is generally one eighth of the cross-sectional circle of the air-cooling tower, and the periphery is mainly composed of a number of triangular cooling columns and cement walls.

[0025] S12: Preprocess the data obtained in S11, including removing abnormal values, data cleaning and data normalization. The ambient temperature and ambient wind speed are the average of the four sensors in the east, west, south and north; the circulating water flow is the average of the eight sector sensors.

[0026] The specific method of data normalization described in S12 is as follows: introduce characteristic quantities to perform dimensionless processing on the physical quantities of ambient temperature, ambient wind speed, circulating water flow, temperature, wind speed and pressure, and use the normalization method of minimum and maximum values ​​to map the dimensionless processed physical data to the interval [0,1]. The formula is as follows: in, is the normalized physical quantity data; x is the dimensionless physical data; and are the maximum and minimum values ​​of the physical data set after dimensionless processing.

[0027] S13: Arrange the preprocessed data into a measurement data set, and divide it into a training set and a test set. In the present invention, the training set is 70% of the measurement data set, and the test set is 30% of the measurement data set.

[0028] S2: Based on the control equations of heat conduction and air flow problems, loss functions are established respectively using initial conditions and boundary conditions, thereby constructing a neural network embedded with physical prior loss functions.

[0029] Specifically, step S2 includes: S21: Construct a fully connected neural network and set the number of layers, initial conditions and hyperparameters; enable the fully connected neural network to accurately approximate complex nonlinear partial differential equation systems.

[0030] In order to adapt to the prediction of multiple working conditions under different environmental conditions, the present invention introduces ambient temperature, ambient wind speed and circulating water flow rate as additional inputs in the neural network to realize the prediction of the temperature field of the air-cooling tower and the air flow field in the tower under different environmental conditions.

[0031] The fully connected neural network is represented as follows: in, is the input layer of the neural network; t is the time information; x,y,z Represents spatial coordinates; d is the circulating water flow rate; is the ambient temperature; is the ambient wind speed; The neural network i layer; and Represent the weight and bias of the network respectively; The neural network i -1st floor; is the activation function; H is the number of hidden layers of the neural network; is the output layer; and denote the output layer weight and bias respectively, For the H-1 layer; T Indicates temperature information; U express x Axis wind speed information; V for y Axis wind speed information; W for z Axis wind speed information; P For pressure information.

[0032] The ReLU activation function is selected in the present invention.

[0033] Furthermore, the number of layers, initial conditions and initial values ​​of hyperparameters of the neural network are set according to actual conditions to initialize the model. The neural network of the present invention has 5 hidden layers, each with 8 neurons, and then randomly initialized.

[0034] S22: Determine the heat conduction equation, air flow field equation, initial conditions and boundary conditions of heat conduction, and initial conditions and boundary conditions of the air flow field satisfied by the air cooling tower of the indirect air cooling unit.

[0035] S23: The residual formed by the heat conduction equation, the initial conditions of heat conduction, and the boundary conditions of heat conduction are used as the loss function of the temperature field. The residual formed by the air flow field control equation, the initial conditions of the air flow field, and the boundary conditions of the air flow field are used as the loss function of the air flow field. Both are embedded in the physical information neural network as part of the loss function.

[0036] The heat conduction equation satisfied by the air cooling tower of the indirect air cooling unit is the diffusion equation of the heat conduction problem, as follows: in, t For time, x, y, z represents the three-dimensional space coordinates of the sampling point, is the temperature corresponding to a certain time and space coordinate of the air cooling tower of the indirect air cooling unit, , and are the thermal conductivity, density and specific heat capacity of the material, respectively. Indicates the heat source of the air cooling tower of the indirect air cooling unit.

[0037] The present invention , and The values ​​are 48W / (m・K), 7.85g / m3, and 480J / (kg・K) respectively.

[0038] Furthermore, the initial conditions satisfied by the temperature field of the air cooling tower of the indirect air cooling unit are: The initial condition is the temperature at the start of sampling. and heat source ,Right now .

[0039] Furthermore, the temperature boundary conditions satisfied by the temperature field of the air cooling tower of the indirect air cooling unit are: Where: is the temperature on the specified boundary, which can be measured by the temperature sensor at the boundary; The residual formed by the heat conduction equation, initial conditions, and boundary conditions is further embedded into the physical information neural network as part of the loss function. The loss function of the temperature field includes four parts: partial differential equation loss, initial condition loss, boundary condition loss, and the error between the internal temperature of the air cooling tower actually measured by the temperature sensor and the temperature of the corresponding measuring point obtained by the physical information neural network, that is, the numerical loss. The specific expression is as follows: in, As shown below: In the formula, represents the temperature field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; Respectively j The measured temperature and calculated temperature of each measuring point; and Respectively represent j The initial temperature and boundary temperature of each data point; is the residual of the diffusion equation for the heat conduction problem.

[0040] The air flow field equation satisfied by the air cooling tower of the indirect air cooling unit is the three-dimensional Navier Stokes equation, which is in the following form: Where Re is the dimensionless Reynolds number, , is the air density, is the air viscosity, l is the reference length, v is the reference speed; u for x Axial wind speed; v for y Axial wind speed; w for z Axial wind speed; p For pressure.

[0041] Furthermore, the initial conditions for the air flow field in the air cooling tower of the indirect air cooling unit are: The initial condition is selected at the start time of sampling. x Axial wind speed , y Axial wind speed , z Axial wind speed ,pressure ,Right now , , , .

[0042] Furthermore, the boundary conditions of the wind speed in the x-axis direction, the wind speed in the y-axis direction, the wind speed in the z-axis direction, and the pressure are considered as follows: In the formula, 、 、 Measured by anemometer at the boundary.

[0043] In the formula, Measured by a differential pressure gauge at the boundary.

[0044] The residual formed by the air flow field control equation, initial conditions, and boundary conditions is further embedded into the physical information neural network as part of the loss function. The loss function of the air flow field in the tower includes four parts: partial differential equation loss, initial condition loss, boundary condition loss, and; the error between the wind speed and pressure inside the air cooling tower actually measured by the anemometer and differential pressure gauge and the wind speed and pressure at the corresponding measuring point obtained by the physical information neural network, that is, the numerical loss. The specific expression is as follows: in As shown below: In the formula, represents the air flow field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; 、 、 、 、 、 、 、 Respectively j of measuring points x Axial wind speed, y Axial wind speed, z Measured and predicted values ​​of axial wind speed and pressure; 、 、 、 、 、 、 、 Respectively represent j data points x Axial wind speed, y Axial wind speed, z Initial and boundary values ​​of axial wind speed and pressure; are the four residuals of the Navier-Stokes equations.

[0045] The total loss function consists of two parts: the temperature field loss function and the air flow field loss function inside the tower.

[0046] Where L is the total loss function of the physical information neural network, is the weight of the temperature field loss function, is the weight of the air flow field loss function in the tower. and Select 0.5, which can be adjusted according to the actual situation and The value of .

[0047] S3: The physical information neural network of the air cooling tower of the indirect air cooling unit is trained by minimizing the loss function using the training set obtained after preprocessing. The training parameters are updated in the back propagation process using the Adam optimization function.

[0048] Specifically, step S3 includes: The Adam optimization function is used to update the training parameters in the back propagation process through the data set preprocessed in step S1. After multiple cycles, the network parameters are optimized. W , b , so that the total loss function of the network drops below the threshold or reaches the maximum round. Epoch >10000, threshold Pick .

[0049] S4: Input the test set into the trained physical information neural network, compare the differences between the predicted results and the actual situation, evaluate the generalization ability of the physical information neural network prediction model, and evaluate the performance of the prediction model.

[0050] Specifically, step S4 includes: S41: Based on the neural network trained in step S3, the corresponding data of the test set is input, and the difference between the predicted results and the actual measured data is compared. 2The generalization ability and prediction performance of the physical information neural network prediction model are evaluated by using indicators such as prediction accuracy and speed. If the performance of the neural network evaluation result is good, it can be used to predict the temperature field of the air cooling tower and the air flow field in the tower; otherwise, return to step S3 to repeatedly train the physical information neural network until the prediction accuracy, speed and other performance indicators are met.

[0051] S42: When the prediction of the temperature field and the air flow field in the tower by the physical information neural network meets the performance indicators such as accuracy and rapidity, the physical information neural network training is considered to be completed. At this time, the physical information neural network only needs to use a small number of sensors and other measuring elements to complete the inversion prediction of the temperature field and the air flow field in the tower and the real-time error correction, so the arrangement of sensors and other measuring elements should be adjusted to ensure accuracy and efficiency while saving costs.

[0052] Furthermore, the measurement point arrangement positions and quantities of the flow sensor, external temperature sensor, and external anemometer remain unchanged; the measurement points of the internal temperature sensor, internal anemometer, and differential pressure meter are arranged in four layers in the air cooling tower, the arrangement positions and quantities of the temperature sensors, anemometers, and differential pressure meters on the first layer remain unchanged, the second, third, and fourth layers are dismantled on the basis of the original sensor arrangement to reduce the arrangement quantity, all sensors in the even-numbered sectors are dismantled and only all sensors in the odd-numbered sectors are retained; the first layer is 10 meters from the ground, the second layer is 50 meters from the ground, the third layer is 100 meters from the ground, and the fourth layer is 148 meters from the ground.

[0053] The fan-shaped working area is generally one eighth of the cross-sectional circle of the air-cooling tower, and the periphery is mainly composed of a number of triangular cooling columns and cement walls.

[0054] S5. Through the trained high-fidelity physical information neural network, the temperature field of the air cooling tower and the air flow field inside the tower of the indirect air cooling unit are predicted.

[0055] Specifically, step S5 includes: Based on the high-precision physical information neural network prediction model obtained in steps S3 and S4, the real-time data of ambient temperature, ambient wind speed, circulating water flow, wind speed in all directions in the tower and pressure measured by a limited number of sensors and other measuring devices are input, and the vector sum of wind speed in all directions, that is, the wind speed in the tower, is obtained using a three-dimensional vector synthesis method. In combination with the wind pressure and temperature in the tower, the temperature field of the air-cooling tower under multiple operating conditions and the air flow field in the tower are accurately constructed under different environmental conditions, so as to reflect the operation status inside the air-cooling tower in real time.

[0056] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the present disclosure are still within the scope of the present disclosure.

Claims

1. A method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network, characterized in that: include: S1: According to the flow sensors, internal and external temperature sensors, internal and external anemometers and differential pressure meters arranged on the air cooling tower of the indirect air cooling unit, the ambient temperature, ambient wind speed, circulating water flow rate of the air cooling tower of the indirect air cooling unit, the temperature inside the tower, the wind speed inside the tower and the pressure inside the tower are collected and preprocessed to construct a data set; S2: Based on the control equations of heat conduction and air flow problems, using initial conditions and boundary conditions, loss functions are established respectively, thereby constructing a physical information neural network; S3: Use the training set obtained after preprocessing to train the physical information neural network of the air cooling tower of the indirect air cooling unit by minimizing the loss function, and use the Adam optimization function to update the training parameters in the back propagation process; S4: Input the test set into the trained physical information neural network, compare the difference between the predicted results and the actual situation, evaluate the generalization ability of the physical information neural network prediction model, and evaluate the performance of the prediction model; S5: Based on the real-time data collected by the flow sensors, internal and external temperature sensors, internal and external anemometers and differential pressure meters arranged on the air cooling tower of the indirect air cooling unit, the temperature field and the air flow field in the air cooling tower of the indirect air cooling unit under different environmental conditions are predicted.

2. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 1, characterized in that: The step S1 comprises: S11: According to the flow sensor, internal and external temperature sensors, internal and external anemometers and differential pressure gauge arranged on the air cooling tower of the indirect air cooling unit, the ambient temperature, ambient wind speed, the circulating water flow rate flowing into the air cooling tower and the temperature, wind speed and pressure measurement values ​​corresponding to the time and space coordinates of the measuring element in the tower are obtained at that moment; S12: preprocessing the data obtained in S11, including removing outliers, data cleaning and data normalization; S13: Arrange the preprocessed data into a measurement data set and divide it into a training set and a test set.

3. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 2, characterized in that: In step S11, the measuring point of the flow sensor is arranged at the middle of the edge of each sector-shaped working area in the air cooling tower, and the number of measuring points is recommended to be arranged one for each sector-shaped working area; The measuring points of the external temperature sensor and the external anemometer are arranged in the east, west, south and north directions around the air cooling tower; one measuring point is arranged in each direction; The measuring points of the internal temperature sensor, internal anemometer and differential pressure meter are arranged in the middle and the middle of the edge of each sector-shaped working area in the air cooling tower. The number of measuring points is two temperature sensors, two anemometers and two differential pressure meters for one sector-shaped working area, with a total of four layers from top to bottom.

4. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 3, characterized in that: In the step S12, The ambient temperature and ambient wind speed are the average values ​​of the four sensors in the east, west, south and north. The circulating water flow is the average value of the eight sector sensors. Data normalization includes: The characteristic quantities are introduced to perform dimensionless processing on the physical quantities of ambient temperature, ambient wind speed, circulating water flow, tower temperature, tower wind speed and tower pressure. The dimensionless physical data are mapped to the interval [0,1] using the normalization method of minimum and maximum values. The formula is as follows: in, is the normalized physical quantity data; x It is the physical data after dimensionless processing; and are the maximum and minimum values ​​of the physical data set after dimensionless processing.

5. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 1, characterized in that: The specific process of step S2 is as follows: S21: Build a fully connected neural network and set the number of layers, initial conditions and hyperparameters; S22: Determine the heat conduction equation, air flow field equation, initial conditions and boundary conditions of heat conduction, and initial conditions and boundary conditions of air flow field satisfied by the air cooling tower of the indirect air cooling unit; S23: The residual formed by the heat conduction equation, the initial conditions of heat conduction, and the boundary conditions of heat conduction are used as the loss function of the temperature field. The residual formed by the air flow field control equation, the initial conditions of the air flow field, and the boundary conditions of the air flow field are used as the loss function of the air flow field. Both are embedded in the physical information neural network as part of the loss function.

6. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 5, characterized in that: In the step S21, The fully connected neural network is represented as: in, is the input layer of the neural network; t is the time information; x,y,z Represents spatial coordinates; d is the circulating water flow rate; is the ambient temperature; is the ambient wind speed; The neural network i layer; and Represent the weight and bias of the network respectively; The neural network i -1st floor; is the activation function; H is the number of hidden layers of the neural network; is the output layer; and denote the output layer weight and bias respectively, For the H-1 layer; T Indicates temperature information; U express x Axis wind speed information; V for y Axis wind speed information; W for z Axis wind speed information; P For pressure information.

7. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 5, characterized in that: In the step S22, The heat conduction equation is: in, t For time, x, y, z represents the three-dimensional space coordinates of the sampling point, is the temperature corresponding to a certain time and space coordinate of the air cooling tower of the indirect air cooling unit, , and are the thermal conductivity, density and specific heat capacity of the material, respectively. Indicates the heat source of the air cooling tower of the indirect air cooling unit; The initial conditions for heat conduction are: To select the temperature at the start of sampling, is the heat source at the beginning of sampling; The boundary conditions for heat conduction are: Where: is the temperature on the specified boundary; The air flow field equation is: Where Re is the dimensionless Reynolds number, , is the air density, is the air viscosity, l is the reference length, v is the reference speed; u for x Axial wind speed; v for y Axial wind speed; w for z Axial wind speed; p For pressure; The initial conditions satisfied by the air flow field are: is the sampling start time x Axial wind speed; is the sampling start time y Axial wind speed; is the sampling start time z Axial wind speed, is the pressure at the beginning of sampling; The boundary conditions satisfied by the air flow field are: In the formula, for t The sampling time (x, y, z) coordinate position x Axial boundary wind speed; for t The sampling time (x, y, z) coordinate position y Axial boundary wind speed; for t The sampling time (x, y, z) coordinate position z Axial boundary wind speed; for t The boundary pressure at the sampling time (x, y, z) coordinate position, where , , Measured by the anemometer at the boundary, Measured by a differential pressure gauge at the boundary.

8. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 1, characterized in that: In step S23: The loss function of the temperature field is: in, As shown below: In the formula, represents the temperature field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; Respectively j The measured temperature and calculated temperature of each measuring point; and Respectively represent j The initial temperature and boundary temperature of each data point; is the residual of the diffusion equation for the heat conduction problem; The loss function of the air flow field is: in As shown below: In the formula, represents the air flow field loss function, is the loss of the partial differential equation; is the initial condition loss; is the boundary condition loss; is the residual between the measured temperature and the calculated temperature, i.e., the numerical loss; N is the number of measuring points; is the number of data points in the air cooling tower; and Respectively represent the number of initial condition training data points and the number of boundary condition training data points; 、 、 、 、 、 、 、 Respectively j of measuring points x Axial wind speed, y Axial wind speed, z Measured and predicted values ​​of axial wind speed and pressure; 、 、 、 、 、 、 、 Respectively represent j data points x Axial wind speed, y Axial wind speed, z Initial and boundary values ​​of axial wind speed and pressure; They are the four residuals of the Navier-Stokes equations; The total loss function is: Where L is the total loss function of the physical information neural network, is the weight of the temperature field loss function, is the weight of the air flow loss function in the tower.

9. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 1, characterized in that: The step S4 comprises: S41: Based on the neural network trained in step S3, the corresponding data of the test set is input, and the difference between the predicted results and the actual measured data is compared. 2 Indicators evaluate the generalization ability and prediction performance of physical information neural network prediction models; If the performance of the neural network evaluation result is good, it is used to predict the temperature field of the air cooling tower and the air flow field in the tower; otherwise, return to step S3 to repeatedly train the physical information neural network until the indicators are met; S42: At this time, the physical information neural network only needs to use a small number of sensors and other measuring elements to complete the inversion prediction and real-time error correction of the temperature field and the air flow field in the tower, and adjust the layout of sensors and other measuring elements to ensure accuracy and efficiency while saving costs.

10. The method for predicting the temperature field of an air cooling tower and the air flow field in the tower based on a physical information neural network according to claim 1, characterized in that: The step S5 includes: based on the physical information neural network prediction model obtained in steps S3 and S4, inputting the ambient temperature, ambient wind speed, circulating water flow rate, and real-time data of wind speed and pressure in all directions in the tower measured by the measuring device, using a three-dimensional vector synthesis method to obtain the vector sum of wind speeds in all directions, i.e., the wind speed in the tower, and combining the wind pressure and temperature in the tower to accurately construct the temperature field of the air-cooling tower under different environmental conditions and the air flow field in the tower, so as to reflect the operation status inside the air-cooling tower in real time.

Images