A groundwater environment automatic monitoring super station state supervision method and system

Abstract

The application provides a kind of groundwater environment automatic monitoring super station state supervision method and system, method includes: collection multi-source dynamic time series data set and space-time alignment;With nonlinear dynamics feature extraction method generates dynamics feature index set;Based on this index set, construct adaptive collaborative measurement network for anomaly monitoring, generate anomaly detection results and collaborative control instruction;With dynamic threshold adjustment algorithm generates real-time monitoring early warning information;With neural network adaptive control strategy and resource scheduling optimization algorithm, generate adaptive control instruction and dynamic resource allocation scheme.The application generates dynamics feature index set by the above method, reconstructs high-dimensional phase space using C-C method, calculates correlation index and dimension using Wolf algorithm and G-P algorithm, and analyzes multi-scale fractal pattern using R / S analysis method and wavelet transform.These methods capture the complex dynamic behavior of groundwater system and solve the problem of insufficient nonlinear feature extraction capability of traditional methods.

Description

Technical Field

[0001] This invention relates to the field of groundwater environment monitoring well technology, and in particular to a method and system for monitoring the status of an automatic groundwater environment monitoring superstation. Background Technology

[0002] Currently, the status monitoring of automatic groundwater environment superstations mainly relies on multi-source sensor networks to collect time-series data such as water level, water quality, and temperature, which are then transmitted to a central platform via wired / wireless transmission. Traditional methods often employ statistical models or shallow machine learning (such as SVM and random forest) for anomaly detection, combined with fixed thresholds to trigger early warnings. Some systems integrate simple collaborative mechanisms to achieve data sharing between nodes, but their dynamic adjustment and cross-domain optimization capabilities are limited.

[0003] The existing technology has three major drawbacks: First, it has a weak ability to extract nonlinear dynamic characteristics (such as chaos and fractal dimension), making it difficult to capture the complex dynamic behavior of groundwater systems; second, threshold adjustment relies on human experience and lacks dynamic adaptability, which can easily lead to false alarms or missed alarms; and third, cross-node collaborative decision-making and resource scheduling are inefficient, making it impossible to achieve dynamic collaborative optimization of computing, network and storage resources, and thus difficult to meet the needs of real-time monitoring. Summary of the Invention

[0004] This invention aims to at least solve the technical problem of existing technologies that are difficult to meet the requirements of real-time monitoring, and innovatively proposes a method and system for monitoring the status of groundwater environment automatic monitoring super stations.

[0005] To achieve the above-mentioned objectives of this invention, this invention provides a method for monitoring the status of an automatic groundwater environment superstation, the method comprising: S1. Deploy a multi-source heterogeneous sensor network in the detection area of ​​groundwater environment monitoring wells to collect multi-source dynamic time-series datasets and perform spatiotemporal alignment. S2. Based on the multi-source dynamic time-series dataset, a set of dynamic feature indicators including chaotic indices, long-term memory features, fractal dimension, and critical bifurcation points is generated using a nonlinear dynamic feature extraction method. S3. Based on the dynamic characteristic index set, construct an adaptive collaborative measurement network for anomaly monitoring, and combine it with an improved probabilistic synchronous parallel mechanism for cross-node collaborative decision-making to generate anomaly detection results and collaborative control commands. S4. Based on the anomaly detection results and collaborative control instructions, a dynamic threshold adjustment algorithm is used to generate real-time monitoring and early warning information; S5. Based on the real-time monitoring and early warning information, generate adaptive control commands and dynamic resource allocation schemes using neural network adaptive control strategies and resource scheduling optimization algorithms.

[0006] In another aspect, the present invention also provides a groundwater environment automatic monitoring superstation status supervision system, the system comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement the automatic monitoring method for the status of the super station for groundwater environment when executing the executable instructions.

[0007] The beneficial effects of this invention are as follows: This invention generates a set of dynamic feature indices, including chaotic indices, long-term memory features, fractal dimension, and critical bifurcation points, through a nonlinear dynamic feature extraction method. It reconstructs a high-dimensional phase space using the CC method, calculates the maximum Lyapunov exponent using the Wolf algorithm, calculates the correlation dimension using the GP algorithm, and identifies the Hurst exponent through R / S analysis and analyzes multi-scale fractal patterns using wavelet transform. These techniques effectively capture the complex dynamic behaviors of groundwater systems, such as chaos, long-term memory effects, fractal structures, and critical transition thresholds. This solves the problem of insufficient nonlinear feature extraction capabilities in traditional methods, improving the accuracy of anomaly detection and the scientific rigor of system state identification.

[0008] This invention introduces a Q-learning reinforcement learning algorithm to dynamically adjust detection threshold parameters. It generates an adaptive threshold curve through optimization strategies based on historical false positive and false negative rates, and dynamically corrects and generates a multi-level warning threshold set using real-time data streams. This mechanism achieves adaptive threshold adjustment, avoiding false positive / false negative problems caused by reliance on human experience. For example, when triggering blue, yellow, and red warning signals through the tiered warning threshold set, the system can automatically match the risk level based on the magnitude and duration of exceeding the threshold, significantly reducing the uncertainty caused by human intervention and improving the timeliness and accuracy of warnings.

[0009] This invention constructs an adaptive collaborative measurement network, combining an improved probabilistic synchronous parallel mechanism to achieve asynchronous intelligence synchronization across nodes, and generating the final anomaly detection result through weighted voting using a federated learning framework. At the resource scheduling level, it utilizes an SDN architecture to dynamically configure the network resource pool, optimizes compute node load balancing by combining Kubernetes containerized deployment and HPA automatic scaling mechanism, and optimizes network bandwidth allocation through SDN traffic engineering algorithms. These technologies achieve dynamic collaborative optimization of compute, network, and storage resources, solving the problems of inefficient collaborative decision-making and rigid resource allocation in traditional systems, improving cross-node collaborative efficiency and resource utilization efficiency, and meeting the needs for rapid response and dynamic optimization in real-time monitoring scenarios.

[0010] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0011] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of a method for monitoring the status of an automatic groundwater environment superstation according to the present invention. Detailed Implementation

[0012] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0013] Example 1 like Figure 1 As shown, a method for monitoring the status of an automatic groundwater environment monitoring superstation includes: S1. Deploy a multi-source heterogeneous sensor network in the detection area of ​​the groundwater environment monitoring wells of the field groundwater environment monitoring super station to collect multi-source dynamic time-series datasets and perform spatiotemporal alignment. In step S1, it is necessary to explain in detail that the multi-source heterogeneous sensor network in this embodiment includes: water level sensors (such as pressure level gauges and ultrasonic level gauges) deployed at monitoring points at different depths within the well; multi-parameter water quality sensors (capable of real-time acquisition of pH value, conductivity, dissolved oxygen concentration, turbidity, permanganate index, and heavy metal ion content such as lead and cadmium); water pressure sensors; and water temperature sensors; well structure monitoring sensors (such as tilt angle sensors, crack width sensors, and vibration sensors) installed at key locations on the well wall; and environmental meteorological sensors (such as rainfall sensors, air temperature sensors, humidity sensors, and wind speed sensors) set around the wellhead. These sensors are networked using low-power wide-area communication modules such as LoRa and NB-IoT to provide layered coverage of the vertical profile of groundwater and the dynamic changes in the well body and surrounding environment, enabling real-time acquisition and preliminary preprocessing of multi-dimensional monitoring data.

[0014] S2. Based on a multi-source dynamic time-series dataset, a set of dynamic feature indices including chaotic indices, long-term memory features, fractal dimension, and critical bifurcation points is generated using a nonlinear dynamic feature extraction method. S3. Based on the dynamic characteristic index set, an adaptive collaborative measurement network is constructed for anomaly monitoring, and an improved probabilistic synchronous parallel mechanism is combined to carry out cross-node collaborative decision-making, generating anomaly detection results and collaborative control commands. S4. Based on the anomaly detection results and collaborative control instructions, a dynamic threshold adjustment algorithm is used to generate real-time monitoring and early warning information; S5. Based on real-time monitoring and early warning information, use neural network adaptive control strategies and resource scheduling optimization algorithms to generate adaptive control commands and dynamic resource allocation schemes.

[0015] The principle of a groundwater environment automatic monitoring superstation status supervision method in this embodiment is as follows: This method achieves accurate supervision of the status of groundwater environment monitoring wells by constructing a closed-loop supervision system of "multi-source data acquisition - nonlinear feature mining - distributed collaborative decision-making - dynamic early warning control - intelligent resource scheduling". First, relying on a multi-source heterogeneous sensor network, multi-dimensional time-series data of groundwater level, water quality, well structure and surrounding environment are collected and spatiotemporally aligned to provide comprehensive input for system status analysis. Second, targeting the nonlinear dynamic characteristics of the groundwater system, the CC method is used to reconstruct the high-dimensional phase space, and combined with Wolf algorithm, GP algorithm, R / S analysis method and wavelet transform and other techniques, chaotic indexes, long-term memory features, fractal dimension and critical bifurcation points are extracted to accurately characterize the complexity of the system. The system first identifies dynamic behavior. Then, based on a set of dynamic feature indicators, it uses variational autoencoder dimensionality reduction and fine-grained expansion of the isolated forest to initially detect anomalies. Next, an improved probabilistic synchronization mechanism is used to achieve distributed node intelligence sharing, and a federated learning framework is combined to complete multi-node collaborative decision-making, generating anomaly detection results and collaborative control commands. Subsequently, a Q-learning reinforcement learning algorithm is introduced to dynamically adjust the detection threshold, and combined with real-time data streams to generate a multi-level early warning threshold set, triggering graded early warning signals corresponding to the risk level. Finally, a hybrid strategy combining BP neural networks and model reference adaptive control is used to generate dynamic control commands. Through SDN architecture and Kubernetes containerized deployment, intelligent scheduling of computing nodes and network resources is achieved, forming a collaborative mechanism of adaptive control and resource optimization. This system effectively captures the nonlinear characteristics of groundwater systems, solving the problems of reliance on human experience and low collaborative efficiency in traditional methods, and improving the accuracy of anomaly detection and the level of intelligent system supervision.

[0016] As an optional embodiment of the present invention, optionally, in step S2, based on a multi-source dynamic time-series dataset, a set of dynamic feature indices including chaotic indices, long-term memory features, fractal dimension, and critical bifurcation points is generated using a nonlinear dynamic feature extraction method, including: S201. Based on a multi-source dynamic time series dataset, the optimal delay time and embedding dimension are determined using the CC method. The high-dimensional phase space is reconstructed through time series to generate a reconstructed phase space dataset. In step S201, it is necessary to explain in detail that for each type of univariate monitoring sequence (such as groundwater level, pH value, well wall crack width, etc.) in the multi-source dynamic time series dataset, linear regression or wavelet threshold denoising method is first used for trend detrending and noise reduction preprocessing to eliminate interference terms caused by non-system factors such as sensor zero-point drift and ambient temperature fluctuations, resulting in a stationary univariate time series. Subsequently, the existing CC method is used to calculate the statistical quantity ΔS(τ,T) of the correlation integral: where τ is the candidate delay time, and T=(m-1)τ is the embedding window length (m is the candidate embedding dimension). The combinations of τ∈[1,τmax] (τmax is set according to the sampling frequency and sequence length, such as τmax=20 for sequences with a sampling interval of 1 hour) and T∈[τ,τmax] are traversed, and the ΔS value under each combination is calculated. ΔS reflects the spatial correlation change of the sequence under different delay times and embedding windows. When ΔS(τ,T) reaches its global minimum, the corresponding τ is the optimal delay time τopt, and T is the optimal embedding window length τopt. The optimal embedding dimension mopt is then obtained using mopt = Topt / τopt + 1. For example, for a groundwater level sequence with a sampling frequency of 1 hour / time, after traversing and calculating within the range τ∈[1,20], if ΔS reaches its minimum when τ=5, and the corresponding T=15, then the optimal embedding parameters for this sequence are τopt=5 and mopt=4. Finally, after determining the optimal embedding parameters for all univariate monitoring sequences, the reconstructed phase spaces of each univariate sequence are concatenated dimensionally to form a high-dimensional joint reconstructed phase space dataset covering multi-source features.

[0017] S202. Based on the reconstructed phase space dataset, the Wolf algorithm is used to calculate the maximum Lyapunov exponent and the GP algorithm is used to calculate the correlation dimension. The nonlinear characteristics are verified by combining the BDS test, and a chaotic index set is generated. In step S202, it should be explained in detail that the Wolf algorithm and the GP algorithm are existing algorithms, and their respective calculation logic is as follows: The Wolf algorithm's specific computational logic is as follows: For each univariate time-series reconstruction dataset (such as the mott-dimensional phase space point set of a groundwater level sequence) in the reconstructed phase space, an initial reference point X0 is randomly selected. The nearest neighbor Y0 is searched in the phase space with a distance less than a preset threshold ε (e.g., ε is 1%~5% of the phase space diameter), ensuring that the evolution directions of the two points are not collinear. Then, the reference point X0 is tracked to evolve to Xt with time step t, and its nearest neighbor Y0 evolves to Yt. The Euclidean distance between the two points is calculated as dt = ||Xt - Yt||. When dt exceeds ε or reaches the preset maximum tracking step count, a new nearest neighbor is selected, and the above process is repeated until all time steps are traversed. Finally, a linear fit is performed on the distance growth rate ln(dt / d0) of all effective time steps. The slope of the fitted line is the maximum Lyapunov exponent λ1 of the sequence. If λ1 > 0, it indicates that the sequence has chaotic characteristics. For example, for the reconstructed phase space sequence of groundwater level, the Wolf algorithm calculates λ1=0.023>0, indicating that the water level sequence has chaotic dynamics.

[0018] The GP algorithm's specific computational logic is as follows: For a univariate time-series dataset in the reconstructed phase space, firstly, the Euclidean distance between all points in the phase space is calculated, and the number of point pairs C(r) with a distance less than a given threshold r is counted. Then, by adjusting the range of r (usually gradually expanding from a small proportion of the phase space diameter), the C(r) value under different r values ​​is calculated, and the relationship curve between ln(C(r)) and ln(r) is plotted. The slope of this curve is then fitted, and the correlation dimension D² is the correlation dimension. The correlation dimension is used to characterize the fractal structural complexity of attractors in the phase space. If D² is a non-integer value, it indicates that the sequence has fractal properties. For example, in the phase space reconstruction of a groundwater pH value sequence, if D² = 2.45 is calculated, it indicates that the sequence has significant fractal characteristics.

[0019] The specific calculation logic of the BDS test is as follows: Based on the reconstructed phase space dataset, the BDS statistic is calculated for different embedding dimensions m (e.g., m=2,3,...,5). This statistic measures the degree of deviation of the sequence's distribution in the m-dimensional phase space from the distribution under the independent and identically distributed (i.i.d.) hypothesis. If the p-value of the BDS statistic is less than the significance level α (e.g., α=0.05), the i.i.d. hypothesis is rejected, verifying that the sequence has nonlinear chaotic characteristics. For example, when performing a BDS test on a wellbore fracture width sequence, the p-value is 0.012 < 0.05 when m=3, indicating that the sequence exhibits significant nonlinear chaotic behavior.

[0020] By integrating the maximum Lyapunov exponent obtained by the Wolf algorithm, the correlation dimension obtained by the GP algorithm, and the significance results of the BDS test, a set of chaotic indices is constructed to characterize the chaotic dynamic characteristics of the groundwater environment monitoring well state system.

[0021] S203. Based on the reconstructed phase space dataset, the Hurst exponent is calculated using the R / S analysis method to identify long-term memory effects. Multi-scale fractal analysis is performed by combining wavelet transform to generate a long-term memory feature set containing fluctuation patterns at different time scales. In step S203, it is necessary to explain in detail that the R / S analysis method is an existing technology. In this embodiment, the R / S analysis method is used to analyze univariate time series in the reconstructed phase space dataset to extract long-term memory features. Specifically, for each series, its cumulative deviation sequence is first calculated, and several non-overlapping time windows are divided based on the series. Within each window, the ratio of range (R) to standard deviation (S) is calculated, i.e., the R / S statistic. Subsequently, the R / S statistic under different window lengths is logarithmically transformed, and the relationship curve between log(R / S) and log(window length) is plotted. By linearly fitting this curve, its slope is the Hurst exponent H. If H > 0.5, it indicates that the series has a persistent long-term memory effect; if H = 0.5, it is a random walk; if H < 0.5, it exhibits anti-persistent characteristics. For example, in the groundwater dissolved oxygen concentration series, if H = 0.72 is calculated, it indicates that the series has significant long-term memory characteristics.

[0022] Wavelet transform techniques are used to further explore the multi-scale fractal patterns of sequences. By selecting appropriate wavelet basis functions (such as Morlet wavelets or Mexican Hat wavelets), continuous wavelet transforms are performed on the sequences to generate spectrums of wavelet coefficients changing with time and scale. The wavelet energy distribution and local modulus maxima at different scales are analyzed to identify the wave patterns and corresponding fractal characteristics of the sequences at short, medium, and long time scales. For example, wavelet analysis of well vibration signals may reveal local abrupt changes in the high-frequency band and periodic trends in the low-frequency band, thereby constructing a long-term memory feature set containing multi-scale wave information. These features can capture the complex dynamic behavior of groundwater systems.

[0023] S204. Based on the reconstructed phase space dataset, the fractal dimension is calculated using the box counting method and the information dimension method, and a fractal dimension index set is generated. In step S204, it is necessary to explain in detail that box counting is an existing technology. In this embodiment, box counting is used to calculate the fractal dimension of the reconstructed phase space dataset. Specifically, for the phase space reconstruction dataset of each univariate time series, the phase space is first divided into hypercube grids with side length ε, and the number of non-empty grids N(ε) is counted. Subsequently, by adjusting the range of ε (usually gradually decreasing from a larger scale), the N(ε) values ​​at different scales are recorded, and the relationship curve between ln(N(ε)) and ln(1 / ε) is plotted. A linear fit is performed on this curve, and its slope is the fractal dimension D0. The fractal dimension is used to characterize the geometric complexity of attractors in the phase space. If D0 is a non-integer value, it indicates that the sequence has fractal properties. For example, in the phase space reconstruction of the inclination angle sequence of groundwater well walls, if D0 = 3.12 is calculated, it indicates that the sequence has a significant fractal structure.

[0024] The information dimension method further supplements the analysis results of fractal dimension. The information dimension method measures the inhomogeneity of attractors in phase space by calculating the probability distribution P(ε) of each non-empty grid and combining it with the concept of information entropy. The specific steps include: First, based on the grid divided by box counting, calculate the probability P(ε) = n(ε) / N(ε) of each non-empty grid, where n(ε) is the number of points in that grid; then, calculate the information entropy S(ε) = -ΣP(ε)ln(P(ε)); finally, plot the relationship curve between S(ε) and ln(1 / ε) and perform linear fitting on the curve; the slope of the curve is the information dimension D1. The information dimension can more finely characterize the local properties of attractors, and is particularly suitable for analyzing complex systems with non-uniform distributions. For example, for dissolved oxygen concentration sequences collected by multi-parameter sensors for groundwater quality, if D1 = 2.85 is calculated, it indicates that the phase space attractors have significant probability distribution inhomogeneity in addition to geometric complexity.

[0025] The fractal dimension D0 obtained by the box counting method and the information dimension D1 obtained by the information dimension method are combined to form a fractal dimension index set. This index set can not only reflect the geometric complexity of the groundwater environment monitoring well status system, but also reveal the local characteristics of its probability distribution.

[0026] S205. Based on the reconstructed phase space dataset, construct a parameter sensitivity model using bifurcation theory, and use the parameter sensitivity model to perform numerical simulation and stability analysis to determine the critical parameter threshold for the transition from a steady state to an oscillating state, and generate a critical bifurcation point index set.

[0027] The specific method for constructing the parameter sensitivity model using bifurcation theory in step S205 is as follows: First, based on the physical mechanism and monitoring data of the groundwater system, a set of key control parameters that play a dominant role in the dynamic behavior of the system is selected, including core parameters such as groundwater level fluctuation amplitude, aquifer permeability coefficient, pollutant diffusion coefficient, and well structure deformation rate; Second, combining Darcy's law, the law of conservation of mass, and pollutant migration and transformation equations, a coupled dynamic equation for the state of groundwater environmental monitoring wells is constructed. Dimensionality is eliminated through dimensionless processing to obtain a standardized system dynamic model; Then, numerical simulation tools (such as MATLAB's OdeSuite or COMSOL MULTI) are used. The system employs a physics-based approach to iteratively solve the model, plotting system state response curves under different parameter combinations. Bifurcation diagram analysis identifies the critical parameter ranges that allow the system to transition from a steady state (e.g., steady flow) to an oscillating state (e.g., periodic fluctuations) or a chaotic state. Subsequently, a global sensitivity analysis method (e.g., the Sobol's exponent method) is introduced to calculate the sensitivity coefficients of each key parameter, quantifying the contribution of parameter changes to system state transitions and selecting the most sensitive dominant parameters. Finally, a refined numerical scan is performed on the dominant parameters to determine the precise parameter thresholds for typical bifurcation phenomena such as Hopf bifurcation and saddle-node bifurcation. These thresholds and their corresponding bifurcation types are integrated into a critical bifurcation point index set. For example, for a typical groundwater monitoring well, when the aquifer permeability coefficient k increases to 1.2 × 10⁻⁴ m / s, the system transitions from a stable uniform flow state to a periodically fluctuating unsteady state. This k value is the critical parameter threshold corresponding to the Hopf bifurcation and is included in the critical bifurcation point index set. These indicators can accurately capture the key nodes of sudden changes in the groundwater system state.

[0028] As an optional embodiment of the present invention, optionally, in step S3, an adaptive cooperative measurement network is constructed based on the dynamic characteristic index set for anomaly monitoring, and cross-node cooperative decision-making is performed in conjunction with an improved probabilistic synchronous parallel mechanism to generate anomaly detection results and cooperative control instructions, including: S301. Based on the dynamic feature index set, a variational autoencoder is used to perform feature dimensionality reduction and latent space mapping to generate a low-dimensional latent feature representation set. In step S301, it is necessary to explain in detail that the variational autoencoder is an existing technology. It maps a high-dimensional set of dynamic feature indicators to a low-dimensional latent space through an encoder, and simultaneously reconstructs the original features from the latent space using a decoder, thereby achieving feature dimensionality reduction and information compression. Specifically, for the input set of dynamic feature indicators, a multi-layer neural network is first used as the encoder to map complex features such as chaotic indicators, long-term memory features, fractal dimension, and critical bifurcation points into low-dimensional latent variables. Subsequently, in the decoder, another multi-layer neural network is used to reconstruct the original features from the latent variables. By minimizing the reconstruction error and the KL divergence loss function of the latent space distribution, it is ensured that the latent variables accurately represent the main information of the original features and conform to a preset probability distribution (such as a standard normal distribution). For example, for a high-dimensional dataset containing 20 dynamic feature indicators, after processing by the variational autoencoder, it can be reduced to a 5-dimensional latent feature representation set. These latent features not only retain the key dynamic characteristics of the original data but also significantly reduce the complexity of subsequent calculations. Furthermore, to improve the robustness of the model, a random noise perturbation mechanism is introduced during training, making the latent features more adaptable to small fluctuations in the input data. The final generated set of low-dimensional latent feature representations is then produced.

[0029] S302. Based on a low-dimensional latent feature representation set, a fine-grained extended isolation forest model is used to perform anomaly detection through a multi-hyperplane segmentation algorithm, generating preliminary anomaly detection results. In step S302, it is important to explain in detail that the fine-grained extended isolation forest model in this embodiment is an improved version of the traditional isolation forest algorithm, aiming to enhance the accuracy and efficiency of anomaly detection. Specifically, this model optimizes the data space partitioning method by introducing an existing multi-hyperplane segmentation algorithm. Traditional isolation forests typically use a method of randomly selecting a single feature and segmenting based on its value, while the fine-grained extended isolation forest uses multiple hyperplanes to segment the data simultaneously, thereby achieving a more refined spatial partitioning. This method can better adapt to datasets with complex distributions, especially dynamic feature data with nonlinear structures. For example, when processing low-dimensional latent feature representation sets of groundwater monitoring well status, the multi-hyperplane segmentation algorithm can establish more accurate boundaries between different feature dimensions, effectively distinguishing between normal and abnormal patterns. Furthermore, to further enhance the robustness of the model, this embodiment also introduces an adaptive sampling strategy, dynamically adjusting the sampling probability according to the local density of data points, thereby avoiding false detections or missed detections caused by uneven data distribution. The final preliminary anomaly detection result not only includes the anomaly score for each data point but also provides the location information of the anomaly region.

[0030] S303. Based on the preliminary anomaly detection results, use the improved probabilistic synchronous parallel mechanism to perform distributed node asynchronous intelligence synchronization and generate asynchronous intelligence synchronization results. In step S303, it should be explained in detail that the improved probabilistic synchronization parallel mechanism of this embodiment addresses the problems of resource redundancy, uneven latency, and insufficient consistency in traditional distributed synchronization by achieving efficient cross-node intelligence collaboration through dynamic priority scheduling, adaptive synchronization probability calculation, and conflict resolution strategies. The specific method is as follows: First, each monitoring node deploys a lightweight state awareness module to collect local preliminary anomaly detection results (including anomaly score, feature vector, and timestamp) and its own network bandwidth and computing load data in real time. Nodes are classified according to the urgency of the anomaly (anomaly score ≥ 0.8 is high priority, 0.5 ≤ anomaly score < 0.8 is medium priority, and < 0.5 is low priority). Second, an exponential function-driven dynamic synchronization probability model is introduced. The initial synchronization probability of high-priority nodes is set to 0.9 and increases linearly with the duration of the anomaly. Medium-priority nodes are adaptively adjusted based on network load (0 ≤ Load factor ≤ 1) (range 0.5-0.8), and low-priority nodes use a low probability (0.1-0.3) to save resources. Third, each node maintains an asynchronous message queue with timestamps, uses a timestamp-first strategy to resolve conflicts (retaining the latest intelligence), and uses feature vector cosine similarity (threshold 0.95) to remove duplicates to ensure message integrity. Finally, when the received intelligence covers more than 60% of the nodes, a global consistency check is initiated, the average deviation value of the anomaly features is calculated, and if it is ≤ 0.05, a unified synchronization result is generated. For example, in a 10-node network, the synchronization probability of high-priority node A (anomaly score 0.85) is 0.92, while the synchronization probability of medium-priority node B (anomaly score 0.6) is adjusted to 0.65 due to a load of 0.3. The final global deviation is 0.03. The synchronization result includes the anomaly location, characteristics, and timestamp of each node.

[0031] S304. Based on the asynchronous intelligence synchronization results, a federated learning framework is used to conduct multi-node collaborative decision-making, and the final anomaly detection results are generated through weighted voting and confidence calibration. In step S304, it is necessary to explain in detail that this embodiment adopts a horizontal federated learning architecture to achieve cross-node collaborative decision-making while ensuring the data privacy of each monitoring node. The specific steps are as follows: First, initialize the global anomaly decision-making model, and set the weighted voting weight coefficient (determined by combining the node's historical detection accuracy and current anomaly priority, such as a high-priority node with a historical accuracy of 98% having a weight of 0.25, and a medium-priority node with an accuracy of 82% having a weight of 0.12) and the confidence calibration threshold (such as 0.8 for confirmed anomaly threshold and 0.5 for suspected anomaly threshold); Second, each node trains a local lightweight anomaly classifier (such as a gradient boosting tree model) based on local preliminary anomaly results and asynchronous intelligence synchronization data, and outputs local prediction results including anomaly type, location, and confidence level; Third, through a differential privacy-preserving secure aggregation protocol (such as FedAv... The algorithm (g-algorithm) is used to weight and aggregate the model parameters uploaded by each node to generate a global anomaly decision model. Fourth, the global model is used to perform secondary verification of the anomaly characteristics of each node, and a weighted voting mechanism is initiated. If more than 75% of the nodes are judged as anomalies and the weighted confidence average is ≥0.85, it is marked as "confirmed anomaly". If 40%-75% of the nodes are judged as anomalies or the confidence average is between 0.55 and 0.85, it is marked as "suspected anomaly" and a deep verification process is triggered. If it is less than 40% or the confidence average is <0.55, it is judged as "normal". Fifth, the Platt scaling method is introduced to calibrate the confidence level, correct the prediction bias of different node models, and ensure decision consistency. For example, in a 12-node monitoring network, there are 6 high-priority nodes (total weight 1.5), 4 medium-priority nodes (total weight 0.48), and 2 low-priority nodes (total weight 0.16). The global weighted confidence average is 0.87, and 83% of the nodes detect an anomaly, ultimately generating a "confirmed anomaly" result. The anomaly type is clearly defined as "wellbore fracture propagation anomaly," the location is in the well depth range of 15-20 meters, and the confidence level is 0.89. Simultaneously, collaborative control instructions are generated: a fracture repair warning is issued to nearby maintenance nodes, and the monitoring frequency of this well is increased from once per hour to once every 15 minutes. Finally, the final anomaly detection result and control instructions are synchronized to all relevant nodes through an encrypted channel, completing the cross-node collaborative decision-making closed loop.

[0032] S305. Based on the final anomaly detection results, optimize the computing resource allocation using a dynamic resource scheduling algorithm to generate collaborative control instructions that include node load balancing and bandwidth allocation.

[0033] In step S305, it is necessary to explain in detail that the dynamic resource scheduling algorithm of this embodiment integrates deep learning prediction and reinforcement learning decision-making mechanisms to achieve efficient allocation and dynamic balancing of global resources. The specific steps are as follows: First, a multi-dimensional resource status awareness matrix is ​​constructed, and core indicators such as CPU utilization, memory usage, remaining network bandwidth, current monitoring task queue length, and anomaly handling priority of each monitoring node are collected in real time to form a standardized resource status vector; second, based on the anomaly level (confirmed anomaly / suspected anomaly / normal) in the final anomaly detection result, resource demand priority weights are assigned to each node (confirmed anomaly nodes are weighted at 1.0, suspected anomaly nodes at 0.7, and normal nodes at 0.3); then, a Long Short-Term Memory (LSTM) network model is used to predict the load change trend of each node within the next 10 minutes, and the load prediction curve is output. Next, a Deep Q-Network (DQN) is introduced as the decision engine, with the optimization objectives of maximizing global resource utilization (target value ≥ 90%) and minimizing node load balancing (variance ≤ 5%). Optimal resource scheduling strategies are generated through iterative training, including computation task migration paths, memory expansion ratios, and bandwidth reservation limits. Subsequently, a closed-loop feedback adjustment mechanism is established, collecting resource status data after scheduling every 30 seconds. If the node load deviation exceeds a preset threshold (10%), strategy fine-tuning is triggered. Finally, the scheduling strategy is transformed into executable collaborative control instructions, covering instructions for inter-node computation task migration, dynamic bandwidth allocation, and memory resource expansion. For example, for a monitoring well node marked as "confirmed anomaly" (fracture propagation at a depth of 15-20 meters), its current CPU utilization is 88%, memory usage is 75%, and remaining bandwidth is only 8Mbps. LSTM predicts that the future load will rise to 95%. The DQN algorithm decision is as follows: migrate the non-real-time historical data archiving task of this node to an adjacent idle node (CPU utilization 25%), reserve 25Mbps bandwidth for high-frequency abnormal data transmission, and expand the memory to 16GB. After scheduling, the node's CPU utilization drops to 62%, memory usage is 58%, bandwidth requirements are met, and the global load balancing variance decreases from 0.42 to 0.11. The generated collaborative control instructions specifically include: <Task Migration> Node A → Node C: Historical data archiving task (priority 0.2); <Bandwidth Allocation> Increase Node A's bandwidth to 25Mbps (reserve 20Mbps for anomaly monitoring); <Memory Adjustment> Expand Node A's memory to 16GB; all instructions are synchronized to the resource management center and relevant nodes through an encrypted channel for execution, ensuring stable and reliable resource supply during anomaly handling.

[0034] As an optional embodiment of the present invention, optionally, in step S4, generating real-time monitoring and early warning information based on the anomaly detection result and the collaborative control command using a dynamic threshold adjustment algorithm includes: S401. Based on the anomaly detection results and collaborative control instructions, the detection threshold parameters are dynamically adjusted using the Q-learning reinforcement learning algorithm, and an adaptive threshold curve is generated by optimizing the historical false alarm rate and false negative rate. In step S401, it is necessary to explain in detail that the Q-learning reinforcement learning algorithm dynamically adjusts the detection threshold parameters in the groundwater monitoring system by constructing a state-action-reward model. Specifically, the algorithm first takes the anomaly detection results and collaborative control instructions as input, defining the state space as the anomaly level of the current monitoring node, resource load status, and historical detection performance indicators (such as false alarm rate and false negative rate). The action space includes three operations: increasing, decreasing, or keeping the detection threshold unchanged. The reward function is designed to comprehensively consider the weighted benefits of reducing the false alarm rate and the false negative rate, with the weight coefficients dynamically adjusted according to the actual application scenario. For example, in a certain monitoring node, if the historical false alarm rate is 5% and the false negative rate is 8%, the reward function prioritizes optimizing the false negative rate to improve detection sensitivity. Subsequently, the algorithm explores the optimal action sequence through an ε-greedy strategy, updates the Q-value table using the Bellman equation, and gradually approaches the globally optimal strategy. During the training process, an adaptive threshold curve is generated in each iteration. This curve can dynamically adjust the threshold range according to real-time monitoring data, thereby reducing resource consumption while ensuring detection accuracy. For example, in a well depth of 15-20 meters, the initial threshold was set to 0.65. After 10 iterations, the adaptive threshold curve adjusted it to 0.72, significantly improving the accuracy and stability of the detection. Furthermore, to enhance the algorithm's generalization ability, an experience replay mechanism was introduced. Historical state-action-reward data is stored in a buffer pool and batch trained through random sampling, effectively mitigating data correlation issues. The resulting adaptive threshold curve is not only applicable to the current monitoring task but can also be transferred to other similar scenarios.

[0035] S402. Based on the adaptive threshold curve, the threshold is dynamically corrected using real-time multi-source dynamic time series data to generate a multi-level early warning threshold set. In step S402, it is necessary to explain in detail that this embodiment dynamically corrects the adaptive threshold curve using real-time multi-source dynamic time-series data to generate a multi-level early warning threshold set. Specifically, the system first collects time-series data containing multiple dimensions such as groundwater pressure, temperature, and flow velocity from multiple monitoring nodes, and compares and analyzes it with historical data to identify potential data drift or environmental change trends. For example, in a monitoring well, if groundwater pressure is found to exhibit non-linear fluctuations for three consecutive hours while the temperature remains stable, the system will automatically adjust the pressure-related early warning threshold to avoid false alarms caused by environmental disturbances. Subsequently, the time-series data is segmented using sliding window technology. Each segment of data is initially screened using the adaptive threshold curve, and its deviation degree is further evaluated using the Local Anomaly Factor (LOF) algorithm. Based on this, the system divides the early warning level into three levels—low, medium, and high—according to different threshold ranges. For example, for the monitoring task of fracture propagation anomalies in the 15-20 meter well depth range, the low-level early warning threshold is set to 0.68, the medium-level to 0.72, and the high-level to 0.75. Furthermore, to ensure the real-time performance and accuracy of threshold correction, a Kalman filter-based noise suppression mechanism was introduced to effectively eliminate random noise interference in the sensor data. The resulting multi-level early warning threshold set can not only dynamically adapt to complex and ever-changing monitoring environments, but also significantly reduce false alarm and false negative rates while maintaining detection sensitivity.

[0036] S403. Based on a multi-level early warning threshold set, real-time comparison and analysis of real-time multi-source dynamic time series data is performed, and a graded early warning signal is triggered when the detected value exceeds the corresponding threshold. S404. Utilize the node load information and communication delay data in the collaborative control instructions to optimize the push path and priority ranking of graded early warning signals, and generate real-time monitoring and early warning information that includes spatiotemporal positioning data and risk level.

[0037] In step S404, it is necessary to explain in detail that the priority ranking mechanism of this embodiment comprehensively considers node load information and communication delay data to ensure that the graded early warning signals can be efficiently and accurately pushed to the relevant processing units. Specifically, the system first evaluates the real-time processing capability of each node based on the node load status in the cooperative control command, and constructs a priority ranking model by combining the delay characteristics of the communication link. This model adopts a weighted scoring method, setting the node load weight to 0.6 and the communication delay weight to 0.4, and calculates the comprehensive priority score of each node through linear combination. For example, when a high-load node (load score of 0.9) is combined with a low-latency link (delay score of 0.2), its comprehensive score is 0.6×0.9+0.4×0.2=0.58, while when another low-load node (load score of 0.3) is combined with a high-latency link (delay score of 0.8), its comprehensive score is 0.6×0.3+0.4×0.8=0.5. Based on this, the system sorts all nodes in descending order according to the comprehensive score to determine the order of pushing graded early warning signals. Furthermore, to further improve push efficiency, spatiotemporal location data and risk level are introduced as auxiliary decision-making factors. Spatiotemporal location data is used to mark the location and spread range of abnormal events, while the risk level is dynamically adjusted according to the severity of the anomaly type. For example, for anomalies of fracture propagation in the 15-20 meter well depth range, if its risk level is assessed as "high," the system will prioritize pushing the warning signal to the maintenance node closest to the anomaly area and simultaneously notify the global monitoring center. The final generated real-time monitoring and warning information not only includes the spatiotemporal location data and risk level of the abnormal event, but also clarifies the push path and priority order of the warning signal, thereby ensuring that abnormal events can be responded to and handled in a timely manner.

[0038] As an optional embodiment of the present invention, optionally, in step S403, based on a multi-level early warning threshold set, real-time comparison and analysis are performed on real-time multi-source dynamic time-series data, and when the detected value exceeds the corresponding threshold, a graded early warning signal is triggered, including: S4031. Based on a multi-level early warning threshold set, a sliding window mechanism is used to segment and capture real-time multi-source dynamic time series data and perform edge computing preprocessing to generate real-time data stream slices. In step S4031, it is necessary to explain in detail that the system employs a sliding window mechanism to capture real-time multi-source dynamic time-series data in segments to achieve efficient data processing and analysis. Specifically, the size and step size of the sliding window are dynamically adjusted according to the needs of the monitoring task. For example, in a high-frequency monitoring scenario, the window size can be set to 5 seconds and the step size to 1 second to ensure the real-time performance and continuity of the data. Subsequently, the system performs edge computing preprocessing on each data slice, using lightweight algorithms to extract key features such as mean, variance, peak value, and trend changes, and removes redundant information to reduce transmission load. For example, in a groundwater monitoring task, the system uses edge computing to identify abnormal fluctuations in pressure data in the 15-20 meter well depth range within a short period of time, with peak values ​​exceeding the normal range by more than three times. This feature is marked as a potential risk signal and uploaded to the central processing unit. In addition, to improve the reliability of the data stream slices, a data integrity verification mechanism based on hash check is introduced to ensure that each slice is not tampered with or lost during transmission, and the final generated real-time data stream slices can reflect the dynamic change trend of the monitored object.

[0039] S4032. Based on real-time data stream slices, a dynamic programming algorithm is used to perform fast matching calculations between multi-level thresholds and real-time data stream slices, and a set of parameters for over-threshold amplitude and duration is generated by the difference comparison method. In step S4032, it is necessary to explain in detail that the system uses existing dynamic programming algorithms (such as the Bellman-Ford algorithm) to quickly match and calculate multi-level thresholds with real-time data stream slices to improve detection efficiency and accuracy. Specifically, the algorithm first compares the multi-level warning threshold set with the key features in the real-time data stream slice one by one, and identifies the magnitude and duration of exceeding the threshold through differential comparison. For example, in a certain monitoring task, if the pressure value of the real-time data stream slice exceeds the high-level warning threshold of 0.75 at a certain moment and remains there for more than 3 seconds, the system will record the magnitude of the exceeding threshold event as 0.08 and the duration as 3.2 seconds. Subsequently, these parameters are further analyzed to determine the severity of the anomaly and potential risks. To optimize the performance of the matching calculation, the algorithm introduces a pruning strategy to remove data segments that cannot meet the threshold conditions, thereby significantly reducing computational complexity. In addition, the system also combines the pattern features in historical data to perform contextual correlation analysis on the exceeding threshold events to avoid false alarms caused by isolated events. The final generated set of exceeding threshold magnitude and duration parameters can accurately describe the dynamic characteristics of the abnormal event.

[0040] S4033. Based on the set of parameters for the magnitude and duration of the over-threshold, and using the graded early warning triggering rules, generate a preliminary early warning signal corresponding to the risk level. In step S4033, it is necessary to explain in detail that the system generates a preliminary warning signal corresponding to the risk level based on the set of parameters for the amplitude and duration of the exceedance threshold, combined with the graded warning triggering rules. Specifically, the graded warning triggering rules dynamically evaluate the amplitude deviation and duration of the abnormal event. For example, when the exceedance threshold amplitude is small but the duration is long, the system determines it as a medium risk; while when the amplitude deviates significantly and the duration exceeds the preset threshold, a high-risk warning is triggered. In addition, to improve the accuracy of the warning signal, a multi-dimensional correlation analysis mechanism is introduced, which comprehensively considers the amplitude, duration, and other environmental parameters (such as temperature and flow rate changes). For example, in a certain monitoring task, if the groundwater pressure exceeds the threshold amplitude by 0.08 and the duration is 3.2 seconds, accompanied by a slight temperature fluctuation, the system will mark it as a medium risk and generate a corresponding preliminary warning signal. This process ensures the scientific nature of the warning signal and effectively avoids the problem of false alarms or missed alarms that may be caused by judging by a single indicator.

[0041] S4034. Based on the preliminary warning signal, combined with spatiotemporal positioning data, and using spatial clustering algorithms and time trend analysis, generate a hierarchical warning signal that includes location, time, and amplitude.

[0042] In step S4034, it is necessary to explain in detail that the system further processes the preliminary warning signal using spatial clustering algorithms (such as DBSCAN or K-means) and time trend analysis to generate a hierarchical warning signal containing location, time, and amplitude. Specifically, the spatial clustering algorithm first analyzes the distribution of monitoring nodes, grouping nodes with similar geographical locations and similar abnormal characteristics into the same cluster to identify potential regional risks. For example, in a monitoring task, if the pressure data of multiple adjacent nodes in the well depth range of 15-20 meters all exceed the threshold, the system will classify these nodes into a risk cluster and mark it as the central area. Subsequently, the time trend analysis module deeply mines the time series characteristics of the abnormal event, identifying the start time, peak time, and possible duration of the anomaly. For example, for a certain cluster of nodes, the system uses time trend analysis to find that the pressure anomaly gradually evolves from an initial small fluctuation to a significant deviation, reaching its peak within 30 minutes; this process is recorded as a key feature in the time dimension. Furthermore, to improve the accuracy of the tiered early warning signals, the system integrates spatiotemporal positioning data, combining the location information, time trajectory, and amplitude changes of abnormal events into a unified descriptive model. For example, for a node in the aforementioned risk cluster, the tiered early warning signal generated by the system clearly marks its geographical location (latitude and longitude coordinates), the time of the anomaly (accurate to the second), and the specific value of the threshold amplitude (e.g., 0.08). The resulting tiered early warning signal can clearly reflect the spatial distribution and temporal evolution of abnormal events.

[0043] As an optional embodiment of the present invention, optionally, in step S5, based on real-time monitoring and early warning information, the adaptive control command and dynamic resource allocation scheme are generated using a neural network adaptive control strategy and a resource scheduling optimization algorithm, including: S501. Based on the spatiotemporal positioning data and risk level in the real-time monitoring and early warning information, a preliminary control command containing dynamic adjustment of the sampling frequency is generated using a neural network adaptive control strategy. In step S501, the neural network adaptive control strategy is a hybrid strategy based on a combination of BP neural network and model reference adaptive control. Backpropagation neural networks are used to identify the dynamic nonlinear characteristics of groundwater environment monitoring well status systems online and obtain identification results. Model reference adaptive control optimizes the control law parameters in real time based on the identification results, generating preliminary control commands that include dynamic adjustment of the sampling frequency.

[0044] The specific implementation process of the strategy is as follows: First, the response characteristics of the ideal monitoring system are set as a reference model, and the error signal between the output of the actual monitoring system and the output of the reference model is collected and learned online through a BP neural network. Second, the weights and thresholds of the BP neural network are updated using the gradient descent method to achieve accurate online approximation of the dynamic characteristics of the monitoring system. Finally, the model reference adaptive controller adjusts the control parameters (including but not limited to the sampling frequency adjustment coefficient and sensor trigger threshold) in real time according to the system dynamic model parameters output by the BP neural network, so that the output of the actual monitoring system tracks the expected output of the reference model, thereby generating sampling frequency dynamic adjustment instructions and equipment operation status optimization suggestions that are adapted to the current spatiotemporal positioning data and risk level.

[0045] S502. Based on the risk level and the operating status of the equipment in the monitoring well, the uncertainty of the groundwater environment monitoring well status system is approximated by the RBF neural network, the control parameters are optimized and adaptive control commands including resource demand parameters are generated. In step S502, it is necessary to explain in detail that the existing RBF neural network, through the characteristics of its radial basis function, can efficiently approximate the uncertainties in the groundwater environment monitoring well status system. Specifically, the system first constructs the input feature set of the RBF neural network based on the risk level and equipment operating status in the real-time monitoring and early warning information. These features include, but are not limited to, sensor sensitivity drift and the fluctuation range of environmental parameters. Subsequently, the network uses the existing Gaussian kernel function to perform nonlinear mapping on the input data and optimizes its center point position and width parameters through the least squares method, thereby achieving accurate modeling of system uncertainties. For example, in a monitoring task, if it is found that the sensor sensitivity of a certain monitoring node has decreased due to long-term use, the RBF neural network will automatically adjust its corresponding control parameters to compensate for this deviation. In addition, to further improve the adaptability of the control commands, the system introduces a parameter optimization mechanism based on a genetic algorithm, which ensures that the control parameters maintain optimal performance in a complex and ever-changing monitoring environment through a global search. The finally generated adaptive control commands not only include resource requirement parameters but also specify optimization suggestions for equipment operating status.

[0046] S503, based on resource requirement parameters of adaptive control instructions, utilizes SDN architecture combined with Kubernetes containerized deployment, optimizes the allocation of computing nodes and network bandwidth through HPA automatic scaling mechanism, and generates a preliminary dynamic resource allocation scheme; S504. Based on the initial dynamic resource allocation scheme, the resource utilization efficiency is evaluated using a reinforcement learning-driven mechanism, the resource scheduling strategy is iteratively optimized, and an optimized dynamic resource allocation scheme is generated.

[0047] In step S504, it is necessary to explain in detail that the system employs a reinforcement learning-driven mechanism to dynamically evaluate resource utilization efficiency and generates a final dynamic resource allocation scheme by iteratively optimizing resource scheduling strategies. Specifically, the reinforcement learning model uses resource utilization, task response time, and energy consumption levels as key indicators to construct a reward function to guide strategy optimization. For example, in a monitoring task, if it is found that the CPU utilization of computing nodes is consistently high and network bandwidth is bottlenecked, the reinforcement learning model will adjust the resource allocation weights according to the current state, prioritizing the expansion of the number of container instances on high-load nodes and reallocating bandwidth resources. Simultaneously, the system combines pattern features from historical task data and uses the Deep Q-Network (DQN) algorithm to predict the long-term returns of different resource scheduling strategies, thereby selecting the optimal solution. To further improve optimization efficiency, the model introduces a strategy exploration mechanism based on Monte Carlo tree search, which simulates and evaluates potential scheduling schemes to avoid the occurrence of local optima. Furthermore, the system dynamically updates the environmental parameters of the reinforcement learning model by monitoring the running status of each node in real time, ensuring that it can quickly adapt to changes in the monitoring environment. The resulting optimized dynamic resource allocation scheme not only significantly improved resource utilization efficiency but also effectively ensured the stability and reliability of the monitoring task.

[0048] As an optional embodiment of the present invention, optionally, in step S503, based on the resource requirement parameters of the adaptive control instructions, using the SDN architecture combined with Kubernetes containerized deployment, and optimizing the allocation of computing nodes and network bandwidth through the HPA automatic scaling mechanism, a preliminary dynamic resource allocation scheme is generated, including: S5031. Based on the resource requirement parameters in the adaptive control instructions, the SDN controller dynamically configures the forwarding policies of network switches and routers to generate a programmable network resource pool. In step S5031, the SDN controller, through its centralized management capabilities, dynamically adjusts the forwarding strategies of network switches and routers to meet the network resource demands of the monitoring task. Specifically, the system first analyzes the current network load distribution based on the resource demand parameters in the adaptive control command and identifies high-priority data transmission paths. For example, in a monitoring task, if a significant increase in sensor data uploads is detected in a certain area, the SDN controller will prioritize allocating more bandwidth resources to that area and optimize the packet forwarding path to reduce transmission latency and packet loss rate. Furthermore, to improve network resource utilization, the system introduces a dynamic scheduling mechanism based on traffic prediction. Using historical traffic data and real-time monitoring information, it predicts network load trends over a future period and adjusts forwarding strategies in advance. This process not only ensures efficient transmission of critical data but also effectively avoids network congestion.

[0049] S5032. Based on a programmable network resource pool and combined with the Pod resource model of Kubernetes containerized deployment, a container-level resource allocation template is generated by mapping resource requirement parameters to container resource quotas through custom resource definitions. In step S5032, Kubernetes, through its flexible container orchestration capabilities, maps the resource requirement parameters in the adaptive control instructions to specific container resource quotas, thereby generating a container-level resource allocation template. Specifically, the system first defines custom resources (CRDs) based on the configuration of the programmable network resource pool to describe the computing, storage, and network resource requirements of different monitoring tasks. For example, in a high-load monitoring task, if a container needs to process a large amount of real-time data streams, the system will allocate higher CPU and memory quotas to it while ensuring its network bandwidth priority. Subsequently, the Kubernetes scheduler dynamically adjusts the Pod deployment strategy according to the generated resource allocation template to achieve efficient resource utilization. To further improve the accuracy of resource allocation, the system introduces a feedback mechanism based on real-time monitoring data. By continuously tracking the container's running status, it dynamically updates the resource quota configuration to avoid resource waste or insufficiency. In addition, to cope with sudden changes in resource demand, the system also incorporates the HPA (HorizontalPodAutoscaler) mechanism to automatically expand or shrink the number of Pod instances based on load fluctuations, thereby ensuring the stable operation of monitoring tasks. The final generated container-level resource allocation template can meet the needs of complex and ever-changing monitoring tasks.

[0050] S5033: Based on the container-level resource allocation template, the HPA automatic scaling mechanism is used to dynamically adjust the number of Pod replicas and resource limits to generate a compute node load balancing optimization scheme. In step S5033, it is necessary to explain in detail that the system uses the HPA (Hyper-Package Scale-Up) mechanism to dynamically adjust the number of Pod replicas and resource limits based on the load metrics in the container-level resource allocation template, thereby generating a compute node load balancing optimization scheme. Specifically, the HPA mechanism uses key performance indicators such as CPU utilization and memory usage as a basis to monitor the running status of each Pod in real time, and automatically expands the number of Pod replicas when the load exceeds a preset threshold, and reduces it otherwise. For example, in a monitoring task, if the CPU utilization of a compute node is found to be consistently higher than 80%, the system will immediately trigger the HPA mechanism to add Pod instances to share the load pressure. At the same time, to ensure the rationality of resource allocation, the system combines historical load data and real-time monitoring information to predict the load change trend in the future period and adjust the number of Pod replicas and resource limits in advance. In addition, to further improve the load balancing effect, the system introduces a weight-based scheduling strategy, which dynamically adjusts the task allocation weight according to the hardware performance and current load of different compute nodes to avoid resource bottlenecks or uneven load. The resulting compute node load balancing optimization scheme can significantly improve the overall operating efficiency of the system while ensuring the stability and reliability of the monitoring tasks.

[0051] S5034. Based on the computing node load balancing optimization scheme, combined with SDN traffic engineering algorithms, network bandwidth allocation is optimized to generate a network resource topology optimization map. In step S5034, it is necessary to explain in detail that the system optimizes network bandwidth allocation using existing SDN traffic engineering algorithms to generate a network resource topology optimization map. Specifically, the SDN controller analyzes bottleneck nodes and links in the current network topology based on the resource allocation in the compute node load balancing optimization scheme, and dynamically adjusts the bandwidth allocation strategy to improve overall transmission efficiency. For example, in a monitoring task, if the bandwidth utilization of a critical link is found to be close to saturation, the system will prioritize expanding the capacity of that link and replan the data flow path to avoid congestion caused by concentrated traffic. In addition, to further improve the utilization efficiency of network resources, the system introduces a traffic scheduling mechanism based on multi-objective optimization, comprehensively considering indicators such as bandwidth utilization, transmission latency, and packet loss rate to select the optimal data forwarding strategy. This process not only ensures the rapid transmission of high-priority monitoring data but also effectively improves the overall carrying capacity of the network. The final generated network resource topology optimization map can intuitively reflect the current distribution of network resources and the optimized data flow paths.

[0052] S5035. Integrate the computing node load balancing optimization scheme with the network resource topology optimization diagram to form a preliminary dynamic resource allocation scheme that includes joint node-network optimization.

[0053] In step S5035, it is necessary to explain in detail that the system integrates the computing node load balancing optimization scheme with the network resource topology optimization diagram to form a preliminary dynamic resource allocation scheme that includes joint node-network optimization. Specifically, this scheme coordinates the collaborative optimization of computing and network resources from a global perspective, ensuring that the monitoring task can still maintain efficient operation under high load or sudden anomalies. For example, in a complex monitoring task, if it is found that some computing nodes are experiencing processing delays due to insufficient resources, and their associated network links are experiencing bandwidth bottlenecks, the system will reallocate computing tasks based on the joint optimization strategy and adjust the traffic distribution of related links, thereby maximizing resource utilization. In addition, to improve the adaptability of the scheme, the system introduces a dynamic adjustment mechanism based on real-time feedback. By continuously monitoring the status of nodes and the network, the optimization strategy is updated in a timely manner to avoid the problem of lag in resource allocation. The final generated preliminary dynamic resource allocation scheme can not only meet the needs of the current monitoring task.

[0054] As an optional embodiment of the present invention, the method may further include: S6. Based on the execution feedback data of adaptive control commands and dynamic resource allocation schemes, a multi-indicator comprehensive evaluation model is used to conduct quantitative analysis of regulatory effects, and the model parameters are updated by combining an incremental learning mechanism to continuously optimize the method.

[0055] Based on execution feedback data (including core indicators such as anomaly handling success rate, resource utilization rate, early warning response time, and false alarm / missed alarm rate), a multi-dimensional evaluation index system covering regulatory efficiency, decision-making accuracy, and system stability is constructed. The Analytic Hierarchy Process (AHP) is used to determine the weights of each index, and the fuzzy comprehensive evaluation method is combined to quantify the regulatory effect and generate a comprehensive evaluation result. Based on the comprehensive evaluation result, bottlenecks in the method execution process (such as resource scheduling delays and insufficient feature extraction accuracy) are identified, and targeted optimization suggestions are output. An incremental learning mechanism is used to merge newly collected feedback data with historical datasets, updating the key parameters of the nonlinear dynamics feature extraction model, anomaly detection model, and resource scheduling algorithm online, enhancing the method's adaptability to complex groundwater environment changes. Regulatory effect evaluation reports are generated periodically, including comprehensive scores, optimization suggestions, and system performance improvement paths.

[0056] In step S6, it is also necessary to explain in detail the construction of a multi-dimensional evaluation indicator system covering regulatory efficiency, decision-making accuracy, and system stability. This multi-dimensional evaluation indicator system needs to be developed from three core dimensions: regulatory efficiency, decision-making accuracy, and system stability. Each dimension is further refined into specific quantifiable and monitorable indicators, and their definitions and calculation methods are clearly defined. The regulatory efficiency dimension includes: ① resource utilization (average CPU / memory utilization of computing nodes, average ratio of actual network bandwidth occupancy to allocated bandwidth), ② task response time (average time from completion of monitoring data collection to generation of adaptive control commands), and ③ anomaly handling turnaround time (total time from triggering an anomaly to completion of resource adjustment or equipment parameter compensation). For example, if the task response time of a certain monitoring well area exceeds a preset threshold (e.g., 3 seconds) five times consecutively, it is determined that there is a bottleneck in the regulatory efficiency of that area.

[0057] Decision accuracy dimensions include: ① anomaly detection accuracy (the proportion of correctly identified anomalies out of the total number of anomalies), ② control command adaptation rate (the proportion of times adjusted equipment parameters or resource configurations effectively resolve system deviations out of the total number of adjustments), and ③ false alarm / missed alarm rate (number of false anomaly alerts / total number of alerts, number of unidentified real anomalies / total number of real anomalies). For example, if the anomaly detection false alarm rate is higher than 1.5% in a certain quarter, it is necessary to backtrack on the input feature set and parameter optimization logic of the RBF neural network.

[0058] System stability dimensions include: ① Service availability (the percentage of time compute nodes and network links remain fault-free out of the total monitoring time), ② Resource allocation volatility (the standard deviation of resource quota adjustments per unit time), and ③ Fault self-healing recovery time (the average time for the system to automatically resume normal operation after a compute node crash or network link interruption). For example, service availability should be maintained above 99.9%. If it falls below this threshold, the redundant link configuration of the SDN architecture or the Kubernetes Pod fault tolerance mechanism needs to be checked.

[0059] Furthermore, this indicator system supports dynamic weight adjustment: for different monitoring scenarios (such as the emergency period of a sudden rise in groundwater level and the routine steady-state monitoring period), the weights of each dimension are redistributed through the analytic hierarchy process. For example, during the emergency period, the weight of anomaly handling turnaround time is increased from 15% to 30% to prioritize ensuring system response speed; during the routine period, the weight of resource utilization rate is increased to 25% to optimize resource cost-effectiveness.

[0060] Example 2 A groundwater environment automatic monitoring superstation status monitoring system includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement a method for monitoring the status of an automatic groundwater environment monitoring superstation when executing executable instructions.

[0061] It should be noted that the computer device includes a processor, a memory, and may also include one or more of a multimedia component, an input / output (I / O) interface, and a communication component.

[0062] The processor controls the overall operation of the computer device to complete all or part of the steps in the above-mentioned method for monitoring the status of the super station for automatic monitoring of the groundwater environment.

[0063] Memory is used to store various types of data to support the operation of the computer device. This data may include, for example, instructions for any application or method used to operate on the computer device, as well as application-related data. Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0064] The multimedia component may include a screen and an audio component, wherein the screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals; for example, the audio component may include a microphone for receiving external audio signals, the received audio signals may be further stored in memory or transmitted via a communication component; the audio component may also include at least one speaker for outputting audio signals.

[0065] I / O interfaces provide interfaces between the processor and other interface modules, such as keyboards, mice, buttons, etc.; these buttons can be virtual buttons or physical buttons.

[0066] The communication component is used for wired or wireless communication between the computer device and other devices; wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G or 5G, or one or more combinations thereof, and the corresponding communication component may include: Wi-Fi module, Bluetooth module, NFC module, mobile communication module.

[0067] As a preferred embodiment, the computer device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described automatic groundwater environment monitoring superstation status supervision method.

[0068] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for monitoring the status of an automatic groundwater environment monitoring superstation, characterized in that, The method includes: A multi-source heterogeneous sensor network is deployed in the groundwater environment monitoring well detection area to collect multi-source dynamic time-series datasets and perform spatiotemporal alignment; the multi-source dynamic time-series datasets include pH value, conductivity, dissolved oxygen concentration, turbidity, permanganate index, lead heavy metal ion content and cadmium heavy metal ion content that can be collected in real time by water quality multi-parameter sensors. Based on the multi-source dynamic time-series dataset, a set of dynamic feature indices including chaotic indices, long-term memory features, fractal dimension, and critical bifurcation points is generated using a nonlinear dynamic feature extraction method. Based on the aforementioned set of dynamic characteristic indicators, an adaptive collaborative measurement network is constructed for anomaly monitoring, and an improved probabilistic synchronous parallel mechanism is combined to perform cross-node collaborative decision-making, generating anomaly detection results and collaborative control commands. Based on the anomaly detection results and collaborative control instructions, a dynamic threshold adjustment algorithm is used to generate real-time monitoring and early warning information; Based on the real-time monitoring and early warning information, an adaptive control command and a dynamic resource allocation scheme are generated using a neural network adaptive control strategy and a resource scheduling optimization algorithm. The improved probabilistic synchronous parallel mechanism includes: First, each monitoring node deploys a lightweight state-aware module to collect local preliminary anomaly detection results and its own network bandwidth and computing load data in real time, classifying nodes according to the urgency of the anomaly. Second, an exponential function-driven dynamic synchronization probability model is introduced: high-priority nodes have an initial synchronization probability of 0.9 that increases linearly with the duration of the anomaly; medium-priority nodes adjust adaptively based on network load; and low-priority nodes use a low probability to conserve resources. Third, each node maintains an asynchronous message queue with timestamps, using a timestamp-first strategy to resolve conflicts and deduplication through feature vector cosine similarity to ensure message integrity. Finally, when the received intelligence covers more than 60% of the nodes, a global consistency check is initiated, calculating the average deviation value of the anomaly features; if it is ≤0.05, a unified synchronization result is generated. The generation of real-time monitoring and early warning information includes: Based on the anomaly detection results and collaborative control instructions, the detection threshold parameters are dynamically adjusted using the Q-learning reinforcement learning algorithm, and an adaptive threshold curve is generated through the optimization strategy based on historical false alarm rate and false negative rate. Based on the adaptive threshold curve, the threshold is dynamically corrected using real-time multi-source dynamic time-series data to generate a multi-level early warning threshold set. Based on the multi-level early warning threshold set, real-time comparison and analysis are performed on real-time multi-source dynamic time series data, and a graded early warning signal is triggered when the detected value exceeds the corresponding threshold. By utilizing node load information and communication delay data in the collaborative control commands, the push path and priority ranking of graded early warning signals are optimized to generate real-time monitoring and early warning information that includes spatiotemporal positioning data and risk level.

2. The method for monitoring the status of an automatic groundwater environment superstation as described in claim 1, characterized in that, The generation of a set of dynamic characteristic indicators, including chaos indices, long-term memory features, fractal dimension, and critical bifurcation points, includes: Based on the multi-source dynamic time series dataset, the optimal delay time and embedding dimension are determined using the CC method, and a high-dimensional phase space is reconstructed through time series to generate a reconstructed phase space dataset. Based on the reconstructed phase space dataset, the Wolf algorithm is used to calculate the maximum Lyapunov exponent and the GP algorithm is used to calculate the correlation dimension. The nonlinear characteristics are verified by combining the BDS test, and a set of chaotic indexes is generated. Based on the reconstructed phase space dataset, the Hurst exponent is calculated using the R / S analysis method to identify long-term memory effects. Multi-scale fractal analysis is then performed using wavelet transform to generate a long-term memory feature set containing fluctuation patterns at different time scales. Based on the reconstructed phase space dataset, the fractal dimension is calculated using the box counting method and the information dimension method, and a fractal dimension index set is generated. Based on the reconstructed phase space dataset, a parameter sensitivity model is constructed using bifurcation theory. The parameter sensitivity model is then used for numerical simulation and stability analysis to determine the critical parameter thresholds for the transition from a steady state to an oscillating state, thereby generating a critical bifurcation point index set.

3. The method for monitoring the status of an automatic groundwater environment monitoring superstation as described in claim 1, characterized in that, The generation of anomaly detection results and collaborative control instructions includes: Based on the dynamic feature index set, a low-dimensional latent feature representation set is generated by using a variational autoencoder to perform feature dimensionality reduction and latent space mapping. Based on the low-dimensional latent feature representation set, a fine-grained extended isolation forest model is used, and anomaly detection is performed through a multi-hyperplane segmentation algorithm to generate preliminary anomaly detection results. Based on the preliminary anomaly detection results, an improved probabilistic synchronous parallel mechanism is used to perform distributed node asynchronous intelligence synchronization to generate asynchronous intelligence synchronization results. Based on the asynchronous intelligence synchronization results, a federated learning framework is used for multi-node collaborative decision-making, and the final anomaly detection results are generated through weighted voting and confidence calibration. Based on the final anomaly detection results, a dynamic resource scheduling algorithm is used to optimize the computational resource allocation and generate collaborative control instructions that include node load balancing and bandwidth allocation.

4. The method for monitoring the status of an automatic groundwater environment monitoring superstation as described in claim 1, characterized in that, The following are the triggering signals for graded early warning: Based on the multi-level early warning threshold set, a sliding window mechanism is used to segment and capture real-time multi-source dynamic time-series data and perform edge computing preprocessing to generate real-time data stream slices. Based on the real-time data stream slices, a dynamic programming algorithm is used to perform fast matching calculations between multi-level thresholds and real-time data stream slices, and a set of parameters for over-threshold amplitude and duration is generated by the difference comparison method. Based on the set of parameters for the magnitude and duration of the over-threshold, a preliminary warning signal corresponding to the risk level is generated using the graded warning triggering rules. Based on the preliminary warning signal, combined with spatiotemporal positioning data, a hierarchical warning signal containing location, time, and amplitude is generated using spatial clustering algorithms and time trend analysis.

5. The method for monitoring the status of an automatic groundwater environment superstation as described in claim 1, characterized in that, The neural network adaptive control strategy is a hybrid strategy based on a combination of BP neural network and model reference adaptive control; The BP neural network is used to identify the dynamic nonlinear characteristics of the groundwater environment monitoring well status system online and obtain the identification results. The model reference adaptive control optimizes the control law parameters in real time based on the identification results, generating preliminary control commands that include dynamic adjustment of the sampling frequency.

6. The method for monitoring the status of an automatic groundwater environment superstation as described in claim 1, characterized in that, The initial dynamic resource allocation scheme includes: Based on the resource requirement parameters in the adaptive control instructions, the SDN controller dynamically configures the forwarding strategies of network switches and routers to generate a programmable network resource pool. Based on the programmable network resource pool and combined with the Pod resource model of Kubernetes containerized deployment, a container-level resource allocation template is generated by mapping resource requirement parameters to container resource quotas through custom resource definitions. Based on the container-level resource allocation template, the HPA automatic scaling mechanism is used to dynamically adjust the number of Pod replicas and resource limits, generating a compute node load balancing optimization scheme. Based on the aforementioned computing node load balancing optimization scheme, network bandwidth allocation is optimized by combining SDN traffic engineering algorithms to generate a network resource topology optimization map. By integrating the aforementioned computing node load balancing optimization scheme with the network resource topology optimization diagram, a preliminary dynamic resource allocation scheme is formed, which includes joint node-network optimization.

7. The method for monitoring the status of an automatic groundwater environment superstation as described in claim 1, characterized in that, The method further includes: Based on the execution feedback data of the adaptive control commands and dynamic resource allocation scheme, a multi-index comprehensive evaluation model is used to quantitatively analyze the regulatory effect, and the model parameters are updated by combining an incremental learning mechanism to continuously optimize the method.

8. The method for monitoring the status of an automatic groundwater environment monitoring superstation as described in claim 1, characterized in that, The generation of adaptive control commands and dynamic resource allocation schemes includes: Based on the spatiotemporal positioning data and risk level in the real-time monitoring and early warning information, a preliminary control command containing dynamic adjustment of the sampling frequency is generated using a neural network adaptive control strategy. Based on the risk level and the operating status of the equipment in the monitoring well, the uncertainty of the groundwater environment monitoring well status system is approximated by the RBF neural network, the control parameters are optimized, and adaptive control commands including resource demand parameters are generated. Based on the resource requirement parameters of the adaptive control instructions, using the SDN architecture combined with Kubernetes containerized deployment, and optimizing the allocation of computing nodes and network bandwidth through the HPA automatic scaling mechanism, a preliminary dynamic resource allocation scheme is generated. Based on the preliminary dynamic resource allocation scheme, a reinforcement learning-driven mechanism is used to evaluate resource utilization efficiency, iteratively optimize the resource scheduling strategy, and generate an optimized dynamic resource allocation scheme.

9. A groundwater environment automatic monitoring superstation status supervision system, characterized in that, The system includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement the groundwater environment automatic monitoring superstation status supervision method according to any one of claims 1 to 8 when executing the executable instructions.

Images