A CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism

Abstract

This invention discloses a CNN-based geothermal ORC working fluid selection method based on an attention-focusing mechanism, specifically including the following steps: S1, constructing a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2, extracting the spatial correlation of working fluid features through a convolutional neural network; S3, introducing a self-attention mechanism to dynamically adjust feature weights; S4, selecting the optimal working fluid based on the output layer prediction results using a loss function. This invention relates to the field of geothermal energy utilization technology. This CNN-based geothermal ORC working fluid selection method based on an attention-focusing mechanism features simple feature coupling, and the working fluid performance is highly linearly coupled with parameters such as heat source temperature, condensation conditions, and heat transfer area. It allows for better modeling using existing methods, significantly reduces computational costs, eliminates the need for repeated iterations of the thermodynamic model, greatly saves time, and does not rely on expert experience. It also exhibits good dynamic adaptability, providing a rapid response to geothermal resource fluctuations.

Description

Technical Field

[0001] This invention relates to the field of geothermal energy utilization technology, specifically to a CNN-based geothermal ORC working fluid optimization method based on an attention-focusing mechanism. Background Technology

[0002] The Organic Rankine Cycle (ORC) is a core technology for low-temperature geothermal power generation, and its performance is highly dependent on the choice of working fluid. Traditional working fluid optimization methods are based on thermodynamic cycle analysis (such as net power output and thermal efficiency) or empirical formulas, requiring manual feature extraction and multi-objective optimization, which has the following problems: Complex feature coupling: The working fluid performance is highly nonlinearly coupled with parameters such as heat source temperature, condensation conditions, and heat transfer area, making it difficult to fully model using traditional methods; High computational cost: It requires repeated iterations of thermodynamic models (such as narrow-point analysis, APR / VPR optimization), which is time-consuming and relies on expert experience; Poor dynamic adaptability: unable to respond quickly to fluctuations in geothermal resources (such as changes in heat source temperature or the combination of mixed working fluids).

[0003] In recent years, artificial intelligence technologies (such as deep learning) have been introduced into the field of working fluid optimization, but existing methods mostly use a single fully connected neural network or support vector machine (SVM), which is difficult to effectively capture the spatial correlation of multi-dimensional features. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a CNN-based geothermal ORC working fluid optimization method based on an attention-focusing mechanism. By automatically learning the mapping relationship between working fluid characteristics and system performance, it achieves efficient optimization of low-temperature geothermal ORC working fluids.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a CNN-based geothermal ORC working fluid optimization method based on an attention-focusing mechanism, specifically comprising the following steps: S1. Construct a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2. Extract the spatial correlation of working fluid features through a convolutional neural network; S3. Introduce a self-attention mechanism to dynamically adjust feature weights; S4. Select the optimal working fluid based on the output layer prediction results using the loss function.

[0006] Preferably, the working fluid properties in step S1 include critical temperature, critical pressure, latent heat of vaporization, viscosity, and thermal conductivity, and the heat source conditions are geothermal water temperature, geothermal water flow rate, and cooling water temperature. The economic indicators are APR and VPR.

[0007] Preferably, in step S1, discrete working fluids (such as R245fa and R601a) are One-Hot encoded before constructing the multidimensional dataset, and continuous parameters are normalized.

[0008] Preferably, the multidimensional data construction in step S1 specifically involves constructing a two-dimensional matrix (such as a time-parameter matrix or a space-parameter matrix) from the working fluid properties and system operating parameters, which serves as the input to the CNN.

[0009] Preferably, the convolutional neural network in step S2 has a multi-layer convolutional structure, as detailed below: The first layer of convolutional kernels extracts local features (such as the matching between the critical temperature of the working fluid and the temperature of the heat source), and subsequent layers capture higher-order nonlinear relationships (such as the coupling effect between the viscosity of the working fluid and the expansion ratio) by stacking convolutional kernels.

[0010] Preferably, the attention concentration mechanism in step S3 is as follows: T1. A self-attention module is introduced after the convolutional layer to dynamically adjust the feature weights and enhance the influence of key parameters (such as APR and VPR) on the working fluid performance. T2, attention weights are calculated using learnable parameters, and the specific formula is as follows: ; Where Q, K, and V are the query, key, and value matrices, and d k This is the scaling factor.

[0011] Preferably, the output layer in step S4 includes a classification task and a regression task: The classification task is used to predict the optimal working fluid category (such as R152a, R245fa, etc.). The regression task is used to predict the system's net output power or economic indicators (such as unit investment cost).

[0012] Preferably, the loss function in step S4 is specifically: a combination of mean squared error (MSE) and cross-entropy loss, balancing accuracy and diversity. ; Where α∈[0,1] is the balance coefficient.

[0013] Preferably, in step S4, a dataset is constructed based on historical experimental data and simulation data (such as Aspen Plus / ThermoCalc simulation results) for data training.

[0014] Preferably, validation is performed after data training. The specific validation process is as follows: cross-validation is used to evaluate the generalization ability of the model, and the performance indicators (such as accuracy and computation time) are compared with those of the genetic algorithm optimization.

[0015] (III) Beneficial Effects This invention provides a CNN-based geothermal ORC working fluid selection method based on an attention-focusing mechanism. Compared with existing technologies, it has the following advantages: This CNN-based geothermal ORC working fluid selection method based on an attention-focusing mechanism specifically includes the following steps: S1, constructing a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2, extracting the spatial correlation of working fluid features through a convolutional neural network; S3, introducing a self-attention mechanism to dynamically adjust feature weights; S4, selecting the optimal working fluid based on the output layer prediction results through a loss function. Feature coupling is simple, and the working fluid performance is highly linearly coupled with parameters such as heat source temperature, condensation conditions, and heat transfer area. Existing methods can be used for better modeling, significantly reducing computational costs. It eliminates the need for repeated iterations of thermodynamic models (such as narrow-point analysis and APR / VPR optimization), greatly saving time and eliminating reliance on expert experience. It also exhibits good dynamic adaptability: it responds quickly and effectively to geothermal resource fluctuations (such as changes in heat source temperature and mixed working fluid combinations). Attached Figure Description

[0016] Figure 1 This is a flowchart of the CNN geothermal ORC working fluid selection method based on the attention concentration mechanism of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 The present invention provides three technical solutions: a CNN-based geothermal ORC working fluid optimization method based on an attention concentration mechanism, specifically including the following embodiments: Example 1: Data preparation: Collect the physical property parameters of 10 typical working fluids (such as R152a, R245fa, R601a, etc.) and the corresponding ORC system operating data; Construct a dataset containing 1000 samples and divide it into training, validation, and test sets in an 8:1:1 ratio.

[0019] Model building: CNN structure: Input layer: 128×128 two-dimensional matrix (including normalized working fluid parameters and system conditions); Convolutional layers: 3 convolutional layers (64, 128, and 256 filters respectively, kernel size 3×3); Attention layer: Self-attention module (number of heads = 8, dimension = 64); Output layer: Fully connected layer (Softmax activation, outputs the probability of working fluid category).

[0020] Training process: Optimizer: Adam (learning rate = 0.001); Batch size: 32; Training rounds: 100; Early stopping strategy: Terminate training when the validation set loss shows no improvement for 5 consecutive rounds.

[0021] Results analysis: The model achieved a classification accuracy of 92% on the test set, which is better than traditional methods (such as SVM's 78% and random forest's 85%). Attention weighting analysis shows that APR and VPR account for more than 60% of the total weight.

[0022] Example 2: A CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism, specifically including the following steps: S1. Construct a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2. Extract the spatial correlation of working fluid features through a convolutional neural network; S3. Introduce a self-attention mechanism to dynamically adjust feature weights; S4. Select the optimal working fluid based on the output layer prediction results using the loss function.

[0023] In this embodiment of the invention, the working fluid physical properties in step S1 include critical temperature, critical pressure, latent heat of vaporization, viscosity and thermal conductivity, and the heat source conditions are geothermal water temperature, geothermal water flow rate and cooling water temperature. The economic indicators are APR and VPR.

[0024] In this embodiment of the invention, before constructing the multidimensional dataset, discrete working fluids (such as R245fa and R601a) are One-Hot encoded, and continuous parameters are normalized.

[0025] In this embodiment of the invention, the multidimensional data construction in step S1 specifically involves constructing a two-dimensional matrix (such as a time-parameter matrix or a space-parameter matrix) from the working fluid properties and system operating parameters, which serves as the input to the CNN.

[0026] In this embodiment of the invention, the structure of the convolutional neural network in step S2 is a multi-layer convolutional neural network, as detailed below: The first layer of convolutional kernels extracts local features (such as the matching between the critical temperature of the working fluid and the temperature of the heat source), and subsequent layers capture higher-order nonlinear relationships (such as the coupling effect between the viscosity of the working fluid and the expansion ratio) by stacking convolutional kernels.

[0027] In this embodiment of the invention, the attention concentration mechanism in step S3 is as follows: T1. A self-attention module is introduced after the convolutional layer to dynamically adjust the feature weights and enhance the influence of key parameters (such as APR and VPR) on the working fluid performance. T2, attention weights are calculated using learnable parameters, and the specific formula is as follows: ; Where Q, K, and V are the query, key, and value matrices, and d k This is the scaling factor.

[0028] Example 3: A CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism, specifically including the following steps: S1. Construct a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2. Extract the spatial correlation of working fluid features through a convolutional neural network; S3. Introduce a self-attention mechanism to dynamically adjust feature weights; S4. Select the optimal working fluid based on the output layer prediction results using the loss function.

[0029] In this embodiment of the invention, the working fluid physical properties in step S1 include critical temperature, critical pressure, latent heat of vaporization, viscosity and thermal conductivity, and the heat source conditions are geothermal water temperature, geothermal water flow rate and cooling water temperature. The economic indicators are APR and VPR.

[0030] In this embodiment of the invention, before constructing the multidimensional dataset, discrete working fluids (such as R245fa and R601a) are One-Hot encoded, and continuous parameters are normalized.

[0031] In this embodiment of the invention, the multidimensional data construction in step S1 specifically involves constructing a two-dimensional matrix (such as a time-parameter matrix or a space-parameter matrix) from the working fluid properties and system operating parameters, which serves as the input to the CNN.

[0032] In this embodiment of the invention, the structure of the convolutional neural network in step S2 is a multi-layer convolutional neural network, as detailed below: The first layer of convolutional kernels extracts local features (such as the matching between the critical temperature of the working fluid and the temperature of the heat source), and subsequent layers capture higher-order nonlinear relationships (such as the coupling effect between the viscosity of the working fluid and the expansion ratio) by stacking convolutional kernels.

[0033] In this embodiment of the invention, the attention concentration mechanism in step S3 is as follows: T1. A self-attention module is introduced after the convolutional layer to dynamically adjust the feature weights and enhance the influence of key parameters (such as APR and VPR) on the working fluid performance. T2, attention weights are calculated using learnable parameters, and the specific formula is as follows: ; Where Q, K, and V are the query, key, and value matrices, and d k This is the scaling factor.

[0034] In this embodiment of the invention, the output layer in step S4 includes a classification task and a regression task: The classification task is used to predict the optimal working fluid category (such as R152a, R245fa, etc.). The regression task is used to predict the system's net output power or economic indicators (such as unit investment cost).

[0035] In this embodiment of the invention, the loss function in step S4 is specifically: combining mean squared error (MSE) and cross-entropy loss, taking into account both accuracy and diversity. ; Where α∈[0,1] is the balance coefficient.

[0036] In this embodiment of the invention, in step S4, a dataset is constructed based on historical experimental data and simulation data (such as simulation results from Aspen Plus / ThermoCalc) for data training. After data training, validation is performed. Specifically, the validation process involves using cross-validation to evaluate the model's generalization ability and comparing its performance metrics (such as accuracy and computation time) with those optimized by a genetic algorithm.

[0037] In summary, this invention features simple coupling, and the working fluid performance is highly linearly coupled with parameters such as heat source temperature, condensation conditions, and heat transfer area. It can be better modeled using existing methods, significantly reducing computational costs. It eliminates the need for repeated iterations of thermodynamic models (such as narrow-point analysis and APR / VPR optimization), saving considerable time and eliminating reliance on expert experience. It also exhibits good dynamic adaptability, providing excellent and rapid response to geothermal resource fluctuations (such as changes in heat source temperature and mixed working fluid combinations).

[0038] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism, characterized in that: Specifically, the following steps are included: S1. Construct a multidimensional dataset containing working fluid physical properties, heat source conditions, and economic indicators; S2. Extract the spatial correlation of working fluid features through a convolutional neural network; S3. Introduce a self-attention mechanism to dynamically adjust feature weights; S4. Select the optimal working fluid based on the output layer prediction results using the loss function.

2. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: The working fluid properties in step S1 include critical temperature, critical pressure, latent heat of vaporization, viscosity, and thermal conductivity. The heat source conditions are geothermal water temperature, geothermal water flow rate, and cooling water temperature. The economic indicators are APR and VPR.

3. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: In step S1, the discrete working fluid is One-Hot encoded before constructing the multidimensional dataset, and the continuous parameters are normalized.

4. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: In step S1, the multidimensional data construction specifically involves constructing a two-dimensional matrix from the working fluid properties and system operating parameters, which serves as the input to the CNN.

5. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: The convolutional neural network in step S2 has a multi-layer convolutional structure, as detailed below: The first layer of convolutional kernels extracts local features, and subsequent layers capture higher-order nonlinear relationships by stacking convolutional kernels.

6. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: The attention concentration mechanism in step S3 is as follows: T1. A self-attention module is introduced after the convolutional layer to dynamically adjust the feature weights and enhance the influence of key parameters on the working fluid performance. T2, attention weights are calculated using learnable parameters, and the specific formula is as follows: ； Where Q, K, and V are the query, key, and value matrices, and d k This is the scaling factor.

7. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: The output layer in step S4 includes classification and regression tasks: The classification task is used to predict the optimal working fluid category; The regression task is used to predict the system's net output power or economic indicators.

8. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: The loss function in step S4 is specifically designed to combine mean squared error (MSE) and cross-entropy loss, balancing accuracy and diversity. ； Where α∈[0,1] is the balance coefficient.

9. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 1, characterized in that: In step S4, a dataset is constructed based on historical experimental data and simulation data for data training.

10. The CNN-based geothermal ORC working fluid optimization method based on attention concentration mechanism according to claim 9, characterized in that: After training on the data, validation is performed. The specific validation process is as follows: cross-validation is used to evaluate the generalization ability of the model, and the performance indicators are compared with those optimized by the genetic algorithm.

Images