A method, device and medium for joint prediction of dam cracks and steel bar stress

Abstract

The present invention discloses a method for jointly predicting dam cracks and steel bar stress, comprising the following steps: S1. collecting dam monitoring data and weighting temperature and reservoir water level data after preprocessing; S2. constructing multi-scale features from the data using a rolling window technique; S3. inputting the multi-scale features at several consecutive moments, constructing time series samples using a sliding window method, and using the crack data and steel bar stress data at the last moment of the window as target variables for joint prediction; S4. constructing a multi-task time series model and training the time series samples to independently output predicted values ​​for cracks and steel bar stress based on the time series samples; S5. evaluating and optimizing the parameters of the multi-task time series model using specified performance evaluation indicators, and applying the trained model to provide early warning information on dam cracks and steel bar stress using the jointly predicted target variables. This method can improve the accuracy and reliability of predictions through a weighting mechanism and a multi-task learning model.

Description

Technical Field

[0001] The present invention relates to the technical field of building safety monitoring, and in particular to a method, device and storage medium for jointly predicting dam cracks and steel bar stress based on multi-scale feature fusion. Background Art

[0002] As important water conservancy projects, the structural safety of reservoir dams is directly related to flood control in downstream areas and the protection of life and property. During dam operation, the formation of cracks and changes in reinforcement stress are key factors affecting dam structural safety. Therefore, real-time and accurate monitoring and prediction of dam cracks and reinforcement stress are crucial for preventing structural failure and formulating maintenance strategies.

[0003] Traditional dam monitoring methods primarily rely on data from a single type of sensor, such as piezometers and strain gauges. However, these methods often suffer from a single data dimension and insufficient information, making it difficult to timely and comprehensively reflect the true state of the dam structure. Furthermore, traditional methods typically analyze cracks and steel stress independently, failing to fully consider the coupling relationship between the two, which can lead to inaccurate prediction results.

[0004] Therefore, existing technologies use time series models such as long short-term memory networks (LSTM) or gated recurrent units (GRU) to analyze dam monitoring data. However, these methods usually only target a single target variable and fail to consider the joint prediction requirements of cracks and steel stress. At the same time, a multi-task learning model with a shared encoder and a dual-branch decoder is also used in the analysis of monitoring data. Therefore, in specific applications, the design of the model structure needs to be adjusted according to actual needs, and there are still deficiencies in its versatility and generalization.

[0005] To this end, this application specifically proposes a method for jointly predicting dam cracks and steel bar stress to solve the above technical problems. Summary of the Invention

[0006] The main purpose of the present invention is to provide a method for jointly predicting dam cracks and steel bar stresses to solve the technical problems raised in the background technology.

[0007] The present invention adopts the following technical solutions to solve the above technical problems:

[0008] A method for jointly predicting dam cracks and steel bar stresses, comprising:

[0009] S1. Collect dam monitoring data, and after pre-processing, weight the temperature data and reservoir water level data according to preset weights;

[0010] S2. Use rolling window technology to construct multi-scale features for weighted temperature and water level data;

[0011] S3. Input multi-scale features at several consecutive moments, construct a time series sample using a sliding window method, and use the crack data and steel bar stress data at the last moment of the window as the target variables for joint prediction;

[0012] S4. Construct a multi-task time series model and train the time series samples, wherein the multi-task time series model is used to independently output the predicted values ​​of cracks and steel stress through time series samples;

[0013] S5. Evaluate and optimize the parameters of the multi-task time series model using specified performance evaluation indicators, and apply the trained model to the prediction of real-time dam safety monitoring data to provide early warning information on dam cracks and steel bar stress through the joint prediction of target variables.

[0014] Preferably, the specific operation process of step S1 includes:

[0015] S11. Collect crack data and steel stress data. The amount of crack data N within the specified time period is c , the amount of steel bar stress data N s And the amount of water level data N H ;

[0016] S12. The crack data and the steel bar stress data are internally connected and merged based on time t to form joint monitoring data:

[0017] JointData(t)={(t,C(t),S(t))|t∈T C ∩T S}

[0018] Where t is the data acquisition time, T C and T S They represent the collection time sets of crack data and steel bar stress data respectively, C(t) is the data set of crack size, and S(t) is the data set of steel bar stress value;

[0019] S13. For each time t in the joint monitoring data, align the water level data using a time-based approximate merging method;

[0020] S14. The temperature weight is preset to α, and the reservoir water level weight is preset to β = 1-α. The temperature and reservoir water level data are weighted separately to obtain the weighted temperature data M′(t) and water level data H′(t). The calculation formula is:

[0021] M′(t)=α·M(t)

[0022] H′(t)=(1-α)H(t)

[0023] Among them, M(t) is the temperature data collected at time t, and H(t) is the reservoir water level data collected at time t.

[0024] Preferably, the specific operation process of aligning the water level data in step S13 includes:

[0025] For each joint monitoring time t, find the time t closest to t in the water level data * ,have:

[0026]

[0027] Among them, T H is the time collection of water level data;

[0028] Let the water level at this moment be H(t * ) is the water level at time t of the joint monitoring data, and the reservoir water level data H(t) = H(t * ).

[0029] Preferably, the multi-scale features in step S2, including but not limited to the cumulative values, means and standard deviations of short-term features, medium-term features and long-term features, are used to form the input feature set. For any weighted data X′(t) in the corresponding window at time t, the statistical features are defined as follows:

[0030]

[0031]

[0032]

[0033] Among them, Sum n (t) is the cumulative value of multi-scale features, n is the size of the rolling window, μ n (t) is the mean of multi-scale features, σ n (t) is the standard deviation of the multi-scale feature;

[0034] The above formulas are applied to the weighted temperature data M′(t) and water level data H′(t) at acquisition time t to construct a multi-scale temperature and water level statistical feature set for subsequent joint prediction model training.

[0035] Preferably, the specific operation process of constructing time series samples using the sliding window method in step S3 includes:

[0036] S31. Set the multi-scale input feature vector obtained at each time point t in the joint monitoring data to F(t)∈R d , where F(t) includes the statistical characteristics of temperature, reservoir water level and their rolling window calculations at various scales, R dis the d-dimensional vector feature map;

[0037] S32. At time t, set the target variables to the crack size C(t) and the steel bar stress S(t);

[0038] S33. Use the sliding window method to construct time series samples. Assume that the window size is n. Then the input X of the i-th time series sample is i With target y i Respectively expressed as:

[0039] X i =[F(t i ),F(t i+1 ),…,F(t i+n-1 )]∈R n×d ,

[0040] y i =[C(t i+n-1 ),S(t i+n-1 )]∈R 2 .

[0041] Where i = 1, 2, ..., τ, where τ = N-n+1, ​​N is the total number of data moments, R n×d is the n×d dimensional channel vector feature map.

[0042] Preferably, the multi-task timing model in step S4 includes a shared encoder module and a dual-branch decoder module, wherein:

[0043] (1) Shared encoder, input is time series sample X i ∈R n×d , the local and global dynamic features of the input time series data are extracted through the convolutional layer and the recurrent neural network. The process is recorded as:

[0044] h i =f enc (X i )

[0045] Among them, h i ∈R p is the shared feature representation, f enc represents the encoder function composed of convolutional neural network and recurrent neural network, R n×d is the n×d-dimensional channel vector feature map, R p is the p-dimensional vector feature map;

[0046] (2) The dual-branch decoder is used to process the shared features separately. The shared features h i Then they are fed into two independent decoders, which independently output the predicted values ​​of cracks and steel bar stress, which are:

[0047]

[0048] Among them, f dec1 and f dec2 Denote the decoder modules for predicting cracks and reinforcement stress respectively. The decoder modules are composed of fully connected layers. and are the crack size and steel bar stress predicted by the model, respectively.

[0049] Preferably, the training objective of the multi-task time series model in step S4 is to minimize the mean square error of all samples, and its overall loss function is expressed as:

[0050]

[0051] Among them, γ1 and γ2 are weight parameters to ensure a reasonable balance in the task contributions of crack size C(t) and steel bar stress S(t) in the loss function.

[0052] Preferably, the data processing operation flow of the multi-task timing model in step S5 includes:

[0053] S51. For the i-th sample, output the predicted target y at the last moment of the window i And the predicted value after multiple rounds of prediction have:

[0054]

[0055] Among them C i and represent the actual and predicted crack sizes, S i and represent the true and predicted reinforcement stresses, respectively;

[0056] S52. The root mean square error and the coefficient of determination R 2 As an indicator to evaluate the model performance, and based on the evaluation results, the parameters of the multi-task timing model are optimized. The calculation formula of the indicator evaluation is:

[0057]

[0058]

[0059] in, is the mean vector of all true targets;

[0060] S53. After model training and parameter optimization, the trained multi-task time series model is used to predict real-time dam safety monitoring data. For new input time series data X new , prediction results Expressed as:

[0061]

[0062] Where f(·) represents the trained joint prediction model, and the crack size prediction result and steel bar stress prediction results They are used to warn of dam cracks and abnormal steel bar stress respectively.

[0063] In another aspect, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0064] On the other hand, the present invention further discloses a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0065] As can be seen from the above technical solution, the present invention provides a method for jointly predicting dam cracks and steel bar stress. Compared with the existing technology, the present invention has the following advantages:

[0066] 1. This invention significantly improves the integrity and reliability of data through time alignment, missing value filling and weighted fusion, ensuring accurate matching of multi-source monitoring data such as crack monitoring, steel bar stress and reservoir water level.

[0067] 2. The present invention constructs multi-scale statistical features and adopts a multi-task time series model that combines a shared encoder with a dual-branch decoder. It can simultaneously predict crack expansion trends and steel bar stress changes. At the same time, it extracts local features through a convolutional neural network and combines it with a recurrent neural network to capture the long-term dependencies of time series data, thereby improving the model's ability to model complex dynamic changes.

[0068] 3. The present invention extracts the local and global dynamic features of the input time series data through a shared encoder, and adopts a dual-branch decoder to predict cracks and steel bar stress respectively. This not only ensures the efficiency of feature sharing, but also takes into account the characteristics of different physical quantities, thereby improving the prediction accuracy and stability.

[0069] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easy to understand through the following description. Of course, it is not necessary to achieve all of the above-mentioned advantages simultaneously in order to implement any product of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0071] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0072] Figure 2 This is a schematic diagram of the multi-task timing model training optimization process of the present invention;

[0073] Figure 3 This is a flow chart of the multi-task timing model structure operation method of the present invention;

[0074] Figure 4 This is a schematic diagram of the model prediction experimental results of monitoring point J01-01 [1# pile|crack] according to an embodiment of the present invention;

[0075] Figure 5 Schematic diagram of the experimental results of the model prediction of monitoring point R02-02 [2# pile | steel bar stress] according to an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0077] In the embodiment, see Figures 1 to 5 .

[0078] like Figure 1 and Figure 2 The embodiment of the present invention proposes a joint prediction method for dam cracks and steel bar stress based on multi-scale feature fusion, which can improve the accuracy and reliability of prediction through a reasonable weighting mechanism and a multi-task learning model. The specific implementation steps include:

[0079] S1. Collect dam monitoring data, and after pre-processing, perform weighted processing on the temperature data and reservoir water level data according to preset weights to form weighted temperature data and water level data.

[0080] The specific operation process at this time includes:

[0081] S11. Data reading and sorting: Collect crack data and steel bar stress data, including:

[0082] The crack data is where t i Indicates the acquisition time, M i Indicates temperature, C i Indicates the size of the crack;

[0083] The steel bar stress data is Among them S j Indicates the stress value of steel bar;

[0084] Reservoir water level data is Among them, H k Indicates the water level,

[0085] Due to the difference in acquisition frequency of different devices, there is a certain amount of crack data N in a specified time. c , the amount of steel bar stress data N s And the amount of water level data N H ;

[0086] S12. The crack data and the steel bar stress data are internally connected and merged based on time t to form joint monitoring data:

[0087] JointData(t)={(t,C(t),S(t))|t∈T C ∩T S}

[0088] Where t is the data acquisition time, T C and T S They represent the collection time sets of crack data and steel bar stress data respectively, C(t) is the data set of crack size, and S(t) is the data set of steel bar stress value;

[0089] S13. For each time t in the joint monitoring data, the water level data are aligned using the time approximate merge (merge_asof) method. The specific operation process includes:

[0090] Let the water level data be H(t). For each joint monitoring time t, find the time t closest to t in the water level data. * ,have:

[0091]

[0092] Among them, T H is the time collection of water level data;

[0093] Let the water level at this moment be H(t * ) is the water level at time t of the joint monitoring data, and the reservoir water level data H(t) = H(t * );

[0094] The final result includes time t, crack size C(t), steel bar stress S(t) and aligned water level H t Joint monitoring data.

[0095] After completing the above steps, the following data table structure is obtained:

[0096] Time t Temperature M Crack size C Steel bar stress S Reservoir water level H <![CDATA[t1]]> <![CDATA[M1]]> <![CDATA[C1]]> <![CDATA[S1]]> <![CDATA[H1]]> <![CDATA[t2]]> <![CDATA[M2]]> <![CDATA[C2]]> <![CDATA[S2]]> <![CDATA[H2]]> <![CDATA[t3]]> <![CDATA[M3]]> <![CDATA[C3]]> <![CDATA[S3]]> <![CDATA[H3]]> … … … … …

[0097] At this time, the integrity and reliability of the data are significantly improved through time alignment, missing value filling and weighted fusion of the data, ensuring the accurate matching of multi-source monitoring data such as crack monitoring, steel bar stress and reservoir water level.

[0098] S14. The temperature weight is preset to α = 0.3, and the reservoir water level weight is preset to β = 1-α. The temperature and reservoir water level data are weighted separately to obtain the weighted temperature data M′(t) and water level data H′(t). The calculation formula is:

[0099] M′(t)=α·M(t)

[0100] H′(t)=(1-α)H(t)

[0101] Among them, M(t) is the temperature data collected at time t, and H(t) is the reservoir water level data collected at time t.

[0102] S2. Use rolling window technology to construct multi-scale features for weighted temperature and water level data.

[0103] The multi-scale features at this time, including but not limited to the cumulative value, mean and standard deviation of short-term features (the last 3 moments), medium-term features (the last 6 moments) and long-term features (the last 21 moments), are used to form the input feature set;

[0104] To fully capture the changing trends of data over time, a rolling window technique is used to calculate statistical features at different scales for weighted temperature M′(t) and reservoir water level H′(t). Assuming the rolling window size is n (n=3 for short-term, n=6 for medium-term, and n=21 for long-term), the statistical features of any weighted data X′(t) (where X represents M or H) within the corresponding window at time t are defined as follows:

[0105]

[0106]

[0107]

[0108] Among them, Sum n (t) is the cumulative value (accumulated value) of the multi-scale feature, n is the size of the rolling window, μn (t) is the mean of multi-scale features, σ n (t) is the standard deviation of the multi-scale feature;

[0109] The above formulas are applied to the weighted temperature data M′(t) and water level data H′(t) at acquisition time t to construct a multi-scale temperature and water level statistical feature set for subsequent joint prediction model training.

[0110] S3. Input multi-scale features from several consecutive moments, construct a time series sample using a sliding window approach, and use the crack and steel stress data at the last moment of the window as the target variables for joint prediction. For example, environmental characteristics from a certain period of time (e.g., 30 time steps, approximately 10 days or longer) are used as model input, and the target variables are the crack and steel stress data corresponding to the end of the window.

[0111] The specific operation process of constructing time series samples using the sliding window method includes:

[0112] S31. Set the multi-scale input feature vector obtained at each time point t in the joint monitoring data to F(t)∈R d , where F(t) includes temperature, reservoir water level and its statistical characteristics (cumulative value, mean, standard deviation) of each scale calculated by rolling window, R d is the d-dimensional vector feature map;

[0113] S32. At time t, set the target variables to the crack size C(t) and the steel bar stress S(t);

[0114] S33. Use the sliding window method to construct the time series sample. Assume that the window size is n (n = 21 represents the observation data of the past 21 moments). Then the input X of the i-th time series sample is i With target y i Respectively expressed as:

[0115] X i =[F(t i ),F(t i+1 ),…,F(t i+n-1 )]∈R n×d ,

[0116] y i =[C(t i+n-1 ),S(t i+n-1 )]∈R 2 .

[0117] Where i = 1, 2, ..., τ, where τ = N-n+1, ​​N is the total number of data moments, R n×d is the n×d dimensional channel vector feature map.

[0118] S4. Build a multi-task time series model and train it on time series samples, such as Figure 3 As shown in the figure, after normalization, the monitoring data are jointly predicted using a multi-task time series model, which is used to independently output the predicted values ​​of cracks and steel bar stress through time series samples.

[0119] Specifically, the multi-task timing model includes a shared encoder module and a dual-branch decoder module, where:

[0120] (1) Shared encoder, input is time series sample X i ∈R n×d , extracting local and global dynamic features of input time series data through convolutional layers and recurrent neural networks (LSTM or GRU). The process is recorded as:

[0121] h i =f enc (X i )

[0122] Among them, h i ∈R p is the shared feature representation, f enc represents the encoder function composed of a convolutional neural network (CNN) and a recurrent neural network (LSTM or GRU), R n×d is the n×d-dimensional channel vector feature map, R p is the p-dimensional vector feature map;

[0123] (2) The dual-branch decoder is used to process the shared features separately. The shared features h i Then they are fed into two independent decoders, which independently output the predicted values ​​of cracks and steel bar stress, which are:

[0124]

[0125] Among them, f dec1 and f dec2 They represent the decoder modules for predicting cracks and reinforcement stress respectively. The decoder module consists of a fully connected layer. and are the crack size and steel bar stress predicted by the model, respectively.

[0126] Therefore, a shared encoder is used to extract the local and global dynamic features of the input time series data, and a dual-branch decoder is used to predict cracks and steel bar stresses respectively. This not only ensures the efficiency of feature sharing, but also takes into account the characteristics of different physical quantities, improving the prediction accuracy and stability. Furthermore, by constructing multi-scale statistical features and adopting a multi-task time series model that combines a shared encoder with a dual-branch decoder, it is possible to simultaneously predict the crack expansion trend and steel bar stress changes. At this time, local features are extracted through a convolutional neural network, and combined with a recurrent neural network to capture the long-term dependencies of time series data, thereby improving the model's ability to model complex dynamic changes.

[0127] At this time, the training goal of the multi-task time series model is to minimize the mean square error of all samples, and its overall loss function is expressed as:

[0128]

[0129] Among them, γ1=γ2=1 is the weight parameter, which ensures a reasonable balance in the task contributions of crack size C(t) and steel bar stress S(t) in the loss function.

[0130] In summary, the sliding window method is used to construct time series samples, and the joint monitoring data is trained through a multi-task time series model (shared encoder and dual-branch decoder structure) to achieve joint prediction of dam cracks and steel stress, providing accurate early warning basis for dam safety monitoring.

[0131] S5. Use a multi-task time series model to predict the entire dataset and display the prediction results in a timeline format. Then, use specified performance evaluation metrics such as mean squared error to evaluate and optimize the multi-task time series model parameters. The trained model is then applied to the prediction of real-time dam safety monitoring data, providing early warning information on dam cracks and steel stress through the joint prediction of target variables.

[0132] The data processing operation flow of the multi-task timing model includes:

[0133] S51. Model output and true value representation: For the i-th sample, output its target value (the predicted target y at the last moment of the window i ) and the predicted value after multiple rounds of prediction have:

[0134]

[0135] Among them C i and represent the actual and predicted crack sizes, S i and represent the true and predicted reinforcement stresses, respectively;

[0136] S52. Performance evaluation: In the model evaluation stage, the root mean square error (RMSE) and the coefficient of determination R 2 As an indicator to evaluate the model performance, and based on the evaluation results, the parameters of the multi-task timing model are optimized. The calculation formula of the indicator evaluation is:

[0137]

[0138]

[0139] in, is the mean vector of all true targets;

[0140] S53. Real-time prediction application: After model training and parameter optimization, the trained multi-task time series model is used to predict real-time dam safety monitoring data. new , prediction results Expressed as:

[0141]

[0142] Where f(·) represents the trained joint prediction model, and the crack size prediction result and steel bar stress prediction results They are used to warn of dam cracks and abnormal steel stress, and assist in dam safety management and risk warning.

[0143] In a specific application of this embodiment, the present invention applies the trained model to the prediction task of real-time data. The experimental results of Reservoir B in City A are as follows: Figure 3 and Figure 4 As shown, the R 2 The results reached 0.9747 and 0.9968 respectively. This result shows that this method can achieve the joint prediction of the two monitoring data and achieve better results, providing a scientific and reliable basis for dam safety management and risk warning.

[0144] In summary, the joint prediction method of dam cracks and steel bar stress based on multi-scale feature fusion disclosed by the present invention adopts the coefficient of determination (R 2 ) and other indicators to evaluate the model performance and optimize the parameters to ensure that the model has good robustness and prediction accuracy. Through the precise joint prediction of dam monitoring data and the effective integration of multi-source monitoring data, the joint prediction of key dam safety indicators is achieved, which provides a scientific decision-making basis for dam safety management and maintenance, and realizes real-time early warning of dam cracks and steel stress, thereby greatly improving the safety of dam operation and risk prevention and control capabilities.

[0145] In another aspect, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0146] On the other hand, the present invention further discloses a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0147] In another embodiment provided in the present application, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute any of the methods for jointly predicting dam cracks and steel bar stresses in the above embodiments.

[0148] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts of the above method.

[0149] The embodiment of the present application further provides an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus.

[0150] Memory for storing computer programs;

[0151] The processor is configured to implement the above-mentioned method for jointly predicting dam cracks and steel bar stress when executing the program stored in the memory.

[0152] The communication bus mentioned in the above electronic device can be a peripheral component interconnect standard bus or an extended industry standard architecture bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0153] The communication interface is used for communication between the above electronic device and other devices.

[0154] The memory may include a random access memory, or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0155] The above-mentioned processor can be a general-purpose processor, including a central processing unit, a network processor, etc.; it can also be a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component.

[0156] It should also be noted that electronic devices also include terminal devices, which can also be called terminals, user equipment, mobile stations, mobile terminals, etc. Terminal devices can be mobile phones, smart TVs, wearable devices, tablet computers, computers with wireless transceiver functions, virtual reality terminal devices, augmented reality terminal devices, wireless terminals in industrial control, wireless terminals in unmanned driving, wireless terminals in remote surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc. The embodiments of this application do not limit the specific technology and specific device form used by the terminal devices.

[0157] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium, or a semiconductor medium (e.g., a solid-state hard disk).

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0159] In addition, it should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0160] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the meaning of "and / or" appearing throughout the text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or schemes in which A and B are satisfied at the same time. In addition, in the embodiments of the present invention, "multiple" refers to more than two. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A method for jointly predicting dam cracks and steel bar stress, characterized in that: include: S1. Collect dam monitoring data, and after pre-processing, weight the temperature data and reservoir water level data according to preset weights; S2. Use rolling window technology to construct multi-scale features for weighted temperature and water level data; S3. Input multi-scale features at several consecutive moments, construct a time series sample using a sliding window method, and use the crack data and steel bar stress data at the last moment of the window as the target variables for joint prediction; S4. Construct a multi-task time series model and train the time series samples, wherein the multi-task time series model is used to independently output the predicted values ​​of cracks and steel stress through time series samples; S5. Evaluate and optimize the parameters of the multi-task time series model using specified performance evaluation indicators, and apply the trained model to the prediction of real-time dam safety monitoring data to provide early warning information on dam cracks and steel bar stress through the joint prediction of target variables; The specific operation process of step S1 includes: S11. Collect crack data and steel stress data, and the amount of crack data within the specified time , the amount of steel bar stress data and water level data volume ; S12. Combine crack data and steel bar stress data based on time Perform inner join and merge to form joint monitoring data: in, is the data collection time, and Respectively represent the collection time sets of crack data and steel bar stress data, is the data set of crack size, is the data set of steel bar stress values; S13. For each moment in the joint monitoring data ,The water level data are aligned using the time-based approximate merging method; S14. The preset temperature weight is , the reservoir water level weight is , weight the temperature and reservoir water level data respectively to obtain the weighted temperature data and water level data , the calculation formula is: in, The collection time is Temperature data at The collection time is Reservoir water level data at time ; The specific operation process of aligning the water level data in step S13 includes: For each joint monitoring moment , find the water level data with The most recent moment ,have: in, is the time collection of water level data; Let the water level value at this moment As joint monitoring data at all times Water level, with reservoir water level data .

2. The method for joint prediction of dam cracks and steel bar stress according to claim 1, characterized in that: The multi-scale features in step S2, including but not limited to the cumulative value, mean and standard deviation of short-term features, medium-term features and long-term features, are used as the input feature set. For any weighted data In time The statistical feature definitions in the corresponding window are: in, is the cumulative value of multi-scale features, is the scroll window size, is the mean of multi-scale features, is the standard deviation of the multi-scale feature; Apply the above formula to the acquisition time Time-weighted temperature data and water level data , to construct a multi-scale temperature and water level statistical feature set for subsequent joint prediction model training.

3. The method for joint prediction of dam cracks and steel bar stress according to claim 1, characterized in that: The specific operation process of constructing time series samples using the sliding window method in step S3 includes: S31. Combine the joint monitoring data at each time point The obtained multi-scale input feature vector is set as ,in Including temperature, reservoir water level and its statistical characteristics at various scales calculated through rolling windows, for dimensional vector feature map; S32.In time At this point, the target variable is set to the crack size and steel bar stress ; S33. Use the sliding window method to construct time series samples, and set the window size to , then Input of time series samples With the goal Respectively expressed as: , . in, in , is the total number of data moments, for dimensional channel vector feature map.

4. The method for joint prediction of dam cracks and steel bar stress according to claim 1, characterized in that: The multi-task timing model in step S4 includes a shared encoder module and a dual-branch decoder module, wherein: (1) Shared encoder, input is time series samples , the local and global dynamic features of the input time series data are extracted through the convolutional layer and the recurrent neural network. The process is recorded as: in, is the shared feature representation, represents the encoder function composed of a convolutional neural network and a recurrent neural network, for dimensional channel vector feature map, for dimensional vector feature map; (2) Dual-branch decoder, used to process shared features separately. Then they are fed into two independent decoders, which independently output the predicted values ​​of cracks and steel bar stress, which are: in, and Denote the decoder modules for predicting cracks and reinforcement stress respectively. The decoder modules are composed of fully connected layers. and are the crack size and steel bar stress predicted by the model, respectively.

5. The method for joint prediction of dam cracks and steel bar stress according to claim 4, characterized in that: The training objective of the multi-task time series model in step S4 is to minimize the mean square error of all samples, and its overall loss function is expressed as: in is the weight parameter to ensure that the crack size in the loss function and steel bar stress The contribution of tasks is reasonably balanced.

6. The method for joint prediction of dam cracks and steel bar stress according to claim 1, characterized in that: The data processing operation flow of the multi-task timing model in step S5 includes: S51. For samples, output the predicted target at the last moment of the window And the predicted value after multiple rounds of prediction ,have: ， in and represent the actual and predicted crack sizes, and represent the true and predicted reinforcement stresses, respectively; S52. The root mean square error and the coefficient of determination As an indicator to evaluate the model performance, and based on the evaluation results, the parameters of the multi-task timing model are optimized. The calculation formula of the indicator evaluation is: in, is the mean vector of all true targets; S53. After model training and parameter optimization, the trained multi-task time series model is used to predict real-time dam safety monitoring data. For new input time series data , prediction results Expressed as: in Represents the trained joint prediction model and the crack size prediction results and steel bar stress prediction results They are used to warn of dam cracks and abnormal steel bar stress respectively.

7. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 6.

8. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 6.

Images