CN-121998421-A - Digital twinning-based hydraulic engineering intelligent monitoring method and system

CN121998421ACN 121998421 ACN121998421 ACN 121998421ACN-121998421-A

Abstract

The invention discloses a digital twinning-based hydraulic engineering intelligent monitoring method and system, which are characterized by comprising the steps of collecting initial multi-source heterogeneous data, establishing a dual-channel deep learning network model, optimizing super parameters of the dual-channel deep learning network model by utilizing a GA genetic algorithm to obtain a target dual-channel deep learning network model, inputting the initial multi-source heterogeneous data into the target dual-channel deep learning network model to output a hydraulic engineering construction risk index, constructing a hydraulic engineering digital twinning model based on BIM+GIS, inputting the hydraulic engineering construction risk index into the hydraulic engineering digital twinning model to perform simulation operation, and outputting a construction calibration report. The safety and the fine management level of hydraulic engineering construction are effectively improved, and the accident rate of construction risks is reduced.

Inventors

ZHU JIANHAI
CAO HAIDI
HU BO
YU YANG
MIAO XIN
SHAN HONGYU
AI RUQUAN
LI XIANGHONG
CHEN NAN

Assignees

莱格科技服务有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. The intelligent hydraulic engineering monitoring method based on digital twinning is characterized by comprising the following steps of: acquiring time sequence sensor data, video image data, environment data and historical working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, and preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data; Establishing a two-channel deep learning network model, processing time sequence data by utilizing an improved LSTM-transducer mixed network, processing image data by adopting YOLOv & lt7+ & gt ResNet & lt 50 & gt double-branch network, and carrying out weighted fusion on the two-channel characteristics by a DWA dynamic weighted attention mechanism; Optimizing super parameters of the two-channel deep learning network model by utilizing a GA genetic algorithm to obtain a target two-channel deep learning network model, inputting the initial multi-source heterogeneous data into the target two-channel deep learning network model, and outputting a hydraulic engineering construction risk index; And constructing a hydraulic engineering digital twin model based on the BIM+GIS, inputting the hydraulic engineering construction risk index into the hydraulic engineering digital twin model for simulation operation, and outputting a construction calibration report.
2. The intelligent monitoring method for hydraulic engineering based on digital twinning according to claim 1, wherein the acquiring time sequence sensor data, video image data, environment data and history working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data comprises: acquiring time sequence sensor data, video image data, environment data and historical working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data; Detecting abnormal values of time sequence sensor data by using a3 sigma criterion, judging the data exceeding the mean value by +/-3 times of standard deviation as the abnormal values, replacing the abnormal values by using a linear interpolation method of the data at adjacent moments, and filling the missing values by using a mean filling method.
3. The intelligent monitoring method for hydraulic engineering based on digital twinning according to claim 1, wherein the acquiring time sequence sensor data, video image data, environment data and history working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data comprises: Cleaning video image data through an image quality evaluation algorithm, eliminating repeated records of environmental data and historical working condition data, and correcting error data through logic verification; Converting the time sequence data of different dimensions into standard normal distribution data with the mean value of 0 and the standard deviation of 1 by adopting a Z-score standardization method, normalizing the pixel value of video image data to a [0,1] interval, and carrying out numerical conversion on classified historical working condition data by adopting independent heat coding; and (3) aligning acquisition time of the video image data and the environment data by taking a time stamp of the time sequence sensor data as a reference to obtain initial multi-source heterogeneous data.
4. The intelligent monitoring method for hydraulic engineering based on digital twinning according to claim 1, wherein the establishing a dual-channel deep learning network model, processing time sequence data by using an improved LSTM-transducer mixed network, processing image data by using YOLOv7+ ResNet50 dual-branch network, and performing weighted fusion of dual-channel characteristics by using a DWA dynamic weighted attention mechanism comprises: On the basis of forgetting gate, input gate and output gate of traditional LSTM, adding time sequence attention gate, and enhancing time sequence feature extraction of key time node by calculating attention weights of different time step data; The sequence of timing features output by the LSTM layer is input into a transducer encoder that includes a plurality of attention heads, each having dimensions of 64, capturing long-range dependencies between timing features by a multi-head attention mechanism.
5. The intelligent monitoring method for hydraulic engineering based on digital twinning according to claim 1, wherein the establishing a dual-channel deep learning network model, processing time sequence data by using an improved LSTM-transducer mixed network, processing image data by using YOLOv7+ ResNet50 dual-branch network, and performing weighted fusion of dual-channel characteristics by using a DWA dynamic weighted attention mechanism comprises: The target detection branch is based on YOLOv networks, 1 small target detection head is added, the size of an anchor frame is optimized, the characteristic enhancement branch adopts ResNet networks, and residual connection is introduced to solve the problem of gradient disappearance of a deep network; inputting the time sequence feature vector and the image feature vector into a module through a DWA dynamic weighted attention mechanism, and determining the initial weight of a channel by calculating the information entropy and mutual information of the two features; Based on the initial weight, dynamic adjustment is carried out through an attention mechanism, time sequence feature weights and image feature weights are obtained through calculation, and a double-channel feature output fusion feature vector is fused in a weighted summation mode.
6. The intelligent monitoring method for hydraulic engineering based on digital twin according to claim 1, wherein optimizing the super parameters of the two-channel deep learning network model by using a GA genetic algorithm to obtain a target two-channel deep learning network model, inputting the initial multi-source heterogeneous data into the target two-channel deep learning network model, and outputting a hydraulic engineering construction risk index, comprises: the super parameters comprise the number of hidden layer units, the dropout rate and the number of transducer attention heads of the improved LSTM-transducer network, the learning rate, the batch size and the anchor frame size parameters of the YOLOv & lt+ ResNet50 & gt double-branch network and the weight adjustment coefficient of the DWA attention mechanism; and constructing an adaptability function by taking the minimum hydraulic engineering construction risk prediction error as an optimization target, executing the selection, crossing and mutation operations of a genetic algorithm, and iteratively outputting an optimal super-parameter combination.
7. The intelligent monitoring method for hydraulic engineering based on digital twinning according to claim 1, wherein the construction of the hydraulic engineering digital twinning model based on the bim+gis, inputting the hydraulic engineering construction risk index into the hydraulic engineering digital twinning model for analog operation, and outputting a construction calibration report, comprises: And accessing the hydraulic engineering construction risk index into the digital twin model in real time, establishing the association relation between the hydraulic engineering construction risk index and the corresponding monitoring area and the components in the model, performing simulation operation to obtain an operation result, and compiling a construction calibration report based on the simulation operation result.
8. Digital twinning-based hydraulic engineering intelligent monitoring system is characterized in that the hydraulic engineering intelligent monitoring system comprises the following modules: The multi-source data acquisition module is used for acquiring time sequence sensor data, video image data, environment data and history working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, and preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data; The dual-channel model building module is used for building a dual-channel deep learning network model, processing time sequence data by utilizing an improved LSTM-transducer mixed network, processing image data by adopting YOLOv7+ ResNet dual-branch network, and carrying out weighted fusion on dual-channel characteristics by a DWA dynamic weighted attention mechanism; The construction risk prediction module is used for optimizing super parameters of the two-channel deep learning network model by utilizing a GA genetic algorithm to obtain a target two-channel deep learning network model, inputting the initial multi-source heterogeneous data into the target two-channel deep learning network model, and outputting a hydraulic engineering construction risk index; The digital twin simulation module is used for constructing a hydraulic engineering digital twin model based on the BIM+GIS, inputting the hydraulic engineering construction risk index into the hydraulic engineering digital twin model for simulation operation, and outputting a construction calibration report.
9. The intelligent monitoring system for hydraulic engineering based on digital twinning according to claim 8, wherein the construction risk prediction module comprises the following submodules: The parameter submodule is used for improving the hidden layer unit number, the dropout rate and the transducer attention head number of the LSTM-transducer network, the learning rate, the batch size and the anchor frame size parameter of the YOLOv7+ ResNet double-branch network and the weight adjustment coefficient of the DWA attention mechanism; And the optimization sub-module is used for constructing a fitness function by taking the minimum hydraulic engineering construction risk prediction error as an optimization target, executing the selection, crossing and mutation operations of a genetic algorithm and iteratively outputting an optimal super-parameter combination.
10. The intelligent monitoring system for hydraulic engineering based on digital twinning according to claim 8, wherein the construction risk prediction module comprises the following submodules: The simulation sub-module is used for accessing the hydraulic engineering construction risk index into the digital twin model in real time, establishing the association relation between the hydraulic engineering construction risk index and the corresponding monitoring area and the components in the model, performing simulation operation to obtain an operation result, and compiling a construction calibration report based on the simulation operation result.

Description

Digital twinning-based hydraulic engineering intelligent monitoring method and system Technical Field The invention relates to the technical field of hydraulic engineering monitoring, in particular to a digital twinning-based hydraulic engineering intelligent monitoring method and system. Background The hydraulic engineering is used as an important infrastructure of national economy, and the safety and stability of the construction process are directly related to engineering quality and life and property safety of surrounding masses. The current manual inspection is low in efficiency and strong in subjectivity, all-weather and full-coverage monitoring is difficult to realize, potential safety hazards exist in inspection of dangerous areas such as deep foundation pits and high slopes, fine structural deformation and risk symptoms cannot be captured timely, the single sensor is single in monitoring data dimension, the association relation between construction working conditions, environmental changes and engineering structural states cannot be comprehensively reflected, and the data utilization rate is low. Meanwhile, the existing monitoring technology lacks an effective multi-source data fusion means, the characteristic information of time sequence data and image data is difficult to complement, and risk assessment accuracy is insufficient. In addition, the traditional monitoring has poor cooperativity with engineering design and construction management, visual presentation and dynamic simulation prediction of risks cannot be realized, and accurate and real-time support for construction decisions is difficult to provide. Disclosure of Invention The invention aims to solve the problems, and designs a hydraulic engineering intelligent monitoring method and system based on digital twinning. The technical scheme for achieving the purpose is that in the intelligent monitoring method of the hydraulic engineering based on digital twinning, the intelligent monitoring method of the hydraulic engineering comprises the following steps: acquiring time sequence sensor data, video image data, environment data and historical working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, and preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data; Establishing a two-channel deep learning network model, processing time sequence data by utilizing an improved LSTM-transducer mixed network, processing image data by adopting YOLOv & lt7+ & gt ResNet & lt 50 & gt double-branch network, and carrying out weighted fusion on the two-channel characteristics by a DWA dynamic weighted attention mechanism; Optimizing super parameters of the two-channel deep learning network model by utilizing a GA genetic algorithm to obtain a target two-channel deep learning network model, inputting the initial multi-source heterogeneous data into the target two-channel deep learning network model, and outputting a hydraulic engineering construction risk index; And constructing a hydraulic engineering digital twin model based on the BIM+GIS, inputting the hydraulic engineering construction risk index into the hydraulic engineering digital twin model for simulation operation, and outputting a construction calibration report. Further, in the digital twinning-based hydraulic engineering intelligent monitoring method, the acquiring time sequence sensor data, video image data, environment data and history working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data comprises the following steps: acquiring time sequence sensor data, video image data, environment data and historical working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data; Detecting abnormal values of time sequence sensor data by using a3 sigma criterion, judging the data exceeding the mean value by +/-3 times of standard deviation as the abnormal values, replacing the abnormal values by using a linear interpolation method of the data at adjacent moments, and filling the missing values by using a mean filling method. Further, in the digital twinning-based hydraulic engineering intelligent monitoring method, the acquiring time sequence sensor data, video image data, environment data and history working condition data in hydraulic engineering construction to obtain multi-source heterogeneous data, preprocessing the multi-source heterogeneous data to obtain initial multi-source heterogeneous data comprises the following steps: Cleaning video image data through an image quality evaluation algorithm, eliminating repeated records of environmental data and historical working condition data, and correcting error data through logic verification; Converting the time sequence data of different dimensions into standard normal distributi