An intelligent monitoring system and method for water conservancy projects

Abstract

The application relates to the technical field of water conservancy engineering monitoring, and discloses a water conservancy engineering intelligent monitoring system and method. The method comprises the following steps: acquiring water level fluctuation sequences, flow rate change sequences and gate opening change sequences and other multi-source monitoring data of water conservancy engineering monitoring points; performing space-time feature extraction processing on the multi-source monitoring data to generate a water conservancy engineering operation state feature matrix containing water level fluctuation features, flow rate change features and gate opening change features and space-time correlation data; subsequently inputting the feature matrix into an anomaly detection model to generate a water conservancy engineering abnormal operation state identifier containing abnormal type, abnormal level and abnormal position information; and finally generating a water conservancy engineering regulation and control instruction set containing a water level regulation scheme, a flow rate control scheme and a gate opening adjustment scheme based on the abnormal operation state identifier. The method can comprehensively monitor the engineering operation state, accurately identify abnormalities, quickly generate regulation and control schemes, and improve the intelligence and efficiency of water conservancy engineering monitoring.

Description

Technical Field

[0001] This invention relates to the field of water conservancy project monitoring technology, specifically to an intelligent monitoring system and method for water conservancy projects. Background Technology

[0002] In the operation and management of water conservancy projects, real-time monitoring and effective control of the project's operational status are crucial for ensuring its safe and stable operation. Traditional water conservancy project monitoring methods largely rely on manual inspections and data collection from single monitoring devices, which have significant limitations. Manual inspections are constrained by the experience, physical strength, and frequency of inspection personnel, making it difficult to achieve real-time monitoring of the entire water conservancy project around the clock. This often results in monitoring blind spots or untimely monitoring, causing potential operational anomalies to go undetected. Furthermore, data recording during manual inspections often relies on paper records or simple spreadsheets, leading to inefficient data processing and analysis, and hindering the rapid and accurate assessment of the project's operational status.

[0003] Data collection using a single monitoring device typically only acquires one of the following data points: water level, flow velocity, or gate opening. This approach fails to enable collaborative acquisition and comprehensive analysis of multi-source data. The operational status of a water conservancy project is a complex systemic issue; water level, flow velocity, and gate opening are closely interrelated, and a single data point cannot fully reflect the actual operational status of the project. For example, when only an abnormal rise in water level is detected, without corresponding analysis of flow velocity and gate opening data, it is difficult to determine whether the abnormal water level is caused by increased upstream water flow, gate malfunction, or changes in flow velocity, thus hindering the development of targeted control measures.

[0004] Traditional monitoring methods also have significant shortcomings in data processing and anomaly detection. Traditional methods often rely on simple threshold judgments for anomaly identification, setting a fixed threshold range and classifying data as abnormal when it exceeds that range. This approach fails to consider the spatiotemporal correlation of data, ignoring the trends of data changes over time during the operation of water conservancy projects and the spatial relationships between data from different monitoring points, easily leading to misjudgments or missed detections. For example, during the flood season, water levels gradually rise over time. Using a fixed threshold might mistake a normal rise in water level for anomalies, or fail to identify anomalies in the early stages of an abnormal rise because the threshold has not yet been reached. Furthermore, traditional methods struggle to quickly generate scientifically sound control instructions after anomaly detection. They typically rely on the experience of technical personnel to manually formulate control plans, which compromises the efficiency and scientific rigor of these plans, potentially delaying optimal control measures and impacting the safe operation of water conservancy projects. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent monitoring system and method for water conservancy projects to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent monitoring method for water conservancy projects, the method comprising:

[0007] Acquire multi-source monitoring data from water conservancy project monitoring points, including water level fluctuation sequences, flow velocity change sequences, and gate opening change sequences;

[0008] The multi-source monitoring data is processed by spatiotemporal feature extraction to generate a water conservancy project operation status feature matrix. The water conservancy project operation status feature matrix contains spatiotemporal correlation data of water level fluctuation features, flow velocity change features, and gate opening change features.

[0009] The water conservancy project operation status feature matrix is ​​input into the anomaly detection model to generate an abnormal operation status identifier for the water conservancy project. The abnormal operation status identifier includes anomaly type, anomaly level and anomaly location information.

[0010] A set of water conservancy project control instructions is generated based on the abnormal operation status identifier. The set of control instructions includes water level adjustment schemes, flow velocity control schemes, and gate opening adjustment schemes.

[0011] Preferably, the spatiotemporal feature extraction processing of the multi-source monitoring data includes:

[0012] Establish the spatiotemporal correlation matrix between the water level fluctuation sequence and the flow velocity change sequence, and calculate the transmission coefficient of the influence of water level change on flow velocity;

[0013] Construct response functions for the gate opening change sequence and the water level fluctuation sequence, and determine the lag time parameter of gate regulation to water level change;

[0014] Multi-scale features of water level fluctuation sequences are extracted using a spatiotemporal convolutional neural network to generate water level fluctuation feature vectors.

[0015] An attention mechanism is used to fuse local and global features of the flow velocity change sequence to generate a flow velocity change feature vector;

[0016] The water level fluctuation feature vector, flow velocity change feature vector, and gate opening change feature vector are spatiotemporally aligned to generate a water conservancy project operation status feature matrix.

[0017] Preferably, establishing the spatiotemporal correlation matrix between the water level fluctuation sequence and the flow velocity change sequence includes:

[0018] Divide the water level fluctuation sequence into time windows and calculate the rate of water level change within each time window;

[0019] Establish a regression model between the rate of change of water level and the rate of change of flow velocity within the corresponding time window, and determine the transmission coefficient of the influence of water level change on flow velocity;

[0020] Construct a water level-flow velocity influence weight matrix based on the spatial relationship of monitoring points, and calculate the influence weight of upstream monitoring points on downstream monitoring points;

[0021] The transmission coefficients are multiplied by the influence weight matrix to generate the spatiotemporal correlation matrix between the water level fluctuation sequence and the flow velocity change sequence.

[0022] Preferably, inputting the feature matrix of the water conservancy project's operational status into the anomaly detection model includes:

[0023] Construct a sample library containing historical anomaly cases and extract feature patterns from these anomaly cases;

[0024] A deep belief network is used to perform feature dimensionality reduction on the feature matrix of water conservancy project operation status to generate a low-dimensional feature representation.

[0025] The distance metric between the low-dimensional feature representation and the normal operation features is calculated using an anomaly detection algorithm;

[0026] Based on the comparison between the distance metric and the preset threshold, an abnormal operation status identifier for the water conservancy project is generated.

[0027] Preferably, the step of using a deep belief network to perform feature dimensionality reduction processing on the feature matrix of the water conservancy project's operational status includes:

[0028] Construct a multi-layer restricted Boltzmann machine network structure and set the number of hidden layer nodes and learning rate parameters;

[0029] The feature matrix of the water conservancy project's operational status is input into the first-layer restricted Boltzmann machine for feature encoding processing.

[0030] By optimizing network parameters through layer-by-layer training, a stable deep belief network model is generated.

[0031] The trained deep belief network model is used to perform a nonlinear transformation on the input features and output a low-dimensional feature representation.

[0032] Preferably, the step of generating a set of water conservancy project control instructions based on the abnormal operating status identifier includes:

[0033] Based on the anomaly type, a preset control strategy template is matched to obtain the basic control plan;

[0034] Based on the anomaly level, adjust the parameter intensity of the basic control scheme to generate an intensity-adaptive control scheme.

[0035] By combining abnormal location information, the implementation scope of the control plan is determined, and spatial positioning control instructions are generated;

[0036] The water level regulation scheme, flow velocity control scheme, and gate opening adjustment scheme are coordinated in a timely manner to generate a set of water conservancy project control instructions.

[0037] Preferably, the parameter intensity of the basic control scheme adjusted based on the anomaly level includes:

[0038] Establish a mapping table between anomaly levels and control intensity, and determine the intensity adjustment coefficient corresponding to each level;

[0039] Calculate the product of the basic control scheme parameter values ​​and the intensity adjustment coefficient to generate intensity-adapted control parameters;

[0040] The intensity adjustment coefficient is corrected based on historical control effect data to optimize the calculation accuracy of control parameters;

[0041] The output includes a control scheme that incorporates intensity adaptation parameters.

[0042] Preferably, the method further includes:

[0043] Collect engineering response data after the execution of control commands, including actual water level changes, actual flow velocity changes, and actual gate opening changes;

[0044] Calculate the deviation measure between the engineering response data and the expected control effect;

[0045] The detection threshold parameters of the anomaly detection model are updated based on the deviation metric.

[0046] Optimize the algorithm for generating the set of water conservancy project control instructions.

[0047] Preferably, the deviation metric between the calculated engineering response data and the expected control effect includes:

[0048] Establish a difference matrix between engineering response data and expected control data, and calculate the local deviation value of each monitoring point;

[0049] A deviation distribution map of the entire region is generated using spatiotemporal interpolation methods to identify deviation hotspot areas.

[0050] Calculate the overall deviation index and the maximum local deviation value to generate a comprehensive deviation metric.

[0051] The overall deviation metric is compared and analyzed with historical deviation data to evaluate the control effect.

[0052] Preferably, the present invention also includes an intelligent monitoring system for water conservancy projects, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described intelligent monitoring method for water conservancy projects.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] This intelligent monitoring method for water conservancy projects effectively solves the problem that traditional monitoring methods, relying on single-source data acquisition, cannot comprehensively reflect the operational status of the project. The multi-source monitoring data includes water level fluctuation sequences, flow velocity change sequences, and gate opening change sequences, enabling the collection of operational data from multiple dimensions and comprehensively capturing key information during the project's operation. This data covers the core parameters of water conservancy project operation, allowing monitoring of the project's operational status to move beyond single indicators and grasp the overall operational situation, providing a comprehensive and rich data foundation for subsequent data analysis and status assessment.

[0055] In the process of generating a feature matrix of water conservancy project operation status by extracting spatiotemporal features from multi-source monitoring data, the spatiotemporal correlation of the data was fully considered, overcoming the limitation of traditional data processing methods that ignore spatiotemporal correlation. Spatiotemporal feature extraction can deeply explore the spatiotemporal correlation data between water level fluctuation characteristics, flow velocity change characteristics, and gate opening change characteristics. It not only focuses on the changing trends of various data in the time dimension, but also analyzes the mutual influence of data from different monitoring points in the spatial dimension, thereby more accurately reflecting the inherent laws of water conservancy project operation status. Through this processing method, the originally scattered and independent multi-source data can be transformed into a feature matrix with inherent correlation, so that the engineering operation information contained in the data can be fully explored and integrated, providing more valuable analytical basis for subsequent anomaly detection.

[0056] Inputting the operational status feature matrix of a water conservancy project into an anomaly detection model to generate anomaly status identifiers significantly improves the accuracy and comprehensiveness of anomaly detection. The anomaly detection model can analyze integrated spatiotemporal correlation feature data, and compared to traditional fixed threshold judgment methods, it can more accurately identify abnormal situations in project operation. The generated anomaly status identifier includes anomaly type, anomaly level, and anomaly location information, enabling technicians to clearly and accurately understand the specific circumstances of the anomaly, rather than simply determining whether an anomaly exists without specifying detailed information, as is the case with traditional methods. A clearly defined anomaly type helps to quickly locate the cause of the anomaly, the anomaly level provides a basis for judging the severity of the anomaly, and the anomaly location information facilitates technicians' rapid on-site verification and handling, greatly improving the efficiency of anomaly handling.

[0057] This system generates a set of control instructions for water conservancy projects based on abnormal operating status identifiers, enabling rapid integration from anomaly detection to control scheme formulation. This solves the problems of low efficiency and insufficient scientific rigor in traditional methods of formulating control schemes after anomaly detection. The control instruction set includes water level regulation schemes, flow velocity control schemes, and gate opening adjustment schemes. These schemes are formulated based on accurate anomaly information and can provide targeted control measures for different types and levels of anomalies. Instead of relying on manual experience to formulate control schemes, it generates scientifically sound control instructions through systematic analysis and calculation, ensuring the scientific validity and effectiveness of the control schemes. Simultaneously, the rapid generation of control instructions enables timely responses to anomalies in project operation, preventing further deterioration due to untimely control, ensuring that water conservancy projects can quickly return to normal operation when anomalies occur, and maintaining the safety and stability of the project.

[0058] The entire intelligent monitoring method forms a complete process from data acquisition, data processing, anomaly detection, and control command generation, realizing the automation and intelligence of water conservancy project monitoring. Compared with traditional manual monitoring and simple data processing methods, this method significantly reduces human intervention and mitigates the errors and inefficiencies caused by manual operation. Automated data acquisition and processing enable real-time monitoring of water conservancy projects, avoiding the lag of manual inspections; intelligent anomaly detection and control command generation can quickly respond to various situations in project operation, improving the overall efficiency and level of water conservancy project operation and management, and ensuring that water conservancy projects can stably perform their functions of flood control, irrigation, and water supply in the long term. Attached Figure Description

[0059] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent monitoring method for water conservancy projects described in this invention.

[0060] Figure 2 This is a schematic diagram illustrating the working principle of spatiotemporal feature extraction and processing of multi-source monitoring data.

[0061] Figure 3 The diagram illustrates the working principle of the anomaly detection model that inputs the feature matrix of the water conservancy project's operational status. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Please see Figure 1This invention provides an intelligent monitoring system and method for water conservancy projects. The method integrates multi-source monitoring data and intelligent algorithms to achieve real-time monitoring and control of the project's operation. The method acquires multi-source monitoring data from monitoring points of the water conservancy project, including water level fluctuation sequences, flow velocity change sequences, and gate opening change sequences. These sequences are continuously collected by sensors deployed at key locations to form time-series data. Then, spatiotemporal feature extraction processing is performed on the multi-source monitoring data. This processing aims to capture the temporal dynamics and spatial correlations in the data, generating a water conservancy project operation status feature matrix. This matrix integrates the spatiotemporal correlation information of water level fluctuation characteristics, flow velocity change characteristics, and gate opening characteristics. To provide structured input for subsequent analysis, the characteristic matrix of the water conservancy project's operating status is then input into the anomaly detection model. This model identifies deviations by comparing with historical normal operating modes and outputs anomaly operation status identifiers for the water conservancy project. The identifiers include details such as anomaly type (e.g., water level or flow velocity anomaly), anomaly level (e.g., mild or severe), and anomaly location information (e.g., specific monitoring point number). Finally, based on the anomaly operation status identifiers, a set of water conservancy project control instructions is generated. This set includes water level regulation schemes (e.g., adjusting gate opening to control water level), flow velocity control schemes (e.g., adjusting flow rate through pumping stations), and gate opening adjustment schemes (e.g., precisely setting gate position). These schemes achieve automatic control of the project through control actuators.

[0064] Example 1: See Figure 2 The acquired water level fluctuation sequence, flow velocity change sequence, and gate opening change sequence are first sent to the preprocessing module for data cleaning and standardization to eliminate the influence of sensor errors and dimensional differences. The preprocessed data forms a regular time series dataset, preparing for subsequent deep feature extraction. Establishing the spatiotemporal correlation matrix between the water level fluctuation sequence and the flow velocity change sequence is the first step in feature extraction. This process captures the stage-specific characteristics of hydrological element changes by dividing reasonable time windows. For example, based on the regulation cycle of water conservancy projects, continuous monitoring data is divided into segments of several hours or days. Within each time window, the average rate of water level change is calculated. This rate reflects the overall trend of water level change during that period. Simultaneously, features related to… Based on the data on the rate of change of flow velocity within a given time period, a statistical regression model between the rate of change of water level and the rate of change of flow velocity is constructed. The least squares method is used to fit the mathematical relationship between the two, thereby determining the quantitative transmission coefficient of the impact of water level changes on flow velocity. This coefficient expresses the amount of flow velocity change caused by a unit change in water level. Furthermore, the impact of the spatial layout of monitoring points needs to be considered. Based on the relationship between water flow direction and distance, influence weights are assigned to monitoring points at different locations. Changes at upstream stations usually have a stronger impact on downstream stations. A water level-flow velocity influence weight matrix is ​​generated through a spatial interpolation algorithm. Finally, the calculated transmission coefficient and the spatial influence weight matrix are combined to generate a matrix structure that comprehensively expresses the spatiotemporal correlation characteristics of water level and flow velocity.

[0065] Constructing response functions for the gate opening change sequence and the water level fluctuation sequence requires analyzing the dynamic relationship between gate operation and water level response. Gate opening and closing directly cause changes in upstream and downstream water levels, but these changes exhibit a certain time lag. By analyzing the correspondence between gate opening adjustment events and subsequent water level changes in historical data, a dynamic response model is established using a system identification method. This model accurately describes the time delay parameter required from gate action to a significant change in water level. The establishment of the response function considers various factors such as gate type and flow conditions. The impact range and duration of gate regulation on water level vary significantly depending on the scale of the gate operation; therefore, appropriate models need to be determined based on specific engineering characteristics. The function form of the method extracts multi-scale features of water level fluctuation sequences using a spatiotemporal convolutional neural network, employing deep learning methods. This network structure contains multiple convolutional and pooling layers to capture fluctuation patterns at different time scales. Shallow convolutional kernels focus on identifying small fluctuations in water level in the short term, while deep networks can capture the long-term trends and periodic patterns of water level changes. The network scans the input sequence step by step using a sliding window to extract representative local features. During the generation of multi-scale features, the network automatically learns important features in different time dimensions, transforming the original water level sequence into an information-rich feature vector. This vector contains both the instantaneous change characteristics of the water level and retains its long-term evolution patterns.

[0066] The use of an attention mechanism to fuse local and global features of flow velocity change sequences aims to balance the importance of recent changes and long-term patterns. This mechanism adaptively weights features by calculating the weight of each time point's contribution to the current state. Local features reflect instantaneous fluctuations in flow velocity, which may foreshadow sudden events. Global features reflect the overall trend and periodicity of flow velocity changes. The attention mechanism dynamically adjusts the contribution of features at different time points by evaluating their importance. This weighted fusion method strengthens features at key time points while appropriately suppressing secondary features. The resulting flow velocity change feature vector focuses more on flow velocity change patterns that significantly impact the operational status of the project, including water level... Spatiotemporal alignment of the fluctuation feature vector, flow velocity change feature vector, and gate opening change feature is a key step in generating a unified feature matrix. Due to differences in the acquisition frequency and transmission delay of different monitoring data, time synchronization is necessary. A timestamp alignment method is used to unify various features to the same time reference. For missing data points, appropriate interpolation methods are used to fill in the gaps. Spatial alignment integrates discrete point features into a continuous spatial distribution based on the geographical location information of the monitoring points. The aligned feature data is organized into a regular three-dimensional tensor structure, in which the time dimension, spatial dimension, and feature dimension together constitute a complete feature matrix of the water conservancy project's operating status. This matrix provides high-quality input data for subsequent anomaly detection.

[0067] The implementation of the spatiotemporal feature extraction process relies on an efficient computing architecture and algorithm optimization. To address the large volume and high real-time requirements of water conservancy project monitoring data, a distributed computing framework is employed for parallel processing. The feature extraction module is designed as a configurable pipeline structure, allowing adjustments to the feature extraction strategy and parameters based on specific project needs. This flexibility enables the method to adapt to monitoring scenarios of different types and scales of water conservancy projects. The entire process emphasizes the interpretability and physical meaning of features. The extracted features are not only statistically significant but also consistent with the actual operating mechanism of water conservancy projects. Special attention is paid to handling common noise and outlier issues in monitoring data during feature extraction. Robust statistical methods and filtering techniques are used to eliminate data anomalies caused by occasional sensor failures or transmission interference. Simultaneously, the differences in feature performance patterns under different seasons and hydrological conditions are considered, and environmental factors are introduced as auxiliary information to improve the adaptability of feature expression. This ensures that the generated feature matrix accurately reflects the project's operating status under various conditions. The results of spatiotemporal feature extraction not only serve real-time monitoring tasks but can also be used for the analysis and summarization of historical operating patterns.

[0068] Example 2: See Figure 3 To construct a sample library containing a wealth of historical anomaly cases, the system needs to collect and organize various abnormal operating condition data accumulated during the long-term operation of water conservancy projects. This data originates from historical records of multiple monitoring parameters such as water level, flow velocity, and gate opening. The selected anomaly cases cover different types of operational anomalies, including abnormal rises or falls in water level, sudden changes in flow velocity, and abnormal gate operation. Each case has been carefully annotated and confirmed by domain experts to ensure the accuracy and representativeness of the sample data. During the construction of the sample library, feature engineering methods are used to extract meaningful feature patterns from the original monitoring data. These patterns can capture the typical characteristics and patterns of anomaly occurrences, providing a reliable reference benchmark for subsequent anomaly detection. A deep belief network is employed to analyze the operation of water conservancy projects. Dimensionality reduction of the row state feature matrix is ​​a crucial step in deep learning. Deep belief networks (DBNs) are composed of multiple stacked Restricted Boltzmann Machines (RBMs) and can automatically learn distributed representations of input data through unsupervised learning. The network structure design needs to consider the dimensionality and complexity of the input features. The number of nodes in the hidden layers is determined based on the size of the feature matrix. Typically, the optimal configuration is selected by comparing the performance of different structures through experiments. During training, the contrastive divergence algorithm is used to update the network parameters layer by layer. Each layer of the RBM learns different abstract representations of the input data, ultimately transforming the high-dimensional feature matrix into a low-dimensional but information-rich feature representation. This dimensionality reduction not only reduces computational complexity but also eliminates redundant information in the original features, improving the efficiency and accuracy of subsequent anomaly detection.

[0069] The distance metric between low-dimensional feature representations and normal operation features is calculated using anomaly detection algorithms. This is achieved through a variety of statistical and machine learning methods. Normal operation features are obtained by analyzing a large amount of historical normal operating data and are typically represented as feature vectors. The distance metric can be calculated using Euclidean distance, Mahalanobis distance, or other adaptive distance indicators. These indicators can quantify the degree of deviation between the current operating state and the normal mode. The distance value calculation considers the weights and correlations of different feature dimensions, with important features given higher weights, making the distance metric more reflective of actual anomalies. A time sliding window mechanism is also introduced in the calculation process, which not only considers the feature values ​​at the current moment but also incorporates the changing trends of recent historical data, thereby providing a more comprehensive assessment of the operating state.

[0070] Generating abnormal operating status indicators for water conservancy projects based on the comparison between distance metrics and preset thresholds requires establishing a scientific threshold setting mechanism. The choice of threshold directly affects the sensitivity and specificity of anomaly detection and needs to be optimized and adjusted based on historical data and actual project conditions. When the distance metric exceeds the preset threshold, the system determines that there is an anomaly in the current operating status and determines the severity of the anomaly based on the magnitude of the distance value. Anomaly levels are usually divided into multiple levels, from minor anomalies to severe anomalies. Anomaly type identification is achieved by analyzing the contribution of each dimension in the feature vector to find the main feature dimension that causes the distance increase, thereby determining whether the anomaly is related to water level, flow velocity, or gate opening. The determination of anomaly location information combines spatial information in the feature matrix. By analyzing the distribution of anomaly degree at different monitoring points, the specific area or equipment where the anomaly occurred can be accurately located.

[0071] The implementation of the anomaly detection model emphasizes real-time performance and adaptability. The model can process continuously input monitoring data online, promptly issue anomaly alerts, and provide decision support for engineering operators. It also possesses self-learning capabilities, continuously updating and optimizing detection parameters based on newly emerging anomaly cases to improve its ability to identify future anomalies. The detection process considers the seasonal and periodic characteristics of water conservancy project operation, employing appropriate detection strategies for different operating conditions to avoid misjudging normal conditions as anomalies. Anomaly detection results are output in a structured format, including detailed information such as anomaly type, level, and location, providing accurate and reliable input for subsequent control command generation. The entire anomaly detection process emphasizes reliability and robustness, employing various technical means to reduce the risk of false alarms and missed alarms, including setting multi-level detection thresholds and introducing voting mechanisms. The detection system also has a fault self-diagnosis function, capable of identifying external problems such as sensor malfunctions or data transmission interruptions, avoiding misjudging these technical issues as operational anomalies. The anomaly detection model is tightly integrated with the upstream feature extraction module and the downstream control command generation module, forming a complete monitoring closed loop to ensure the safe and stable operation of water conservancy projects.

[0072] Example 3: The construction of the network structure requires determining the appropriate network depth and width based on the dimension of the input feature matrix. The number of hidden layer nodes is usually set according to the principle of decreasing layer by layer. The number of nodes in each layer is determined after multiple experimental comparisons to ensure that the network has sufficient expressive power without overfitting. The learning rate parameter adopts an adaptive adjustment strategy, using a larger value in the early stage of training to accelerate convergence and gradually decreasing it in the later stage to improve the accuracy of parameter optimization. The introduction of the momentum parameter helps to smooth the training process and avoid getting trapped in local optima. When the feature matrix of the water conservancy project operation status is input into the first layer of the restricted Boltzmann machine for feature encoding, the machine realizes the nonlinear transformation of data through the bidirectional connection between the visible layer and the hidden layer. The visible layer nodes receive the input feature vector and calculate the activation probability of the hidden layer nodes through the weight matrix and bias parameters. The hidden layer nodes then reconstruct the output of the visible layer. The entire process learns the probability distribution of the input data by minimizing the energy function. The activation function used in the feature encoding process is usually the sigmoid function, which can convert continuous values ​​into probabilistic representations, facilitating subsequent sampling and reconstruction calculations.

[0073] The network parameters are optimized through a layer-by-layer training approach using a greedy layered pre-training strategy. Each layer of the Restricted Boltzmann Machine (RBM) undergoes independent unsupervised pre-training, learning the distribution characteristics of the input data and using its output as the input for the next layer. During training, the contrastive divergence algorithm is used to quickly approximate maximum likelihood estimation. Network parameters are updated through a finite number of Gibbs sampling steps, significantly improving training efficiency. The network parameter updates follow the gradient descent principle, and the adjustment amounts of the weight matrix and bias vector are jointly determined by the reconstruction error and the learning rate. After multiple iterations, the network gradually converges to a stable state. After generating a stable deep belief network model, overall fine-tuning training is still required, using the backpropagation algorithm for supervised fine-tuning. By fine-tuning all layer parameters, the quality of feature representation is further improved. When the trained deep belief network model is used to perform nonlinear transformation on the input features, the original high-dimensional feature vector is passed through multiple layers of the network. Each layer performs linear transformation and nonlinear activation, gradually extracting more abstract and essential feature representations. The final output low-dimensional feature representation retains the most discriminative information in the original data while eliminating redundancy and noise. Its feature dimension is usually reduced to one-tenth to one-fifth of the original dimension. Low-dimensional feature representation not only reduces the cost of subsequent computation but also improves the accuracy of anomaly detection because removing irrelevant feature dimensions reduces the complexity of the data distribution.

[0074] The training quality of deep belief networks directly affects the effect of feature dimensionality reduction. Therefore, multiple measures are needed to ensure the stability of the training process. Preprocessing of training data includes standardization and denoising to ensure that the input data meets the requirements of the network. During training, the changes in reconstruction error and likelihood function value are monitored, and hyperparameters are adjusted in a timely manner to prevent overfitting. Regularization techniques such as Dropout are used to improve the generalization ability of the model. The choice of network depth needs to balance representation power and training difficulty. Overly deep networks may lead to gradient vanishing problems, while overly shallow networks may not be able to fully extract the hierarchical features of the data.

[0075] The energy function in this deep belief network model is in the form of:

[0076]

[0077] in: This represents the state vector of the visible layer unit. It is the first visible layer Units, This represents the state vector of a hidden layer unit. It is the first hidden layer Units, Visible layer unit The bias parameters, It is a hidden layer unit The bias parameters, Visible layer unit With hidden layer unit Connection weights between them This indicates the total number of visible layer cells. This indicates the total number of hidden layer units.

[0078] The feature dimensionality reduction process takes into account the special characteristics of water conservancy project data. Monitoring data has obvious temporal and spatial correlations. Therefore, the network design introduces corresponding mechanisms to preserve these characteristics. In the temporal dimension, recursive connections are used to capture long-term dependencies, and in the spatial dimension, convolutional operations are used to preserve local feature structures. This ensures that the dimensionality-reduced features still contain rich spatiotemporal information. The dimensionality reduction results not only serve the current anomaly detection task but can also be used for other data analysis purposes, such as operational status clustering and trend prediction, providing multifaceted support for the intelligent management of water conservancy projects. The entire feature dimensionality reduction module adopts a modular design, allowing for flexible adjustment of the network structure and parameters according to different engineering needs. It also provides interfaces for real-time processing and historical data analysis. By continuously collecting new operational data and periodically retraining the network, the feature dimensionality reduction model can adapt to changes in engineering operating conditions, maintaining long-term effectiveness and accuracy. The dimensionality-reduced feature representation maintains an interpretable correspondence with the original features, facilitating engineers to understand the analysis results and make correct decisions.

[0079] Example 4: Suppose that the monitoring system of a water conservancy project detects an abnormal rise in water level at monitoring point S12. The anomaly type is determined to be water level exceeding the limit, the anomaly level is level two (moderate severity), and the anomaly location is accurate to monitoring point S12 and its downstream area of ​​500 meters. The system first matches a preset control strategy template according to the anomaly type and retrieves the basic control scheme corresponding to the water level exceeding the limit type from the knowledge base. This scheme includes basic operating parameters in three dimensions: water level regulation, flow velocity control, and gate opening adjustment. The matching process of the control strategy template adopts a rule-based retrieval mechanism. The water level anomaly type corresponds to the control logic. In the basic scheme, the gate opening adjustment value is to increase the opening degree by 5%, the flow velocity control target is to decrease by 0.2 m / s, and the water level regulation target is to decrease by 15 cm. These basic parameters are derived from the historical operation experience and expert knowledge of the project, forming a standardized response measure library. The retrieval process also considers the current overall operation status of the project, including background factors such as upstream water inflow and downstream water demand, to ensure that the extracted basic scheme is consistent with the actual situation.

[0080] The parameter intensity of the basic control scheme based on anomaly level adjustment needs to refer to the preset level-intensity mapping relationship. This project established a five-level anomaly level system, with an intensity adjustment coefficient corresponding to each level. The intensity adjustment coefficient for level two anomalies is 1.2, indicating a 20% increase in control intensity based on the basic scheme parameter values. During the calculation process, the basic gate opening adjustment value of 5% became 6% after intensity adaptation, the flow velocity control target was adjusted from 0.2m / s to 0.24m / s, and the water level adjustment target was adjusted from 15cm to 18cm. The determination of the intensity adjustment coefficient takes into account the nonlinear relationship between the severity of the anomaly level and the control effect. Higher-level anomalies require a larger proportion of intensity increase. The mapping relationship between anomaly level and control intensity is continuously corrected based on historical control effect data. The system records the actual effect of each control action, analyzes the correlation between control intensity and anomaly elimination speed, and optimizes the coefficient values ​​in the mapping relationship table through machine learning algorithms to make the control intensity setting more accurate and effective. The intensity adjustment calculation adopts a piecewise linear interpolation method to maintain a smooth transition of the adjustment coefficient within different level intervals and avoid drastic jumps in control parameters.

[0081] Determining the scope of the control plan based on anomaly location information requires analyzing the spatial structure characteristics of the water conservancy project. Monitoring point S12 is located in the middle section of the main channel, with three diversion gates and two drainage outlets within 500 meters downstream. The system calculates the impact range of the anomaly based on the water flow propagation model, determining that the gates requiring adjustment include the main control gate upstream of point S12 and the two diversion gates downstream. The spatial positioning process utilizes the project's geographic information system data to construct the correspondence between control operations and spatial locations, generating targeted control instructions. The spatial positioning of the control instructions considers the temporal and spatial characteristics of water flow. Adjustments to the upstream gates will affect downstream points after a certain time delay; therefore, the instructions include an operational time sequence. For the water level anomaly at point S12, the main control gate 200 meters upstream is first adjusted, increasing its opening by 6%; 10 minutes later, the first downstream diversion gate is adjusted, reducing the water diversion by 15%; 20 minutes later, the second diversion gate is adjusted, correspondingly adjusting the water diversion ratio. This spatiotemporally coordinated control plan ensures the stable recovery of the water level in the entire affected area.

[0082] Coordinating the timing of water level regulation, flow velocity control, and gate opening adjustment schemes requires establishing a multi-objective optimization model. This model must consider the mutual influence and temporal dependencies among the various control measures. Water level regulation is achieved through gate operation; changes in gate opening affect flow velocity, which in turn affects water level fluctuations. This coupling relationship needs to be properly handled in the timing arrangement. The coordination algorithm aims to ensure overall stability, avoid conflicts between control measures, and ensure a smooth transition of the project to normal operation. The timing coordination process employs model predictive control, predicting the implementation effects of different control schemes based on the hydraulic model of the water conservancy project. The optimal operation time sequence is selected. For detected water level anomalies, the system generates a set of instructions containing specific operation times, operation objects, and operation parameters. Reasonable time intervals are maintained between instructions, allowing for subsequent adjustments after each operation produces an effect. The entire set of instructions constitutes a complete control process, guiding engineers or automatic control systems to execute corresponding operations.

[0083] Table 1: Adaptation Table for Abnormal Control Parameters

[0084]

[0085] The generation of the control command set also considers the constraints of engineering operation, including physical constraints such as gate mechanical limitations, maximum allowable flow velocity, and minimum operating water level. The system checks whether the generated control parameters are within the allowable range and automatically corrects commands that exceed the limits to ensure the safety and feasibility of the control scheme. The output format of the command set adopts a standardized structure, including fields such as operating equipment number, operation type, operation parameters, and execution time, which facilitates the parsing and execution by the automatic control system. The entire control command generation process emphasizes practicality and reliability. The generated commands are not only based on theoretical calculations but also incorporate practical experience in engineering operation. The system provides a manual review and modification interface, allowing experienced operators to fine-tune the automatically generated commands, giving full play to the advantages of human-machine collaboration. Through continuous accumulation of control cases and effect data, the command generation algorithm is continuously optimized to improve the ability to cope with complex abnormal situations.

[0086] Example 5: Assume that after a water level anomaly control action, the system begins collecting actual operational data following the execution of the control command. This data is acquired in real time through a sensor network deployed at key locations throughout the project, including monitoring devices such as water level gauges, flow meters, and gate opening sensors. The data acquisition frequency is synchronized with the monitoring system, typically set to once per minute. For critical locations, the data acquisition interval may be shortened to the second level to ensure that the dynamic response process generated by the control measures can be captured. The collected data includes actual water level changes, actual flow velocity changes, and actual gate opening changes. These data, along with the expected target values ​​in the control command, form a correlation. In accordance with the relationship, it is necessary to establish a scientific and reasonable comparison method to measure the deviation between the engineering response data and the expected control effect. The expected control effect comes from the theoretical prediction value of the hydraulic calculation model, including the expected water level change, the expected flow velocity change, and the expected gate opening change. The deviation calculation is first carried out independently at each monitoring point. For each monitoring parameter, the absolute difference and relative difference between the actual value and the expected value are calculated to form a preliminary local deviation assessment. Taking the water level parameter as an example, if the expected water level drop is 15 cm at a certain monitoring point and the actual drop is 13 cm, the absolute deviation is 2 cm and the relative deviation is about 13.3%. Similar calculation methods are applied to parameters such as flow velocity and gate opening.

[0087] The calculation of local deviation values ​​takes into account the effects of measurement errors and random fluctuations. A moving average filtering technique is used to preprocess the raw data to eliminate interference from short-term fluctuations. Establishing a difference matrix between the engineering response data and the expected control data requires consideration of the spatiotemporal dimension. The rows of the difference matrix represent the locations of different monitoring points, and the columns represent continuous time points. Matrix elements record the deviation values ​​of each point at each time point. This matrix is ​​constructed based on a geographic information system platform, combining the deviation data of discrete monitoring points with the spatial layout of the engineering project to form a spatially continuous deviation distribution map. The blank areas between monitoring points are processed using spatiotemporal interpolation methods. The Kriging interpolation algorithm is used to estimate the deviation of unmonitored areas based on the deviation values ​​of adjacent points, generating a deviation distribution map covering the entire project. The deviation distribution map visually displays the spatial variation characteristics of the control effect in the form of a heat map, with color intensity indicating the magnitude of the deviation, facilitating the identification of hotspot areas with larger deviations.

[0088] The overall deviation index and the maximum local deviation value are calculated using a multi-indicator comprehensive evaluation method. The overall deviation index is obtained by weighted averaging of the deviation values ​​at all time points across all monitoring points. The weighting coefficients are determined based on the importance of the monitoring points and the reliability of the data. The maximum local deviation value is selected from all monitoring points based on the largest absolute deviation value, reflecting the least ideal situation in terms of control effect. The overall deviation index characterizes the overall implementation effect of the control measures; the smaller the value, the closer the actual effect is to the expected target. The maximum local deviation value reveals the uniformity of the control effect; an excessively large value indicates a significant imbalance in the engineering response. The combination of these two indicators can comprehensively evaluate the quality of the control actions, providing a basis for subsequent... Continuing optimization provides directional guidance. Comparing and analyzing the comprehensive deviation metric with historical deviation data requires querying historical databases to retrieve historical deviation records under similar control scenarios, including background information such as control type, engineering conditions, and hydrological conditions. The comparative analysis focuses on examining the relative level of the current control effect in the historical sequence to determine whether the expected improvement target has been achieved. Historical data provides a benchmark reference to help distinguish whether the current deviation is caused by systematic or accidental factors. If the deviation value is consistently higher than the historical average, it indicates that there are problems that need to be improved in the control strategy or execution process. The analysis results are displayed in the form of trend charts, which intuitively present the long-term trend of the control effect.

[0089] The anomaly detection model's detection threshold parameters, updated based on deviation metrics, employ an adaptive adjustment mechanism. These threshold parameters include key parameters such as the distance threshold for anomaly detection and the confidence level. When the actual adjustment effect deviates little from the expected target, it indicates that the current anomaly detection model has good accuracy, and the detection threshold can be appropriately lowered to improve sensitivity. Conversely, when the deviation is large, the threshold may need to be increased to avoid false alarms caused by oversensitivity. Threshold adjustments follow a gradual principle, with each adjustment controlled within a reasonable range to avoid drastic threshold fluctuations due to a single anomaly. The adjusted threshold parameters need to be verified through simulation testing to ensure that model performance improves rather than deteriorates under the new threshold settings. The algorithm for generating the control command set of the water conservancy project focuses on improving the parameter calculation logic and timing arrangement strategy. The algorithm optimization is based on the results of deviation analysis to identify weak links. If it is found that the actual water level change after gate adjustment is generally less than the expected value, the correlation coefficient between gate opening and water level change is adjusted, and the expected water level adjustment amount corresponding to the unit opening change is appropriately increased. If the flow velocity control effect shows obvious non-uniformity in space, the spatial allocation algorithm of the flow velocity control scheme is optimized to strengthen the coordination between different parts. The timing arrangement optimization adjusts the execution timing and interval of the control commands, and resets the time delay between each command according to the actual response speed, so that the multi-step control action can form a better synergistic effect.

[0090] The optimization process employs a data-driven approach, establishing a mapping relationship between control parameters and implementation effects through training on a large number of historical cases. This continuously refines model parameters and improves prediction accuracy. The optimization algorithm incorporates machine learning techniques, enabling the system to learn autonomously from both successful and unsuccessful cases, gradually improving the quality of generated control commands. After each optimization update, the system records performance comparison data between the old and new algorithms, establishing a version management mechanism to ensure the traceability and reversibility of the optimization process. Through continuous iterative optimization, the control command generation algorithm gradually adapts to changes in engineering characteristics, maintaining long-term effectiveness and reliability. The system periodically generates optimization effect evaluation reports, summarizing the effects and experiences of recent improvements to provide a reference for subsequent optimizations. The optimization process emphasizes robustness; any adjustment to algorithm parameters is thoroughly verified to avoid system performance fluctuations due to over-optimization. Through this continuous improvement mechanism, the water conservancy project monitoring system can continuously adapt to changes in operating conditions, improve the accuracy and efficiency of anomaly handling, and provide reliable guarantees for the safe and stable operation of the project.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art 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 appended claims and their equivalents.

Claims

1. A method for intelligent monitoring of hydraulic structures, characterized in that, The method comprises the following steps: obtaining multi-source monitoring data of a water conservancy monitoring point, the multi-source monitoring data comprising a water level fluctuation sequence, a flow rate change sequence and a gate opening change sequence; performing spatio-temporal feature extraction processing on the multi-source monitoring data to generate a water conservancy operation state feature matrix, the water conservancy operation state feature matrix containing spatio-temporal correlation data of water level fluctuation features, flow rate change features and gate opening change features; inputting the water conservancy operation state feature matrix into an anomaly detection model to generate a water conservancy abnormal operation state identifier, the abnormal operation state identifier containing abnormal type, abnormal level and abnormal position information; generating a water conservancy control instruction set based on the abnormal operation state identifier, the control instruction set containing a water level adjustment scheme, a flow rate control scheme and a gate opening adjustment scheme; the spatio-temporal feature extraction processing on the multi-source monitoring data comprises: establishing a spatio-temporal correlation matrix of the water level fluctuation sequence and the flow rate change sequence, and calculating a transfer coefficient of the influence of water level change on flow rate; constructing a response function of the gate opening change sequence and the water level fluctuation sequence, and determining a lag time parameter of the influence of gate adjustment on water level change; extracting multi-scale features of the water level fluctuation sequence through a spatio-temporal convolutional neural network to generate a water level fluctuation feature vector; fusing local features and global features of the flow rate change sequence by using an attention mechanism to generate a flow rate change feature vector; performing spatio-temporal alignment processing on the water level fluctuation feature vector, the flow rate change feature vector and the gate opening change feature to generate a water conservancy operation state feature matrix; the generation of the water conservancy control instruction set based on the abnormal operation state identifier comprises: matching a preset control strategy template according to the abnormal type to obtain a basic control scheme; adjusting the parameter intensity of the basic control scheme based on the abnormal level to generate an intensity-adapted control scheme; determining the implementation range of the control scheme in combination with the abnormal position information to generate a spatially positioned control instruction; performing time sequence coordination processing on the water level adjustment scheme, the flow rate control scheme and the gate opening adjustment scheme to generate the water conservancy control instruction set.

2. The method for intelligent monitoring of hydraulic engineering according to claim 1, characterized in that, the establishment of the spatio-temporal correlation matrix of the water level fluctuation sequence and the flow rate change sequence comprises: dividing the time window of the water level fluctuation sequence, and calculating the water level change rate in each time window; establishing a regression model of the water level change rate and the flow rate change rate in the corresponding time window to determine the transfer coefficient of the influence of water level change on flow rate; constructing a water level-flow rate influence weight matrix according to the spatial position relationship of the monitoring points to calculate the influence weight of the upstream monitoring points on the downstream monitoring points; performing product operation on the transfer coefficient and the influence weight matrix to generate the spatio-temporal correlation matrix of the water level fluctuation sequence and the flow rate change sequence.

3. The method for intelligent monitoring of hydraulic engineering according to claim 1, characterized in that, the inputting of the water conservancy operation state feature matrix into the anomaly detection model comprises: constructing a sample library containing historical abnormal cases, and extracting feature patterns of the abnormal cases; performing feature dimension reduction processing on the water conservancy operation state feature matrix by using a deep belief network to generate a low-dimensional feature representation; calculating the distance measurement value of the low-dimensional feature representation and the normal operation feature by using an anomaly detection algorithm; Based on the comparison result of the distance metric value and the preset threshold value, an abnormal operation state identifier of the water conservancy project is generated.

4. The method for intelligent monitoring of hydraulic engineering according to claim 3, characterized in that, The feature dimension reduction processing of the water conservancy project operation state feature matrix by using the deep belief network comprises: A multi-layer restricted Boltzmann machine network structure is constructed, and the number of hidden layer nodes and the learning rate parameter are set; The water conservancy project operation state feature matrix is input into the first layer of the restricted Boltzmann machine for feature coding processing; The network parameters are optimized by layer-by-layer training method to generate a stable deep belief network model; The input features are nonlinearly transformed by using the trained deep belief network model to output low-dimensional feature representation.

5. The method for intelligent monitoring of hydraulic engineering according to claim 4, characterized in that, The parameter intensity of the basic regulation scheme is adjusted based on the abnormal level, comprising: A mapping relationship table of abnormal level and regulation intensity is established to determine the intensity adjustment coefficient corresponding to each level; The product of the basic regulation scheme parameter value and the intensity adjustment coefficient is calculated to generate the intensity-adapted regulation parameter; The intensity adjustment coefficient is corrected according to the historical regulation effect data to optimize the calculation accuracy of the regulation parameter; The regulation scheme containing the intensity-adapted parameter is output.

6. The method for intelligent monitoring of hydraulic engineering according to claim 1, characterized in that, It also includes: Collecting project response data after executing the regulation instruction, the project response data including actual water level change, actual flow rate change and actual gate opening change; Calculating the deviation metric value of the project response data and the expected regulation effect; Updating the detection threshold parameter of the abnormal detection model based on the deviation metric value; Optimizing the generation algorithm of the water conservancy project regulation instruction set.

7. The method for intelligent monitoring of hydraulic engineering according to claim 6, characterized in that, The calculation of the deviation metric value of the project response data and the expected regulation effect comprises: A difference matrix of the project response data and the expected regulation data is established to calculate the local deviation value of each monitoring point; The deviation distribution map of the whole region is generated by the space-time interpolation method to determine the deviation hot spot area; The overall deviation index and the maximum local deviation value are calculated to generate the comprehensive deviation metric value; The comprehensive deviation metric value is compared and analyzed with the historical deviation data to evaluate the regulation effect.

8. A hydraulic engineering intelligent monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, realizes the steps of the water conservancy project intelligent monitoring method according to any one of claims 1 to 7.

Images