Lightweight construction method and device for dam safety monitoring system

Abstract

The invention provides a dam safety monitoring system lightweight construction method and device, and the method comprises the steps: deploying a protocol conversion gateway, carrying out the protocol analysis and data standardization processing of an existing heterogeneous sensor, generating data in a unified format, and supporting the network disconnection caching and automatic supplementary transmission; deploying a data processing module on a local server or an industrial personal computer based on an edge computing architecture, performing feature extraction and multi-scale fusion analysis on acquired data, and performing data compression by adopting an improved revolving door algorithm; constructing a lightweight anomaly detection model, training by using historical data to generate a normal mode template library, calculating a dynamic time warping distance between new data and a template in real time, and triggering early warning when the distance of a plurality of continuous cycles exceeds a preset threshold value; the sampling frequency of the sensor is dynamically adjusted according to the real-time data analysis result, key parameters are collected in a high-frequency mode, non-key parameters are collected in a low-frequency or intermittent mode according to the change rate prediction result, and therefore the system resource use efficiency is optimized.

Description

Technical Field

[0001] This invention relates to the field of integrated dam safety monitoring and control, and in particular to a lightweight construction method and device for a dam safety monitoring system. Background Technology

[0002] Dam safety monitoring, as a crucial component of smart hydropower system construction, is widely applied in various hydropower projects, playing a key role, especially in small and medium-sized hydropower stations. With the continuous tightening of national policies on the safety supervision of water conservancy facilities, dam monitoring systems are gradually evolving towards intelligence and integration. Among related technologies, a complete monitoring system covering data perception, transmission, processing, analysis, and application has been constructed through the collaborative operation of sensor networks, data acquisition terminals, and cloud analysis platforms. Specifically, this system encompasses the entire process from physical parameter acquisition to risk warning, including key aspects such as hardware deployment, protocol compatibility, edge computing, data compression, AI modeling, and visualization, aiming to achieve real-time control of the dam's structural status and intelligent decision support.

[0003] However, existing dam monitoring methods, which directly employ dedicated sensors, independent data acquisition terminals, and centralized cloud analysis architectures, fail to adequately consider the resource constraints of small hydropower stations. This can lead to high hardware procurement costs, heavy maintenance burdens, and a shortage of professional personnel. Initial hardware investments can easily reach hundreds of thousands of yuan, with annual maintenance costs exceeding 30% of the power station's profits. Furthermore, most power stations lack dedicated maintenance teams, resulting in equipment failure response times exceeding 72 hours, severely impacting the integrity of monitoring data and system stability. In addition, traditional systems have high requirements for equipment protocol consistency, making them incompatible with various heterogeneous sensors and limiting the reuse of existing equipment. This further exacerbates the high-risk, low-efficiency contradiction in safety monitoring for small hydropower stations. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this invention is to propose a lightweight construction method for dam safety monitoring systems.

[0006] The second objective of this invention is to provide a lightweight construction device for a dam safety monitoring system.

[0007] To achieve the above objectives, the first aspect of this invention proposes a lightweight construction method for a dam safety monitoring system, comprising: S1, deploying a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generating data in a unified format and supporting offline caching and automatic retransmission; S2, deploying a data processing module on a local server or industrial control computer based on an edge computing architecture to perform feature extraction and multi-scale fusion analysis on the collected data, and using an improved rotating door algorithm for data compression to reduce data transmission and storage costs; S3, constructing a lightweight anomaly detection model, using historical data to train and generate a normal mode template library, calculating the dynamic time warping distance between new data and templates in real time, and triggering an early warning when the distance exceeds a preset threshold for multiple consecutive cycles; S4, dynamically adjusting the sampling frequency of the sensors based on real-time data analysis results, using high-frequency acquisition for key parameters, and low-frequency or intermittent acquisition for non-key parameters based on the rate of change prediction results to optimize system resource utilization efficiency.

[0008] In one embodiment of the present invention, the deployment protocol conversion gateway performs protocol parsing and data standardization processing on existing heterogeneous sensors, generates data in a unified format, and supports offline caching and automatic retransmission. It also includes: S11, dynamically loading DLL / SO libraries through an embedded device to adapt to private protocols without the need for cooperation from sensor manufacturers; S12, converting the raw data into JSON format containing timestamps, device IDs, values, and quality codes, and caching the data locally for 72 hours when the network is offline, and automatically retransmitting it after the network is restored.

[0009] In one embodiment of the present invention, the data processing module deployed on a local server or industrial control computer based on the edge computing architecture performs feature extraction and multi-scale fusion analysis on the collected data, and uses an improved rotating door algorithm for data compression to reduce data transmission and storage costs. The method further includes: S21, constructing a multi-scale feature vector, including time-domain features, frequency-domain features, and correlation features, to improve data analysis accuracy; S22, combining the improved rotating door algorithm with LZW lossless compression technology to achieve a data compression ratio of 10:1, significantly reducing transmission bandwidth requirements.

[0010] In one embodiment of the present invention, the step of constructing a lightweight anomaly detection model, using historical data to train and generate a normal pattern template library, calculating the dynamic time warping distance between new data and templates in real time, and triggering an early warning when the distance exceeds a preset threshold for multiple consecutive periods, further includes: S31, using an improved isolated forest algorithm and setting a lightweight configuration of n_estimators=50 to reduce computational resource consumption; S32, calculating the DTW dynamic time warping distance between new data and templates in real time, and triggering an early warning if the distance exceeds the threshold for three consecutive periods.

[0011] In one embodiment of the present invention, the step of dynamically adjusting the sampling frequency of the sensor based on real-time data analysis results, using high-frequency acquisition for key parameters and low-frequency or intermittent acquisition for non-key parameters based on the prediction results of the rate of change, in order to optimize the system resource utilization efficiency, further includes: S41, predicting the parameter change trend based on Kalman filtering, and if the parameter change rate Δ is greater than a preset threshold TH1, increasing the sampling frequency to 1Hz; S42, if the parameter change rate Δ is less than a preset threshold TH2, reducing the sampling frequency to 0.1Hz, otherwise maintaining the default frequency of 0.5Hz.

[0012] In one embodiment of the present invention, it further includes: S5, realizing mobile terminal synchronization through WeChat or DingTalk robot, pushing dam health briefings daily, providing immediate alarms for threshold exceeding the limit, and supporting remote confirmation and handling process.

[0013] To achieve the above objectives, a second aspect of the present invention proposes a lightweight construction device for a dam safety monitoring system, comprising: a protocol adaptation and caching module, used to deploy a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generate data in a unified format, and support offline caching and automatic retransmission; an edge feature fusion and compression module, used to deploy a data processing module on a local server or industrial control computer based on an edge computing architecture, to perform feature extraction and multi-scale fusion analysis on the collected data, and to use an improved rotating door algorithm for data compression to reduce data transmission and storage costs; an anomaly detection and early warning module, used to construct a lightweight anomaly detection model, use historical data to train and generate a normal mode template library, calculate the dynamic time warp distance between new data and templates in real time, and trigger an early warning when the distance exceeds a preset threshold for multiple consecutive cycles; and a sampling frequency dynamic adjustment module, used to dynamically adjust the sampling frequency of the sensors according to the real-time data analysis results, with key parameters being collected at high frequency and non-key parameters being collected at low frequency or intermittently based on the rate of change prediction results, in order to optimize the efficiency of system resource utilization.

[0014] The methods and apparatus of this invention significantly reduce the total lifecycle cost of small hydropower station dam safety monitoring systems, improve the utilization rate of existing equipment and operation and maintenance response efficiency, and effectively ensure the integrity and security of monitoring data.

[0015] 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

[0016] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0017] Figure 1This is a flowchart of a lightweight construction method for a dam safety monitoring system according to an embodiment of the present invention;

[0018] Figure 2 This is an architecture diagram of a lightweight construction method for a dam safety monitoring system according to an embodiment of the present invention;

[0019] Figure 3 This is a structural diagram of a lightweight construction device for a dam safety monitoring system according to an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0022] The following describes, with reference to the accompanying drawings, a lightweight construction method and apparatus for a dam safety monitoring system according to an embodiment of the present invention.

[0023] Example 1

[0024] Figure 1 This is a flowchart of a lightweight construction method for a dam safety monitoring system according to an embodiment of the present invention, such as... Figure 1 As shown, it includes:

[0025] S1 deploys a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generating data in a unified format and supporting offline caching and automatic retransmission.

[0026] Specifically, this step involves deploying a protocol conversion gateway to achieve protocol parsing and data standardization processing for heterogeneous sensors, and to support offline caching and automatic retransmission functions. It serves as a crucial bridge between the sensing layer and the data layer in this invention. In some implementations, the protocol conversion gateway uses an embedded hardware platform (such as a Raspberry Pi or an industrial-grade ARM development board) in conjunction with industrial communication modules (such as RS485, CAN, Ethernet, etc.), achieving multi-protocol compatibility through software configuration. The gateway has a built-in protocol parsing engine that supports mainstream industrial protocols such as Modbus RTU / TCP, CANopen, and DL / T645, and can extend support for proprietary protocols by loading dynamic link libraries (DLLs / SOs), thereby achieving a protocol-independent architecture that does not rely on sensor manufacturer cooperation.

[0027] Regarding data standardization, the gateway converts raw sensor data into a unified JSON format, including fields such as timestamp (ISO 8601 standard), device ID (16-bit MAC address or custom encoding), numerical value (floating-point or integer), and quality code (QoS, e.g., 0x00 for normal, 0x01 for abnormal), ensuring the data structure conforms to the IEC 61131-3 standard. In the event of a network outage, the gateway's built-in cache queue can locally store monitoring data for up to 72 hours, typically on eMMC or SD cards, supporting FAT32 or ext4 file systems. Once the network is restored, the gateway automatically retransmits data via MQTT or HTTP protocols, ensuring data integrity and continuity.

[0028] This step is applicable to small hydropower station dam monitoring systems, especially in remote areas with unstable communication or limited network bandwidth. Through protocol conversion and data standardization, the system is compatible with over 85% of existing sensors, significantly reducing hardware replacement costs. Simultaneously, the offline caching mechanism effectively handles intermittent communication interruptions, ensuring no data loss and improving system reliability. This technology not only solves the problem of heterogeneous device access but also provides structured, high-quality data input for subsequent edge computing and AI analysis, forming a crucial foundation for integrated dam safety monitoring.

[0029] Furthermore, S1 includes:

[0030] The S11 dynamically loads DLL / SO libraries through embedded devices to adapt to proprietary protocols, without requiring cooperation from sensor manufacturers.

[0031] Specifically, the step of "dynamically loading DLL / SO libraries through embedded devices to adapt to proprietary protocols without the need for sensor manufacturer cooperation" in this invention is a key technical means to achieve heterogeneous sensor protocol compatibility. In some implementations, the embedded device (such as a Raspberry Pi based on ARM architecture or an industrial-grade embedded module) runs customized protocol conversion gateway software. This software supports dynamically loading dynamic link libraries (DLLs, Windows platform) or shared object libraries (SOs, Linux platform) at runtime, thereby achieving flexible adaptation to different sensor proprietary protocols. This mechanism is based on a modular design concept, encapsulating the communication protocols of various sensors into independent plug-in libraries and calling them through a unified interface, avoiding the limitations of relying on manufacturer-provided SDKs or firmware upgrades in traditional methods.

[0032] At the parameter level, this step supports multiple industrial communication protocols, including but not limited to Modbus RTU / TCP, CANopen, and DL / T 645. It defines key parameters such as frame structure, verification methods, and data parsing rules for private protocols through configuration files (e.g., YAML or JSON format). During dynamic loading, the system initializes and registers protocol plugins through reflection mechanisms or dynamic link library interfaces (e.g., dlopen / dlsym on Linux, LoadLibrary / GetProcAddress on Windows), ensuring that communication parameters (such as baud rate, data bits, stop bits, and verification methods) are configurable and highly compatible. Furthermore, the system supports automatic protocol version identification and rollback, ensuring communication stability even in the event of protocol changes or device malfunctions.

[0033] In application scenarios, this technology is particularly suitable for the reality of small hydropower stations where there are diverse sensor models and inconsistent protocols. Through locally deployed embedded devices, new and old sensors can be quickly connected, achieving seamless data acquisition and transmission without requiring dedicated drivers or interface support from manufacturers, significantly reducing the difficulty and cost of system integration.

[0034] The technical effect of this step is that it constructs a protocol-independent communication architecture, improves the system's compatibility and scalability, lays the foundation for subsequent data standardization processing and edge computing analysis, and is one of the core supporting technologies for achieving low-cost and high-efficiency operation of the dam safety monitoring system.

[0035] S12 converts the raw data into JSON format containing timestamps, device IDs, values, and quality codes, and caches the data locally for 72 hours when the network is down, automatically re-uploading it after the network is restored.

[0036] Specifically, this step involves converting the raw monitoring data into a standardized JSON format and implementing local caching in the event of a network outage and automatic retransmission after network recovery. The technology is based on an edge computing architecture and data protocol adaptation mechanism, aiming to improve data processing compatibility and system reliability.

[0037] In some implementations, raw data often comes from various heterogeneous sensors, such as level gauges, piezometers, and displacement sensors, whose communication protocols may include Modbus RTU / TCP, CANopen, DL / T645, etc. The system deploys an embedded protocol conversion gateway (such as an industrial communication module integrated on a Raspberry Pi platform) to perform protocol parsing and format conversion of the raw data. Specifically, the gateway loads the parsing logic for different protocols through a configuration file, encapsulating the collected raw data into a JSON object containing a timestamp (ISO 8601 format), device ID (16-bit unique identifier), numerical value (floating-point, retaining 4 decimal places), and quality code (QoS, using a 4-digit hexadecimal code defined by the IEEE 1451 standard), ensuring the uniformity and parsability of the data structure.

[0038] In terms of parameters, the timestamp accuracy requirement is millisecond-level (±1ms), the device ID must conform to the device identification specifications in the IEC 62443 standard, and the quality code is used to identify data validity (e.g., 0x0000 indicates normal operation, 0x1000 indicates sensor failure). In the event of a network outage, the system employs a local caching mechanism based on SQLite, with a maximum cache duration of 72 hours. The cache capacity is dynamically adjusted based on the data acquisition frequency (0.1Hz-1Hz) and the number of sensors to ensure no data loss under extreme conditions.

[0039] In terms of application scenarios, this step is suitable for small hydropower station dam safety monitoring systems, especially in remote areas or environments with unstable communication, ensuring the continuity of data acquisition and transmission. After the network is restored, the system identifies the connection status through a heartbeat detection mechanism, automatically initiates the data retransmission process, and uses HTTP / HTTPS protocol to communicate asynchronously with the cloud server to ensure data integrity and timeliness.

[0040] In terms of technical effectiveness, this step effectively solves the problem of inconsistent data formats from heterogeneous devices, improving system compatibility and data processing efficiency. Simultaneously, the local caching and automatic retransmission mechanisms significantly enhance the system's robustness in network outage scenarios, providing a high-quality, continuous data input foundation for subsequent edge computing and AI analysis. This is a key step in realizing lightweight and intelligent dam safety monitoring.

[0041] S2 deploys a data processing module on a local server or industrial control computer based on an edge computing architecture. It performs feature extraction and multi-scale fusion analysis on the collected data and uses an improved rotating door algorithm to compress the data, thereby reducing data transmission and storage costs.

[0042] Specifically, this step involves deploying a data processing module on a local server or industrial control computer within an edge computing architecture. This module performs feature extraction and multi-scale fusion analysis of the collected data and employs an improved rotating door algorithm for data compression to reduce data transmission and storage costs. This step is a crucial component in the collaborative optimization of the data layer and analysis layer in this invention, exhibiting significant lightweight and intelligent characteristics.

[0043] At the technical implementation level, the data processing module is deployed on edge computing nodes (such as industrial control computers or embedded servers) and implemented using high-performance languages ​​such as Python or C++. First, feature extraction is performed on the raw data from heterogeneous sensors, including time-domain features (such as mean, variance, kurtosis, and number of times the threshold is exceeded), frequency-domain features (such as FFT main frequency amplitude and wavelet packet energy entropy), and correlation features (such as the correlation coefficient between water level and seepage pressure, and cumulative displacement). These feature vectors are integrated using multi-scale fusion analysis techniques. Methods such as weighted fusion or principal component analysis (PCA) are used to extract more discriminative comprehensive features, improving the accuracy and robustness of subsequent analyses.

[0044] At the parameter level, the improved version of the rotating door algorithm introduces a maximum permissible error ε as a compression threshold, for example, setting ε = 0.5 kPa in osmotic pressure data. The algorithm only retains inflection point data in the data sequence that exceeds the error range, thereby achieving efficient data compression. Combined with the LZW lossless compression algorithm, the overall compression ratio can reach 10:1, significantly reducing data storage space and transmission bandwidth requirements. In addition, the response latency of the edge processing module is controlled within 50ms, ensuring real-time requirements.

[0045] At the application level, this step is suitable for small hydropower station dam safety monitoring systems, especially in edge environments with limited bandwidth and scarce storage resources. Local preprocessing and compression can effectively alleviate the computational and communication pressure on the central cloud platform, making it suitable for monitoring scenarios with frequent water level fluctuations and significant seepage pressure changes during the rainy season.

[0046] The technical benefits of this step are that it improves the expressive power and analysis efficiency of data through feature extraction and multi-scale fusion; and it achieves high-precision, low-redundancy data compression through the improved rotating door algorithm, thereby significantly reducing system operating costs while ensuring monitoring accuracy, enhancing the autonomous processing capabilities of edge computing nodes, and providing key technical support for building an economical and lightweight monitoring system.

[0047] Furthermore, S2 includes:

[0048] S21, construct multi-scale feature vectors, including time-domain features, frequency-domain features, and correlation features, to improve the accuracy of data analysis.

[0049] Specifically, constructing multi-scale feature vectors is one of the core steps in the analysis layer of this invention. Its purpose is to improve the analytical accuracy and model generalization ability of dam safety monitoring data by fusing time-domain, frequency-domain, and correlated features. In some implementations, this step employs a multi-dimensional feature extraction and fusion strategy, combining the time-series characteristics of key dam monitoring parameters (such as water level, seepage pressure, and displacement) to construct a comprehensive feature vector with both physical meaning and statistical properties, providing high-quality input for subsequent anomaly detection and trend prediction.

[0050] From a technical implementation perspective, time-domain features include mean, variance, kurtosis, skewness, and number of times the threshold is crossed, used to characterize the statistical distribution and fluctuation characteristics of the data. Frequency-domain features extract the dominant frequency amplitude through Fast Fourier Transform (FFT) or obtain the energy entropy of different frequency bands using wavelet packet decomposition, thereby revealing periodic changes and potential structures in the data. Correlation features are based on the statistical relationships between multiple variables, such as the Pearson correlation coefficient between water level and seepage pressure, and the hysteretic correlation between displacement and temperature, reflecting the coupling mechanism between environmental factors and structural responses. Furthermore, the construction process of feature vectors needs to consider the balance and redundancy elimination of feature dimensions, and dimensionality reduction can be performed using Principal Component Analysis (PCA) or feature selection algorithms (such as sparsification methods based on LASSO).

[0051] At the parameter level, the calculation period for time-domain features is typically set to 1 to 24 hours to accommodate the dynamic characteristics of different monitoring parameters. In frequency-domain analysis, the recommended number of sampling points for FFT is 2048, with the frequency resolution controlled within 0.01 Hz. Wavelet packet decomposition can use the db4 wavelet basis function, with 5 decomposition levels covering the key frequency band from 0.1 Hz to 10 Hz. The calculation of correlation features requires setting the sliding window length (e.g., 72 hours) and lag step size (e.g., 1-3 hours) to capture the temporal dependencies between variables.

[0052] In practical applications, this step is mainly used to support the training and inference of lightweight AI models, especially in resource-constrained small hydropower station environments. Through multi-scale feature fusion, high-precision anomaly detection and trend prediction can be achieved with a smaller amount of data. For example, during the rapid rise in water levels during the rainy season, the frequency domain energy entropy change of seepage pressure data can serve as an early warning indicator, while a decrease in the correlation coefficient between water level and seepage pressure may indicate structural anomalies.

[0053] In terms of technical effectiveness, the construction of multi-scale feature vectors significantly improves the model's adaptability to non-stationary signals and enhances its sensitivity to identifying minute anomalies. Experiments show that, with the same amount of data, this method can improve anomaly detection accuracy by 15%-20% compared to a single feature extraction method, providing more reliable intelligent analysis support for dam safety.

[0054] S22, combining an improved rotating door algorithm with LZW lossless compression technology, achieves a data compression ratio of 10:1, significantly reducing transmission bandwidth requirements.

[0055] Specifically, this step involves a data compression method that combines an improved rotating door algorithm with LZW lossless compression technology on edge computing nodes to achieve a compression ratio of up to 10:1, thereby significantly reducing the transmission bandwidth requirements of dam safety monitoring data. This technical solution is a key optimization method at the data layer in this invention, aiming to improve the resource utilization and communication efficiency of small hydropower station monitoring systems.

[0056] At the technical implementation level, the improved Rotating Door Algorithm is a lossy data compression method based on error tolerance. Its core idea is to reduce the number of data points by setting a maximum permissible error ε and retaining only inflection point data points in the data sequence that exceed this error range. In this invention, for key monitoring parameters such as dam seepage pressure and water level, ε is set to 0.5 kPa (e.g., seepage pressure data) to ensure that the compressed data remains analyzable within the engineering accuracy range. The Rotating Door Algorithm preprocesses the original data at the edge, removing redundant data points to form a sparse but crucial data sequence.

[0057] Subsequently, the data sequence processed by the rotating door algorithm is input into the LZW (Lempel-Ziv-Welch) lossless compression algorithm for secondary compression. The LZW algorithm constructs a dynamic dictionary to encode recurring data patterns into shorter codes, thereby achieving efficient compression. In this invention, the LZW algorithm runs on an edge server, employs fixed-length encoding (e.g., 12-bit encoding length), and combines this with the sparsity characteristics of the data sequence to further improve compression efficiency. Practical testing has verified that this combined compression strategy achieves a compression ratio of 10:1 on a typical dam monitoring dataset, significantly outperforming traditional compression methods.

[0058] At the parameter level, the error threshold ε of the rotating door algorithm needs to be set according to the engineering accuracy requirements of the monitored parameters. For example, seepage pressure data usually requires an error of less than 0.5 kPa, while water level data can be set to ±1 cm. The encoding length, dictionary size, and compression window parameters of the LZW algorithm need to be optimized according to the characteristics of the data stream to balance compression efficiency and computational overhead.

[0059] In application scenarios, this step is deployed on the edge computing nodes of small hydropower stations, such as industrial control computers or existing servers, without requiring additional hardware investment. During the rainy season or when reservoir capacity changes drastically, the system will collect key parameters at high frequency. This compression strategy effectively reduces the amount of data uploaded, alleviating bandwidth bottlenecks, and is particularly suitable for 4G / 5G wireless networks or bandwidth-constrained industrial Ethernet environments.

[0060] The technical effect of this step is that, through the synergistic effect of lossy and lossless compression, the amount of data transmitted is greatly reduced, communication costs and latency are lowered, while ensuring the integrity and availability of key data. This provides high-quality data input for subsequent real-time analysis and early warning, and is an important supporting technology for realizing a lightweight, low-cost dam monitoring system.

[0061] S3 constructs a lightweight anomaly detection model, uses historical data to train and generate a normal pattern template library, calculates the dynamic time warp distance between new data and templates in real time, and triggers an alert when the distance exceeds a preset threshold for multiple consecutive periods.

[0062] Specifically, this step involves constructing a lightweight anomaly detection model. Its core lies in generating a normal pattern template library through training with historical data, and using the Dynamic Time Warping (DTW) algorithm in real-time monitoring to calculate the distance between new data and the templates, thereby enabling the identification of anomalies in the dam's operational status. In some implementations, this model is trained based on an improved Isolation Forest algorithm to adapt to the environment of limited edge computing resources at small hydropower stations.

[0063] At the technical implementation level, multi-scale feature vectors are first extracted from historical monitoring data collected by the perception layer. These vectors include time-domain features (such as mean, variance, kurtosis, and number of times the threshold is exceeded), frequency-domain features (such as FFT main frequency amplitude and wavelet packet energy entropy), and correlation features (such as the correlation coefficient between water level and seepage pressure, and cumulative displacement). These feature vectors are used to train the Isolation Forest model, enabling it to identify normal operating modes and to build a template library. After the model training is completed, it is deployed on edge servers or industrial control computers to reduce reliance on cloud computing resources.

[0064] At the parameter level, the number of trees in the isolated forest (n_estimators) in the model is set to 50 to achieve a balance between computational efficiency and detection accuracy. When calculating the DTW distance, a dynamic window constraint (such as Sakoe-Chiba bandwidth) is used to reduce computational complexity; the window width is typically set to 10% of the time series length. The anomaly trigger threshold is set when the DTW distance exceeds a preset threshold (e.g., 0.8) for three consecutive monitoring periods, thereby avoiding false alarms caused by instantaneous fluctuations.

[0065] In application scenarios, this step is suitable for real-time anomaly detection of key dam parameters (such as seepage pressure, displacement, and water level), especially during the rainy season or when reservoir capacity changes frequently, enabling timely detection of potential structural anomalies. Combined with edge computing and dynamic sampling strategies, it can operate stably under low bandwidth and low computing power conditions.

[0066] The technical effect of this step is that, through a lightweight model and an efficient distance calculation method, it achieves highly sensitive and low false alarm rate anomaly detection of dam operation status, providing an economical and efficient intelligent monitoring method for small hydropower stations, and effectively improving the real-time performance and reliability of dam safety early warning.

[0067] Furthermore, S3 includes:

[0068] S31 employs an improved isolated forest algorithm with a lightweight configuration of n_estimators=50 to reduce computational resource consumption.

[0069] Specifically, this step employs an improved Isolation Forest algorithm with a lightweight configuration of `n_estimators=50` to achieve efficient detection of abnormal data in the dam safety monitoring system while significantly reducing computational resource consumption. Isolation Forest is an unsupervised anomaly detection algorithm based on ensemble learning. Its core principle is to isolate data points by randomly selecting features and split points. Anomalies, due to their sparse distribution in the feature space, are typically isolated more quickly, resulting in shorter path lengths. This invention optimizes the traditional Isolation Forest algorithm by introducing historical data clustering to generate a normal pattern template library and combining it with the Dynamic Time Warping (DTW) algorithm to calculate the distance between real-time data and the templates, thereby improving detection accuracy and adaptability.

[0070] In its implementation, the algorithm first trains on historical monitoring data (such as water level, seepage pressure, and displacement) to build a model containing 50 isolated trees. Each tree is constructed using a random partitioning strategy, with a partitioning depth not exceeding 256 layers to control model complexity. During the real-time detection phase, the system calculates the DTW distance between the current sampled data and the template library. If the distance exceeds a set threshold (e.g., 0.8) for three consecutive sampling periods, it is considered an anomaly and triggers an early warning mechanism. In this step, setting `n_estimators=50` ensures detection performance while keeping model memory usage below 500MB and CPU utilization below 15%, making it suitable for deployment on edge computing devices (such as industrial control computers or embedded servers).

[0071] This technical solution is suitable for resource-constrained environments at small hydropower stations, especially for deploying edge analytics modules on existing server infrastructure without requiring additional high-performance computing platforms. Through lightweight model design, the system can achieve real-time anomaly detection of key dam parameters under low power consumption and low bandwidth conditions, improving system response speed and stability. Furthermore, this step, as one of the core algorithms of the analysis layer, works in conjunction with dynamic sampling strategies and data compression modules to optimize the allocation of computing and communication resources, providing crucial technical support for building an economical, practical, and sustainable dam safety monitoring system.

[0072] S32 calculates the DTW dynamic time warping distance between new data and the template in real time. If the distance exceeds the threshold for three consecutive periods, an early warning is triggered.

[0073] Specifically, this step involves a real-time anomaly detection mechanism based on the Dynamic Time Warping (DTW) algorithm, which is one of the key technologies of the analysis layer in the dam safety monitoring system of this invention. Its core principle is to determine whether the current data deviates from the normal state by calculating the DTW distance between the real-time acquired time-series data and historical normal pattern templates. In some implementations, the system first trains an Isolation Forest model using historical data, extracting multiple time series as a normal pattern template library. Each template represents the typical behavioral characteristics of a certain type of monitoring parameter (such as water level, seepage pressure, displacement, etc.) under specific operating conditions.

[0074] During the real-time detection phase, the system calculates the DTW distance between the newly collected data from the current period and the historical templates with corresponding parameters in the template library. The DTW algorithm, through non-linear alignment of two time series, allows for similarity comparison of data of different lengths or with time offsets, making it particularly suitable for addressing the issue of incomplete synchronization of time series due to environmental disturbances or equipment response delays in dam monitoring. Specifically, the FastDTW algorithm is employed to reduce computational complexity while meeting real-time requirements (response time < 1 second).

[0075] If the DTW distance exceeds a preset threshold (e.g., distance > 0.8) for three consecutive sampling periods (each period interval can be set to 5-10 minutes), the system determines it to be in an abnormal state and triggers an early warning mechanism. This threshold can be adaptively adjusted according to different parameter types and historical data distribution to improve detection accuracy. In practical applications, this step is deployed on edge computing nodes, combined with a lightweight AI model, effectively reducing dependence on cloud computing resources and improving system response speed and stability, especially suitable for small hydropower station environments with limited network conditions. Its technical value lies in achieving highly sensitive anomaly identification of key dam parameters, providing a reliable basis for safety early warning.

[0076] S4 dynamically adjusts the sensor sampling frequency based on real-time data analysis results. Key parameters are acquired at high frequency, while non-key parameters are acquired at low frequency or intermittently based on the predicted rate of change, in order to optimize the efficiency of system resource utilization.

[0077] Specifically, this step involves dynamically adjusting the sensor sampling frequency based on real-time data analysis results to optimize system resource allocation. In some implementations, this strategy is based on the distinction between critical and non-critical parameters, combined with a predictive model of parameter change rate, to adaptively adjust the sensor sampling frequency. Specifically, critical parameters (such as water level, seepage pressure, and displacement) have a direct impact on dam structural safety and require high-frequency sampling (e.g., 1Hz) to ensure data real-time performance and accuracy; while non-critical parameters (such as ambient temperature and humidity) are sampled at low frequencies (e.g., 0.1Hz) or intermittently based on their rate of change prediction results, thereby reducing resource consumption for data acquisition and transmission.

[0078] At the technical implementation level, this step relies on a lightweight data analysis model deployed on edge computing nodes. The system predicts parameter change trends using a Kalman filter algorithm, calculates the rate of change Δ of the current parameters, and compares it with preset thresholds TH1 (e.g., 0.5%) and TH2 (e.g., 0.1%) to determine whether to adjust the sampling frequency. Simultaneously, the system incorporates a built-in sensor health assessment model, which quantitatively evaluates sensor data quality by calculating metrics such as signal-to-noise ratio (SNR) and zero-drift, further optimizing the sampling strategy.

[0079] Regarding parameters, the sampling frequency can be adjusted from 0.1Hz to 1Hz, and the rate of change thresholds TH1 and TH2 can be configured according to specific application scenarios. The error tolerance ε (e.g., 0.5kPa for seepage pressure data) is used to control the data compression accuracy. This strategy complies with power system safety and monitoring standards such as IEC 62443 and DL / T 578, and is suitable for resource-constrained small hydropower station environments.

[0080] This step is of significant value in the system. Through dynamic sampling strategies, it can effectively reduce the energy consumption and bandwidth requirements for data acquisition and transmission, improve the overall operating efficiency of the system, and at the same time ensure the monitoring accuracy of key safety parameters, providing reliable data support for dam safety assessment and early warning.

[0081] Furthermore, S4 includes:

[0082] S41, based on the Kalman filter to predict the parameter change trend, if the parameter change rate Δ is greater than the preset threshold TH1, the sampling frequency will be increased to 1Hz.

[0083] Specifically, this step involves predicting the change trend of the key monitoring parameters of the dam based on the Kalman filter algorithm, and dynamically adjusting the sampling frequency according to the prediction results to achieve a balance between resource optimization and data quality assurance. In some implementations, the system first collects the raw sensor data, removes noise and outliers through the preprocessing module, and then inputs the data into the Kalman filter for state estimation and trend prediction. The Kalman filter constructs a state space model, combines the system model and the observation model, optimally estimates the parameter value at the current moment, and predicts the change rate Δ at the next moment. This change rate Δ can be calculated by taking the difference between the state estimate value and its previous moment value, that is where is the estimated value at the current moment, and Δt is the time step.

[0084] At the parameter index level, the system sets two dynamic adjustment thresholds TH1 and TH2. For example, TH1 = 0.5 mm / min (for displacement parameters) or TH1 = 0.3 kPa / min (for seepage pressure parameters), TH2 = 0.1 mm / min or TH2 = 0.1 kPa / min. When Δ > TH1, the system determines that the parameter is in a rapidly changing state, which may indicate potential risks. Therefore, the sampling frequency is increased from the default 0.5 Hz to 1 Hz to improve the real-time performance and resolution of data acquisition. Conversely, if Δ < TH2, it is considered that the parameter changes slowly, and the sampling frequency can be reduced to 0.1 Hz, thereby reducing the data volume and system load.

[0085] This step is mainly used for dynamic monitoring of key parameters of the dam (such as displacement, seepage pressure, water level, etc.) in practical applications. For example, during the rainy season or when the reservoir water level rises rapidly, the change rate Δ of seepage pressure may increase significantly. At this time, the system automatically increases the sampling frequency to ensure that abnormal changes can be captured in a timely manner. During the stable operation stage, the system can reduce the sampling frequency to save storage and communication resources.

[0086] From the technical effect perspective, this step achieves an adaptive balance between the data acquisition accuracy and resource consumption of the monitoring system. By introducing the Kalman filter prediction mechanism, the system can intelligently adjust the sampling strategy according to the parameter change trend without manual intervention, thereby improving the data quality during key periods, while reducing the energy consumption and storage pressure during non-critical periods, which conforms to the core design concept of "lightweight, adaptive, low-cost" of this invention.

[0087] S42, if the parameter change rate Δ is less than the preset threshold TH2, then reduce the sampling frequency to 0.1 Hz, otherwise maintain the default frequency of 0.5 Hz.

[0088] Specifically, the step "if the parameter change rate Δ is less than the preset threshold TH2, then reduce the sampling frequency to 0.1Hz, otherwise keep the default frequency of 0.5Hz" is one of the core logics of the dynamic sampling frequency control strategy in this invention, which aims to achieve a balance between resource optimization and data acquisition efficiency in the dam safety monitoring system.

[0089] This step, based on real-time analysis of the changing trends of the monitored parameters, uses a Kalman filter algorithm to dynamically model and predict the time-series data collected by the sensor, thereby calculating the rate of change Δ of the current parameter. The rate of change Δ is typically defined as the ratio of the difference between two adjacent sampling points to the sampling interval, i.e., Δ = |x(t) - x(t-1)| / Δt. When the Δ value is below a preset threshold TH2 for several consecutive periods (e.g., 3 periods), the system determines that the parameter is in a stable state and does not require high-frequency acquisition, thus triggering a mechanism to reduce the sampling frequency. In some implementations, this frequency adjustment can be achieved through a control logic module (such as a Python script or embedded C program) in the edge computing node, communicating with the sensor's acquisition controller to dynamically modify its sampling interval.

[0090] The TH2 setting needs to be combined with the physical characteristics of the specific monitoring parameters and engineering experience. For example, for a pressure sensor, TH2 can be set to 0.05 kPa / s; for a displacement sensor, TH2 can be set to 0.01 mm / s. The sampling frequency adjustment range is 0.1 Hz to 1 Hz, where 0.1 Hz corresponds to sampling once every 10 seconds, and 0.5 Hz is the default frequency, corresponding to sampling once every 2 seconds. This strategy significantly reduces the sampling burden during non-critical periods while ensuring data continuity, thereby saving power, bandwidth, and storage resources.

[0091] This step is applicable to the long-term stable operation phase of small hydropower station dams, especially during the non-rainy season or periods of stable reservoir water levels. In actual deployment, edge computing nodes are deployed on local servers or industrial control computers at the power station, receiving data streams from heterogeneous sensors in real time and dynamically adjusting the sampling frequency according to the rate of change Δ. For example, under conditions of no rainfall and no flood discharge, seepage pressure changes slowly, and the system can automatically reduce the sampling frequency from 0.5Hz to 0.1Hz, thereby reducing data transmission pressure and server load.

[0092] This step enables the system to intelligently reduce the data acquisition frequency during non-critical states, effectively reducing the continuous consumption of hardware resources, extending equipment lifespan, and decreasing data storage and transmission costs. Simultaneously, this strategy improves the overall energy efficiency ratio of the system without affecting the trend analysis of key parameters, providing crucial technical support for building an economical and lightweight monitoring system for small hydropower stations.

[0093] The lightweight construction method for dam safety monitoring system in this invention enables low-cost and efficient construction of the dam safety monitoring system. Through protocol compatibility, dynamic data acquisition, and edge computing optimization, it reduces hardware and communication resource consumption and improves data processing efficiency and anomaly detection capabilities.

[0094] Also includes:

[0095] The S5 can synchronize with mobile devices via WeChat or DingTalk robots, push daily dam health reports, provide immediate alerts for exceeding threshold limits, and support remote confirmation of handling procedures.

[0096] Specifically, this step achieves mobile synchronization through WeChat or DingTalk robots, pushes dam health briefings daily at set times, and issues immediate alarms when monitoring parameters exceed limits. It also supports remote confirmation and handling processes and is a key communication and interaction module in the application layer of this invention.

[0097] At the technical implementation level, the system integrates with the backend monitoring platform using the API interfaces of WeChat Work or DingTalk, and proactively pushes messages through a Webhook mechanism. Daily health reports are automatically generated by edge computing nodes or the central server based on preset templates. The reports include statistical values, trends, and health scores for key monitoring parameters (such as water level, seepage pressure, and displacement), in structured text or visual charts (such as PNG images exported from Grafana). The alarm mechanism is based on real-time data analysis results. When a parameter exceeds a set threshold for three consecutive sampling periods (e.g., the seepage pressure threshold is set to 95% of the design value), the system automatically triggers an alarm message containing the device ID, parameter name, current value, threshold, timestamp, and preliminary judgment suggestions.

[0098] Regarding parameters and metrics, briefings are pushed out daily, alarm response latency is controlled within 5 seconds, and message format must conform to the Markdown or JSON standards of WeChat / DingTalk, supporting mixed text and image layouts and link jumps. Alarm thresholds are dynamically adjusted based on dam structural characteristics and historical data distribution. Initial thresholds can be set based on statistical control charts (such as the 3σ principle), and subsequently optimized using transfer learning models. The remote confirmation process supports a multi-level approval mechanism, embedding confirmation buttons or jump links in alarm messages. Maintenance personnel can complete confirmation, status updates, and other operations on mobile devices, and the system records operation logs and synchronizes them to the database.

[0099] This step is deployed in the operation and maintenance management scenario of small hydropower stations, suitable for operation and maintenance models that lack dedicated IT teams and rely on mobile office work. Through mobile synchronization, it enables 24 / 7 monitoring information access, improves response efficiency, and reduces the frequency of manual inspections. Its technical value lies in building a low-latency, highly available mobile alarm and response closed loop, which, combined with edge computing and AI model output, significantly improves the intelligence and real-time performance of dam safety monitoring.

[0100] The lightweight construction method for the dam safety monitoring system in this invention further enhances the real-time response capability and ease of operation and maintenance of the dam safety monitoring system through a mobile terminal synchronization mechanism and an instant alarm push function, enabling rapid perception and remote closed-loop handling of abnormal events, and enhancing the overall intelligence and collaborative management efficiency of the system.

[0101] The lightweight construction method for dam safety monitoring system in this invention enables low-cost and efficient construction of the dam safety monitoring system. Through protocol compatibility, dynamic data acquisition, and edge computing optimization, it reduces hardware and communication resource consumption and improves data processing efficiency and anomaly detection capabilities.

[0102] Example 2

[0103] like Figure 2 The diagram shows the architecture of the lightweight construction method for the dam safety monitoring system of the present invention.

[0104] I. System Design Principles:

[0105] 1. Maximize the utilization of existing equipment: Reuse existing sensors (coverage ≥85%) through protocol adaptation;

[0106] 2. Adaptive data acquisition strategy: dynamically adjust the sampling frequency (adjustable from 0.1Hz to 1Hz), acquire key parameters at high frequency, and record secondary parameters at low frequency;

[0107] 3. Lightweight edge computing architecture: Deploy edge analytics modules on existing servers / industrial control computers to reduce reliance on the cloud;

[0108] 4. Multi-source data fusion analysis: Combine environmental data (such as rainfall and reservoir capacity changes) with equipment operating status to build a comprehensive evaluation model.

[0109] II. Technical Implementation Path

[0110] 1. Hardware Layer: Heterogeneous Device Compatibility Solution

[0111] Protocol Conversion Gateway: Deploy an embedded protocol conversion device (such as a Raspberry Pi + industrial communication module) to achieve the following functions:

[0112] Protocol parsing: Supports common protocols such as Modbus RTU / TCP, CANopen, DL / T645, etc., and adapts to private protocols through configuration files;

[0113] Data standardization: Converts raw data into a unified JSON format, including fields such as timestamp, device ID, value, quality code, etc.;

[0114] Cache queue: Locally stores data for 72 hours when the network is disconnected, and automatically retransmits it after the network is restored.

[0115] Sensor efficiency evaluation:

[0116] Use Python to establish a sensor health evaluation model:<​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Then, edge-side data compression is performed using an improved rotating door algorithm: a maximum allowable error ε is set (e.g., ε = 0.5 kPa for osmotic pressure data); only inflection point data exceeding the error range is stored; combined with LZW lossless compression, a compression ratio of 10:1 is achieved.

[0130] 3. Analysis Layer: Lightweight AI Model

[0131] Feature extraction and fusion: Constructing multi-scale feature vectors: Time domain features: mean, variance, kurtosis, number of times the threshold is exceeded; Frequency domain features: FFT main frequency amplitude, wavelet packet energy entropy; Correlation features: water level seepage pressure correlation coefficient, cumulative displacement.

[0132] Lightweight anomaly detection: Employs an improved isolated forest algorithm: Clusters historical data to generate a normal pattern template library; DTW dynamic time warping distance between new data and templates is calculated in real time; If the distance exceeds the threshold for three consecutive periods, an alert is triggered.

[0133] 4. Application Layer: Visualization and Decision Support

[0134] A web visualization platform was built, and a monitoring panel was built based on the open-source Grafana. Key functions include: 3D status map of the dam: with overlaid heat maps of parameters such as displacement and seepage; health trend curve: displaying the changes in the health index over 7 days / 30 days; and early warning classification management: setting three levels of response: blue (observation), yellow (early warning), and red (emergency).

[0135] Mobile devices can synchronize with WeChat / DingTalk robots to achieve: daily health briefing push; instant alarms for exceeding thresholds; and remote confirmation of handling procedures.

[0136] In summary, compared with the original monitoring scheme, this method features a protocol-independent architecture: it supports private protocols by dynamically loading DLL / SO libraries without requiring vendor cooperation; optimized computation and communication: 80% of data processing is completed at the edge, reducing bandwidth requirements; and knowledge transfer learning: it utilizes pre-trained models to quickly adapt to new power plants, reducing the need for labeled data.

[0137] Example 3

[0138] To achieve the above embodiments, such as Figure 3 As shown, this embodiment also provides a lightweight construction device 10 for a dam safety monitoring system, including:

[0139] The protocol adaptation and caching module 100 is used to deploy a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generate data in a unified format, and support offline caching and automatic retransmission.

[0140] The edge feature fusion and compression module 200 is used to deploy a data processing module on a local server or industrial control computer based on an edge computing architecture. It performs feature extraction and multi-scale fusion analysis on the collected data, and uses an improved rotating door algorithm to compress the data to reduce data transmission and storage costs.

[0141] The anomaly detection and early warning module 300 is used to build a lightweight anomaly detection model, use historical data to train and generate a normal pattern template library, calculate the dynamic time warp distance between new data and templates in real time, and trigger an early warning when the distance exceeds a preset threshold for multiple consecutive periods.

[0142] The sampling frequency dynamic adjustment module 400 is used to dynamically adjust the sampling frequency of the sensor based on the real-time data analysis results. Key parameters are acquired at high frequency, while non-key parameters are acquired at low frequency or intermittently based on the prediction results of the rate of change, so as to optimize the efficiency of system resource utilization.

[0143] Furthermore, the protocol adaptation and caching module is also used for:

[0144] Dynamically load DLL / SO libraries into embedded devices to adapt to proprietary protocols, without requiring cooperation from sensor manufacturers;

[0145] The raw data is converted into JSON format containing timestamps, device IDs, values, and quality codes, and the data is cached locally for 72 hours when the network is down, and automatically re-uploaded when the network is restored.

[0146] Furthermore, the edge feature fusion and compression module is also used for:

[0147] Construct multi-scale feature vectors, including time-domain features, frequency-domain features, and correlation features, to improve the accuracy of data analysis;

[0148] By combining the improved rotating door algorithm with LZW lossless compression technology, a data compression ratio of 10:1 is achieved, significantly reducing the transmission bandwidth requirements.

[0149] Furthermore, the anomaly detection and early warning module is also used for:

[0150] An improved isolated forest algorithm is adopted, and a lightweight configuration of n_estimators=50 is set to reduce the consumption of computing resources;

[0151] The DTW dynamic time warping distance between new data and the template is calculated in real time. If the distance exceeds the threshold for three consecutive periods, an early warning is triggered.

[0152] Furthermore, the sampling frequency dynamic adjustment module is also used for:

[0153] Based on the Kalman filter to predict the parameter change trend, if the parameter change rate Δ is greater than the preset threshold TH1, the sampling frequency will be increased to 1Hz.

[0154] If the parameter change rate Δ is less than the preset threshold TH2, the sampling frequency will be reduced to 0.1Hz; otherwise, the default frequency of 0.5Hz will be maintained.

[0155] Furthermore, it also includes:

[0156] The mobile synchronization module is used to achieve mobile synchronization via WeChat or DingTalk robots, push daily dam health briefings, provide immediate alarms for threshold exceeding situations, and support remote confirmation of handling procedures.

[0157] The lightweight construction device for the dam safety monitoring system in this embodiment of the invention further enhances the real-time response capability and ease of operation and maintenance of the dam safety monitoring system through a mobile terminal synchronization mechanism and an instant alarm push function, enabling rapid perception and remote closed-loop handling of abnormal events, and enhancing the overall intelligence and collaborative management efficiency of the system.

[0158] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0159] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A lightweight construction method for a dam safety monitoring system, characterized in that, include: S1 deploys a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generating data in a unified format and supporting offline caching and automatic retransmission. S2 deploys a data processing module on a local server or industrial control computer based on an edge computing architecture. It performs feature extraction and multi-scale fusion analysis on the collected data and uses an improved rotating door algorithm to compress the data in order to reduce data transmission and storage costs. S3: Construct a lightweight anomaly detection model, use historical data to train and generate a normal pattern template library, calculate the dynamic time warp distance between new data and templates in real time, and trigger an alert when the distance exceeds a preset threshold for multiple consecutive periods. S4 dynamically adjusts the sensor sampling frequency based on real-time data analysis results. Key parameters are acquired at high frequency, while non-key parameters are acquired at low frequency or intermittently based on the predicted rate of change, in order to optimize the efficiency of system resource utilization.

2. The method as described in claim 1, characterized in that, The deployment protocol conversion gateway performs protocol parsing and data standardization processing on existing heterogeneous sensors, generates data in a unified format, and supports offline caching and automatic retransmission. It also includes: S11 dynamically loads DLL / SO libraries through embedded devices to adapt to proprietary protocols, without requiring cooperation from sensor manufacturers; S12 converts the raw data into JSON format containing timestamps, device IDs, values, and quality codes, and caches the data locally for 72 hours when the network is down, automatically re-uploading it after the network is restored.

3. The method as described in claim 1, characterized in that, The edge computing architecture deploys a data processing module on a local server or industrial control computer to perform feature extraction and multi-scale fusion analysis on the collected data, and uses an improved rotating door algorithm for data compression to reduce data transmission and storage costs. It also includes: S21, construct multi-scale feature vectors, including time-domain features, frequency-domain features, and correlation features, to improve the accuracy of data analysis; S22, combining an improved rotating door algorithm with LZW lossless compression technology, achieves a data compression ratio of 10:1, significantly reducing transmission bandwidth requirements.

4. The method as described in claim 1, characterized in that, The construction of a lightweight anomaly detection model, which uses historical data to train and generate a normal pattern template library, calculates the dynamic time-warped distance between new data and the templates in real time, and triggers an alert when the distance exceeds a preset threshold for multiple consecutive periods, also includes: S31 adopts an improved isolated forest algorithm and sets a lightweight configuration of n_estimators=50 to reduce the consumption of computing resources; S32 calculates the DTW dynamic time warping distance between new data and the template in real time. If the distance exceeds the threshold for three consecutive periods, an early warning is triggered.

5. The method as described in claim 1, characterized in that, The method of dynamically adjusting the sensor sampling frequency based on real-time data analysis results, with key parameters acquired at high frequency and non-key parameters acquired at low frequency or intermittently based on the rate of change prediction results, in order to optimize system resource utilization efficiency, also includes: S41, based on the Kalman filter to predict the parameter change trend, if the parameter change rate Δ is greater than the preset threshold TH1, the sampling frequency will be increased to 1Hz; S42, if the parameter change rate Δ is less than the preset threshold TH2, then the sampling frequency is reduced to 0.1Hz; otherwise, the default frequency of 0.5Hz is maintained.

6. The method as described in claim 1, characterized in that, Also includes: The S5 can synchronize with mobile devices via WeChat or DingTalk robots, push daily dam health reports, provide immediate alerts for exceeding threshold limits, and support remote confirmation of handling procedures.

7. A lightweight construction device for a dam safety monitoring system, characterized in that, include: The protocol adaptation and caching module is used to deploy a protocol conversion gateway to perform protocol parsing and data standardization processing on existing heterogeneous sensors, generate data in a unified format, and support offline caching and automatic retransmission. The edge feature fusion and compression module is used to deploy data processing modules on local servers or industrial control computers based on edge computing architecture. It performs feature extraction and multi-scale fusion analysis on the collected data, and uses an improved rotating door algorithm to compress the data to reduce data transmission and storage costs. The anomaly detection and early warning module is used to build a lightweight anomaly detection model, use historical data to train and generate a normal pattern template library, calculate the dynamic time warp distance between new data and templates in real time, and trigger an early warning when the distance exceeds a preset threshold for multiple consecutive periods. The sampling frequency dynamic adjustment module is used to dynamically adjust the sampling frequency of the sensor based on the real-time data analysis results. Key parameters are acquired at high frequency, while non-key parameters are acquired at low frequency or intermittently based on the prediction results of the rate of change, so as to optimize the efficiency of system resource utilization.

8. The apparatus as claimed in claim 7, characterized in that, The protocol adaptation and caching module is also used for: Dynamically load DLL / SO libraries into embedded devices to adapt to proprietary protocols, without requiring cooperation from sensor manufacturers; The raw data is converted into JSON format containing timestamps, device IDs, values, and quality codes, and the data is cached locally for 72 hours when the network is down, and automatically re-uploaded when the network is restored.

9. The apparatus as claimed in claim 7, characterized in that, The edge feature fusion and compression module is also used for: Construct multi-scale feature vectors, including time-domain features, frequency-domain features, and correlation features, to improve the accuracy of data analysis; By combining the improved rotating door algorithm with LZW lossless compression technology, a data compression ratio of 10:1 is achieved, significantly reducing the transmission bandwidth requirements.

10. The apparatus as claimed in claim 7, characterized in that, The anomaly detection and early warning module is also used for: An improved isolated forest algorithm is adopted, and a lightweight configuration of n_estimators=50 is set to reduce the consumption of computing resources; The DTW dynamic time warping distance between new data and the template is calculated in real time. If the distance exceeds the threshold for three consecutive periods, an early warning is triggered.

11. The apparatus as claimed in claim 7, characterized in that, The sampling frequency dynamic adjustment module is also used for: Based on the Kalman filter to predict the parameter change trend, if the parameter change rate Δ is greater than the preset threshold TH1, the sampling frequency will be increased to 1Hz. If the parameter change rate Δ is less than the preset threshold TH2, the sampling frequency will be reduced to 0.1Hz; otherwise, the default frequency of 0.5Hz will be maintained.

12. The apparatus as claimed in claim 7, characterized in that, Also includes: The mobile synchronization module is used to achieve mobile synchronization via WeChat or DingTalk robots, push daily dam health briefings, provide immediate alarms for threshold exceeding situations, and support remote confirmation of handling procedures.

Images