Method and system for evaluating water sealing performance of underground water-sealed cave depot

Abstract

The method comprises the following steps: constructing a data set which takes the circumferential average monitoring data of a plurality of water curtain holes near a target vertical osmometer in a construction period as an input feature and takes the circumferential average osmotic pressure data of the target osmometer as an output target; cleaning the data set and creating a sliding time window to generate a training sample; dividing a training set and a test set; an LSTM-KAN prediction model fusing a long short-term memory network (LSTM) and a Coriolgov-Arnod network (KAN) is constructed and trained; and inputting the monitoring data of the water level hole in the operation period and the osmometer osmotic pressure value into the trained LSTM-KAN prediction model to evaluate the water seal performance. By adopting the technical scheme of the invention, the problems of insufficient seepage field characterization precision, lack of quantitative evaluation of water curtain efficiency, insufficient development of a long-term operation and maintenance data driving model and the like in the existing evaluation method for the water sealing performance of the underground water-sealed cave depot are solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of water-sealed cavern monitoring and measurement, and particularly relates to a water-sealing evaluation method and system for a water-sealed underground cavern. BACKGROUND

[0002] A water-sealed cavern is an important facility for underground oil and gas storage, and its water-sealing performance is the core to ensure safe operation. Due to its excellent regional adaptability, large storage scale and significant reduction of environmental risk, the underground water-sealed cavern has become an ideal choice for large-scale oil and gas storage. However, as a key scientific problem in the construction of the cavern, the water-sealing evaluation needs to comprehensively consider the complex coupling of multiple factors such as engineering and hydrogeological conditions, rock mass permeability characteristics and excavation disturbance, and its research system and method need to be systematized and refined. The existing water-sealing evaluation methods mainly include empirical method, numerical method and experimental method. The empirical method such as geological analysis method and engineering analogy method has strong subjectivity and limited precision; the numerical method such as hydrodynamic method, finite element method, discrete element method and fracture network method has complex calculation, insufficient representation precision of seepage field and difficulty in efficient processing of time sequence dynamics; the experimental method such as field monitoring and gas tightness test has high cost and long cycle. Although the current water-sealing evaluation research focuses on the construction of multi-dimensional method system and technical innovation, the traditional methods have strong subjectivity and great limitations, such as the empirical method can only provide preliminary estimation, and the numerical method needs to rely on many assumptions and simplifications. However, the existing data-driven methods have the problem of insufficient capture of complex dynamic changes of seepage field when predicting the water-sealing performance of the underground water-sealed cavern, such as relying only on the data of the piezometer in the same vertical direction, which is difficult to fully reflect the dynamic changes in the horizontal direction and has insufficient spatial representation. These deficiencies make it difficult for the existing methods to achieve high-precision dynamic prediction and comprehensive evaluation of water-sealing performance. SUMMARY

[0003] To solve the problems of insufficient representation precision of seepage field, lack of quantitative evaluation of water curtain effectiveness and insufficient development of long-term operation data-driven models in the existing water-sealing evaluation methods for underground water-sealed caverns, the present application provides a water-sealing evaluation method and system for underground water-sealed caverns, aiming to improve the precision and reliability of water-sealing evaluation and provide strong technical support for the safe and stable operation of the cavern.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solution:

[0005] A water-sealing evaluation method for an underground water-sealed cavern, comprising:

[0006] Step S1, constructing a data set containing the average monitoring data of multiple water curtain holes near the target vertical piezometer during the construction period as input features, and the average seepage pressure data of the target piezometer as output target;

[0007] Step S2, cleaning the data set and creating a sliding time window to generate training samples; dividing the training set and the test set; constructing and training an LSTM-KAN prediction model that integrates long short-term memory network (LSTM) and Kolmogorov-Arnold network (KAN); wherein LSTM is used to capture the long-term dependence of time series, and KAN network is used to enhance the fitting ability of complex nonlinear relationship;

[0008] Step S3, inputting the water level hole monitoring data and osmotic pressure value of the osmotic pressure meter in the running period into the trained LSTM-KAN prediction model to evaluate the water sealing performance.

[0009] As a preferred, the monitoring data of the water curtain hole includes: weekly average flow and pressure data.

[0010] As a preferred, the size of the sliding time window is 4 weeks.

[0011] The application also provides a groundwater sealed cavern water sealing evaluation system, comprising:

[0012] The first processing module is used to construct a data set containing the weekly average monitoring data of a plurality of water curtain holes near the target vertical osmotic pressure meter in the construction period as input features, and the weekly average osmotic pressure data of the target osmotic pressure meter as output targets;

[0013] The second processing module is used to clean the data set and create a sliding time window to generate training samples; divide the training set and the test set; construct and train an LSTM-KAN prediction model that integrates long short-term memory network (LSTM) and Kolmogorov-Arnold network (KAN); wherein LSTM is used to capture the long-term dependence of time series, and KAN network is used to enhance the fitting ability of complex nonlinear relationship;

[0014] The third processing module is used to input the water level hole monitoring data and osmotic pressure value of the osmotic pressure meter in the running period into the trained LSTM-KAN prediction model to evaluate the water sealing performance.

[0015] As a preferred, the monitoring data of the water curtain hole includes: weekly average flow and pressure data.

[0016] As a preferred, the size of the sliding time window is 4 weeks.

[0017] The application realizes continuous monitoring and evaluation from the construction period to the operation period by constructing a water seal multi-source multi-dimensional long-time monitoring data set of the whole life cycle of the cave. In the construction period, the water curtain hole flow, pressure and seepage pressure data are obtained through the collaborative monitoring of the water curtain hole and the seepage pressure gauge, providing rich input features and output targets for model training. In the operation period, the long-term evolution of the water seal performance is continuously tracked through the monitoring of the water level monitoring hole and the seepage pressure gauge, ensuring that the model can reflect the water seal state of the cave in real time. The underground water seal cave water seal evaluation method and system provided by the application is as follows: first, a water seal multi-source multi-dimensional long-time monitoring data set is constructed through the water curtain hole-seepage pressure gauge-water level monitoring hole collaborative monitoring network, and the data set is subjected to time series data preprocessing and feature engineering; then a time series prediction model is established, that is, the KAN network and the LSTM model are fused by using the LSTM-KAN model, so as to effectively capture the dynamic changes of the seepage field of the underground water seal cave. The application has unique advantages in processing water seal performance data with time dependence by using the time series model based on LSTM-KAN, and can accurately quantify the water seal performance of the underground water seal cave, providing a scientific basis for the safe operation and maintenance of the cave. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0019] Figure 1 The flow chart of the underground water seal cave water seal evaluation method of the embodiment of the present application;

[0020] Figure 2 The data acquisition and monitoring planar arrangement drawing of the embodiment of the present application;

[0021] Figure 3 The data acquisition and monitoring longitudinal section arrangement drawing of the embodiment of the present application;

[0022] Figure 4 The time series model architecture drawing of the embodiment of the present application;

[0023] Figure 5 The prediction model loss iteration drawing of the embodiment of the present application;

[0024] Figure 6 The comparison drawing of the predicted value and the actual value of the embodiment of the present application. DETAILED DESCRIPTION

[0025] 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.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1:

[0028] like Figure 1 As shown, this embodiment of the invention provides a method for evaluating the water-sealing performance of underground water-sealed caverns, including:

[0029] S1. Obtain flow and pressure data of water curtain holes and corresponding piezometer data during the construction period of the underground water-sealed cavern; obtain water level data of water level monitoring holes and piezometer data of key locations during the operation period of the underground water-sealed cavern; and construct a multi-source, multi-dimensional, long-term monitoring dataset of water seal performance throughout the entire life cycle.

[0030] Furthermore, such as Figure 2 As shown, the data includes the weekly average flow rate and pressure data of the eight water curtain orifices, as well as the data from the vertical piezometers PC2-5 to PC2-1.

[0031] Specifically, flow and pressure data of the water curtain orifice, as well as data from the piezometer, are obtained through on-site monitoring.

[0032] This study aims to construct a comprehensive, multi-source, multi-dimensional, long-term monitoring dataset for water-sealing performance in underground water-sealed caverns, providing a rich and reliable data foundation for evaluating their water-sealing performance. It comprehensively considers various monitoring methods, including water curtain orifices, piezometers, and water level monitoring wells, fully integrating data resources from each monitoring point to achieve a holistic mapping of the water-sealing performance of underground water-sealed caverns. During the construction phase of the underground water-sealed cavern, the monitoring dataset is primarily constructed using a collaborative analysis approach combining water curtain orifice and piezometer data. This results in a construction-phase water-sealing performance monitoring dataset with water curtain orifice flow and pressure data as input features and piezometer water pressure data as the output target. During the operational phase, the monitoring dataset is mainly constructed using piezometers and water level monitoring wells, jointly providing direct evidence for assessing the water-sealing safety of the cavern and ensuring its sealing performance and the safety of the groundwater environment.

[0033] A coordinated monitoring network of water curtain orifice-piezometer-water level monitoring orifice, such as Figure 2 , Figure 3As shown: from the construction period water curtain hole pressure and flow, osmotic pressure monitoring data screening collection, the average value of each week monitoring data. Select the average flow and pressure data of 8 water curtain holes near the vertical osmotic pressure gauge arrangement position as input characteristics, and 5 osmotic pressure gauges in the vertical direction as output targets. From the water level monitoring hole and osmotic pressure gauge data in the operation period, screening and collection are carried out, and the average value of each week monitoring data is taken. The average water level data of the permanent water level monitoring hole around the cavern is selected as the input characteristic, and the 5 osmotic pressure gauges in the vertical direction are selected as the output target.

[0034] S2, integrate the data into a data set and pre-process;

[0035] As an embodiment of the embodiment of the application, the data is integrated into a data set, and data cleaning, creation of time series window and the like are generated to generate a training sample preprocessing step. Specifically, data preprocessing includes:

[0036] Data cleaning: checking and processing missing values, abnormal values or error values;

[0037] Creating a time series window: setting the window size to 4, taking the data in each window as the input feature of the model, and the corresponding target value as the output label of the model.

[0038] The pre-processed data set is divided into a training set and a test set, and the ratio is 7:3.

[0039] S3, using the pre-processed data set to train the LSTM-KAN model (cavern water sealing evaluation time series model);

[0040] In the LSTM-KAN, the KAN network is fused with the LSTM model to overcome the shortcomings of the LSTM in processing complex nonlinear relationships and high-dimensional data. In the model training process, the pre-processed data set is divided into a training set and a test set. The training set is used to train the LSTM-KAN model to learn the patterns and rules in the data; the test set is used to evaluate the performance and generalization ability of the model.

[0041] Specifically, the training steps of the LSTM-KAN model are as follows:

[0042] Data preparation: the pre-processed data set is randomly divided into a training set and a test set, and the ratio is 7:3;

[0043] Model initialization: initialize the parameters of the LSTM model and the KAN network layer;

[0044] Loss function and optimizer selection: using mean square error (MSE) as the loss function and Adam as the optimizer;

[0045] Model training: input the training set data into the model, calculate the predicted value and loss value by forward propagation, and then calculate the parameter gradient using the back propagation algorithm. The optimizer updates the parameters according to the gradient;

[0046] Model evaluation: use the test set to evaluate the trained model and calculate the performance indicators.

[0047] Further, as shown in the following formula, the update process of the LSTM model can be simplified: Figure 4

[0048]

[0049] Where i t , f t , o t are the outputs of the input gate, the forget gate and the output gate, respectively; is the candidate memory cell state; C t is the updated memory cell state; h t is the output hidden state; W and b are the weight matrix and bias term, respectively; σ is the sigmoid activation function; and ⊙ represents element-wise multiplication.

[0050] Further, based on the Kolmogorov-Arnold theorem, the KAN network can decompose a multi-variable function into a linear combination of single-variable functions, and its basic form is:

[0051]

[0052] S4, using the trained LSTM-KAN model to predict and evaluate the water sealing performance of the underground water-sealed cavern.

[0053] Further, the performance of the optimized LSTM-KAN model in evaluating the water sealing performance of the underground water-sealed cavern is analyzed. Through the performance evaluation indicators of the model, including the coefficient of determination (R 2 ), mean absolute error (MAE), and mean absolute percentage error (MAPE).

[0054]

[0055] Specifically, taking the osmotic pressure data of PA1-1 as the prediction object, selecting the flow and pressure data of the water curtain hole near the vertical measurement of PA1-X as the input features, a 62-week data set with a time span from October 2023 to December 2024 and a week as a unit is constructed, and it is randomly divided into training set and test set according to the ratio of 7:3. LSTM and LSTM-KAN models are used for training, and the loss value change trend of the models within 500 iterations is analyzed. The performance difference of the two models at different stages can be clearly seen, as shown in​Figure 5 As shown in the figure, the results demonstrate that the LSTM-KAN model outperforms the LSTM model in both the speed of loss descent and the final loss level, further validating the superior predictive performance of the LSTM-KAN model. The LSTM-KAN model not only exhibits faster convergence during training but also demonstrates higher accuracy and stability in its final predictive performance.

[0056] In terms of prediction performance, the LSTM-KAN model performs better than MAE, MAPE, and R. 2 In key metrics, it outperforms the LSTM model. Specifically, the MAE of the LSTM-KAN model is 0.0149, lower than the 0.0182 of the LSTM model; in terms of MAPE, the LSTM-KAN model is 2.3842%, lower than the 2.8976% of the LSTM model; and in R... 2 In terms of accuracy, the LSTM-KAN model scored 0.8762, higher than the LSTM model's 0.8400. This indicates that the LSTM-KAN model in this embodiment outperforms the traditional LSTM model in terms of prediction accuracy, stability, and fit, enabling it to more accurately assess the water seal performance of underground water-sealed caverns and provide a more reliable basis for the safe operation of the caverns.

[0057] Based on the embodiments of the present invention, the generalization performance of the LSTM-KAN model was further verified, such as... Figure 6 As shown, based on a cavern project under construction, 53 weeks of data from 2024 were selected. PA2-1 to PD2-1 were chosen as output targets, and the pressure and flow data of eight water curtain orifices near each target were selected as input features to construct a time series prediction dataset. These four locations are located in different areas of the water curtain tunnel and can represent the water seal condition of the oil storage cavern within a certain range.

[0058] By making predictions at these four locations, water seal information for different areas of the water curtain tunnel can be obtained, thus comprehensively evaluating the water seal performance of the entire oil storage chamber. Table 1 shows a comprehensive evaluation of the prediction performance of different models (PA2-1, PB2-1, PC2-1, and PD2-1). The PC2-1 model performs best in terms of prediction accuracy, stability, and fit, making it the best performing of the four models. Although the PB2-1 model performs well in terms of MAE, its high MAPE indicates poor prediction stability. The PD2-1 model has the highest MAE but lower prediction accuracy, although its MAPE and R² are relatively high. 2 The performance is acceptable. The PA2-1 model shows relatively balanced performance across various metrics and good predictive capabilities.

[0059] Table 1

[0060] Evaluation index PA2-1 PB2-1 PC2-1 PD2-1 MAE 0.00467 0.00092 0.00246 0.00772 MAPE 1.5872％ 37.885％ 2.1008％ 1.7103％ [R 2 ]] 0.8768 0.7445 0.8968 0.8555

[0061] Therefore, the framework based on multi-location seepage pressure data prediction can comprehensively evaluate the water seal performance of underground water-sealed caverns, providing reliable technical support for the long-term safe operation of such caverns. By making predictions at different locations, the local bias that may result from predictions at a single location can be effectively avoided, ensuring a comprehensive assessment of the water seal performance of oil storage caverns. This embodiment fully demonstrates the effectiveness and reliability of the method of the present invention in practical applications, proving its broad application prospects in the field of water seal performance evaluation of underground water-sealed caverns.

[0062] Based on the embodiments of this invention, the parameters of the LSTM-KAN model were optimized and adjusted. By adjusting parameters such as the number of neurons in the LSTM layer, the structure of the KAN network layer, and the activation function, the predictive performance of the model was further improved. For example, the number of neurons in the LSTM layer was increased from 128 to 256, and the structure of the KAN network layer was adjusted to better adapt to the characteristics of water-sealed cavern data. The optimized LSTM-KAN model showed a significant improvement in predictive performance, with its MAE reduced to 0.0135, MAPE reduced to 2.1534%, and R... 2 The accuracy was increased to 0.8927. This indicates that the method of the present invention has a certain degree of flexibility and scalability. By optimizing and adjusting the model parameters, the prediction accuracy and stability of the model can be further improved, better meeting the needs of practical applications.

[0063] Example 2:

[0064] This invention also provides a system for evaluating the water-sealing performance of underground water-sealed caverns, comprising:

[0065] The first processing module is used to construct a dataset that includes weekly average monitoring data of multiple water curtain holes near the target vertical piezometer during the construction period as input features and weekly average piezometer data as output targets.

[0066] The second processing module is used to clean the dataset and create a sliding time window to generate training samples; divide the training set and the test set; construct and train the LSTM-KAN prediction model that integrates the Long Short-Term Memory Network (LSTM) and the Kolmogorov-Arnold Network (KAN); where LSTM is used to capture long-term dependencies in time series and KAN network is used to enhance the fitting ability of complex nonlinear relationships.

[0067] The third processing module is used to input the water level monitoring data and piezometer seepage pressure values ​​during operation into the trained LSTM-KAN prediction model to evaluate the water seal performance.

[0068] As one embodiment of the present invention, the monitoring data of the water curtain orifice includes weekly flow and pressure data.

[0069] As one embodiment of the present invention, the sliding time window size is 4 weeks.

[0070] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for evaluating the water-sealing performance of underground water-sealed caverns, characterized in that, include: Step S1: Construct a dataset that includes weekly average monitoring data of multiple water curtain holes near the target vertical piezometer during the construction period as input features and weekly average piezometer pressure data as output targets. Step S2: Clean the dataset and create a sliding time window to generate training samples; divide the training set and test set; construct and train the LSTM-KAN prediction model that integrates the Long Short-Term Memory Network (LSTM) and the Kolmogorov-Arnold Network (KAN); where LSTM is used to capture long-term dependencies in time series and KAN network is used to enhance the fitting ability of complex nonlinear relationships. Step S3: Input the water level monitoring data and piezometer pressure values ​​during operation into the trained LSTM-KAN prediction model to evaluate the water seal performance.

2. The method for evaluating the water-sealing performance of underground water-sealed caverns as described in claim 1, characterized in that, The monitoring data from the water curtain orifice includes weekly average flow and pressure data.

3. The method for evaluating the water-sealing performance of underground water-sealed caverns as described in claim 1, characterized in that, The sliding time window size is 4 weeks.

4. A system for evaluating the water-sealing performance of underground water-sealed caverns, characterized in that, include: The first processing module is used to construct a dataset that includes weekly average monitoring data of multiple water curtain holes near the target vertical piezometer during the construction period as input features and weekly average piezometer data as output targets. The second processing module is used to clean the dataset and create a sliding time window to generate training samples; divide the training set and the test set; construct and train the LSTM-KAN prediction model that integrates the Long Short-Term Memory Network (LSTM) and the Kolmogorov-Arnold Network (KAN); where LSTM is used to capture long-term dependencies in time series and KAN network is used to enhance the fitting ability of complex nonlinear relationships. The third processing module is used to input the water level monitoring data and piezometer seepage pressure values ​​during operation into the trained LSTM-KAN prediction model to evaluate the water seal performance.

5. The underground water-sealed cavern water-sealing performance evaluation system as described in claim 4, characterized in that, The monitoring data from the water curtain orifice includes weekly average flow and pressure data.

6. The underground water-sealed cavern water-sealing performance evaluation system as described in claim 5, characterized in that, The sliding time window size is 4 weeks.

Images