Oil reservoir dynamic monitoring method and device

Abstract

The invention discloses an oil reservoir dynamic monitoring method and device. The method comprises the steps that sound wave data and temperature data collected by distributed optical fiber sensors in real time are obtained; the distributed optical fiber sensors are deployed in a plurality of injection and production wells according to a preset deployment rule; the sound wave data and the temperature data are input into an oil reservoir dynamic monitoring model, an oil reservoir fluid three-dimensional dynamic graph is output, and the oil reservoir fluid three-dimensional dynamic graph displays oil reservoir fluid state data; the reservoir dynamic monitoring model is obtained by training a Transform model by using historical sound wave data, temperature data and reservoir multi-source information in advance; in the training process, time and space grid alignment processing is carried out on historical sound wave data, temperature data and oil reservoir multi-source information, time-space sequence features are obtained, and the model identifies oil reservoir fluid state data in the time-space sequence features based on self-attention and multi-attention mechanisms. The accuracy and reliability of oil reservoir monitoring and analysis can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum exploration, and in particular to a method and a device for dynamic monitoring of oil reservoirs. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] The main purpose of reservoir monitoring technology is to understand the dynamic changes of reservoirs and improve oil and gas recovery. Traditional reservoir monitoring technology mainly obtains real-time data from short-term dynamic detection or a few low-precision ground sensors, and then uses these data for dynamic reservoir monitoring. However, when using these data for reservoir monitoring, there are obvious deficiencies in accuracy, richness, density, coverage, and analysis methods. There are also methods for reservoir monitoring based on convolutional neural networks in the prior art, but the data modality processed by this method is relatively single, and the parallel processing capability of the model is also poor, resulting in low prediction accuracy of the flow state and distribution law of fluids inside the reservoir. Summary of the invention

[0004] The embodiment of the present invention provides a reservoir dynamic monitoring method for accurately predicting the flow state and distribution law of fluid inside the reservoir and improving the reliability of reservoir dynamic prediction. The method includes:

[0005] Acquiring acoustic wave data collected in real time by distributed acoustic wave sensors and temperature data collected in real time by distributed temperature sensors; the distributed acoustic wave sensors and distributed temperature sensors are deployed in multiple injection and production wells according to preset deployment rules, wherein the preset deployment rules include deployment location requirement information;

[0006] Input the acoustic wave data and temperature data into a pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic image of the reservoir fluid, wherein the three-dimensional dynamic image of the reservoir fluid displays the reservoir fluid state data, and the reservoir fluid state data includes reservoir fluid properties, distribution law, flow state, and non-flow state remaining oil distribution; the reservoir dynamic monitoring model pre-uses historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model;

[0007] During the training process: historical acoustic wave data, temperature data and multi-source reservoir information are grid-aligned in time and space to obtain spatiotemporal sequence features. The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the multi-source reservoir information includes geological characteristics, formation characteristics, reservoir evaluation, and production engineering information.

[0008] The embodiment of the present invention further provides a reservoir dynamic monitoring device for accurately predicting the flow state and distribution law of fluid inside the reservoir and improving the reliability of reservoir dynamic prediction. The device includes:

[0009] A data acquisition module is used to acquire the acoustic wave data collected in real time by the distributed acoustic wave sensor and the temperature data collected in real time by the distributed temperature sensor; the distributed acoustic wave sensor and the distributed temperature sensor are deployed in a plurality of injection and production wells according to a preset deployment rule, and the preset deployment rule includes deployment position requirement information;

[0010] The intelligent prediction module is used to input the acoustic wave data and temperature data into a pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic map of the reservoir fluid, wherein the three-dimensional dynamic map of the reservoir fluid displays the reservoir fluid state data, wherein the reservoir fluid state data includes reservoir fluid properties, distribution law, flow state, and non-flowing residual oil distribution; the reservoir dynamic monitoring model pre-uses historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model; during the training process, the historical acoustic wave data, temperature data, and reservoir multi-source information are grid-aligned in time and space to obtain spatiotemporal sequence features, and the Transformer model identifies the reservoir fluid state data in the spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the reservoir multi-source information includes geological features, formation features, reservoir evaluation, and production engineering information.

[0011] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned reservoir dynamic monitoring method when executing the computer program.

[0012] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned reservoir dynamic monitoring method is implemented.

[0013] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned reservoir dynamic monitoring method is implemented.

[0014] The embodiment of the present invention realizes dynamic monitoring of oil reservoirs based on the Transformer model. The Transformer model is trained using historical acoustic wave data, temperature data, and multi-source information of oil reservoirs. During the training process, the historical acoustic wave data, temperature data, and multi-source information of oil reservoirs are grid-aligned in time and space to obtain spatiotemporal sequence characteristics, forming a unified multi-level, multi-dimensional, multi-vector, and multi-modal data set for oil reservoirs, greatly improving the consistency and availability of data; the Transformer model can perform overall fusion analysis of various types of oil reservoir data through attention mechanism, multi-modal feature fusion, and machine autonomous learning model optimization technology, effectively capture and analyze the correlation between different data modes, and realize ultra-fine analysis of the geological physical characteristics and fluid flow state inside the oil reservoir, thereby improving the prediction accuracy of the fluid flow state and distribution law inside the oil reservoir, and improving the reliability of oil reservoir dynamic prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0016] Figure 1 Schematic diagram of the process of the reservoir dynamic monitoring method in an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of sensor deployment requirements in an embodiment of the present invention;

[0018] Figure 3 Schematic diagram of multimodal data integration and processing flow in an embodiment of the present invention;

[0019] Figure 4 A schematic diagram of the data processing flow of the trained reservoir dynamic monitoring model in an embodiment of the present invention;

[0020] Figure 5 This is a schematic diagram of prediction results of a reservoir dynamic monitoring model in an embodiment of the present invention;

[0021] Figure 6 Schematic diagram of a reservoir dynamic monitoring device in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0023] In the existing technology, reservoir monitoring has obvious deficiencies in accuracy, richness, density, coverage and analytical methods. Especially in complex reservoir environments, it is often impossible to accurately predict the flow state and distribution law of fluids inside the reservoir. Therefore, there is an urgent need for a technical solution that can integrate multiple types of data to provide more comprehensive monitoring and interpretation results.

[0024] The embodiments of the present invention aim to solve the limitation problem of using only single modality data in existing reservoir dynamic analysis technology. In complex reservoir environments, this limitation cannot meet the high-precision, real-time and comprehensive requirements of oilfield development. Therefore, the embodiments of the present invention propose a more comprehensive, real-time and efficient multi-modal AI deep learning reservoir dynamic monitoring and rapid analysis solution.

[0025] The embodiment of the present invention relates to a method for dynamic monitoring and dynamic analysis of oil reservoirs, and constructs an oil reservoir dynamic monitoring model based on distributed optical fiber sensing multimodal fusion and Transformer model framework. The oil reservoir dynamic monitoring model makes full use of the self-attention mechanism, multimodal feature fusion and machine autonomous learning model optimization technology to achieve deep fusion between the sound wave and temperature data collected by the distributed optical fiber sensor and multi-source data such as oil reservoir geological characteristics, production dynamics, and injection and production process measures. The embodiment of the present invention can accurately explain the fluid distribution law, flow state and characteristics inside the oil reservoir, as well as the distribution of remaining oil, and thus achieve high-precision oil reservoir dynamic prediction, injection and production process optimization, production capacity dynamic evaluation and development plan adjustment suggestions.

[0026] Figure 1 FIG. 1 is a flow chart of a method for dynamic monitoring of oil reservoirs in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0027] Step 101, obtaining acoustic wave data collected in real time by a distributed acoustic wave sensor and temperature data collected in real time by a distributed temperature sensor; the distributed acoustic wave sensor and the distributed temperature sensor are deployed in a plurality of injection and production wells according to a preset deployment rule, wherein the preset deployment rule includes deployment position requirement information;

[0028] Step 102: input the acoustic wave data and temperature data into a pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic map of the reservoir fluid, wherein the three-dimensional dynamic map of the reservoir fluid displays the reservoir fluid state data, wherein the reservoir fluid state data includes reservoir fluid properties, distribution patterns, flow states, and non-flowing residual oil distribution; the reservoir dynamic monitoring model pre-uses historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model; during the training process, the historical acoustic wave data, temperature data, and reservoir multi-source information are grid-aligned in time and space to obtain spatiotemporal sequence features, and the Transformer model identifies the reservoir fluid state data in the spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the reservoir multi-source information includes geological features, formation features, reservoir evaluation, and production engineering information.

[0029] from Figure 1 It can be seen from the process shown that the embodiment of the present invention realizes dynamic monitoring of oil reservoirs based on the Transformer model. The Transformer model is trained using historical acoustic wave data, temperature data, and multi-source information of oil reservoirs. During the training process, the historical acoustic wave data, temperature data, and multi-source information of oil reservoirs are grid-aligned in time and space to obtain spatiotemporal sequence characteristics, forming a unified multi-level, multi-dimensional, multi-vector, and multi-modal data set for oil reservoirs, greatly improving the consistency and availability of data; the Transformer model can perform overall fusion analysis of various types of oil reservoir data through attention mechanism, multi-modal feature fusion, and machine autonomous learning model optimization technology, effectively capture and analyze the correlation between different data modes, and realize ultra-fine analysis of the geological physical characteristics and fluid flow state inside the oil reservoir, thereby improving the prediction accuracy of the fluid flow state and distribution law inside the oil reservoir, and improving the reliability of oil reservoir dynamic prediction.

[0030] In specific implementation, in the embodiment of the present invention, the distributed acoustic wave sensors and distributed temperature sensors are deployed in multiple injection and production wells according to preset deployment rules, and the preset deployment rules include deployment position requirement information.

[0031] For example, refer to Figure 2 , Figure 2Schematic diagram of sensor deployment requirements in an embodiment of the present invention. In an embodiment of the present invention, based on the geomechanical model and fluid dynamics analysis, the sensor sensing accuracy is optimized, and at least three injection and production wells distributed in a triangle are set to be equipped with distributed optical fiber DAS (Distributed Acoustic Sensing, DAS) sound and DTS (Distributed Temperature Sensing, DTS) temperature sensors with millisecond sampling frequencies that meet the ultra-low frequency (for example, <0.01 Hz) shear wave detection accuracy at the centimeter level.

[0032] In the embodiment of the present invention, requirements are also set for sensor detection, which requires ensuring that the ultra-low frequency acoustic wave detection range can cover the triangular deployment area, and that each spatial micro-grid (for example, 1 cubic meter) in the area is covered by at least clear DAS acoustic wave sensor data.

[0033] After obtaining the acoustic wave data collected in real time by the distributed acoustic wave sensor and the temperature data collected in real time by the distributed temperature sensor, the acoustic wave data and the temperature data are input into a pre-trained reservoir dynamic monitoring model, and a three-dimensional dynamic map of the reservoir fluid is output. The three-dimensional dynamic map of the reservoir fluid displays the reservoir fluid status data, and the reservoir fluid status data includes the reservoir fluid properties, distribution law, flow state, and non-flowing residual oil distribution.

[0034] For example, after low-pass filtering and low-frequency feature enhancement preprocessing, the distributed optical fiber DAS acoustic wave data and DTS temperature data collected in real time are input into the reservoir dynamic monitoring model pre-trained based on the Transformer model to perform dynamic deductive analysis of the reservoir fluid flow state and output the final result. The result can be displayed in a three-dimensional dynamic analysis diagram, showing the overall flow state of the reservoir fluid, the reservoir fluid properties, distribution law, flow state, non-flowing residual oil distribution, and the reservoir micro-grid fluid properties, flow velocity, and flow direction, and dynamically outputs the dynamic distribution and changes of reservoir micro-vibration, flow state, fluid properties, and residual oil.

[0035] Among them, the dynamic monitoring model of the reservoir pre-uses historical acoustic data, temperature data, and reservoir multi-source information to train the Transformer model; during the training process: the historical acoustic data, temperature data and reservoir multi-source information are grid-aligned in time and space to obtain spatiotemporal sequence characteristics, and the Transformer model identifies the reservoir fluid state data in the spatiotemporal sequence characteristics based on self-attention and multi-head attention mechanisms; the reservoir multi-source information includes geological characteristics, formation characteristics, reservoir evaluation, and production engineering information.

[0036] Specifically, in one embodiment, the reservoir dynamic monitoring model can be trained as follows:

[0037] Acquire historical: acoustic data, temperature data, and multi-source reservoir information;

[0038] Using historical acoustic wave data, temperature data, and multi-source information of oil reservoirs, a spatial microgrid is constructed; the spatial microgrid includes a plurality of grids of specified sizes, each grid corresponds to a specified position of the oil reservoir, and each grid includes dynamic data of the corresponding specified position of the oil reservoir, and the dynamic data includes acoustic characteristic data, seepage characteristic data, rock physical property data, geological structure data, fluid property data, and reservoir condition data;

[0039] Use spatial microgrids as training data to train the Transformer model.

[0040] During implementation, a large amount of acoustic wave data, temperature data, and multi-source reservoir information are first obtained. Multi-source information refers to geological, geophysical, engineering and production data from different sources and different types. These data together constitute information for a comprehensive understanding of the reservoir and its surrounding environment, such as reservoir evaluation. Reservoir evaluation involves evaluating the reserves, production capacity and economic value of the reservoir. Reservoir evaluation can include logging data, core analysis data, well test data, seismic data, etc. Geological characteristics can reflect the stratigraphic structure and sedimentary environment of the area where the reservoir is located, mainly including geological maps, seismic data interpretation, underground structure models, etc. Production dynamics can reflect the performance of oil wells or oil fields in the production process, such as wellhead pressure, temperature, production (oil, gas, water), water content, production plan, water injection plan, etc.

[0041] Then, the spatial micro-grid is constructed using the large amount of original sample data obtained.

[0042] In order to improve the efficiency of constructing the spatial micro-grid, the acquired data is first pre-processed and integrated. In one embodiment, the historical acoustic wave data, temperature data, and reservoir multi-source information are used. Before constructing the spatial micro-grid, the following steps may also be included:

[0043] The historical acoustic data, temperature data, and multi-source reservoir information are annotated in time and space to obtain a multi-modal data set after time and space alignment; the multi-modal data set after time and space alignment includes annotated acoustic data and annotated non-acoustic data;

[0044] Based on the wavelet analysis method, the labeled sound wave data is separated into spatial, time and frequency domains, and the separated signals are enhanced to obtain enhanced sound wave data;

[0045] The enhanced acoustic wave data is processed by the time reverse migration (TRM) method based on AI deep learning to obtain multiple sound source signals; the sound source signal represents the sound source where the fluid seepage occurs at any spatial position inside the reservoir;

[0046] Using historical acoustic data, temperature data, and reservoir multi-source information, a spatial microgrid is constructed, including:

[0047] The spatial micro-grid is constructed using the annotated non-acoustic data and multiple sound source signals.

[0048] refer to Figure 3 , Figure 3 Schematic diagram of multimodal data integration and processing flow in an embodiment of the present invention. Figure 3 It shows how to obtain data through various channels, for example, through downhole fiber optic sensing systems, single well condition monitoring systems, oil well dynamic monitoring, injection and production well dynamic monitoring, reservoir geological results and other channels. Finally, multimodal data is integrated and processed in time and space based on AI deep learning, and spatial micro-grids are constructed through static serialization and dynamic serialization.

[0049] When integrating multimodal data in time and space, the DAS acoustic wave data, DTS temperature data, and multi-source information such as reservoir evaluation, geological structure, and production dynamics of all distributed fiber optic monitoring wells are integrated and synchronized to form various reservoir graphics, documents, data tables, and multi-level, multi-dimensional, multi-vector, and multi-modal data sets such as audio, video, and images with unified time and space annotation standards.

[0050] When enhancing the low-frequency signal, the spatial, temporal and frequency domain signal separation method of the distributed optical fiber sensor DAS acoustic wave data of oil and water wells based on wavelet analysis is used to perform reasonable spatial, temporal and frequency domain signal separation, and necessary signal enhancement is performed on the separated low-frequency characteristic signals of the fluid flow.

[0051] When restoring the spatial sound source, the TRM sound restoration technology based on AI deep learning is used to simulate the focusing phenomenon of the signal at the original sound source position during the reverse propagation process on the basis of spatial, temporal and frequency domain signal separation, thereby effectively separating and enhancing multiple specific sound source signals generated by fluid seepage in any space inside the reservoir.

[0052] Then, a spatial microgrid is constructed. According to the complexity of the geological characteristics of the reservoir, a spatial matrix microgrid (≤5 cubic meters) time-space sequence coding system that can clearly express the flow state of the reservoir fluid is constructed. The purpose is to accurately divide and encode the target reservoir production dynamics through the microgrids of the time-space dimension, realize the orderly organization of multi-source and multi-modal data in any space of the reservoir, and lay the foundation for the application of the Transformer model with multi-modal self-attention mechanism for reservoir dynamic monitoring and intelligent analysis. The microgrid time-space sequence coding system is required to cover the accuracy requirements of reservoir dynamic data analysis from centimeters to meters, that is, to deploy data with centimeter accuracy to ensure that the data can be aligned and integrated in the spatial microgrid. When constructing a spatial microgrid, a spatial matrix is ​​constructed, and the spatial matrix is ​​divided into multiple grids of specified sizes. The static data is first serialized using the annotated non-acoustic data, and the annotated non-acoustic data is mapped to the spatial grid. Then, multiple specific sound source signals are used to serialize the dynamic data, and dynamic information is given to the static data already in the spatial grid.

[0053] First, static data serialization is performed, and AI deep learning algorithms are used to fully combine the specific sound source signals generated by fluid seepage in any space inside the reservoir. The multimodal data set after preprocessing time and space integration is used to perform spatial micro-grid optimization, high-dimensional multi-parameter data adaptive interpolation, adaptive filtering denoising, self-identification feature annotation and other grid data recovery (restoring the multimodal data subset in the grid), and construct a multimodal normalized static serialization data set of acoustic characteristics, seepage characteristics, rock properties, geological structure, fluid properties, reservoir conditions, etc. in each spatial micro-grid. In specific processing, the results of the Kriging interpolation method, the adaptive interpolation method and the convolutional neural network interpolation method are integrated through the deep learning algorithm, and the parameters in the regression task are optimized by the deep learning optimization regression algorithm, such as the back propagation algorithm and the gradient descent optimization method, to automatically minimize the prediction error.

[0054] Then, dynamic data serialization is performed: using a deep learning algorithm, the specific sound source signal generated by fluid seepage in any space inside the reservoir is used, combined with the micro-grid multimodal normalized static serialized data set, to perform spatial sound source restoration processing again. For example, the signal that has been processed by spatial sound source restoration is separated again in the micro-grid in the spatial, temporal, and frequency domains using a deep learning algorithm. The spatial, temporal, and frequency domain signal separation results in the micro-grid are further subjected to noise suppression and feature signal enhancement to complete the feature restoration of the continuous acoustic wave signal in the micro-grid. The purpose is to separate more identifiable feature signals of fluid properties and flow states, and restore and enhance the acoustic spectrum, flow velocity, flow direction, pressure change, temperature change and other fluid flow state information in each spatial micro-grid according to the time series, and construct a high-dimensional multimodal spatiotemporal serialized gridded data set for overall spatial analysis of reservoir fluid flow states.

[0055] The constructed spatial microgrid includes multiple networks of specified sizes, each grid corresponds to a specified position of the reservoir, and each grid includes dynamic data of the corresponding specified position of the reservoir, and the dynamic data at least includes acoustic characteristic data, seepage characteristic data, rock physical property data, geological structure data, fluid property data, and reservoir condition data.

[0056] Afterwards, the spatial microgrids are used as training data to train the Transformer model.

[0057] In the embodiment of the present invention, an open source big model is preferred. A multimodal Transformer model of reservoir time-space gridding that allows customized development is constructed based on the open source big model API. It is required to have a training foundation in audio signal processing, and can solve the feature processing problem of audio signals through its unique attention mechanism and global feature capture capability, and can deeply integrate audio signals with other multimodal data features through its self-attention and multi-head attention mechanisms.

[0058] Choose a mature open source framework as the technical foundation, and implement model building in combination with tools such as service platforms for building, training, deploying and managing artificial intelligence models or data science assistants for deploying models.

[0059] When training a model in an embodiment of the present invention, a multimodal data input microservice is developed. Specifically, a normalized static data serialization data set and a high-dimensional fluid flow state dynamic data serialization data set are obtained by using a spatial microgrid. On the basis of receiving and processing the spatial microgrid multimodal normalized static data serialization data set and the microgrid high-dimensional fluid flow state dynamic data serialization data set, a Transformer model training multimodal data input microservice is developed. Various types of geological research, dynamic analysis, program evaluation, injection and production well basic information, production dynamics and other multi-source information of the reservoir and the distributed optical fiber DAS acoustic wave data of the monitoring well, DTS temperature original and preprocessing result data are all used as training data after space and time alignment processing, and a consistent spatiotemporal serialization representation is performed for use in Transformer model training.

[0060] During the model training process, the historical acoustic wave data, temperature data and reservoir multi-source information are grid-aligned in time and space to obtain spatiotemporal sequence features. The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, wherein the Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, including:

[0061] Step 1-recognizable feature pattern learning: The Transformer model identifies multiple feature combinations in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, and the feature combinations include any combination of reservoir fluid properties, distribution patterns, flow states, and non-flowing remaining oil distribution;

[0062] Step 2-Automatic screening of identifiable features: using a feature importance evaluation method and a feature selection algorithm to screen multiple feature combinations to obtain multiple screened feature combinations; the feature importance evaluation method includes a mutual information method or a recursive feature elimination method, and the feature selection algorithm includes a LASSO (Least Absolute Shrinkage and Selection Operator, a regression analysis algorithm) regression algorithm or a random forest algorithm;

[0063] Step 3-Adaptive feature recognition combination: using an intelligent optimization algorithm to adjust multiple feature combinations to obtain multiple adjusted feature combinations; the intelligent optimization algorithm includes a genetic algorithm or a particle swarm optimization algorithm; the adjustment includes adjusting the correlation between features and the correlation between feature combinations;

[0064] Step 4-Automatic labeling of microgrid fluid state: Based on the adjusted multiple feature combinations and the pre-established global self-attention weight matrix, the basic equations related to the basic knowledge of reservoir fluid dynamics are used as constraints to perform recognition training on reservoir fluid state data; the model parameters are optimized during the recognition training process; the global self-attention weight matrix is ​​used to capture the dependency relationship between the adjusted multiple feature combinations, and the basic equations related to the basic knowledge of reservoir fluid dynamics include one or any combination of the continuity equation, momentum equation, and energy equation.

[0065] Specifically, step 1 - recognizable feature pattern learning: the self-attention mechanism and multi-head attention mechanism of the Transformer model are used to capture the correlation between high-dimensional typical features of reservoir spatial micro-grid data, and the typical feature pattern is automatically discovered and defined for the combination relationship of related features summarized by training.

[0066] Step 2 - Automatic screening of identifiable features: Utilizing the self-feedback mechanism in the Transformer model, feature importance evaluation methods and feature selection algorithms are used to automatically screen feature combination patterns that have a significant impact on reservoir fluid flow, reduce unnecessary feature dimensions in the combination patterns, and improve model training efficiency and generalization ability.

[0067] Step 3-Adaptive feature recognition combination: Apply the long-distance dependency calculation mechanism of the Transformer model, preset the reservoir fluid flow state and fluid property identification feature pattern combination strategy, combine with genetic algorithm, particle swarm optimization and other intelligent optimization algorithms, automatically optimize and adjust the fluid flow time-space feature combination pattern, find the optimal feature pattern combination subset, so that the reservoir fluid flow state model can automatically adjust the correlation of the reservoir fluid flow state feature combination pattern according to the different training data, and enhance the pattern recognition accuracy and generalization ability of the reservoir fluid flow state. Identify the reservoir fluid state through different feature combinations.

[0068] Step 4 - Automatic labeling of micro-grid fluid states: Pre-build a multi-layer Transformer model, establish a global self-attention weight matrix, and based on the feature combination pattern of reservoir fluid flow state, use the basic equations related to the basic knowledge of reservoir fluid dynamics as constraints to conduct combined training and clarity parameter optimization of feature engineering such as micro-grid reservoir fluid properties, distribution trends, flow states, and non-flowing residual oil characteristics.

[0069] Figure 4 FIG. 1 is a schematic diagram of the data processing flow of the trained reservoir dynamic monitoring model in an embodiment of the present invention. Figure 4 As shown in the figure, based on the spatial micro-grid, the spatiotemporal sequence dataset A and the spatiotemporal sequence dataset n are input respectively. For each input spatiotemporal sequence dataset, it is processed by Wav2Vec2 (Wav2Vec2 is a speech recognition model based on self-supervised learning for processing sound wave data) and BERT (a pre-trained model based on Transformer architecture for processing non-sound wave data) to obtain an embedding matrix, which is then processed by the encoder, combined attention mechanism (QKV), full connection, and multi-head interactive attention matrix to reach the feedforward neural network processing, and residual connection, summation, and layer normalization are performed in the middle, and finally maximum pooling and decoder processing are performed, and then the fully connected feedforward neural network, residual connection, summation, normalization, and linear maximum layer processing are performed until the prediction result is finally output.

[0070] Figure 5 Schematic diagram of prediction results of reservoir dynamic monitoring model in the embodiment of the present invention. Figure 5 As shown, different colors represent different fluid properties, and can be displayed in layers according to different reservoir layers or seepage layers.

[0071] In one embodiment, after inputting the acoustic wave data and the temperature data into a pre-established reservoir dynamic monitoring model and outputting a three-dimensional dynamic diagram of the reservoir fluid, the following may also be included:

[0072] Determine the change trend of injection and production capacity based on the three-dimensional dynamic diagram;

[0073] According to the injection-production capacity change trend, the injection-production process measures corresponding to the optimal recovery factor and the optimization suggestion information of the current injection-production process and measures are determined; the optimal recovery factor indicates that the recovery factor exceeds the specified value.

[0074] In the specific implementation, the intelligent generation capability of the Transformer model is used to predict the current injection and production capacity change trend of the current reservoir fluid flow state, infer the best injection and production process measures for the optimal recovery rate, dynamically generate optimization suggestions for the current injection and production processes and measures, and predict future capacity improvement conditions.

[0075] In summary, the embodiments of the present invention have the following beneficial effects:

[0076] 1. Improvement of monitoring and analysis capabilities:

[0077] By introducing a reservoir dynamic monitoring model based on distributed fiber optic sensing multimodal fusion, the embodiment of the present invention significantly improves the accuracy and real-time performance of reservoir monitoring. Distributed fiber optic sensing technology can obtain high-precision, high-density real-time data, including sound waves, temperature and other information. Combined with advanced signal processing technology and AI deep learning methods, low-frequency signals are enhanced and data quality is significantly improved, thereby improving the accuracy of reservoir monitoring.

[0078] In terms of data analysis, through the spatiotemporal integration and preprocessing of multimodal data, a unified multi-level, multi-dimensional, multi-vector, and multi-modal data set of the reservoir has been formed, which greatly improves the consistency and availability of the data. The Transformer model of multimodal processing effectively captures and analyzes the correlation between different data modes, further improving the explanatory power and prediction accuracy of the model.

[0079] (II) Optimization of injection and production technology and improvement of production capacity:

[0080] The embodiment of the present invention can dynamically analyze the flow state of reservoir fluid and provide scientific and accurate decision support for capacity optimization. Through intelligent analysis of the current flow state of reservoir fluid, the optimal injection and production process measures can be predicted and implemented, thereby improving the capacity and recovery rate.

[0081] 3. Risk and cost reduction:

[0082] By accurately predicting the fluid distribution law, flow state and remaining oil distribution inside the reservoir, the embodiment of the present invention significantly reduces the risk of blind exploitation. At the same time, it realizes the dynamic prediction of the reservoir and the optimization of the injection and production process, which helps to reasonably plan the development plan, thereby reducing unnecessary exploration and production costs.

[0083] (IV) Improvement of intelligence level and enhancement of work efficiency:

[0084] The technical solution of the embodiment of the present invention provides a new intelligent solution for reservoir monitoring and promotes the progress of reservoir development technology. Through the combination of automated feature screening and adaptive feature recognition, the data processing process is simplified and the work efficiency is significantly improved. At the same time, a real-time processing and prediction system is established to provide more scientific and accurate support for reservoir management.

[0085] 5. Enhancement of sustainability:

[0086] Through intelligent dynamic prediction and optimization suggestions, the embodiments of the present invention help to achieve more scientific and reasonable reservoir management, thereby extending the life cycle of the reservoir, while reducing resource waste and improving resource utilization.

[0087] (VI) Improvement of data fusion depth and spatiotemporal resolution:

[0088] The embodiment of the present invention deeply integrates multiple data types through the Transformer model of multimodal processing, reveals more hidden information, and improves the accuracy and comprehensiveness of prediction. At the same time, it realizes fine division and long-term prediction in space and time series, and enhances the foresight and response speed of decision-making.

[0089] (VII) Formation of continuous optimization mechanism:

[0090] The embodiment of the present invention adapts to reservoir environmental changes through feature correlation analysis and continuous model optimization, bringing long-term technical benefits. This continuous optimization mechanism ensures the long-term effectiveness and competitiveness of the embodiment of the present invention in practical applications.

[0091] In summary, the embodiments of the present invention not only significantly improve the accuracy and reliability of reservoir monitoring and analysis by introducing a reservoir dynamic monitoring model based on distributed fiber optic sensing multimodal fusion and its application, but also optimize the injection and production process, improve production capacity and recovery rate, reduce development risks and costs, improve the intelligence level of reservoir development, and enhance the sustainability of reservoir development.

[0092] The present invention also provides a reservoir dynamic monitoring device, as described in the following embodiments. Since the principle of the device to solve the problem is similar to that of the reservoir dynamic monitoring method, the implementation of the device can refer to the implementation of the reservoir dynamic monitoring method, and the repeated parts will not be repeated.

[0093] Figure 6 FIG. 1 is a schematic diagram of a reservoir dynamic monitoring device according to an embodiment of the present invention. Figure 6 As shown, the device comprises:

[0094] The data acquisition module 601 is used to acquire the acoustic wave data collected in real time by the distributed acoustic wave sensor and the temperature data collected in real time by the distributed temperature sensor; the distributed acoustic wave sensor and the distributed temperature sensor are deployed in multiple injection and production wells according to a preset deployment rule, and the preset deployment rule includes deployment position requirement information;

[0095] The intelligent prediction module 602 is used to input the acoustic wave data and temperature data into a pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic map of the reservoir fluid, wherein the three-dimensional dynamic map of the reservoir fluid displays the reservoir fluid status data, and the reservoir fluid status data includes reservoir fluid properties, distribution patterns, flow status, and non-flowing residual oil distribution; the reservoir dynamic monitoring model pre-uses historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model; during the training process, the historical acoustic wave data, temperature data, and reservoir multi-source information are grid-aligned in time and space to obtain spatiotemporal sequence features, and the Transformer model identifies the reservoir fluid status data in the spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the reservoir multi-source information includes geological characteristics, formation characteristics, reservoir evaluation, and production engineering information.

[0096] In one embodiment, the reservoir dynamic monitoring model is trained as follows:

[0097] Acquire historical: acoustic data, temperature data, and multi-source reservoir information;

[0098] Using historical acoustic wave data, temperature data, and multi-source information of oil reservoirs, a spatial microgrid is constructed; the spatial microgrid includes a plurality of grids of specified sizes, each grid corresponds to a specified position of the oil reservoir, and each grid includes dynamic data of the corresponding specified position of the oil reservoir, and the dynamic data includes acoustic characteristic data, seepage characteristic data, rock physical property data, geological structure data, fluid property data, and reservoir condition data;

[0099] Use spatial microgrids as training data to train the Transformer model.

[0100] In one embodiment, before constructing a spatial microgrid using historical acoustic wave data, temperature data, and reservoir multi-source information, the following steps are also included:

[0101] The historical acoustic data, temperature data, and multi-source reservoir information are annotated in time and space to obtain a multi-modal data set after time and space alignment; the multi-modal data set after time and space alignment includes annotated acoustic data and annotated non-acoustic data;

[0102] Based on the wavelet analysis method, the labeled sound wave data is separated into spatial, time and frequency domains, and the separated signals are enhanced to obtain enhanced sound wave data;

[0103] The enhanced acoustic wave data is processed by the time reverse smoothing migration (TRM) method based on AI deep learning to obtain multiple sound source signals; the sound source signal represents the sound source where the fluid seepage occurs at any spatial position inside the reservoir;

[0104] Using historical acoustic data, temperature data, and reservoir multi-source information, a spatial microgrid is constructed, including:

[0105] The spatial micro-grid is constructed using the annotated non-acoustic data and multiple sound source signals.

[0106] In one embodiment, the Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, including:

[0107] The Transformer model identifies multiple feature combinations in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, wherein the feature combinations include any combination of reservoir fluid properties, distribution patterns, flow states, and non-flowing remaining oil distribution;

[0108] A feature importance evaluation method and a feature selection algorithm are used to screen multiple feature combinations to obtain multiple feature combinations after screening; the feature importance evaluation method includes a mutual information method or a recursive feature elimination method, and the feature selection algorithm includes a LASSO regression algorithm or a random forest algorithm;

[0109] Adopting an intelligent optimization algorithm to adjust multiple feature combinations to obtain multiple adjusted feature combinations; the intelligent optimization algorithm includes a genetic algorithm or a particle swarm optimization algorithm; the adjustment includes adjusting the correlation between the features and the feature combinations;

[0110] Based on the adjusted multiple feature combinations and the pre-established global self-attention weight matrix, the basic equations related to the basic knowledge of reservoir fluid dynamics are used as constraints to perform recognition training on reservoir fluid state data; the model parameters are optimized during the recognition training process; the global self-attention weight matrix is ​​used to capture the dependency relationship between the adjusted multiple feature combinations, and the basic equations related to the basic knowledge of reservoir fluid dynamics include one or any combination of the continuity equation, the momentum equation, and the energy equation.

[0111] In one embodiment, after inputting the acoustic wave data and the temperature data into a pre-established reservoir dynamic monitoring model and outputting a three-dimensional dynamic diagram of the reservoir fluid, the method further includes:

[0112] Determine the change trend of injection and production capacity based on the three-dimensional dynamic diagram;

[0113] According to the injection-production capacity change trend, the injection-production process measures corresponding to the optimal recovery factor and the optimization suggestion information of the current injection-production process and measures are determined; the optimal recovery factor indicates that the recovery factor exceeds the specified value.

[0114] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned reservoir dynamic monitoring method when executing the computer program.

[0115] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned reservoir dynamic monitoring method is implemented.

[0116] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned reservoir dynamic monitoring method is implemented.

[0117] The embodiment of the present invention realizes dynamic monitoring of oil reservoirs based on the Transformer model. The Transformer model is trained using historical acoustic wave data, temperature data, and multi-source information of oil reservoirs. During the training process, the historical acoustic wave data, temperature data, and multi-source information of oil reservoirs are grid-aligned in time and space to obtain spatiotemporal sequence characteristics, forming a unified multi-level, multi-dimensional, multi-vector, and multi-modal data set for oil reservoirs, greatly improving the consistency and availability of data; the Transformer model can perform overall fusion analysis of various types of oil reservoir data through attention mechanism, multi-modal feature fusion, and machine autonomous learning model optimization technology, effectively capture and analyze the correlation between different data modes, and realize ultra-fine analysis of the geological physical characteristics and fluid flow state inside the oil reservoir, thereby improving the prediction accuracy of the fluid flow state and distribution law inside the oil reservoir, and improving the reliability of oil reservoir dynamic prediction.

[0118] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0122] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring of oil reservoirs, characterized in that: include: Acquire the acoustic wave data collected in real time by the distributed acoustic wave sensor and the temperature data collected in real time by the distributed temperature sensor; Distributed acoustic wave sensors and distributed temperature sensors are deployed in multiple injection and production wells according to preset deployment rules, wherein the preset deployment rules include deployment location requirement information; Input the acoustic wave data and temperature data into a pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic image of the reservoir fluid, wherein the three-dimensional dynamic image of the reservoir fluid displays the reservoir fluid state data, and the reservoir fluid state data includes reservoir fluid properties, distribution law, flow state, and non-flow state remaining oil distribution; the reservoir dynamic monitoring model pre-uses historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model; During the training process: historical acoustic wave data, temperature data and multi-source reservoir information are grid-aligned in time and space to obtain spatiotemporal sequence features. The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the multi-source reservoir information includes geological characteristics, formation characteristics, reservoir evaluation, and production engineering information.

2. The reservoir dynamic monitoring method according to claim 1, characterized in that: The reservoir dynamic monitoring model is trained as follows: Acquire historical: acoustic data, temperature data, and multi-source reservoir information; Use historical acoustic data, temperature data, and reservoir multi-source information to construct a spatial microgrid; The spatial microgrid includes a plurality of grids of specified sizes, each grid corresponds to a specified position of the reservoir, and each grid includes dynamic data of the corresponding specified position of the reservoir, and the dynamic data includes acoustic characteristic data, seepage characteristic data, rock physical property data, geological structure data, fluid property data, and reservoir condition data; Use spatial microgrids as training data to train the Transformer model.

3. The reservoir dynamic monitoring method according to claim 2, characterized in that: Using historical acoustic wave data, temperature data, and reservoir multi-source information, before constructing the spatial microgrid, it also includes: The historical acoustic data, temperature data, and multi-source reservoir information are annotated in time and space to obtain a multi-modal data set after time and space alignment; the multi-modal data set after time and space alignment includes annotated acoustic data and annotated non-acoustic data; Based on the wavelet analysis method, the labeled sound wave data is separated into spatial, time and frequency domains, and the separated signals are enhanced to obtain enhanced sound wave data; The enhanced acoustic wave data is processed by the time reverse smoothing migration (TRM) method based on AI deep learning to obtain multiple sound source signals; the sound source signal represents the sound source where the fluid seepage occurs at any spatial position inside the reservoir; Using historical acoustic data, temperature data, and reservoir multi-source information, a spatial microgrid is constructed, including: The spatial micro-grid is constructed using the annotated non-acoustic data and multiple sound source signals.

4. The reservoir dynamic monitoring method according to claim 1, characterized in that: The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, including: The Transformer model identifies multiple feature combinations in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, wherein the feature combinations include any combination of reservoir fluid properties, distribution patterns, flow states, and non-flowing remaining oil distribution; A feature importance evaluation method and a feature selection algorithm are used to screen multiple feature combinations to obtain multiple feature combinations after screening; the feature importance evaluation method includes a mutual information method or a recursive feature elimination method, and the feature selection algorithm includes a LASSO regression algorithm or a random forest algorithm; Adopting an intelligent optimization algorithm to adjust multiple feature combinations to obtain multiple feature combinations after adjustment; the intelligent optimization algorithm includes a genetic algorithm or a particle swarm optimization algorithm; the adjustment includes adjusting the correlation between features and the correlation between feature combinations; Based on the adjusted multiple feature combinations and the pre-established global self-attention weight matrix, the basic equations related to the basic knowledge of reservoir fluid dynamics are used as constraints to perform recognition training on reservoir fluid state data; the model parameters are optimized during the recognition training process; the global self-attention weight matrix is ​​used to capture the dependency relationship between the adjusted multiple feature combinations, and the basic equations related to the basic knowledge of reservoir fluid dynamics include one or any combination of the continuity equation, the momentum equation, and the energy equation.

5. The reservoir dynamic monitoring method according to claim 1, characterized in that: After inputting the acoustic wave data and temperature data into the pre-established reservoir dynamic monitoring model and outputting the three-dimensional dynamic diagram of the reservoir fluid, it also includes: Determine the change trend of injection and production capacity based on the three-dimensional dynamic diagram; According to the injection-production capacity change trend, the injection-production process measures corresponding to the optimal recovery factor and the optimization suggestion information of the current injection-production process and measures are determined; the optimal recovery factor indicates that the recovery factor exceeds the specified value.

6. A reservoir dynamic monitoring device, characterized in that: include: A data acquisition module is used to acquire the acoustic wave data collected in real time by the distributed acoustic wave sensor and the temperature data collected in real time by the distributed temperature sensor; the distributed acoustic wave sensor and the distributed temperature sensor are deployed in a plurality of injection and production wells according to a preset deployment rule, and the preset deployment rule includes deployment position requirement information; The intelligent prediction module is used to input the acoustic wave data and temperature data into the pre-trained reservoir dynamic monitoring model, and output a three-dimensional dynamic diagram of the reservoir fluid, wherein the three-dimensional dynamic diagram of the reservoir fluid displays the reservoir fluid state data, and the reservoir fluid state data includes the reservoir fluid properties, distribution law, flow state, and non-flow state remaining oil distribution; the reservoir dynamic monitoring model is pre-trained by using the historical acoustic wave data, temperature data, and reservoir multi-source information to train the Transformer model; During the training process: historical acoustic wave data, temperature data and multi-source reservoir information are grid-aligned in time and space to obtain spatiotemporal sequence features. The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms; the multi-source reservoir information includes geological characteristics, formation characteristics, reservoir evaluation, and production engineering information.

7. The reservoir dynamic monitoring device according to claim 6, characterized in that: The reservoir dynamic monitoring model is trained as follows: Acquire historical: acoustic data, temperature data, and multi-source reservoir information; Use historical acoustic data, temperature data, and reservoir multi-source information to construct a spatial microgrid; The spatial microgrid includes a plurality of grids of specified sizes, each grid corresponds to a specified position of the reservoir, and each grid includes dynamic data of the corresponding specified position of the reservoir, and the dynamic data includes acoustic characteristic data, seepage characteristic data, rock physical property data, geological structure data, fluid property data, and reservoir condition data; Use spatial microgrids as training data to train the Transformer model.

8. The reservoir dynamic monitoring device according to claim 7, characterized in that: Using historical acoustic wave data, temperature data, and reservoir multi-source information, before constructing the spatial microgrid, it also includes: The historical acoustic data, temperature data, and multi-source reservoir information are annotated in time and space to obtain a multi-modal data set after time and space alignment; the multi-modal data set after time and space alignment includes annotated acoustic data and annotated non-acoustic data; Based on the wavelet analysis method, the labeled sound wave data is separated into spatial, time and frequency domains, and the separated signals are enhanced to obtain enhanced sound wave data; The TRM method based on AI deep learning processes the enhanced acoustic wave data to obtain multiple sound source signals; the sound source signals represent the sound source where the fluid seepage occurs at any spatial position inside the reservoir; Using historical acoustic data, temperature data, and reservoir multi-source information, a spatial microgrid is constructed, including: The spatial micro-grid is constructed using the annotated non-acoustic data and multiple sound source signals.

9. The reservoir dynamic monitoring device according to claim 6, characterized in that: The Transformer model identifies reservoir fluid state data in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, including: The Transformer model identifies multiple feature combinations in spatiotemporal sequence features based on self-attention and multi-head attention mechanisms, wherein the feature combinations include any combination of reservoir fluid properties, distribution patterns, flow states, and non-flowing remaining oil distribution; A feature importance evaluation method and a feature selection algorithm are used to screen multiple feature combinations to obtain multiple feature combinations after screening; the feature importance evaluation method includes a mutual information method or a recursive feature elimination method, and the feature selection algorithm includes a LASSO regression algorithm or a random forest algorithm; Adopting an intelligent optimization algorithm to adjust multiple feature combinations to obtain multiple feature combinations after adjustment; the intelligent optimization algorithm includes a genetic algorithm or a particle swarm optimization algorithm; the adjustment includes adjusting the correlation between features and the correlation between feature combinations; Based on the adjusted multiple feature combinations and the pre-established global self-attention weight matrix, the basic equations related to the basic knowledge of reservoir fluid dynamics are used as constraints to perform recognition training on reservoir fluid state data; the model parameters are optimized during the recognition training process; the global self-attention weight matrix is ​​used to capture the dependency relationship between the adjusted multiple feature combinations, and the basic equations related to the basic knowledge of reservoir fluid dynamics include one or any combination of the continuity equation, the momentum equation, and the energy equation.

10. The reservoir dynamic monitoring device according to claim 6, characterized in that: After inputting the acoustic wave data and temperature data into the pre-established reservoir dynamic monitoring model and outputting the three-dimensional dynamic diagram of the reservoir fluid, it also includes: Determine the change trend of injection and production capacity based on the three-dimensional dynamic diagram; According to the injection-production capacity change trend, the injection-production process measures corresponding to the optimal recovery factor and the optimization suggestion information of the current injection-production process and measures are determined; the optimal recovery factor indicates that the recovery factor exceeds the specified value.

11. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the reservoir dynamic monitoring method according to any one of claims 1 to 5 is implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for dynamic monitoring of oil reservoirs according to any one of claims 1 to 5 is implemented.

13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method for dynamic monitoring of oil reservoirs according to any one of claims 1 to 5 is implemented.

Images