Integrated rockfill dam deformation analysis method based on integration of multiple professional models, and rockfill dam deformation analysis system and device based on integration of multiple professional models

Abstract

The present invention relates to the field of computers and the field of hydraulic structures. Disclosed is an integrated rockfill dam deformation analysis method based on the integration of multiple professional models, the method comprising the following step: S1, using a parameter back-analysis method to establish a constitutive model for rockfill materials. Further disclosed is a rockfill dam deformation analysis system based on the integration of multiple professional models, the system comprising a parameter back-analysis module, which is used for establishing a constitutive model for rockfill materials. Moreover, further disclosed is a rockfill dam deformation analysis device based on the integration of multiple professional models, the device comprising a cluster management host, which is responsible for resource allocation and management of an entire device network. Using containerized WebSocket technology greatly simplifies real-time data exchange and communication between containers; moreover, in a complex data processing flow, an output of step S1 can immediately serve as an input of step S2, and an output of step S2 can immediately serve as an input of step S3; and the "single communication, persistent connection" advantage based on the WebSocket technology reduces the communication load of servers.

Description

Rockfill dam deformation analysis integration method, system and device based on multi-specialty model integration TECHNICAL FIELD

[0001] The present application relates to the field of computer and the field of hydraulic structure, and specifically to a rockfill dam deformation analysis integration method, system and device based on multi-specialty model integration. BACKGROUND

[0002] In the construction of rockfill dams, ensuring the structural safety is of great importance, and deformation control plays a key role. With the rapid development of monitoring technology and data acquisition platforms, the amount of data generated by monitoring has significantly increased. When using data-driven methods to process and analyze data, the richness of multi-source monitoring data provides data support for the health monitoring of rockfill dams. Currently, the deformation analysis methods for rockfill dam projects can be divided into three categories.

[0003] The first category includes time series prediction, which usually uses modern mathematical methods such as grey time series model, machine learning and neural networks. Machine learning algorithms can automatically learn deformation patterns from a large amount of monitoring data, reducing the need for human intervention. By training the model to identify trends and anomalies in the data, machine learning can effectively predict the future deformation behavior of rockfill dams. At the same time, machine learning models can adapt to changing monitoring data and have the ability to continuously learn. The second category involves spatial interpolation techniques. The most commonly used methods include inverse distance weighting and ordinary kriging. Spatial interpolation methods can estimate deformation between monitoring points, generating continuous deformation fields that make deformation distribution easier to visualize and understand. Interpolation methods can provide accurate deformation estimates within local areas. The third category uses spatio-temporal models to estimate the overall deformation of rockfill dams. Spatio-temporal models can integrate time and spatial variables to provide more comprehensive deformation analysis, allowing for a better understanding of the dynamic development process of rockfill dam deformation. By integrating time series and spatial analysis, the model can predict the deformation state of the rockfill dam at future time points, enabling long-term monitoring of the rockfill dam.

[0004] However, single methods often have many problems, and the performance of machine learning models is highly dependent on the quality and quantity of training data. Monitoring data often has large deviations, is incomplete or has too much noise, and data problems will directly affect the accuracy and reliability of the model. Spatial interpolation methods are usually based on the assumption of spatial autocorrelation, that is, points close in distance have similar properties. However, in practice, the deformation of rockfill dams may be influenced by a variety of complex factors, and the spatial distribution of these factors may not meet this assumption. When the monitoring points are sparse, the uncertainty of the interpolation results increases, which may lead to misunderstanding of the deformation field. The effect of spatial interpolation is limited by the observation scale, which is scale sensitive. Different scales may lead to completely different interpolation results, which is particularly evident in large-scale rockfill dam structures. Rockfill dam deformation is a dynamic process influenced by a variety of factors, and spatiotemporal models need to be constantly adjusted and optimized to accurately capture these dynamic changes. At the same time, the effectiveness of the model is limited by the spatiotemporal scale, and different observation times and spatial ranges may require different model configurations and parameter adjustments.

[0005] Rockfill dam deformation calculation needs to process a large amount of monitoring data, including but not limited to displacement, stress, strain and other parameters of the dam body, foundation and surrounding geological body. With the continuous improvement of monitoring data collection frequency and accuracy, the amount of data has shown exponential growth. The processing and analysis of these data pose extremely high requirements on computing resources. The deformation calculation model of rockfill dams is usually nonlinear, and needs to consider various physical and mechanical processes. The coupling of these factors makes the calculation model very complex, usually relying on large-scale finite element analysis or machine learning methods, etc. Finite element calculation method has a large calculation volume, and has high requirements on computer memory and processing power. In order to ensure the accuracy of rockfill dam deformation calculation, finer meshes and more calculation nodes need to be used in machine models, further increasing the complexity of calculation and the required resources, so that serious load overload problems often occur during calculation.

[0006] Due to the obvious spatiotemporal correlation of rockfill dam deformation and the significant variability of monitoring data in time and space, it is difficult to describe with a unified deformation analysis model, and professional model calculation usually needs to process a large number of data points and complex data relationships, which makes the model's calculation demand high and the calculation cost huge, so relying only on single machine system to perform single method deformation evaluation and prediction still needs to be further improved.

[0007] SUMMARY

[0008] In view of the deficiencies of the prior art, the present application provides a rockfill dam deformation analysis integration method, system and device based on multi-specialty model integration, which solves the problems of insufficient data processing capacity, low analysis accuracy and poor model adaptability of single method in rockfill dam deformation analysis.

[0009] To achieve the above object, the application is implemented by the following technical solutions: a rockfill dam deformation analysis integration method based on multi-specialty model integration, comprising the following steps:

[0010] S1, establishing a rockfill material constitutive model by a parameter inversion analysis method;

[0011] S2, performing finite element calculation based on the constitutive model to generate a rockfill dam deformation field;

[0012] S3, reconstructing the rockfill dam deformation field by a Bayesian Kriging interpolation method.

[0013] Preferably, the steps of the parameter inversion analysis method specifically comprise the following steps:

[0014] selecting an agent calculation model for training, the agent calculation model including an ANN neural network, a support vector machine SVR, a gradient boosting machine XGBOOST and a random forest RF;

[0015] performing parameter inversion by a non-dominated sorting genetic algorithm to obtain the rockfill material constitutive model.

[0016] Preferably, the steps of the finite element calculation specifically comprise the following steps:

[0017] obtaining constitutive model parameters by containerized WebSocket full-duplex communication;

[0018] performing finite element calculation of the rockfill dam deformation field by a containerized ABAQUS platform.

[0019] Preferably, the steps of the Bayesian Kriging interpolation method specifically comprise the following steps:

[0020] obtaining finite element calculation results by containerized WebSocket full-duplex communication;

[0021] reconstructing the rockfill dam deformation field by Bayesian Kriging interpolation.

[0022] A rockfill dam deformation analysis system based on multi-specialty model integration, comprising:

[0023] a parameter inversion analysis module for establishing a rockfill material constitutive model;

[0024] a finite element calculation module for calculating a deformation field based on the constitutive model;

[0025] a deformation field reconstruction module for reconstructing the deformation field by Bayesian Kriging interpolation;

[0026] a data processing unit for processing and caching data.

[0027] Preferably, the system is designed based on a micro-service architecture, adopts containerized deployment, the parameter inversion analysis module utilizes a multi-objective strategy for parameter inversion, and the finite element calculation module is realized by containerized packaging and secondary development of an ABAQUS platform.

[0028] Preferably, the data processing unit comprises:

[0029] a processor for being responsible for core computing tasks of rockfill dam deformation analysis, including parameter inversion analysis of a constitutive model, finite element calculation, and reconstruction of a deformation field, and establishing WebSocket full-duplex communication of a client with an algorithm module and a database;

[0030] a cache connected with the processor, for temporarily storing computing data and intermediate results, improving data transmission quality and guaranteeing stability of a computing process, and reducing overload risk of the system.

[0031] The rockfill dam deformation analysis device based on multi-specialty model integration comprises:

[0032] a cluster management host for being responsible for resource allocation and management of the entire device network, and being responsible for management of method and data allocation in the rockfill dam deformation analysis system based on multi-specialty model integration, and supporting addition of existing algorithm models;

[0033] a data management host connected with the cluster management host, for being responsible for persistent storage of computing data and historical data;

[0034] an algorithm management host connected with the cluster management host, for being responsible for management and execution of various algorithms involved in the system.

[0035] The rockfill dam deformation analysis device based on multi-specialty model integration can be applied to the data management host.

[0036] A chip system comprises a processor for reading and executing a system program stored in a memory.

[0037] The present application provides a rockfill dam deformation analysis integrated method, system and device based on multi-specialty model integration.

[0038] 1. The application greatly simplifies real-time data exchange and communication between containers through containerized WebSocket technology. WebSocket allows containers to send and receive data in real time, which is beneficial for frequently interactive rockfill dam deformation analysis model system. In a complex data processing flow, the output of S1 step can become the input of S2 step, and the output of S2 step can become the input of S3 step. The advantage of "one communication, persistent connection" based on WebSocket technology reduces the server communication load.

[0039] 2. The application supports users to upload existing rockfill dam constitutive model or directly upload finite element calculation results, enhancing the flexibility of the platform and the operation convenience of users. Users can flexibly select data sources and calculation methods according to actual needs, optimizing the calculation process, reducing the operation difficulty of non-professionals in rockfill dam finite element calculation, while ensuring the accuracy and reliability of the calculation results.

[0040] 3. The application can flexibly call the calculation results of each integrated module, automatically input the calculation results and customize the input of user's existing files, effectively integrating and utilizing existing calculation resources and data, improving the consistency and efficiency of calculation.

[0041] 4. The application significantly improves the accuracy and efficiency of deformation analysis through highly integrated calculation process, and significantly improves the efficiency and reliability of analysis through systematic processing flow.

[0042] 5. The application can automatically allocate the required computing resources according to the complexity of the deformation analysis task through the advanced resource management strategy of the cluster management host. The adaptive strategy based on Kafka message queue fusion VPA+HPA is proposed. For modules with large computing resource occupancy or access, the host will automatically schedule more computing resources, reducing the delay and waste of calculation caused by improper resource allocation, and ensuring the stability of complex model calculation and finite element analysis.

[0043] 6. The application can efficiently process a large amount of calculation data from rockfill dam deformation analysis through the data management host, and through its efficient storage and fast access capability, ensures the speed and accuracy of calculation data processing, making large-scale data analysis more rapid and reliable.

[0044] 7. The application allows rapid deployment and update of rockfill dam deformation analysis algorithms such as finite element model and Kriging interpolation algorithm through the use of containerization technology by the algorithm management host. These algorithms can run in independent and isolated environments, increasing the security and portability of the calculation process, while also facilitating algorithm testing and optimization.

[0045] 8、The application decouples the algorithm and data for management on different hosts, enhances the flexibility and scalability of the system, optimizes the cost-effectiveness, improves fault tolerance and reliability, and simplifies data security and compliance measures, in addition, the decoupling architecture helps performance optimization, easy disaster recovery, and also makes system management and maintenance more efficient and convenient, which is particularly suitable for rockfill dam deformation analysis and calculation that needs to process large-scale data and complex computing tasks. BRIEF DESCRIPTION OF DRAWINGS

[0046] Fig. 1 is a flowchart of the method of the application;

[0047] Fig. 2 is a general flowchart of the method of the application;

[0048] Fig. 3 is a schematic diagram of the deformation calculation process of the rockfill dam deformation analysis system of the application;

[0049] Fig. 4 is a schematic diagram of the containerization process of the WebSocket service of the application;

[0050] Fig. 5 is a schematic diagram of the execution process of the parameter inversion analysis method of the system package of the application;

[0051] Fig. 6 is a schematic diagram of the execution process of the finite element calculation method of the system package of the application;

[0052] Fig. 7 is a schematic diagram of the execution process of the rockfill dam deformation field reconstruction method of the system package of the application;

[0053] Fig. 8 is a schematic diagram of the equipment of the application;

[0054] Fig. 9 is a cluster management host management architecture diagram of the equipment of the application;

[0055] Fig. 10 is a load balancing architecture diagram of the cluster management host of the application;

[0056] Fig. 11 is a schematic diagram of the load balancing result of the application;

[0057] Fig. 12 is a data management architecture diagram of the data management host of the application;

[0058] Fig. 13 is a flowchart of the algorithm deployment of the algorithm management host of the application;

[0059] Fig. 14 is an algorithm management architecture diagram of the algorithm management host of the application;

[0060] Fig. 15 is a framework diagram of the system of the application;

[0061] Fig. 16 is a system and equipment execution flowchart of the method of the application;

[0062] Fig. 17 is a flowchart of the first embodiment of the application. DETAILED DESCRIPTION

[0063] The technical solutions of the present application will be described clearly and completely in combination with the drawings of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0064] Referring to FIG. 1-FIG. 4, the embodiment of the present application provides a rockfill dam deformation analysis integration method based on multi-specialty model integration, including the following steps:

[0065] S1, establishing a rockfill material constitutive model by using a parameter inversion analysis method;

[0066] S2, performing finite element calculation based on the constitutive model to generate a rockfill dam deformation field;

[0067] S3, reconstructing the rockfill dam deformation field by a Bayesian Kriging interpolation method.

[0068] Specifically, the present application efficiently integrates the above-mentioned three rockfill dam deformation analysis processes through containerization integration technology, realizes mutual communication and result fusion of each algorithm process and tool component, and aims to solve the limitations in the prior art of single means for rockfill dam deformation analysis. The full-duplex communication combined with the containerization technology and the configuration of the load balancing strategy realize efficient integration and high-performance management of the relevant professional models in each stage of the rockfill dam deformation analysis.

[0069] The present application combines WebSocket full-duplex communication technology, containerization technology and container orchestration tools to manage the integration and operation of each model of the system, simplifies the integration difficulty of each algorithm model, and establishes data communication between each model. The WebSocket service is deployed in one or more containers. The WebSocket service can run in an isolated environment to ensure stability and security. The container is configured to automatically run the WebSocket service code when starting.

[0070] The present application first performs inversion calculation on the physical properties of the rockfill material through the S1 step model to obtain the accurate constitutive model. This step provides the physical accuracy and calculation reliability of the model for subsequent finite element calculation.

[0071] Then, the constitutive model obtained by inversion is used to simulate the overall deformation structure of the rockfill dam through the S2 step of the present application. This not only helps to predict the response of the rockfill dam under various working conditions, but also guides the decision-making in the engineering design and construction process.

[0072] Finally, the rock-fill dam deformation field is reconstructed by the S3 step of the application, so that the deformation behavior of the rock-fill dam can be more accurately analyzed and predicted.

[0073] In the communication process of each module, WebSocket full-duplex communication means is adopted, and the containerized WebSocket full-duplex communication process is shown in FIG. 3.

[0074] In general, the overall process of the application is as follows:

[0075] In the S1 method calculation, the S1 calculation task related parameters are transmitted into the Kafka cache queue through WebSocket full-duplex communication, the multi-Kafka consumer nodes of S2 on the algorithm management host obtain the calculation task parameters and calculate the rock-fill dam constitutive model, the cluster management host increases or reduces the number of S1 calculation nodes according to the actual calculation demand, and the obtained constitutive model is persistently stored to the data management host;

[0076] In the S2 method calculation, the S1 calculation constitutive model file is obtained from the data management host through WebSocket full-duplex communication and imported into the method; in the S2 method calculation, the S2 calculation task related parameters are transmitted into the Kafka cache queue through WebSocket, the multi-Kafka consumer nodes of S2 on the algorithm management host obtain the calculation task parameters and calculate the rock-fill dam deformation field file, the cluster management host increases or reduces the number of S2 calculation nodes according to the actual calculation demand, and the obtained rock-fill dam deformation field file is persistently stored to the data management host;

[0077] In the S3 method calculation, the S2 calculation rock-fill dam deformation field file is obtained from the data management host through WebSocket full-duplex communication and imported into the calculation method; in the S3 method calculation, the S3 calculation task related parameters are transmitted into the Kafka cache queue through WebSocket, the multi-Kafka consumer nodes of S3 on the algorithm management host obtain the calculation task parameters and calculate the rock-fill dam deformation field reconstruction file, the cluster management host increases or reduces the number of S3 calculation nodes according to the actual calculation demand, and the obtained rock-fill dam deformation field reconstruction file is persistently stored to the data management host.

[0078] The rock-fill dam rock-fill material constitutive model calculation S1 is the first step of the rock-fill dam deformation analysis, and in the application, the rock-fill dam constitutive model is calculated in the first step, and the rock-fill material constitutive model inversion parameters need to be provided. b, the bulk modulus exponent m; the alpha rheological rate in the Merchant seven-parameter model, the b, c, d rheological deformation parameters, and the c in the three-parameter wetting model w , n w , b w The model parameters are taken as the inversion parameters.

[0079] Referring to FIGS. 3 and 5, the steps of the parameter inversion analysis method specifically include the following steps:

[0080] The agent calculation model is selected for training, and the agent calculation model includes an ANN neural network, a support vector machine SVR, a gradient boosting machine XGBOOST, and a random forest RF.

[0081] The non-dominated sorting genetic algorithm is used for parameter inversion to obtain the rockfill material constitutive model.

[0082] Specifically, the rockfill dam parameter inversion calculation is divided into three steps:

[0083] Step one, the system realizes the rockfill dam parameter inversion calculation, and the calculation process includes agent model accuracy verification and parameter inversion calculation. The user selects the agent calculation model, inputs the calculation file, which includes the model training file and the model test file, clicks the calculation, and the system background starts the asynchronous calculation of the agent model. The asynchronous calculation accuracy result can be transmitted to the front-end chart in real time through the WebSocket full-duplex communication constructed by the system to realize the visual display of the model accuracy. In order to provide system robustness, the system automatically detects the calculated dam area model to avoid repeated model calculation of the same dam area. Each agent model must meet all dam section calculation results before inversion calculation.

[0084] Step two, input the parameter inversion calculation model. First, input the user's basic information for result retrieval. According to the calculation agent model calculated in step one, input the agent model accuracy, and simultaneously input the deformation file of the measuring point by using the non-dominated sorting genetic algorithm NSGA-III. The system provides a user-defined interactive interface for genetic algorithm parameters, including the number of particle swarms and the number of iterations, wherein the number of iterations is positively correlated with the system model calculation time.

[0085] Step three, system robustness calculation verification process. Before calculation, the inversion calculation condition needs to be checked first. The system will feedback the missing data to the user. After meeting the conditions, the calculation function is opened and can be normally run. Clicking the operation system will display the calculation process. After the calculation is completed, the system outputs the parameter inversion calculation model parameters, and marks the completion of the inversion calculation. At the same time, the system persistently stores the obtained rockfill dam constitutive file to the data management host.

[0086] S2 is the second step of the rockfill dam deformation analysis, and the system finite element calculation management module is packaged and developed based on ABAQUS, as shown in FIG. 6. Users can publish finite element calculation tasks in the system. Compared with the complex operation interface of directly using finite element calculation software, the system can simplify finite element calculation to the greatest extent and reduce the threshold of using finite element software.

[0087] Referring to FIGS. 3 and 6, the steps of finite element calculation specifically include the following steps:

[0088] The constitutive model parameters are obtained through containerized WebSocket full-duplex communication.

[0089] The finite element calculation of the rockfill dam deformation field is performed by using the containerized ABAQUS platform.

[0090] Specifically, the finite element calculation task of the rockfill dam is submitted in four steps:

[0091] Step 1: When the finite element calculation management module starts to calculate, automatically establish WebSocket full-duplex communication with the background to realize automatic input of the first step algorithm calculation result file as the second step calculation parameter. Users can also choose to upload an existing rockfill dam constitutive model to the finite element calculation management module for finite element calculation. Input the monitoring data time and click submit. The system prompts that the task has been submitted, and then step 2 can be executed.

[0092] Step 2: Finite element calculation text information input. According to the task number submitted in step 1, the user inputs the text explanation of the finite element calculation related parameter information.

[0093] Step 3: Calculation task starts. On the premise of normal submission of steps 1 and 2, the user inputs the custom task ID according to the task number, and clicks operation to start finite element calculation. The calculation task running state is divided into four categories, of which green represents that the task is completed, yellow represents that the task is in progress, blue represents that the task is idle, and red represents that the system operation is delayed. Among them, the button state changes from blue to yellow after the operation starts, and the normal operation is green after the operation ends. Red is the calculation delay (i.e. error operation).

[0094] Step 4 is the finite element calculation result extraction. The finite element calculation can realize complete step extraction. According to the task number, the user inputs the calculation extraction step according to the monitoring data time and the constitutive file name of steps 1 to 3, and clicks submit. The background can execute the task. The user can download the result file of the finite element calculation step extraction. At the same time, the user can upload the existing finite element result file.

[0095] This module can perform three-dimensional chart visualization display on the uploaded data. The three-dimensional chart shows the deformation degree of the rockfill dam by color classification of the calculation or uploaded results.

[0096] The S3 of the reconstruction of the deformation field of the rock-fill dam is the third step of the deformation analysis of the rock-fill dam, adopts a reconstruction method of the deformation field of the rock-fill dam based on the Bayesian Kriging interpolation, and is applied to the rock-fill dam deformation analysis system. The method adopts a loose coupling integration strategy, modularly encapsulates algorithm input, processing, and output, guarantees the usability and intuitive operation of the algorithm, calculates the reconstruction deformation field of the rock-fill dam through the Bayesian Kriging interpolation method of the reconstruction of the deformation field of the rock-fill dam, and obtains the reconstruction deformation field of the rock-fill dam as the final result of the deformation analysis and calculation of the rock-fill dam.

[0097] Please refer to FIG. 3 and FIG. 7, the steps of the Bayesian Kriging interpolation method specifically include the following steps:

[0098] The finite element calculation result is obtained through the containerized WebSocket full-duplex communication;

[0099] The deformation field of the rock-fill dam is reconstructed through the Bayesian Kriging interpolation.

[0100] Specifically, the three steps of the deformation field reconstruction module integrated by the system are as follows:

[0101] Step 1: input the calculation parameters, including the calculation period and the monitoring time and the like, input the finite element calculation file and the monitoring settlement value file that need to be reconstructed, wherein the deformation field reconstruction module starts to calculate, automatically establishes the WebSocket full-duplex communication with the background, realizes the automatic input of the calculation result file of step 2, takes the calculation file parameter of step 3, and the finite element calculation file can also be input by the user.

[0102] Step 2: select the deformation field reconstruction model integrated by the system, the system provides three types of models including the Gaussian model, the spherical model, and the exponential model, and calculates the precision of each model.

[0103] Step 3: input the calculation period and the monitoring time parameter to start the calculation, the system calculates the state and feeds back to the user in real time, under the premise of considering the parallel operation of multiple users, the information box includes the name of the submitter and the model selection information, the result is automatically transmitted to the client three-dimensional chart for display through the established full-duplex communication after the calculation is completed, and the model calculation precision is output. At the same time, the system stores the obtained rock-fill dam deformation field reconstruction file persistently to the data management host.

[0104] Please refer to FIG. 15, the rock-fill dam deformation analysis system based on the integration of multiple professional models includes:

[0105] The parameter inversion analysis module is used for establishing the constitutive model of the rock-fill material.

[0106] The finite element calculation module is used for calculating the deformation field based on the constitutive model.

[0107] The deformation field reconstruction module is configured to perform Bayesian Kriging interpolation reconstruction on the deformation field.

[0108] The data processing unit is configured to process and cache data.

[0109] The system is designed based on a micro-service architecture and adopts containerized deployment, the parameter inversion analysis module uses a multi-objective strategy for parameter inversion, and the finite element calculation module is realized by containerizing and encapsulating the ABAQUS platform and secondary development.

[0110] Specifically, the system of the present application integrates multiple professional analysis models based on a micro-service architecture, and for different stages of rockfill dam deformation analysis, machine learning algorithms, spatial interpolation methods and spatio-temporal models are combined and applied. In this way, the present application can integrate the advantages of various models to improve the comprehensiveness and accuracy of rockfill dam deformation monitoring and analysis.

[0111] The data processing unit includes:

[0112] The processor is configured to perform core computing tasks for rockfill dam deformation analysis, including parameter inversion analysis of the constitutive model, finite element calculation and reconstruction of the deformation field.

[0113] The cache is connected to the processor and is configured to temporarily store calculation data and intermediate results, improve data transmission quality and ensure stable calculation process, and reduce the risk of system overload.

[0114] Specifically, the processor realizes communication between the integrated method calculation and the data management host and the algorithm management host through WebSocket full-duplex communication, realizes "one communication, persistent connection", and reduces the internal communication cost of the system.

[0115] The cache temporarily stores the calculation task submission data of the rockfill dam deformation analysis integrated method based on multi-specialty model integration using Kafka message queue, reduces the system load under high peak value calculation, improves data access speed and supports fast calculation processing, and reduces the response time of the system. The message queue, as middleware between the client and the server, can temporarily store incoming requests and then pass them to the back-end cluster one by one for processing to smooth the load peak and avoid overloading the cluster service due to sudden increase in requests.

[0116] Referring to FIGS. 8-14, the rockfill dam deformation analysis device based on multi-specialty model integration includes:

[0117] The cluster management host is configured to be responsible for resource allocation and management of the entire device network, and is also responsible for managing method and data allocation in the rockfill dam deformation analysis system based on multi-specialty model integration, and supports adding existing algorithm models.

[0118] a data management host connected with the cluster management host, which is responsible for the persistent storage of calculation data and historical data;

[0119] an algorithm management host connected with the cluster management host, which is responsible for the management and execution of various algorithms involved in the system.

[0120] Specifically, the device is used for optimizing and accelerating the model calculation of rockfill dam deformation analysis, realizes the containerized management of various professional models, and guarantees the stability of integrated model calculation by combining the load balancing means of the message queue. Therefore, the device not only supports efficient data processing and complex algorithm operation, but also provides a stable and reliable computing platform that can support multiple users.

[0121] It mainly uses a container orchestration tool to manage containers in each host. The cluster management host mainly includes a control plane including a cluster management component, a cluster scheduling component, a cluster control component and an ETCD database. By integrating efficient scheduling algorithms and load balancing mechanisms in the cluster management host, resource management and load balancing of the data management host and the algorithm management host in the cluster are realized. The data management host and the algorithm management host mainly include an agent component, a network agent component and a container engine.

[0122] When a user submits a new algorithm or data, the cluster management host will deploy the Pod to the algorithm management host, and the related data will be deployed to the data management host. Deploying different types of workloads on different hosts can achieve resource isolation, which helps to prevent one overloaded workload from affecting another workload. Intensive data processing tasks will not interfere with algorithm calculation.

[0123] The cluster management host collects resource usage data of all nodes and Pods through MetricsServer, such as resource access volume and memory usage. The data is collected through the / metrics / resource interface of Kubelet. A Pod is a container collection in the content of a K8s cluster. In order to guarantee the load balancing of cluster resources, only one container is included in each Pod. The dynamic scheduler adjusts the resource allocation according to the preset rules and real-time monitoring data. Containers that generate burst traffic will be allocated more resources, and more Pod replicas will be generated according to the access volume to optimize resource usage of each module. The number of Pods is automatically adjusted according to real-time load and preset performance indicators to ensure load balancing of each node and improve overall performance and stability of the cluster.

[0124] The cluster management host first receives the Pod configuration information submitted by the user through the API Server, and then stores the cluster state through ETCD. As shown in FIG. 9 and FIG. 10, the automatic scaling controller monitors the resource usage of the Pod, executes the adaptive strategy based on the fusion of the Kafka message queue, VPA and HPA, and dynamically adjusts the computing resources and the number of Pod replicas according to the set threshold, wherein the Kafka consumer and the model algorithm are deployed in the Pod, so as to guarantee the processing efficiency and response speed.

[0125] Implementation steps:

[0126] Step 1: In the present application, the Kafka message queue is selected to be deployed outside the K8s cluster, and the persistent storage and high-availability configuration are set inside the K8s cluster. At the same time, the queue management computing task message is created, and the message retention policy and consumer group are set. The Kafka message queue is used as a client-to-server intermediate cache;

[0127] Step 2: The Kafka consumer module and the model algorithm are simultaneously deployed in the Pod node of the algorithm management host of the K8s cluster, the computing task data sent by the queue is imported into the model algorithm through the Kafka consumer module, the computing task message is pulled from the message queue, and the computing task resources are managed through VPA+HPA;

[0128] Step 3: The Deployment for monitoring the computing task service is specified, and the updateMode in the VPA configuration file is set to Auto;

[0129] Step 4: The VPA+HPA is used to automatically adjust the resource request and the number of the Pod according to the load.

[0130] The data management host is responsible for managing all data in the cluster and has high-speed data processing capability to meet high-concurrency data operations. In order to ensure that the data management host has high-speed data processing capability and fast data access, the present application is equipped with a server with powerful CPU and large amount of RAM and large-capacity SSD to provide faster read and write speed. At the same time, the network interface card supports more than 50 GbE, realizing the demand of high-concurrency data transmission. As shown in FIG. 12, the data management host establishes NFS as a data storage node, allowing multiple clients to access the same file system through the network, realizing the application of shared data. In the cluster environment, multiple Pods can simultaneously read and write to the same file system, without worrying about data synchronization and data consistency problems.

[0131] The readable storage medium based on the multi-specialty model integrated rock-fill dam deformation analysis device is applied to the multi-specialty model integrated rock-fill dam deformation analysis device.

[0132] The input data is processed in batches at each step of the rockfill dam deformation analysis, and then analyzed using integrated algorithms through system modules, with the results stored in NFS shared volumes for subsequent analysis and visualization. The implementation steps are as follows:

[0133] Step one: Create a PersistentVolume in the cluster, pointing to the NFS server. Build a container image for the algorithm model.

[0134] Step two: Create a PersistentVolumeClaim (PVC) to ensure that the algorithm calculation can use the PVC.

[0135] Step three: Deploy a Pod containing data analysis tools, which includes customized algorithms and commonly used data analysis software packages.

[0136] Step four: Run data analysis, load data from external sources or internal databases, perform analysis through integrated algorithms, and finally output the processing results to the mounted NFS storage / data.

[0137] All calculation outputs are centrally managed and stored on the data management host, which not only improves data security and accessibility, but also makes data backup and recovery simple and efficient. This architecture ensures the rockfill dam deformation analysis's need to handle large amounts of data and generate large amounts of output.

[0138] The algorithm management host has high computing power to handle algorithm operations, supports rapid deployment and updating of algorithms, and realizes isolation of the algorithm environment. As shown in Figure 13, the host generates an image for each algorithm and its necessary environment, generates a container based on the image, and deploys it to the production environment while uploading the image to the remote image repository for storage. Each Pod contains only one container, which contains professional algorithms and Kafka consumer function modules. To support efficient machine learning algorithm calculations, the host is equipped with professional GPUs to support algorithm training. The host only runs algorithm-related Pods. As shown in Figure 14, Deployment is used to deploy and manage algorithm applications, with 20 Pods allocated for each algorithm base. In subsequent calculation task submissions, the number of Pods is automatically increased or decreased based on the resource usage and algorithm access requirements of different algorithms, based on the resource allocation mechanism of the cluster management host. At the same time, the algorithm management host is equipped with algorithm update and rollback mechanisms to ensure algorithm updates and fault tolerance.

[0139] The rockfill dam deformation analysis requires the use of deep learning algorithms such as image recognition to quickly identify cracks on the dam surface. This algorithm needs to handle high-concurrency picture uploads and quickly provide classification results. The implementation steps are as follows:

[0140] Step one: Deploy the machine learning model, create a container image containing the machine learning model. The image contains all the necessary code and libraries, such as TensorFlow or PyTorch.

[0141] Step two: Create K8sDeployment at the same time, deploy the model in the algorithm management host.

[0142] Step three: Create Service and Ingress to ensure external access to the service.

[0143] Step four: Send pictures to the / recognize endpoint of the service through the Websocket protocol. The service will run the image recognition model on the GPU-accelerated node, and quickly return the classification result.

[0144] The multi-specialty model integrated rockfill dam deformation analysis equipment readable storage medium is applied to the data management host.

[0145] A chip system includes a processor for reading and executing a system program stored in a memory.

[0146] Embodiment one:

[0147] Please refer to FIG. 16 and FIG. 17. In an optional example, the application of the present application is combined with a certain high core wall rockfill dam deformation analysis engineering project.

[0148] Step one: Please refer to FIG. 5, FIG. 16 and FIG. 17. Calculate the rockfill dam constitutive parameters through the integrated method S1 process of the present application.

[0149] Input the dam filling material contractor and the triaxial test results of the detection, and determine the value range of the inversion parameters of each partition as shown in the table.

[0150] Considering that the number of target measurement points in each partition is large, the support vector machine model has low processing efficiency for large-scale training sets. Therefore, in step one of S1, the present example selects three machine learning algorithms, artificial neural network, gradient boosting machine and random forest, to train the proxy model. The system can automatically input the evaluation indicators, among which the smaller the mean absolute percentage error MAPE and the root mean square error RMSE or the larger the correlation coefficient R 2 The smaller the mean absolute percentage error MAPE and the root mean square error RMSE or the larger the correlation coefficient R

[0151] NSGA-III algorithm is selected for iteration optimization in S1 process step two. The input parameters in the system are as follows, in this example, the initial population size of the algorithm is set to 500, the maximum evolution generation is 1000; the mutation probability is 0.1, and the crossover and mutation parameters are 30 and 20 respectively. The final multi-objective optimization parameters are calculated, and the optimal constitutive model of the rockfill dam is shown in the following table.

[0152] Step two: please refer to FIG. 6, FIG. 16 and FIG. 17, the finite element deformation field of the rockfill dam is calculated by the integrated method S2 process of the application.

[0153] The calculation results of step one are used as the calculation data of step two, which are directly imported from the system. After inputting the basic text information (which can be ignored) of the finite element calculation, the calculation is started by clicking the calculation button, which changes from blue to yellow. When the button turns green, it means that the finite element calculation is completed. The finite element calculation of a certain high core wall rockfill dam is divided into 146692 grids, and the calculation results are shown in the table:

[0154] Step three: please refer to FIG. 7, FIG. 16 and FIG. 17, the reconstruction of the finite element deformation field of the rockfill dam is calculated by the integrated method S3 process of the application.

[0155] It is difficult to fully consider various engineering characteristics of rockfill materials in the deformation process through the macro constitutive model of rockfill body, and it is also difficult to accurately determine all the model parameters. Therefore, the deformation field of the rockfill dam calculated by the finite element module based on the constitutive model calculated in step two of the integrated method needs to be further optimized and reconstructed.

[0156] The calculation results of step two are used as the calculation data of step three, which are directly imported from the system. At the same time, the finite element calculation file and the monitoring settlement value file that need to be reconstructed are inputted, and the deformation field reconstruction model provided by the system is selected. The system provides three types of models including Gaussian model, spherical model and exponential model, and the deformation field reconstruction model calculation is three times. By comparing the calculation results of the three models, the optimal model is selected according to the return accuracy.

[0157] The calculation accuracy of each model is shown in the following table:

[0158] The Gaussian model is finally selected as the deformation field reconstruction calculation model, and the calculation period and monitoring time parameters are inputted to start the calculation to obtain the deformation field reconstruction results as shown in the table:

[0159] Experimental summary:

[0160] The final result selects a four-way verification method, randomly selects 25% of the monitoring points as verification points, and uses the average error and root mean square error indicators to evaluate the comprehensive error of the monitoring points. The error comparison is shown in the table:

[0161] The results show that the accuracy of the reconstructed deformation field of the rockfill dam calculated by the multi-specialty model integrated rockfill dam deformation analysis integration method is greatly improved compared to single method, and the calculation result more accurately reflects the actual deformation of the rockfill dam.

[0162] Example two:

[0163] Under the emergency construction period, the calculation task of the rockfill dam calculation system is submitted, two users are simulated to submit multiple calculation tasks at the same time, the background uses the Kafak message queue and VPA+HPA to cooperatively process high-flow events, solves the system overload problem caused by the submission of a large number of tasks of the rockfill dam under the emergency construction period, please refer to the attached drawing 10, and the load balancing result in the algorithm management host is shown in the drawing 11.

[0164] The balanced strategy execution result is as follows:

[0165] Experimental summary:

[0166] Through the above, it is concluded that the load balancing strategy under the load coordination of the present application can effectively handle the high load problem in each computing node when the amount of computing tasks is large. By decoupling the user operation and the computing task processing process, and the message queue buffers the task data under high load, combined with VPA automatically adjusting the Pod resources of the algorithm model node and HPA automatically adjusting the Pod replica number of the algorithm model node. When the algorithm model computing resource demand increases, more CPU and memory resources and more Pod replica data can be obtained to process multiple-user high-resource load computing tasks.

[0167] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An integrated method for rockfill dam deformation analysis based on multi-specialty model integration, characterized in that, The method comprises the following steps: S1, establishing a rockfill material constitutive model by using a parameter inversion analysis method; S2, performing finite element calculation based on the constitutive model to generate a rockfill dam deformation field; S3, reconstructing the rockfill dam deformation field by using a Bayesian Kriging interpolation method.

2. The multi-specialty model integrated rockfill dam deformation analysis integrated method of claim 1, wherein, The parameter inversion analysis method comprises the following steps: selecting an agent calculation model for training, the agent calculation model comprising an ANN neural network, a support vector machine SVR, a gradient boosting machine XGBOOST, and a random forest RF; performing parameter inversion by using a non-dominated sorting genetic algorithm to obtain a rockfill material constitutive model.

3. The multi-specialty model integrated rockfill dam deformation analysis integrated method of claim 1, wherein, The finite element calculation comprises the following steps: obtaining constitutive model parameters through containerized WebSocket full-duplex communication; performing finite element calculation of the rockfill dam deformation field by using a containerized ABAQUS platform.

4. The multi-specialty model integrated rockfill dam deformation analysis integrated method of claim 1, wherein, The Bayesian Kriging interpolation method comprises the following steps: obtaining finite element calculation results through containerized WebSocket full-duplex communication; reconstructing the rockfill dam deformation field by using Bayesian Kriging interpolation.

5. A rockfill dam deformation analysis system based on multi-specialty model integration, characterized in that, The system comprises: a parameter inversion analysis module for establishing a rockfill material constitutive model; a finite element calculation module for calculating a deformation field based on the constitutive model; a deformation field reconstruction module for reconstructing the deformation field by using Bayesian Kriging interpolation; a data processing unit for processing and caching data.

6. The rockfill dam deformation analysis system based on multi-specialty model integration of claim 5, wherein, The system is designed based on a micro-service architecture and is deployed by using containerization, the parameter inversion analysis module performs parameter inversion by using a multi-objective strategy, and the finite element calculation module is realized by containerizing and encapsulating an ABAQUS platform and performing secondary development.

7. The rockfill dam deformation analysis system based on multi-specialty model integration of claim 5, wherein, The data processing unit comprises: a processor for performing core calculation tasks of rockfill dam deformation analysis, including parameter inversion analysis of the constitutive model, finite element calculation, and reconstruction of the deformation field, establishing WebSocket full-duplex communication between the client and the algorithm module and the database; a cache connected to the processor, which is used to temporarily store calculation data and intermediate results, improve data transmission quality and ensure the stability of the calculation process, and reduce the risk of system overload.

8. A device for rockfill dam deformation analysis based on multi-specialty model integration, characterized in that, The rockfill dam deformation analysis system based on multi-specialty model integration according to any one of claims 5-7 comprises: a cluster management host for responsible for resource allocation and management of the entire device network, and responsible for managing the methods and data allocation in the system of claims 5-7, and supporting the addition of existing algorithm models; a data management host connected to the cluster management host, which is used for the persistent storage of calculation data and historical data; an algorithm management host connected to the cluster management host, which is used for managing and executing various algorithms involved in the system.

9. A rockfill dam deformation analysis apparatus readable storage medium based on multi-specialty model integration, characterized in that, The data management host according to claim 8 is applied.

10. A chip system, characterized by The rockfill dam deformation analysis system based on multi-specialty model integration according to any one of claims 5-7 comprises a processor for reading and executing system programs stored in the memory.

Images