A calculation method and system for the thermodynamic properties of working fluids based on RBF neural network

Abstract

The present invention discloses a calculation method and system for the thermodynamic properties of a working fluid based on an RBF neural network, comprising the following steps: obtaining sample data of the thermodynamic parameters of the working fluid in a gas turbine model; using the fuel-air ratio and the enthalpy value or entropy value of the working fluid as the model inputs, and the corresponding working fluid temperature as the model output, optimizing the initial RBF neural network model to obtain a target RBF neural network model; and using the target RBF neural network model to calculate the thermodynamic properties of the working fluid. The present invention introduces an RBF neural network model into the calculation of the thermodynamic properties of the working fluid during the simulation process of the thermodynamic system, directly inputs the fuel-air ratio and the enthalpy value or entropy value of the working fluid into the RBF neural network model, and the model directly outputs the working fluid temperature, without the need for the iterative calculation process of the thermodynamic properties of the working fluid in the existing thermodynamic system simulation, thereby improving the simulation efficiency of the thermodynamic system.

Description

Technical Field

[0001] The present invention relates to the field of gas turbine simulation, and particularly to a method and system for calculating the thermodynamic properties of a working fluid based on an RBF neural network. Background Art

[0002] The data of the thermodynamic properties of the working fluid is not only the original information for the simulation analysis of the thermodynamic system, but also an important parameter for characterizing the system performance and simulation results. Currently, the methods for obtaining the data of the thermodynamic properties of the working fluid generally include querying the gas thermodynamic property table, using the temperature fitting formula, or invoking a dedicated calculation program for the thermodynamic properties of the working fluid. On the one hand, the accuracy of the data of the thermodynamic properties of the working fluid has a great influence on the simulation results of the thermodynamic system; on the other hand, during the simulation analysis of the thermodynamic system, important parameters such as temperature and enthalpy value need to be calculated through repeated iterations, which seriously affects the speed of the simulation calculation. Usually, the calculation of the thermodynamic properties of the working fluid often accounts for more than 80% of the total time consumed by the entire system. Therefore, when performing simulation calculations on the thermodynamic system, it is very necessary to establish a fast and accurate calculation program for the thermodynamic properties of the working fluid.

[0003] In order to accelerate the simulation speed of the thermodynamic system, it is necessary to solve the iterative problem of inverse calculating the temperature based on the known enthalpy value or entropy value and the fuel-air ratio during the calculation of the thermodynamic properties of the working fluid in the gas turbine model. For this purpose, the present invention proposes a method and system for calculating the thermodynamic properties of the working fluid based on an RBF neural network. Summary of the Invention

[0004] The object of the present invention is to provide a method and system for calculating the thermodynamic properties of a working fluid based on an RBF neural network, which can use the RBF neural network to calculate the thermodynamic properties of the working fluid, eliminate the iterative calculation process in the simulation of thermodynamic systems such as gas turbines, and improve the efficiency of the thermodynamic system simulation while ensuring the calculation accuracy of the thermodynamic properties of the working fluid.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for calculating the thermodynamic properties of a working fluid based on an RBF neural network, the method comprising:

[0007] Obtaining sample data of the thermodynamic parameters of the working fluid in the gas turbine model; the sample data of the thermodynamic parameters of the working fluid includes the fuel-air ratio, the working fluid temperature, and the working fluid enthalpy value or the working fluid entropy value;

[0008] Setting the initial propagation speed of the initial RBF neural network model, and using the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value as the model inputs, and using the corresponding working fluid temperature as the model output, optimizing the initial RBF neural network model to obtain a target RBF neural network model;

[0009] Use the target RBF neural network model to calculate the thermodynamic properties of the working fluid.

[0010] Optionally, optimizing the initial RBF neural network model specifically includes:

[0011] Divide the sample data of the working fluid thermodynamic parameters into a training set, a test set, and a validation set according to a preset ratio;

[0012] Use the training set to train the initial RBF neural network model;

[0013] Use the test set to test the RBF neural network model that has passed the training;

[0014] Use the validation set to validate the RBF neural network model that has passed the test, and obtain a validated RBF neural network model, that is, the target RBF neural network model.

[0015] Optionally, using the training set to train the initial RBF neural network model specifically includes:

[0016] Use the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the training set as the model input, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model;

[0017] Calculate the first root mean square error of the RBF neural network model after the current iteration, and determine whether the first root mean square error is less than or equal to a first preset value to obtain a first judgment result;

[0018] If the first judgment result is yes, obtain the RBF neural network model that has passed the training;

[0019] If the first judgment result is no, make the currently iterated RBF neural network model the initial RBF neural network model, and return to the step "Use the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the training set as the model input, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model" until the first root mean square error is less than or equal to the first preset value.

[0020] Optionally, using the test set to test the RBF neural network model that has passed the training specifically includes:

[0021] Input the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the test set into the RBF neural network model that has passed the training, and output the test value of the working fluid temperature;

[0022] Calculate the second root mean square error of the RBF neural network model after the current test iteration according to the measured value of the working fluid temperature and the actual value of the working fluid temperature, and determine whether the second root mean square error is less than or equal to a second preset value to obtain a second judgment result;

[0023] If the second judgment result is yes, obtain the RBF neural network model that passes the test, and record the propagation speed of the RBF neural network model that passes the test;

[0024] If the second judgment result is no, add a preset value to the propagation speed of the current iterated RBF neural network model to obtain a test-optimized RBF neural network model, set the test-optimized RBF neural network model as the RBF neural network model that passes the training, and return to the step of "inputting the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the test set into the RBF neural network model that passes the training, and outputting the measured value of the working fluid temperature" until the second root mean square error is less than or equal to the second preset value.

[0025] Optionally, use the validation set to validate the RBF neural network model that passes the test, specifically including:

[0026] Input the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the validation set into the RBF neural network model that passes the test, and output the validation value of the working fluid temperature;

[0027] Calculate the third root mean square error of the RBF neural network model after the current validation iteration according to the validation value of the working fluid temperature and the actual value of the working fluid temperature, and determine whether the third root mean square error is less than or equal to the second preset value to obtain a third judgment result;

[0028] If the third judgment result is yes, obtain the RBF neural network model that passes the validation, that is, the target RBF neural network model;

[0029] If the third judgment result is no, set the propagation speed of the RBF neural network model that passes the test as the initial propagation speed, and return to the step of "training the initial RBF neural network model with the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the training set as the model input and the corresponding working fluid temperature as the model output" until the third root mean square error is less than or equal to the second preset value.

[0030] Optionally, the first preset value is greater than the second preset value.

[0031] The present invention also provides a working fluid thermodynamic property calculation system based on an RBF neural network, and the system includes:

[0032] A sample data acquisition module for acquiring sample data of the working medium thermodynamic parameters in a gas turbine model; the sample data of the working medium thermodynamic parameters includes the fuel-air ratio, the working medium temperature, and the working medium enthalpy value or the working medium entropy value;

[0033] A model optimization module for setting the initial propagation speed of the initial RBF neural network model, and using the fuel-air ratio and the working medium enthalpy value or the working medium entropy value as model inputs, and the corresponding working medium temperature as the model output to optimize the initial RBF neural network model to obtain a target RBF neural network model;

[0034] A calculation module for calculating the thermodynamic properties of the working medium by using the target RBF neural network model.

[0035] Optionally, the model optimization module includes a data set division sub-module, a training sub-module, a testing sub-module, and a verification sub-module;

[0036] The data set division sub-module is used to divide the sample data of the working medium thermodynamic parameters into a training set, a testing set, and a verification set according to a preset ratio;

[0037] The training sub-module is used to train the initial RBF neural network model by using the training set;

[0038] The testing sub-module is used to test the RBF neural network model that has passed the training by using the testing set;

[0039] The verification sub-module is used to verify the RBF neural network model that has passed the testing by using the verification set to obtain a verified RBF neural network model, that is, the target RBF neural network model.

[0040] Optionally, the training sub-module specifically includes:

[0041] A training unit for training the initial RBF neural network model by using the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as model inputs and the corresponding working medium temperature as the model output;

[0042] A first judgment unit for calculating the first root mean square error of the RBF neural network model after the current iteration and judging whether the first root mean square error is less than or equal to a first preset value to obtain a first judgment result;

[0043] If the first judgment result is yes, the trained RBF neural network model is obtained;

[0044] If the first judgment result is negative, set the currently iterated RBF neural network model as the initial RBF neural network model, and return to the step of "using the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the training set as the model input, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model" until the first root mean square error is less than or equal to the first preset value.

[0045] Optionally, the test sub-module specifically includes:

[0046] A test unit for inputting the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the test set into the trained RBF neural network model to output a test value of the working fluid temperature;

[0047] A second judgment unit for calculating the second root mean square error of the RBF neural network model after the current test iteration according to the test value of the working fluid temperature and the actual value of the working fluid temperature, and judging whether the second root mean square error is less than or equal to a second preset value to obtain a second judgment result;

[0048] If the second judgment result is positive, obtain the RBF neural network model that has passed the test, and record the propagation speed of the RBF neural network model that has passed the test;

[0049] If the second judgment result is negative, add a preset value to the propagation speed of the currently iterated RBF neural network model to obtain a test-optimized RBF neural network model, set the test-optimized RBF neural network model as the trained RBF neural network model, and return to the step of "inputting the fuel-air ratio and the enthalpy value or entropy value of the working fluid in the test set into the trained RBF neural network model to output a test value of the working fluid temperature" until the second root mean square error is less than or equal to the second preset value.

[0050] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0051] The present invention provides a method and system for calculating the thermodynamic properties of a working fluid based on an RBF neural network. By applying the RBF neural network in the calculation of the thermodynamic properties of the working fluid during the simulation calculation of the thermodynamic system, only the fuel-air ratio and the enthalpy value or entropy value of the working fluid need to be input into the RBF neural network, and the working fluid temperature is directly output by the neural network, without the need for the iterative calculation process of the thermodynamic properties of the working fluid in the existing thermodynamic system simulation calculation. This solves the problem of long calculation time due to the iterative calculation of the thermodynamic properties of the working fluid during the simulation calculation of the thermodynamic system, and improves the simulation efficiency of the thermodynamic system. Description of the Drawings

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0053] Figure 1 Flowchart of a method for calculating the thermodynamic properties of a working fluid based on an RBF neural network provided in Embodiment 1 of the present invention;

[0054] Figure 2 Schematic diagram of a method for calculating the thermodynamic properties of a working fluid based on an RBF neural network provided in Embodiment 1 of the present invention;

[0055] Figure 3 Accuracy of fitting enthalpy value - temperature based on an RBF neural network provided in Embodiment 1 of the present invention;

[0056] Figure 4 Accuracy of fitting entropy value - temperature based on an RBF neural network provided in Embodiment 1 of the present invention;

[0057] Figure 5 Graph of the relationship between the enthalpy value and temperature of the working fluid obtained by applying the iterative method and the neural network method provided in Embodiment 1 of the present invention;

[0058] Figure 6 Graph of the relationship between the entropy value and temperature of the working fluid obtained by applying the iterative method and the neural network method provided in Embodiment 1 of the present invention. Detailed implementation manners

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0060] The object of the present invention is to provide a method and system for calculating the thermodynamic properties of a working fluid based on an RBF neural network, which can use the RBF neural network to calculate the thermodynamic properties (temperature) of the working fluid, eliminate the iterative calculation process in the simulation of thermal systems such as gas turbines, and improve the efficiency of thermal system simulation while ensuring the calculation accuracy of the thermodynamic properties of the working fluid.

[0061] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0062] Embodiment 1

[0063] As Figure 1 and 2 shown, this embodiment provides a calculation method for the thermodynamic properties of the working fluid based on the RBF neural network. The method includes:

[0064] S1: Obtain the sample data of the thermodynamic parameters of the working fluid in the gas turbine model; the sample data of the thermodynamic parameters of the working fluid includes the fuel-air ratio, the temperature of the working fluid, and the enthalpy value or entropy value of the working fluid.

[0065] Considering that the temperature and fuel-air ratio data may be known in the existing database, but the enthalpy value (or entropy value) of the working fluid is unknown, or the enthalpy value (or entropy value) of the working fluid and the fuel-air ratio data are known, but the temperature data of the working fluid is unknown. Therefore, in order to make the sample data complete, the unknown parameter values need to be calculated. For example, methods such as looking up tables and temperature polynomials are used to write the calculation code for solving the enthalpy value and entropy value of the working fluid with known temperature and fuel-air ratio. Iterative algorithms such as the bisection method and the Newton downhill method are used to write the calculation code for iteratively solving the temperature of the working fluid with known enthalpy value / entropy value and fuel-air ratio. Using the written code, the required enthalpy value / entropy value and fuel-air ratio range are respectively input, and the temperature value is iteratively calculated to obtain the sample data required for training the neural network. Here, the iterative calculation can be understood as follows: in the calculation of the thermodynamic properties of the working fluid in the thermal system simulation, a series of formulas for calculating the enthalpy value from the fuel-air ratio and temperature are generally given. When calculating the temperature of the working fluid with known fuel-air ratio and enthalpy value, a temperature value needs to be initialized first, and it is brought into a series of formulas to compare the difference between the calculated enthalpy value and the actual value. According to whether the difference meets the requirements, the temperature value is adjusted. When the difference does not meet the conditions, the adjusted temperature value needs to be brought into a series of formulas again to obtain the calculated enthalpy value again, and the difference is calculated again to determine whether the difference meets the conditions. By continuously iterating, the temperature value when the difference meets the conditions is found, which is the required temperature value.

[0066] In this embodiment, diesel (C8H 16 ) can be used as the fuel, and the RBF neural network is used to fit its gas thermodynamic properties. The value range of the fuel-air ratio is 0:0.01:0.05, the value range of the enthalpy value is 300:5:2200, and the value range of the entropy value is 6.5:0.005:8.8. In the form of A:B:C, A and C are the endpoints of the numerical range, and B represents the step size of the value.

[0067] S2: Set the initial propagation speed of the initial RBF neural network model, and use the fuel-air ratio and the enthalpy value or entropy value of the working fluid as the model input, and the corresponding temperature of the working fluid as the model output to optimize the initial RBF neural network model to obtain the target RBF neural network model.

[0068] As Figure 3 and Figure 4, respectively show the accuracy of the RBF neural network fitting enthalpy - temperature and entropy - temperature when optimizing the RBF neural network using the training set, test set, and validation set. It can be intuitively seen that the fitting accuracy of the neural network is high.

[0069] Among them, in step S2, optimizing the initial RBF neural network model specifically includes:

[0070] S21: Divide the working fluid thermodynamic parameter sample data into a training set, a test set, and a validation set according to a preset ratio.

[0071] For the fitting process of solving temperature from known enthalpy and fuel - air ratio: Take the enthalpy and fuel - air ratio in the sample data as input samples, and the corresponding temperature value as the output sample. Randomly divide the sample data into a training set, a test set, and a validation set, with their proportions being 70%, 15%, and 15% respectively.

[0072] For the fitting process of solving temperature from known entropy and fuel - air ratio: Take the entropy and fuel - air ratio in the sample data as input samples, and the corresponding temperature value as the output sample. Randomly divide the sample data into a training set, a test set, and a validation set, with their proportions being 70%, 15%, and 15% respectively; Use the training set data to create and train an RBF neural network.

[0073] S22: Use the training set to train the initial RBF neural network model.

[0074] For the fitting process of solving temperature from known enthalpy and fuel - air ratio: Use the training set data, set the initial propagation speed to 1, create and train an RBF neural network; Use the root - mean - square error as the evaluation function to detect the performance of the neural network. If the root - mean - square error is greater than 0.5, retrain the RBF neural network until the set conditions are met.

[0075] For the fitting process of solving temperature from known entropy and fuel - air ratio: Use the training set data, set the initial propagation speed to 1, create and train an RBF neural network; Use the root - mean - square error as the evaluation function to detect the performance of the neural network. If the root - mean - square error is greater than 0.1, retrain the RBF neural network until the set conditions are met.

[0076] Specifically, step S22 (using the training set to train the initial RBF neural network model) specifically includes:

[0077] S221: Use the fuel - air ratio and the working fluid enthalpy value or the working fluid entropy value in the training set as the model input, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model.

[0078] S222: Calculate the first root mean square error of the RBF neural network model after the current iteration, and determine whether the first root mean square error is less than or equal to the first preset value to obtain a first judgment result.

[0079] The calculation of the first root mean square error is based on the output working medium temperature of the RBF neural network model after the current iteration and the corresponding working medium temperature (training data) in the training set.

[0080] If the first judgment result is yes, obtain the RBF neural network model that has passed the training.

[0081] If the first judgment result is no, set the RBF neural network model after the current iteration as the initial RBF neural network model, and return to step S221 until the first root mean square error is less than or equal to the first preset value ε1.

[0082] S23: Use the test set to test the RBF neural network model that has passed the training.

[0083] For the fitting process of solving the temperature with known enthalpy value and fuel-air ratio: Use the test set data to test the neural network fitting result. Also use the root mean square error as the evaluation function to detect the performance of the neural network. If the root mean square error is greater than 0.2, then iteratively optimize the propagation speed of the RBF neural network to make the root mean square error of the neural network model meet its set conditions.

[0084] For the fitting process of solving the temperature with known entropy value and fuel-air ratio: Use the test set data to test the neural network fitting result. Also use the root mean square error as the evaluation function to detect the performance of the neural network. If the root mean square error is greater than 0.02, then iteratively optimize the propagation speed of the RBF neural network to make the root mean square error of the neural network model meet its set conditions.

[0085] Among them, step S23 (using the test set to test the RBF neural network model that has passed the training) specifically includes:

[0086] S231: Input the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the test set into the RBF neural network model that has passed the training, and output the test value of the working medium temperature.

[0087] S232: Calculate the second root mean square error of the RBF neural network model after the current test iteration according to the test value of the working medium temperature and the actual value of the working medium temperature, and determine whether the second root mean square error is less than or equal to the second preset value to obtain a second judgment result.

[0088] If the second judgment result is yes, obtain the RBF neural network model that has passed the test, and record the propagation speed of the RBF neural network model that has passed the test.

[0089] If the second judgment result is negative, then increase the propagation speed of the currently iterated RBF neural network model by a preset value (here the preset value is set according to requirements, for example, it can be set to 1, corresponding to Figure 2 in which the optimized propagation speed S = S + 1), to obtain a tested and optimized RBF neural network model, make the tested and optimized RBF neural network model be the RBF neural network model that has passed the training, and return to step S231 until the second root mean square error is less than or equal to the second preset value ε2.

[0090] To ensure the accuracy of model training, it can be required that the first preset value is greater than the second preset value.

[0091] S24: Use the validation set to verify the RBF neural network model that has passed the test, to obtain a verified RBF neural network model, that is, the target RBF neural network model.

[0092] Use the validation set data to verify the fitting result accuracy of the neural network that has passed the test. If the root mean square error does not meet the set requirements (the enthalpy value is 0.2 and the entropy value is 0.02), then use the propagation speed obtained by testing and optimizing as the initial propagation speed, and repeat steps S22, S23, and S24 until the root mean square error meets the requirements.

[0093] Among them, step S24 (using the validation set to verify the RBF neural network model that has passed the test, to obtain a verified RBF neural network model, that is, the target RBF neural network model) specifically includes:

[0094] S241: Input the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the validation set into the RBF neural network model that has passed the test, and output the verified value of the working medium temperature.

[0095] S242: Calculate the third root mean square error of the RBF neural network model after the current verification iteration according to the verified value of the working medium temperature and the actual value of the working medium temperature, and judge whether the third root mean square error is less than or equal to the second preset value, to obtain a third judgment result.

[0096] If the third judgment result is positive, then obtain the verified RBF neural network model, that is, the target RBF neural network model.

[0097] If the third judgment result is negative, then make the propagation speed of the RBF neural network model that has passed the test (that is, Figure 2 the optimal propagation speed S in opt [[ID=३२]]) be the initial propagation speed, and return to step S22 until the third root mean square error is less than or equal to the second preset value.

[0098] S3: Calculate the thermodynamic properties of the working fluid using the target RBF neural network model.

[0099] Such as Figure 5 and Figure 6 , which shows the relationship diagrams of the enthalpy - temperature and entropy - temperature of the working fluid obtained by using the iterative method (a commonly used iterative method for calculating the thermodynamic properties of the working fluid in a gas turbine model) and the neural network method provided in this embodiment. It can be intuitively seen that the calculation accuracy of the neural network method for calculating the thermodynamic properties of the working fluid in this embodiment is high.

[0100] In this embodiment, based on the existing database, complete sample data (where the enthalpy value or entropy value, fuel - air ratio, and the corresponding temperature value are all known) is obtained through iterative calculation. The RBF neural network is trained using the complete sample data, and the thermodynamic properties (temperature) of the working fluid are calculated using the trained neural network model. The input data is directly input into the RBF neural network, and the temperature data is directly output by the model, eliminating the iterative calculation process in the calculation of the thermodynamic properties of the working fluid in the gas turbine model, and improving the efficiency of the thermal system simulation analysis. In addition, the thermodynamic properties of the working fluid are calculated by fitting based on the RBF neural network, and the RBF neural network is optimized using the test set data, with high calculation accuracy. And the method of this embodiment is applicable to a wide variety of fuel types and temperature ranges, with strong practicability.

[0101] Embodiment 2

[0102] This embodiment provides a system for calculating the thermodynamic properties of a working fluid based on an RBF neural network. The system includes:

[0103] A sample data acquisition module T1, configured to acquire the sample data of the working fluid thermodynamic parameters in the gas turbine model; the sample data of the working fluid thermodynamic parameters includes the fuel - air ratio, the working fluid temperature, and the enthalpy value or entropy value of the working fluid.

[0104] A model optimization module T2, configured to set the initial propagation speed of the initial RBF neural network model, and use the fuel - air ratio and the enthalpy value or entropy value of the working fluid as the model input, and the corresponding working fluid temperature as the model output to optimize the initial RBF neural network model to obtain the target RBF neural network model.

[0105] The model optimization module T2 includes a data set division sub - module, a training sub - module, a testing sub - module, and a verification sub - module.

[0106] The data set division sub - module is configured to divide the sample data of the working fluid thermodynamic parameters into a training set, a testing set, and a verification set according to a preset ratio.

[0107] The training sub - module is configured to train the initial RBF neural network model using the training set.

[0108] The training sub-module specifically includes:

[0109] A training unit, configured to use the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the training set as model inputs, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model;

[0110] A first judgment unit, configured to calculate the first root mean square error of the RBF neural network model after the current iteration, and judge whether the first root mean square error is less than or equal to a first preset value to obtain a first judgment result;

[0111] If the first judgment result is yes, the trained RBF neural network model is obtained;

[0112] If the first judgment result is no, the RBF neural network model after the current iteration is set as the initial RBF neural network model, and the step "using the fuel-air ratio and the enthalpy value or the entropy value of the working fluid in the training set as model inputs, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model" is returned until the first root mean square error is less than or equal to the first preset value.

[0113] The testing sub-module is configured to test the trained RBF neural network model by using the test set.

[0114] The testing sub-module specifically includes:

[0115] A testing unit, configured to input the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the test set into the trained RBF neural network model to output a working fluid temperature test value;

[0116] A second judgment unit, configured to calculate the second root mean square error of the RBF neural network model after the current test iteration according to the working fluid temperature test value and the actual working fluid temperature, and judge whether the second root mean square error is less than or equal to a second preset value to obtain a second judgment result;

[0117] If the second judgment result is yes, the tested RBF neural network model is obtained, and the propagation speed of the tested RBF neural network model is recorded;

[0118] If the second judgment result is negative, increase the propagation speed of the currently iterated RBF neural network model by a preset value to obtain a tested and optimized RBF neural network model. Set the tested and optimized RBF neural network model as the RBF neural network model that has passed the training, and return to the step of "inputting the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the test set into the RBF neural network model that has passed the training, and outputting the working medium temperature test value" until the second root mean square error is less than or equal to the second preset value.

[0119] The verification sub-module is used to verify the RBF neural network model that has passed the test by using the verification set to obtain a verified RBF neural network model, that is, the target RBF neural network model.

[0120] Among them, the model optimization module further includes a model propagation speed setting sub-module for setting the initial propagation speed of the initial RBF neural network model.

[0121] The verification sub-module specifically includes:

[0122] A verification unit for inputting the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the verification set into the RBF neural network model that has passed the test, and outputting the working medium temperature verification value;

[0123] A third judgment unit for calculating the third root mean square error of the RBF neural network model after the current verification iteration according to the working medium temperature verification value and the actual working medium temperature, and judging whether the third root mean square error is less than or equal to the second preset value to obtain a third judgment result.

[0124] If the third judgment result is positive, obtain the verified RBF neural network model;

[0125] If the third judgment result is negative, set the propagation speed of the RBF neural network model that has passed the test as the initial propagation speed, and return to the step of "training the initial RBF neural network model with the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as the model input and the corresponding working medium temperature as the model output" until the third root mean square error is less than or equal to the second preset value.

[0126] The calculation module T3 is used to calculate the thermodynamic properties of the working medium by using the target RBF neural network model.

[0127] Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0128] In this article, specific examples are used to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. To sum up, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A calculation method for the thermodynamic properties of working fluids based on RBF neural networks, characterized in that The method includes: Obtaining sample data of the working medium thermodynamic parameters in a gas turbine model; the sample data of the working medium thermodynamic parameters includes the fuel-air ratio, the working medium temperature, and the working medium enthalpy value or the working medium entropy value; Setting the initial propagation speed of the initial RBF neural network model, using the fuel-air ratio and the working medium enthalpy value or the working medium entropy value as the model inputs, and using the corresponding working medium temperature as the model output, optimizing the initial RBF neural network model to obtain a target RBF neural network model; Calculating the thermodynamic properties of the working medium using the target RBF neural network model; Among them, optimizing the initial RBF neural network model specifically includes: Dividing the sample data of the working medium thermodynamic parameters into a training set, a test set, and a validation set according to a preset ratio; Training the initial RBF neural network model using the training set; Testing the RBF neural network model that has passed the training using the test set; Validating the RBF neural network model that has passed the test using the validation set to obtain a validated RBF neural network model, that is, the target RBF neural network model; Among them, training the initial RBF neural network model using the training set specifically includes: Using the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as the model inputs, and using the corresponding working medium temperature as the model output to train the initial RBF neural network model; Calculating the first root mean square error of the RBF neural network model after the current iteration, and determining whether the first root mean square error is less than or equal to a first preset value to obtain a first judgment result; If the first judgment result is yes, obtaining the RBF neural network model that has passed the training; If the first judgment result is no, setting the currently iterated RBF neural network model as the initial RBF neural network model, and returning to the step "Using the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as the model inputs, and using the corresponding working medium temperature as the model output to train the initial RBF neural network model" until the first root mean square error is less than or equal to the first preset value; Among them, testing the RBF neural network model that has passed the training using the test set specifically includes: Inputting the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the test set into the RBF neural network model that has passed the training, and outputting the working medium temperature test value; Calculating the second root mean square error of the RBF neural network model after the current test iteration according to the working medium temperature test value and the actual working medium temperature value, and determining whether the second root mean square error is less than or equal to a second preset value to obtain a second judgment result; If the second judgment result is yes, obtaining the RBF neural network model that has passed the test, and recording the propagation speed of the RBF neural network model that has passed the test; If the second judgment result is negative, increase the propagation speed of the currently iterated RBF neural network model by a preset value to obtain a tested and optimized RBF neural network model. Let the tested and optimized RBF neural network model be the RBF neural network model that has passed the training, and return to the step of "inputting the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the test set into the RBF neural network model that has passed the training, and outputting the tested value of the working fluid temperature" until the second root mean square error is less than or equal to the second preset value; Among them, using the validation set to validate the RBF neural network model that has passed the test specifically includes: Inputting the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the validation set into the RBF neural network model that has passed the test, and outputting the verified value of the working fluid temperature; Calculating the third root mean square error of the RBF neural network model after the current validation iteration according to the verified value of the working fluid temperature and the actual value of the working fluid temperature, and judging whether the third root mean square error is less than or equal to the second preset value to obtain a third judgment result; If the third judgment result is positive, obtain the RBF neural network model that has passed the validation, that is, the target RBF neural network model; If the third judgment result is negative, set the propagation speed of the RBF neural network model that has passed the test to the initial propagation speed, and return to the step of "using the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value in the training set as the model input, and the corresponding working fluid temperature as the model output to train the initial RBF neural network model" until the third root mean square error is less than or equal to the second preset value.

2. The method according to claim 1, wherein The first preset value is greater than the second preset value.

3. A working medium thermodynamic property calculation system based on an RBF neural network according to the method described in any one of claims 1 to 2, characterized in that, The system includes: A sample data acquisition module for acquiring sample data of the working fluid thermodynamic parameters in the gas turbine model; the sample data of the working fluid thermodynamic parameters includes the fuel-air ratio, the working fluid temperature, and the working fluid enthalpy value or the working fluid entropy value; A model optimization module for setting the initial propagation speed of the initial RBF neural network model, and using the fuel-air ratio and the working fluid enthalpy value or the working fluid entropy value as the model input, and the corresponding working fluid temperature as the model output to optimize the initial RBF neural network model to obtain the target RBF neural network model; A calculation module for calculating the thermodynamic properties of the working fluid using the target RBF neural network model; The model optimization module includes a data set division sub-module, a training sub-module, a testing sub-module, and a validation sub-module; The data set division sub-module is used to divide the sample data of the working fluid thermodynamic parameters into a training set, a test set, and a validation set according to a preset ratio; The training sub-module is used to train the initial RBF neural network model using the training set; The testing sub-module is used to test the RBF neural network model that has passed the training using the test set; The validation sub-module is used to validate the RBF neural network model that has passed the test using the validation set to obtain the RBF neural network model that has passed the validation, that is, the target RBF neural network model; Among them, the training sub-module specifically includes: A training unit, configured to use the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as model inputs, and the corresponding working medium temperature as the model output to train the initial RBF neural network model; A first judgment unit, configured to calculate the first root mean square error of the currently trained RBF neural network model, and judge whether the first root mean square error is less than or equal to a first preset value to obtain a first judgment result; If the first judgment result is yes, obtain the RBF neural network model that passes the training; If the first judgment result is no, set the currently trained RBF neural network model as the initial RBF neural network model, and return to the step "use the fuel-air ratio and the enthalpy value or the entropy value of the working medium in the training set as model inputs, and the corresponding working medium temperature as the model output to train the initial RBF neural network model" until the first root mean square error is less than or equal to the first preset value; Among them, the testing sub-module specifically includes: A testing unit, configured to input the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the testing set into the RBF neural network model that passes the training, and output a working medium temperature test value; A second judgment unit, configured to calculate the second root mean square error of the RBF neural network model after the current testing iteration according to the working medium temperature test value and the actual working medium temperature, and judge whether the second root mean square error is less than or equal to a second preset value to obtain a second judgment result; If the second judgment result is yes, obtain the RBF neural network model that passes the testing, and record the propagation speed of the RBF neural network model that passes the testing; If the second judgment result is no, increase the propagation speed of the currently iterated RBF neural network model by a preset value to obtain a tested and optimized RBF neural network model, set the tested and optimized RBF neural network model as the RBF neural network model that passes the training, and return to the step "input the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the testing set into the RBF neural network model that passes the training, and output a working medium temperature test value" until the second root mean square error is less than or equal to the second preset value; Among them, using the validation set to validate the RBF neural network model that passes the testing specifically includes: Input the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the validation set into the RBF neural network model that passes the testing, and output a working medium temperature validation value; Calculate the third root mean square error of the RBF neural network model after the current validation iteration according to the working medium temperature validation value and the actual working medium temperature, and judge whether the third root mean square error is less than or equal to the second preset value to obtain a third judgment result; If the third judgment result is yes, obtain the RBF neural network model that passes the validation, that is, the target RBF neural network model; If the result of the third determination is negative, set the propagation speed of the RBF neural network model that has passed the test to the initial propagation speed, and return to the step of "training the initial RBF neural network model with the fuel-air ratio and the working medium enthalpy value or the working medium entropy value in the training set as the model input and the corresponding working medium temperature as the model output" until the third root mean square error is less than or equal to the second preset value.

Images