Hydrogeological dynamic monitoring and analysis system and method based on big data

Abstract

The application relates to the technical field of hydrogeological monitoring, in particular to a hydrogeological dynamic monitoring and analysis system and method based on big data; specifically, hydrogeological data is collected in real time, data denoising, space-time calibration and error correction are carried out in combination with edge computing, noise and drift error are eliminated by adopting variational mode decomposition, geological model constraint and wavelet transform, and the storage and transmission efficiency is optimized by increment encoding and Huffman encoding, and a quadratic form check code is generated to ensure data reliability; a space-time deep learning model and a self-attention mechanism are fused to predict a hydrological evolution trend, adaptive anomaly detection is realized by combining DBSCAN clustering and prediction error analysis; water resource allocation strategies are optimized through reinforcement learning, risk levels are evaluated in layers, and management schemes are generated, and finally, monitoring data, abnormal scores and decision results can be visually displayed. The application integrates multi-source data fusion, space-time modeling and intelligent decision-making, and improves hydrological monitoring precision and response efficiency.

Description

Technical Field

[0001] This invention relates to the field of hydrogeological monitoring technology, specifically to a hydrogeological dynamic monitoring and analysis system and method based on big data. Background Technology

[0002] With the increasing impact of global climate change, urbanization, and human activities on natural resources, the dynamic changes in the hydrogeological environment are becoming increasingly complex. Traditional hydrogeological monitoring methods can no longer meet the needs of modern society for efficient, accurate, and real-time data analysis. Changes in hydrogeology not only involve important parameters such as groundwater level, flow velocity, and pressure, but are also closely related to water quality changes, meteorological conditions, and geological structures. In order to effectively address this challenge, big data-based hydrogeological monitoring and analysis methods have emerged.

[0003] A method for monitoring hydrogeological risks, disclosed in Chinese invention patent application CN117609900A, includes: constructing a risk monitoring model; monitoring the mountain to be monitored using the risk monitoring model; the method for constructing the risk monitoring model is as follows: constructing a threshold assessment sub-model for the mountain to be monitored to determine the threshold for disasters occurring in the mountain to be monitored; constructing a weather prediction sub-model to predict weather data within a preset time period; acquiring the current hydrogeological data of the mountain to be monitored through automatic scanning; and constructing a risk monitoring model based on a neural network model according to the weather data, hydrogeological data, and the threshold for disasters occurring in the mountain to be monitored, thus solving the problem of low accuracy in predicting hydrogeological risks in mountainous areas.

[0004] However, existing methods rely on single sensors to collect hydrological parameters, which are susceptible to noise interference and equipment drift, resulting in low data reliability. When fusing multi-source data, there is a lack of dynamic weight adjustment mechanisms, making it difficult to balance the spatiotemporal differences between sensor, remote sensing, and meteorological data. Anomaly detection often relies on fixed thresholds or single algorithms, which cannot adapt to dynamic changes in complex geological environments, resulting in high rates of missed detections and false positives. This restricts the real-time performance, prediction accuracy, and emergency response capabilities of hydrological monitoring. There is an urgent need to build an intelligent monitoring system that integrates edge computing, spatiotemporal deep learning, and reinforcement learning to improve data reliability, anomaly identification sensitivity, and decision-making adaptability. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a hydrogeological dynamic monitoring and analysis system and method based on big data.

[0006] The technical solution of this invention: a method for dynamic monitoring and analysis of hydrogeology based on big data, comprising the following specific implementation steps:

[0007] S1. Collect hydrogeological data, use edge computing for denoising, error compensation and spatiotemporal calibration, remove low-frequency drift, introduce geological model to correct spatial error, remove high-frequency noise, compensate sensor drift error through dynamic deviation regression, calculate and optimize processing efficiency, and finally generate quadratic check code.

[0008] S2. Verify the availability and rationality of hydrogeological data using a quadratic check code, including data binary conversion, data reset code calculation, and the construction and comparison of two check coefficients. Use HDFS to store historical data and NoSQL database to store real-time data.

[0009] S3. By dynamically adjusting data weights through an adaptive data fusion mechanism, a spatiotemporal deep learning model is constructed to integrate multi-scale convolution, LSTM, and self-attention mechanisms to predict hydrological evolution trends. Based on an adaptive anomaly detection method, dynamic threshold anomaly detection is performed by combining DBSCAN clustering and prediction error analysis, and a comprehensive score is generated.

[0010] S4. By integrating comprehensive scores, historical anomaly frequencies and time factors through a hierarchical framework, risk scores are calculated, low, medium and high risk levels are divided and monitoring or scheduling schemes are automatically selected. Q-learning reinforcement learning models are used to define state and action spaces, and decision-making is optimized by combining reward mechanisms and dynamic learning rates. Hydrogeological management schemes are output, and prediction accuracy is improved by adjusting weights through intelligent feedback.

[0011] S5. Visualize and display the processed hydrogeological data, comprehensive score, and hydrogeological management plan.

[0012] Preferably, the noise reduction process is as follows:

[0013] S21. Decompose the raw hydrological data into n modal components and set a low-frequency threshold ω. thresh Components with frequencies below the threshold are filtered out, and the remaining modal components are combined to form the signal after primary denoising.

[0014] Define the objective function as: Used to minimize bandwidth;

[0015] In the formula, u k (t) represents the k-th modal component; w k δ(t) represents the center frequency of the k-th mode; δ(t) represents the Dirac impulse function. This represents the first derivative with respect to time t; Represents a complex exponential function; Represents the square of the 2-norm; j represents the imaginary unit;

[0016] S22. Establish a geological model and predict theoretical hydrological parameters G. model(t), calculate the error, determine the deviation between the measured data and the predicted data, and use the correction coefficient λ to adjust the data, defining the digital model: X corrected (t)=X raw (t)+λ(G model (t)-X raw (t)), using geological models for error correction;

[0017] Among them, X raw (t) represents the raw hydrological data collected by the sensor; G model (t) represents the theoretical value calculated based on the hydrogeological model; λ represents the correction coefficient. α represents the adjustment parameter;

[0018] S23. The signal is decomposed into multi-level sub-signals using the optimal wavelet basis, and high-frequency noise is removed using an improved soft-threshold denoising function. The denoised wavelet coefficients are then subjected to inverse transform to obtain a smoothed signal. Adaptive threshold control is used to control the smoothness of denoising;

[0019] In the formula, c j T represents wavelet coefficients; r β represents the classic threshold; β represents the adjustment parameter that controls the smoothness of denoising; T′ represents the adaptive threshold.

[0020] S24. Define the long-term error compensation digital model: X final (t)=X filtered (t)+γ(P pred (t)-X filtered (t));

[0021]

[0022] Among them, X final (t) represents the final corrected data; X filtered (t) represents the denoised data; P pred (t) represents the predicted value; γ represents the correction coefficient; β represents the parameter controlling the correction sensitivity.

[0023] Preferably, the generation process of the quadratic checksum is as follows:

[0024] S31. Convert the processed hydrogeological data data into a binary string datab;

[0025] S32. Calculate the data encapsulation code CE = H(datab);

[0026] Where H() represents a predefined hash function, n is the integer field modulo p; p is a predefined 1024-bit large prime number; n is a predefined integer dimension.

[0027] S33. Construct a system of linear equations: A·s=CE-v mod p;

[0028] Where A is a predefined, randomly chosen invertible matrix A∈GL(n,p); GL(n,p) is a finite field. The general linear group on the x-axis; v is a predefined, randomly chosen vector.

[0029] S34. Solving for the quadratic check code c = A -1 (CE-v)mod p.

[0030] Preferably, the verification process for checking the usability and rationality of hydrogeological data using a quadratic checksum is as follows:

[0031] S41. Convert the processed hydrogeological data data into a binary string datab'.

[0032] S42. Calculate the data reset code CA = H(datab′);

[0033] S43. Construct the first verification coefficient fⅠ=s T Bs+2s T u;

[0034] Where T represents the transpose of the matrix; B is a predefined parse key, B = A T A mod p is a symmetric matrix; u represents the predefined first analytic coefficient, u = Av mod p;

[0035] S44. Construct the second verification coefficient fⅡ=(CA) T (CA)-w mod p;

[0036] Where w represents a predefined second analytical coefficient, w = v T v mod p;

[0037] S45. If fⅠ = fⅡ, the verification is successful, indicating that the received processed hydrogeological data is usable and reasonable; otherwise, an immediate warning is issued.

[0038] Preferred spatiotemporal deep learning models include:

[0039] Input layer, the input data is a spatiotemporal feature matrix;

[0040] Multi-scale convolutional layers employ multi-scale convolution, using several convolutional kernels of sizes 3×3, 5×5, and 7×7 to extract spatial features and capture multi-level feature information in space;

[0041] LSTM layers learn time series features and model time dependencies;

[0042] The spatiotemporal fusion layer fuses the spatial features extracted by the convolutional layer with the temporal features learned by the LSTM layer, and uses a self-attention mechanism to calculate the importance weights W of the spatial and temporal features. f And perform weighted fusion X;

[0043] The output layer feeds the spatiotemporally fused features X into the fully connected layer to generate the final prediction result, outputting the predicted value X of the future hydrogeological state. pred .

[0044] Preferably, the implementation process based on the adaptive anomaly detection method is as follows:

[0045] S61. The DBSCAN density clustering algorithm is used to separate outlier data points from normal data points based on the density differences. The density of the data is measured by the distance between data points, and the areas with lower density are identified as outlier areas. And calculate the anomaly score S based on DBSCAN clustering. DBSCAN ;

[0046] Where C is the cluster center; X i C(X) represents the i-th data point; i () represents the cluster center of the i-th data point;

[0047] S62. By quantifying the error between the current observed value and the predicted value, anomalies can be accurately determined. The error calculation formula is as follows:

[0048]

[0049] Among them, X current Indicates the current observation value; X pred This represents the model's predicted value, predicted using the ST-ConvLSTM model; σ represents the standard deviation of the current dataset.

[0050] When the error value S anomaly When the value exceeds the preset threshold τ, the current data point is considered an outlier.

[0051] S63. Construct a multi-dimensional comprehensive evaluation model, combining the confidence level of the data source and the correlation between the data sources, and identify outliers by calculating the comprehensive score of each outlier: S total =λ1S DBSCAN +λ2Sanomaly +λ3S confidence ;

[0052]

[0053] Among them, S total S represents the overall score. DBSCAN This represents the anomaly score based on DBSCAN clustering; S anomaly S represents anomaly scoring based on prediction error; confidence λ1, λ2, and λ3 represent the confidence score of the data source; λ1, λ2, and λ3 represent the weight coefficients, i.e., the importance of each evaluation dimension.

[0054] Preferably, the anomaly score S based on DBSCAN clustering DBSCAN The calculation process is as follows:

[0055] S71. For each point X i Calculate the number of points in its neighborhood, that is, the neighborhood density ρ(X) of that point. i For a point X i If there are k points in its neighborhood, then the density can be expressed as:

[0056] Where k represents point X i The number of neighbors within radius ∈; V ∈ (X i ) represents point X i The neighborhood volume;

[0057] S72, Regarding the core point X i Its density ρ(X) i ) is used as a reference value to calculate the density difference at other points. For other points X j Calculate their density difference Δρ ij Δρ ij =ρ(X) i )-ρ(X j );

[0058] S73, Based on the density difference Δρ at the points ij The anomaly score S is calculated by combining the clustering results of DBSCAN. DBSCAN (X i ):

[0059] Where, N ∈ (X i ) represents point X i All points within the neighborhood of .

[0060] Preferably, the hierarchical framework includes:

[0061] S81, Data Input Layer: Based on comprehensive score S total Extract the historical anomaly frequency H and obtain the time factor T;

[0062] S82, Risk Assessment Layer: Based on comprehensive score S total Calculate the abnormal risk score R for hydrological events to determine whether intervention measures are necessary: ​​R = w1S total +w2H+w3T;

[0063] Among them, S total H represents the overall score; T represents the frequency of historical anomalies; w1, w2, and w3 represent the weighting coefficients.

[0064] Hydrogeological conditions are classified according to the risk score R;

[0065] S83. Optimized Inference Layer: Automatically selects the optimal hydrogeological management plan D based on the risk level. * Furthermore, it automatically learns the relationship between the environment and decision-making through a Q-learning reinforcement learning model, thereby optimizing the decision-making process;

[0066] S84, Intelligent Feedback Layer: Performs retrospective analysis of historical decisions and adjusts decision weights accordingly.

[0067] In the formula, R real Represents the true risk value; R pred Indicates the predicted risk value; η represents the learning rate;

[0068] The optimization logic is as follows: if past decisions were accurate, maintain the current weights; if the prediction error is large, adjust the risk calculation weights to improve the accuracy of future decisions.

[0069] The preferred optimization process for automatically learning the relationship between the environment and decision-making using the Q-learning reinforcement learning model is as follows:

[0070] S91. Define the state space S = {R, E, D} history};

[0071] Where R represents the current risk score; E represents the current environmental status; D history Information representing a combination of historical decision-making options and their effects;

[0072] S92. Define the action space A = {a1, a2, a3, a4};

[0073] Floodgate discharge a1: When the water level is too high, floodgate discharge measures are taken.

[0074] Water storage scheduling a2: When the water level is too low, water storage is used to regulate the water level;

[0075] Strengthen monitoring a3: Increase monitoring frequency;

[0076] Adjusting irrigation strategies a4: Adjusting irrigation measures during droughts;

[0077] S93, Set the reward function R t+1 =f(S) t ,a t ,S t+1 );

[0078] Among them, S t Indicates the current state; a t Indicates the current decision or action taken; S t+1 Indicates the next state, i.e., executing 'a'. t Post-environmental changes; f() assesses the contribution of current actions to environmental change;

[0079] S94. In Q-learning, the Q-value represents the long-term expected return of taking a certain action in a given state. The update formula is as follows: Q(s,a)=(1-α′)Q(s,a)+α′[r+γ′max a′ Q(s′,a′)];

[0080] Where Q(s,a) represents the Q-value of taking action a in the current state s; α′ represents the learning rate; r represents the reward after taking the action; γ′ represents the discount factor; max a′ Q(s′,a′) represents the optimal Q value for the next state s′;

[0081] S95. Adjust the dynamic learning rate and intelligent exploration strategy.

[0082] The technical solution of this invention: A hydrogeological dynamic monitoring and analysis system based on big data, which is used to execute the above-mentioned hydrogeological dynamic monitoring and analysis method based on big data, including:

[0083] The hydrogeological data acquisition module is used to acquire hydrogeological data using an Internet of Things sensor network and perform local data preprocessing through edge computing technology.

[0084] The big data management and storage module is used to store historical hydrogeological data using a distributed storage architecture;

[0085] The dynamic monitoring and analysis module is used to combine spatiotemporal big data analysis, use deep learning models to predict groundwater level change trends and geological disaster risk assessment, and provide multi-level data fusion analysis.

[0086] The intelligent decision support module is used to provide auxiliary decision support and optimize water resource allocation strategies using a reinforcement learning model.

[0087] The user interaction and visualization module is used to display monitoring data, early warning information, and historical trends.

[0088] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0089] This invention designs a hydrogeological dynamic monitoring and analysis system and method based on big data. Through technological innovation and system integration, it solves the problems of poor data quality, low prediction accuracy, rigid anomaly detection, and insufficient decision-making efficiency in traditional hydrological monitoring, providing a high-precision, high-response, and intelligent solution for water resource management.

[0090] (1) Enhanced data quality and reliability: Edge computing combined with variational mode decomposition (VMD), wavelet transform and geological model constraint (GCF) is used to achieve multi-stage denoising and error correction, effectively eliminating sensor noise, drift and spatial offset problems; incremental coding and Huffman coding are used to optimize storage and transmission efficiency and reduce redundant data; and quadratic check codes are introduced to verify data integrity. The rationality of data is verified by comparing the symmetric matrix with the predefined parsing code, which significantly improves the reliability of data transmission and storage and avoids data failure due to interference or tampering.

[0091] (2) Improved monitoring accuracy and prediction capability: Construct a spatiotemporal deep learning model (ST-ConvLSTM) with self-attention mechanism, dynamically allocate spatiotemporal feature weights, and combine adaptive data fusion (ADF) to dynamically adjust the confidence of multi-source data, reduce groundwater level prediction error, and improve the accuracy of geological disaster risk assessment.

[0092] (3) Optimization of anomaly detection robustness: Constructing a multi-dimensional comprehensive scoring model (S total By integrating DBSCAN clustering, prediction error analysis, and data source confidence, the threshold is dynamically adjusted to adapt to different hydrological conditions, thereby reducing the false alarm rate and false negative rate of anomaly detection.

[0093] (4) Intelligent decision-making and efficient resource management: Water resource allocation strategies are optimized based on a hierarchical reinforcement learning framework (Q-learning), and management plans are adaptively generated by risk scoring (R), which shortens the decision response time; the system's adaptive capability is enhanced and the long-term decision accuracy is significantly improved by continuously optimizing the model through an intelligent feedback layer.

[0094] (5) Enhanced system scalability and practicality: The distributed storage architecture (HDFS+NoSQL) is adopted to support massive data management. The modular design is compatible with multiple types of sensors and data sources, and can be extended to hydrological monitoring scenarios of different scales and geographical environments, reducing operation and maintenance costs. Attached Figure Description

[0095] Figure 1 This is a system architecture diagram of a hydrogeological dynamic monitoring and analysis system based on big data proposed in this invention;

[0096] Figure 2 This is a flowchart of a hydrogeological dynamic monitoring and analysis method based on big data proposed in this invention. Detailed Implementation

[0097] Example 1, as Figure 1 As shown, the present invention proposes a hydrogeological dynamic monitoring and analysis system based on big data, comprising: a hydrogeological data acquisition module, a big data processing and storage module, a dynamic monitoring and analysis module, an intelligent decision support module, and a user interaction and visualization module;

[0098] The hydrogeological data acquisition module uses an Internet of Things sensor network, including but not limited to groundwater level monitoring sensors, water quality monitoring sensors, weather stations, and remote sensing data sources, to acquire hydrogeological data and perform local data preprocessing through edge computing technology.

[0099] The big data management and storage module adopts a distributed storage architecture (including but not limited to HDFS and NoSQL databases) to store historical hydrogeological data;

[0100] The dynamic monitoring and analysis module combines spatiotemporal big data analysis and uses deep learning models to predict groundwater level change trends and geological hazard risk assessment. It provides multi-level data fusion analysis, including but not limited to surface water and groundwater linkage analysis and precipitation and groundwater dynamic relationship modeling.

[0101] The intelligent decision support module provides auxiliary decision support and uses reinforcement learning (RL) models to optimize water resource allocation strategies and improve management efficiency.

[0102] The user interaction and visualization module displays monitoring data, early warning information, and historical trends.

[0103] Example 2, as Figure 2 As shown, the present invention proposes a method for dynamic monitoring and analysis of hydrogeology based on big data, which is applied to a dynamic monitoring and analysis system for hydrogeology based on big data proposed in Example 1. The specific implementation steps are as follows:

[0104] S1. The hydrogeological data acquisition module adopts an Internet of Things (IoT) sensor network, including but not limited to groundwater level monitoring sensors, water quality monitoring equipment, weather stations, and remote sensing images, to acquire hydrogeological data in real time.

[0105] S11. Suppose that at a certain time t, the hydrogeological state vector of a certain region A is X(t)=[H(t),V(t),P(t),Q(t),T(t),C(t)];

[0106] Where H(t) represents the groundwater level (m), measured by a water level sensor; V(t) represents the groundwater flow velocity (m / s), measured by a flow velocity sensor; P(t) represents the aquifer pressure (Pa), directly measured by a pressure sensor; and Q(t) represents the groundwater flow rate (m³ / s). 3 / s), calculated by the flow meter; T(t) represents the water temperature (°C), measured by the temperature sensor; C(t) represents water quality parameters, including but not limited to pH, conductivity, and dissolved oxygen, measured by a multi-channel water quality sensor;

[0107] S12. Deploy edge computing units at the data acquisition end to perform noise reduction, data completion, and anomaly detection on the acquired data, reducing invalid data and improving data quality. Specifically:

[0108] S1201. Based on the variational mode decomposition (VMD) method, the original signal is decomposed into different frequency modes, and key frequency band signals are selected through adaptive optimization to remove low-frequency drift. That is, the original hydrological data is decomposed into n modal components (IMFs), and a low-frequency threshold ω is set. thresh Components with frequencies below the threshold are filtered out, and the remaining modal components are combined to form the denoised signal.

[0109] The objective function for VMD is:

[0110] In the formula, u k (t) represents the k-th modal component, i.e., the signal in a different frequency band; w k δ(t) represents the center frequency of the k-th mode; δ(t) represents the Dirac impulse function. It represents the first derivative with respect to time t, used to obtain the instantaneous frequency of the signal; This represents a complex exponential function, i.e., the frequency component of each mode; represents the squared L2 norm, i.e., the sum of squares of a vector. This part calculates the energy error for each mode, with the goal of minimizing the overlap between modal components; j represents the imaginary unit.

[0111] Therefore: the objective function is used to minimize bandwidth and improve the separation effect of modal components;

[0112] S1202. Introduce geological structure constraints (GCF) and perform spatial error correction based on the geological model, i.e.: establish a geological model and predict theoretical hydrological parameters G. model(t), calculate the error, determine the deviation between the measured data and the predicted data, and use the correction coefficient λ to adjust the data to ensure that the measured data conforms to the geological characteristics;

[0113] The digital model is: X corrected (t)=X raw (t)+λ(G model (t)-X raw (t));

[0114] Among them, X raw (t) represents the raw hydrological data collected by the sensor; G model (t) represents the theoretical value calculated based on the hydrogeological model, that is, the prediction of hydrological parameters at a specific time t based on known geological characteristics (including but not limited to groundwater flow direction, rock layer thickness, etc.); λ represents the correction coefficient. α represents the adjustment parameter; the larger the error, the stronger the correction. X corrected (t) represents the corrected data, the hydrogeological data after spatial error correction;

[0115] Therefore: use geological models to correct errors and improve data reliability;

[0116] S1203. Sensor signals are often affected by high-frequency noise (including but not limited to electromagnetic interference and equipment vibration), which affects data quality. Wavelet transform is used to decompose the signal into components of different scales, and an adaptive thresholding method is used to remove high-frequency noise. That is, the optimal wavelet basis is used to decompose the signal into multi-level sub-signals, and an improved soft thresholding denoising function is used to remove high-frequency noise. The denoised wavelet coefficients are then subjected to inverse transform to obtain a smooth signal.

[0117] In the formula, c j T represents wavelet coefficients; r β represents the classic threshold; β represents the adjustment parameter that controls the smoothness of denoising; T′ represents the adaptive threshold.

[0118] Therefore: an adaptive threshold is used to control the smoothness of denoising and reduce the loss of high-frequency information;

[0119] S1204. Due to noise reduction, sensor drift error may still exist, requiring long-term error compensation. A digital model needs to be established: X final (t)=X filtered (t)+γ(P pred (t)-X filtered (t));

[0120]

[0121] Among them, X final(t) represents the final corrected data, the hydrogeological data after dynamic bias regression compensation; X filtered (t) represents the denoised data, specifically the hydrogeological data after removing low-frequency and high-frequency noise; P pred (t) represents the predicted value, which is the regression prediction based on historical data and the model, i.e., the sensor drift in the long-term trend; γ represents the correction coefficient, which is used to determine how to adjust the denoised data; β represents the parameter that controls the correction sensitivity, which is used to adjust the correction strength of the sensor drift error.

[0122] S13. Due to the time delay and geographical offset of data from different sensors, spatiotemporal calibration is required to ensure spatial consistency of the data. Specifically:

[0123] By using NTP (Network Time Protocol), the time of all sensors is unified to UTC standard time, reducing time drift error;

[0124] The position of each sensor is calibrated using a coordinate transformation matrix: P′ i =RP i +T v ;

[0125] Among them, P i Represents the original coordinates (GPS / WGS84 format); P′ i Represents the transformed coordinates (UTM format); R represents the rotation matrix, correcting for sensor tilt error; T v This represents the translation vector, used to correct for positional deviations.

[0126] S14. Incremental encoding (Delta Encoding) is adopted to store only the changes in hydrological parameters, reducing storage redundancy. Huffman coding is used to optimize transmission efficiency, and Flink / Spark Streaming is used to perform streaming computation on the data to improve real-time response capabilities.

[0127] S15. Output the processed hydrogeological data (data), and generate a quadratic checksum for the processed hydrogeological data (data). The generation process is as follows:

[0128] S1501. Convert the processed hydrogeological data data into a binary string datab;

[0129] S1502, Calculate the data encapsulation code CE = H(datab);

[0130] Where H() represents a predefined hash function, (i.e., the integer field modulo p); p is a predefined 1024-bit large prime number; n is a predefined integer dimension (n = 128);

[0131] S1503, Construct a linear system of equations A·s=CE-v mod p;

[0132] Where A is a predefined, randomly chosen invertible matrix A∈GL(n,p) (ensuring the determinant is non-zero); GL(n,p) is a finite field. The general linear group on the x-axis; v is a predefined, randomly chosen vector.

[0133] S1504, Solving for the quadratic check code c = A -1 (CE-v)mod p;

[0134] S15. Transmit {secondary checksum s, processed hydrogeological data data} to the big data management and storage module and the dynamic monitoring and analysis module;

[0135] S2. Big Data Management and Storage Module: Establishes an efficient and scalable data storage system to ensure long-term storage, rapid retrieval, and data consistency of massive hydrogeological data. Specifically:

[0136] Obtain {quadratic verification code s, processed hydrogeological data data}, extract the quadratic verification code s and the processed hydrogeological data data from it, and verify the usability and rationality of the processed hydrogeological data data transmitted by the hydrogeological data acquisition module based on the quadratic verification code s. The verification process is as follows:

[0137] S21. Convert the processed hydrogeological data data into a binary string datab';

[0138] S22. Calculate the data reset code CA = H(datab′);

[0139] S23. Construct the first verification coefficient fⅠ=s T Bs+2s T u;

[0140] Here, T represents the transpose of the matrix; B is a predefined parse key, B = A. T A mod p is a symmetric matrix; u represents the predefined first analytic coefficient, u = Av mod p;

[0141] S24. Construct the second verification coefficient fⅡ=(CA) T (CA)-w mod p;

[0142] Where w represents a predefined second analytical coefficient, w = v T v mod p;

[0143] S25. If fⅠ = fⅡ, the check passes, indicating that the received processed hydrogeological data is usable and reasonable; otherwise, an immediate warning is issued.

[0144] A distributed storage architecture of HDFS+NoSQL database is adopted to store historical data (including but not limited to hydrological records over many years) in HDFS (Hadoop Distributed File System) and real-time data (including but not limited to hydrological changes in the last 24 hours) in NoSQL database (non-relational database) to improve query efficiency.

[0145] The S3 dynamic monitoring and analysis module utilizes big data analysis and intelligent algorithms to construct an efficient, accurate, and intelligent hydrogeological dynamic monitoring and analysis system. This system enables real-time monitoring, trend prediction, anomaly detection, and decision support for key hydrological indicators. The specific implementation process is as follows:

[0146] S31. Obtain real-time hydrogeological data data={X sensor (t),X remote (t),X weather (t)}, and obtain historical hydrogeological data X history (t);

[0147] To improve monitoring accuracy, an adaptive data fusion (ADF) mechanism is constructed to dynamically adjust the weights of different data sources, so that high-confidence data contributes more to the final monitoring results.

[0148] X fused (t)=w1X sensor (t)+w2X remote (t)+w3X weather (t)+w4X history (t);

[0149]

[0150] Among them, X sensor (t) represents the sensor data at time t; X remote (t) represents the remote sensing data at time t; X weather (t) represents the weather data at time t; X history (t) represents historical data; w1, w2, w3, and w4 represent adaptive weights; μ is used to control the adjustment magnitude; Q i This represents the confidence level of the i-th data source; X represents the average confidence level across all data sources; fused (t) represents the fused data;

[0151] It should be noted that the performance of each data source is evaluated based on historical data, i.e., the confidence level Q of the data source:

[0152] In the formula, Q i Represents the confidence level of the i-th data source; θ controls the sensitivity of the adjustment and determines the magnitude of the confidence level change; error i This represents the average error of the i-th data source over the past time window T; merror represents the average error of all data sources over the same time window.

[0153] S32. Construct a spatiotemporal deep learning model (ST-ConvLSTM) to predict hydrological evolution, specifically:

[0154] S3201, Constructing the Spatiotemporal Feature Matrix FX t ={X fused (t-2), X fused (t-1), X fused (t)};

[0155] S3202, the spatiotemporal feature matrix FX t The input is fed into the ST-ConvLSTM model, and the model structure is as follows:

[0156] (1) Input layer: The input data is the spatiotemporal feature matrix FX t ;

[0157] (2) Multi-scale convolutional layer: Multi-scale convolution is used to extract spatial features through convolutional kernels of different sizes (including but not limited to 3×3, 5×5, 7×7), capture multi-level feature information in space, and each convolutional block processes data at different scales to help the model recognize spatial patterns of different sizes.

[0158] (3) LSTM layer (learning time series features and modeling time dependencies): The LSTM layer models the time dependencies of data and captures long-term memory in time series data;

[0159] (4) Spatiotemporal Fusion Layer (ST-fusion): The spatial features extracted by the convolutional layer are fused with the temporal features learned by the LSTM layer. To this end, a self-attention mechanism is used to calculate the importance weights W of the spatial and temporal features. f And perform weighted fusion X;

[0160] Specifically, for the spatial features X output by the convolutional layer... conv and the temporal features h of the LSTM layer output t The fusion weight W is calculated using a self-attention mechanism. f :W f =softmax(Wa *[X conv ,h t ]+b a ), and combine spatial and temporal features to generate a spatiotemporal fusion feature X = W. f ·X conv +(1-W f )·h t ;

[0161] Among them, W a This represents the weight matrix of the self-attention layer, used to calculate the fusion weights of spatial and temporal features; b a This represents the bias term of the self-attention layer; softmax() represents the softmax function;

[0162] Accordingly, by using the self-attention mechanism, the model can adaptively adjust the weights of spatial and temporal features according to the spatiotemporal dependence of the input data, thereby capturing the evolutionary laws of the hydrogeological system more accurately.

[0163] (5) Output layer: The spatiotemporally fused feature X is fed into the fully connected layer to generate the final prediction result, outputting the predicted value X of the future hydrogeological state (including but not limited to groundwater level, flow velocity, etc.). pred ;

[0164] S33. Construct an adaptive anomaly detection (AAE) method, which combines cluster analysis and prediction error analysis to achieve dynamic and intelligent anomaly identification, specifically:

[0165] S3301. Preliminary Screening of Outliers (Based on DBSCAN Density Clustering): The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) density clustering algorithm is used to separate outlier data points from normal data points based on the density differences. The distance between data points is used to measure the density of the data, and areas with lower density are identified as outlier areas.

[0166] Where C is the cluster center; X i C(X) represents the i-th data point; i () represents the cluster center of the i-th data point;

[0167] Therefore, using the DBSCAN algorithm for preliminary screening can distinguish outliers from regular data points based on changes in data point density, avoiding the problem of manually setting fixed thresholds, and calculating the outlier score S based on DBSCAN clustering. DBSCAN :

[0168] (1) Calculate the neighborhood density of each data point: For each point X i First, calculate the number of points in its neighborhood, that is, the neighborhood density ρ(X) of that point. i For a point X i If there are k points in its neighborhood, then the density can be expressed as:

[0169] Where k represents point X i The number of neighbors within radius ∈; V ∈ (X i ) represents point X i The neighborhood volume depends on the dimension of the data space;

[0170] (2) Calculate the density difference at each point: Assume point X i For a core point (i.e., its neighborhood contains at least MinPts neighbors), its density ρ(X) i This is used as a reference value to calculate the density difference at other points, for other points X. j Calculate their density difference Δρ ij The formula is: Δρ ij =ρ(X) i )-ρ(X j );

[0171] (3) Calculate the anomaly score: based on the density difference Δρ of the points. ij The anomaly score S is calculated by combining the clustering results of DBSCAN. DBSCAN (X i ):

[0172] Where, N ∈ (X i ) represents point X i All points within the neighborhood;

[0173] S3302. Dynamic Anomaly Assessment Based on Prediction Error: After initial screening by DBSCAN, some anomalies may still not be accurately identified. Further analysis using prediction error is conducted by quantifying the error between the current observed value and the predicted value to accurately determine whether an anomaly exists. The error calculation formula is as follows:

[0174]

[0175] Among them, X current Indicates the current observation value; X pred This represents the model's predicted value, predicted using the ST-ConvLSTM model; σ represents the standard deviation of the current dataset.

[0176] When the error value S anomalyWhen the value exceeds the preset threshold τ, the current data point is considered an outlier.

[0177] It should be noted that, in order to cope with different hydrological conditions and data distribution changes, the threshold τ is not fixed, but dynamically adjusted. The threshold is adjusted by calculating based on the distribution of historical data, so that the model can adapt to the data characteristics of different time periods and regions. The threshold adjustment formula is: τ(t)=μ+η·σ(t);

[0178] Where μ represents the mean of historical data; σ(t) represents the standard deviation of the data at the current moment; and η represents the dynamic adjustment coefficient, which controls the sensitivity of the threshold.

[0179] S3303. Construct a multi-dimensional comprehensive evaluation model, combining the confidence level of data sources and the correlation between data sources, to comprehensively assess the probability of anomalies. By calculating the comprehensive score of each anomaly, the most likely anomalies are identified first: S total =λ1S DBSCAN +λ2S anomaly +λ3S confidence ;

[0180]

[0181] Among them, S total S represents the overall score. DBSCAN This represents the anomaly score based on DBSCAN clustering; S anomaly S represents anomaly scoring based on prediction error; confidence λ1, λ2, and λ3 represent the confidence score of the data source; λ1, λ2, and λ3 represent the weight coefficients, i.e., the importance of each evaluation dimension.

[0182] S34, the overall score S total Transmitted to the intelligent decision support module;

[0183] S4, the intelligent decision support module receives the comprehensive score S from each monitoring point. total This is used to measure the degree of data anomaly and credibility. An Intelligent Decision Support System (IDSS) is introduced, utilizing machine learning prediction, risk assessment, and adaptive optimization strategies to generate optimal hydrogeological management plans, assisting decision-makers in making precise interventions.

[0184] To adapt to different decision-making scenarios, a hierarchical decision-making framework is adopted, forming a closed-loop optimization from data analysis and risk assessment to final strategy generation, including:

[0185] Data input layer (input S) total The hydrogeological analysis results from step S3 are used as the basis for decision-making data.

[0186] Risk assessment layer (calculating anomaly risk score R): assessing potential risks through big data analysis and historical data modeling;

[0187] Optimize the inference layer (compute the optimal decision D) * ): Generate optimal management solutions based on machine learning models;

[0188] Intelligent feedback layer (adaptive optimization): continuously optimizes the decision-making model through reinforcement learning to improve decision-making accuracy;

[0189] Specifically:

[0190] S41, Data Input Layer: Receives the comprehensive score S total Extract the frequency of historical anomalies H (including but not limited to hydrological observation data under similar past conditions for trend analysis) and obtain the time factor T (including but not limited to meteorological data affecting hydrogeology over a certain period of time);

[0191] S42, Risk Assessment Layer: Based on comprehensive score S total Calculate the abnormal risk score R for hydrological events to determine whether intervention measures are necessary: ​​R = w1S total +w2H+w3T;

[0192] Among them, S total The overall score is represented by H, which represents the frequency of historical anomalies (high-frequency anomaly areas have higher weight); T represents the time factor (flood season anomalies have higher weight); and w1, w2, and w3 represent weighting coefficients.

[0193] Based on the risk score R, the hydrogeological conditions are divided into different levels:

[0194]

[0195] Among them, R low and R high These represent predefined low and high thresholds, respectively.

[0196] S43. Optimize the inference layer:

[0197] S4301. Automatically select the optimal hydrogeological management plan D based on the risk level. * :

[0198] Option 1: Low Risk (R) <R low D1: Continuous monitoring, normal monitoring frequency, no additional intervention required;

[0199] Option 2: Medium Risk (R) low ≤R <R highD2: Issue early warning information, strengthen monitoring, and notify relevant departments;

[0200] Option 3: High risk (R≥R) high D3: Emergency dispatching, including but not limited to opening floodgates for flood discharge and water storage dispatching;

[0201] S4302. The relationship between the environment and decision-making is automatically learned through a Q-learning reinforcement learning model, enabling the selection of the optimal solution based on risk assessment and continuous optimization of the decision-making process through continuous learning. Specifically:

[0202] (1) Define the state space S: S = {R, E, D} history};

[0203] Where R represents the current risk score; E represents the current environmental status; D history Information representing a combination of historical decision-making options and their effects;

[0204] (2) In hydrogeological decision-making, the system needs to select different operational schemes as decisions, defining the action space A:

[0205] Floodgate discharge a1: When the water level is too high, floodgate discharge measures are taken.

[0206] Water storage scheduling a2: When the water level is too low, water storage is used to regulate the water level;

[0207] Enhanced monitoring a3: By increasing the frequency of monitoring, the risk of unpredictable risks in the future can be reduced;

[0208] Adjusting irrigation strategies a4: Adjusting irrigation measures during droughts to improve water resource utilization efficiency;

[0209] Therefore, the action space A is: A = {a1, a2, a3, a4};

[0210] (3) Define the reward function R t+1 =f(S) t ,a t ,S t+1 );

[0211] Among them, S t Indicates the current state; a t Indicates the current decision or action taken; S t+1 Indicates the next state, i.e., executing 'a'. t The change in the post-environment; f() evaluates the contribution of the current action to the change in the environment (if an action leads to system stability or achievement of the goal, the reward is positive; if it leads to an adverse result, the reward is negative);

[0212] It should be noted that the reward mechanism is designed as follows: Reward for successful risk mitigation: if the decision taken reduces hydrological risk, the reward is positive; Penalty for failure to effectively mitigate risk: if the decision taken fails to mitigate risk, a negative reward is given; Penalty for over-scheduling or management: if water resource management measures are excessive and lead to resource waste, a negative reward will also be imposed.

[0213] (4) In Q-learning, the Q-value represents the long-term expected return of taking a certain action in a certain state. The update formula is as follows: Q(s,a)=(1-α′)Q(s,a)+α′[r+γ′max a′ Q(s′,a′)];

[0214] Where Q(s,a) represents the Q-value of taking action a in the current state s; α′ represents the learning rate; r represents the reward after taking the action; γ′ represents the discount factor; max a′ Q(s′,a′) represents the optimal Q value for the next state s′;

[0215] (5) Dynamic learning rate and intelligent exploration strategy:

[0216] Dynamic learning rate: The learning rate is dynamically adjusted as the number of training iterations increases. It is larger in the early stages to enable the model to learn quickly; as training progresses to a certain stage, it is gradually reduced to avoid over-adjustment.

[0217] Intelligent exploration mechanism: In the initial stage, an exploration strategy (including but not limited to ε-greedy) is adopted to try new actions with a higher probability; as learning progresses, the utilization of the current best decision is gradually increased and exploration is reduced;

[0218] S44, Intelligent Feedback Layer: Used to optimize the decision-making system, enabling it to continuously improve decision accuracy as data is updated. Specifically:

[0219] S4401. Conduct back-analysis of historical decisions and adjust decision weights accordingly:

[0220] In the formula, R real Represents the true risk value (actual data feedback); R pred Indicates the predicted risk value; η represents the learning rate;

[0221] S4402. The optimization logic is as follows: if past decisions are accurate, maintain the current weights; if the prediction error is large, adjust the risk calculation weights to improve the accuracy of future decisions.

[0222] S45. Hydrogeological Management Plan D * Transmitted to the user interaction and visualization module;

[0223] S5, User Interaction and Visualization Module: Visualizes the processed hydrogeological data (data) and comprehensive score (S). total Hydrogeological Management Plan D * .

[0224] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A hydrogeological dynamic monitoring and analysis method based on big data, characterized in that, The specific implementation steps include the following: S1, collect hydrogeological data, use edge computing for denoising, error compensation and space-time calibration, remove low-frequency drift, introduce a geological model to correct spatial errors, remove high-frequency noise, compensate for sensor drift errors through dynamic deviation regression, and calculate the optimization processing efficiency, and finally generate a quadratic check code; S2, check the availability and reasonableness of the hydrogeological data using the quadratic check code, including data binary conversion, data reset code calculation, and the construction and comparison of two check coefficients, using HDFS to store historical data and NoSQL database to store real-time data; S3, dynamically adjust the data weight through an adaptive data fusion mechanism, build a spatio-temporal deep learning model to fuse multi-scale convolution, LSTM and self-attention mechanism to predict the hydrological evolution trend, dynamically detect thresholds based on an adaptive anomaly detection method combined with DBSCAN clustering and prediction error analysis, and generate a comprehensive score; The implementation process based on the adaptive anomaly detection method is as follows: A1, adopt DBSCAN density clustering algorithm, according to the density difference of data, separate abnormal data points from normal data points, measure the density of data by the distance between data points, and the area with lower density is the abnormal area: And calculate the abnormal score S based on DBSCAN clustering DBSCAN ; where C is the cluster center; X i represents the i-th data point; C(X i ) represents the cluster center of the i-th data point; A2, accurately determine whether it is abnormal by quantifying the error between the current observation value and the predicted value, and the error calculation formula is: ； wherein X current represents the current observation value; X pred represents the model prediction value, which is predicted by the ST-ConvLSTM model; and σ represents the standard deviation of the current data set. When the error value S anomaly is greater than a preset threshold τ, the current data point is considered as an abnormal point. A3, a multi-dimensional comprehensive evaluation model is constructed, the confidence of the data source and the correlation between the data sources are combined, the comprehensive score of each abnormal point is calculated, and the abnormal point is identified: S total = λ1S DBSCAN + λ2S anomaly + λ3S confidence ; ； where S total represents the comprehensive score; S DBSCAN represents the anomaly score based on DBSCAN clustering; S anomaly represents the anomaly score based on prediction error; S confidence represents the confidence score of the data source; λ1, λ2, λ3 represent the weight coefficients, i.e., the importance of each evaluation dimension; S4, calculate the risk score by integrating the comprehensive score, historical abnormal frequency and time factors through a hierarchical framework, divide the low, medium and high risk levels and automatically select the monitoring or scheduling scheme, define the state and action space using the Q-learning reinforcement learning model, optimize the decision-making by combining the reward mechanism and dynamic learning rate, output the hydrogeological management scheme, and adjust the weight through intelligent feedback to improve the prediction accuracy; S5, visually display the processed hydrogeological data, comprehensive score and hydrogeological management scheme.

2. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 1, characterized in that, The denoising process is as follows: S21, decompose the original hydrological data into n modal components, set a low frequency threshold ω thresh , filter out the components with frequency lower than the threshold, combine the remaining modal components to form the primary denoised signal: The objective function is defined as: for minimizing the bandwidth; where u k (t) denotes the kth modal component; w k denotes the center frequency of the kth mode; δ(t) denotes the Dirac impulse function; denotes the first order derivative with respect to time t; denotes the complex exponential function; denotes the squared two-norm; j denotes the imaginary unit; S22, establish a geological model, predict the theoretical hydrological parameters G model (t), calculate the error, determine the deviation of the measured data and the predicted data, and adjust the data using the correction coefficient λ, define the digital model: X corrected (t)=X raw (t)+λ(G model (t)-X raw (t)) Error correction using geological model; Wherein, X raw (t) represents the original hydrological data collected by the sensor; G model (t) represents the theoretical value calculated based on the hydrogeological model; λ represents the correction coefficient, ; α represents the adjustment parameter; S23, using the optimal wavelet base to decompose the signal into multi-level sub-signals, using the improved soft threshold denoising function to remove high frequency noise, and using the inverse transform of the denoised wavelet coefficients to obtain a smooth signal: , using the adaptive threshold control to control the denoising smoothness; where c j denotes the wavelet coefficients; T r denotes the classical threshold; β denotes an adjustment parameter, controlling the denoising smoothness; T' denotes the adaptive threshold; S24, define long-term error compensation digital model: X final (t) = X filtered (t) + γ(P pred (t) - X filtered (t)). ； wherein X final (t) denotes the final corrected data; X filtered (t) denotes the data that has been de-noised; P pred (t) denotes the predicted value; γ denotes a correction coefficient; and β denotes a parameter that controls the sensitivity of the correction.

3. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 1, characterized in that, The generation process of the quadratic check code is as follows: S31, convert the processed hydrogeological data data into a binary string datab; S32, calculate the data encapsulation code CE=H(datab); where H() represents a predefined hash function, H: {0,1} * → ; is an integer field modulo p; p is a predefined large prime number of 1024 bits; n is a predefined integer dimension; S33, construct a linear equation system A·s=CE-v mod p; wherein A is a predefined randomly selected invertible matrix A e GL(n,p); GL(n,p) is the general linear group over the finite field of the above; v is a predefined randomly selected vector, v e ; S34, solve for quadratic form check code c = A -1 (CE-v) mod p.

4. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 3, characterized in that, The checking process for checking the availability and reasonableness of the hydrogeological data using the quadratic check code is as follows: S41, convert the processed hydrogeological data data into a binary string databʹ; S42, calculate the data reset code CA=H(databʹ); S43、constructing the first correction coefficient f I = s T Bs+2s T u; wherein T denotes a transposed matrix of the matrix; B is a predefined analytical code, B = A T A mod p, is a symmetric matrix; u denotes a predefined first analytical coefficient, u = Av mod p; S44, constructing a second check coefficient f II = (CA) T (CA)-w mod p; where w denotes a predefined second resolution factor, w = v T v mod p; S45, if fⅠ=fⅡ, the check passes, indicating that the received processed hydrogeological data data has availability and reasonableness; otherwise, an alarm is immediately raised.

5. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 1, characterized in that, The spatio-temporal deep learning model includes: The input layer, the input data is a spatio-temporal feature matrix; The multi-scale convolution layer uses multi-scale convolution to extract spatial features through several convolution kernels with sizes of 3x3, 5x5 and 7x7 to capture multi-level feature information in space; The LSTM layer learns time series features and models time-dependent relationships; The spatio-temporal fusion layer fuses the spatial features extracted by the convolution layer and the temporal features learned by the LSTM layer, and adopts a self-attention mechanism to calculate the importance weight W of the spatial and temporal features f and weighted fusion X; An output layer, which sends the spatio-temporally fused feature X into a fully connected layer to generate a final prediction result, and outputs a predicted value of the future hydrogeological state X pred .

6. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 1, characterized in that, Anomaly score S based on DBSCAN clustering DBSCAN The calculation process is as follows: S71, for each point X i , the number of points in its neighborhood, i.e. the neighborhood density p(X i ), for a point X i , if there are k points in its neighborhood, the density can be expressed as: ； where k denotes the point X i Number of neighbors within radius e; V ϵ (X i ) denotes the neighborhood volume of the point X i ​ S72, for the core point X i , the density p(X i ) of which is calculated as a reference value for calculating the density difference of other points, for other points X j , their density difference Δρ ij is calculated: Δρ ij = p(X i )- p(X j ); S73, according to the density difference of points Δρ ij , the clustering result of DBSCAN is combined to calculate the anomaly score S DBSCAN (X i ): ； where N ϵ (X i ) denotes all points in the neighborhood of point X i .

7. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 1, characterized in that, The hierarchical framework includes: S81, data input layer: based on the comprehensive score S total , extract the historical abnormal frequency H, and obtain the time factor T; S82, risk assessment layer: based on the comprehensive score S total An abnormal risk score R of the hydrological event is calculated for determining whether intervention measures are needed: R = w1S total +w2H + w3T; wherein S total represents the comprehensive score; H represents the historical abnormal occurrence frequency; T represents the time factor; w1, w2, w3 represent the weight coefficient; According to the risk score R, the hydrogeological conditions are classified; S83, optimize the reasoning layer: according to the risk level, automatically select the optimal hydrogeological management scheme D ∗ And through the Q-learning reinforcement learning model, automatically learn the relationship between environment and decision, optimize the decision process; S84, intelligent feedback layer: backtracking analysis on historical decisions, adjust decision weight: ; where R real represents the true risk value; R pred represents the predicted risk value; η represents the learning rate; The optimization logic is: if the past decision is accurate, maintain the current weight; if the prediction error is large, adjust the risk calculation weight to improve the accuracy of future decision-making.

8. The hydrogeological dynamic monitoring and analysis method based on big data according to claim 7, characterized in that, The optimization process of automatically learning the relationship between the environment and the decision through the Q-learning reinforcement learning model is as follows: S91, define state space S: S = {R, E, D history}; wherein R represents a current risk score; E represents a current environment state; D history represents combined information of historical decision schemes and effects; S92, define the action space A={a1, a2, a3, a4}; Gate opening and flood discharge a1: when the water level is too high, take measures to open the gate and discharge the flood; Water storage regulation a2: when the water level is too low, adjust the water level by storing water; Strengthen monitoring a3: by increasing the monitoring frequency, reduce the risk of future unpredictability; Adjust irrigation strategy a4: adjust irrigation measures during drought to improve water resource utilization efficiency; S93, set reward function R t+1 = f(S t , a t , S t+1 ); where S t represents the current state; a t represents the decision action taken currently; S t+1 represents the next state, i.e. the change of the environment after performing a t f() evaluates the contribution of the action taken currently to the change of the environment; S94、In Q-learning, the Q-value represents the long-term expected return of taking an action in a state, and the update formula is as follows: Q(s, a) = (1 - aʹ)Q(s, a) + aʹ[r + yʹmax aʹ Q(sʹ, aʹ)] wherein Q(s, a) represents the Q value of performing action a in current state s; a' represents a learning rate; r represents a reward after performing the action; g' represents a discount factor; max aʹ Q(s', a') represents the optimal Q value of the next state s'. S95, adjust the dynamic learning rate and intelligent exploration strategy.

9. A big data-based hydrogeological dynamic monitoring and analysis system for performing the big data-based hydrogeological dynamic monitoring and analysis method of any one of claims 1 to 8, characterized in that, Including: Hydrogeological data acquisition module, for using Internet of Things sensor network to obtain hydrogeological data, and through edge computing technology for local data preprocessing; Big data management and storage module, for using distributed storage architecture to store historical hydrogeological data; Dynamic monitoring and analysis module, for combining spatio-temporal big data analysis, using deep learning model to predict groundwater level change trend, geological disaster risk assessment, providing multi-level data fusion analysis; Intelligent decision support module, for providing auxiliary decision support, using reinforcement learning model to optimize water resource allocation strategy; User interaction and visualization module, for displaying monitoring data, warning information, historical trend.

Images