Dam deformation prediction method, device, storage medium and product
By performing deep feature extraction and cluster analysis on dam monitoring data, a Transformer-LSTM model was constructed, which solved the problems of complex multivariate data processing and insufficient information utilization in traditional dam deformation prediction methods, and achieved high-precision dam deformation prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- POWERCHINA ZHONGNAN ENG
- Filing Date
- 2025-04-28
- Publication Date
- 2026-06-26
AI Technical Summary
Traditional methods for predicting dam deformation suffer from problems such as complex multivariate data processing, insufficient information utilization, and slow response speed, resulting in poor prediction accuracy.
We acquired dam monitoring data, preprocessed it, and then performed deep feature extraction and cluster analysis. We constructed a Transformer-LSTM model to predict dam deformation, optimized the initial cluster centers using deep fuzzy clustering and particle swarm optimization algorithms, and found the optimal hyperparameters using a grid search algorithm.
The accuracy of dam deformation prediction has been improved by mining nonlinear relationships through deep learning algorithms, which has enhanced the accuracy of sample dataset construction and prediction.
Smart Images
Figure CN120492800B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of water conservancy engineering dam deformation monitoring technology, and particularly relates to a dam deformation prediction method, equipment, storage medium and product. Background Technology
[0002] As a crucial component of water conservancy projects, the structural safety of dams directly impacts the safety and socio-economic development of downstream areas. Therefore, dam deformation monitoring is of paramount importance. Deformation monitoring primarily involves real-time or periodic monitoring and analysis of deformation data of the dam body and its surrounding environment to promptly identify potential safety hazards and implement appropriate measures.
[0003] Traditional deformation prediction methods mainly include statistical methods, finite element analysis, and shallow machine learning methods. However, traditional methods have the following problems:
[0004] (1) Multivariate data processing is complex and it is difficult to analyze the data characteristics of the dam during different operating periods, resulting in poor prediction accuracy;
[0005] (2) Insufficient information utilization makes it difficult to capture the deep nonlinear relationship between long-term series deformation and monitoring factors, resulting in poor prediction accuracy;
[0006] (3) The response speed is slow, the model training time is long, and it is difficult to make predictions quickly.
[0007] These problems lead to significant limitations in the practical application of traditional dam deformation monitoring and prediction methods, making it difficult to meet the requirements of high reliability. Therefore, the research and development of new deformation monitoring and prediction methods is particularly urgent and important. Summary of the Invention
[0008] The purpose of this invention is to provide a method, device, storage medium and product for predicting dam deformation, so as to solve the problem of poor prediction accuracy of traditional prediction methods.
[0009] This invention solves the above-mentioned technical problems through the following technical solution: a method for predicting dam deformation, comprising:
[0010] Acquire dam monitoring data; wherein, the dam monitoring data includes reservoir water level, environmental data, time-dependent factors, and dam measuring point displacement; the time-dependent factors refer to the stepwise regression equation of the time-dependent parameters, and the time-dependent parameters refer to the rate of change of the time from the monitoring time to the start time of dam operation;
[0011] The dam monitoring data is preprocessed;
[0012] Deep feature extraction is performed on the preprocessed environmental data to obtain environmental features;
[0013] Cluster analysis was performed on the pre-treated reservoir water level and the environmental characteristics, and the time-domain division of the time-effect factors and the displacement of the dam measuring points was carried out based on the cluster analysis results to obtain the cluster division results for different time periods.
[0014] A sample dataset is constructed based on each clustering result; the number of sample datasets is the same as the number of clustering results.
[0015] Construct a Transformer-LSTM model; wherein, the Transformer-LSTM model is obtained by replacing the decoder of the Transformer model with an LSTM.
[0016] The Transformer-LSTM model was trained and tested using each sample dataset to obtain the dam deformation prediction model for each time period.
[0017] Obtain the data to be predicted, and use the dam deformation prediction model corresponding to the time period to which the data to be predicted belongs to predict the data to obtain the dam deformation prediction result.
[0018] Furthermore, the environmental data includes air temperature, water temperature, dam surface temperature, and dam foundation temperature.
[0019] Furthermore, the Transformer model is used to extract deep features from the preprocessed environmental data to obtain environmental features.
[0020] Furthermore, a deep fuzzy clustering algorithm is used to perform cluster analysis on the preprocessed reservoir water level and the environmental features.
[0021] Furthermore, when using the deep fuzzy clustering algorithm to perform cluster analysis on the preprocessed reservoir water level and the environmental features, the particle swarm optimization algorithm is used to optimize the randomly assigned initial cluster centers to avoid the deep fuzzy clustering algorithm getting trapped in local optima.
[0022] Furthermore, during the training of the Transformer-LSTM model, a grid search algorithm is used to globally solve the hyperparameters of the Transformer-LSTM model based on the sample dataset in order to find the optimal hyperparameters of the Transformer-LSTM model.
[0023] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program / instructions stored in the memory, wherein the processor executes the computer program / instructions to implement the dam deformation prediction method as described above.
[0024] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the dam deformation prediction method as described above.
[0025] Based on the same concept, the present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the dam deformation prediction method as described above.
[0026] Beneficial effects
[0027] Compared with the prior art, the advantages of the present invention are as follows:
[0028] This invention performs deep feature mining on environmental data, learns the nonlinear relationship between air temperature and dam-related temperature at different time periods, and then performs cluster analysis on the mined environmental features and reservoir water level, which improves the clustering accuracy and thus improves the accuracy of constructing sample datasets for each time period. It also uses an improved deep learning algorithm (i.e., Transformer-LSTM model) to mine the deep correlation between each clustering result, which improves the prediction accuracy of dam deformation. Attached Figure Description
[0029] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 This is a flowchart of the dam deformation prediction method in an embodiment of the present invention;
[0031] Figure 2 This is a flowchart of cluster analysis and time-domain partitioning in an embodiment of the present invention;
[0032] Figure 3 This is a schematic diagram of the time-domain division of the deformation of the dam measuring points in an embodiment of the present invention;
[0033] Figure 4 This is a structural diagram of the Transformer-LSTM model in an embodiment of the present invention;
[0034] Figure 5 This is a schematic diagram comparing the evaluation results of different models in an embodiment of the present invention; wherein, the vertical axis represents the evaluation index value;
[0035] Figure 6 This is a schematic diagram comparing the prediction effect of the test set with the actual test data in an embodiment of the present invention. Detailed Implementation
[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. 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.
[0037] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.
[0038] Example 1
[0039] Traditional methods for processing large amounts of monitoring data are typically based on linear dimensionality reduction, which fails to effectively capture the nonlinear relationships between different monitoring data points, resulting in poor prediction accuracy. To address these technical problems, this invention provides a dam deformation prediction method based on clustering and a Transformer-LSTM model. Figure 1 A flowchart of the dam deformation prediction method provided by the present invention is shown, as follows: Figure 1 As shown, the dam deformation prediction method includes the following steps:
[0040] Step 1: Obtain dam monitoring data.
[0041] The monitoring data for the dam includes reservoir water level, environmental data, time-dependent factors, and displacement of dam measuring points; environmental data includes air temperature, water temperature, dam surface temperature, and dam foundation temperature; time-dependent factors refer to the stepwise regression equation of time-dependent parameters, which are the rate of change of time from the monitoring time to the start of dam operation, and are specifically expressed as follows:
[0042] θ=(t-t0) / 100(1)
[0043] δ θ =c1θ+c2lnθ(2)
[0044] Where θ represents the time-dependent parameter, t represents the monitoring time, t0 represents the start time of dam operation, and δ θ Here, c1 and c2 represent statistical coefficients, respectively, indicating the time-sensitivity factor. When constructing the sample dataset, only the time-sensitivity parameter θ needs to be obtained; the statistical coefficients c1 and c2 are automatically determined during the training of the Transformer-LSTM model.
[0045] Factors influencing dam deformation typically include hydraulic and temperature factors. However, for concrete dams, the longer the time, the more prone the concrete is to failure. Therefore, aging is also a crucial factor affecting dam deformation. This invention incorporates aging factors into the monitoring data, thereby improving the accuracy of dam deformation prediction.
[0046] At each monitoring moment, a set of dam monitoring data can be acquired. This set of dam monitoring data includes the reservoir water level, air temperature, dam surface temperature, dam foundation temperature, time factor calculated based on the current monitoring moment, and dam measuring point displacement (i.e., dam deformation).
[0047] For example, a dam located on the eastern source of the Pihe River, a tributary of the Huai River, primarily serves flood control, while also providing irrigation, power generation, and navigation. Construction of the dam began in 1954, with a maximum height of 75.9 meters and a crest length of 510 meters. It is equipped with an automatic monitoring system capable of long-term monitoring of data such as deformation, settlement, stress and strain, and temperature (including air, water, and internal temperatures). The dam has 20 cycloidal lines (PL) and 3 inverted cycloidal lines (IP) installed on its body and foundation for deformation monitoring, and numerous thermometers are also installed inside the dam for temperature monitoring.
[0048] Step 2: Preprocess the dam monitoring data.
[0049] To improve data quality, the present invention also preprocesses the dam monitoring data. The preprocessing in this embodiment includes cleaning, interpolation, denoising, and normalization.
[0050] Step 3: Perform deep feature extraction on the preprocessed environmental data to obtain environmental features.
[0051] The environmental data in this embodiment includes air temperature and dam-related temperatures (i.e., dam surface temperature and dam foundation temperature). The variation range of air temperature and the variation range of dam-related temperatures differ across different time periods (e.g., different seasons), indicating a non-linear relationship between them. To further explore the non-linear relationship between air temperature variation and dam-related temperature variation and improve the accuracy of constructing sample datasets for each time period, this invention performs deep feature extraction on the preprocessed environmental data to obtain environmental features.
[0052] In a specific embodiment of the present invention, the Transformer model is used to extract deep features from the preprocessed environmental data to obtain environmental features.
[0053] The Transformer model is a neural network model based on a self-attention mechanism. It consists of a multi-layered encoder and decoder, with each layer containing multiple attention mechanism modules and feedforward neural network modules. The encoder encodes the air temperature, dam surface temperature, and dam foundation temperature at each monitoring moment into a high-dimensional feature vector representation, while the decoder decodes the high-dimensional feature vector representation output by the encoder into environmental features. The Transformer model employs a parallel mechanism, making its structure more lightweight. It also uses techniques such as residual connections and layer normalization to accelerate model convergence and improve performance. By deeply exploring the nonlinear relationship between the magnitude of air temperature changes and the magnitude of changes in dam-related temperatures, the Transformer model facilitates more accurate clustering and the segmentation of different time periods.
[0054] Step 4: Perform cluster analysis on the pre-processed reservoir water level and environmental characteristics, and divide the pre-processed time-dependent factors and dam measuring point displacements into time domains based on the cluster analysis results to obtain cluster division results for different time periods.
[0055] To accurately segment dam monitoring data across different time periods, cluster analysis is performed on the preprocessed reservoir water level and environmental characteristics. Then, based on the cluster analysis results, time-domain factors and dam measuring point displacements are further segmented. In a specific embodiment of this invention, a deep fuzzy C-Means (DF) clustering algorithm is used to perform cluster analysis on the preprocessed reservoir water level and the environmental characteristics obtained in step 3.
[0056] Compared to K-means clustering, deep fuzzy clustering algorithms offer superior clustering performance and improve data partitioning accuracy. However, the clustering results of deep fuzzy clustering are significantly influenced by the initial cluster centers, which are typically randomly assigned based on the data. To avoid the algorithm getting trapped in local optima due to random initial center assignment, this invention uses Particle Swarm Optimization (PSO) to optimize the randomly assigned initial cluster centers.
[0057] In this embodiment, the prediction effect is best when the number of clusters is 4. That is, cluster analysis is performed on the preprocessed reservoir water level and environmental characteristics, resulting in 4 cluster analysis results; then, based on the 4 cluster analysis results, the displacement of the dam measuring points and time-related factors are divided into time domains, resulting in 4 cluster partitioning results, such as... Figure 3As shown, the displacements of measuring points in similar time domains belong to similar clusters. Each clustering result corresponds to a time period, and each clustering result includes the reservoir water level, environmental data, time-related factors, and measuring point displacements for that time period.
[0058] Step 5: Construct a sample dataset based on the results of each clustering.
[0059] To improve the accuracy of dam deformation prediction, a separate dam deformation prediction model is needed for each time period. Therefore, a sample dataset needs to be constructed based on each clustering result, with the number of sample datasets matching the number of clustering results. In this embodiment, there are four clustering results, resulting in four sample datasets. Each sample dataset is further divided into a training set and a test set to facilitate the subsequent training and testing of the Transformer-LSTM model.
[0060] Step 6: Construct the Transformer-LSTM model.
[0061] The Transformer model employs a parallel mechanism, making its structure more lightweight. However, when applied to time series data prediction, the input sequence may lack temporal information, affecting the prediction results. To address this issue, this invention replaces the attention mechanism module in the Transformer model's decoder with an LSTM (Long Short-Term Memory) network to obtain a Transformer-LSTM model, as follows: Figure 4 As shown, LSTM, as a recurrent neural network, is more suitable for modeling time series data prediction tasks. It can better capture the dynamic patterns of dam deformation changes and improve the accuracy of dam deformation prediction.
[0062] Step 7: Train and test the Transformer-LSTM model using each sample dataset to obtain the dam deformation prediction model for each time period.
[0063] This embodiment uses four sample datasets. The Transformer-LSTM model is trained and tested using each sample dataset, resulting in a dam deformation prediction model corresponding to the time period of that sample dataset. This yields four dam deformation prediction models. In a specific implementation of this invention, during training, a grid search algorithm is used to globally solve for the hyperparameters of the Transformer-LSTM model based on the sample datasets, seeking the optimal hyperparameters.
[0064] After training, each sample in the test set is input into the trained Transformer-LSTM model to obtain prediction results. All prediction results are then summarized in chronological order, and the trained Transformer-LSTM model is evaluated using an evaluation metric to obtain the dam deformation prediction model. In this embodiment, the evaluation metric used is the coefficient of determination R0. 2 Root mean square error (RMSE) and mean absolute percentage error (MAPE).
[0065] Figure 5 The comparison of evaluation results between the model of this invention (i.e., the DF-Transformer-LSTM model) and other models is shown. The DF-Transformer model represents the model obtained based on cluster analysis and Transformer, the DF-LSTM model represents the model obtained based on cluster analysis and LSTM, and the Transformer-LSTM model represents the model obtained based on Transformer-LSTM. Figure 5 It can be seen that the DF-Transformer-LSTM model of this invention has the best performance across all evaluation metrics. In the comparison of various models on the test set, the DF-Transformer-LSTM model outperforms the Transformer-LSTM model, indicating that the clustering method effectively analyzes the characteristics of environmental data and effectively improves prediction accuracy. The DF-Transformer-LSTM model outperforms the DF-Transformer model, demonstrating that the Transformer-LSTM model built based on the Transformer effectively improves prediction accuracy.
[0066] Figure 6 A schematic diagram comparing the prediction results of the test set with the measured data at the test points is shown. Figure 6 It can be seen that the prediction results of the present invention are similar to the measured data, indicating that it has high accuracy in predicting dam deformation.
[0067] Step 8: Obtain the data to be predicted, and use the dam deformation prediction model corresponding to the time period to which the data to be predicted belongs to predict the data to obtain the dam deformation prediction result.
[0068] The data to be predicted includes reservoir water level, environmental data, and time-related factors. By inputting the data to be predicted into the dam deformation prediction model corresponding to the time period to which the data belongs, the dam displacement can be output, thus realizing the prediction of dam deformation.
[0069] Example 2
[0070] This invention also provides an electronic device, which includes a memory, a processor, and a computer program / instructions stored in the memory. The processor executes the computer program / instructions to implement the dam deformation prediction method in this application embodiment.
[0071] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.
[0072] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.
[0073] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the dam deformation prediction method of the present application embodiments.
[0074] Readable storage media include both permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient media, such as modulated data signals and carrier waves.
[0075] Although not shown, embodiments of the present invention also provide a computer program product, including: a computer program / instructions that, when executed by a processor, implement the dam deformation prediction method in the embodiments of this application.
[0076] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting dam deformation, characterized in that, The prediction method includes: Acquire dam monitoring data; wherein, the dam monitoring data includes reservoir water level, environmental data, time-dependent factors, and dam measuring point displacement; the time-dependent factors refer to the stepwise regression equation of the time-dependent parameters, and the time-dependent parameters refer to the rate of change of the time from the monitoring time to the start time of dam operation; The dam monitoring data is preprocessed; Deep feature extraction is performed on the preprocessed environmental data to obtain environmental features; The deep fuzzy clustering algorithm is used to perform cluster analysis on the preprocessed reservoir water level and the environmental features. Based on the cluster analysis results, the time-domain factors and the displacement of the dam measuring points are divided into time domains to obtain the cluster division results for different time periods. In the process of using the deep fuzzy clustering algorithm to perform cluster analysis on the preprocessed reservoir water level and the environmental features, the particle swarm optimization algorithm is used to optimize the randomly assigned initial cluster centers. A sample dataset is constructed based on each clustering result; the number of sample datasets is the same as the number of clustering results. Construct a Transformer-LSTM model; wherein, the Transformer-LSTM model is obtained by replacing the decoder of the Transformer model with an LSTM. The Transformer-LSTM model was trained and tested using each sample dataset to obtain the dam deformation prediction model for each time period. Obtain the data to be predicted, and use the dam deformation prediction model corresponding to the time period to which the data to be predicted belongs to predict the data to obtain the dam deformation prediction result.
2. The dam deformation prediction method according to claim 1, characterized in that, The environmental data includes air temperature, water temperature, dam surface temperature, and dam foundation temperature.
3. The dam deformation prediction method according to claim 1, characterized in that, The Transformer model is used to extract deep features from the preprocessed environmental data to obtain environmental features.
4. The method for predicting dam deformation according to any one of claims 1 to 3, characterized in that, During the training of the Transformer-LSTM model, the hyperparameters of the Transformer-LSTM model are solved globally using a grid search algorithm based on the sample dataset.
5. An electronic device comprising a memory, a processor, and a computer program / instructions stored in the memory, characterized in that, The processor executes the computer program / instructions to implement the dam deformation prediction method as described in any one of claims 1 to 4.
6. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the dam deformation prediction method as described in any one of claims 1 to 4.
7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the dam deformation prediction method as described in any one of claims 1 to 4.