Method for establishing associated model of high reliability and wide operating conditions steam turbine

Abstract

A method for establishing a high-reliability, wide-operating-condition steam turbine associated model comprises: 1) constructing a traditional steam turbine model and training a steam turbine machine learning model; 2) inputting an input parameter set A from existing steam turbine operating data into the traditional steam turbine model and the trained steam turbine machine learning model, respectively, to obtain output parameter datasets B and F; inputting an input parameter dataset D into the traditional steam turbine model and the trained steam turbine machine learning model, respectively, to obtain output parameter datasets E and G; and 3) training the input parameter set A, output parameter dataset B, output parameter dataset F, input parameter dataset D, output parameter dataset E, and output parameter dataset G from the existing steam turbine operating data in the trained steam turbine machine learning model according to different weights, to obtain a high-reliability, wide-operating-condition steam turbine associated model. The present invention has low reliance on data quality and has a faster operating speed.

Description

Technical Field

[0001] The present invention belongs to the field of safe operation of steam turbines and relates to a method for establishing a steam turbine associated model, and in particular to a method for establishing a high-reliability, wide-operating-condition steam turbine associated model combined with machine learning. Background Art

[0002] There are four main types of existing steam turbine associated simulation technologies.

[0003] One type of technology involves first-principles modeling. These are physical simulation models based on the energy, mass, and momentum conservation equations. Their advantages are minimal data requirements, requiring only key turbine parameters, and a wide range of applicable operating conditions. Their disadvantages include poor accuracy, slow model execution, and difficulty synchronizing with the turbine when time-varying parameters are involved, making it difficult to achieve all the primary objectives of the accompanying model.

[0004] The second category primarily relies on empirical formulas and is often used in combination with the first category's accompanying steam turbine simulation technology. Compared to pure first-principles techniques, these methods offer the advantage of faster model execution, but their applicable operating conditions are narrower. The first two categories are generally referred to as traditional models.

[0005] The third category involves machine learning technologies such as neural networks trained on big data. This involves using machine learning simulations with existing operating data to predict inputs and outputs. The advantages are high runtime speed after training, the ability to operate simultaneously with the turbine even when time-varying parameters are involved, and the ability to serve as a companion model within the available data. However, the disadvantages are the requirement for extensive existing data; high reliance on data quality; relatively poor versatility, with significant variations across turbines; and limited simulation capabilities within the existing operating data, limiting the applicable operating range.

[0006] The fourth category is based on a combination of the three. There are two main types. The first uses first principles or empirical formulas to simulate results, and then uses machine learning to simulate. This is a generalized surrogate model. This technology improves the simulation speed of time-varying models while retaining the advantages of the first and second types of models, but other disadvantages still exist. The second type is based on the third type of technology, using data generated by the first and second types of technologies for training outside the data range. This technology expands the range of operating conditions that the third type of technology can be used for and has a certain degree of versatility. Other advantages and disadvantages are basically the same as the third type of technology.

[0007] However, during power plant operation, steam turbines require testing to evaluate performance, provide guidance for performance optimization, improve variable-condition performance, and enhance unit efficiency, and provide data for turbine maintenance and overhaul. However, in actual turbine operation, many key parameter locations cannot be monitored by sensors, or installing sensors would result in power and efficiency losses. Alternatively, sensors may not be pre-installed in areas prone to failure, making it impossible to directly obtain parameters. Therefore, companion models can provide refined analytical data for turbine optimization, maintenance, and overhaul, and provide early warning of incidents such as turbine failures. Summary of the Invention

[0008] In order to solve the above technical problems existing in the background technology, the present invention provides a method for establishing a steam turbine associated model with high reliability and wide operating conditions, which has less dependence on data quality and faster operation speed.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for establishing a high-reliability, wide-operating-condition steam turbine associated model, characterized in that the method comprises the following steps:

[0011] 1) Build a traditional steam turbine model and build and train a steam turbine machine learning model;

[0012] 2) Inputting the input parameter data set A from the existing steam turbine operation data into the traditional steam turbine model to obtain the output parameter data set B;

[0013] Input the input parameter data set A from the existing steam turbine operation data into the training steam turbine machine learning model to obtain the output parameter data set F;

[0014] Inputting an input parameter data set D having a wider range of operating conditions than an input parameter data set A in existing steam turbine operation data into a traditional steam turbine model to obtain an output parameter data set E;

[0015] Inputting an input parameter dataset D with a wider range of operating conditions than the input parameter dataset A in the existing steam turbine operation data into the training steam turbine machine learning model to obtain an output parameter dataset G;

[0016] 3) The input parameter data set A, output parameter data set B, output parameter data set F, input parameter data set D, output parameter data set E and output parameter data set G in the existing steam turbine operation data are trained in the training turbine machine learning model according to different weights to obtain a high-reliability wide-operating-condition steam turbine associated model.

[0017] In the above step 3), the input parameter data set A and the output parameter data set B in the existing steam turbine operation data form a data set 1; the input parameter data set A and the output parameter data set F in the existing steam turbine operation data form a data set 2;

[0018] The input parameter data set D and the output parameter data set E form a data set three;

[0019] The input parameter data set D and the output parameter data set G form a data set four;

[0020] In step 3), data set 1, data set 2, data set 3 and data set 4 are trained in the steam turbine machine learning model according to different weights to obtain a high-reliability, wide-operating-condition steam turbine associated model.

[0021] The weight of the dataset one is smaller than that of the dataset two; the weight of the dataset four is smaller than that of the dataset three.

[0022] The above-mentioned traditional steam turbine model is constructed based on the design parameters of the steam turbine, according to the first principles or empirical formulas, or according to the first principles and empirical formulas with or without time-varying parameters.

[0023] The above-mentioned training turbine machine learning model is constructed based on the machine learning method using the input parameter data set A in the existing turbine operation data and the existing turbine output parameter data set C.

[0024] The above-mentioned machine learning methods are neural network algorithms or reinforcement learning methods.

[0025] The input parameter data set A in the above-mentioned existing turbine operation data is the result of direct measurement or the parameters obtained by solving the direct measurement results; the direct measurement results are the turbine inlet flow, turbine inlet pressure and / or temperature, turbine outlet flow, turbine outlet pressure and / or temperature, flow at the turbine monitoring point, and pressure and / or temperature at the turbine monitoring point; the parameters obtained by solving the direct measurement results are enthalpy, entropy and / or dryness.

[0026] The above-mentioned output data set B is the result of direct measurement or the parameters calculated from the direct measurement results; the direct measurement result is the pressure, temperature and / or flow rate of a certain measuring point; the parameters calculated from the direct measurement results are enthalpy, entropy and / or dryness.

[0027] The above-mentioned input parameter data set D has a wider range of operating conditions than the input parameter data set A in the existing steam turbine operation data, including but not limited to the steam turbine input parameters when the steam turbine is in a fault condition or a failure condition.

[0028] The advantages of the present invention are:

[0029] The present invention provides a method for establishing a high-reliability, wide-operating-condition steam turbine associated model, comprising: 1) constructing a traditional steam turbine model and constructing a training steam turbine machine learning model; 2) inputting an input parameter data set A in existing steam turbine operating data into the traditional steam turbine model to obtain an output parameter data set B; inputting the input parameter data set A in existing steam turbine operating data into the training steam turbine machine learning model to obtain an output parameter data set F; inputting an input parameter data set D with a wider operating condition than the input parameter data set A in existing steam turbine operating data into the traditional steam turbine model to obtain an output parameter data set E; inputting an input parameter data set D with a wider operating condition than the input parameter data set A in existing steam turbine operating data into the training steam turbine machine learning model to obtain an output parameter data set G; 3) training the input parameter data set A, output parameter data set B, output parameter data set F, input parameter data set D, output parameter data set E and output parameter data set G in the existing steam turbine operating data in the training steam turbine machine learning model according to different weights to obtain a high-reliability, wide-operating-condition steam turbine associated model. The present invention addresses existing issues such as poor accuracy, slow model execution, and the inability to fully implement the companion model's functions. It also addresses issues such as the requirement for a large amount of existing data, high reliance on data quality, and a narrow operating range, limiting simulation capabilities to existing operating data. The present invention combines the advantages of existing technologies while avoiding their primary drawbacks. While maintaining high accuracy, it achieves faster execution, potentially matching or exceeding actual time. It is relatively less dependent on data quality and has a wide operating range, allowing for use outside the range of existing data. Compared to traditional model methods, the present model method offers high accuracy and fast execution after training. It can operate simultaneously with the steam turbine even in situations involving time-varying parameters and can be used as a companion model within the range of existing data. Compared to general machine learning model methods, the present model method requires less data, is less dependent on data quality, and can accommodate operating conditions outside the range of existing data. Overall, while maintaining similar accuracy, it is relatively less dependent on data quality and requires a smaller amount of data. It achieves faster execution, potentially matching or exceeding actual time, and has a wider operating range, allowing for use outside the range of existing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the method for establishing a high-reliability, wide-operating-condition steam turbine associated model provided by the present invention. DETAILED DESCRIPTION

[0031] See also Figure 1 The present invention provides a method for establishing a high-reliability, wide-operating-condition steam turbine associated model, the method comprising the following steps:

[0032] 1) Constructing a traditional steam turbine model and a training steam turbine machine learning model; the traditional steam turbine model is constructed based on the turbine design parameters, established according to first principles or empirical formulas, or constructed according to first principles and empirical formulas with or without time-varying parameters. The training steam turbine machine learning model is constructed using an input parameter dataset A from existing steam turbine operating data and an existing steam turbine output parameter dataset C based on a machine learning method (e.g., a neural network algorithm or reinforcement learning method). The input parameter dataset A from the existing steam turbine operating data is a direct measurement result or a parameter calculated from the direct measurement result; the direct measurement result is the turbine inlet flow rate, turbine inlet pressure and / or temperature, turbine outlet flow rate, turbine outlet pressure and / or temperature, turbine monitoring point flow rate, and turbine monitoring point pressure and / or temperature; the parameters calculated from the direct measurement result are common parameters such as enthalpy, entropy, and / or dryness.

[0033] 2) Inputting an input parameter data set A from existing steam turbine operating data into a traditional steam turbine model to obtain an output parameter data set B; the output parameter data set B is a direct measurement result or a parameter calculated from the direct measurement result; the direct measurement result is pressure, temperature, and / or flow at a certain measuring point; the parameter calculated from the direct measurement result is a common parameter such as enthalpy, entropy, and / or dryness;

[0034] Input the input parameter data set A from the existing steam turbine operation data into the training steam turbine machine learning model to obtain the output parameter data set F;

[0035] An input parameter dataset D with a wider range of operating conditions than the input parameter dataset A in the existing steam turbine operation data is input into a traditional steam turbine model to obtain an output parameter dataset E; the input parameter dataset D includes but is not limited to the turbine input parameters when the steam turbine is in a fault condition or a failure condition.

[0036] An input parameter dataset D with a wider operating condition than the input parameter dataset A in the existing steam turbine operation data is input into the training steam turbine machine learning model to obtain an output parameter dataset G.

[0037] 3) The input parameter dataset A, output parameter dataset B, output parameter dataset F, input parameter dataset D, output parameter dataset E, and output parameter dataset G in the existing steam turbine operating data are trained in a training steam turbine machine learning model according to different weights to obtain a high-reliability, wide-operating-condition steam turbine associated model. The input parameter dataset A and output parameter dataset B in the existing steam turbine operating data form dataset one; the input parameter dataset A and output parameter dataset F in the existing steam turbine operating data form dataset two; the input parameter dataset D and output parameter dataset E form dataset three; and the input parameter dataset D and output parameter dataset G form dataset four. In step 3), dataset one, dataset two, dataset three, and dataset four are trained in a training steam turbine machine learning model according to different weights to obtain a high-reliability, wide-operating-condition steam turbine associated model. The weight of dataset one is less than the weight of dataset two; and the weight of dataset four is less than the weight of dataset three.

[0038] The present invention addresses existing issues such as poor accuracy, slow model execution, and the inability to fully implement the companion model's functions. It also addresses issues such as the requirement for a large amount of existing data, high reliance on data quality, and a narrow operating range, limiting simulation capabilities to existing operating data. The present invention combines the advantages of existing technologies while avoiding their primary drawbacks. While maintaining high accuracy, it achieves faster execution, potentially matching or exceeding actual time. It is relatively less dependent on data quality and has a wide operating range, allowing for use outside the range of existing data. Compared to traditional model methods, the present model method offers high accuracy and fast execution after training. It can operate simultaneously with the steam turbine even in situations involving time-varying parameters and can be used as a companion model within the range of existing data. Compared to general machine learning model methods, the present model method requires less data, is less dependent on data quality, and can accommodate operating conditions outside the range of existing data. Overall, while maintaining similar accuracy, it is relatively less dependent on data quality and requires a smaller amount of data. It achieves faster execution, potentially matching or exceeding actual time, and has a wider operating range, allowing for use outside the range of existing data.

Claims

1. A method for establishing a high-reliability, wide-operating-condition steam turbine associated model, characterized by: The method for establishing a high-reliability, wide-operating-condition steam turbine associated model comprises the following steps: 1) Constructing a traditional steam turbine model and constructing a machine learning model for training a steam turbine; the traditional steam turbine model is constructed based on the design parameters of the steam turbine, based on first principles or empirical formulas, or based on first principles and empirical formulas with or without time-varying parameters; 2) Inputting the input parameter data set A from the existing steam turbine operating data into the traditional steam turbine model to obtain the output parameter data set B; Input the input parameter data set A from the existing steam turbine operation data into the training steam turbine machine learning model to obtain the output parameter data set F; Inputting an input parameter data set D having a wider range of operating conditions than an input parameter data set A in existing steam turbine operation data into a traditional steam turbine model to obtain an output parameter data set E; Inputting an input parameter dataset D with a wider range of operating conditions than the input parameter dataset A in the existing steam turbine operation data into the training steam turbine machine learning model to obtain an output parameter dataset G; 3) The input parameter dataset A, output parameter dataset B, output parameter dataset F, input parameter dataset D, output parameter dataset E, and output parameter dataset G in the existing steam turbine operation data are trained in the steam turbine machine learning model according to different weights to obtain a high-reliability, wide-operating-condition steam turbine associated model; The input parameter data set A and the output parameter data set B in the existing steam turbine operation data form a data set 1; the input parameter data set A and the output parameter data set F in the existing steam turbine operation data form a data set 2; The input parameter data set D and the output parameter data set E form a data set three; The input parameter data set D and the output parameter data set G form a data set four; The data set 1, data set 2, data set 3 and data set 4 are trained in the steam turbine machine learning model according to different weights.

2. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 1, characterized in that: The weight of the data set 1 is smaller than the weight of the data set 2; the weight of the data set 4 is smaller than the weight of the data set 3.

3. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 2, characterized in that: The training steam turbine machine learning model is constructed based on a machine learning method using an input parameter data set A in existing steam turbine operation data and an existing steam turbine output parameter data set C.

4. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 3, characterized in that: The machine learning method is a neural network algorithm or a reinforcement learning method.

5. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 4, characterized in that: The input parameter data set A in the existing turbine operation data is the result of direct measurement; the direct measurement results are the turbine inlet flow, turbine inlet pressure and / or temperature, turbine outlet flow, turbine outlet pressure and / or temperature, flow at the turbine monitoring point, and pressure and / or temperature at the turbine monitoring point.

6. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 4, characterized in that: The input parameter data set A in the existing steam turbine operation data is a parameter calculated from a direct measurement result; the parameter calculated from the direct measurement result is enthalpy, entropy and / or dryness.

7. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 6, characterized in that: The output parameter data set B is the result of direct measurement or the parameters calculated from the direct measurement results; the direct measurement result is the pressure, temperature and / or flow rate of a certain measuring point; the parameters calculated from the direct measurement results are enthalpy, entropy and / or dryness.

8. The method for establishing a high-reliability, wide-operating-condition steam turbine associated model according to claim 7, characterized in that: The input parameter data set D includes the steam turbine input parameters when the steam turbine is in a fault condition or a failure condition.

Images