A geothermal exploitation dynamic optimization method, device, equipment and medium

Abstract

This application discloses a dynamic optimization method, apparatus, equipment, and medium for geothermal extraction, relating to the field of engineering optimization and control. The method includes: inputting target input data for the heating season of the target area into a preset heat-fluid coupling numerical model; training the preset hybrid model using a target training dataset generated based on the target input data and the obtained simulated injection-production pressure difference data to obtain a target hybrid model; extracting data features of the target input data through long short-term memory network branches and time-series convolutional network branches of the target hybrid model to predict the injection-production pressure difference based on these features; determining whether the model performance of the target hybrid model meets preset performance conditions based on the predicted target injection-production pressure difference; and optimizing the reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model. Therefore, large-scale modeling and multi-objective optimization can replace numerical simulation, thereby improving the efficiency of reinjection parameter decision-making in geothermal heating.

Description

Technical Field

[0001] This invention relates to the field of engineering optimization control, and in particular to a dynamic optimization method, apparatus, equipment and medium for geothermal extraction. Background Technology

[0002] Geothermal energy, as a clean and stable renewable energy source, plays an important role in the global energy transition. To ensure the long-term sustainability and economic feasibility of the exploitation of medium-deep geothermal systems, it is essential to implement precise and real-time dynamic control of reinjection parameters (such as reinjection temperature and reinjection flow rate) to prevent thermal breakthroughs in production wells and minimize the energy consumption of water pumps.

[0003] Traditional physics-based numerical simulation methods, while highly accurate, are often limited by complex geological grid divisions and massive computational demands, requiring hours or even days of computation time. Currently, transforming dynamic production forecasting into a time series forecasting problem and utilizing deep learning to construct surrogate models has become an effective way to overcome these computational bottlenecks. However, this approach suffers from limitations: a single model may not meet forecasting requirements, hybrid networks suffer from feature dilution, and current optimizations are often limited to a single objective, resulting in reinjection strategies that lack comprehensive economic benefits and thermal stability guarantees in practical engineering. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a dynamic optimization method, apparatus, equipment, and medium for geothermal extraction, which can replace numerical simulation with large-scale modeling and multi-objective optimization, thereby improving the efficiency of reinjection parameter decision-making in geothermal heating. The specific solution is as follows: In a first aspect, this application discloses a dynamic optimization method for geothermal extraction, including: The target input data for the heating season in the target area is input into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and a target training dataset is generated through the target input data and the simulated injection-production pressure difference data; the target input data consists of real-time meteorological temperature data and preset field reinjection parameters for the heating season in the target area. The preset hybrid model is trained using the target training dataset to obtain the target hybrid model. The data features of the target input data are extracted through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model. Based on the data features, the injection-production pressure difference is predicted to obtain the target predicted injection-production pressure difference. Based on the target predicted injection-production pressure difference, determine whether the model performance of the target hybrid model meets the preset performance conditions; If the target hybrid model meets the preset performance conditions, the system reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model; the system reinjection parameters include reinjection flow rate and reinjection temperature.

[0005] Optionally, the step of inputting the target input data for the heating season of the target area into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and generating a target training dataset using the target input data and the simulated injection-production pressure difference data, includes: Real-time meteorological temperature data of the target area during the heating season is collected, and the real-time meteorological temperature data and preset field reinjection parameters are used as input boundary conditions to input the preset heat flow coupling numerical model to generate simulated injection-production pressure difference data. The real-time meteorological temperature data, the preset field reinjection parameters, and the simulated injection-production pressure difference data are subjected to standardization processing to obtain processed data, and a target training dataset is constructed based on the processed data.

[0006] Optionally, the step of training a preset hybrid model using the target training dataset to obtain a target hybrid model, and extracting data features of the target input data through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, to predict the injection-production pressure difference based on the data features, and obtaining the target predicted injection-production pressure difference, includes: The preset hybrid model is trained based on the target training dataset, preset training rounds, and preset training parameters to obtain the target hybrid model; The target input data is input into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and perform injection-production pressure difference prediction based on the data features to obtain the target predicted injection-production pressure difference.

[0007] Optionally, the step of inputting the target input data into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and to predict the injection-import pressure difference based on the data features to obtain the target predicted injection-import pressure difference, includes: The target input data is respectively input into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model; The first data feature of the target input data is extracted through the forget gate, input gate and output gate in the branch of the long short-term memory network; The second data feature of the target input data is extracted by the causal dilated convolution branch of the temporal convolutional network and the residual block; The first data feature and the second data feature are fused by the feature fusion layer of the target hybrid model to obtain fused features. The injection-production pressure difference is then predicted by the fully connected layer of the target hybrid model based on the fused features to obtain the target predicted injection-production pressure difference.

[0008] Optionally, determining whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference includes: The model error of the target hybrid model is determined based on the target predicted injection-production pressure difference, and it is determined whether the model error is less than a preset error threshold. If the model error is less than the preset error threshold, then the model performance of the target hybrid model is determined to meet the preset performance conditions.

[0009] Optionally, optimizing the reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model includes: The target fitness function is set based on minimizing the injection-production pressure difference, minimizing the thermal power difference, and maximizing the net revenue of the system in the preset medium-deep geothermal system. The target fitness function is solved by the target hybrid model based on preset constraints to obtain the corresponding target optimal solution set; The reinjection parameters of the preset medium-deep geothermal system are optimized based on the target optimal solution set.

[0010] Optionally, the step of solving the target fitness function based on preset constraints using the target hybrid model to obtain the corresponding target optimal solution set includes: By using a pre-set multi-objective genetic algorithm during population iteration, the target hybrid model is invoked to solve the target fitness function based on pre-set constraints, so as to obtain several solutions; By performing non-dominated sorting, crowding calculation, and crossover mutation on the solutions, several target optimal solutions are selected, and a target optimal solution set is generated based on the solutions.

[0011] Secondly, this application discloses a dynamic optimization device for geothermal extraction, comprising: The dataset generation module is used to input the target input data of the heating season in the target area into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and to generate a target training dataset through the target input data and the simulated injection-production pressure difference data; the target input data is the real-time meteorological temperature data of the heating season in the target area and preset field reinjection parameters; The data prediction module is used to train a preset hybrid model using the target training dataset to obtain a target hybrid model, and to extract data features of the target input data through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, so as to predict the injection-production pressure difference based on the data features and obtain the target predicted injection-production pressure difference. The condition judgment module is used to determine whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference; The parameter optimization module is used to optimize the system reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model if the target hybrid model meets the preset performance conditions; the system reinjection parameters include reinjection flow rate and reinjection temperature.

[0012] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned dynamic optimization method for geothermal extraction.

[0013] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned dynamic optimization method for geothermal extraction.

[0014] In this application, target input data for the heating season in the target area can be input into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data. A target training dataset is then generated using the target input data and the simulated injection-production pressure difference data. The target input data consists of real-time meteorological temperature data for the heating season in the target area and preset field reinjection parameters. A preset hybrid model is trained using the target training dataset to obtain a target hybrid model. Data features of the target input data are extracted using the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model. Injection-production pressure difference prediction is then performed based on these features to obtain the target predicted injection-production pressure difference. Based on the target predicted injection-production pressure difference, it is determined whether the model performance of the target hybrid model meets preset performance conditions. If the target hybrid model meets the preset performance conditions, the system reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model. The system reinjection parameters include reinjection flow rate and reinjection temperature.

[0015] Therefore, the method of this application requires generating simulated injection-production pressure difference data based on the target input data of the target region's heating season using a preset heat-fluid coupling numerical model. Then, a target training dataset is generated using the target input data and the simulated injection-production pressure difference data. The preset hybrid model is then trained using the target training dataset so that the obtained target hybrid model has the accuracy and processing efficiency to adapt to geothermal heating optimization. Furthermore, the data features of the target input data need to be extracted through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model to predict the injection-production pressure difference based on the data features, thereby obtaining the target predicted injection-production pressure difference. Then, the model accuracy is determined by the target predicted injection-production pressure difference. If the accuracy is satisfied, the reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model. In this way, on the one hand, the feature dilution problem in the traditional serial hybrid model can be avoided by using a target hybrid model composed of a long short-term memory network and a temporal convolutional network. The parallel branch can retain the sensitivity of the temporal convolutional network to the extreme high-frequency peaks of the injection-production pressure difference, and can also retain the global memory of the long short-term memory network for long-term seasonal thermal decay, so as to effectively improve the prediction accuracy and processing efficiency. On the other hand, by performing multi-objective optimization on the reinjection flow rate and reinjection temperature in the system reinjection parameters, the optimization effect can be guaranteed. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a flowchart of a dynamic optimization method for geothermal extraction disclosed in this application; Figure 2 This application discloses a flowchart of a dynamic optimization process for geothermal extraction. Figure 3 This is a schematic diagram of a target hybrid model structure disclosed in this application; Figure 4 This application discloses a specific dynamic optimization method for geothermal extraction, which is shown in the flowchart below. Figure 5 This is a schematic diagram of the structure of a geothermal extraction dynamic optimization device disclosed in this application; Figure 6 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0018] 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.

[0019] Currently, while traditional physics-based numerical simulation methods offer high accuracy for geothermal heating optimization, they are limited by complex geological grid divisions and massive computational demands, often requiring hours or even days of computation time. Furthermore, surrogate model methods suffer from limitations: single models cannot meet prediction requirements, hybrid networks suffer from feature dilution, and current optimizations are often limited to a single objective, resulting in reinjection strategies that lack comprehensive economic benefits and thermal stability guarantees in practical engineering.

[0020] To overcome the aforementioned technical problems, this application discloses a dynamic optimization method, apparatus, equipment, and medium for geothermal extraction, which can replace numerical simulation with large-scale modeling and multi-objective optimization, thereby improving the efficiency of decision-making on reinjection parameters in geothermal heating.

[0021] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a dynamic optimization method for geothermal extraction, including: Step S11: Input the target input data of the heating season in the target area into the preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and generate a target training dataset through the target input data and the simulated injection-production pressure difference data; the target input data is the real-time meteorological temperature data of the heating season in the target area and the preset field reinjection parameters.

[0022] In this embodiment, a dataset needs to be constructed for training the preset hybrid model. Specifically, such as... Figure 2As shown, real-time meteorological temperature data of the target area during the heating season needs to be collected. This real-time meteorological temperature data, along with preset field reinjection parameters, are then input as boundary conditions into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data. It should be noted that the preset heat-fluid coupling numerical model is a three-dimensional heat-fluid (TH) coupled finite element numerical model of a medium-deep geothermal dual-well system. The reinjection parameters, including reinjection flow rate and reinjection temperature, require the actual ambient meteorological temperature of the target area during historical heating seasons and the field reinjection parameters (reinjection flow rate and reinjection temperature) as input boundary conditions. Through finite element transient solution, the injection-production pressure difference (IPPD) required to maintain system operation is simulated and calculated to obtain the simulated injection-production pressure difference data. Taking a specific scenario as an example, a three-dimensional thermal-fluid coupled finite element model needs to be established using COMSOL Multiphysics. The model's spatial dimensions are set to 2500m × 2500m × 2000m, where the depth range of 500m to 1500m is defined as the geothermal reservoir (carbonate rock), and the overlying 500m caprock and the underlying 500m basement (mudstone) are set as impermeable and thermally insulating boundaries. The model includes one injection well and one production well, with a well spacing of 500m. The mesh and boundary conditions are set as follows: reservoir top temperature is set to 45℃, pore pressure is set to 9.8MPa, geothermal gradient is set to 0.03℃ / m, and pressure gradient is set to 9800Pa / m. Local mesh refinement is applied to the area around the wellbore, with a total mesh element count of approximately 260,000 and a time step of 1 hour.

[0023] Furthermore, it is necessary to standardize the real-time meteorological temperature data, preset field reinjection parameters, and simulated injection-production pressure difference data to obtain processed data, and then construct the target training dataset based on the processed data. Taking a real-world example, real-time meteorological temperature data for the heating season in region A from 2019 to 2025 is collected. Combined with field reinjection flow and reinjection temperature as input features, the corresponding simulated injection-production pressure difference data (IPPD) is obtained through simulation calculation using a preset heat-flow coupling numerical model, generating a total of 17,337 hours of time series data. During standardization, the maximum-minimum normalization method is used to uniformly scale all feature data to the interval [-1, 1]. The heating season data from 2019 to 2024 (14,444 samples) is divided into a training set, and the heating season data from 2024 to 2025 (2,893 samples) is divided into a test set. The target training dataset is composed of the training set and the test set.

[0024] Step S12: Train the preset hybrid model using the target training dataset to obtain the target hybrid model, and extract the data features of the target input data through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, so as to predict the injection-production pressure difference based on the data features and obtain the target predicted injection-production pressure difference.

[0025] In this embodiment, a target hybrid model needs to be trained using the target training dataset, and then the injection-sampling pressure difference is predicted using the trained target hybrid model. Specifically, model training is first required. The preset hybrid model needs to be trained based on the target training dataset, a preset training epoch, and preset training parameters to obtain the target hybrid model. The preset hybrid model is a parallel feature fusion PC-LSTM-TCN deep learning proxy model, where LSTM (Long Short-Term Memory) refers to a long short-term memory network, and TCN (Temporal Convolutional Network) refers to a temporal convolutional network. The structure of the preset hybrid model is as follows: Figure 3 As shown, the input sequence length (time step) is set to 24; the LSTM sub-network branch is set to 1 layer with 64 hidden nodes to extract the global long-period decay trend; the TCN sub-network branch contains 3 residual blocks, with a dilation coefficient of 3, a kernel size of 3, and the number of convolution channels set to [64, 32, 32] to extract local high-frequency mutation features. The output features of the two branches are concatenated through a feature fusion layer and then output as the prediction result through a fully connected layer.

[0026] Furthermore, after obtaining the target hybrid model, the target input data needs to be input into the Long Short-Term Memory (LSTM) network branch and the Temporal Convolutional Network (TCNN) branch of the target hybrid model, respectively. This allows for the extraction of features from the target input data through the LSM and TCNN branches, and the injection-import pressure difference is then predicted based on these features to obtain the target predicted injection-import pressure difference. Specifically, as follows... Figure 3 As shown, the target hybrid model structure requires feature extraction from the input target data through both a Long Short-Term Memory (LTC) network branch and a Temporal Convolutional Network (TCN) branch. Then, the feature vectors extracted from the two branches are concatenated in the fusion layer, and finally, the predicted target injection-sampling pressure difference is output in the fully connected layer. This parallel network structure in the target hybrid model retains both the TCN's sensitivity to extreme high-frequency peaks in the injection-sampling pressure difference and the LSTM's global memory for long-period seasonal thermal decay, effectively improving prediction accuracy. Furthermore, since the feature extraction processes of the LSTM and TCN parts are computed in parallel and concurrently, processing efficiency is significantly improved.

[0027] Step S13: Determine whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference.

[0028] In this embodiment, as Figure 2 As shown, model evaluation is required. Specifically, the model error of the target hybrid model needs to be determined based on the predicted injection-production pressure difference, and it needs to be confirmed whether the model error is less than a preset error threshold. It should be noted that the error between the predicted injection-production pressure difference and the set actual injection-production pressure difference is considered valid. If the model error is less than the preset error threshold, then the model performance of the target hybrid model is determined to meet the preset performance conditions. In this case, the model accuracy has met the requirements, and parameter optimization can be performed.

[0029] Step S14: If the target hybrid model meets the preset performance conditions, then optimize the system reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model; the system reinjection parameters include reinjection flow rate and reinjection temperature.

[0030] In this embodiment, optimization is required by pre-setting the system reinjection parameters of the medium-deep geothermal system using the target hybrid model. Specifically, such as... Figure 2 As shown, a target fitness function needs to be set based on minimizing the injection-production pressure difference, minimizing the thermal power difference, and maximizing the net system benefit for a pre-defined medium-deep geothermal system. Specifically, the objective hybrid model needs to be embedded into a non-dominated sorting genetic algorithm (NSGA-II) for multi-objective reinjection strategy optimization. The target fitness function needs to be set based on minimizing the injection-production pressure difference, minimizing the thermal power difference, and maximizing the net system benefit. Minimizing the injection-production pressure difference (… This value, calculated by the target hybrid model, is used to reduce pump energy consumption. Minimize the thermal power difference ( This is used to ensure heating stability, and the calculation formula is: ; in, (MW) represents the actual thermal power produced by the geothermal system, which is based on the reinjection flow rate. A function for dynamically calculating the supply and return water temperature difference; (MW) represents the actual heat demand of the building, which is based on the building area and real-time weather temperature. (A dynamically calculated function.)

[0031] Specifically, the calculation of actual production thermal power takes into account fluid properties and operating parameters: ; In the formula, The density of water ( ); The specific heat capacity of water ( ); For the recharge flow rate ( ); For reservoir production temperature ( ); Recharge temperature ( The actual heat demand of a building takes into account changes in ambient temperature. ; In the formula, For the total area of ​​the building to be heated ( ); The heat loss coefficient per unit area of ​​the foundation ( ); The heat loss coefficient per unit temperature ( ); Standard indoor heating temperature ( ); Real-time ambient meteorological temperature ( ).

[0032] Maximize system net profit ( Taking into account economic indicators such as heat sales revenue and water pump power consumption costs, the calculation formula is as follows: ; in, (RMB) represents the total revenue from heat sales; (RMB) represents the operating cost. The specific calculation is as follows: ; ; in, The industrial electricity price for pump operation is 0.725 RMB / (kW·h) in this embodiment. (J) The mechanical energy consumed by the water pump required to overcome the injection-production pressure difference (IPPD); The working efficiency of the water pump (0.75 in this embodiment); The unit price for geothermal heating (in this example, it is 0.18 RMB / (kW·h)); This refers to the system uptime.

[0033] Then, the objective fitness function needs to be solved using a target hybrid model based on preset constraints to obtain the corresponding optimal solution set. It should be noted that the preset constraints are engineering constraints: considering the safety of geothermal equipment and prevention of ground cooling damage, the reinjection temperature is set... Reinjection flow rate between 15 and 20℃ Between 10 and 140 m 3The heat supply must be between / h, and the heat power difference ΔW ≥ 0 must be enforced (i.e., ensuring that the heat supply is greater than or equal to the demand, and not allowing insufficient heating). It should be noted that a pre-defined multi-objective genetic algorithm is needed to call the objective hybrid model during population iteration to solve the objective fitness function based on the pre-defined constraints, obtaining several solutions. By performing non-dominated sorting, crowding calculation, and crossover mutation on these solutions, several objective optimal solutions are selected, and an objective optimal solution set is generated based on these optimal solutions. Specifically, the pre-defined multi-objective genetic algorithm, namely the NSGA-II algorithm, needs to call the ultra-fast response objective hybrid model to calculate the fitness value in each population iteration. Through non-dominated sorting, crowding calculation, and crossover mutation, the Pareto optimal solution set is finally output, obtaining the optimal reinjection flow and reinjection temperature dynamic adjustment strategy that changes with real-time air temperature. Furthermore, the NSGA-II population size needs to be set to 200, the maximum number of iterations to 20, the crossover probability to 0.9, and the mutation probability to 0.1. During the iteration process, the algorithm frequently calls the surrogate model to generate the Pareto front. Finally, the reinjection parameters of the preset medium-deep geothermal system are optimized based on the target optimal solution set. That is, the optimal reinjection flow rate and reinjection temperature are determined and applied using the parameters in the target optimal solution set.

[0034] In this embodiment, a simulated injection-production pressure difference (EPP) data is generated based on the target input data of the target region's heating season using a preset heat-fluid coupling numerical model. Then, a target training dataset is generated using the target input data and the simulated EPP data. The preset hybrid model is then trained using the target training dataset to ensure that the obtained target hybrid model has the accuracy and processing efficiency to adapt to geothermal heating optimization. Furthermore, data features of the target input data are extracted using the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model. Based on these data features, EPP prediction is performed to obtain the target predicted EPP. The model accuracy is then determined using the target predicted EPP. If the accuracy is satisfactory, the reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model. In this way, on the one hand, the feature dilution problem in the traditional serial hybrid model can be avoided by using a target hybrid model composed of a long short-term memory network and a temporal convolutional network. The parallel branch can retain the sensitivity of the temporal convolutional network to the extreme high-frequency peaks of the injection-production pressure difference, and can also retain the global memory of the long short-term memory network for long-term seasonal thermal decay, so as to effectively improve the prediction accuracy and processing efficiency. On the other hand, by performing multi-objective optimization on the reinjection flow rate and reinjection temperature in the system reinjection parameters, the optimization effect can be guaranteed.

[0035] As can be seen from the foregoing embodiments, this application requires training a preset hybrid model, and then using the obtained target hybrid model to predict the injection-production pressure difference. Therefore, this embodiment provides a detailed explanation of how to train the model and how to predict the injection-production pressure difference. See [link to documentation]. Figure 4 As shown in the figure, an embodiment of the present invention discloses a dynamic optimization method for geothermal extraction, including: Step S21: Input the target input data of the heating season in the target area into the preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and generate a target training dataset through the target input data and the simulated injection-production pressure difference data; the target input data is the real-time meteorological temperature data of the heating season in the target area and the preset field reinjection parameters.

[0036] Step S22: Train the preset hybrid model based on the target training dataset, preset training rounds, and preset training parameters to obtain the target hybrid model.

[0037] In this embodiment, a target hybrid model needs to be trained using the target training dataset, and then the injection-sampling pressure difference is predicted using the trained target hybrid model. Specifically, model training is first required. The preset hybrid model needs to be trained based on the target training dataset, a preset training epoch, and preset training parameters to obtain the target hybrid model. The preset hybrid model is a parallel feature fusion PC-LSTM-TCN deep learning proxy model, where LSTM (Long Short-Term Memory) refers to a long short-term memory network, and TCN (Temporal Convolutional Network) refers to a temporal convolutional network. The structure of the preset hybrid model is as follows: Figure 3 As shown, the input sequence length (time step) is set to 24; the LSTM sub-network branch is set to 1 layer with 64 hidden nodes to extract the global long-period decay trend; the TCN sub-network branch contains 3 residual blocks, with a dilation coefficient of 3, a kernel size of 3, and the number of convolution channels set to [64, 32, 32] to extract local high-frequency mutation features. The output features of the two branches are concatenated through a feature fusion layer and then output as the prediction result through a fully connected layer.

[0038] It should be noted that the preset training epochs are 100 epochs, with a Dropout rate of 0.2 and an initial learning rate of 0.001. The model needs to be trained on the training set for 100 epochs, based on the set Dropout rate and initial learning rate. To improve the model's robustness to extreme peaks and troughs in geothermal data, the optimal combination of training hyperparameters needs to be determined through grid search. The batch size is set to 32, the loss function is Mean Absolute Error (MAE Loss), and the optimizer is the RMSprop algorithm. This training method aims to eliminate feature dilution, thereby improving model performance.

[0039] Step S23: Input the target input data into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and perform injection-production pressure difference prediction based on the data features to obtain the target predicted injection-production pressure difference.

[0040] In this embodiment, as Figure 3 As shown, the target input data needs to be fed into the Long Short-Term Memory (LSTM) network branch and the Temporal Convolutional Network branch of the target hybrid model, respectively. Then, the first data feature of the target input data is extracted through the forget gate, input gate, and output gate in the LSTM network branch; the second data feature is extracted through causal dilation convolution and residual blocks in the Temporal Convolutional Network branch. It should be noted that the forget gate, input gate, and output gate mechanism in the LSTM network is used to extract long-term global time-dependent trends in geothermal operation data, such as the seasonal inertia of formation heat decay. Simultaneously, the causal dilation convolution and residual block structure in the TCN network are used to extract local high-frequency abrupt changes caused by instantaneous changes in operating parameters, such as transient pressure shocks caused by valve regulation.

[0041] Finally, the first data feature and the second data feature need to be fused through the feature fusion layer of the target hybrid model to obtain the fused feature. Then, the injection-production pressure difference is predicted based on the fused feature through the fully connected layer of the target hybrid model to obtain the target predicted injection-production pressure difference.

[0042] Step S24: Determine whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference.

[0043] Step S25: If the target hybrid model meets the preset performance conditions, then optimize the system reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model; the system reinjection parameters include reinjection flow rate and reinjection temperature.

[0044] It should be noted that the specific contents of steps S21, S24 and S25 above can be referred to the foregoing embodiments, and will not be repeated here.

[0045] Therefore, the method described in this application requires training a pre-defined hybrid model based on the target training dataset, a pre-defined training epoch, and pre-defined training parameters to obtain the target hybrid model. Then, the target input data is fed into the Long Short-Term Memory (LSTM) branch and the Temporal Convolutional Network (TCN) branch of the target hybrid model, respectively. The LSM and TCN branches extract features from the target input data, and injection-sampling pressure difference is predicted based on these features to obtain the target predicted injection-sampling pressure difference. In this way, on the one hand, prediction accuracy can be guaranteed through model training; on the other hand, the parallel processing of LSTM and TCN can ensure the quality of features while improving processing efficiency and prediction accuracy.

[0046] See Figure 5 As shown, an embodiment of the present invention discloses a dynamic optimization device for geothermal extraction, comprising: The dataset generation module 11 is used to input the target input data of the heating season in the target area into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and to generate a target training dataset through the target input data and the simulated injection-production pressure difference data; the target input data is the real-time meteorological temperature data of the heating season in the target area and preset field reinjection parameters; The data prediction module 12 is used to train a preset hybrid model using the target training dataset to obtain a target hybrid model, and to extract data features of the target input data through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, so as to predict the injection-production pressure difference based on the data features and obtain the target predicted injection-production pressure difference. Condition judgment module 13 is used to determine whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference; The parameter optimization module 14 is used to optimize the system reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model if the target hybrid model meets the preset performance conditions; the system reinjection parameters include reinjection flow rate and reinjection temperature.

[0047] In this embodiment, a simulated injection-production pressure difference (EPP) data is generated based on the target input data of the target region's heating season using a preset heat-fluid coupling numerical model. Then, a target training dataset is generated using the target input data and the simulated EPP data. The preset hybrid model is then trained using the target training dataset to ensure that the obtained target hybrid model has the accuracy and processing efficiency to adapt to geothermal heating optimization. Furthermore, data features of the target input data are extracted using the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model. Based on these data features, EPP prediction is performed to obtain the target predicted EPP. The model accuracy is then determined using the target predicted EPP. If the accuracy is satisfactory, the reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model. In this way, on the one hand, the feature dilution problem in the traditional serial hybrid model can be avoided by using a target hybrid model composed of a long short-term memory network and a temporal convolutional network. The parallel branch can retain the sensitivity of the temporal convolutional network to the extreme high-frequency peaks of the injection-production pressure difference, and can also retain the global memory of the long short-term memory network for long-term seasonal thermal decay, so as to effectively improve the prediction accuracy and processing efficiency. On the other hand, by performing multi-objective optimization on the reinjection flow rate and reinjection temperature in the system reinjection parameters, the optimization effect can be guaranteed.

[0048] In some embodiments, the dataset generation module 11 may specifically include: The simulation data generation unit is used to collect real-time meteorological temperature data of the target area during the heating season, and input the real-time meteorological temperature data and preset field reinjection parameters as input boundary conditions into the preset heat flow coupling numerical model to generate simulated injection-production pressure difference data. The dataset generation unit is used to perform standardization processing on the real-time meteorological temperature data, the preset field reinjection parameters, and the simulated injection-production pressure difference data to obtain processed data, and to construct a target training dataset based on the processed data.

[0049] In some embodiments, the data prediction module 12 may specifically include: The model training submodule is used to train the preset hybrid model based on the target training dataset, preset training rounds, and preset training parameters to obtain the target hybrid model. The data prediction submodule is used to input the target input data into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and perform injection-production pressure difference prediction based on the data features to obtain the target predicted injection-production pressure difference.

[0050] In some embodiments, the data prediction submodule may specifically include: The data input unit is used to input the target input data into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, respectively. The first feature extraction unit is used to extract the first data features of the target input data through the forget gate, input gate and output gate in the long short-term memory network branch; The second feature extraction unit is used to extract the second data features of the target input data through the causal dilated convolution branch of the temporal convolutional network and the residual block; The data prediction unit is used to perform feature fusion on the first data feature and the second data feature through the feature fusion layer of the target hybrid model to obtain fused features, and to perform injection-production pressure difference prediction based on the fused features through the fully connected layer of the target hybrid model to obtain the target predicted injection-production pressure difference.

[0051] In some embodiments, the condition judgment module 13 may specifically include: The data comparison unit is used to determine the model error of the target hybrid model based on the target predicted injection-production pressure difference, and to determine whether the model error is less than a preset error threshold. The condition judgment unit is used to determine that the model performance of the target hybrid model meets the preset performance condition if the model error is less than the preset error threshold.

[0052] In some embodiments, the parameter optimization module 14 may specifically include: The function construction submodule is used to set the target fitness function based on the preset medium-deep geothermal system, which minimizes the injection-production pressure difference, minimizes the thermal power difference, and maximizes the system net benefit. The solution set determination submodule is used to solve the target fitness function based on preset constraints using the target hybrid model to obtain the corresponding target optimal solution set; The parameter optimization submodule is used to optimize the reinjection parameters of the preset medium-deep geothermal system based on the target optimal solution set.

[0053] In some embodiments, the solution set determination submodule may further include: The function solving unit is used to call the target hybrid model in the population iteration using a preset multi-objective genetic algorithm to solve the target fitness function based on preset constraints, so as to obtain several solutions; The solution set generation unit is used to select several target optimal solutions by performing non-dominated sorting, crowding calculation and crossover mutation on the several solutions, and generate a target optimal solution set based on the several target optimal solutions.

[0054] Furthermore, embodiments of this application also disclose an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0055] Figure 6 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the dynamic optimization method for geothermal extraction disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0056] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0057] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0058] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the geothermal extraction dynamic optimization method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0059] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned dynamic optimization method for geothermal extraction. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0061] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0062] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0063] Finally, 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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0064] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A dynamic optimization method for geothermal extraction, characterized in that, include: The target input data for the heating season in the target area is input into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and a target training dataset is generated through the target input data and the simulated injection-production pressure difference data; the target input data consists of real-time meteorological temperature data and preset field reinjection parameters for the heating season in the target area. The preset hybrid model is trained using the target training dataset to obtain the target hybrid model. The data features of the target input data are extracted through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model. Based on the data features, the injection-production pressure difference is predicted to obtain the target predicted injection-production pressure difference. Based on the target predicted injection-production pressure difference, determine whether the model performance of the target hybrid model meets the preset performance conditions; If the target hybrid model meets the preset performance conditions, the system reinjection parameters of the preset medium-deep geothermal system are optimized based on the target hybrid model; the system reinjection parameters include reinjection flow rate and reinjection temperature.

2. The dynamic optimization method for geothermal extraction according to claim 1, characterized in that, The step of inputting target input data for the heating season in the target area into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and generating a target training dataset using the target input data and the simulated injection-production pressure difference data, includes: Real-time meteorological temperature data of the target area during the heating season is collected, and the real-time meteorological temperature data and preset field reinjection parameters are used as input boundary conditions to input the preset heat flow coupling numerical model to generate simulated injection-production pressure difference data. The real-time meteorological temperature data, the preset field reinjection parameters, and the simulated injection-production pressure difference data are subjected to standardization processing to obtain processed data, and a target training dataset is constructed based on the processed data.

3. The dynamic optimization method for geothermal extraction according to claim 1, characterized in that, The process of training a preset hybrid model using the target training dataset to obtain a target hybrid model, and extracting data features from the target input data using the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, to predict the injection-mass pressure difference based on the data features, and obtaining the target predicted injection-mass pressure difference, includes: The preset hybrid model is trained based on the target training dataset, preset training rounds, and preset training parameters to obtain the target hybrid model; The target input data is input into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and perform injection-production pressure difference prediction based on the data features to obtain the target predicted injection-production pressure difference.

4. The dynamic optimization method for geothermal extraction according to claim 3, characterized in that, The step of inputting the target input data into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model respectively, so as to extract the target input data features through the long short-term memory network branch and the temporal convolutional network branch, and to predict the injection-import pressure difference based on the data features to obtain the target predicted injection-import pressure difference, includes: The target input data is respectively input into the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model; The first data feature of the target input data is extracted through the forget gate, input gate and output gate in the branch of the long short-term memory network. The second data feature of the target input data is extracted by the causal dilated convolution branch of the temporal convolutional network and the residual block; The first data feature and the second data feature are fused by the feature fusion layer of the target hybrid model to obtain fused features. The injection-production pressure difference is then predicted by the fully connected layer of the target hybrid model based on the fused features to obtain the target predicted injection-production pressure difference.

5. The dynamic optimization method for geothermal extraction according to claim 1, characterized in that, The step of determining whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference includes: The model error of the target hybrid model is determined based on the target predicted injection-production pressure difference, and it is determined whether the model error is less than a preset error threshold. If the model error is less than the preset error threshold, then the model performance of the target hybrid model is determined to meet the preset performance conditions.

6. The dynamic optimization method for geothermal extraction according to any one of claims 1 to 5, characterized in that, The optimization of the reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model includes: The target fitness function is set based on minimizing the injection-production pressure difference, minimizing the thermal power difference, and maximizing the net revenue of the system in the preset medium-deep geothermal system. The target fitness function is solved by the target hybrid model based on preset constraints to obtain the corresponding target optimal solution set; The reinjection parameters of the preset medium-deep geothermal system are optimized based on the target optimal solution set.

7. The dynamic optimization method for geothermal extraction according to claim 6, characterized in that, The step of solving the target fitness function based on preset constraints using the target hybrid model to obtain the corresponding target optimal solution set includes: By using a pre-set multi-objective genetic algorithm during population iteration, the target hybrid model is invoked to solve the target fitness function based on pre-set constraints, so as to obtain several solutions; By performing non-dominated sorting, crowding calculation, and crossover mutation on the solutions, several target optimal solutions are selected, and a target optimal solution set is generated based on the solutions.

8. A dynamic optimization device for geothermal extraction, characterized in that, include: The dataset generation module is used to input the target input data of the heating season in the target area into a preset heat-fluid coupling numerical model to generate simulated injection-production pressure difference data, and to generate a target training dataset through the target input data and the simulated injection-production pressure difference data; the target input data is the real-time meteorological temperature data of the heating season in the target area and preset field reinjection parameters; The data prediction module is used to train a preset hybrid model using the target training dataset to obtain a target hybrid model, and to extract data features of the target input data through the long short-term memory network branch and the temporal convolutional network branch of the target hybrid model, so as to predict the injection-production pressure difference based on the data features and obtain the target predicted injection-production pressure difference. The condition judgment module is used to determine whether the model performance of the target hybrid model meets the preset performance conditions based on the target predicted injection-production pressure difference; The parameter optimization module is used to optimize the system reinjection parameters of the preset medium-deep geothermal system based on the target hybrid model if the target hybrid model meets the preset performance conditions; the system reinjection parameters include reinjection flow rate and reinjection temperature.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the dynamic optimization method for geothermal extraction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the dynamic optimization method for geothermal extraction as described in any one of claims 1 to 7.

Images