CNN-LSTM fused geothermal exploitation productivity optimization prediction system

Abstract

The invention discloses a CNN-LSTM fused geothermal exploitation productivity optimization prediction system, which belongs to the technical field of geothermal productivity prediction, and comprises the following process modules: S1, data acquisition; s1-1, geological exploration equipment; s1-2, geothermal well monitoring equipment is used; s1-3, monitoring equipment of the heating system; s2, data preprocessing; s2-1, a missing value processing unit; s2-2, an abnormal value detection unit; s2-3, carrying out unit normalization; s3, a feature screening module; s3-1, a correlation analysis unit; and S4, a model construction module. The CNN-LSTM fused geothermal exploitation productivity optimization prediction system not only can realize comprehensive data coverage so that the model can accurately capture dynamic changes of a geothermal system, but also can reduce redundant variables through feature screening so as to realize accurate prediction, dynamic optimization and intelligent regulation and control of geothermal exploitation productivity; high-precision, high-efficiency and high-adaptability technical support is provided for scientific development of geothermal energy, and the geothermal energy is promoted to advance towards the predictable, controllable and sustainable direction.

Description

Technical Field

[0001] This invention relates to the field of geothermal production capacity prediction technology, specifically to a geothermal extraction production capacity optimization prediction system that integrates CNN-LSTM. Background Technology

[0002] Geothermal energy is an important renewable and clean energy source. Given the continuous growth in global energy demand and the serious environmental problems and energy security risks caused by the massive consumption of traditional fossil fuels, developing geothermal resources is of great significance for optimizing the energy structure. Geothermal extraction refers to the process of extracting geothermal resources (mainly hot water or steam) from the Earth's internal thermal reservoirs using engineering techniques and converting them into usable energy (such as heating and power generation). Geothermal extraction prediction systems can achieve accurate prediction of geothermal extraction capacity (production temperature, heat extraction power, injection-production pressure difference) and intelligent optimization of operating parameters by integrating deep learning, numerical simulation, and multi-objective optimization technologies. However, when using the equipment, the complexity of the geothermal extraction process, especially the multi-field coupling effects of seepage, heat transfer, stress deformation, and water-rock reaction involved in heat extraction from the working fluid within the geothermal reservoir, results in a large computational load and long operating time. Furthermore, hydrothermal geothermal wells are often deep and located in complex high-temperature and high-pressure environments, making it impossible to directly obtain some geological parameters of the thermal reservoir through conventional methods. This fails to meet engineering needs and reduces the overall extraction efficiency and utilization rate of geothermal resources.

[0003] To overcome the above-mentioned shortcomings, prior art 1 (Chinese patent application number 202311771116.0, application date 2023-12-21) provides a method and system for intelligent prediction of geothermal production capacity based on PINNs and physical boundary condition constraints. The method involves creating a dataset based on time coordinates, spatial coordinates, and boundary / initial conditions; inputting the dataset into a PINNs model and constructing a loss function using physical constraint formulas; obtaining the optimal model based on the loss function and the optimizer built into the PINNs model; inputting the data to be tested into the optimal model to obtain the predicted value. This device uses physical boundary conditions as input data to build the PINNs model and as constraints for model training. The trained model can achieve truly intelligent prediction of production capacity, enhancing the model's transferability and considering physical mechanisms, thus enabling real-time and accurate prediction of geothermal production capacity. Prior art 2 (application number 202... (Chinese Patent No. 211635103.6, filed on 2022-12-19) A method for predicting the temperature of extraction fluid in a karst geothermal well-injection system is proposed. This method involves collecting relevant geological data and operational data of the geothermal system, determining the tracer composition and calculating the dosage, conducting tracer experiments in the karst well-injection system, obtaining the results, and plotting curves. Based on the results, Qtracer2 software is used to calculate the tracer recovery rate, cross-sectional area, and height of the karst channel. Boundary conditions for the karst geothermal heat transfer model are determined, a time period is set, and the temperature change of the extraction well is calculated based on the heat transfer model. This method for predicting the temperature of extraction fluid is accurate, reliable, highly operable, convenient, and fast. It solves the problem of predicting the temperature of extraction fluid in a well-injection system under long-term reinjection conditions and provides important guidance for karst geothermal extraction and reinjection schemes, as well as the selection of well locations for later projects.

[0004] Geothermal reservoirs exhibit spatial heterogeneity; for example, the temperature field and permeability are not uniformly distributed spatially. This makes it difficult for a single model to fully understand the characteristics of geothermal reservoirs from a spatial perspective, resulting in insufficient analysis of factors affecting production capacity. For instance, it is difficult to determine the degree of geothermal enrichment and the direction of heat conduction in different areas, making predictions unable to accurately reflect the impact of spatial changes in reservoir capacity. Furthermore, the aforementioned device cannot achieve fusion of CNN-LSTM with the prediction system during use, making it difficult for the model to fully mine and utilize this spatial structural information. This not only leads to biases in the assessment of the production capacity contribution of different wells but also results in a lack of accuracy in well network optimization strategies, which in turn seriously affects overall production efficiency. Consequently, the model cannot provide solid and reliable support for production decisions, significantly increasing potential risks during the production process. Summary of the Invention

[0005] The purpose of this invention is to provide a geothermal extraction capacity optimization prediction system that integrates CNN-LSTM, in order to solve the problem mentioned in the background art that the inability to integrate CNN-LSTM with the prediction system makes it difficult for the model to fully explore and utilize these spatial structure information. This not only leads to deviations in the assessment of the capacity contribution of different wells, but also makes the well network optimization strategy lack accuracy, which seriously affects the overall extraction efficiency. It also makes the model unable to provide solid and reliable support for extraction decisions, and greatly increases the potential risks in the extraction process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a geothermal extraction capacity optimization prediction system integrating CNN-LSTM, comprising the following process modules: S1, data acquisition; S1-1, Geological exploration equipment; S1-2, Geothermal well monitoring equipment; S1-3, Heating system monitoring equipment; S2, Data Preprocessing; S2-1, Missing Value Processing Unit; S2-2, Outlier Detection Unit; S2-3, Normalized Unit; S3, Feature Filtering Module; S3-1, Correlation Analysis Unit; S4, Model Building Module; S4-1, CNN submodule; S4-2, LSTM module; S5, Hyperparameter Optimization Module; S5-1, NSGA-Optimizer; S6, Model Evaluation Module; S6-1, Error Calculation Unit; S6-2, Comparison and Verification Unit; S7, Real-time Prediction Module; S7-1, Dynamic Control Unit.

[0007] The aforementioned structural design utilizes various sensors to collect geological, engineering, and operational parameter data. This data is then transmitted in real-time to a data center via wired or wireless means, forming a raw dataset that provides the data foundation for subsequent modules. This real-time approach ensures the timeliness and effectiveness of the data, enabling subsequent modules to analyze and process information based on the latest data. This provides a solid foundation for accurate prediction and decision-making. Furthermore, correlation analysis using the Pearson correlation coefficient accurately identifies features closely related to geothermal performance. This targeted feature selection method reduces the model's input dimensions and computational complexity, while simultaneously increasing the model's focus on key factors, enhancing its predictive accuracy and stability. The CNN-LSTM model integrates spatial and temporal features, fully leveraging the advantages of both. CNN's extraction of reservoir spatial features allows the model to understand the spatial distribution patterns of geothermal reservoirs, while LSTM's capture of long-term time-series dependencies allows the model to grasp the dynamic trends during geothermal extraction. The combined end-to-end prediction model comprehensively and deeply characterizes the changing mechanisms of geothermal production capacity, achieving accurate predictions and providing a reliable basis for the optimized regulation of geothermal extraction.

[0008] Preferably, the S1 data acquisition includes real-time transmission of data to a data center via wired / wireless means to form a raw dataset, providing basic data support for subsequent modules.

[0009] The aforementioned structural design is the primary prerequisite for ensuring model reliability. It can be implemented from four dimensions: data timeliness, integrity, transmission reliability, and engineering adaptability. It provides the model with richer features, while the real-time, complete, and high-resolution raw dataset provides an unbiased foundation for subsequent S2 data preprocessing (cleaning and standardization) and S3 model training, avoiding the "garbage in, garbage out" problem. The hybrid transmission scheme controls the deployment cost to less than 50,000 yuan per well (fiber + LoRa), which is only 30% of the cost of the all-fiber scheme, while meeting the requirement of 99.9% transmission reliability.

[0010] Preferably, the S2-1 missing value processing unit includes a mean / median imputation algorithm, and the S2 data preprocessing includes cleaning and standardizing the original data, and dividing it into training set / test set to ensure the data quality of the input model.

[0011] The above structural design can systematically improve data quality, provide reliable input for model training, eliminate outliers that violate geological-mining theory, and preserve data trends through linear interpolation.

[0012] Preferably, the S3-1 correlation analysis unit includes Pearson correlation coefficient calculation for linear correlation between variables, and the S3 characteristic screening module includes...

[0013] The above structural design can achieve high efficiency, scientificity and engineering practicality in feature selection. The Pearson coefficient (r) accurately measures the strength of the linear correlation between two variables by the ratio of covariance to standard deviation, avoiding the bias of subjective experience judgment.

[0014] Preferably, the features that are strongly correlated with heat extraction performance (|r|>0.7) are used as model inputs, such as reservoir temperature, permeability, and injection-production well spacing.

[0015] The above structural design can eliminate redundant features, reduce the risk of overfitting, and enhance the expression of key physical mechanisms. Through feature importance analysis (such as SHAP value), the contribution of each feature to the prediction results can be quantified. It was found that the contribution of reservoir temperature accounts for more than 60%, which is consistent with geological theory.

[0016] Preferably, the S4 model building module includes concatenating CNN features and LSTM outputs, outputting predicted values ​​through a fully connected layer, and fusing spatial and temporal features to build an end-to-end prediction model, thereby achieving accurate prediction of geothermal production capacity.

[0017] In the above structural design, the CNN extracts spatial local patterns from geothermal monitoring data layer by layer through a sliding window of the convolution kernel (such as 3×3 or 5×5), avoiding overfitting caused by parameter redundancy in fully connected layers.

[0018] Preferably, the S5 hyperparameter optimization module includes adjusting the hyperparameters of the CNN-LSTM model through a multi-objective optimization algorithm to improve the overall performance of the model under metrics such as RMSE and MAE.

[0019] The above structural design allows the multi-objective optimization algorithm to simultaneously optimize both objectives through the Pareto front, finding a non-dominated solution set and avoiding the local optimum trap caused by traditional single-objective optimization.

[0020] Preferably, the S6-1 error calculation unit includes indicators such as RMSE, MAE, and LogRMSE to quantify the prediction deviation, and the S6-2 comparison and verification unit includes comparing the prediction results with actual field data to verify the effectiveness of the model.

[0021] The above structural design is sensitive to outliers and is suitable for evaluating the model's ability to capture extreme production fluctuations (such as geothermal reservoir water inrush and equipment failure). In geothermal extraction, RMSE can quickly locate the model's prediction deviation under extreme conditions such as high pressure / high temperature, and guide the optimization of the gating mechanism or attention weight allocation of CNN-LSTM.

[0022] Preferably, the S7 real-time prediction module includes a dynamic cycle of "prediction-control-feedback" to optimize geothermal resource extraction efficiency and extend the life of geothermal reservoirs.

[0023] The above structural design achieves precise optimization of geothermal resource extraction efficiency and significant extension of reservoir life. By using LSTM to model the long-term dependence of historical pressure data, the impact of groundwater level rise caused by seasonal precipitation on production capacity can be accurately predicted.

[0024] Preferably, the S7 real-time prediction module includes a prediction engine and a feedback mechanism, which periodically evaluates prediction deviations, updates model parameters or optimizes strategies, and forms a closed-loop control.

[0025] The aforementioned structural design not only improves the model's prediction accuracy but also achieves a better balance among different indicators, enhancing its stability and generalization ability. This allows the model to maintain good performance even in complex and ever-changing geothermal extraction environments. Furthermore, evaluation through multiple indicators provides a more comprehensive and accurate understanding of the model's predictive performance. The S6-2 comparison and verification unit directly verifies the model's effectiveness in practical applications by comparing the prediction results with actual field data. This data-driven verification method ensures that the model accurately reflects the actual situation of geothermal extraction, providing a reliable basis for decision-making. The S7 real-time prediction module implements a dynamic "prediction-control-feedback" cycle. The prediction engine provides a basis for extraction control through real-time predictions, while the feedback mechanism adjusts model parameters or optimizes strategies promptly based on prediction deviations. This closed-loop control can optimize geothermal resource extraction in real time according to actual conditions, continuously improving extraction efficiency, reducing resource waste, and extending the lifespan of geothermal reservoirs through reasonable regulation, thus achieving the sustainable development and utilization of geothermal resources.

[0026] Compared with existing technologies, the beneficial effects of this invention are: By adopting a novel structural design and evaluating multiple indicators, the predictive performance of the model can be understood more comprehensively and accurately, ensuring that the model can truly reflect the actual situation of geothermal extraction and providing a reliable basis for decision-making. Simultaneously, it can optimize geothermal resource extraction in real time according to actual conditions, continuously improving extraction efficiency and reducing resource waste. Furthermore, by rationally regulating and extending the lifespan of geothermal reservoirs, it achieves the sustainable development and utilization of geothermal resources. The specific details are as follows: This geothermal extraction capacity optimization and prediction system, which integrates CNN-LSTM, not only provides comprehensive data coverage, enabling the model to accurately capture the dynamic changes of the geothermal system, but also reduces redundant variables through feature selection. This achieves accurate prediction, dynamic optimization, and intelligent control of geothermal extraction capacity, providing high-precision, high-efficiency, and highly adaptable technical support for the scientific development of geothermal energy and promoting the development of geothermal energy towards a "predictable, controllable, and sustainable" direction.

[0027] Furthermore, it can effectively extract the spatial heterogeneity characteristics of reservoirs, which helps to understand the distribution pattern of geothermal energy underground and provides spatial dimension information support for production capacity prediction. This enables the model to accurately characterize the dynamic changes in the geothermal extraction process, such as predicting the impact of changes in injection and production volume on production capacity at different time stages.

[0028] This geothermal extraction capacity optimization prediction system, which integrates CNN-LSTM, uses a dynamic control unit to adjust parameters such as injection / extraction volume and reinjection temperature in real time based on the prediction results, forming a closed-loop management mechanism of "prediction-control-feedback". This allows the system to make timely adjustments based on the actual situation during the geothermal extraction process, ensuring that geothermal extraction is always in the optimal state.

[0029] Furthermore, it has achieved a 15% increase in resource utilization, effectively improved the efficiency of geothermal resource extraction, reduced resource waste, and extended the thermal storage life by 20%, which is conducive to the sustainable development of geothermal energy, ensures the efficient and stable operation of the system, and meets the energy needs of users.

[0030] (3) The geothermal extraction capacity optimization prediction system integrating CNN-LSTM provides sufficient information for the model. Compared with the traditional data collection method, the prediction accuracy is improved by more than 20%, thus laying a solid foundation for the accurate prediction of geothermal extraction capacity. At the same time, it can strictly control the data quality and improve the reliability and consistency of the data. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the system flow structure of the present invention.

[0032] Figure 2 This is a schematic diagram of the data preprocessing structure of the present invention.

[0033] Figure 3 This is a schematic diagram of the CNN-LSTM fusion prediction structure of the present invention.

[0034] Figure 4 This is a schematic diagram of the dual-layer optimization mechanism structure of the present invention. Detailed Implementation

[0035] 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.

[0036] This invention discloses the following technical solution: a geothermal extraction capacity optimization and prediction system integrating CNN-LSTM, comprising the following process modules: S1, data acquisition; S1-1, geological exploration equipment; S1-2, geothermal well monitoring equipment; S1-3, heating system monitoring equipment; S2, data preprocessing; S2-1, missing value processing unit; S2-2, outlier detection unit; S2-3, normalization unit; S3, feature selection module; S3-1, correlation analysis unit; S4, model building module; S4-1, CNN sub-module; S4-2, LSTM module; S5, hyperparameter optimization module; S5-1, NSGA-optimizer; S6, model evaluation module; S6-1, error calculation unit; S6-2, comparison and verification unit; S7, real-time prediction module; S7-1, dynamic control unit.

[0037] Staff utilized geological exploration, geothermal well monitoring, and heating system monitoring equipment to comprehensively collect geological, engineering, and operational parameters. To address missing values, outliers, and dimensional differences in the raw data, they employed various methods, including mean or median interpolation, Z-score or box plot methods, and Min-Max normalization. The dataset was also divided according to specific proportions to improve data quality and usability. Furthermore, Pearson correlation coefficients were used to calculate variable correlations, and strongly correlated features were selected based on set thresholds to reduce redundant information and improve model efficiency and prediction accuracy. Finally, the collected data was processed using a CNN submodule. By extracting reservoir spatial heterogeneity features through convolutional layers, the LSTM module captures long-term time-series dependencies through its network. The outputs of both are concatenated and processed by a fully connected layer to achieve end-to-end production capacity prediction through spatial-temporal feature fusion. Furthermore, constrained by actual reservoir conditions, and aiming to minimize error, maximize robustness, and balance computational efficiency, the NSGA-III optimizer is used to determine the optimal hyperparameter combination, improving model performance. Simultaneously, long-term future production capacity prediction is achieved based on historical data. Extraction parameters are dynamically adjusted according to the prediction results, and the model is optimized based on system response feedback, forming a closed loop to ensure efficient and stable operation of geothermal extraction (e.g., Figure 1 and Figure 2 (As shown).

[0038] S1 Data Acquisition includes real-time data transmission to the data center via wired / wireless methods to form a raw dataset, providing basic data support for subsequent modules. S2-1 Missing Value Processing Unit includes using mean / median interpolation algorithms. S2 Data Preprocessing includes cleaning and standardizing the raw data, dividing it into training and test sets to ensure the data quality of the input model. S3-1 Correlation Analysis Unit includes calculating the linear correlation between variables using the Pearson correlation coefficient. S3 Feature Screening Module includes screening features strongly correlated with geothermal performance (|r|>0.7) as model input, such as reservoir temperature, permeability, and injection-production well spacing. S4 Model Building Module includes concatenating CNN features and LSTM outputs, outputting predicted values ​​through a fully connected layer, and fusing spatial and temporal features to build an end-to-end prediction model, achieving accurate prediction of geothermal production capacity.

[0039] Various sensors are used to collect geological, engineering, and operational parameter data. This data is then transmitted in real-time to a data center via wired or wireless methods to form a raw dataset. A mean or median imputation algorithm is used to fill in missing values ​​in the raw dataset. Simultaneously, the raw data undergoes comprehensive cleaning to remove noise and errors, and standardization is performed to eliminate dimensional differences between different data features, making the data more comparable. The preprocessed data is divided into training and testing sets to prepare for model training and evaluation, ensuring the reliability of the input data. Data strongly correlated with heat extraction performance (|r|>) is selected. Features such as reservoir temperature, permeability, and injection-production well spacing (0.7) are used as model inputs to reduce redundant information and improve model efficiency. The fused features are processed through a fully connected layer to output predicted values, constructing an end-to-end prediction model for accurate geothermal production capacity prediction. This model not only collects multi-dimensional data covering geology, engineering, and operation, comprehensively reflecting the state of the geothermal extraction system, but also ensures data timeliness and effectiveness through real-time data transmission. This allows the system to analyze and predict based on the latest information, providing strong support for accurate decision-making. Furthermore, the CNN-LSTM model integrates spatial and temporal features, fully leveraging the advantages of both (e.g., Figure 3 As shown, CNN's extraction of reservoir spatial features enables the model to understand the spatial distribution patterns of geothermal reservoirs, while LSTM's capture of long-term time series dependencies allows the model to grasp the dynamic trends during geothermal extraction. The end-to-end prediction model constructed by combining the two can more comprehensively and deeply depict the changing mechanism of geothermal production capacity, achieve accurate prediction, and provide a reliable basis for the optimized regulation of geothermal extraction.

[0040] The S5 hyperparameter optimization module adjusts the hyperparameters of the CNN-LSTM model through a multi-objective optimization algorithm to improve the model's overall performance under metrics such as RMSE and MAE. The S6-1 error calculation unit quantifies prediction deviations using metrics such as RMSE, MAE, and LogRMSE. The S6-2 comparison and verification unit compares the prediction results with actual field data to verify the model's effectiveness. The S7 real-time prediction module implements a dynamic cycle of "prediction-control-feedback" to optimize geothermal resource extraction efficiency and extend the lifespan of geothermal reservoirs. The S7 real-time prediction module includes a prediction engine and a feedback mechanism to periodically evaluate prediction deviations, update model parameters or optimization strategies, and form a closed-loop control.

[0041] By employing a multi-objective optimization algorithm to adjust the hyperparameters of the CNN-LSTM model, the goal is to improve the model's overall performance under metrics such as RMSE (Root Mean Square Error) and MAE (Mean Absolute Error), finding the optimal combination of hyperparameters to achieve a better balance between accuracy and stability. Furthermore, the error calculation unit uses metrics such as RMSE, MAE, and LogRMSE to quantify the deviation between the model's predicted values ​​and actual values, measuring the accuracy of the model's predictions from different perspectives. Simultaneously, the comparison and validation unit compares the model's prediction results with actual field data to verify the model's effectiveness and determine whether the model can accurately reflect the actual situation. The S7 real-time prediction module (such as...) Figure 4 As shown, the prediction engine uses an optimized and evaluated model to predict geothermal production capacity in real time. Based on the prediction results, it adjusts geothermal resource extraction parameters, such as injection volume, extraction volume, and reinjection temperature. The feedback mechanism periodically evaluates prediction deviations and updates model parameters or optimizes extraction strategies based on the deviations, forming a dynamic cycle of "prediction-regulation-feedback". This continuously optimizes geothermal resource extraction efficiency and extends reservoir life. This not only improves the model's prediction accuracy but also achieves a better balance among different indicators, enhancing the model's stability and generalization ability. It enables the model to maintain good performance in complex and ever-changing geothermal extraction environments. Furthermore, the closed-loop regulation can optimize geothermal resource extraction in real time according to actual conditions, continuously improving extraction efficiency, reducing resource waste, and extending reservoir life through reasonable regulation, thus achieving the sustainable development and utilization of geothermal resources.

[0042] The above is the entire working process of the device, and all contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A geothermal extraction capacity optimization prediction system integrating CNN-LSTM, comprising the following process modules: S1, data acquisition; S1-1, Geological exploration equipment; S1-2, Geothermal well monitoring equipment; S1-3, Heating system monitoring equipment; S2, Data Preprocessing; S2-1, Missing Value Processing Unit; S2-2, Outlier Detection Unit; S2-3, Normalized Unit; S3, Feature Filtering Module; S3-1, Correlation Analysis Unit; S4, Model Building Module; S4-1, CNN submodule; S4-2, LSTM module; S5, Hyperparameter Optimization Module; S5-1, NSGA-Optimizer; S6, Model Evaluation Module; S6-1, Error Calculation Unit; S6-2, Comparison and Verification Unit; S7, Real-time Prediction Module; S7-1, Dynamic Control Unit.

2. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 1, characterized in that: The S1 data acquisition includes transmitting data to the data center in real time via wired / wireless means to form a raw dataset, providing basic data support for subsequent modules.

3. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 2, characterized in that: The S2-1 missing value processing unit includes a mean / median imputation algorithm, and the S2 data preprocessing includes cleaning and standardizing the original data, dividing it into training set / test set to ensure the data quality input to the model.

4. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 3, characterized in that: The S3-1 correlation analysis unit includes the Pearson correlation coefficient to calculate the linear correlation between variables, and the S3 characteristic screening module includes...

5. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 4, characterized in that: The features that are strongly correlated with heat extraction performance (|r|>0.7) are used as model inputs, such as reservoir temperature, permeability, and injection-production well spacing.

6. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 5, characterized in that: The S4 model building module includes concatenating CNN features and LSTM outputs, outputting predicted values ​​through a fully connected layer, and fusing spatial and temporal features to build an end-to-end prediction model, thereby achieving accurate prediction of geothermal production capacity.

7. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 6, characterized in that: The S5 hyperparameter optimization module includes adjusting the hyperparameters of the CNN-LSTM model through a multi-objective optimization algorithm to improve the model's overall performance under metrics such as RMSE and MAE.

8. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 7, characterized in that: The S6-1 error calculation unit includes indicators such as RMSE, MAE, and LogRMSE to quantify the prediction deviation. The S6-2 comparison and verification unit includes comparing the prediction results with actual field data to verify the effectiveness of the model.

9. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 8, characterized in that: The S7 real-time prediction module enables a dynamic cycle of "prediction-control-feedback" to optimize geothermal resource extraction efficiency and extend the lifespan of geothermal reservoirs.

10. The geothermal extraction capacity optimization and prediction system fused with CNN-LSTM according to claim 9, characterized in that: The S7 real-time prediction module includes a prediction engine and a feedback mechanism, which periodically evaluates prediction deviations, updates model parameters or optimizes strategies, and forms a closed-loop control.

Images