A method and system for intelligent monitoring of water conservancy project data

Abstract

This invention relates to the field of water conservancy engineering technology, specifically to an intelligent monitoring method and system for water conservancy engineering data. The method includes: real-time acquisition of multi-dimensional data within a target water area at the current moment, forming a current data point; inputting the current data point into a network model and outputting the corresponding first anomaly probability; acquiring multi-dimensional data from multiple moments prior to the current moment, with each dimension containing data from all sampling moments of the current moment forming a sequence; performing frequency domain analysis on each dimension sequence to determine the anomaly probability of the current moment's data in the corresponding dimension, and using the maximum anomaly probability across all dimensions at the current moment as a second anomaly probability; integrating the first and second anomaly probabilities to obtain a comprehensive probability; and, in response to the comprehensive probability exceeding a set threshold, indicating an anomaly at the current data point, requiring notification to staff for appropriate measures. In other words, the solution of this invention can accurately monitor water conservancy engineering data.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to an intelligent monitoring method and system for water conservancy engineering data. Background Technology

[0002] Water conservancy projects (such as dams, reservoirs, dikes, and canals) are crucial infrastructure for ensuring water resource management, flood control, power generation, and irrigation. These projects are often located in complex natural environments and are affected by factors such as water pressure, temperature variations, geological activity, and material aging, which may lead to subtle anomalies (such as cracks, leaks, deformation, and settlement). If these subtle changes are not identified in time, they may develop into major safety hazards, causing project failure or even catastrophic consequences. Therefore, the identification of subtle anomalies is crucial in the monitoring and maintenance of water conservancy projects.

[0003] Currently, with the rapid development of IoT, cloud computing and big data technologies, digital twin technology is gradually becoming an important means to improve the management efficiency of water conservancy projects. By creating virtual models of physical water conservancy facilities and monitoring and analyzing them in real time, digital twins can effectively integrate various sensor data and reflect information such as hydrological, meteorological and water resource changes in real time.

[0004] By using high-precision sensors and remote sensing technology, real-time operational data of water conservancy projects can be collected, including key parameters such as water level, flow velocity, and pressure. Using this data to build a digital twin model, the physical state of water conservancy facilities can be dynamically reflected. The digital twin model can provide real-time monitoring capabilities, enabling managers to view the operational status of water conservancy facilities at any time.

[0005] By using data analysis algorithms, potential faults and risks can be identified, early warning capabilities can be improved, and the safe and stable operation of facilities can be ensured. Furthermore, by integrating environmental data such as meteorology and hydrology into digital twin models, the adaptability to changes in external factors can be enhanced. The models can simulate engineering performance under different environmental conditions, supporting decision-makers in formulating more scientific management strategies.

[0006] Although existing technologies can perform real-time monitoring of water conservancy project data, they still have the following problems:

[0007] 1) Data acquisition accuracy limitations: Current data acquisition equipment such as sensors have limited accuracy, making it difficult to accurately capture some minute changes in water conservancy projects.

[0008] 2) Data transmission and processing delays: In the digital twin system of water conservancy projects, data needs to be collected from the physical entity and then transmitted to the digital model for processing and analysis. This process may involve certain delays. For example, the speed of data transmission and processing may not keep up with the speed of change in the actual physical process. In this case, the analysis of water conservancy project data may not reflect the current operating status in real time, affecting the timeliness of decision-making. Summary of the Invention

[0009] The purpose of this invention is to propose an intelligent monitoring method and system for water conservancy project data, in order to solve the problem that existing technologies are unable to reflect the changes in the state of water conservancy projects in real time, thus making it impossible to accurately monitor water conservancy projects intelligently; to this end, this invention provides solutions in the following two aspects.

[0010] In a first aspect, the present invention provides an intelligent monitoring method for water conservancy project data, comprising:

[0011] Real-time data collection of multiple dimensions within the target water area at the current moment, forming the current data point; the multiple dimensions of data include dam leakage, ambient temperature, rainfall, water flow, water pressure, and water level;

[0012] Input the current data point into the network model and output the corresponding first anomaly probability;

[0013] Data from multiple dimensions at multiple time points prior to the current time point are obtained. Data from all sampling time points at the current time point in the same dimension constitute a sequence. Frequency domain analysis is performed on each dimension sequence to determine the anomaly probability of the current time point data in the corresponding dimension. The maximum anomaly probability among all dimensions at the current time point is taken as the second anomaly probability.

[0014] The first anomaly probability and the second anomaly probability are integrated to obtain a comprehensive probability;

[0015] If the overall probability exceeds a set threshold, the data point at the current moment is considered abnormal, and staff must be notified to take appropriate measures.

[0016] The above-mentioned scheme takes two approaches: one is time-domain analysis, which uses network models to analyze the data and determine the probability of anomalies at each sampling time; the other is frequency-domain analysis to capture subtle abnormal changes. Finally, the data from both perspectives are fused to obtain the final comprehensive probability of the water conservancy project data, thereby identifying anomalies at each sampling time. In other words, the scheme of this invention performs anomaly analysis from two perspectives, accurately identifying anomalies in water conservancy project data and improving the reliability of monitoring.

[0017] Optionally, the second weight of the integrated second anomaly probability is:

[0018] STL decomposition is performed on the dimensional sequence corresponding to the second anomaly probability to obtain the sum of seasonal components and residual components under each dimension;

[0019] The second weight is negatively correlated with the sum of the seasonal component and the remainder;

[0020] The sum of the first weight and the second weight of the first anomaly probability is 1.

[0021] The above scheme adaptively obtains the weights of the first and second anomaly probabilities based on the characteristics of the data points at the current moment, which makes the fused comprehensive probability more accurate.

[0022] Optionally, the process of obtaining the anomaly probability is as follows:

[0023] The time-frequency diagram is obtained by processing the sequences of each dimension using short-time Fourier transform.

[0024] Based on the absolute value of the difference between the target sampling time and the sampling spectrum amplitude of the target frequency on the time-frequency graph and the previous adjacent time, as well as the mean value of the spectrum amplitude of the target frequency, anomaly indicators are obtained, and the anomaly indicators are normalized to obtain anomaly probability; the target sampling time is any sampling time.

[0025] By analyzing from a frequency domain perspective, we can assess anomalies in the target water area and ensure the stability and safety of the system.

[0026] Optionally, the anomaly index is the average of the global deviation and the local deviation;

[0027] , ;

[0028] in, For global deviation, This is a local deviation; The first on the time-frequency graph The average frequency amplitude of each frequency; The first on the time-frequency graph The spectral amplitude of a frequency at time n. For the first The spectral amplitude of a frequency at time n-1.

[0029] The above scheme provides a method for accurately obtaining abnormal indicators.

[0030] Optionally, the network model is a fully connected network model.

[0031] Optionally, the training process of the network model is as follows:

[0032] Obtain a training set, which includes historical data points from multiple moments within a historical period; the labels are the normal or abnormal conditions of each data point at each moment.

[0033] The network model is trained using the training set to obtain a trained network model.

[0034] Optionally, it also includes: performing time alignment processing on the dimensional data in the historical data points at each time point to obtain aligned historical data points.

[0035] The above approach aligns the data across different dimensions over time, providing data support for obtaining an accurate network model.

[0036] Optionally, the normalization process employs... Function or maximum / minimum value normalization method.

[0037] Optionally, the dam leakage, water flow, water pressure, and water level are obtained by measuring a piezometer, flow meter, pressure gauge, and water level gauge, respectively; the ambient temperature and rainfall are obtained from meteorological data.

[0038] In the second aspect, a smart monitoring system for water conservancy project data includes:

[0039] processor;

[0040] The memory stores computer instructions for intelligent monitoring of water conservancy project data. When the computer instructions are executed by the processor, the system performs the aforementioned intelligent monitoring method for water conservancy project data.

[0041] The beneficial effects of this invention are as follows:

[0042] The solution of this invention, from both time and frequency domain perspectives, can accurately analyze water conservancy project data and improve the reliability of monitoring. Attached Figure Description

[0043] Figure 1 This illustration schematically shows a flowchart of the steps of an intelligent monitoring method for water conservancy project data in this embodiment;

[0044] Figure 2 The schematic diagram illustrates the structural block diagram of an intelligent monitoring system for water conservancy project data in this embodiment. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0046] This invention targets the monitoring of water conservancy project data of a target reservoir, that is, the monitoring of abnormal situations in water conservancy project data based on digital twin technology.

[0047] Digital twin technology is a cutting-edge technology that combines physical entities with virtual digital models. Through real-time data interaction and advanced analysis, it enables accurate simulation, optimization, and prediction of physical entities.

[0048] The technical architecture of digital twins includes a physical layer, a data layer, an analytics layer, a model layer, and an application layer. Since this technical architecture is existing technology, it will not be elaborated upon here.

[0049] The following section provides a detailed description of an intelligent monitoring method for water conservancy project data based on digital twin technology.

[0050] The digital twin system in this embodiment includes:

[0051] Physical layer: Deploy IoT sensors at monitoring points within the target reservoir area, such as water level gauges, piezometers, accelerometers, and flow meters at different monitoring points, to collect data from the corresponding monitoring points in real time, including water level, pressure, flow rate, vibration, etc.

[0052] Data layer: Sensor data is preprocessed (e.g., denoised) through edge computing nodes and then uploaded to the cloud-based digital twin platform.

[0053] Analysis layer: Performs data processing and analysis on the collected data to achieve anomaly detection and prediction.

[0054] Model layer: Construct digital twin models of water conservancy projects and use computer simulation technology to simulate and predict the behavior and performance of physical entities.

[0055] The digital twin model in this embodiment includes a geometric model (3D modeling) and a physical model (water level change model, hydrodynamic model, and dam structural model) to simulate water conservancy facilities and achieve dynamic simulation of actual facilities.

[0056] Application layer: Provides a visual interface, early warning notifications, and optimization suggestions (such as scheduling strategies and maintenance plans).

[0057] It should be noted that the solution in this embodiment focuses on the data processing and analysis of the collected data in the analysis layer. Therefore, the following uses the monitoring of water conservancy data of the target water area as an example to introduce the data processing and analysis in the analysis layer.

[0058] Specifically, such as Figure 1 As shown in this embodiment, a method for intelligent monitoring of water conservancy project data includes the following steps:

[0059] Step S1: Obtain multiple historical data points within the historical time period and multiple data points within the current time period within the area where the target reservoir is located.

[0060] Specifically, the data points at any given sampling time include data from multiple dimensions, such as dam leakage, ambient temperature, rainfall, water flow, water pressure, and water level.

[0061] The aforementioned historical period can be six months, one year, or longer, and the current period can be several consecutive days, specifically three days, five days, or more.

[0062] Data at different sampling times for the same dimension constitute a sequence. Among them, multiple dimensions collected in historical time periods are the same as the dimensions in the current time period.

[0063] The ambient temperature and rainfall data mentioned above were obtained from meteorological data sources.

[0064] Furthermore, the sequence composed of historical data and current data from multiple dimensions is standardized.

[0065] Step S2: Determine the first anomaly probability of the current data point.

[0066] In this embodiment, the data within the current time period is analyzed to determine the first probability of anomaly in the current water conservancy project data.

[0067] The process for obtaining the first anomaly probability is as follows:

[0068] Step S21: Construct the network model.

[0069] The network model in this embodiment is a fully connected network model (such as the FCN model).

[0070] Step S22: Obtain the training set and use the training set to train the network model to obtain the trained network model.

[0071] In this embodiment, the training set consists of multiple historical data points and their corresponding labels within the target reservoir's location over a historical period. The labels indicate whether each historical data point is normal or abnormal, with 1 representing normal and 0 representing abnormal.

[0072] The training process is as follows:

[0073] The training set is input into the network model for training, and the loss value is calculated using a loss function. The parameters of the network model are adjusted using the gradient descent algorithm until the loss value between the theoretical output value and the label is less than the threshold or the number of training iterations reaches the set number. At this point, training stops, and the trained network model is obtained.

[0074] The loss function described above is the mean squared error loss. Since the specific training process is based on existing technology, it will not be elaborated upon here.

[0075] Furthermore, in order to obtain a more accurate trained network model, it also includes: performing time alignment processing on the sequences of data in each dimension.

[0076] In water conservancy project data monitoring, when collecting data from different dimensions at the same time, there will be a certain delay between the data. For example, changes in water level usually precede changes in flow rate, changes in rainfall precede changes in reservoir water level (after rainfall, rainwater needs to enter the reservoir through processes such as surface runoff and groundwater infiltration), and changes in ambient temperature precede changes in water level, and so on. Therefore, when analyzing water conservancy project data, it is necessary to first calculate the time delay between different parameters.

[0077] In this embodiment, the cross-correlation function is used to calculate the delay time between different dimension parameters. Since the cross-correlation function describes the degree of correlation between the values ​​of different random signals at any two different times, the delay time between different dimension parameters in historical parameter data is determined by the cross-correlation function.

[0078] Specifically, taking the sequence of data in any dimension as the target sequence, calculate the cross-correlation coefficient of each target data in the target sequence with the data in the sequence of data in any other dimension. Select the time corresponding to the data with the largest cross-correlation coefficient as the delayed time. Thus, the difference between the delayed time and the time corresponding to the target data can be obtained, and this difference is the delay time.

[0079] At this point, we can obtain a sequence of time-aligned data across multiple dimensions, and then obtain the time-aligned historical data points and the current data points.

[0080] The purpose of obtaining the aforementioned delay time is to acquire data from different dimensions with high consistency in change, which facilitates accurate analysis of data from different dimensions in the future.

[0081] Step S23: Use the trained network model to perform anomaly detection on the current data points and obtain the first anomaly probability.

[0082] Furthermore, due to the issue of model accuracy, the product of model accuracy and the first anomaly probability can be used as the corrected first anomaly probability to avoid significant errors caused by inaccurate model accuracy.

[0083] Step S3: Determine the second anomaly probability of the current data point.

[0084] In this embodiment, considering that data with abnormal frequency components may still appear normal in the time domain (without obvious abnormalities), it is not comprehensive to only consider the abnormalities in the amplitude of each dimension of data points in the time domain. Therefore, in order to detect data with abnormal frequency components that are not obvious in the time domain, it is also necessary to analyze the frequency domain of each dimension of data.

[0085] Specifically, the process for the second anomaly probability is as follows:

[0086] First, the sequence of data in each dimension of each data point is processed by short-time Fourier transform to obtain a time-frequency diagram.

[0087] Specifically, the time-frequency plot of the data sequence in each dimension is obtained by short-time Fourier transform. The time-frequency plot uses frequency as the vertical axis and time as the horizontal axis to reflect the magnitude of the frequency amplitude of the data at different sampling times at different frequencies.

[0088] Secondly, based on the absolute value of the difference between the target sampling time and the sampling spectrum amplitude of the target frequency on the time-frequency graph and the previous adjacent time, as well as the mean value of the spectrum amplitude of the target frequency, anomaly indicators are obtained, and the anomaly indicators are normalized to obtain the anomaly probability.

[0089] The target sampling time mentioned above can be any sampling time.

[0090] In one embodiment, the anomaly index is the average of the global deviation and the local deviation.

[0091] , ;

[0092] in, For global deviation, This is a local deviation; The first on the time-frequency graph The average frequency amplitude of each frequency; The first on the time-frequency graph The spectral amplitude of a frequency at time n. For the first The spectral amplitude of a frequency at time n-1.

[0093] when This reflects the fluctuation of the spectral amplitude of the k-th frequency across all time frames, i.e., the overall deviation. The larger the value, the greater the deviation at time n. The impact of anomalies is greater at certain frequencies; It reflects the local fluctuations or abrupt changes of the k-th frequency in adjacent time frames. The larger the abrupt change, the more it proves that the k-th frequency is abnormal at time n.

[0094] In another embodiment, the anomaly index can also be obtained by comparing the similarity between the time-frequency plots of the sequences of data in each dimension and the standard time-frequency plots of the corresponding normal dimension sequences. The anomaly index is negatively correlated with the similarity; for example, the anomaly index can be calculated as 1 minus the similarity.

[0095] At this point, the abnormal indicators can be directly used as the second abnormal probability without the need for normalization.

[0096] The aforementioned normal dimensional data sequences refer to dimensional data under anomaly-free conditions, which can be obtained from historical data using big data technology. It should be noted that, since the sequences of each dimensional data are also related to seasonality, the normal dimensional data sequences obtained above need to be the same as or similar to the current dimensional data sequences.

[0097] Among them, the maximum anomaly probability among all dimensions at the current time is selected as the second anomaly probability.

[0098] In the above embodiments, the short-time Fourier transform can link time and frequency, thereby analyzing anomalies in the changes of each frequency over time. Furthermore, since the energy distribution varies across frequencies, anomalies at higher energy frequencies have a greater impact, thus attracting more attention to anomalies at high-energy frequencies.

[0099] The above normalization function can be adopted Functions, maximum and minimum value normalization methods.

[0100] Step S4: Integrate the first and second anomaly probabilities to obtain a comprehensive probability; if the comprehensive probability is greater than a set threshold, it indicates that there is an anomaly in the current data point, and staff need to be notified to take measures.

[0101] In this embodiment, the first and second anomaly probabilities of each data point are integrated together to obtain a comprehensive probability.

[0102] The process of obtaining the first weight of the first anomaly probability and the second weight of the second anomaly probability during integration is as follows:

[0103] First, the dimensional sequence corresponding to the second anomaly probability is decomposed using STL to obtain the sum of the seasonal components and residual components under each dimension.

[0104] STL decomposition is a commonly used method in time series decomposition algorithms. Based on LOESS (Locally Weighted Regression), it decomposes the time series into trend components, seasonal components, and residual terms.

[0105] Next, calculate the second weight, and the sum of the first weight and the second weight is 1.

[0106] In this embodiment, the second weight is negatively correlated with the sum of the seasonal component and the remainder.

[0107] Specifically, the second weight is ; Let be the absolute value of the sum of the seasonal component and the remainder of the k-th dimension, and exp() be an exponential function with the natural constant e as the base.

[0108] In time series analysis, the trend component represents the continuous and long-term change of the series mean. Therefore, if the absolute value of the sum of the seasonal component and the remainder is larger, it proves that the trend component is smaller, the second anomaly probability obtained from the time series is less important, and its weight should be smaller; conversely, the second weight should be larger.

[0109] In this embodiment, the threshold value can be 0.8. Of course, in other implementations, it can also be set based on experience.

[0110] Furthermore, in this embodiment, the water conservancy project data at future moments can also be predicted to determine whether the predicted data points are abnormal.

[0111] In one embodiment, an LSTM model can also be used for prediction. The input consists of the data point at the current time step and the data points before the current time step, to obtain the data point at the next time step, and to determine the combined probability of the data point at the next time step, so as to determine whether the data point at the next time step is an anomaly.

[0112] Before using the LSTM model for prediction, it is necessary to train the LSTM model. Since the training process is a current technology, it will not be described in detail here.

[0113] The solution of this invention can accurately analyze the anomalies in hydraulic engineering data at the current moment, thereby improving the security of water conservancy facility monitoring.

[0114] This invention also provides an intelligent monitoring system for water conservancy project data. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the intelligent monitoring method for water conservancy engineering data according to the present invention.

[0115] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0116] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0117] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0118] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for intelligent monitoring of water conservancy project data, characterized in that, include: Real-time data collection of multiple dimensions within the target water area at the current moment, forming the current data point; the multiple dimensions of data include dam leakage, ambient temperature, rainfall, water flow, water pressure, and water level; Input the current data point into the network model and output the corresponding first anomaly probability; Data from multiple dimensions at multiple time points prior to the current time point are obtained. Data from all sampling time points at the current time point in the same dimension constitute a sequence. Frequency domain analysis is performed on each dimension sequence to determine the anomaly probability of the current time point data in the corresponding dimension. The maximum anomaly probability among all dimensions at the current time point is taken as the second anomaly probability. The first anomaly probability and the second anomaly probability are integrated to obtain a comprehensive probability; If the overall probability is greater than a set threshold, then the data point at the current moment is abnormal, and staff need to be notified to take measures. The second weight of the integrated second anomaly probability is: the STL decomposition of the dimension sequence corresponding to the second anomaly probability is performed to obtain the sum of the seasonal component and the residual component; the second weight is negatively correlated with the sum of the seasonal component and the remainder term; wherein, the sum of the first weight and the second weight of the first anomaly probability is 1; The second weight is ; For the first The absolute value of the sum of the seasonal components and remainders of each dimension, exp() is an exponential function with the natural constant e as its base.

2. The intelligent monitoring method for water conservancy project data according to claim 1, characterized in that, The process of obtaining the anomaly probability is as follows: The time-frequency diagram is obtained by processing the sequences of each dimension using short-time Fourier transform. Based on the absolute value of the difference between the sampled spectral amplitude of any frequency at the target sampling time and the previous adjacent time on the time-frequency graph, and the mean spectral amplitude of any frequency, an anomaly index is obtained, and the anomaly index is normalized to obtain the anomaly probability; the target sampling time is any sampling time.

3. The intelligent monitoring method for water conservancy project data according to claim 2, characterized in that, The abnormal index is the average of the global deviation and the local deviation; ， ； in, For global deviation, This is a local deviation; The first on the time-frequency graph The average frequency amplitude of each frequency; The first on the time-frequency graph The spectral amplitude of a frequency at time n. For the first The spectral amplitude of a frequency at time n-1.

4. The intelligent monitoring method for water conservancy project data according to claim 1, characterized in that, The network model is a fully connected network model.

5. The intelligent monitoring method for water conservancy project data according to claim 4, characterized in that, The training process of the network model is as follows: Obtain a training set, which includes historical data points from multiple moments within a historical period; the labels are the normal or abnormal conditions of each data point at each moment. The network model is trained using the training set to obtain a trained network model.

6. The intelligent monitoring method for water conservancy project data according to claim 5, characterized in that, Also includes: Time alignment is performed on the dimensional data of historical data points at each time point to obtain aligned historical data points.

7. The intelligent monitoring method for water conservancy project data according to claim 2, characterized in that, The normalization process adopts Function or maximum / minimum value normalization method.

8. The intelligent monitoring method for water conservancy project data according to claim 1, characterized in that, The dam leakage, water flow, water pressure, and water level are measured using a piezometer, flow meter, pressure gauge, and water level gauge, respectively; the ambient temperature and rainfall are obtained from meteorological data.

9. A smart monitoring system for water conservancy project data, characterized in that, include: processor; A memory storing computer instructions for intelligent monitoring of water conservancy project data, wherein when the computer instructions are executed by the processor, the system performs an intelligent monitoring method for water conservancy project data according to any one of claims 1-8.

Images