Digital twin arch dam behavior detection blind source risk factor identification method based on ICA

Abstract

The invention discloses an ICA (independent component analysis)-based blind source risk factor identification method for digital twin arch dam behavior detection. A comprehensive monitoring and analysis system consisting of a Wavelet Scattering Network wavelet scattering network, a PCA (principal component analysis) dimension reduction module, an ICA independent component analysis module, an Apache Kafka data transmission module and a digital twin system is adopted. According to the method, the arch dam operation data collected in real time is processed and analyzed, potential blind source risk factors are identified, and the evaluation result is output. The system firstly performs multi-scale feature extraction by using a wavelet scattering network, then performs noise reduction and dimension reduction processing on extracted high-dimensional data by using PCA, and then identifies independent risk factor signals through an ICA module. And finally, transmitting the processed data to a digital twinborn platform through Apache Kafka so as to realize real-time monitoring and health assessment of the whole life cycle of the arch dam.

Description

Technical Field

[0001] The present invention belongs to the technical field of dam engineering safety monitoring, and in particular relates to a blind source risk factor identification method for digital twin arch dam performance detection based on ICA. Background Art

[0002] Digital twin technology enables the water conservancy and hydropower industry to closely connect the physical entity of the dam with the digital model, allowing managers to manage and maintain the dam more accurately and comprehensively. By acquiring and analyzing data in real time, digital twin technology can simulate the behavior of the dam under different conditions, thereby identifying potential risks in advance, preventing disasters, improving the efficiency of daily management, optimizing maintenance strategies, reducing maintenance costs, and extending the life of the dam. When responding to emergencies, digital twins can provide a virtual environment to simulate the effects of multiple emergency plans, quickly make the best decision, and protect the safety of the dam. In addition, digital twin technology also plays a vital role in the design and construction stage of the dam. It can verify the rationality of the design plan and ensure the quality and safety of the construction. Digital twin technology can not only better control the operating status of the dam, but also provide more scientific data support and decision-making basis for future development, change the traditional way of interaction between dam managers and dams, and enhance the engineering management capabilities of the water conservancy industry.

[0003] An arch dam is a type of dam that uses an arch structure to disperse water pressure. Its design focuses on how to effectively disperse hydraulic pressure using the curved shape rather than relying solely on the weight of the dam. This characteristic of using structural morphology to disperse pressure means that any factors that affect the stability of the dam, such as settlement of the foundation, aging of materials, or cracks in the dam, will directly affect the safety of the entire dam. The safety assessment of an arch dam must pay special attention to the contact surface between the dam and the foundation, the stress distribution inside the dam, and its behavior under extreme conditions such as earthquakes or floods. In addition, due to the structural characteristics of an arch dam, once a problem occurs, its impact range is correspondingly larger. Therefore, the monitoring and maintenance of an arch dam must not only ensure the integrity of its structure, but also consider the gradual cumulative effects in long-term operation. In short, the unique safety issue of an arch dam lies in its high dependence on structural morphology and foundation conditions, which requires special attention to these aspects in design, construction, and maintenance to ensure the long-term stability and safety of the dam.

[0004] Among the safety issues of arch dams, the uniqueness of blind source problems is that potential risks are difficult to detect and evaluate directly. Blind source problems refer to those hidden dangers that exist during the operation of the dam but are not easily identified by traditional monitoring methods, such as tiny cracks, material degradation, or small changes in the foundation. These problems may not be significant in the early stage, but over time, they will gradually accumulate and eventually pose a threat to the overall structural safety of the dam. Since arch dams rely on their structural form to disperse water pressure, once blind source problems occur, they are often difficult to detect in time through surface phenomena or conventional detection methods. Such hidden dangers will accumulate unknowingly and suddenly manifest themselves at a certain critical point, which may lead to serious consequences. Therefore, the uniqueness of blind source problems lies in their concealment and potential explosiveness, requiring us to strengthen multi-faceted monitoring and prevention of potential risks during the design, construction and operation of arch dams to ensure the long-term safe operation of the dam.

[0005] In summary, the technical problem to be solved by the present invention is that it is difficult for traditional arch dam monitoring methods to effectively identify hidden dangers that exist during the operation of the dam but are not easily detected directly, such as small cracks, material degradation or small changes in the foundation, etc. These blind source risks may not be significant in the early stage, but as time accumulates, they will pose a threat to the overall structural safety of the dam. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method for identifying blind source risk factors for digital twin arch dam performance detection based on ICA, and to construct a system consisting of Wavelet Scattering Network, blind source analysis PCA principal component analysis, blind source analysis ICA independent component analysis, Apache Kafka data transmission technology and digital twin technology. This system can not only realize multi-scale feature extraction and blind source risk identification of arch dam performance, but also monitor and evaluate the status of the arch dam in real time through a digital twin platform, effectively solving the problem that the arch dam digital twin technology in the prior art cannot accurately identify blind source risks.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: The blind source risk factor identification method for digital twin arch dam performance detection based on ICA includes the following steps: S1. Collect data through the real-time data acquisition system of the arch dam, input it into the wavelet scattering network, and extract multi-scale features of the data; S2. Use the principal component analysis method to pre-process the extracted data to remove noise and reduce dimension, so as to extract effective feature data; S3, use the independent component analysis module to perform independent component analysis on the preprocessed data to obtain key feature data; S4. Use the Apache Kafka distributed stream processing platform to transmit real-time data streams to the digital twin platform for subsequent evaluation and processing. Preferably, in step S1, data collection is performed by collecting physical information including stress, vibration, and temperature using a variety of sensors arranged inside and outside the dam structure; The wavelet scattering network is used to process the collected physical information including stress, vibration, and temperature; and edge computing devices are used to realize local collection, preliminary processing and wireless transmission of sensor data to reduce data delay and improve collection efficiency.

[0008] Preferably, the wavelet scattering network uses the wavelet scattering transform as the feature extraction layer to decompose the input signal or image into multiple invariant and stable feature maps, automatically extract invariant features, capture local details and overall trends of the monitoring signal, especially for the detection of sudden events and small deformations. The mother wavelet is a basic function that generates a series of sub-wavelets by scaling and shifting. : ; In the formula, is the scale parameter, controlling the wavelet width. is the translation parameter controlling the position of the wavelet, Indicates time, is an average operator. Wavelet is similar to the biological waveform of the human visual cortex and cochlea in response to stimulation, and the result of wavelet scattering network calculation retains high-frequency information. This makes it very stable to deformation and invariant to translation, making it a good feature extractor for classification. These characteristics of the wavelet transform network enable it to better analyze the acoustic and optical information from the arch dam.

[0009] Preferably, the wavelet scattering network includes a layer modulus operation layer, which is used to capture the local amplitude information of the signal and calculate the amplitude (absolute value) of each wavelet coefficient to detect the intensity of the signal change. The formula is: ; Signal With wavelet function The absolute value of the convolution result at a certain scale and position is taken. The modular operation is used to capture the local amplitude information of the signal, making the signal characteristics insensitive to translation, thereby enhancing its translation invariance.

[0010] The formula for low-pass filtering is: ; The low-pass filter function is used to remove the high-frequency components in the signal and retain only the low-frequency part (that is, the overall trend of the signal). Common low-pass filters include Gaussian filters. It is the signal after low-pass filtering, which indicates the low-frequency part of the signal and represents the global characteristics of the signal.

[0011] Downsample the data: ; Indicates the downsampling step size.

[0012] For the multi-level recursive feature extraction formula: ; Represents the input signal. ∗ is the convolution operator, which represents the convolution of the signal and the wavelet function. Represents wavelet functions of different scales and positions. Features at different scales are extracted through multiple convolutions. |⋅| is a modular operation, which is used to take absolute values ​​and retain amplitude information. Through multi-level processing, the wavelet transform network can extract multi-scale features of signals, which is particularly suitable for capturing local details and global trends of signals.

[0013] Preferably, the principal component analysis method is used for noise reduction preprocessing. The high-dimensional data is reduced in dimension while retaining the characteristic information in the data. The object data is projected into the subspace composed of the selected principal components to obtain the noise-reduced data: ; In the formula, Represents the denoised data matrix. represents the original data matrix. is a matrix containing the first k eigenvectors, T represents the transpose, for The transposed matrix of .

[0014] Preferably, the independent component analysis module uses ICA independent component analysis for the optimized feature data, using the feature of ICA to separate independent components from blind sources. ICA separates independent feature components from mixed signals to complete identification by assuming that the original signals are statistically independent.

[0015] Preferably, ICA independent component analysis includes whitening, independent component separation and maximizing non-Gaussianity. , these signals have several independent original signals, This process can be obtained by linear mixing using a mixing matrix express: ; In the formula, is the vector of observed signals. is a vector of independent source signals, which are statistically independent. It is an n×m mixing matrix, which represents the linear mixing process from source signal to observation signal.

[0016] For the signal after PCA preprocessing Whitening is performed so that the covariance matrix of the signal becomes the unit matrix (that is, the components of the signal are uncorrelated and have unit differences): ; In the formula, is the whitening matrix, satisfying , D is the eigenvalue diagonal matrix of the oblique variance matrix.

[0017] The Fast ICA iterative algorithm is used on the whitened signal to maximize the non-Gaussianity through a certain number of iterations. The iterative equation for achieving the maximum non-Gaussianization is: ; In the formula, the symbol E[⋅] represents the expected value, which is used to calculate the average value on the given data to optimize the algorithm. It is a nonlinear function used to enhance non-Gaussianity. Common ones include and .

[0018] This process continues iteratively until Convergence, through iteration, the independent components after maximizing non-Gaussianity are obtained, and the separated target independent signal source is obtained : ; In the formula, is the separation matrix obtained by iterative optimization, is an independent component matrix, each column represents an independent component. Using nonlinear functions in the iterative process and its derivative To adjust the separation matrix The isolated independent signal , Preferably, in step S4, to solve the problem of independent signal The transmission problem between the digital twin and the independent feature signal separated from ICA Apache Kafka is used as a real-time data stream processing and published to the topic through the Kafka producer. The topic can be divided into multiple partitions, each of which stores a part of the data. This allows data to be processed in parallel, where the proxy point is responsible for transmitting data from the producer to the consumer digital twin system. Kafka also ensures high availability and fault tolerance of data through data replication and automatic failover, providing a safe, automated and highly resilient method for the arch dam's digital twin system to conduct dam health assessment (stress analysis, structural deformation). The system combines with the arch dam's digital twin system to achieve real-time monitoring of key parameters such as dam stress, deformation, and seepage, and can quickly respond to abnormal situations and identify potential risks in a timely manner. The early warning function is realized through central decision-making. When a risk signal is detected, an alarm is automatically triggered, and a detailed risk analysis report and emergency treatment suggestions are generated. The report is published to the public platform through a portable terminal, which reduces manual intervention in actual operation and realizes a fully automated process from data collection, processing, analysis, evaluation to decision-making.

[0019] The ICA-based digital twin arch dam performance detection blind source risk factor identification system adopts the ICA-based digital twin arch dam performance detection blind source risk factor identification method.

[0020] The present invention can achieve the following beneficial effects: 1. The present invention uses the ICA algorithm to perform blind source separation on multi-dimensional monitoring data, which can effectively extract independent risk sources in mixed data, overcoming the limitation that traditional methods are difficult to identify blind sources. Fusion of digital twins and ICA: The present invention combines digital twin technology with the ICA algorithm, performs blind source separation based on the real-time status information provided by the digital twin model, and realizes more accurate risk factor identification. Adaptive risk assessment: Combining the dam's operating data and independent component analysis results, an adaptive risk assessment model is constructed, which can dynamically adjust the assessment strategy to adapt to changes in different environments and working conditions.

[0021] 2. By integrating the Wavelet Scattering Network, PCA dimensionality reduction module and ICA independent component analysis module, the multi-scale feature extraction of arch dam performance and the accurate identification of blind source risks are realized. The system effectively identifies potential structural risk factors by collecting and processing the operation data of the arch dam in real time, and transmits the processed data to the digital twin platform through Apache Kafka data transmission technology, realizing real-time monitoring and health assessment of the entire life cycle of the arch dam. This integrated method ensures the accurate identification of blind source risks and significantly improves the reliability and effectiveness of arch dam performance detection.

[0022] 3. By optimizing the system architecture and technical processes, the safety and management efficiency of arch dam operation are effectively improved. Utilizing the real-time monitoring and simulation capabilities of digital twin technology, the system can quickly respond to and identify abnormal situations, and provide accurate early warning and decision support. Compared with the existing technology, the present invention can not only accurately control and detect the potential blind source risks of arch dams, but also reduce operating and maintenance costs, improve the intelligent level of management, and provide a more economical and efficient solution for the long-term stability and safety of arch dams. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the present invention; Figure 2 This is a diagram showing the effect of implementing this embodiment. DETAILED DESCRIPTION

[0024] The preferred solution is Figure 1-2 As shown, the blind source risk factor identification method for digital twin arch dam performance detection based on ICA is used for real-time status monitoring, blind source risk identification and health assessment of arch dams, so as to improve the safety of arch dam operation and management efficiency. The present invention adopts a comprehensive monitoring and analysis system composed of Wavelet Scattering Network, PCA dimension reduction module, ICA independent component analysis module, Apache Kafka data transmission module and digital twin system. The system processes and analyzes the real-time collected arch dam operation data, identifies potential blind source risk factors, and outputs the evaluation results. The system first uses the wavelet scattering network to extract multi-scale features, then uses PCA to reduce noise and dimension of the extracted high-dimensional data, and then identifies independent risk factor signals through the ICA module. Finally, the processed data is transmitted to the digital twin platform through Apache Kafka to realize real-time monitoring and health assessment of the entire life cycle of the arch dam. The specific steps include: S1 collects data through the real-time data acquisition system of the arch dam and inputs it into the Wavelet Scattering Network to extract multi-scale features of the data; S2 uses PCA (principal component analysis) to pre-process the extracted data for denoising and dimensionality reduction, and further extracts effective feature data; S3 uses the ICA (independent component analysis) module to perform independent component analysis on the preprocessed data to obtain key feature data; S4 will transmit real-time data streams to the digital twin platform via Apache Kafka for subsequent evaluation and processing.

[0025] The present invention integrates a variety of advanced technologies to achieve monitoring and management of the entire life cycle of the arch dam. The digital twin and its extended system can identify potential structural risks in real time, optimize maintenance strategies, reduce operating costs, and improve the overall safety and stability of the dam through intelligent means. It can achieve accurate identification and real-time monitoring of blind source risks of arch dams, and improve the reliability and effectiveness of arch dam performance detection.

[0026] Step S1 uses the wavelet scattering network to process physical information such as stress, vibration, temperature, etc. collected by various sensors arranged inside and outside the dam structure, and uses edge computing devices to realize local collection, preliminary processing and wireless transmission of sensor data to reduce data delay and improve collection efficiency. Utilize its wavelet transform, multi-scale analysis capabilities and hierarchical feature extraction ideas of deep learning. These include wavelet transform, layer module operation layer, low-pass filtering and downsampling layer, and multi-level recursive feature extraction. Using wavelet scattering transform as the feature extraction layer, the input signal or image is decomposed into multiple invariant and stable feature maps, and the invariant features are automatically extracted to capture the local details and overall trends of the monitoring signal, especially the detection of emergencies and small deformations. Among them, the mother wavelet is a basic function that generates a series of sub-wavelets by scaling and shifting. : ; In the formula, is the scale parameter, controlling the wavelet width. is the translation parameter controlling the position of the wavelet, Indicates time, is an average operator. Wavelet is similar to the biological waveform of the human visual cortex and cochlea in response to stimulation, and the result of wavelet scattering network calculation retains high-frequency information. This makes it very stable to deformation and invariant to translation, making it a good feature extractor for classification. These characteristics of the wavelet transform network enable it to better analyze the acoustic and optical information from the arch dam.

[0027] The layer modulus operation layer is used to capture the local amplitude information of the signal and calculate the amplitude (absolute value) of each wavelet coefficient to detect the intensity of signal changes. The formula is: ; Signal With wavelet function The absolute value of the convolution result at a certain scale and position is taken. The modular operation is used to capture the local amplitude information of the signal, making the signal characteristics insensitive to translation, thereby enhancing its translation invariance.

[0028] The formula for low-pass filtering is: ; The low-pass filter function is used to remove the high-frequency components in the signal and retain only the low-frequency part (that is, the overall trend of the signal). Common low-pass filters include Gaussian filters. It is the signal after low-pass filtering, which indicates the low-frequency part of the signal and represents the global characteristics of the signal.

[0029] Downsample the data: ; Indicates the downsampling step size.

[0030] For the multi-level recursive feature extraction formula: ; Represents the input signal. ∗ is the convolution operator, which represents the convolution of the signal and the wavelet function. Represents wavelet functions of different scales and positions. Features at different scales are extracted through multiple convolutions. |⋅| is a modular operation, which is used to take absolute values ​​and retain amplitude information. Through multi-level processing, the wavelet transform network can extract multi-scale features of signals, which is particularly suitable for capturing local details and global trends of signals. Based on this, it has unique advantages as the information extraction layer of the arch dam digital twin system.

[0031] Table 1 Wavelet reference values ​​in arch dam digital twin monitoring Parameter Type Mother wavelet (ψ) Scale parameter (a) Translation parameter (b) Number of wavelet layers (L) Wavelet transform accuracy Parameter Value 1-10 Hz 1-8 1-6 3 0.1-1Hz Parameter Type Filter length (K) Convolution kernel size Sampling frequency Downsampling ratio Feature Extraction Window Parameter Value 6-10 5×5 100-1000 Hz 2 / 4 128 points PCA and ICA are both blind source separation technologies. The result data processed by Wavelet Scattering Network is usually high-dimensional. In order to facilitate ICA data processing and reduce data redundancy, PCA is selected for noise reduction preprocessing. High-dimensional data is reduced to low dimensionality while retaining the characteristic information in the data. The object data is projected into the subspace composed of the selected principal components to obtain the denoised data: ; In the formula, Represents the denoised data matrix. represents the original data matrix. is a matrix containing the first k eigenvectors, T represents the transpose, for The transposed matrix of .

[0032] Table 2 Matrix dimensions in arch dam digital twin monitoring Parameter Type Number of sampling points (N) Number of sensors (M) Original data matrix dimensions Feature map size (L) Number of scales of wavelet transform (S) Wavelet feature extraction matrix dimensions Parameter Value 60,000 50 60,000×50 30,000 3 30,000×50×3 Parameter Type The number of principal components retained by PCA (K) Matrix dimensions after PCA dimensionality reduction ICA independent component number (C) Matrix dimensions after ICA processing Batch size for each data transfer (B) Data transfer matrix dimensions Parameter Value 10 6,000×50 8 60,000×8 1,000 1,000×50 In the step S3, ICA independent component analysis is used for the optimized feature data, and the feature of ICA is used to separate independent components from blind sources. ICA separates independent feature components from mixed signals by assuming that the original signals are statistically independent. Its main steps include whitening, independent component separation and maximizing non-Gaussianity. , these signals have several independent original signals, This process can be obtained by linear mixing using a mixing matrix express: ; In the formula, is the vector of observed signals. is a vector of independent source signals, which are statistically independent. It is an n×m mixing matrix, which represents the linear mixing process from source signal to observation signal.

[0033] For the signal after PCA preprocessing Whitening is performed so that the covariance matrix of the signal becomes the unit matrix (that is, the components of the signal are uncorrelated and have unit differences): ; In the formula, is the whitening matrix, satisfying , D is the eigenvalue diagonal matrix of the oblique variance matrix.

[0034] The Fast ICA iterative algorithm is used on the whitened signal to maximize the non-Gaussianity through a certain number of iterations. The iterative equation for achieving the maximum non-Gaussianization is: ; In the formula, the symbol E[⋅] represents the expected value, which is used to calculate the average value on the given data to optimize the algorithm. It is a nonlinear function used to enhance non-Gaussianity. Common ones include and .

[0035] This process continues iteratively until Convergence, through iteration, the independent components after maximizing non-Gaussianity are obtained, and the separated target independent signal source is obtained : ; In the formula, is the separation matrix obtained by iterative optimization, is an independent component matrix, each column represents an independent component. Using nonlinear functions in the iterative process and its derivative To adjust the separation matrix .

[0036] This results in the separation of independent signals , to solve the independent signal To solve the transmission problem between the digital twin and the real-time CNN, we proceed as in step S4. For the independent feature signals separated from ICA, Apache Kafka is used as a real-time data stream processing and published to the topic through the Kafka producer. The topic can be divided into multiple partitions, each of which stores a part of the data. This allows data to be processed in parallel, where the proxy point is responsible for transmitting data from the producer to the consumer digital twin system. Kafka also ensures high availability and fault tolerance of data through data replication and automatic failover, providing a safe, automated and highly resilient method for the arch dam's digital twin system to conduct dam health assessment (stress analysis, structural deformation). The system combines with the arch dam's digital twin system to achieve real-time monitoring of key parameters such as dam stress, deformation, and seepage, and can quickly respond to abnormal situations and identify potential risks in a timely manner. The early warning function is realized through central decision-making. When a risk signal is detected, an alarm is automatically triggered, and a detailed risk analysis report and emergency treatment suggestions are generated. The report can be published to the public platform through a portable terminal, which reduces manual intervention in actual operation and realizes a fully automated process from data collection, processing, analysis, evaluation to decision-making, significantly improving the efficiency of dam monitoring and management. Adapt to changing environmental conditions through digital analysis of digital twins, ensure personnel safety, and improve the ability to deal with environmental factors, achieving efficiency, safety, intelligence, and convenience.

[0037] A blind source risk factor identification system for digital twin arch dam performance detection based on ICA adopts the blind source risk factor identification method for digital twin arch dam performance detection based on ICA. It also includes: The data acquisition module includes a variety of sensors arranged inside and outside the arch dam structure to collect real-time physical parameters such as stress, vibration, temperature, etc. These data are initially processed by edge computing devices, including data filtering, noise reduction and compression, and then transmitted to the central server for further analysis.

[0038] The wavelet scattering network is used to extract multi-scale features from the data after preliminary processing. The data is transformed by the mother wavelet function to capture the feature information at different scales, especially the slight deformation of the high-frequency part and the overall trend of the low-frequency part. Specifically, the wavelet scattering network includes multiple layers of convolution blocks and recursive feature extraction layers. These hierarchical structures can extract local and global features of the signal while maintaining high-frequency information.

[0039] The PCA module reduces the dimensionality and denoises the high-dimensional data extracted by the wavelet scattering network. PCA calculates the covariance matrix of the data, selects the first k principal components with larger eigenvalues, and projects the high-dimensional data into a low-dimensional space, retaining key features while reducing redundant information. This step helps reduce the complexity of subsequent ICA processing and improves analysis efficiency.

[0040] The ICA module uses the FastICA algorithm to perform independent component analysis on the data after PCA dimension reduction. Through whitening and maximizing non-Gaussianity, potential blind source risk factor signals are separated. This process includes multiple iterative optimizations until the separation matrix converges, and finally independent signal sources are obtained, which are used to identify potential structural problems of arch dams.

[0041] The Apache Kafka data transmission module is responsible for transmitting independent component signals to the digital twin platform in real time. The high concurrency and fault tolerance of Kafka ensure stable data transmission. Kafka publishes data to the specified topic and optimizes the efficiency of data transmission through partitioned parallel processing.

[0042] The digital twin system receives and processes data streams from Apache Kafka to perform real-time simulation and health assessment of the arch dam's operating status. The digital twin system receives real-time monitoring data of the arch dam from the specified Topic through Kafka consumers. These data come from various sensors such as stress, vibration, temperature, deformation, etc.

[0043] The system can simulate the dynamic operation of the dam based on the 3D geometric model of the arch dam and physical models such as stress distribution model, deformation model, seepage model, etc. The model can recalculate key parameters based on real-time sensor data.

[0044] Combining historical data with current status, the system can predict the health status of the dam in the future. For example, if the system detects that the dam has a tendency to continue to deform, structural failure may occur at some point in the future. Through the prediction model, the system can issue an early warning.

[0045] The system combines the structural model of the dam with real-time monitoring data to generate an assessment report on key parameters such as stress distribution and deformation trend. When an abnormal signal is detected, the system automatically triggers an early warning mechanism and generates a detailed risk analysis report and emergency response recommendations.

[0046] As more data accumulates, the system can optimize and adjust the digital twin model to make it more consistent with the actual situation. Model parameters are continuously updated and calibrated based on actual monitoring data to improve the accuracy of the simulation. Maintenance recommendations can be made based on the real-time status and health assessment results of the dam, such as whether local repairs are needed, whether structural reinforcement is needed, etc. At the same time, the system can optimize the operation plan and reduce unnecessary maintenance costs, ensuring the safety and operational efficiency of the dam.

[0047] Through the system architecture and algorithm optimization of the present invention, the blind source risk identification capability of the arch dam has been significantly improved. The system performs well in dealing with complex working conditions and potential risks, greatly improving the safety and management efficiency of the arch dam operation.

[0048] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. ICA-based blind source risk factor identification method for digital twin arch dam performance detection, characterized by The following steps are involved: S1. Collect data through the real-time data acquisition system of the arch dam, input it into the wavelet scattering network, and extract multi-scale features of the data; S2. Use the principal component analysis method to pre-process the extracted data to remove noise and reduce dimension, so as to extract effective feature data; S3, use the independent component analysis module to perform independent component analysis on the preprocessed data to obtain key feature data; S4. Use the Apache Kafka distributed stream processing platform to transmit real-time data streams to the digital twin platform for subsequent evaluation and processing.

2. According to claim 1, the method for identifying blind source risk factors of digital twin arch dam performance detection based on ICA is characterized by: In step S1, data is collected by various sensors arranged inside and outside the dam structure to obtain physical information including stress, vibration, and temperature; The wavelet scattering network is used to process the collected physical information including stress, vibration, and temperature; and edge computing devices are used to realize local collection, preliminary processing and wireless transmission of sensor data to reduce data delay and improve collection efficiency.

3. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 2 is characterized by: The wavelet scattering network uses the wavelet scattering transform as the feature extraction layer to decompose the input signal or image into multiple invariant and stable feature maps, automatically extract invariant features, capture the local details and overall trends of the monitoring signal, and detect sudden events and small deformations at the same time; the mother wavelet is a basic function that generates a series of daughter wavelets through scaling and shifting. : ； In the formula, is the scale parameter, controlling the wavelet width; is the translation parameter controlling the position of the wavelet, Indicates time, It is an average operator; the wavelet is similar to the biological waveform of the human visual cortex and cochlea in response to stimulation, and the result of the wavelet scattering network calculation retains high-frequency information; it has good stability to deformation and is translation-invariant, making it a good feature extractor for classification; These characteristics of the wavelet transform network enable it to better analyze the acoustic and optical information from the arch dam.

4. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 1 is characterized by: The wavelet scattering network includes a layer modulus operation layer, which is used to capture the local amplitude information of the signal and calculate the amplitude of each wavelet coefficient to detect the intensity of the signal change; the formula is: ； Signal With wavelet function The absolute value of the convolution result at a certain scale and position is taken; the modular operation is used to capture the local amplitude information of the signal, making the signal characteristics insensitive to translation, thereby enhancing its translation invariance; The formula for low-pass filtering is: ； The low-pass filter function is used to remove the high-frequency components in the signal and only retain the low-frequency part; It is the signal after low-pass filtering, indicating the low-frequency part of the signal and representing the global characteristics of the signal; Downsample the data: ； Indicates the downsampling step size; For the multi-level recursive feature extraction formula: ； represents the input signal; ∗ is the convolution operator, which represents the convolution of the signal and the wavelet function; Represents wavelet functions of different scales and positions; extracts features at different scales through multiple convolutions; |⋅| is a modular operation, which is used to take absolute values ​​and retain amplitude information; through multi-level processing, the wavelet transform network can extract multi-scale features of the signal.

5. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 1 is characterized by: The principal component analysis method is used for noise reduction preprocessing; it reduces the dimensionality of high-dimensional data while retaining the characteristic information in the data; it projects the object data into the subspace composed of the selected principal components to obtain the noise-reduced data: ； In the formula, represents the data matrix after denoising; represents the original data matrix; is a matrix containing the first k eigenvectors, T represents the transpose, for The transposed matrix of .

6. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 1 is characterized by: The independent component analysis module uses ICA independent component analysis for the optimized feature data, taking advantage of the ability of ICA to separate independent components from blind sources; ICA separates independent feature components from mixed signals to complete identification by assuming that the original signals are statistically independent.

7. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 6 is characterized by: ICA independent component analysis includes whitening, independent component separation and maximizing non-Gaussianity; for a set of preprocessed observation signals , the observed signal There are several independent original signals. This process is obtained by linear mixing using a mixing matrix express: ； In the formula, is the vector of observed signals; is the vector of independent source signals; is an n×m mixing matrix, which represents the linear mixing process from source signal to observation signal; For the signal after PCA preprocessing Perform whitening so that the covariance matrix of the signal becomes the identity matrix: ； In the formula, is the whitening matrix, satisfying , D is the eigenvalue diagonal matrix of the oblique variance matrix; The Fast ICA iterative algorithm is used on the whitened signal to maximize the non-Gaussianity through a certain number of iterations. The iterative equation for achieving the maximum non-Gaussianization is: ； In the formula, the symbol E[⋅] represents the expected value, which is used to calculate the average value on the given data to optimize the algorithm. It is a nonlinear function used to enhance non-Gaussianity; common ones include and ; This process continues iteratively until Convergence, through iteration, the independent components after maximizing non-Gaussianity are obtained, and the separated target independent signal source is obtained : ； In the formula, is the separation matrix obtained by iterative optimization, is an independent component matrix, each column represents an independent component; Using nonlinear functions in the iterative process and its derivative To adjust the separation matrix ; The separated independent signals are obtained .

8. The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to claim 6 is characterized by: In step S4, to solve the problem of independent signal The transmission problem between the digital twin and the independent feature signal separated from ICA Processed by Apache Kafka as a real-time data stream and published to a Topic through a Kafka producer; A Topic can be divided into multiple partitions, each of which stores a portion of the data; this allows data to be processed in parallel, with the proxy point responsible for transmitting data from the producer to the consumer digital twin system; and Kafka ensures high availability and fault tolerance of data through data replication and automatic failover; the system combines with the digital twin system of the arch dam to achieve real-time monitoring of key parameters such as dam stress, deformation, and seepage, and can quickly respond to abnormal situations and identify potential risks in a timely manner; the early warning function is realized through central decision-making, and an alarm is automatically triggered when a risk signal is detected, and a detailed risk analysis report and emergency response recommendations are generated, and the report is published to the public platform through a portable terminal, reducing manual intervention in actual operations and realizing a fully automated process from data collection, processing, analysis, evaluation to decision-making.

9. The ICA-based digital twin arch dam performance detection blind source risk factor identification system is characterized by: The blind source risk factor identification method for digital twin arch dam performance detection based on ICA according to any one of claims 1 to 8 is adopted.

Images