Intelligent early warning method for geological disasters caused by rising of underground water level

Abstract

The invention relates to an intelligent early warning method for geological disasters caused by underground water level rise, which comprises the following steps: acquiring underground water level data of a target area, inputting the underground water level data into an underground water level prediction model, and acquiring a water level change trend prediction result; the underground water level prediction model is obtained by training an LSTM neural network model through a training set; the training set comprises a timestamp of the original underground water digit data and a corresponding water level value; according to the water level change trend prediction result, a water level change curve is drawn, whether the predicted water level fluctuates abnormally or exceeds a preset threshold value or not is judged, and when the predicted water level fluctuates abnormally or exceeds the preset threshold value, an alarm is triggered. The problem of accuracy of underground water level monitoring in a complex environment is solved, large-range real-time data acquisition and transmission are realized, data processing efficiency is improved through edge calculation and cloud analysis, accurate early warning and risk area division of geological disasters are finally realized, and powerful support is provided for disaster prevention and reduction.

Description

Technical Field

[0001] The present invention relates to the field of disaster early warning technology, and in particular to an intelligent early warning method for geological disasters caused by rising groundwater levels. Background Art

[0002] In the early warning of geological disasters caused by rising groundwater levels, how to achieve real-time, continuous and high-precision monitoring of groundwater levels in key areas is a key technical issue. The traditional manual inspection method is inefficient and difficult to meet the timeliness requirements of early warning. Although various types of water level sensors are currently available, they still face many challenges in actual deployment. First, the groundwater level monitoring environment is complex, and factors such as temperature have a significant impact, which places high demands on the protection level and long-term stability of the sensor. Secondly, the field monitoring points are widely distributed, and data transmission is difficult. It is necessary to study the Internet of Things communication technology with wide coverage, flexible networking and reliable transmission. Finally, the monitoring data is large in volume and high in dimension. How to achieve efficient storage, real-time processing and intelligent analysis of data and build an effective early warning model is also a major technical challenge. The solution to these problems depends on the collaborative innovation of multiple fields such as sensing technology, the Internet of Things, and big data, which is a complex system engineering. Summary of the invention

[0003] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to propose an intelligent early warning method for geological disasters caused by rising groundwater levels, which realizes accurate early warning and risk area division of geological disasters, and provides strong support for disaster prevention and mitigation.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] An intelligent early warning method for geological disasters caused by rising groundwater levels, comprising:

[0006] Obtaining groundwater level data of the target area, inputting the groundwater level data into a groundwater level prediction model, and obtaining a water level change trend prediction result; the groundwater level prediction model is obtained by training an LSTM neural network model using a training set; the training set includes: a timestamp of the original groundwater level data and a corresponding water level value;

[0007] According to the water level change trend prediction result, a water level change curve is drawn to determine whether the predicted water level has abnormal fluctuations or exceeds a preset threshold. When the predicted water level has abnormal fluctuations or exceeds the preset threshold, an alarm is triggered.

[0008] Optionally, obtaining the groundwater level data includes:

[0009] Obtain output data of the groundwater level sensor at different temperatures, establish a relationship model between temperature and output error based on the output data, obtain a temperature compensation function fitted by a quadratic polynomial, embed the temperature compensation function into a microprocessor of the groundwater level sensor, and obtain an optimized water level sensor;

[0010] The optimized water level sensor is used to obtain the groundwater level data of the target area.

[0011] Optionally, obtaining the temperature compensation function of the quadratic polynomial fitting includes:

[0012] E(T)=a0+a1T+a2T 2

[0013] Among them, a0, a1, a2 are temperature compensation coefficients, and T is temperature.

[0014] Optionally, determining the temperature compensation coefficient includes:

[0015]

[0016] Among them, E i is the sensor at the i-th temperature point T i The difference between the output value and the true value under , n is the number of data points.

[0017] Optionally, obtaining the training set includes:

[0018] Preprocessing the original groundwater level data to obtain preliminarily processed groundwater level data; the preprocessing includes: denoising and outlier removal;

[0019] The timestamps and corresponding water level values ​​of the preliminarily processed groundwater level data are extracted to construct a time series data set, and the time series data set is used as a training set.

[0020] Optionally, training the LSTM neural network model using the training set includes:

[0021] A single LSTM layer is trained using the training set, and LSTM layers are gradually added. Attention weights are introduced at the front end of the fully connected layer of multiple LSTM layers for weighted summation with the output features of the LSTM layer until the model performance does not improve. Then, the LSTM layer is stopped from being added and the training is completed.

[0022] Optionally, the method further comprises:

[0023] The water level warning results are used in combination with the distribution map of geological disaster-prone areas in the geographic information to conduct overlay analysis and identify high-risk areas affected by water levels.

[0024] Optionally, determining the high-risk area includes:

[0025] According to the water level warning level and the susceptibility of geological disasters, the geological disaster risk index of each region is calculated, and regions with similar risk indexes are divided into the same category to form geological disaster risk areas of different levels;

[0026] Based on the geological disaster risk areas, a geological disaster risk area map is generated, and different risk levels are marked with different colors to form the high-risk areas.

[0027] Optionally, the high-risk areas include: landslide areas, sand liquefaction areas, land salinization areas and surface swampy areas.

[0028] The beneficial effects of the present invention are:

[0029] The present invention collects groundwater level data in complex environments by deploying corrosion-resistant and waterproof sensors, and uses LoRa or NB-IoT technology to build an Internet of Things network to achieve real-time data transmission. Edge computing modules are deployed at monitoring nodes to perform preliminary data processing, and cloud servers receive the processed water level information and perform time series analysis. LSTM neural network models are used to train and predict water level change curves, and a geological disaster risk area map is generated in combination with a geographic information system, thereby solving the accuracy problem of groundwater level monitoring in complex environments, achieving large-scale real-time data collection and transmission, and improving data processing efficiency through edge computing and cloud analysis. Finally, accurate early warning of geological disasters and risk area division are achieved, providing strong support for disaster prevention and mitigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 The present invention is a flowchart of an intelligent early warning method for geological disasters caused by rising groundwater levels according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] like Figure 1 As shown, this embodiment discloses an intelligent early warning method for geological disasters caused by rising groundwater levels, including: obtaining groundwater level data of a target area, inputting the groundwater level data into a groundwater level prediction model, and obtaining a water level change trend prediction result; the groundwater level prediction model is obtained by training an LSTM neural network model using a training set; the training set includes: the timestamp of the original groundwater level data and the corresponding water level value; according to the water level change trend prediction result, a water level change curve is drawn to determine whether the predicted water level has abnormal fluctuations or exceeds a preset threshold, and when the predicted water level has abnormal fluctuations or exceeds the preset threshold, an alarm is triggered.

[0035] Specifically: This embodiment discloses an intelligent early warning method for geological disasters caused by rising groundwater levels, which mainly includes: adopting a corrosion-resistant, waterproof and dust-proof sensor housing design, combined with a temperature compensation algorithm, to obtain groundwater level data that adapts to complex environments, and the groundwater level data is used as input for subsequent analysis; through LoRa or NB-IoT communication technology, a widely covered Internet of Things network is built to obtain real-time water level data of monitoring points; using edge computing technology, a data preprocessing module is deployed at the monitoring node to filter noise data in the real-time water level data to obtain water level information after preliminary processing; receiving the water level information after preliminary processing through a cloud server, combining a time series analysis method to judge the water level change trend and generate a water level change curve; using an LSTM neural network model to train the water level change curve, predict future water level changes, and obtain a water level early warning result; based on the water level early warning result and a preset threshold, combined with a geographic information system, a geological disaster risk area map is generated to determine high-risk areas.

[0036] Furthermore, obtaining groundwater level data includes: obtaining output data of a groundwater level sensor at different temperatures, establishing a relationship model between temperature and output error based on the output data, obtaining a temperature compensation function of a quadratic polynomial fitting, embedding the temperature compensation function into a microprocessor of the groundwater level sensor, and obtaining an optimized water level sensor; and using the optimized water level sensor to obtain groundwater level data of a target area.

[0037] Further, obtaining a temperature compensation function of a quadratic polynomial fit includes:

[0038] E(T)=a0+a1T+a2T 2

[0039] Among them, a0, a1, a2 are temperature compensation coefficients, and T is temperature.

[0040] Further, determining the temperature compensation coefficient includes:

[0041]

[0042] Among them, E i is the sensor at the i-th temperature point T i The difference between the output value and the true value under , n is the number of data points.

[0043] Specifically: A corrosion-resistant, waterproof, and dust-proof sensor housing design is adopted, combined with a temperature compensation algorithm, to obtain groundwater level data that adapts to complex environments. The groundwater level data is used as input for subsequent analysis.

[0044] According to the technical parameters and environmental requirements of the sensor, a corrosion-resistant, waterproof and dust-proof sensor housing is designed to ensure that the sensor can adapt to the complex underground environment. By collecting and analyzing the sensor output data under different temperature conditions, a mathematical model between the sensor output and the ambient temperature is established to obtain the temperature compensation function. The temperature compensation function is embedded in the data processing unit of the sensor to obtain the ambient temperature data in real time, and the groundwater level data output by the sensor is temperature compensated to eliminate the influence of temperature change on the groundwater level measurement. The Kalman filter algorithm is used to filter the groundwater level data after temperature compensation to remove random noise and outliers, thereby improving the accuracy and reliability of the groundwater level data. The filtered groundwater level data is uploaded to the cloud server through the wireless communication module and stored in the time series database to provide data support for subsequent data analysis and application. Based on the groundwater level data and other environmental parameters, a machine learning algorithm (such as support vector machine, random forest, etc.) is used to establish a groundwater level prediction model to predict the groundwater level in the future. The real-time monitoring data and prediction results of the groundwater level are presented to the user in a visual way, and the water level is judged to be abnormal based on the preset threshold. If it exceeds the threshold, an alarm is triggered and relevant personnel are notified to handle it in time.

[0045] For example, the sensor housing design is the key to ensure the normal operation of the equipment in a complex underground environment. Taking the water level sensor as an example, 316L stainless steel can be selected as the housing material, which has excellent corrosion resistance. The housing adopts IP68 protection grade design and is completely waterproof through sealing rings and waterproof joints. A groove structure is designed on the surface of the housing and filled with dustproof materials to effectively block dust from entering. Temperature compensation is an important means to improve the accuracy of the sensor. By collecting sensor output data under different temperature conditions (such as -20℃ to 60℃, one test point every 10℃), a relationship model between temperature and output error can be established. For example, it may be found that the sensor output is 1% lower at 0℃ and 0.5% higher at 40℃. Based on this, a quadratic polynomial fitting temperature compensation function can be obtained. This function is embedded in the sensor's microprocessor, the temperature sensor data is read in real time, and the groundwater level data is corrected to eliminate the influence of temperature drift. The Kalman filter algorithm can effectively remove random noise and improve data reliability. Assuming that there is Gaussian white noise in the water level measurement value, the initial state estimation and error covariance can be set, and then two stages of prediction and update are performed. In the prediction phase, the current state is predicted based on the state at the previous moment, and in the update phase, the prediction result is corrected in combination with the actual observation value. Through repeated iterations, the optimal estimate is finally obtained, which effectively smoothes the data curve and filters out abnormal fluctuations. Wireless communication and data storage are the basis for remote monitoring. Low-power wide area network technologies such as NB-IoT can be used to regularly upload filtered water level data to the cloud server. On the server side, time series databases such as InfluxDB are used to store water level data, supporting efficient time series query and analysis. This method not only saves storage space and energy consumption on the sensor side, but also provides convenience for subsequent big data analysis. Machine learning algorithms can be used for groundwater level prediction. Taking support vector machine (SVM) as an example, historical water level data, rainfall, evaporation, etc. can be used as input features, and the average daily water level in the next week can be used as the output label. The kernel function is used to map low-dimensional features to high-dimensional space, find the optimal separation hyperplane, and realize nonlinear regression prediction. After the model training is completed, the latest environmental parameters can be input to obtain the future water level forecast. Visualization and alarm systems are important links in data application. Web applications can be developed to display real-time groundwater level data and forecast results in the form of line graphs using chart libraries such as ECharts. At the same time, groundwater level warning thresholds can be set. For example, a yellow warning is triggered when the groundwater level exceeds the 90% quantile of history, and a red warning is triggered when it exceeds the 95% quantile. Warning information can be pushed to relevant personnel via SMS, email, etc. to ensure timely response measures.

[0046] Furthermore, through LoRa or NB-IoT communication technology, a widely covered IoT network can be built to obtain real-time groundwater level data at monitoring points.

[0047] By comparing and analyzing the technical characteristics of LoRa and NB-IoT, combined with the needs of actual application scenarios, suitable communication technologies are selected to build a wide-area Internet of Things network. According to the geographical distribution of monitoring points, the deployment of base stations and gateways is reasonably planned to ensure network coverage and communication quality, and to achieve stable and reliable data transmission. Low-power water level sensors are used to regularly collect groundwater level data at monitoring points, and the data is uploaded to the cloud server through wireless communication modules. Data transmission protocols and data formats are designed to compress and encrypt the collected groundwater level data to reduce the amount of data transmission and ensure data security. A data receiving and storage platform is built on the cloud server to parse and store the received groundwater level data and build a groundwater level monitoring database. Time series algorithms are used to analyze and predict groundwater level data, identify abnormal water level change trends, and timely warn of potential flood disaster risks. A water level monitoring visualization platform is developed to intuitively display the real-time groundwater level status of each monitoring point in the form of curves, heat maps, etc., so that managers can grasp the dynamics of water conditions.

[0048] For example, in the construction of a wide-area IoT network, it is crucial to choose the right communication technology. LoRa and NB-IoT are two commonly used low-power wide-area network technologies, each with its own advantages. LoRa has a long transmission distance and low power consumption, and is suitable for deployment in remote areas or places with insufficient power supply. NB-IoT relies on the existing cellular network infrastructure, has a wide coverage range, and is suitable for cities and densely populated areas. For example, in mountain reservoir monitoring, LoRa technology can be used to achieve coverage of a large range of dispersed monitoring points by deploying gateways at commanding heights. In urban waterlogging monitoring, NB-IoT is more applicable, and can directly use the operator network to quickly deploy the monitoring system. Reasonable planning of base stations and gateways is the key to ensuring network coverage and communication quality. Taking the LoRa network as an example, the best gateway location can be determined through terrain analysis and signal strength simulation. In complex terrain areas, relay stations can be used to expand coverage. For example, in a mountainous basin with an area of ​​100 square kilometers, LoRa gateways can be deployed at 3-5 points with higher altitudes, and each gateway covers a radius of about 5-8 kilometers, achieving effective coverage of the entire basin. The selection and deployment of low-power water level sensors are equally important. Pressure or ultrasonic water level sensors can be used, combined with a solar power supply system, to achieve long-term stable operation. The frequency of data collection can be set according to actual needs, such as collecting data every 5 minutes during the flood season and once an hour during the flat water season, which not only ensures data timeliness but also saves energy. The design of the data transmission protocol needs to consider security and efficiency. The lightweight MQTT protocol can be used, and the AES encryption algorithm can be used to protect data security. In terms of data compression, differential encoding and other technologies can be used to transmit only the amount of water level changes, greatly reducing the amount of data. For example, compressing the original 32-bit floating point number to an 8-bit integer can save 75% of the transmission bandwidth. The construction of the cloud data platform is the core of the system. Time series databases such as InfluxDB can be used to store water level data, supporting efficient time series query and analysis. Combined with message middleware such as ApacheKafka, real-time data reception and processing can be achieved. In terms of data analysis, time series models such as ARIMA can be used for water level prediction, and abnormal water level changes can be detected in time by setting dynamic thresholds. The development of a visualization platform is crucial for intuitively displaying water dynamics. Open source tools such as Grafana can be used to build a real-time monitoring screen. Multi-dimensional data display, such as water level curves and watershed heat maps, can help managers quickly grasp the overall water situation. At the same time, GIS maps can be integrated to achieve the integrated display of underground and geographic information, providing more intuitive decision support.

[0049] Furthermore, obtaining the training set includes: preprocessing the original groundwater level data to obtain the groundwater level data after preliminary processing; the preprocessing includes: denoising and outlier removal; extracting the timestamp and corresponding water level value of the groundwater level data after preliminary processing, constructing a time series data set, and using the time series data set as the training set.

[0050] Furthermore, using the training set to train the LSTM neural network model includes: using the training set to train a single LSTM layer, and gradually adding LSTM layers, introducing attention weights at the front end of the fully connected layer of multiple LSTM layers for weighted summation with the output features of the LSTM layer, until the model performance does not improve, stop adding LSTM layers, and complete the training.

[0051] Specifically: Edge computing technology is used to deploy a data preprocessing module at the monitoring node to filter out noise data in real-time groundwater level data and obtain preliminary processed groundwater level information.

[0052] Edge computing technology is deployed on the monitoring node to obtain real-time groundwater level data and transmit the data to the data preprocessing module. The data preprocessing module uses the Kalman filter algorithm to filter the noise of the groundwater level data and filter out the outliers and interference signals in the groundwater level data. According to the preset water level threshold, it is judged whether the filtered groundwater level data exceeds the normal range. If it exceeds, the early warning mechanism is triggered. The groundwater level information after preliminary processing is compressed and encoded to reduce the bandwidth occupation of data transmission and improve the efficiency of data transmission. The compressed groundwater level information is uploaded to the cloud server through the wireless communication module to realize remote storage and analysis of data. The machine learning algorithm is deployed on the cloud server to train the historical water level data, establish a water level prediction model, and realize the trend prediction and anomaly detection of the water level. When the predicted water level trend is abnormal, the cloud server sends an early warning instruction to the monitoring node, and the monitoring node takes corresponding emergency measures according to the instruction, such as increasing the monitoring frequency, starting the drainage equipment, etc., to ensure the safety of the water level.

[0053] For example, the deployment of edge computing technology in groundwater level monitoring nodes can greatly improve data processing efficiency. By installing microprocessors with computing capabilities at monitoring points, such as the ARMCortex-M series, data can be processed on-site. This method not only reduces the burden on the central server, but also reduces data transmission delays. The Kalman filter algorithm plays an important role in the preprocessing of groundwater level data. The algorithm effectively removes random errors in groundwater level measurement through two steps: prediction and correction. For example, the original groundwater level data of a monitoring point is 5.2 meters, 5.1 meters, and 5.3 meters. After Kalman filter processing, smoother values ​​of 5.18 meters, 5.21 meters, and 5.23 meters are obtained, which improves data reliability. The setting of groundwater level thresholds needs to take into account historical data and geographical characteristics. For example, in mountainous rivers, the warning groundwater level can be set to 1.5 times the average water level; while in plain areas, a more conservative 1.3 times may be required. When the measured groundwater level exceeds the threshold, the system automatically triggers an early warning and notifies relevant departments to take flood prevention measures. Data compression technology is crucial to improving transmission efficiency. Using methods such as differential coding, the original groundwater level data can be compressed to about 30% of the original. For example, continuous groundwater level data of 5.21 meters, 5.23 meters, and 5.26 meters can be encoded as 5.21, +0.02, and +0.03, which greatly reduces the amount of data. Machine learning algorithms deployed in the cloud, such as long short-term memory networks (LSTM), can effectively capture the long-term dependencies of water level changes. By analyzing historical data, LSTM can predict the trend of groundwater level changes in the next 24 hours. If the groundwater level is predicted to rise by more than 0.5 meters within 6 hours, the system will issue an early warning. The issuance of early warning instructions adopts a priority mechanism. For minor anomalies, the system may only increase the sampling frequency, such as from once an hour to once every 15 minutes. For serious anomalies, the system will initiate emergency plans, such as opening flood gates and evacuating surrounding residents. This hierarchical response ensures the rational allocation of resources and the timely implementation of emergency measures. The design of the entire system embodies the closed-loop management idea of ​​"monitoring-analysis-early warning-response". From edge computing to processing raw data, to cloud-based intelligent analysis and prediction, to intelligent control of on-site equipment, a complete water level monitoring and flood control management system has been formed. This not only improves the accuracy and real-time performance of water level monitoring, but also provides scientific decision-making support for flood control and disaster reduction, ultimately achieving the goal of protecting people's lives and property.

[0054] The water level information after preliminary processing is received through the cloud server, and the water level change trend is determined by combining the time series analysis method to generate a water level change curve.

[0055] The original groundwater level data collected by the sensor is received through the cloud server, and the original groundwater level data is preprocessed, including denoising, outlier removal and other operations, to obtain the groundwater level information after preliminary processing. According to the groundwater level information after preliminary processing, the timestamp and corresponding water level value of the groundwater level data are extracted to construct a time series data set. The time series analysis algorithm, such as the ARIMA model or the LSTM neural network, is used to model and predict the time series data to obtain the groundwater level change trend in the future. According to the groundwater level change trend prediction results, it is judged whether the groundwater level has abnormal fluctuations or exceeds the preset threshold. If there is an abnormal situation, the early warning mechanism is triggered. The groundwater level change trend prediction results are compared with the actual groundwater level data, and the prediction error is calculated. The time series analysis model is optimized and adjusted through error feedback to improve the accuracy of the groundwater level change trend prediction. Based on the groundwater level change trend prediction results, the groundwater level change curve is drawn to intuitively display the changes in the water level in the future period of time, providing decision support for water conservancy management departments. The water level change curve and warning information are presented to users through a visual interface, and the data is stored in a cloud database for subsequent query and analysis.

[0056] Exemplarily, after receiving the raw groundwater level data collected by the sensor, the cloud server first performs preprocessing. This step includes denoising and outlier removal to ensure data quality. For example, the median filter method can be used to remove sudden noise, or the moving average method can be used to smooth the data curve. For outliers, a reasonable threshold can be set. For example, data points that exceed ±3 times the standard deviation of the historical groundwater level record can be regarded as abnormal and removed. After preprocessing, the timestamp and corresponding water level value of the groundwater level data are extracted to construct a time series data set. This step lays the foundation for subsequent analysis. For example, the data can be organized into a two-dimensional table of "time-water level" to facilitate subsequent modeling and analysis. Next, the time series analysis algorithm is used for modeling and prediction. The ARIMA model is suitable for groundwater level data with seasonality and trend, while the LSTM neural network is more suitable for processing long-term dependencies. Based on the prediction results, it is determined whether the groundwater level has abnormal fluctuations or exceeds the preset threshold. For example, a warning water level line can be set, and an early warning is triggered when the predicted groundwater level exceeds the line. Or by comparing the predicted value with the historical water level of the same period, an early warning can also be triggered when a significant deviation is found. To improve the accuracy of prediction, the prediction results are compared with the actual groundwater level data, the prediction error is calculated and the optimization model is fed back. For example, the rolling prediction method can be used. After each new actual groundwater level data is obtained, the model parameters are updated to continuously improve the prediction accuracy. The groundwater level change curve is drawn based on the prediction results to intuitively display the future groundwater level changes. For example, the groundwater level prediction curve for the next 7 days can be drawn and the warning water level line can be marked to facilitate the water conservancy management department to quickly judge the risk. Finally, the groundwater level change curve and warning information are presented through a visual interface, and the data is stored in the cloud database. For example, a Web application can be developed to display the groundwater level prediction curve in real time, and when a warning occurs, it is marked with a striking color. At the same time, historical data, prediction results, etc. are stored in a distributed database to support high-concurrency queries and big data analysis. This series of steps forms a complete groundwater level monitoring and early warning system. From raw data collection and preprocessing to time series analysis, prediction, and result visualization and storage, each link is closely connected. Through such a system, water conservancy management departments can grasp the trend of groundwater level changes in a timely manner, take preventive measures in advance, and effectively reduce the risk of flood disasters. At the same time, the system's self-optimization mechanism ensures continuous improvement in prediction accuracy, providing reliable support for long-term water resources management.

[0057] Time series analysis algorithms, such as ARIMA model or LSTM neural network, are used to model and predict time series data to obtain the water level change trend in the future.

[0058] Obtain the water level data of the historical time series, preprocess the data, remove outliers and missing values, and normalize the data. According to the characteristics of the time series, select a suitable time series analysis algorithm, such as the ARIMA model or the LSTM neural network. If the ARIMA model is selected, the order parameters p, d, and q of the model are determined according to the autocorrelation and partial autocorrelation of the time series. If the LSTM neural network is selected, the parameters such as the number of layers, the number of neurons, and the activation function of the neural network are designed according to the length and characteristic dimension of the time series. The preprocessed historical water level data is divided into a training set and a test set. The selected algorithm model is trained with the training set data, and the prediction performance of the model is evaluated with the test set data. According to the trained algorithm model, input a time point in the future to predict the corresponding groundwater level change trend. Compare the predicted future groundwater level change trend with the set warning water level. If the predicted groundwater level exceeds the warning water level, an early warning is triggered, and the corresponding flood control measures are determined according to the predicted groundwater level change amplitude.

[0059] For example, the analysis of time series water level data is a key task in water conservancy projects and is of great significance for flood prevention and disaster reduction. First, when obtaining historical groundwater level data, the integrity and representativeness of the data must be considered. For example, a river monitoring station may have ten years of daily groundwater level records, which reflect the hydrological characteristics of the river section. The data preprocessing stage is crucial and directly affects the accuracy of subsequent analysis. When removing outliers, the box plot method can be used to treat data that exceeds 1.5 times the interquartile range of the upper and lower quartiles as abnormal. For missing values, interpolation methods such as linear interpolation or spline interpolation can be used to fill them. Normalization can use the minimum-maximum scaling method to map the data to the [0,1] interval to eliminate the dimension effect. The selection of a suitable time series analysis algorithm requires consideration of data characteristics. The ARIMA model is suitable for groundwater level data with obvious seasonality and trend. For example, if a groundwater level shows an annual cycle change and has a slow upward trend in recent years, the ARIMA model can be considered. When determining the model order, you can assist in the judgment by drawing the autocorrelation function (ACF) and partial autocorrelation function (PACF) graphs. LSTM neural networks are more suitable for dealing with complex groundwater level changes with long-term dependencies. For example, a groundwater level is affected by multiple upstream tributaries and is related to factors such as rainfall and temperature. In this case, the LSTM model may perform better. When designing the network structure, you can first try a simple single-layer LSTM, and then gradually increase the complexity, such as adding multiple LSTM layers and introducing an attention mechanism. Model training and evaluation are key steps to ensure prediction accuracy. Usually 80% of the data can be used for training and 20% for testing. The evaluation indicators can be selected as root mean square error (RMSE) or mean absolute percentage error (MAPE). If MAPE is less than 10%, the model performance is generally considered to be good. When predicting future groundwater level trends, you need to carefully choose the prediction period. Short-term predictions (such as within 7 days) are generally more reliable, while long-term predictions (such as more than 30 days) have greater uncertainty. The prediction results can be used for flood warning. For example, if the water level is predicted to exceed the warning line by 0.5 meters in three days, flood prevention measures such as deploying sandbags and evacuating residents in low-lying areas can be taken in advance. In addition, continuous optimization of the model is also important. The model can be updated regularly with new data, and abnormal prediction results can be manually reviewed in combination with expert experience to continuously improve the adaptability and accuracy of the model. This method can fully utilize the experience and knowledge of water conservancy engineering experts while ensuring scientificity, and realize intelligent water level prediction and flood control decision support with human-machine collaboration.

[0060] Step S105, using the LSTM neural network model to train the water level change curve, predict future water level changes, and obtain water level warning results.

[0061] Obtain historical groundwater level change data, preprocess the data, and extract the characteristics of the groundwater level change curve. According to the extracted curve characteristics, build an LSTM neural network model, set the model parameters and training parameters. Divide the preprocessed historical water level data into a training set and a validation set, and input them into the LSTM model for training. During the training process, adjust the model parameters and optimize the model performance according to the prediction results of the validation set. After the training is completed, use the trained LSTM model to predict the newly input groundwater level data. According to the prediction results, determine whether the future groundwater level exceeds the preset warning water level threshold. If it exceeds, generate warning information. Send the warning information to relevant departments and personnel, and store the prediction results and warning information in the database for subsequent analysis and decision-making.

[0062] For example, the prediction of groundwater level change is a key task in water conservancy projects and is of great significance for flood prevention and disaster reduction. First, obtain historical groundwater level data, which usually comes from monitoring stations of rivers, lakes or reservoirs. The data preprocessing stage includes removing outliers, filling missing values ​​and standardization. For example, in the daily groundwater level data of a river monitoring station in the past five years, there are abnormally high values ​​caused by equipment failure, which need to be identified and eliminated by statistical methods. Extracting the characteristics of the groundwater level change curve is the basis of modeling. These characteristics may include periodicity, trend and mutation. For example, the analysis found that the groundwater level showed obvious seasonal changes, and the water level rose significantly during the flood season from June to August each year. In addition, in recent years, due to the regulation of upstream reservoirs, the fluctuation amplitude of groundwater levels has decreased. These characteristics provide important information for subsequent model training. LSTM (Long Short-Term Memory) neural network is a special recurrent neural network suitable for processing and predicting time series data. When building an LSTM model, it is necessary to determine parameters such as the number of network layers and the number of neurons in each layer. For example, a network structure with two LSTM layers and one fully connected layer can be designed, with 64 LSTM units in the first layer and 32 in the second layer. Training parameters include learning rate, batch size, and number of training rounds. The initial learning rate can be set to 0.001 and gradually reduced as the training progresses to achieve better convergence. The preprocessed historical groundwater level data is divided into a training set and a validation set, usually in an 8:2 ratio. During the training process, the model continuously adjusts the weights through the back propagation algorithm to minimize the error between the predicted value and the actual value. The model performance is evaluated on the validation set, and common indicators include root mean square error (RMSE) and mean absolute error (MAE). If the model is found to perform poorly on the validation set, there may be an overfitting problem. At this time, you can consider adding a dropout layer or adjusting the regularization parameter. After training, the model is tested with new groundwater level data. Assume that the model predicts the groundwater level changes in the next 7 days. The results show that the groundwater level will reach 145 meters on the 5th day, exceeding the preset warning water level of 144 meters. At this point, the system will automatically generate warning information, including key information such as the expected time to exceed the warning level and the highest groundwater level. The warning information is sent to relevant units such as the water conservancy department and the flood control command center through various channels such as text messages and emails. At the same time, the prediction results and warning information are stored in the database to provide a basis for subsequent analysis and decision-making. For example, the accuracy and reliability of the model can be evaluated based on historical warning records, and the warning threshold and response mechanism can be further optimized. The implementation of this groundwater level prediction and warning system can predict potential flood risks in advance and win precious time for flood control and flood prevention work. By continuously accumulating data and optimizing models, the prediction accuracy of the system will continue to improve, providing strong support for water resources management and disaster prevention and mitigation.

[0063] After the training is completed, the trained LSTM model is used to predict the newly input groundwater level data. Based on the prediction results, it is determined whether the future groundwater level exceeds the preset warning water level threshold. If it exceeds the threshold, an early warning message is generated.

[0064] Obtain a pre-trained LSTM model for predicting groundwater level data; obtain newly input groundwater level data as input to the LSTM model; use the LSTM model to predict the newly input groundwater level data to obtain groundwater level prediction results for a period of time in the future; obtain a preset warning water level threshold as a condition for determining whether an early warning is required; compare the future groundwater level prediction results obtained by the LSTM model with the preset warning water level threshold; if the future groundwater level prediction results exceed the preset warning water level threshold, determine that early warning information needs to be generated; based on the judgment result, automatically generate corresponding water level early warning information and send it to relevant personnel or systems so that preventive measures can be taken in a timely manner.

[0065] For example, the groundwater level prediction and early warning system is an important tool for flood prevention and disaster reduction. The system first needs to obtain a pre-trained LSTM model, which has learned the law of groundwater level changes through a large amount of historical groundwater level data. For example, a groundwater LSTM model may be trained based on daily water level data for the past 10 years, learning factors such as seasonal changes and rainfall effects. Obtaining new input groundwater level data is key to the operation of the system. This data may come from groundwater level monitoring stations distributed throughout the target area, uploading the latest water level data in real time every hour. For example, during the flood season, the system may obtain the latest groundwater level data every 15 minutes to provide more timely predictions and early warnings. Using the LSTM model to predict new data is the core function of the system. The model may predict water level changes in the next 24 hours, 48 ​​hours, or even a week. The prediction results are usually presented in the form of a time series, showing the predicted water level value at each time point. The preset warning water level threshold is determined based on the characteristics of the river and flood control requirements. For example, the warning level of a groundwater may be set at 50 meters, which is usually determined by the water conservancy department based on historical flood data and local flood control capabilities. Comparing the forecast results with the warning level is a key step in early warning. The system will check the groundwater level value at each time point in the forecast one by one. As long as one time point exceeds the warning level, an early warning will be triggered. For example, if the forecast shows that the groundwater level will reach 50.5 meters in 18 hours, exceeding the warning level, the system will generate an early warning message. The generation of early warning information needs to consider many factors. In addition to the time when the warning level is exceeded and the expected height of the groundwater level, it may also include information such as the speed of water level rise and duration. For example, the early warning message may be: "It is expected that the water level will exceed the warning line by 0.5 meters at 2 am tomorrow. Please make flood prevention preparations for relevant departments." Early warning information is usually sent through multiple channels to ensure that the information is delivered in time. It may include sending text messages to the flood control headquarters, pushing mobile phone APP notifications to coastal residents, and issuing early warnings through the broadcasting system. This multi-channel release ensures that even if a communication channel fails, the early warning information can still be conveyed in time. The workflow of the entire system embodies the intelligent flood prevention concept of "prediction-judgment-warning". Through the accurate prediction of the LSTM model, combined with the reasonably set warning threshold, the system can warn of possible flood risks in advance. This not only buys valuable preparation time for the flood control department, but also provides important protection for the safety of life and property of coastal residents. At the same time, the automated operation of the system greatly improves the efficiency of flood control and reduces the delays or errors that may be caused by human judgment. In addition, this system has the potential for continuous optimization. By comparing the prediction results with the actual water level changes, the LSTM model can be continuously adjusted and improved to improve the prediction accuracy. At the same time, according to the actual situation of each flood, the warning water level threshold can also be appropriately adjusted to make the warning more accurate and effective.This dynamic optimization mechanism enables the system to continuously adapt to new challenges brought about by climate change and river basin changes, providing strong support for long-term flood prevention and disaster reduction work.

[0066] Furthermore, the method also includes: using the water level warning results, combining them with the distribution map of geological disaster-prone areas in the geographic information to perform overlay analysis and identify high-risk areas affected by the water level.

[0067] Furthermore, determining high-risk areas includes: calculating the geological disaster risk index of each area according to the water level warning level and the susceptibility of geological disasters, dividing areas with similar risk indices into the same category, and forming geological disaster risk areas of different levels; generating a geological disaster risk area map based on the geological disaster risk areas, and marking different risk levels with different colors to form high-risk areas.

[0068] Specifically: Based on the groundwater level warning results and the preset thresholds, combined with the geographic information system, a geological disaster risk area map is generated to determine the high-risk areas.

[0069] The real-time groundwater level data of the groundwater level monitoring point is obtained and transmitted to the groundwater level early warning system. The groundwater level early warning system compares and analyzes the real-time groundwater level data according to the preset water level threshold. If the groundwater level exceeds the early warning threshold, the water level early warning is triggered. The water level early warning results are superimposed and analyzed with the distribution map of geological disaster-prone areas in the geographic information system to identify high-risk areas affected by the groundwater level. The fuzzy comprehensive evaluation method is used to calculate the geological disaster risk index of each area based on factors such as the groundwater level early warning level and the degree of geological disaster susceptibility. Through the cluster analysis algorithm, areas with similar risk indexes are divided into the same category to form geological disaster risk areas of different levels. In the geographic information system, a geological disaster risk area map is generated based on the risk area division results, and different risk levels are marked with different colors. The location coordinates, risk level and other information of the high-risk area are extracted to form a list of high-risk areas, providing a basis for subsequent early warning response and disaster prevention measures.

[0070] For example, the groundwater level early warning system is an important tool for disaster prevention and mitigation, and its core lies in real-time monitoring and analysis of groundwater level data. Taking a river basin as an example, the system obtains real-time groundwater level data through multiple monitoring stations distributed along the coast. These data are transmitted to the central processing system in real time through a wireless sensor network to ensure the timeliness and accuracy of the information. The setting of the early warning threshold is a key link in the system. According to historical data and expert experience, the early warning water level of the river is set to 10 meters. When the real-time groundwater level reaches 9.5 meters, the system will issue a yellow warning; when it reaches 10 meters, a red warning will be issued. This hierarchical early warning mechanism helps relevant departments take corresponding preventive measures. The distribution map of geological disaster-prone areas is drawn based on geological surveys and historical disaster data. By superimposing the water level early warning results with this map for analysis, high-risk areas affected by groundwater levels can be identified. For example, a mountainous area is prone to landslides during the rainy season due to its loose geological structure. When the groundwater level early warning in the area is triggered, the system will automatically mark it as a key focus. Fuzzy comprehensive evaluation method plays an important role in risk assessment. The method considers multiple factors, such as groundwater level warning level, geological conditions, and vegetation coverage. Each factor is assigned a different weight, and the risk index is calculated comprehensively. For example, if the groundwater level warning level in a certain area is red (weight 0.4), the geological disaster susceptibility is high (weight 0.3), and the vegetation coverage is low (weight 0.3), the risk index of the area is calculated to be 0.85, which belongs to the high risk level. Cluster analysis algorithm is used to classify areas with similar risk index. The commonly used K-means algorithm can divide the area into three risk levels: low, medium, and high. This classification method helps to allocate resources rationally and ensure that high-risk areas receive priority attention. Geographic Information System (GIS) plays a key role in visualizing risk areas. The system marks the risk areas on the electronic map with different colors: green for low risk, yellow for medium risk, and red for high risk. This intuitive display method helps decision makers quickly understand the overall risk distribution. The list of high-risk areas is an important basis for disaster prevention actions. The list contains detailed information about each high-risk area, such as geographic coordinates, risk level, and main threat type. For example, the information of a high-risk area may be: coordinates (longitude 116.123, latitude 39.456), risk level: high, main threat: landslide. This information provides precise guidance for emergency response and evacuation plans. The design of the entire system embodies the concept of "prevention first, comprehensive prevention and control". Through real-time monitoring, multi-dimensional analysis and visual display, the system can not only issue early warnings in a timely manner, but also provide a scientific basis for disaster prevention and mitigation decisions, and minimize the losses caused by geological disasters.

[0071] Furthermore, high-risk areas include: landslide areas, sand liquefaction areas, land salinization areas and surface swamping areas.

[0072] Landslide area: Groundwater level warning level: Generally, when the water level warning level is high, such as reaching orange or red warning, the groundwater level rises significantly, which increases the possibility of landslides. This is because the rise in groundwater levels will increase the pore water pressure of the rock and soil, thereby reducing the shear strength of the rock and soil, reducing the stability of the slope, and easily causing landslides.

[0073] Factors that affect the susceptibility of geological disasters: areas with complex geological structures, loose rock and soil structures, large terrain slopes, and free-falling surfaces are more prone to landslides. For example, slopes with developed joints, fissures, layers, and faults, especially when steeply inclined structural surfaces parallel to and perpendicular to the slope and gently inclined structural surfaces along the slope are developed, are most prone to landslides.

[0074] Sand liquefaction area: Groundwater level warning level: The groundwater level has an important impact on sand liquefaction. When the groundwater level is high and under the action of external forces such as earthquakes, the pore water pressure in the sand will increase sharply, resulting in a decrease in the effective stress between sand particles, thereby causing sand liquefaction. Generally speaking, when the water level warning level is high, such as reaching orange or red warning, the groundwater level rises more, and the risk of sand liquefaction will increase accordingly.

[0075] Factors affecting the susceptibility of geological disasters: In terms of geological conditions, loose sand and silt layers are prone to liquefaction. In addition, areas with stable geological structures and frequent earthquakes also have a higher risk of liquefaction. This is because violent vibrations such as earthquakes can redistribute the pore water pressure in the sand. When the pore water pressure exceeds the effective stress of the sand, liquefaction will occur.

[0076] Land salinization areas:

[0077] Groundwater level warning level: The rise of groundwater level is one of the important factors of land salinization. When the water level warning level is high and the groundwater level is close to or above the surface, after the water in the soil evaporates, salt will accumulate on the surface, leading to land salinization. For example, in arid and semi-arid areas, the rise of groundwater level to a certain height is likely to cause soil salinization.

[0078] Factors affecting the susceptibility of geological disasters: Topographic conditions have an important impact on land salinization. In low-lying areas with poor drainage, groundwater is easy to accumulate, and the salt in the soil is also more likely to migrate and accumulate upward, thus forming saline soil. In addition, soils with fine texture and poor permeability are also prone to salinization, because these soils have a strong barrier effect on groundwater, making it easier for the salt in the groundwater to accumulate on the soil surface.

[0079] Surface swamping area: Groundwater level warning level: A high groundwater level is one of the important conditions for surface swamping. When the water level warning level is high and the groundwater level is close to or higher than the surface, the surface is wet for a long time and swamps are likely to form. For example, in some low-lying areas, when the groundwater level rises to near the surface, surface water cannot be discharged in time, and swamps will form.

[0080] Factors affecting the susceptibility of geological disasters: In terms of topographic conditions, low-lying areas, basins, valleys, plains, etc. are prone to swamping. These areas have low terrain, poor drainage, and are prone to water accumulation. In addition, climate conditions also have an important impact on swamping. Warm and humid climates are conducive to the growth of vegetation and the accumulation of organic matter, thereby promoting the formation and development of swamps.

[0081] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent early warning method for geological disasters caused by rising groundwater levels, characterized in that: include: Obtaining groundwater level data of a target area, inputting the groundwater level data into a groundwater level prediction model, and obtaining a water level change trend prediction result; The groundwater level prediction model is obtained by training the LSTM neural network model using a training set; the training set includes: the timestamp of the original groundwater level data and the corresponding water level value; According to the water level change trend prediction result, a water level change curve is drawn to determine whether the predicted water level has abnormal fluctuations or exceeds a preset threshold. When the predicted water level has abnormal fluctuations or exceeds the preset threshold, an alarm is triggered.

2. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 1 is characterized in that: Acquiring the groundwater level data includes: Obtain output data of the groundwater level sensor at different temperatures, establish a relationship model between temperature and output error based on the output data, obtain a temperature compensation function fitted by a quadratic polynomial, embed the temperature compensation function into a microprocessor of the groundwater level sensor, and obtain an optimized water level sensor; The optimized water level sensor is used to obtain the groundwater level data of the target area.

3. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 2 is characterized in that: Obtaining the temperature compensation function of the quadratic polynomial fitting includes: E(T)=a0+a1T+a2T 2 Among them, a0, a1, a2 are temperature compensation coefficients, and T is temperature.

4. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 3 is characterized in that: Determining the temperature compensation coefficient includes: Among them, E i is the sensor at the i-th temperature point T i The difference between the output value and the true value under , n is the number of data points.

5. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 1 is characterized in that: Acquiring the training set includes: Preprocessing the original groundwater level data to obtain preliminarily processed groundwater level data; the preprocessing includes: denoising and outlier removal; The timestamps and corresponding water level values ​​of the preliminarily processed groundwater level data are extracted to construct a time series data set, and the time series data set is used as a training set.

6. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 1 is characterized in that: Training the LSTM neural network model using the training set includes: A single LSTM layer is trained using the training set, and LSTM layers are gradually added. Attention weights are introduced at the front end of the fully connected layer of multiple LSTM layers for weighted summation with the output features of the LSTM layer until the model performance does not improve. Then, the LSTM layer is stopped from being added and the training is completed.

7. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 1 is characterized in that: The method further includes: The water level warning results are used in combination with the distribution map of geological disaster-prone areas in the geographic information to conduct overlay analysis and identify high-risk areas affected by water levels.

8. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 7, characterized in that: The high-risk areas identified include: According to the water level warning level and the susceptibility of geological disasters, the geological disaster risk index of each region is calculated, and regions with similar risk indexes are divided into the same category to form geological disaster risk areas of different levels; Based on the geological disaster risk areas, a geological disaster risk area map is generated, and different risk levels are marked with different colors to form the high-risk areas.

9. The intelligent early warning method for geological disasters caused by rising groundwater levels according to claim 7, characterized in that: High-risk areas include: landslide areas, sand liquefaction areas, land salinization areas and surface swamping areas.

Images