GNSS-assisted dam displacement monitoring algorithm

Abstract

The invention relates to the field of hydraulic engineering safety, and discloses a GNSS-assisted dam displacement monitoring algorithm, which comprises the following steps: S1, data collection: collecting three-dimensional coordinate time sequence data and environmental data of each measuring point on a dam; s2, data preprocessing: using a low-pass filtering method to remove noise, identifying and removing abnormal values, adopting linear interpolation to fill missing data, and using Z-score standardization to zoom the data to the same range; s3, feature engineering: extracting the characteristics of displacement speed, acceleration and the like, analyzing the periodic change in the data, selecting the characteristics which have the most influence on displacement prediction, and reducing redundancy. According to the GNSS-assisted dam displacement monitoring algorithm, various environmental factors such as humidity, water level and seismic activity are comprehensively incorporated into the prediction model, the actual working state of the dam can be reflected more comprehensively, the accuracy and reliability of the algorithm are enhanced, and the accuracy and reliability of the dam displacement monitoring algorithm are improved by analyzing the interaction among different factors. The algorithm can better capture the internal law of dam deformation.

Description

Technical Field

[0001] The present invention relates to the field of water conservancy project safety, and specifically to a GNSS-assisted dam displacement monitoring algorithm. Background Art

[0002] GNSS (Global Navigation Satellite System) is a technical system that uses satellites for global positioning and navigation. GNSS sends signals to ground receivers through a group of satellites operating in different orbits, and calculates information such as the geographical location, speed, and time of an object using the principle of triangulation; the dam displacement monitoring algorithm is a calculation and analysis algorithm used to monitor and predict the displacement or deformation of the dam structure in real time, used to evaluate the safety performance of the dam, and timely detect potential structural problems, so as to take corresponding maintenance and repair measures;

[0003] GNSS provides high-precision geolocation data, which is crucial for dam displacement monitoring. GNSS technology plays a key role in dam displacement monitoring, providing high-precision, real-time, and sustainable data support. Combined with the dam displacement monitoring algorithm, GNSS technology makes the monitoring work more efficient and accurate, providing a scientific basis for the safety and management of the dam, and is an important tool for ensuring the safety of the dam;

[0004] Existing dam displacement monitoring algorithms ignore many problems, such as the factor of humidity. Specifically, dam materials such as concrete will expand and contract when the humidity changes, resulting in deformation. High humidity will increase the water content of the foundation soil, causing the soil to expand and affecting the supporting structure of the dam, resulting in deformation; humidity changes are often associated with temperature changes, forming a complex pattern that affects dam deformation; similarly, for the dam displacement monitoring algorithm, the factor of water level is also crucial. An increase in water level will increase water pressure, which will cause the dam to move downstream. Changes in water level will affect the seepage pressure of the dam body and the foundation, and thus affect deformation. Changes in water temperature may affect the expansion and contraction of the dam body material, and may also cause deformation; there is also earthquake. Earthquakes will cause the dam structure to vibrate, resulting in deformation, damaging the dam body, and affecting its long-term stability. Existing prediction systems do not consider these potential problems, and there is much room for improvement. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] In view of the deficiencies of the prior art, the present invention provides a GNSS-assisted dam displacement monitoring algorithm to solve the above problems.

[0007] (2) Technical Solutions

[0008] To achieve the above object, the present invention provides the following technical solution: A GNSS-assisted dam displacement monitoring algorithm, comprising the following steps:

[0009] S1. Data collection: Collect the three-dimensional coordinate time series data and environmental data of each measuring point on the dam;

[0010] S2. Data preprocessing: Use low-pass filtering method to remove noise, identify and remove outliers, adopt linear interpolation to fill in missing data, and use Z-score normalization to scale the data to the same range;

[0011] S3. Feature engineering: Extract features such as the velocity and acceleration of displacement, analyze the periodic changes in the data, select the features that have the most influence on displacement prediction, and reduce redundancy;

[0012] S4. Model design: Design a multi-layer LSTM structure, each layer contains an appropriate number of units, and add a dropout layer to prevent overfitting. Add a fully connected layer after the LSTM to further extract features;

[0013] S5. Model training: Select the Adam optimizer, use the mean square error as the loss function, monitor the validation set loss, and prevent overfitting;

[0014] S6. Model evaluation and tuning: Divide the data set into a training set and a validation set, perform cross-validation, use indicators such as MSE, RMSE, and R-squared value to evaluate the model performance, and adjust hyperparameters such as the learning rate and batch size to optimize the model performance;

[0015] S7. Anomaly detection: Set a threshold to detect displacement anomalies, and use algorithms such as Isolation Forest for anomaly detection;

[0016] S8. Data storage and transmission: Use a database to efficiently store data, ensure the security and reliability of the data, and adopt the MQTT or HTTPS protocol to ensure the efficiency and security of data transmission;

[0017] S10. Model interpretability: Use LIME or SHAP to explain the decision-making process of the LSTM model and improve the interpretability of the model.

[0018] Preferably, determine the previous and next known data points of the missing data point. If the displacement data at time t 2 is missing, find the displacement data x 1 at t 3 and x 1 at t 3 , calculate the time difference Δt total =t 3 -t 1 , calculate the time difference Δt between the missing point and the previous known point1 = t 2 -t 1 , calculate the value of the missing point according to the linear relationship This formula means that the change in displacement is distributed according to the ratio of time; for multiple consecutive missing values, linear interpolation is performed in segments.

[0019] Preferably, extract the velocity and acceleration of the displacement, and specifically analyze the periodic changes in the data. Specifically, the displacement data is evenly sampled and the missing values are processed. The interpolation method is used to fill in the missing values, and the displacement data is smoothed to reduce the influence of noise. A low-pass filter is used, and the velocity is estimated by numerical differentiation: where dt is the time interval; the velocity sequence is further smoothed, and the acceleration is estimated by numerical differentiation: where dt is the time interval; the acceleration sequence is further smoothed, the Fourier transform is performed on the displacement time series to obtain the spectrum, and the frequencies corresponding to the peaks in the power spectrum are found. The time periods corresponding to these frequencies are the periodic changes in the data. The wavelet transform is used for multi-scale analysis to identify the periodicity at different time scales.

[0020] Preferably, LSTM layer design:

[0021] Number of layers: Select 2 - 3 layers of LSTM to capture temporal features at different levels;

[0022] Number of units: Each layer of LSTM contains 64 - 128 units, adjusted according to the data complexity;

[0023] Dropout layer: Add a Dropout layer between the LSTM layers, set the dropout rate to 0.2 - 0.5 to prevent overfitting;

[0024] Fully connected layer design:

[0025] Number of layers: Add 1 - 2 fully connected layers;

[0026] Number of units: The first fully connected layer contains 64 neurons, and the second fully connected layer contains 32 neurons;

[0027] Activation function: Use the ReLU activation function to improve the nonlinear fitting ability of the model;

[0028] Output layer:

[0029] Number of units: The number of units in the output layer depends on the task. If it is a regression task, output one value; if it is a multi-step prediction, output multiple values;

[0030] Activation function: Linear activation function (no activation function), directly output the predicted value.

[0031] Preferably, the specific process of model training is as follows:

[0032] Select the Adam optimizer, set the initial learning rate to 0.001, use the mean squared error (MSE) as the loss function, which is suitable for regression tasks, use the root mean squared error to evaluate the model performance, batch size: select 64, adjust according to the computing resources, train for 100 - 200 epochs, adjust according to the performance on the validation set. At the end of each epoch, calculate the loss of the validation set, monitor the generalization ability of the model. When the validation loss does not improve for several consecutive epochs, stop training to prevent overfitting. Use the ModelCheckpoint callback function to save the model with the lowest validation loss;

[0033] The specific steps of the Isolation Forest algorithm for anomaly detection are as follows:

[0034] Construct 500 isolation trees, set the maximum depth of each tree, specify the proportion of abnormal samples in the data, sample the data with replacement to generate multiple different training sets, construct an isolation tree on each training set, split the data by randomly selecting features and randomly selecting feature values. For each sample, calculate the path length (i.e., the number of splits) by which it is isolated on all isolation trees. According to the path length, calculate the anomaly score. The shorter the path length, the higher the anomaly score, indicating that it is more likely to be an anomaly point. According to the distribution of the anomaly scores and business requirements, set a threshold to distinguish normal and abnormal samples, and compare the anomaly score of each sample with the threshold. Samples exceeding the threshold are marked as abnormal.

[0035] Preferably, collect historical humidity data and corresponding dam deformation data, conduct a correlation analysis to determine the relationship between humidity and deformation, use a non - linear model to capture complex relationships, pre - process the humidity data to ensure data quality, construct lag features and rolling averages to reflect the temporal impact of humidity on deformation, consider the cumulative effect of humidity change trends on deformation, introduce time - series analysis methods, evaluate the improvement of humidity on model performance by comparing the prediction effects of models with and without humidity variables, ensure that the acquisition frequency of humidity data matches the dam deformation monitoring frequency, and perform data interpolation or estimation if necessary.

[0036] Preferably, real-time water level data is collected as one of the inputs of the prediction algorithm. Relevant features are extracted according to different water level ranges, such as the water level change rate and historical water level fluctuations, to enhance the prediction ability of the algorithm. For the different characteristics of high water level, medium water level, and low water level, multiple sub-models are constructed. Each sub-model is specifically trained for the data within a specific water level range. The water level change may cause non-linear responses of the dam materials. A neural network is used to capture this non-linear relationship. Through the correlation analysis of historical water level and deformation data, potential patterns and laws between water level changes and dam deformation are mined to provide more useful training data for the algorithm. The weights of the prediction model are adjusted in a timely manner through water level changes to ensure the response ability and early warning ability of the algorithm during drastic water level changes. Uncertainty factors such as water level fluctuations caused by climate change are considered in the model design, and the robustness of the model under different water levels is evaluated through methods such as Monte Carlo simulation.

[0037] Preferably, real-time earthquake monitoring data is accessed, including the epicenter location, magnitude, depth, and vibration intensity. These data are used as input features of the algorithm. Historical earthquake activity data is utilized to analyze the relationship between it and dam deformation, and potential laws and features are extracted for model training. A finite element model is used to simulate the propagation of seismic waves and their impact on the dam, capturing the instantaneous deformation of the dam under seismic action. The time series data is analyzed to enable the model to identify the deformation patterns before and after an earthquake. The attenuation characteristics of seismic waves during propagation are analyzed, and considering factors such as the distance between the dam and the epicenter, the impact of a specific dam under seismic activity is evaluated. Earthquake activities are incorporated into the considerations of dam design to predict the possible maximum deformation to ensure the safety of the dam under extreme seismic conditions. After the dam has experienced one or more earthquakes, the deformation data is collected and fed back into the model.

[0038] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable medium. Computer-executable instructions are stored in the storage medium. When the computer-executable instructions are loaded and executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

[0039] According to another aspect of the embodiments of the present invention, there is also provided a computer program product. The computer program product is tangibly stored on a non-transitory computer-readable medium and includes machine-executable instructions. When the machine-executable instructions are executed, the machine executes the steps of the method according to any one of claims 1 to 8.

[0040] (III) Beneficial effects

[0041] Compared with the prior art, the present invention provides a GNSS-assisted dam displacement monitoring algorithm, which has the following beneficial effects:

[0042] The GNSS-assisted dam displacement monitoring algorithm incorporates various environmental factors such as humidity, water level, and seismic activity into the prediction model, which can more comprehensively reflect the actual working state of the dam, enhance the accuracy and reliability of the algorithm. By analyzing the interactions between different factors, the algorithm can better capture the internal laws of dam deformation, improve the prediction accuracy, and timely identify potential deformation risks. By real-time monitoring and analyzing the data of humidity, water level, and seismic activity, it can identify in advance the potential safety hazards faced by the dam, provide a scientific basis for formulating risk control strategies. The algorithm can handle the changes in real-time data, implement dynamic adjustment and optimization, ensure that in the face of extreme weather, earthquakes and other emergencies, it can quickly respond and conduct deformation prediction. Using the prediction results generated by the algorithm, it can provide a basis for engineers and managers to help them formulate targeted maintenance and emergency plans to ensure the safe operation of the dam. Through the comprehensive analysis of humidity, water level, and seismic activity, it helps to conduct scientific water resource management and ecological environment protection, and promote the sustainable development of the areas around the dam. The algorithm can be used as part of a long-term monitoring system, accumulating data over time, which helps to establish a long-term monitoring and evaluation mechanism for dam deformation and form a closed-loop management. Detailed implementation manners

[0043] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0044] It should be understood that in various embodiments of the present invention, the order numbers of the various processes do not mean the order of execution, and the execution order of the various processes should be determined by their functions and internal logics, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0045] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0046] A GNSS-assisted dam displacement monitoring algorithm includes the following steps:

[0047] S1. Data collection: Collect the three-dimensional coordinate time series data and environmental data of each measuring point on the dam;

[0048] S2. Data Preprocessing: Use low-pass filtering method to remove noise, identify and remove outliers, adopt linear interpolation to fill in missing data, and use Z-score normalization to scale the data to the same range;

[0049] S3. Feature Engineering: Extract features such as velocity and acceleration of displacement, analyze the periodic changes in the data, select the features that have the most influence on displacement prediction, and reduce redundancy;

[0050] S4. Model Design: Design a multi-layer LSTM structure, with each layer containing an appropriate number of units, and add a dropout layer to prevent overfitting. Add a fully connected layer after the LSTM to further extract features;

[0051] S5. Model Training: Select the Adam optimizer, use the mean squared error as the loss function, monitor the validation set loss, and prevent overfitting;

[0052] S6. Model Evaluation and Tuning: Divide the dataset into a training set and a validation set, perform cross-validation, use metrics such as MSE, RMSE, and R-squared value to evaluate the model performance, and adjust hyperparameters such as the learning rate and batch size to optimize the model performance;

[0053] S7. Anomaly Detection: Set a threshold to detect displacement anomalies, and use algorithms such as Isolation Forest for anomaly detection;

[0054] S8. Data Storage and Transmission: Use a database to efficiently store data, ensure the security and reliability of the data, and adopt protocols such as MQTT or HTTPS to ensure the efficiency and security of data transmission;

[0055] S10. Model Interpretability: Use LIME or SHAP to explain the decision-making process of the LSTM model and improve the interpretability of the model.

[0056] It should be noted that before determining the missing data points, find the previous and next known data points. If the displacement data at time t 2 is missing, find the displacement data x 1 and x 3 at t 1 and t 3 , calculate the time difference Δt total =t 3 -t 1 between the previous and next known data points, calculate the time difference Δt 1 =t 2 -t 1 between the missing point and the previous known point, and calculate the value of the missing point according to the linear relationship This formula means that the change in displacement is allocated according to the proportion of time; for multiple consecutive missing values, linear interpolation is performed in segments.

[0057] It should be noted that to extract the velocity and acceleration of displacement, the periodic changes in the data are specifically analyzed. Specifically, the displacement data is evenly sampled and missing values are processed. The interpolation method is used to fill in the missing values, and the displacement data is smoothed to reduce the influence of noise. A low-pass filter is used, and the velocity is estimated through numerical differentiation: where dt is the time interval; the velocity sequence is further smoothed, and the acceleration is estimated through numerical differentiation: where dt is the time interval; the acceleration sequence is further smoothed, the Fourier transform is performed on the displacement time series to obtain the frequency spectrum, and the frequencies corresponding to the peaks in the power spectrum are found. The time periods corresponding to these frequencies are the periodic changes in the data. The wavelet transform is used for multi-scale analysis to identify the periodicity at different time scales.

[0058] It should be noted that the design of the LSTM layer:

[0059] Number of layers: Select 2 - 3 layers of LSTM to capture the time series features at different levels;

[0060] Number of units: Each layer of LSTM contains 64 - 128 units, which is adjusted according to the data complexity;

[0061] Dropout layer: A Dropout layer is added between the LSTM layers, and the dropout rate is set to 0.2 - 0.5 to prevent overfitting;

[0062] Design of the fully connected layer:

[0063] Number of layers: Add 1 - 2 fully connected layers;

[0064] Number of units: The first fully connected layer contains 64 neurons, and the second fully connected layer contains 32 neurons;

[0065] Activation function: Use the ReLU activation function to improve the non-linear fitting ability of the model;

[0066] Output layer:

[0067] Number of units: The number of units in the output layer depends on the task. If it is a regression task, one value is output; if it is a multi-step prediction, multiple values are output;

[0068] Activation function: Linear activation function (no activation function), directly output the predicted value.

[0069] It should be noted that the specific process of model training is as follows:

[0070] Select the Adam optimizer with an initial learning rate of 0.001. Use the mean squared error (MSE) as the loss function, which is suitable for regression tasks. Use the root mean squared error to evaluate the model performance. Batch size (batch_size): Select 64 and adjust according to the computing resources. Train for 100 - 200 epochs and adjust according to the performance on the validation set. At the end of each epoch, calculate the loss on the validation set and monitor the generalization ability of the model. When the validation loss does not improve for several consecutive epochs, stop training to prevent overfitting. Use the ModelCheckpoint callback function to save the model with the lowest validation loss;

[0071] The specific steps for anomaly detection using the Isolation Forest algorithm are as follows:

[0072] Build 500 isolation trees, set the maximum depth of each tree, specify the proportion of abnormal samples in the data, sample the data with replacement to generate multiple different training sets, build an isolation tree on each training set, split the data by randomly selecting features and randomly selecting feature values. For each sample, calculate the isolation path length (i.e., the number of splits) on all isolation trees. According to the path length, calculate the anomaly score. The shorter the path length, the higher the anomaly score, indicating that it is more likely to be an anomaly point. According to the distribution of the anomaly scores and business requirements, set a threshold to distinguish normal and abnormal samples, and compare the anomaly score of each sample with the threshold. Samples exceeding the threshold are marked as abnormal.

[0073] It should be noted that collect historical humidity data and corresponding dam deformation data, conduct a correlation analysis to determine the relationship between humidity and deformation, use a non - linear model to capture complex relationships, pre - process the humidity data to ensure data quality, construct lag features and rolling averages to reflect the temporal impact of humidity on deformation, consider the cumulative effect of humidity change trends on deformation, introduce time - series analysis methods, and evaluate the improvement of model performance by humidity by comparing the prediction effects of models with and without humidity variables. Ensure that the acquisition frequency of humidity data matches the dam deformation monitoring frequency, and perform data interpolation or estimation if necessary.

[0074] It should be noted that real-time water level data is collected as one of the inputs of the prediction algorithm. Relevant features are extracted according to different water level ranges, such as the water level change rate and historical water level fluctuations, to enhance the prediction ability of the algorithm. For the different characteristics of high water level, medium water level and low water level, multiple sub-models are constructed. Each sub-model is specifically trained for the data within a specific water level range. The water level change may cause non-linear responses of the dam materials. A neural network is used to capture this non-linear relationship. Through the correlation analysis of historical water level and deformation data, potential patterns and laws between water level changes and dam deformations are mined to provide more useful training data for the algorithm. The weights of the prediction model are adjusted in a timely manner through water level changes to ensure the response ability and early warning ability of the algorithm during drastic water level changes. Uncertainty factors of water level changes, such as water level fluctuations caused by climate change, are considered in the model design, and the robustness of the model under different water levels is evaluated through methods such as Monte Carlo simulation.

[0075] It should be noted that real-time earthquake monitoring data is accessed, including the epicenter location, magnitude, depth and vibration intensity. These data are used as input features of the algorithm. Historical earthquake activity data is utilized to analyze the relationship between it and the dam deformation, and potential laws and features are extracted for model training. A finite element model is adopted to simulate the propagation of seismic waves and their impact on the dam, capturing the instantaneous deformation of the dam under seismic action. The time series data is analyzed to enable the model to identify the deformation patterns before and after the earthquake. The attenuation characteristics of seismic waves during propagation are analyzed, and considering factors such as the distance between the dam and the epicenter, the impact of a specific dam under seismic activities is evaluated. Earthquake activities are incorporated into the considerations of dam design to predict the possible maximum deformation to ensure the safety of the dam under extreme seismic conditions. After the dam has experienced one or more earthquakes, the deformation data is collected and fed back into the model.

[0076] According to another aspect of the embodiments of the present invention, a computer-readable medium is further provided. Computer-executable instructions are stored in the storage medium. When the computer-executable instructions are loaded and executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

[0077] According to another aspect of the embodiments of the present invention, a computer program product is further provided. The computer program product is tangibly stored on a non-transitory computer-readable medium and includes machine-executable instructions. When the machine-executable instructions are executed, the machine is caused to execute the steps of the method according to any one of claims 1 to 8.

[0078] In summary, the present invention provides a GNSS-assisted dam displacement monitoring algorithm that comprehensively incorporates various environmental factors such as humidity, water level, and seismic activity into the prediction model, which can more comprehensively reflect the actual working state of the dam, enhance the accuracy and reliability of the algorithm. By analyzing the interactions between different factors, the algorithm can better capture the internal laws of dam deformation, improve the prediction accuracy, and timely identify potential deformation risks. By real-time monitoring and analyzing the data of humidity, water level, and seismic activity, it is possible to identify in advance the potential safety hazards faced by the dam, provide a scientific basis for formulating risk control strategies. The algorithm can handle the changes in real-time data, implement dynamic adjustment and optimization, ensure that in the face of extreme weather, earthquakes and other emergencies, it can quickly respond and conduct deformation prediction. Using the prediction results generated by the algorithm can provide a basis for engineers and managers to help them formulate targeted maintenance and emergency plans to ensure the safe operation of the dam. Through the comprehensive analysis of humidity, water level, and seismic activity, it helps to conduct scientific water resource management and ecological environment protection, and promotes the sustainable development of the areas around the dam. This algorithm can be used as part of a long-term monitoring system to accumulate data over time, which helps to establish a long-term monitoring and evaluation mechanism for dam deformation and form a closed-loop management.

[0079] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A GNSS-assisted dam displacement monitoring algorithm, characterized in that: The following steps are involved: S1. Data collection: Collect the three-dimensional coordinate time series data and environmental data of each measuring point on the dam; S2. Data preprocessing: Use low-pass filtering to remove noise, identify and remove outliers, use linear interpolation to fill missing data, and use Z-score standardization to scale the data to the same range; S3. Feature engineering: extract features such as displacement velocity and acceleration, analyze periodic changes in the data, select the most influential features for displacement prediction, and reduce redundancy; S4. Model design: Design a multi-layer LSTM structure with an appropriate number of units in each layer, add a dropout layer to prevent overfitting, and add a fully connected layer after LSTM to further extract features; S5. Model training: Select the Adam optimizer, use mean square error as the loss function, monitor the validation set loss, and prevent overfitting; S6. Model evaluation and tuning: Divide the dataset into training set and validation set, perform cross-validation, use indicators such as MSE, RMSE and R-squared value to evaluate model performance, and adjust hyperparameters such as learning rate and batch size to optimize model performance; S7. Anomaly detection: set a threshold to detect displacement anomalies and use algorithms such as isolation forest for anomaly detection; S8. Data storage and transmission: Use database to efficiently store data, ensure data security and reliability, and use MQTT or HTTPS protocol to ensure efficient and secure data transmission; S9. Model interpretability: Use LIME or SHAP to explain the decision-making process of the LSTM model and improve the interpretability of the model.

2. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: Determine the known data points before and after the missing data point. If the displacement data at time t2 is missing, find the displacement data x1 and x3 at t1 and t3, and calculate the time difference Δt between the previous and next known data points. total = t3-t1, calculate the time difference between the missing point and the previous known point Δt1 = t2-t1, calculate the value of the missing point based on the linear relationship This formula indicates that the change in displacement is distributed according to the proportion of time; linear interpolation is performed for multiple consecutive missing value segments.

3. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: Extract the velocity and acceleration of the displacement, and specifically analyze the periodic changes in the data. The specific implementation is to uniformly sample the displacement data and process the missing values, use the interpolation method to fill the missing values, smooth the displacement data to reduce the influence of noise, use a low-pass filter, and estimate the velocity through numerical differentiation: Where dt is the time interval; The velocity series is further smoothed and the acceleration is estimated by numerical differentiation: Where dt is the time interval; the acceleration series is further smoothed, the displacement time series is Fourier transformed to obtain the spectrum, and the frequencies corresponding to the peaks in the power spectrum are found. The time periods corresponding to these frequencies are the periodic changes in the data. Multi-scale analysis is performed using wavelet transform to identify periodicity on different time scales.

4. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: LSTM layer design: Number of layers: Choose 2-3 LSTM layers to capture different levels of temporal features; Number of units: Each LSTM layer contains 64-128 units, which is adjusted according to the complexity of the data; Dropout layer: Add a Dropout layer between the LSTM layers and set the dropout rate to 0.2-0.5 to prevent overfitting; Fully connected layer design: Number of layers: Add 1-2 fully connected layers; Number of units: The first fully connected layer contains 64 neurons, and the second fully connected layer contains 32 neurons; Activation function: Use ReLU activation function to improve the nonlinear fitting ability of the model; Output layer: Number of units: The number of units in the output layer depends on the task. If it is a regression task, it outputs a value. If it is a multi-step prediction, output multiple values; Activation function: Linear activation function (no activation function), directly output the predicted value.

5. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: The specific process of model training is as follows: Select Adam optimizer, set the initial learning rate to 0.001, use mean square error (MSE) as the loss function, suitable for regression tasks, use root mean square error to evaluate model performance, batch size (batch_size): select 64, adjust according to computing resources, train 100-200 epochs, adjust according to the performance of the validation set, calculate the loss of the validation set at the end of each epoch, monitor the generalization ability of the model, stop training when the validation loss no longer improves within several consecutive epochs to prevent overfitting, use ModelCheckpoint callback function to save the model with the lowest validation loss; The specific steps of the isolation forest algorithm for anomaly detection are as follows: Build 500 isolation trees, set the maximum depth of each tree, specify the proportion of abnormal samples in the data, sample the data with replacement, generate multiple different training sets, build an isolation tree on each training set, split the data by randomly selecting features and randomly selecting feature values, for each sample, calculate the length of its isolated path (i.e., the number of splits) on all isolation trees, calculate the anomaly score based on the path length, the shorter the path length, the higher the anomaly score, indicating that it is more likely to be an anomaly point, set a threshold to distinguish normal and abnormal samples based on the distribution of anomaly scores and business needs, compare the anomaly score of each sample with the threshold, and mark samples exceeding the threshold as abnormal.

6. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: Collect historical humidity data and corresponding dam deformation data, conduct correlation analysis, determine the relationship between humidity and deformation, use nonlinear models to capture complex relationships, preprocess humidity data to ensure data quality, construct lag characteristics and rolling averages to reflect the temporal impact of humidity on deformation, consider the cumulative effect of humidity change trends on deformation, introduce time series analysis methods, compare the model prediction effects with and without humidity variables, evaluate the improvement of model performance due to humidity, ensure that the frequency of humidity data acquisition matches the frequency of dam deformation monitoring, and perform data interpolation or estimation when necessary.

7. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: Real-time water level data is collected as one of the inputs of the prediction algorithm. Relevant features, such as the water level change rate and historical water level fluctuations, are extracted according to different water level ranges to enhance the prediction ability of the algorithm. Multiple sub-models are constructed according to the different characteristics of high water level, medium water level and low water level. Each sub-model is trained specifically for data within a specific water level range. Water level changes may cause nonlinear responses of dam materials. Neural networks are used to capture this nonlinear relationship. Through the correlation analysis of historical water level and deformation data, the potential patterns and laws between water level changes and dam deformation are explored to provide more useful training data for the algorithm. The weights of the prediction model are adjusted in time according to water level changes to ensure the responsiveness and early warning capabilities of the algorithm when the water level changes drastically. The uncertainty factors of water level changes and water level fluctuations caused by climate change are considered in the model design, and the robustness of the model at different water levels is evaluated through methods such as Monte Carlo simulation.

8. The GNSS-assisted dam displacement monitoring algorithm according to claim 1, characterized in that: Access real-time earthquake monitoring data, including epicenter location, magnitude, depth and vibration intensity. These data are used as input features of the algorithm. Use historical earthquake activity data to analyze the relationship between it and dam deformation, extract potential laws and characteristics for model training, and use finite element models to simulate seismic wave propagation and its impact on the dam, capture the instantaneous deformation of the dam under the action of an earthquake, analyze time series data, enable the model to identify deformation patterns before and after an earthquake, analyze the attenuation characteristics of seismic waves during propagation, and evaluate the impact of specific dams under seismic activity in combination with factors such as the distance between the dam and the epicenter. Incorporate seismic activity into dam design considerations and predict the possible maximum deformation to ensure the safety of the dam under extreme seismic conditions. After the dam has experienced one or more earthquakes, collect deformation data and feed it back into the model.

9. A computer-readable medium, characterized in that The storage medium stores computer executable instructions, and when the computer executable instructions are loaded and executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer program product, characterized in that The computer program product is tangibly stored on a non-transitory computer readable medium and comprises machine executable instructions which, when executed, cause a machine to perform the steps of the method according to any one of claims 1 to 8.

Images