Digital twin-oriented multi-dimensional data fusion analysis method and device

Abstract

The invention provides a digital twinborn-oriented multi-dimensional data fusion analysis method and device, and relates to the technical field of data analysis, and the method comprises the steps: constructing M fault prediction models in a target power plant; performing multi-level data fusion modeling on the plurality of power plant systems; loading the M fault prediction models to a plurality of system-level digital twin bodies; performing multi-stage failover analysis on the target power plant; performing fault conduction rendering on the initial twin of the power plant according to the fault state transition diagram of the power plant to obtain a digital twin of the power plant; and loading the M pieces of real-time equipment operation data to the M fault prediction models, and after M equipment operation states are obtained, carrying out visualization on the M equipment operation states in the power plant digital twin. Through the method and the device, the technical problem of low operation and maintenance efficiency of the power plant due to lack of effective fusion of multi-dimensional data in the prior art can be solved, and the operation and maintenance efficiency of the power plant is improved by constructing the digital twin of the power plant to realize visualization of the operation state of the equipment.

Description

Technical Field

[0001] The present application relates to the field of data analysis technology, and in particular to a multidimensional data fusion analysis method and device for digital twins. Background Art

[0002] As the scale of modern power plants and other complex industrial systems continues to expand, the number of equipment is increasing, and the operating conditions are becoming more and more complex, traditional operation and maintenance management methods can no longer meet the needs of efficiency and real-time. As an innovative equipment management method, digital twin technology can map the operating status of actual equipment to a digital platform through a virtual model, providing real-time visualization. However, most existing digital twins focus on virtual mapping at the equipment level, lacking integrated processing of cross-device and multi-dimensional data, and are prone to problems such as equipment safety distance exceeding the limit and inaccurate fault location, which in turn affects the operation and maintenance management efficiency of power plants.

[0003] In summary, the existing technology has the technical problem that the power plant system involves complex multi-source heterogeneous data and lacks deep integration across devices and systems, resulting in low power plant operation and maintenance efficiency. Summary of the invention

[0004] The purpose of this application is to provide a multidimensional data fusion analysis method and device for digital twins, so as to solve the technical problem in the prior art that the power plant system involves complex multi-source heterogeneous data and lacks deep integration across devices and systems, resulting in low power plant operation and maintenance efficiency.

[0005] In view of the above problems, the present application provides a multi-dimensional data fusion analysis method and device for digital twins.

[0006] In the first aspect, the present application provides a multidimensional data fusion analysis method for digital twins, which is implemented by a multidimensional data fusion analysis device for digital twins, wherein the multidimensional data fusion analysis method for digital twins includes: using an LSTM network to construct M fault prediction models for M power plant equipment in a target power plant, wherein the M power plant equipment have M equipment source identifiers; after performing multidimensional data collection on multiple power plant systems to obtain multiple system-level multidimensional data, using the multiple system-level multidimensional data to perform multi-level data fusion modeling to generate multiple a system-level digital twin; according to the M device source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant; a multi-level fault transfer analysis is performed on the target power plant to obtain a power plant fault state transfer diagram; according to the power plant fault state transfer diagram, fault conduction rendering is performed on the initial twin of the power plant to obtain a power plant digital twin; after loading the M real-time device operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M device operation states, the M device operation states are visualized in the power plant digital twin.

[0007] In the second aspect, the present application also provides a multidimensional data fusion analysis device for digital twins, which is used to execute a multidimensional data fusion analysis method for digital twins as described in the first aspect, wherein the multidimensional data fusion analysis device for digital twins includes: a fault prediction model construction module, which is used to use an LSTM network to construct M fault prediction models for M power plant equipment in the target power plant, wherein the M power plant equipment have M equipment source identifiers; a digital twin modeling module, which is used to collect multidimensional data from multiple power plant systems and obtain multiple system-level multidimensional data, and then use the multiple system-level multidimensional data to perform multi-level data fusion modeling to generate multiple system-level digital twins; a power plant twin construction module, which is used to build a fault prediction model for M power plant equipment in the target power plant using an LSTM network, wherein the M power plant equipment has M equipment source identifiers; a digital twin modeling module, which is used to collect multidimensional data from multiple power plant systems and obtain multiple system-level multidimensional data, and then use the multiple system-level multidimensional data to perform multi-level data fusion modeling to generate multiple system-level digital twins. The body construction module is used to load the M fault prediction models into the multiple system-level digital twins according to the M device source identifiers to complete the construction of the initial twin of the power plant; the fault transfer analysis module is used to perform multi-level fault transfer analysis on the target power plant to obtain the power plant fault state transfer diagram; the fault conduction rendering module is used to perform fault conduction rendering on the initial twin of the power plant according to the power plant fault state transfer diagram to obtain the power plant digital twin; the operation status visualization module is used to load the M real-time device operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin to obtain the M device operation status, and then visualize the M device operation status in the power plant digital twin.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: By adopting the LSTM network, M fault prediction models of M power plant equipment in the target power plant are constructed, wherein the M power plant equipment has M equipment source identifiers; after multi-dimensional data collection is performed on multiple power plant systems to obtain multiple system-level multi-dimensional data, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate multiple system-level digital twins; according to the M equipment source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant; a multi-level fault transfer analysis is performed on the target power plant to obtain a power plant fault state transfer diagram; according to the power plant fault state transfer diagram, fault conduction rendering is performed on the initial twin of the power plant to obtain a power plant digital twin; after loading the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M equipment operation status, the M equipment operation status is visualized in the power plant digital twin. That is to say, by constructing LSTM fault prediction models for multiple devices respectively, multi-level data fusion is used to model and fuse multiple system-level multidimensional data, multiple system-level digital twins are generated, and the established fault prediction models are integrated into the digital twins to achieve real-time monitoring and optimization of the entire power plant system. Fault conduction rendering is performed through multi-level fault transfer analysis, and the real-time operation data of the equipment is loaded into the fault prediction model for status visualization. The latest status of each device is displayed in the digital twin, and the equipment operation status is visualized, thereby improving the operation and maintenance efficiency of the power plant.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. 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 application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0011] Figure 1 A flowchart of a multidimensional data fusion analysis method for digital twins in this application; Figure 2This is a structural schematic diagram of a multidimensional data fusion analysis device for digital twins in this application.

[0012] Explanation of the accompanying drawings: fault prediction model construction module 11, digital twin modeling module 12, power plant twin construction module 13, fault transfer analysis module 14, fault conduction rendering module 15, operation status visualization module 16. DETAILED DESCRIPTION

[0013] This application provides a multi-dimensional data fusion analysis method and device for digital twins, which solves the technical problem in the prior art that the power plant system involves complex multi-source heterogeneous data and lacks deep integration across devices and systems, resulting in low power plant operation and maintenance efficiency. By constructing LSTM fault prediction models for multiple devices respectively, multi-level data fusion is used to model and fuse multiple system-level multi-dimensional data, generate multiple system-level digital twins, integrate the established fault prediction model into the digital twin, realize real-time monitoring and optimization of the entire power plant system, perform fault conduction rendering through multi-level fault transfer analysis, load the real-time operation data of the equipment into the fault prediction model, perform status visualization, and display the latest status of each device in the digital twin, realize the visualization of the equipment operation status, and improve the operation and maintenance efficiency of the power plant.

[0014] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0015] For example, please refer to the attached Figure 1 The present application provides a multidimensional data fusion analysis method for digital twins, wherein the multidimensional data fusion analysis method for digital twins is applied to a multidimensional data fusion analysis device for digital twins, and the multidimensional data fusion analysis method for digital twins specifically includes the following steps: S100: constructing M fault prediction models for M power plant equipment in a target power plant using an LSTM network, wherein the M power plant equipment have M equipment source identifiers.

[0016] Specifically, there are M power plant equipment in the target power plant, and each equipment has different working environments, failure modes, and operating characteristics. In order to build personalized prediction models for different equipment, it is necessary to build a separate LSTM network model for each equipment, so that the fault prediction capability of each equipment will be more accurate. Each device will have a unique device source identifier, which can be the device ID number, device type, device location, etc., which is used to distinguish data from different devices and ensure that the data source of each device is correctly identified and processed. During model training and prediction, these identifiers are used to index the data corresponding to each device to ensure that the model does not confuse the operating characteristics of different devices.

[0017] For each device, prepare the corresponding historical fault operation data. The historical fault data of each device contains multiple features (such as temperature, pressure, vibration, load, etc.), which can be used to predict the fault trend and fault type of the device. According to the node where the fault occurred, the historical fault data of each device is segmented, and the segmentation results are normalized using a dynamic slice window to obtain M groups of sample fault operation data. For each power plant device, define an LSTM network structure, including the number of layers, the number of neurons, the activation function, etc. The size of the input layer should be consistent with the size of the sliding window. Set up multiple LSTM layers to capture long-term dependencies in time series data. The size of the output layer depends on the prediction task, including the fault trend and fault type of the equipment.

[0018] The data set of each device is divided into a training set, a validation set, and a test set. The training set data is used to train the LSTM model of each device, and the validation set is used to adjust the model parameters, such as the learning rate, batch size, etc., and the test set is used to evaluate the performance of the model, such as accuracy, recall, etc. The specific training process is explained in detail in the subsequent refinement of the corresponding steps. For the sake of brevity in the manual, it is only briefly introduced here. According to the M groups of sample fault operation data, the construction of M fault evolution prediction units and M fault type identification units is completed, and M fault prediction models are cascaded. By independently constructing a fault prediction model for each device, customized predictions can be made based on the working characteristics and failure modes of the equipment, and maintenance or replacement of equipment can be carried out in advance based on the fault prediction results, avoiding unnecessary downtime, improving equipment utilization, and reducing maintenance costs.

[0019] S200: After performing multi-dimensional data collection on multiple power plant systems to obtain multiple system-level multi-dimensional data, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate multiple system-level digital twins.

[0020] Specifically, the target power plant includes multiple power systems, that is, multiple systems with different functions or areas, each of which is responsible for a part of the power plant's functions, such as power generation system, cooling system, control system, etc. Each power plant system contains multiple devices, and the operating data of each device is closely related to the overall performance of the system. Data is collected from multiple power plant systems through sensors, equipment monitoring systems, environmental monitoring systems, and other methods to obtain multiple system-level multidimensional data, including point cloud data, real-scene data, power plant design information, etc. corresponding to each power system.

[0021] Multi-level data fusion modeling is performed on multiple system-level multidimensional data. That is to say, on the basis of multiple system-level multidimensional data, different data sources from various systems and equipment are integrated through data processing and fusion at different levels. Through the comprehensive processing of these multi-level data, an efficient model reflecting the overall status of the power plant is constructed.

[0022] Since the power plant system is large and complex, a unified large digital twin may be difficult to process the data of all devices and systems. In order to improve computing efficiency and support real-time data updates, the digital twin can be split into multiple sub-twins, each of which is responsible for the modeling and simulation of a specific subsystem or device. A distributed computing architecture is adopted, that is, a large digital twin is decomposed into multiple small, independent sub-twins, each of which is responsible for the modeling and simulation of a specific power system or subsystem. Each subsystem can run independently on different servers or computing nodes, and these servers exchange data and work together through high-speed networks. In short, each power plant system corresponds to a digital twin, which is constructed through the multidimensional data corresponding to the power plant system. The corresponding subordinate right in this step takes the power generation system as an example to construct the digital twin. The construction of digital twins for other power systems in the target power plant is similar to that of the power generation system, and no repeated explanation is given.

[0023] Through the above steps, multiple sub-twins are processed and integrated to form multiple system-level digital twins, providing power plants with real-time monitoring, performance analysis, fault diagnosis and other functions for each subsystem or the entire power plant. Through the distributed computing architecture, the huge data processing tasks of the power plant can be distributed to multiple computing nodes for parallel processing, which greatly improves the computing efficiency and response speed, allowing the power plant system to update its digital twin in real time to reflect the latest operating status. The distributed computing architecture makes the digital twin of the power plant have good scalability. With the increase of power plant equipment and systems, new sub-twins can be easily added without large-scale modification or reconstruction of the existing system. Each sub-twin can display the equipment status, operating efficiency and key performance indicators through a visual interface, helping operation and maintenance personnel to understand the operating status of the power plant in a timely manner and make corresponding decisions, which not only improves the operation and maintenance efficiency, but also effectively prevents failures and reduces equipment downtime.

[0024] S300: According to the M device source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant.

[0025] Further, the present application S300 includes: A fault prediction complexity analysis is performed based on the M historical fault operation data, and multiple system-level prediction complexities are calculated and output based on the analysis results; with reference to the multiple system-level prediction complexities, computing nodes are allocated to the multiple power plant systems in a distributed computing architecture to obtain K groups of power systems with K computing nodes; based on the M equipment source identifiers, after the M fault prediction models are loaded into the multiple system-level digital twins, the K groups of power systems are used as replication guides, and the multiple system-level digital twins are replicated to the K computing nodes to complete the construction of the initial twin of the power plant.

[0026] Specifically, complexity analysis is performed on M historical fault operation data to evaluate the difficulty and complexity of fault prediction. For example, by statistically analyzing the frequency, type, and distribution of faults, the amount of data that needs to be processed is predicted, as well as the complexity problems that the model may encounter during operation (such as too large data dimensions, too long calculation time, etc.), to evaluate the difficulty of fault prediction. Based on the analysis results, multiple system-level prediction complexities are calculated and output. Based on the analysis results of multiple system-level prediction complexities, computing nodes are allocated to each power plant system in the distributed computing architecture. Based on the prediction complexity of each power plant system, computing resources are dynamically allocated. For systems with higher complexity, more computing nodes are allocated, and for systems with lower complexity, fewer computing nodes are allocated, which helps to improve resource utilization and reduce computing time. Based on this, K computing nodes of K groups of power systems are obtained.

[0027] According to the source identification of the equipment, M fault prediction models are loaded into multiple system-level digital twins. The fault prediction model of each device corresponds to its source identification, ensuring that a personalized prediction model is provided for each power plant equipment. Subsequently, according to the K groups of power systems, multiple system-level digital twins are copied to the corresponding K computing nodes, and each node will run the corresponding fault prediction model. The constructed multiple system-level digital twins are distributed to K computing nodes according to the K groups of power systems of the power plant. Each computing node runs one or more groups of digital twins of the power plant system, and performs collaborative computing through a distributed computing architecture to ensure efficient cooperation between computing nodes and complete real-time modeling and simulation of the power plant system.

[0028] Through the above steps, the construction of the initial digital twin of the power plant is completed. Each computing node simulates the operating status, fault prediction, performance evaluation, etc. of the power plant in real time by running the digital twin model. The digital twin of the entire power plant works together through distributed computing nodes to form a comprehensive and accurate power plant system simulation model. The distributed computing architecture enables the system to work in parallel on multiple nodes, ensuring that when some nodes fail, other nodes can still work normally, thereby enhancing the stability and reliability of the digital twin of the entire power plant.

[0029] S400: Perform multi-level fault transfer analysis on the target power plant to obtain a power plant fault state transfer diagram.

[0030] Further, the present application S400 includes: Perform fault cross-system conduction analysis on the target power plant to obtain a system-level state transition diagram; perform fault cross-equipment conduction analysis on the target power plant based on the M historical fault operation data to obtain multiple equipment-level state transition diagrams for multiple power plant systems; and obtain a power plant fault state transition diagram by performing multi-dimensional data fusion on the system-level state transition diagram and multiple equipment-level state transition diagrams.

[0031] Specifically, according to the equipment and system structure of the target power plant, a fault cross-system conduction analysis is conducted. By modeling the interaction between equipment and the fault conduction path in multiple power systems, the chain of possible fault conduction between equipment is identified. For example, a boiler failure may cause a steam supply interruption, which in turn affects the normal operation of the steam turbine. Fault cross-system conduction analysis is to analyze the fault conduction mechanism between different power systems and study how the operation of other systems is affected when a system fails. Since the target power plant contains multiple independent systems, the equipment in each system may also cause a chain reaction due to failures, causing other equipment or systems to fail. The system-level state transition diagram represents the transition process between different states of various systems in the power plant (such as power generation system, transmission system, etc.). Each state represents a working state of the system (such as normal, fault, repair, etc.), and the transition represents the process of the system transferring from one state to another.

[0032] The device-level state transition diagram is similar to the system-level state transition diagram, except that the device-level state transition diagram is a state transition analysis for specific equipment in the power plant (such as boilers, steam turbines, transformers, etc.). Each power plant equipment will have different states under different working conditions, and the transition between these states is triggered by the working conditions and fault conditions of the equipment. Perform fault conduction analysis on each power plant equipment (such as boilers, steam turbines, transformers, etc.) and establish a device-level state transition diagram. Each device may have different behavior patterns and fault transmission mechanisms in different states (such as normal, warning, fault, etc.).

[0033] Through multi-dimensional data fusion, the data of the device-level state transition diagram and the system-level state transition diagram are integrated to obtain the power plant fault state transition diagram. In the fusion process, it is necessary to combine the fault data, operating status data and historical fault data from different systems and equipment to obtain a comprehensive fault conduction path. After completing the data fusion, the power plant fault state transition diagram is generated, which integrates the fault state transfer information of multiple devices, shows the mutual influence between devices, the fault conduction path between systems, and the possible fault transfer process of the entire power plant during operation. By constructing the power plant fault state transfer diagram, we can fully understand the mutual influence and fault conduction path between various systems and equipment, and can respond quickly when a fault occurs and make rapid adjustments according to the fault state transfer diagram to minimize downtime and economic losses.

[0034] S500: Perform fault conduction rendering on the initial twin of the power plant according to the fault state transition diagram of the power plant to obtain a digital twin of the power plant.

[0035] Specifically, the constructed power plant fault state transfer diagram is rendered onto the initial twin of the power plant to obtain the power plant digital twin. In other words, each fault state, equipment state and its transfer path in the power plant fault state transfer diagram are also integrated into the initial twin of the power plant, and the fault conduction process is represented in the digital twin. According to the data in the fault state transfer diagram, a specific device or system is selected as the starting point of the fault in the digital twin. According to the fault state transfer rules of the device, the fault conduction process is simulated in the digital twin. Through graphical rendering technology, the fault conduction process will be presented in a visual way in the digital twin. The rendering engine can display information such as the working status of the equipment, the time node of the fault occurrence, and the fault conduction path. Fault conduction simulation is performed in the digital twin to help power plant operation and maintenance personnel identify key equipment and fault conduction paths, and timely discover potential fault chains. The power plant digital twin is a virtual copy of the power plant constructed by digital technology, which simulates the structure, equipment, system and operating status of the power plant, can reflect the operating status of the power plant, the health status of the equipment and fault information in real time, and can be used for fault warning, diagnosis, maintenance and optimization. By rendering the fault conduction of the initial twin of the power plant, the conduction process of the fault state can be simulated and displayed in the digital twin, and real-time visualization of various systems and equipment of the power plant can be performed, helping the power plant to improve the efficiency of fault diagnosis and emergency response.

[0036] S600: After loading the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M equipment operation statuses, the M equipment operation statuses are visualized in the power plant digital twin.

[0037] Furthermore, the present application S600 includes: Perform pixel data collection on the M power plant equipment to obtain the M real-time equipment operation data; in the power plant digital twin, use the M fault prediction models to perform fault feature analysis on the M real-time equipment operation data, and output the M equipment operation statuses; perform fault feature aggregation on the M historical fault operation data to obtain M groups of fault operation performances; use the M equipment operation statuses to traverse the M groups of fault operation performances and schedule M real-time operation performances; use the real-time rendering engine to visualize the M equipment operation statuses of the M real-time operation performances in the power plant digital twin; perform fault transfer analysis on the M real-time operation performances in the power plant fault state transfer diagram, and visualize the fault state transfer trend of the analysis results in the power plant digital twin.

[0038] Specifically, pixel data collection is performed on M power plant equipment, that is, various real-time data of M power plant equipment is collected through sensors and other equipment to obtain M real-time equipment operation data, including temperature, pressure, load, vibration, etc. In the power plant digital twin, the M fault prediction models corresponding to the M power plant equipment are used to analyze the fault characteristics of the M real-time equipment operation data, determine the equipment failure risk, and predict the possible failure time, and obtain the M equipment operation status, including the failure risk (the possibility of failure in the current operation state) and the predicted failure time (the predicted time of failure).

[0039] Aggregate the fault characteristics of M historical fault operation data and analyze the performance of the equipment in the historical faults, including the equipment's working parameters and the time point when the fault occurred. Fault operation performance refers to the performance of the equipment when a fault occurs, including the changes in the equipment's working data and the characteristics of the fault, and presents the abnormal situation of the equipment when the fault occurs in a graphical or pictorial manner. M groups of fault operation performance refer to the external performance of the equipment at different fault stages obtained through the analysis of historical fault data, such as the different characteristics of the equipment in the early, middle and late stages of the fault. According to the operating status of the M devices, traverse the M groups of fault operation performance and schedule the real-time operation performance to monitor the status changes and potential fault risks of the equipment.

[0040] Using a real-time rendering engine, M real-time operating tables are displayed in the power plant digital twin, and the status of the equipment is displayed in a visual form, which helps operation and maintenance personnel to intuitively understand the operating status of each device. Fault transfer analysis is performed on the real-time operating performance of the equipment, and the fault propagation path of the equipment and system is modeled and analyzed through the fault state transition diagram to identify how the fault may be transmitted from one device to another. The results of the fault state transition analysis are presented in the digital twin, and the evolution of the fault state is displayed in a dynamic graphic. For example, the transition process of the equipment from normal state to warning state and then to fault state will be displayed in the visual interface, helping operation and maintenance personnel to better diagnose and prevent faults.

[0041] Through real-time equipment operation data collection, fault feature analysis and equipment operation status output, historical fault data analysis and real-time operation performance scheduling, equipment operation status visualization and fault status transfer trend analysis, the operation status of power plant equipment can be fully monitored and managed, potential fault risks can be discovered and responded to in a timely manner, and the operation and maintenance efficiency and reliability of the power plant can be improved.

[0042] Furthermore, the present application S100 includes: Equipment data is collected from the target power plant to obtain M historical fault operation data of M power plant equipment; after segmenting the M historical fault operation data based on the fault occurrence node, a sliding slice window is used to normalize the segmentation results to obtain M groups of sample fault operation data, wherein the sliding slice window has K sample segmentation steps; an LSTM network is used to construct M fault evolution prediction units of the M power plant equipment; a CNN network is used to construct M fault type identification units of the M power plant equipment; according to the K sample segmentation steps, the M groups of sample fault operation data are decomposed, and the decomposition results are used to optimize the parameters of the M fault evolution prediction units and the M fault type identification units; the M fault evolution prediction units and the M fault type identification units after mapping cascade optimization are constructed to complete the construction of the M fault prediction models.

[0043] Specifically, data collection is performed on the equipment of the target power plant to obtain the fault operation data of M power plant equipment and obtain M historical fault operation data, including the equipment's operating status, environmental conditions, historical fault records, sensor data, etc. The target power plant is a complex facility composed of multiple interrelated power systems, each of which contains multiple power equipment, such as the power generation system including boilers, steam turbines, generators, etc., the transmission system including transformers, cables, distribution equipment, etc., the cooling system including cooling towers, water pumps, etc., and the auxiliary system including pump stations, fans, compressors, etc.

[0044] For the historical fault operation data of M devices, the data is segmented according to the node where the fault occurred. The fault node is the fault occurrence point marked in the historical fault data. Whenever a device fails, the time node of the fault, the fault type, and the fault impact are recorded. The segmentation results are normalized using a sliding slice window to convert each data slice to a unified scale for subsequent analysis. The K sample segmentation step is the number of samples spanned each time the sliding window slides. K is the window size, that is, the length of data considered each time the slice is sliced ​​(usually the number of samples in the time series). The sliding slice window slides in the data segment according to the given step size and is divided into multiple small time windows (samples). These small windows will be normalized to ensure that the data is compared and trained at the same scale. The sliding slice window is a time series data processing method that segments the time series data by setting the window size (for example, K samples) and continuously sliding the window in the data series. Each slide will generate a new window to process the data in it.

[0045] Each sample data after cutting is normalized. The purpose of normalization is to eliminate the dimensional differences between different devices and different measurement data, map the data to the [0,1] interval, or standardize the data to have a mean of 0 and a variance of 1. After sliding slice window segmentation and normalization, M groups of sample fault operation data of M devices are finally obtained. Each group of samples represents the status data of a device in a specific time period, and the features of each sample have been standardized, which is suitable for training prediction models.

[0046] For each device (a total of M devices), a fault evolution prediction unit is constructed using an LSTM network. The LSTM network can learn the time evolution law of equipment faults based on historical data (including fault evolution data and steady-state data) and predict the fault evolution process of the equipment in the future. Taking the first fault evolution prediction unit as an example, the corresponding first group of sample fault operation data is obtained from the M groups of sample fault operation data, where the first group of historical fault operation data includes the first group of sample fault evolution data and the first group of sample fault steady-state data, including various sensor data (such as temperature, pressure, vibration, etc.) when the equipment fails and the stable operation state data after the equipment fails.

[0047] The first set of sample fault evolution data and the first set of sample fault steady-state data are used as training data for the LSTM network to train the LSTM model and help the LSTM learn the evolution process after the equipment fault occurs. The training data will form input and output according to the time series, and the LSTM model will learn the time dependency of these data to predict the development trend of future faults. During the training process, the LSTM network will adjust the network weights according to the input data (fault evolution data and steady-state data) to better predict the future fault evolution path of the equipment. The generalization ability of the model is evaluated using the test set, and the prediction error indicators such as the mean square error are calculated. The model is then adjusted and optimized until the model converges. The model at this time is used as the first fault evolution prediction unit. M sets of sample fault operation data are used as training data to complete the construction of M fault evolution prediction units for M power plant equipment.

[0048] Extract the first set of sample fault types from the first set of sample fault steady-state data. The sample fault type refers to the characteristic pattern or type exhibited by the equipment after the fault occurs. Each fault type usually has unique operating characteristics, such as overheating, overload, abnormal vibration, etc., which helps to identify the nature and severity of the fault. CNN is a deep learning algorithm that is widely used in image recognition and classification tasks. In fault type identification, CNN can automatically learn the important features in the equipment fault data and classify different fault types based on these features. According to the first set of sample fault steady-state data, the model structure of CNN is designed, including input layer, convolution layer, pooling layer, and fully connected layer. The input layer is designed according to the dimension of the fault data, and multi-channel input is set according to the input multi-dimensional signal; the size and number of convolution kernels are adjusted according to the needs, and multiple convolution layers are used to extract fault features; a pooling layer (such as maximum pooling) is added after the convolution layer to reduce the feature dimension and extract key information; finally, a fully connected layer is added, and the number of output nodes corresponds to the number of categories of the fault type.

[0049] Initialize the appropriate batch size (such as 32 or 64), training rounds (such as 100 times), and learning rate, and use the first set of sample fault steady-state data and the first set of sample fault types as training data, and input them into the CNN network for training. During the training process, the CNN network will extract the key features in the input data through multiple layers of convolution and pooling layers, and classify the fault types based on these features. Adjust and optimize the model structure and parameters according to the problems encountered in practical applications. Use the validation set to monitor the performance changes during the model training process to prevent overfitting. Deploy the trained CNN model as the first fault type recognition unit, receive new fault steady-state data, and output the type of the equipment fault based on the previously trained model. Use M sets of sample fault operation data as training data to complete the construction of M fault type recognition units for M power plant equipment.

[0050] According to K sample segmentation steps, M groups of sample fault operation data are decomposed. Fault operation data is usually time series data, including the historical operation status of the equipment, such as temperature, pressure, vibration and other parameters. Through the sliding slice window, according to the set step size K, the original time series data is divided into multiple samples, each sample contains K data points. According to the decomposition results, the M fault evolution prediction units (LSTM network) and M fault type identification units (CNN network) are optimized. The purpose of parameter optimization is to improve the prediction accuracy by adjusting the hyperparameters of the model. The parameter adjustment process includes selecting appropriate learning rate, optimizer, number of network layers, number of neurons and other parameters to ensure that the model can effectively learn the features in the data.

[0051] Parameter optimization refers to adjusting the model's hyperparameters (such as learning rate, number of network layers, number of neurons, etc.) during model training to improve model performance. Hyperparameter tuning can be done through methods such as grid search, random search, or Bayesian optimization, with the goal of finding a set of most appropriate parameter configurations to maximize the model's prediction accuracy and robustness. Decomposition results refer to the sub-datasets generated by segmenting multiple fault operation data (for example, using a sliding window method) during data processing, which contain information from different time periods of the original data. For example, in the parameter optimization of the LSTM network, the learning rate is adjusted (such as from 0.001 to 0.01) to select the learning rate that best suits the current data set. Then adjust the number of layers and neurons of the LSTM network and test different network configurations to improve the accuracy and robustness of the model. Similarly, in the CNN network, hyperparameters such as the size of the convolution kernel, the number of pooling layers, and the number of convolution layers can be adjusted.

[0052] After completing the parameter optimization of the fault evolution prediction unit and the fault type identification unit, the two optimized models are cascaded. Cascade optimization means jointly mapping the fault evolution prediction unit and the fault type identification unit after parameter optimization to form a complete fault prediction model. Cascade optimization refers to combining multiple optimized models and cooperating with each other through the optimization process to build the final fault prediction model. In the fault prediction task, the LSTM model is responsible for predicting the evolution process of equipment faults (such as how temperature and pressure change over time), while the CNN model is responsible for identifying the specific type of equipment faults (such as overheating or mechanical failure). The combination of the two can achieve comprehensive fault prediction, that is, it can know when the equipment failure will occur and determine the type of fault. Through the above steps, the construction of M fault prediction models for M devices is finally completed. The fault prediction model of each device consists of a fault evolution prediction unit (LSTM network) and a fault type identification unit (CNN network). These two modules work together to accurately predict the operating status of the equipment and help power plants make fault warnings and operation and maintenance decisions.

[0053] The combined use of LSTM and CNN networks can fully utilize the time dependency and feature information in the equipment fault history data to provide more accurate fault prediction and type identification. The LSTM network can predict the evolution trend of faults, while CNN can accurately classify fault types. The parameter optimization of the model improves the robustness and stability of the fault prediction model. Through accurate fault prediction and type identification, potential faults can be discovered in a timely manner, sudden equipment shutdowns can be reduced, the reliability and stability of power plant equipment can be improved, and safe and efficient operation of power plants can be ensured.

[0054] Furthermore, the present application also includes the following steps: The first group of sample fault operation data is decomposed into a first group of sample fault evolution data and a first group of sample fault steady-state data, wherein the time span of the sample fault evolution data is K-1 sample segmentation steps; the first group of sample fault evolution data and the first group of sample fault steady-state data are used as training data to construct a first fault evolution prediction unit for the first power plant equipment; after extracting the first group of sample fault types of the first group of sample fault steady-state data, the first group of sample fault steady-state data and the first group of sample fault types are used as training data to construct a first fault type identification unit for the first power plant equipment; the first fault evolution prediction unit and the first fault type identification unit are cascaded to complete the construction of the first fault prediction model; and so on, the M historical fault operation data are used as training data, and the M fault prediction models are constructed using an LSTM network.

[0055] Specifically, according to the first set of sample fault operation data, it is decomposed into the first set of sample fault evolution data and the first set of sample fault steady-state data, where the sample fault evolution data refers to the data of the state change of the equipment over time after the fault occurs, reflecting the fault degree and health status of the equipment at different time points; the sample fault steady-state data is the stable operation state data of the equipment after a period of time. Although the equipment has failed, its state is relatively stable and no longer changes dramatically. The time span of the sample fault evolution data is K-1 sample segmentation steps, that is, a longer time period of data is determined by a sliding window to accurately capture the evolution and steady-state characteristics of the equipment fault. The time span is based on the setting of the sliding slice window, and K sample segmentation steps minus 1 are used as the time range of each evolution data set. For example, if K is set to 100, the time span is 99 steps (that is, the process data from the initial stage of the fault to its stable stage).

[0056] The first set of sample fault evolution data and the first set of sample fault steady-state data are used as training data to train the LSTM network model to obtain the first fault evolution prediction unit. The LSTM network can effectively learn the law of equipment fault evolution through its time series modeling ability, and then predict the evolution process of equipment faults. The specific training process is described in detail in the construction steps of the aforementioned M fault evolution prediction units, which is similar to this.

[0057] The first group of sample fault types are extracted from the first group of sample fault steady-state data, including overheating, wear, mechanical failure, etc. The extracted fault types are used as labels and together with the fault steady-state data as training data to train the CNN network and then construct a fault type recognition unit. CNN processes the data through convolutional layers and pooling layers to identify different fault modes. The specific training process is described in detail in the construction steps of the aforementioned M fault type recognition units, which is similar to this.

[0058] After completing the training of the LSTM network and the CNN network, the two units are cascaded, and the comprehensive prediction of equipment failures is achieved through the joint action of the two models. The evolution process of equipment failures and the identification of fault types are combined to form a complete prediction system. The above steps are repeated continuously, and the M historical fault operation data are segmented according to the fault occurrence node, and the segmentation results are normalized using a sliding slice window to obtain M groups of sample fault operation data. The M groups of sample fault operation data are used as training data to build M fault prediction models for M power plant equipment, thereby building a dedicated fault prediction model for each device. The failure mode of each device is different. By training a fault prediction model for each device separately, a personalized prediction model can be generated based on the historical fault data of the device, avoiding a one-size-fits-all approach and improving the pertinence and effectiveness of the prediction.

[0059] Further, the present application S200 includes: The point cloud data of the power generation system is obtained by laser scanning the power generation system; the real scene data of the power generation system is obtained by photogrammetry of the power generation system; after interactively obtaining the power plant design information of the target power plant, a three-dimensional design model of the power generation system is constructed according to the power plant design information, and the point cloud data of the power generation system is used to calibrate the model posture to obtain the power generation system space model, wherein the point cloud data of the power generation system, the real scene data of the power generation system and the power plant design information constitute the multi-dimensional data of the power generation system; SIFT is used to perform feature matching on the spatial model of the power generation system and the real scene data of the power generation system to obtain the real scene space alignment coordinates; according to the real scene space alignment coordinates, the real scene data of the power generation system and the spatial model of the power generation system are virtual-real fused, and a real-time rendering engine is introduced to complete the construction of the digital twin of the power generation system; and by analogy, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate the multiple system-level digital twins.

[0060] Specifically, laser scanning is performed on multiple power generation systems of the target power plant to obtain point cloud data of the power generation system. Laser scanning is a method of obtaining high-precision three-dimensional data on the surface of an object through laser technology. For example, laser radar generates high-density point cloud data by emitting laser beams and receiving reflected signals. The point cloud data of the power generation system is a set of data points with three-dimensional coordinates obtained by laser scanning or other three-dimensional scanning methods. Each point represents a position on the surface of the scanned object, and point cloud data is often used for three-dimensional modeling and digital representation. At the same time, photogrammetry is performed on the power generation system through photography technology, usually using photos from multiple angles, and reconstructing the three-dimensional information of the object through image processing technology. Photogrammetry is performed to obtain real-life data of the power generation system, including the appearance of the equipment, installation location, surrounding environment, etc.

[0061] By interacting with power plant designers or operation and maintenance personnel, the design information of the target power plant, including equipment specifications, installation locations, configurations, system drawings, etc., can be obtained, and detailed data on the overall structure of the power plant can be provided. A three-dimensional design model of the power generation system is constructed based on the power plant design information to ensure that the digital model is consistent with the design documents. A three-dimensional design model refers to a three-dimensional digital model of a device or system constructed based on power plant design information through computer-aided design (CAD) or other modeling software, which represents the geometric shape of the equipment, its installation location, and its relationship with other equipment.

[0062] Use the point cloud data of the power generation system to calibrate the model posture. By comparing the spatial position of the point cloud data with the design model, adjust the position and rotation angle of the design model to make it completely aligned with the actual physical position. After the model posture calibration, an accurate spatial model, that is, the power generation system spatial model, is obtained, which accurately reflects the actual state of the power generation system. Multidimensional data refers to data containing multiple types and sources, such as point cloud data, real-scene data, and design information. The multidimensional data of the power generation system consists of the point cloud data of the power generation system, the real-scene data of the power generation system, and the design information of the power plant.

[0063] The SIFT algorithm is used to perform feature matching between the spatial model of the power generation system and the real-scene data, thereby obtaining the real-scene spatial alignment coordinates, ensuring the precise spatial alignment of the virtual model with the real-scene data actually shot. SIFT (Scale Invariant Feature Transform) is a computer vision algorithm that can extract key feature points from images. These feature points are scale-invariant and rotation-invariant and can be used to match images from different perspectives. Through the SIFT algorithm, key feature points (such as the four corner points of the boiler) can be extracted from the real-scene data of the power generation system, and these features can be matched with the corresponding features in the point cloud data to achieve precise docking between the virtual model and the real-world data. Real-scene spatial alignment coordinates refer to aligning the real-world real-scene data with the spatial data of the virtual three-dimensional model through a feature matching algorithm, thereby ensuring precise matching of the virtual model and the real scene in spatial position.

[0064] The real-world data of the power generation system and the spatial model of the power generation system are integrated, and the real-world data is aligned with the virtual model in space and time, so that the virtual model can reflect the precise state of the real world. Virtual-real integration refers to combining the virtual three-dimensional model with the real data of the real world (such as point cloud data or real-world images obtained through laser scanning and photogrammetry) to form a highly unified and coordinated digital twin. The goal of virtual-real integration is to fully align the virtual model and the real data in terms of space, scale, etc. for further analysis or display. The fused data is rendered using a real-time rendering engine, which is a computer graphics engine that can generate and display three-dimensional images in real time and is usually used in dynamic interactive applications. In the construction of digital twins, the real-time rendering engine can integrate the virtual three-dimensional model and real-time data (such as sensor data, monitoring data, etc.) and present them to the user, which is usually used for simulation, visualization and real-time monitoring.

[0065] Through virtual-real fusion and real-time rendering engine, a complete digital twin of the power generation system is obtained, which not only includes the 3D design model of the power generation system, but also integrates data from the real world (such as real-time sensor data, image data, etc.), which can reflect the operating status of the power generation system in real time and has high interactivity and visualization functions. Repeat this process to obtain multiple system-level multidimensional data of multiple power generation systems in the target power plant, perform multi-level data fusion modeling, and build corresponding multiple system-level digital twins.

[0066] Through the combination of laser scanning, photogrammetry and design information, the three-dimensional data of the power generation system can be accurately obtained, and the virtual model and the actual equipment can be aligned in space through posture calibration and feature matching, thereby achieving high-precision virtual-real fusion. The real-time rendering engine is used to dynamically render the fused digital twin, allowing users to observe the operating status of the equipment, environmental changes, etc. in real time, improving the real-time response capability of operation and maintenance.

[0067] In summary, the multidimensional data fusion analysis method for digital twins provided in this application has the following technical effects: By adopting the LSTM network, M fault prediction models of M power plant equipment in the target power plant are constructed, wherein the M power plant equipment has M equipment source identifiers; after multi-dimensional data collection is performed on multiple power plant systems to obtain multiple system-level multi-dimensional data, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate multiple system-level digital twins; according to the M equipment source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant; a multi-level fault transfer analysis is performed on the target power plant to obtain a power plant fault state transfer diagram; according to the power plant fault state transfer diagram, fault conduction rendering is performed on the initial twin of the power plant to obtain a power plant digital twin; after loading the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M equipment operation status, the M equipment operation status is visualized in the power plant digital twin. That is to say, by constructing LSTM fault prediction models for multiple devices respectively, multi-level data fusion is used to model and fuse multiple system-level multidimensional data, multiple system-level digital twins are generated, and the established fault prediction models are integrated into the digital twins to achieve real-time monitoring and optimization of the entire power plant system. Fault conduction rendering is performed through multi-level fault transfer analysis, and the real-time operation data of the equipment is loaded into the fault prediction model for status visualization. The latest status of each device is displayed in the digital twin, and the equipment operation status is visualized, thereby improving the operation and maintenance efficiency of the power plant.

[0068] Embodiment 2: Based on the same inventive concept as the multidimensional data fusion analysis method for digital twins in the aforementioned embodiment 1, the present application also provides a multidimensional data fusion analysis device for digital twins, please refer to the attached Figure 2 , the multidimensional data fusion analysis device for digital twins comprises: A fault prediction model construction module 11, wherein the fault prediction model construction module 11 is used to use an LSTM network to construct M fault prediction models for M power plant equipment in a target power plant, wherein the M power plant equipment has M equipment source identifiers; a digital twin modeling module 12, wherein the digital twin modeling module 12 is used to perform multi-dimensional data collection on multiple power plant systems to obtain multiple system-level multi-dimensional data, and then use the multiple system-level multi-dimensional data to perform multi-level data fusion modeling to generate multiple system-level digital twins; a power plant twin construction module 13, wherein the power plant twin construction module 13 is used to load the M fault prediction models into the multiple system-level digital twins according to the M equipment source identifiers, Complete the construction of the initial twin of the power plant; a fault transfer analysis module 14, the fault transfer analysis module 14 is used to perform multi-level fault transfer analysis on the target power plant to obtain a power plant fault state transfer diagram; a fault conduction rendering module 15, the fault conduction rendering module 15 is used to perform fault conduction rendering on the initial twin of the power plant according to the power plant fault state transfer diagram to obtain a power plant digital twin; an operation status visualization module 16, the operation status visualization module 16 is used to load the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin, obtain the M equipment operation status, and then visualize the M equipment operation status in the power plant digital twin.

[0069] Furthermore, the fault prediction model construction module 11 in the multidimensional data fusion analysis device for digital twins is also used for: Equipment data is collected from the target power plant to obtain M historical fault operation data of M power plant equipment; after segmenting the M historical fault operation data based on the fault occurrence node, a sliding slice window is used to normalize the segmentation results to obtain M groups of sample fault operation data, wherein the sliding slice window has K sample segmentation steps; an LSTM network is used to construct M fault evolution prediction units of the M power plant equipment; a CNN network is used to construct M fault type identification units of the M power plant equipment; according to the K sample segmentation steps, the M groups of sample fault operation data are decomposed, and the decomposition results are used to optimize the parameters of the M fault evolution prediction units and the M fault type identification units; the M fault evolution prediction units and the M fault type identification units after mapping cascade optimization are constructed to complete the construction of the M fault prediction models.

[0070] Furthermore, the fault prediction model construction module 11 in the multidimensional data fusion analysis device for digital twins is also used for: The first group of sample fault operation data is decomposed into a first group of sample fault evolution data and a first group of sample fault steady-state data, wherein the time span of the sample fault evolution data is K-1 sample segmentation steps; the first group of sample fault evolution data and the first group of sample fault steady-state data are used as training data to construct a first fault evolution prediction unit for the first power plant equipment; after extracting the first group of sample fault types of the first group of sample fault steady-state data, the first group of sample fault steady-state data and the first group of sample fault types are used as training data to construct a first fault type identification unit for the first power plant equipment; the first fault evolution prediction unit and the first fault type identification unit are cascaded to complete the construction of the first fault prediction model; and so on, the M historical fault operation data are used as training data, and the M fault prediction models are constructed using an LSTM network.

[0071] Furthermore, the digital twin modeling module 12 in the multidimensional data fusion analysis device for digital twins is also used for: The point cloud data of the power generation system is obtained by laser scanning the power generation system; the real scene data of the power generation system is obtained by photogrammetry of the power generation system; after interactively obtaining the power plant design information of the target power plant, a three-dimensional design model of the power generation system is constructed according to the power plant design information, and the point cloud data of the power generation system is used to calibrate the model posture to obtain the power generation system space model, wherein the point cloud data of the power generation system, the real scene data of the power generation system and the power plant design information constitute the multi-dimensional data of the power generation system; SIFT is used to perform feature matching on the spatial model of the power generation system and the real scene data of the power generation system to obtain the real scene space alignment coordinates; according to the real scene space alignment coordinates, the real scene data of the power generation system and the spatial model of the power generation system are virtual-real fused, and a real-time rendering engine is introduced to complete the construction of the digital twin of the power generation system; and by analogy, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate the multiple system-level digital twins.

[0072] Furthermore, the power plant twin construction module 13 in the multidimensional data fusion analysis device for digital twins is also used for: A fault prediction complexity analysis is performed based on the M historical fault operation data, and multiple system-level prediction complexities are calculated and output based on the analysis results; with reference to the multiple system-level prediction complexities, computing nodes are allocated to the multiple power plant systems in a distributed computing architecture to obtain K groups of power systems with K computing nodes; based on the M equipment source identifiers, after the M fault prediction models are loaded into the multiple system-level digital twins, the K groups of power systems are used as replication guides, and the multiple system-level digital twins are replicated to the K computing nodes to complete the construction of the initial twin of the power plant.

[0073] Furthermore, the fault transfer analysis module 14 in the multidimensional data fusion analysis device for digital twins is also used for: Perform fault cross-system conduction analysis on the target power plant to obtain a system-level state transition diagram; perform fault cross-equipment conduction analysis on the target power plant based on the M historical fault operation data to obtain multiple equipment-level state transition diagrams for multiple power plant systems; and obtain a power plant fault state transition diagram by performing multi-dimensional data fusion on the system-level state transition diagram and multiple equipment-level state transition diagrams.

[0074] Furthermore, the operation status visualization module 16 in the multidimensional data fusion analysis device for digital twins is also used for: Perform pixel data collection on the M power plant equipment to obtain the M real-time equipment operation data; in the power plant digital twin, use the M fault prediction models to perform fault feature analysis on the M real-time equipment operation data, and output the M equipment operation statuses; perform fault feature aggregation on the M historical fault operation data to obtain M groups of fault operation performances; use the M equipment operation statuses to traverse the M groups of fault operation performances and schedule M real-time operation performances; use the real-time rendering engine to visualize the M equipment operation statuses of the M real-time operation performances in the power plant digital twin; perform fault transfer analysis on the M real-time operation performances in the power plant fault state transfer diagram, and visualize the fault state transfer trend of the analysis results in the power plant digital twin.

[0075] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The multidimensional data fusion analysis method and specific examples for digital twins in the first embodiment are also applicable to a multidimensional data fusion analysis device for digital twins in the present embodiment. Through the above detailed description of a multidimensional data fusion analysis method for digital twins, those skilled in the art can clearly know a multidimensional data fusion analysis device for digital twins in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0076] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0077] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A multidimensional data fusion analysis method for digital twins, characterized in that: include: Using an LSTM network to construct M fault prediction models for M power plant equipment in a target power plant, wherein the M power plant equipment has M equipment source identifiers; After performing multi-dimensional data collection on multiple power plant systems to obtain multiple system-level multi-dimensional data, multi-level data fusion modeling is performed using the multiple system-level multi-dimensional data to generate multiple system-level digital twins; According to the M device source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant; Performing a multi-level fault transfer analysis on the target power plant to obtain a fault state transfer diagram of the power plant; Performing fault conduction rendering on the initial twin of the power plant according to the fault state transition diagram of the power plant to obtain a digital twin of the power plant; After loading the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M equipment operation statuses, the M equipment operation statuses are visualized in the power plant digital twin.

2. A multidimensional data fusion analysis method for digital twins according to claim 1, characterized in that: After multi-dimensional data collection is performed on multiple power plant systems to obtain multiple system-level multi-dimensional data, multi-level data fusion modeling is performed using the multiple system-level multi-dimensional data to generate multiple system-level digital twins, including: By laser scanning the power generation system, point cloud data of the power generation system is obtained; Obtaining real scene data of the power generation system by performing photogrammetry on the power generation system; After interactively obtaining the power plant design information of the target power plant, after constructing the three-dimensional design model of the power generation system according to the power plant design information, the point cloud data of the power generation system is used to calibrate the model posture to obtain the power generation system space model, wherein the point cloud data of the power generation system, the real scene data of the power generation system and the power plant design information constitute the multi-dimensional data of the power generation system; Using SIFT to perform feature matching on the power generation system spatial model and the real scene data of the power generation system to obtain real scene space alignment coordinates; According to the real-scene space alignment coordinates, after the real-scene data of the power generation system and the power generation system space model are virtual-real fused, a real-time rendering engine is introduced to complete the construction of the digital twin of the power generation system; By analogy, the multiple system-level multi-dimensional data are used to perform multi-level data fusion modeling to generate the multiple system-level digital twins.

3. A multidimensional data fusion analysis method for digital twins as claimed in claim 2, characterized in that: The LSTM network is used to build M fault prediction models for M power plant equipment in the target power plant, including: Collect equipment data of the target power plant to obtain M historical fault operation data of M power plant equipment; After segmenting the M historical fault operation data based on the fault occurrence node, a sliding slice window is used to normalize the segmentation result to obtain M groups of sample fault operation data, wherein the sliding slice window has K sample segmentation steps; Using LSTM network to construct M fault evolution prediction units of the M power plant equipment; Using a CNN network to construct M fault type identification units for the M power plant equipment; After decomposing the M groups of sample fault operation data according to the K sample segmentation steps, the decomposition results are used to optimize the parameters of the M fault evolution prediction units and the M fault type identification units; The M fault evolution prediction units and the M fault type identification units after cascade optimization are mapped to complete the construction of the M fault prediction models.

4. A multidimensional data fusion analysis method for digital twins as claimed in claim 3, characterized in that: Performing a multi-level fault transfer analysis on the target power plant to obtain a fault state transfer diagram of the power plant, including: Conduct fault cross-system transmission analysis on the target power plant to obtain a system-level state transition diagram; Performing fault cross-device conduction analysis on the target power plant according to the M historical fault operation data to obtain multiple device-level state transition diagrams of multiple power plant systems; By performing multi-dimensional data fusion on the system-level state transition diagram and multiple device-level state transition diagrams, a power plant fault state transition diagram is obtained.

5. A multidimensional data fusion analysis method for digital twins according to claim 4, characterized in that: After loading the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin and obtaining the M equipment operation states, visualizing the M equipment operation states in the power plant digital twin includes: Collect pixel data for the M power plant equipment to obtain the M real-time equipment operation data; In the power plant digital twin, the M fault prediction models are used to perform fault feature analysis on the M real-time equipment operation data, and the operation status of the M equipment is output; Aggregating fault features of the M historical fault operation data to obtain M groups of fault operation performances; Using the M equipment operating states to traverse the M groups of fault operating performances, scheduling M real-time operating performances; Using the real-time rendering engine to visualize the operating status of the M devices in the digital twin of the power plant by using the M real-time operating tables; A fault transfer analysis is performed on the M real-time operating performances in the power plant fault state transfer diagram, and the analysis results are visualized in the power plant digital twin for fault state transfer trends.

6. The multidimensional data fusion analysis method for digital twins according to claim 3, characterized in that: include: Decomposing the first group of sample fault operation data into the first group of sample fault evolution data and the first group of sample fault steady-state data, wherein the time span of the sample fault evolution data is K-1 sample segmentation steps; Using the first set of sample fault evolution data and the first set of sample fault steady-state data as training data to construct a first fault evolution prediction unit for a first power plant device; After extracting the first group of sample fault types of the first group of sample fault steady-state data, using the first group of sample fault steady-state data and the first group of sample fault types as training data, constructing a first fault type identification unit for the first power plant equipment; Cascading the first fault evolution prediction unit and the first fault type identification unit to complete the construction of the first fault prediction model; By analogy, the M historical fault operation data are used as training data, and the M fault prediction models are constructed using the LSTM network.

7. The multidimensional data fusion analysis method for digital twins according to claim 3, characterized in that: According to the M device source identifiers, the M fault prediction models are loaded into the multiple system-level digital twins to complete the construction of the initial twin of the power plant, including: Performing a fault prediction complexity analysis based on the M historical fault operation data, and calculating and outputting multiple system-level prediction complexities based on the analysis results; Referring to the plurality of system-level prediction complexities, computing nodes are allocated to the plurality of power plant systems in a distributed computing architecture to obtain K groups of power systems with K computing nodes; According to the M device source identifiers, after the M fault prediction models are loaded into the multiple system-level digital twins, the multiple system-level digital twins are copied to the K computing nodes with the K groups of power systems as replication guides to complete the construction of the initial twin of the power plant.

8. A multidimensional data fusion analysis device for digital twins, characterized in that: The steps for implementing the multidimensional data fusion analysis method for digital twins as described in any one of claims 1 to 7, wherein the multidimensional data fusion analysis device for digital twins comprises: A fault prediction model construction module, wherein the fault prediction model construction module is used to construct M fault prediction models of M power plant equipment in the target power plant by using an LSTM network, wherein the M power plant equipment has M equipment source identifiers; A digital twin modeling module, wherein the digital twin modeling module is used to collect multi-dimensional data of multiple power plant systems to obtain multiple system-level multi-dimensional data, and then use the multiple system-level multi-dimensional data to perform multi-level data fusion modeling to generate multiple system-level digital twins; A power plant twin construction module, wherein the power plant twin construction module is used to load the M fault prediction models into the multiple system-level digital twins according to the M device source identifiers, so as to complete the construction of the initial twin of the power plant; A fault transfer analysis module, the fault transfer analysis module is used to perform a multi-level fault transfer analysis on the target power plant to obtain a fault state transfer diagram of the power plant; A fault conduction rendering module, the fault conduction rendering module is used to perform fault conduction rendering on the initial twin of the power plant according to the fault state transition diagram of the power plant to obtain a digital twin of the power plant; An operation status visualization module, wherein the operation status visualization module is used to load the M real-time equipment operation data of the M power plant equipment into the M fault prediction models of the power plant digital twin, obtain the M equipment operation status, and then visualize the M equipment operation status in the power plant digital twin.

Images