Compressed air energy storage power station centrifugal compressor impeller optimization design method and system

Abstract

The invention relates to the technical field of energy storage, and provides a compressed air energy storage power station centrifugal compressor impeller optimization design method and system, and the method comprises the steps: S1, building a centrifugal compressor sample database under various rated conditions; s2, constructing a deep belief neural network model of the centrifugal compressor, training the model by using the sample database, and optimizing model parameters; and S3, giving performance index parameters of the centrifugal compressor to be designed, obtaining optimized structure parameters of the centrifugal compressor by utilizing the trained model, and carrying out modeling and performance verification on the impeller according to the optimized structure parameters. The system comprises a centrifugal compressor sample database unit, a centrifugal compressor deep belief neural network model training unit and an impeller optimization design unit. The method has strong non-linear mapping capability and self-learning capability, and can excavate the non-linear relationship between the structure parameters and the performance parameters of the centrifugal compressor impeller, so that the design parameters of the impeller are optimized more accurately and quickly.

Description

Technical Field

[0001] The present invention relates to the field of energy storage technology, and in particular to a method and system for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station. Background Art

[0002] Compressed air energy storage, as a long-term energy storage, has the characteristics of large capacity, high safety, long service life and low construction cost. It plays an important supporting role in auxiliary services such as peak regulation, frequency regulation, peak shaving and valley filling, and rotational inertia support, and can help the rapid development of new power systems.

[0003] The compressor, a key component of a centrifugal compressor, employs aerodynamics, utilizing an impeller to increase the pressure of the gas in the centrifugal field, simultaneously increasing its kinetic energy. This kinetic energy is then converted into static pressure energy as the gas flows through the diffuser, further increasing the gas pressure. Therefore, as a key working component of a centrifugal compressor, optimizing the blade structural parameters is crucial for improving compressor performance and reducing energy consumption.

[0004] Commonly used design methods include similarity criterion methods, flow path design methods, and model methods. Similarity criterion methods primarily design new impellers based on previous design criteria; flow path design methods predict the performance of new design components based on the performance curves of basic components, thereby determining the impeller's structural parameters; and model methods primarily use complex empirical or theoretical models to find the optimal design parameters. The structural and performance parameters of centrifugal compressors have a nonlinear relationship, and finding a more precise method to establish this nonlinear relationship is of great significance.

[0005] The present invention adopts a data-driven method to establish a centrifugal compressor impeller optimization design method to find the optimal impeller structural parameters. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a method and system for optimizing the impeller design of a centrifugal compressor of a compressed air energy storage power station. The method and system have powerful nonlinear mapping and self-learning capabilities, and can explore the nonlinear relationship between the structural parameters and performance parameters of the centrifugal compressor impeller, thereby optimizing the impeller design parameters more accurately and quickly.

[0007] The present invention adopts the following technical solutions:

[0008] In one aspect, the present invention provides a method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station, comprising:

[0009] S1. Establishing a sample database of centrifugal compressors under various rated operating conditions, wherein the sample database includes structural parameters of the centrifugal compressors and corresponding performance index parameters, and determining constraints on the structural parameters;

[0010] S2. Constructing a deep belief neural network model for a centrifugal compressor, using the sample database established in step S1 to train the deep belief neural network model for the centrifugal compressor, and optimizing model parameters;

[0011] S3. Given the performance index parameters of the centrifugal compressor to be designed, use the deep belief neural network model of the centrifugal compressor trained in step S2 to obtain the optimized structural parameters of the centrifugal compressor, and model the impeller and verify its performance based on the optimized structural parameters.

[0012] Any possible implementation as described above, further provides an implementation, in step S1, the centrifugal compressor structural parameters include inlet rim radius, inlet hub radius, rotor outlet radius, inlet airflow angle and impeller speed; the performance index parameters include efficiency, pressure ratio and flow rate; the constraint condition is the range of values of the centrifugal compressor structural parameters determined according to actual engineering.

[0013] As for any possible implementation described above, a further implementation is provided, in step S1, the multiple rated operating conditions include low temperature operating conditions, normal temperature operating conditions and high temperature operating conditions.

[0014] Any possible implementation as described above, further provides an implementation method, in step S1, the method for establishing the sample database is specifically: based on a standard impeller, different working conditions are set, and by continuously changing the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters, the corresponding centrifugal compressor efficiency, pressure ratio, and flow parameters are obtained through CFD (Computational Fluid Dynamics) software simulation.

[0015] Any possible implementation as described above, further provides an implementation, in step S2, the centrifugal compressor deep belief neural network model is composed of a multi-layer unsupervised restricted Boltzmann machine RBM and a BP neural network, wherein the RBM includes a visual layer, a hidden layer, and an output layer;

[0016] During model training, the input of the centrifugal compressor deep belief neural network is the centrifugal compressor structural parameters in the sample database, including the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed, and the output is the compressor efficiency, pressure ratio, and flow parameters.

[0017] For any of the possible implementations described above, a further implementation is provided, in step S2, the specific method of model establishment and training is:

[0018] S21, randomly dividing the data in the sample database into a training data set and a test data set;

[0019] S22. Establish the energy function between the visible layer and the hidden layer of RBM:

[0020]

[0021] Where E(v,h) is the energy function between the visible layer and the hidden layer; v i is the output value of the i-th visible layer node. The number of nodes is the number of input parameters, that is, the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed correspond to one visible layer node. j is the output value of the jth hidden layer node, w is the weight matrix of the visible layer and the hidden layer, w ji is the element of the weight matrix, b and c are the bias of the visible layer and the hidden layer respectively, m is the number of nodes in the visible layer, that is, the number of input parameters, and n is the number of nodes in the hidden layer;

[0022] S23. Calculate the joint probability function p(v,h) between the visual layer and the hidden layer based on the energy function, and establish each conditional probability distribution p(h|v) and p(v|h);

[0023] When the visible layer or hidden layer is determined, the activation functions of the visible layer and the hidden layer are:

[0024]

[0025] According to the conditional probability distribution p(h|v) and p(v|h), randomly extract samples to reconstruct the visible layer;

[0026] S24. During the training process, the contrastive divergence algorithm is used to update the parameters. The update rules of each parameter are:

[0027] Δw=ε(E data (v i h j )-E recon (v i h j ))

[0028] Δc=ε(E data (h j )-E recon (h j ))

[0029] Δb=ε(E data (v i )-E recon (v i ))

[0030] Where ε is the learning rate, E data (.) and E recon (.) represent the expected values of input data and reconstructed data, respectively, Δw, Δc, and Δb are w ij 、b i and c j According to the above update rules, update the parameter w ij 、b i and c j ;

[0031] S25, fine-tuning:

[0032] A BP network is set at the end of the deep belief neural network model DBN of the centrifugal compressor, which receives the output feature vector of RBM as its input feature vector and uses the BP algorithm to adjust the parameter w ij 、b i and c j Perform fine-tuning to establish a globally optimized prediction model;

[0033] S26, test:

[0034] The trained deep belief neural network is tested using the test dataset.

[0035] Any possible implementation as described above is further provided, in which the deep belief neural network parameters include the weight matrix w of the visible layer and the hidden layer, the visible layer bias b and the hidden layer bias c, and the particle swarm optimization algorithm is used to optimize the deep belief neural network parameters.

[0036] Any possible implementation as described above, further provides an implementation, in step S3, based on the design performance indicators of the centrifugal compressor, using the deep belief neural network model of the centrifugal compressor trained in step S2, reversely calculate and obtain the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters that have the highest efficiency and meet the requirements of pressure ratio and flow rate, and perform three-dimensional modeling of the impeller based on these design parameters, and perform three-dimensional aerodynamic optimization of the impeller using a professional design software package.

[0037] On the other hand, the present invention also provides a system for optimizing the design of centrifugal compressor impellers in a compressed air energy storage power station, wherein the system is used to implement the above method, and the system comprises:

[0038] A centrifugal compressor sample database unit is used to establish centrifugal compressor structural parameters and corresponding performance index parameters, and determine the constraints of the structural parameters;

[0039] a centrifugal compressor deep belief neural network model training unit, configured to construct a centrifugal compressor deep belief neural network model, train the centrifugal compressor deep belief neural network model using the sample database, and optimize model parameters;

[0040] The impeller optimization design unit, given the performance index parameters of the centrifugal compressor to be designed, uses the trained deep belief neural network model of the centrifugal compressor to obtain the optimized structural parameters of the centrifugal compressor, and models the impeller and verifies its performance based on the optimized structural parameters.

[0041] Any possible implementation as described above further provides an implementation, wherein the impeller optimization design unit uses a professional design software package to perform three-dimensional aerodynamic optimization on the impeller and uses CFD software to verify the impeller performance.

[0042] The beneficial effects of the present invention are:

[0043] 1. The optimization design method provided by the present invention has powerful nonlinear mapping and self-learning capabilities, and can explore the nonlinear relationship between the structural parameters and performance parameters of the centrifugal compressor impeller, thereby more accurately predicting and optimizing the performance of the impeller.

[0044] 2. Compared with the conventional CFD simulation process, the optimization design method provided by the present invention has fast design speed and high accuracy, significantly improves the design accuracy and efficiency of the impeller, reduces the trial and error cost and time cost in the design process, significantly shortens the design cycle of the centrifugal compressor impeller, and thus shortens the product's time to market.

[0045] 3. The optimization method provided by the present invention fully considers the performance indicators of the impeller such as efficiency and pressure ratio, and realizes comprehensive optimization design while meeting numerous design constraints.

[0046] 4. The optimization method provided by the present invention has a stronger competitiveness than the products obtained by traditional design methods, and can provide users with more efficient and reliable services. At the same time, it reduces costs and construction time, and can respond to market demand more quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure shows a flow chart of a method for optimizing the impeller design of a centrifugal compressor of a compressed air energy storage power station according to an embodiment of the present invention.

[0048] Figure 2 The figure shows a schematic diagram of the training process of the deep belief neural network model of the centrifugal compressor in the embodiment. DETAILED DESCRIPTION

[0049] The following will describe in detail specific embodiments of the present invention with reference to the accompanying drawings. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated, and they can be combined with each other to achieve better technical effects.

[0050] like Figure 1 As shown, an embodiment of the present invention provides a method for optimizing the impeller design of a centrifugal compressor of a compressed air energy storage power station, comprising:

[0051] S1. Establishing a sample database of centrifugal compressors under various rated operating conditions, wherein the sample database includes structural parameters of the centrifugal compressors and corresponding performance index parameters, and determining constraints on the structural parameters;

[0052] S2. Constructing a deep belief neural network model for a centrifugal compressor, using the sample database established in step S1 to train the deep belief neural network model for the centrifugal compressor, and optimizing model parameters;

[0053] S3. Given the performance index parameters of the centrifugal compressor to be designed, use the deep belief neural network model of the centrifugal compressor trained in step S2 to obtain the optimized structural parameters of the centrifugal compressor, and model the impeller and verify its performance based on the optimized structural parameters.

[0054] In a specific embodiment, in step S1, the centrifugal compressor structural parameters include the inlet rim radius, the inlet hub radius, the rotor outlet radius, the inlet airflow angle and the impeller speed; the performance index parameters include efficiency, pressure ratio and flow rate; the constraint conditions are the value ranges of the centrifugal compressor structural parameters determined according to actual engineering practice.

[0055] In a specific embodiment, the multiple rated operating conditions include low temperature operating conditions, normal temperature operating conditions and high temperature operating conditions.

[0056] In a specific embodiment, the method for establishing the sample database is as follows: based on a standard impeller, different working conditions are set, and by continuously changing the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters, the corresponding centrifugal compressor efficiency, pressure ratio, and flow parameters are obtained through CFD software simulation.

[0057] In a specific embodiment, the data in the sample database also includes: existing centrifugal compressor structural parameters and corresponding performance index parameters.

[0058] In a specific embodiment, in step S2, the centrifugal compressor deep belief neural network model is composed of a multi-layer unsupervised restricted Boltzmann machine RBM and a BP neural network, wherein the RBM includes a visual layer, a hidden layer, and an output layer;

[0059] During model training, the input of the centrifugal compressor deep belief neural network is the centrifugal compressor structural parameters in the sample database, including the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed, and the output is the compressor efficiency, pressure ratio, and flow parameters.

[0060] In a specific embodiment, Figure 2 As shown, in step S2, the specific method of model establishment and training is:

[0061] S21, randomly dividing the data in the sample database into a training data set and a test data set;

[0062] S22. Establish the energy function between the visible layer and the hidden layer of RBM:

[0063]

[0064] Where E(v,h) is the energy function between the visible layer and the hidden layer; v is the output value of the visible layer node, h is the output value of the hidden layer node, w is the weight matrix of the visible layer and the hidden layer, b and c are the visible layer bias and the hidden layer bias respectively, m is the number of visible layer nodes, and n is the number of hidden layer nodes.

[0065] S23. Calculate the joint probability function p(v,h) between the visual layer and the hidden layer based on the energy function, and establish each conditional probability distribution p(h|v) and p(v|h);

[0066] When the visible layer or hidden layer is determined, the activation functions of the visible layer and the hidden layer are:

[0067]

[0068] According to the conditional probability distribution p(h|v) and p(v|h), randomly extract samples to reconstruct the visible layer;

[0069] S24. During the training process, the contrastive divergence algorithm is used to update the parameters. The update rules of each parameter are:

[0070] Δw=ε(E data (v i h j )-E recon (v i h j ))

[0071] Δc=ε(E data (h j )-E recon (h j ))

[0072] Δb=ε(E data(v i )-E recon (v i ))

[0073] Where ε is the learning rate, E data (.) and E recon (.) represent the expected values of input data and reconstructed data respectively; according to the above update rules, update the parameter w ij 、b i and c j ;

[0074] S25, fine-tuning:

[0075] At the end of the deep belief neural network model DBN of the centrifugal compressor, a BP network is set up to receive the output feature vector of RBM as its input feature vector, and the parameter w is adjusted by BP algorithm. ij 、b i and c j Perform fine-tuning to establish a globally optimized prediction model;

[0076] S26, test:

[0077] Use the test dataset to test the trained deep belief neural network.

[0078] In a specific embodiment, the deep belief neural network parameters include the weight matrix w of the visible layer and the hidden layer, the bias c of the visible layer b and the hidden layer, and the particle swarm optimization algorithm is used to optimize the deep belief neural network parameters.

[0079] In a specific embodiment, in step S3, based on the design performance indicators of the centrifugal compressor, the deep belief neural network model of the centrifugal compressor trained in step S2 is used to reversely calculate the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters that have the highest efficiency and meet the pressure ratio and flow requirements. Based on these design parameters, the impeller is three-dimensionally modeled, and a professional design software package is used to perform three-dimensional aerodynamic optimization on the impeller.

[0080] In a specific embodiment, step S3 is specifically as follows:

[0081] S31. Given a design performance indicator, set an error threshold;

[0082] S32, preliminarily select initial values of structural parameters, input them into the trained deep belief neural network model, and obtain network output values of performance indicator parameters;

[0083] S33, comparing the error between the design performance index and the performance index parameter network output value. If the error is within the error threshold range, the structural parameter is the optimized structural parameter; if the error exceeds the error threshold range, go to step S34;

[0084] S34, adjusting the initial values of the structural parameters, then re-inputting the trained deep belief neural network model to obtain the adjusted performance indicator parameter network output value, and then going to step S33;

[0085] Repeat steps S33 and S34 until the optimized structural parameters are found.

[0086] An embodiment of the present invention provides a system for optimizing the impeller design of a centrifugal compressor of a compressed air energy storage power station. The system is used to implement the above-mentioned method, including:

[0087] A centrifugal compressor sample database unit is used to establish centrifugal compressor structural parameters and corresponding performance index parameters, and determine the constraints of the structural parameters;

[0088] a centrifugal compressor deep belief neural network model training unit, configured to construct a centrifugal compressor deep belief neural network model, train the centrifugal compressor deep belief neural network model using the sample database, and optimize model parameters;

[0089] The impeller optimization design unit, given the performance index parameters of the centrifugal compressor to be designed, uses the trained deep belief neural network model of the centrifugal compressor to obtain the optimized structural parameters of the centrifugal compressor, and models the impeller and verifies its performance based on the optimized structural parameters.

[0090] In a specific embodiment, the impeller optimization design unit uses a professional design software package to perform three-dimensional aerodynamic optimization on the impeller and uses CFD software to verify the impeller performance.

[0091] The impeller optimization method provided by the present invention can greatly improve the efficiency and accuracy of impeller optimization design, reduce the trial and error cost and time cost in the design process, significantly shorten the design cycle of centrifugal compressor impellers, and thus shorten the product launch time.

[0092] Although several embodiments of the present invention have been described herein, those skilled in the art will appreciate that modifications may be made to the embodiments herein without departing from the spirit of the present invention. The above embodiments are merely exemplary and should not be used as limitations on the scope of the present invention.

Claims

1. A method for optimizing the impeller design of a centrifugal compressor of a compressed air energy storage power station, characterized in that: The method comprises: S1. Establishing a sample database of centrifugal compressors under various rated operating conditions, wherein the sample database includes structural parameters of the centrifugal compressors and corresponding performance index parameters, and determining constraints on the structural parameters; S2. Constructing a deep belief neural network model for a centrifugal compressor, using the sample database established in step S1 to train the deep belief neural network model for the centrifugal compressor, and optimizing model parameters; S3. Given the performance index parameters of the centrifugal compressor to be designed, use the deep belief neural network model of the centrifugal compressor trained in step S2 to obtain the optimized structural parameters of the centrifugal compressor, and model the impeller and verify its performance based on the optimized structural parameters.

2. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 1, wherein: In step S1, the centrifugal compressor structural parameters include the inlet rim radius, the inlet hub radius, the rotor outlet radius, the inlet airflow angle and the impeller speed; the performance index parameters include efficiency, pressure ratio and flow rate; and the constraint conditions are the value ranges of the centrifugal compressor structural parameters determined according to actual engineering practice.

3. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 1, wherein: In step S1 , the multiple rated operating conditions include low temperature operating conditions, normal temperature operating conditions, and high temperature operating conditions.

4. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 2, wherein: In step S1, the method for establishing the sample database is specifically as follows: based on a standard impeller, different working conditions are set, and by continuously changing the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters, the corresponding centrifugal compressor efficiency, pressure ratio, and flow parameters are obtained through CFD software simulation.

5. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 1, wherein: In step S2, the centrifugal compressor deep belief neural network model is composed of a multi-layer unsupervised restricted Boltzmann machine (RBM) and a BP neural network, wherein the RBM includes a visual layer, a hidden layer, and an output layer; During model training, the input of the centrifugal compressor deep belief neural network is the centrifugal compressor structural parameters in the sample database, including the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed, and the output is the compressor efficiency, pressure ratio, and flow parameters.

6. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 5, characterized in that: In step S2, the specific method of model establishment and training is: S21, randomly dividing the data in the sample database into a training data set and a test data set; S22. Establish the energy function between the visible layer and the hidden layer of RBM: Where E(v,h) is the energy function between the visible layer and the hidden layer; v i Output value of the i-th visible layer node, h j is the output value of the jth hidden layer node, w ji is the element of the weight matrix of the visible layer and the hidden layer, b i and c j are the bias of the i-th visible layer and the j-th hidden layer, respectively, m is the number of visible layer nodes, and n is the number of hidden layer nodes; S23. Calculate the joint probability function p(v,h) between the visual layer and the hidden layer based on the energy function, and establish each conditional probability distribution p(h|v) and p(v|h); When the visible layer or hidden layer is determined, the activation functions of the visible layer and the hidden layer are: According to the conditional probability distribution p(h|v) and p(v|h), randomly extract samples to reconstruct the visible layer; S24. During the training process, the contrastive divergence algorithm is used to update the parameters. The update rules of each parameter are: Δw=ε(E data (v i h j )-E recon (v i h j )); Δc=ε(E data (h j )-E recon (h j )); Δb=ε(E data (v i )-E recon (v i )); Where ε is the learning rate, E data (.) and E recon (.) represent the expected values of input data and reconstructed data, respectively, Δw, Δc, and Δb are w ij 、b i and c j According to the above update rules, update the parameter w ij 、b i and c j ; S25, fine-tuning: A BP network is set at the end of the deep belief neural network model of the centrifugal compressor, which receives the output feature vector of the RBM as its input feature vector and uses the BP algorithm to adjust the parameter w ij 、b i and c j Perform fine-tuning to establish a globally optimized prediction model; S26, test: The trained deep belief neural network is tested using the test dataset.

7. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 1, wherein: The parameters of the deep belief neural network include the weight matrix w of the visible layer and the hidden layer, the visible layer bias b and the hidden layer bias c. The particle swarm optimization algorithm is used to optimize the parameters of the deep belief neural network.

8. The method for optimizing the design of a centrifugal compressor impeller for a compressed air energy storage power station according to claim 1, wherein: In step S3, based on the design performance indicators of the centrifugal compressor, the deep belief neural network model of the centrifugal compressor trained in step S2 is used to reversely calculate the inlet rim radius, inlet hub radius, inlet airflow angle, rotor outlet diameter, and impeller speed parameters that have the highest efficiency and meet the requirements of pressure ratio and flow. Based on these design parameters, the impeller is three-dimensionally modeled, and a professional design software package is used to perform three-dimensional aerodynamic optimization of the impeller.

9. A compressed air energy storage power station centrifugal compressor impeller optimization design system, characterized in that: The system is used to implement the method according to any one of claims 1 to 8, and the system includes: A centrifugal compressor sample database unit is used to establish centrifugal compressor structural parameters and corresponding performance index parameters, and determine the constraints of the structural parameters; a centrifugal compressor deep belief neural network model training unit, configured to construct a centrifugal compressor deep belief neural network model, train the centrifugal compressor deep belief neural network model using the sample database, and optimize model parameters; The impeller optimization design unit, given the performance index parameters of the centrifugal compressor to be designed, uses the trained deep belief neural network model of the centrifugal compressor to obtain the optimized structural parameters of the centrifugal compressor, and models the impeller and verifies its performance based on the optimized structural parameters.

10. The compressed air energy storage power station centrifugal compressor impeller optimization design system according to claim 9, characterized in that: The impeller optimization design unit uses a professional design software package to perform three-dimensional aerodynamic optimization on the impeller and uses CFD software to verify the impeller performance.

Images