Power station equipment characteristic operation state trend analysis method, system and equipment and storage medium

Abstract

The invention provides a power station equipment characteristic operation state trend analysis method and system, equipment and a storage medium, and the method comprises the following steps: S1, obtaining online data of a first characteristic in power station equipment, and carrying out the preprocessing to obtain standard data; s2, carrying out disassembly and model remodeling on the power station equipment to obtain a digital twinborn analysis model; s3, performing adversarial training on different nodes in the analysis model to obtain a nonlinear relationship among the different nodes in the analysis model; and S4, endowing the nonlinear relationship to the analysis model to obtain a prediction model, and inputting the standard data into the prediction model to obtain an online prediction result. A dynamic network model is constructed and a nonlinear relation function between nodes is generated, so that the interaction description precision of the characteristics of each component of the equipment is remarkably improved; meanwhile, through real-time data updating and a deviation correction mechanism, the robustness and the real-time performance of the model are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of trend analysis, and particularly relates to a method, a system, a device and a storage medium for trend analysis of the characteristic operating state of power station equipment. Background Art

[0002] With the rapid development of modern industry and energy systems, power station equipment, as the core component of power generation and transmission, the reliability and stability of its operating state are directly related to the safety and efficiency of the entire power system. In recent years, the progress of digital technology, sensor networks and Internet of Things technology has made it possible to monitor the real-time operating state of equipment and perform big data analysis. Especially digital twin technology, by constructing a dynamic mapping between virtual equipment and physical equipment, provides a new solution for the prediction and analysis of equipment operating state. In addition, the application of machine learning and deep learning technologies in industrial fault detection and state prediction is also becoming increasingly widespread. These technologies can extract complex features from a large amount of historical data and real-time data, so as to realize the intelligent analysis of equipment operating state.

[0003] At present, many technologies have been applied to the monitoring and trend analysis of the operating state of power station equipment. For example, the trend prediction method based on statistical models can perform simple prediction on a single variable, the method based on physical modeling can analyze its operating behavior through the internal principle of the equipment, and the technology based on neural network can learn the complex relationship between multiple variables. However, these methods still have many limitations in terms of the real-time performance of equipment state, multi-node correlation and prediction accuracy. Especially, most traditional state monitoring technologies rely on static models or simple linear analysis, and it is difficult to capture the non-linear change relationship caused by dynamic environment or complex interaction between components during the operation of power station equipment. Summary of the Invention

[0004] The first object of the present invention is to provide a method for trend analysis of the characteristic operating state of power station equipment in view of the above-mentioned problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A method for trend analysis of the characteristic operating state of power station equipment includes the following steps: S1. Obtain the online data of the first feature in the power station equipment, and obtain the standard data through preprocessing; S2. Disassemble and reshape the model of the power station equipment to obtain an analysis model of digital twin; S3. Through adversarial training of different nodes in the analysis model, obtain the non-linear relationship between different nodes in the analysis model; S4. Endow the analysis model with the non-linear relationship to obtain a prediction model, and input the standard data into the prediction model to obtain the result of online prediction.

[0006] While adopting the above technical solution, the present invention can also adopt or combine the following technical solutions: As a preferred technical solution of the present invention: in step S1, the first feature includes temperature information and vibration information.

[0007] As a preferred technical solution of the present invention: step S1 further includes the following sub-steps: S11. Obtain the standard deviation, mean, and range in the historical data, combine them with the on-line data collected by the sensor, and obtain the standardized high-dimensional data. The formula is as follows: In the formula, represents the standard deviation of the first feature, represents the mean of the first feature, represents the range of the first feature, represents the on-line data of the first feature, represents the current moment; S12. At the start of equipment operation, continuously record the on-line data, and at the same time update the real-time standard deviation, mean, and range. Intercept the on-line data within a time window to obtain the representation of the standard data with respect to the time series. The formula is as follows: In the formula, represents the sequence representation from the moment t-a to the moment t, a represents the length of the time window, represents the standard data at the moment t; S13. In the time window, correct the real-time first feature data through a preset verification algorithm to obtain the standard data in each sensor.

[0008] As a preferred technical solution of the present invention: step S2 further includes the following sub-steps: S21. Disassemble the equipment by components, take each component as a node, and represent the existence of the transfer relationship of the first feature between nodes through the connection between nodes to obtain a dynamic graph network; S22. Through continuously working on the situation of each node, perform the transfer of the first feature in the dynamic graph network, so as to realize the trend analysis of the operation state of the equipment features; S23. During the analysis process, by obtaining the on-line data, update the real-time first feature of some nodes, and at the same time, according to the updated first feature, update the first feature of all nodes. Some nodes include the nodes configured with sensors, and the on-line data of the first feature is obtained through the sensors.

[0009] As a preferred technical solution of the present invention: in step S3, the adversarial training includes a generator G and a discriminator D, and includes the following sub-steps: S31. Use the output of the generator as the non-linear relationship function of the first feature between two nodes; The generator includes an input: representing the first feature of node and representing the first feature of node ; the output: representing the generated non-linear relationship function; The discriminator includes an input: the true relationship function obtained by fitting historical data , and the relationship function generated by the generator ; the output: the probability that the input relationship function is the true relationship; S32. Respectively aim at minimizing the loss function to update the parameters of the generator G and the discriminator D; Fix the generator G and update the parameters of the discriminator D to make the discrimination result more accurate; Fix the discriminator D and update the parameters of the generator G to make the generated function closer to the true relationship; S33. Repeat the training until the function generated by the generator cannot be distinguished by the discriminator, reaching an adversarial balance; S34. Use the trained generator to generate a non-linear relationship function to describe the interaction relationship of the first features between real nodes.

[0010] As a preferred technical solution of the present invention: S41. Assign the non-linear relationship function to the analysis model, where the nodes represent each component, and the connection lines represent the non-linear relationship function of the first features between the components; S42. Use the standard data collected by the sensor as the initial data of the first feature, and use the prediction model to obtain the initial value of the first feature of each node; S43. In the continuous working stage, obtain the nodes H={h1, h2,..., hn} that generate the first feature in the device, and through the physical and mathematical model, obtain the first feature generated by each node in the set H at the next moment; where H represents the set of nodes that generate the first feature, hn represents the nth node that generates the first feature, and n represents the number of nodes that generate the first feature; S44. Use the prediction model to transfer the first feature generated by each node in the set H between the nodes according to the non-linear relationship function, so as to obtain the first feature of each node at the next moment t1; S45. Integrate the first features of each node to obtain the first feature set of each component of the device, and perform comprehensive analysis through a preset neural network algorithm to obtain the comprehensive evaluation result of the state trend at the next moment t1; S46. Update the first features of each component of the device at the next moment t1 according to the first feature set of each component of the device; based on the next moment t1, obtain the first features generated by each node in the set H at the next next moment t2, and use the prediction model to complete the prediction of the first features of each node at the t2 moment; perform comprehensive analysis through a preset neural network algorithm to obtain the comprehensive evaluation result of the state trend at the t2 moment; S47. Recursively update the three steps of the first features of each component of the device, the prediction of the first features of each node, and the comprehensive evaluation of the state trend, so as to realize the analysis of the operation state trend of the device features within the set time interval; S48. During the process of analyzing the operation state trend of the device features, by obtaining online data in real time, in the prediction model, update the real-time first features of some nodes, and update the first features of other nodes by using the connections in the prediction model; S49. Use the updated data as the comparison data. While completing the update of the first features of all nodes, analyze the prediction results: if the deviation between the prediction result at the current moment and the comparison data exceeds the preset threshold, then starting from the current moment, re-analyze the operation state trend of the device features within the time interval; If the deviation between the prediction result at the current moment and the comparison data does not exceed the preset threshold, then only update the current comparison data and retain the analysis result of the operation state trend of the device features within the time interval.

[0011] The second object of the present invention is to provide a system for analyzing the operation state trend of power station equipment features, including the following modules: A data acquisition module, configured to obtain online data of the first features in the power station equipment and obtain standard data through preprocessing; An analysis model construction module, configured to disassemble and reshape the model of the power station equipment to obtain an analysis model of digital twin; An analysis module, configured to obtain the non-linear relationship between different nodes in the analysis model through adversarial training of different nodes in the analysis model; A prediction module, configured to endow the analysis model with the non-linear relationship to obtain a prediction model, and input the standard data into the prediction model to obtain the result of online prediction.

[0012] The third object of the present invention is to provide an electronic device, where the electronic device includes a processor, a communication interface, a memory, and a communication bus, and the processor, the communication interface, and the memory complete communication with each other through the communication bus, A memory for storing a computer program; A processor for executing the computer program stored on the memory to implement the steps of the method for analyzing the characteristic operating state trend of power station equipment as described above.

[0013] Another object of the present invention is to provide a non-volatile storage medium storing an executable program, which when executed by a processor, implements the steps of the method for analyzing the characteristic operating state trend of power station equipment as described above.

[0014] Compared with the prior art, the present invention has the following beneficial effects: By combining digital twin technology with generative adversarial network (GAN), high-precision analysis and real-time prediction of the operating state trend of power station equipment are achieved; a dynamic network model is constructed and a non-linear relationship function between nodes is generated, significantly improving the accuracy of the interactive description of the characteristics of each component of the equipment; at the same time, through real-time data update and deviation correction mechanisms, the robustness and real-time performance of the model are enhanced; the overall state trend of the equipment is comprehensively evaluated, providing more reliable technical support for equipment operation and maintenance, and effectively solving the problems of insufficient real-time performance and accuracy in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the method for analyzing the characteristic operating state trend of power station equipment provided by the present invention.

[0016] Figure 2 It is a framework diagram of the system for analyzing the characteristic operating state trend of power station equipment provided by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] As Figure 1 shown, a method for analyzing the characteristic operating state trend of power station equipment specifically includes the following steps: S1. Obtain the online data of the first feature in the power station equipment, and obtain the standard data through preprocessing; the first feature includes but is not limited to temperature information and vibration information.

[0019] S11. Obtain the standard deviation, mean value, and range in the historical data, and combine them with the online data collected by the sensor to obtain the standardized high-dimensional data. The formula is as follows: In the formula, represents the standard deviation of the first feature, represents the mean value of the first feature, represents the range of the first feature, represents the online data of the first feature, Indicates the current moment; S12. When the equipment starts running, the online data is continuously recorded, and the real-time standard deviation, mean, and range are updated. The online data within a time window is intercepted to obtain the representation of the standard data on the time series. The formula is as follows: In the formula, represents the sequence representation from time ta to time t, a represents the length of the time window, Represents the standard data at time t; S13. In the time window, the real-time first characteristic data is corrected by a preset verification algorithm to obtain standard data in each sensor.

[0020] You can use a time series model (such as the sliding average method or the ARIMA model) to model the sensor's historical data, thereby correcting the real-time first feature data. Or you can use a pre-trained machine learning model (such as SVM or neural network) to classify the data, determine whether it is an outlier, and correct it.

[0021] By calculating statistics such as standard deviation, mean and range, combined with online data collection, noise, outliers and data drift are removed to ensure the stability and reliability of input data. High-quality standard data can significantly reduce the interference of data fluctuations on subsequent analysis results and improve the robustness and prediction accuracy of the model.

[0022] Record and update indicators such as standard deviation, mean and range through time windows to reflect the dynamic change trend of the first feature of the equipment in different time periods. Dynamic feature capture helps to describe the real-time changes in the operating status of power station equipment and supports the model to identify abnormal equipment status or trend changes. Intercept serialized standard data within the time window, construct a time series representation, and enhance the model's learning ability for feature time series changes. Correct real-time data through a preset verification algorithm to reduce the impact of sensor measurement errors and environmental interference on the data, and ensure that each sensor provides accurate standard data. The data correction mechanism improves the credibility of input data and avoids model prediction deviations caused by input errors.

[0023] S2. Disassemble and remodel the power station equipment to obtain the analysis model of the digital twin; S21, disassembling the device according to its components, taking each component as a node, and using lines between the nodes to indicate that the first feature has a transfer relationship between the nodes, thereby obtaining a dynamic graph network; S22, by continuously working on the situation of each node, the first feature is transmitted in the dynamic graph network, thereby realizing trend analysis of the characteristic operation status of the equipment; S23. During the analysis process, by obtaining online data, update the first features of some nodes in real time. At the same time, based on the updated first features, update the first features of all nodes. The some nodes include the nodes configured with sensors, and obtain the online data of the first features through the sensors.

[0024] By disassembling the device into components and constructing a dynamic graph network of nodes and connections, the aim is to capture the complex interaction relationships between the device components, especially the dynamic transfer laws of the first features (such as temperature, vibration, etc.) between different components. This method can truly reflect the non-linear feature changes and mutual influences during the device operation, and provide an accurate dynamic modeling basis for the trend analysis of the device feature operation state. It emphasizes updating the features of some nodes through online data, and using the updated feature values to drive the update of other nodes, so as to realize the state adjustment of the global network. This dynamic update mechanism ensures that the model can timely reflect the real-time state of the device, and make accurate predictions about future trends based on the current state, improving the real-time performance and prediction ability of the model.

[0025] In practical applications, some device components may not have the conditions for directly installing sensors (for example, due to space limitations or cost considerations), while some key components can directly obtain the real-time data of the first features through sensors. By designing the mechanism for updating some nodes, the model can deduce the state of the entire network using limited sensor data, greatly improving the analysis efficiency and reducing the dependence on comprehensive sensor deployment.

[0026] S3. Through adversarial training of different nodes in the analysis model, obtain the non-linear relationships between different nodes in the analysis model; the adversarial training includes a generator G and a discriminator D, and includes the following sub-steps: S31. Use the output of the generator as the non-linear relationship function of the first features between two nodes; The generator includes, input: representing the first feature of node and representing the first feature of node ; output: representing the generated non-linear relationship function; The discriminator includes, input: the real relationship function obtained by fitting historical data , and the relationship function generated by the generator ; output: the probability that the input relationship function is the real relationship; S32. Respectively aiming at minimizing the loss function, update the parameters of the generator G and the discriminator D; Fix the generator G and update the parameters of the discriminator D to make the discrimination result more accurate; Fix the discriminator D and update the parameters of the generator G to make the generated function closer to the true relationship; S33. Repeat the training until the function generated by the generator cannot be distinguished by the discriminator, reaching an adversarial balance; S34. Use the trained generator to generate a non-linear relationship function to describe the interaction relationship of the first feature between real nodes.

[0027] This relationship function can effectively capture the complex dynamics of feature transfer between nodes, including non-linearity, time-variation, and coupling effects between nodes. This modeling method avoids the limitations of traditional linear relationship descriptions and provides a solid foundation for the accurate analysis of equipment state trends.

[0028] The training process is based on the overall characteristics of the entire device, rather than simply treating two nodes in isolation. When generating the non-linear relationship function between nodes, not only the transfer law of the first feature (such as temperature or vibration) between two nodes is analyzed, but also the internal characteristics of the nodes are comprehensively considered, such as: energy changes indirectly caused by other factors: mechanical vibrations inside the node may generate heat, affecting the node temperature; heat generation due to friction: friction between components will have a direct impact on the temperature characteristics. Through adversarial training, all these objective influencing factors can be considered in the non-linear relationship function.

[0029] It ensures that the non-linear relationship function generated by the generator can reflect both the interaction relationship between nodes and the internal characteristic changes of the nodes themselves, improving the physical rationality and prediction accuracy of the model.

[0030] Through the adversarial training between the generator and the discriminator, the generator is continuously optimized, and the generated non-linear relationship function gradually approaches the feature relationship between real nodes; at the same time, the discriminator improves the ability of the generator by distinguishing between the true relationship function and the generated function. When the adversarial balance is finally reached, the non-linear relationship function output by the generator has strong expressive ability and can accurately describe the complex feature interaction between nodes.

[0031] S4. Assign the non-linear relationship to the analysis model to obtain a prediction model, and input the standard data into the prediction model to obtain the online prediction result.

[0032] S41. Assign the non-linear relationship function to the analysis model, where the nodes represent each component, and the connection lines represent the non-linear relationship functions of the first feature between components; S42. Use the standard data collected by the sensor as the initial data of the first feature, and use the prediction model to obtain the initial values of the first feature of each node (only some nodes have sensors); S43. During the continuous operation stage, obtain the nodes H = {h1, h2,..., hn} in the device that generate the first feature. Through the physical and mathematical model, obtain the first feature generated by each node in the set H at the next moment. Among them, H represents the set of nodes that generate the first feature, hn represents the nth node that generates the first feature, and n represents the number of nodes that generate the first feature. The nodes that generate the first feature refer to those device components that have an active contribution or driving effect on the first feature (such as temperature, vibration). Their role is the direct source of the first feature, rather than just a feature transfer point. The generation characteristics of these nodes are usually directly related to the working principle of the device. For the feature of temperature, this node is the driving device, not the friction of the gear; however, in this case, the friction of the gear will also generate heat, and this problem is overcome through the adversarial training in the above text.

[0033] S44. Using the prediction model, transfer the first feature generated by each node in the set H among the nodes according to the non-linear relationship function, so as to obtain the first feature of each node at the next moment t1. S45. Integrate the first features of each node to obtain the first feature set of each component of the device. Through the preset neural network algorithm for comprehensive analysis, obtain the comprehensive evaluation result of the state trend at the next moment t1. S46. According to the first feature set of each component of the device, update the first feature of each component of the device at the next moment t1. On the basis of the next moment t1, obtain the first feature generated by each node in the set H at the next next moment t2. Use the prediction model to complete the prediction of the first feature of each node at the t2 moment. Through the preset neural network algorithm for comprehensive analysis, obtain the comprehensive evaluation result of the state trend at the t2 moment. S47. Loop and update the three steps of the first feature of each component of the device, the prediction of the first feature of each node, and the comprehensive evaluation of the state trend, so as to realize the analysis of the running state trend of the device features within the set time interval. S48. During the process of analyzing the running state trend of the device features, by obtaining online data in real time, in the prediction model, update the real-time first feature of some nodes, and use the connection in the prediction model to update the first feature of other nodes. S49. Use the updated data as the comparison data. While completing the update of the first feature of all nodes, analyze the prediction result: If the deviation between the prediction result at the current moment and the comparison data exceeds the preset threshold, then starting from the current moment, re-analyze the running state trend of the device features within the time interval. If the deviation between the prediction result at the current moment and the comparison data does not exceed the preset threshold, only the current comparison data is updated, and the analysis result of the operation state trend of the device features within the time interval is retained.

[0034] The feature transfer between power station equipment components (such as temperature and vibration) often has nonlinearity and dynamics. This transfer is affected by various factors, including mechanical coupling, heat conduction paths, and changes in operating conditions. By endowing the analysis model with a nonlinear relationship function, the prediction model can accurately describe these complex interaction relationships. The nonlinear relationship function enables the prediction model to have stronger expressive power, can capture the nonlinear change laws in feature transfer, avoids the simplified assumptions about feature transfer laws in traditional linear models, and thus improves the accuracy of model prediction.

[0035] By constructing a dynamic graph network with nodes and connections, the operation state of the device can be dynamically updated, the change trend of the first feature of each component can be predicted, and the operation state of the overall device can be reflected. The operation environment of power station equipment is complex, and the states of each component may be affected by external disturbances or changes in operating conditions, requiring the prediction model to have the ability to update and respond in real time. The nonlinear relationship function can dynamically adapt to the changes in real-time data and provide flexibility for the model.

[0036] Through the prediction model, using the nonlinear relationship function between nodes, the node features are dynamically transferred, and the first eigenvalue of each node at the next moment is deduced. The first feature sets of each node are integrated to obtain the first feature set of the overall device. Through a preset neural network algorithm (such as a deep learning model), these features are comprehensively analyzed to obtain a comprehensive evaluation result of the device state. This analysis result is used to predict the state trend of the device at the next moment (such as t1). By continuously updating the feature data of the device components, the feature prediction and comprehensive evaluation of each node are carried out to achieve real-time prediction of the device state. This process is a cyclic process. At each moment t1, t2,..., the state at the next moment is continuously deduced and compared with historical data to optimize the prediction result.

[0037] The core objective of this solution is to accurately predict the operation state trend of the device by combining the nonlinear relationship function between nodes and real-time data update, and be able to make dynamic adjustments according to the actual operation state of the device (such as temperature, vibration, etc.).

[0038] The Deep Gated Recurrent Unit (DGRU) is a time-series data prediction model in deep learning. It belongs to the variant of the Recurrent Neural Network (RNN) and can effectively process sequential data and capture long-term dependencies. In this design, by improving the through DGRU, it can handle the problem of trend prediction more effectively. Due to the complex non-linear interaction relationships between the components of the device, we introduce the Graph Neural Network (GNN) on the basis of DGRU to help process the non-linear transfer relationships between nodes. The Graph Neural Network can transmit information between nodes and model the non-linear relationships between nodes based on the graph structure. This combination can effectively capture the mutual influence between nodes, especially the dynamic cooperation between the components (nodes) of the device.

[0039] The standard DGRU adopts a fixed update and reset gating mechanism. The parameters of each gate (such as the weight matrix of the update gate) are static, obtained through a certain training process, and remain unchanged throughout the training process. However, such a fixed mechanism may not be able to capture the dynamic changes of the device components under different working conditions. Therefore, in this design, the gating mechanism is replaced by non-linear relationships, eliminating the static parameter limitations in the traditional model. Through the generator (G) in the Generative Adversarial Network (GAN), this design introduces non-linear relationship functions for the interaction relationships between each node (device component) of the device. These relationship functions can adaptively adjust the update process of each node state according to different device working states and the dynamic characteristics of the components.

[0040] The output of the non-linear relationship function is not a simple weighted average or gating mechanism, but the complex relationships between nodes learned by the generator, which dynamically adjusts the state of each node. In other words, the non-linear relationship function replaces the role of the traditional update gate and reset gate, directly affecting the state update of each node. When the generator learns appropriate non-linear relationships, it will use these relationship functions to update the state of the device nodes at each time step. By applying the non-linear relationship function to the state transfer between device components, the generator dynamically determines the state changes of each node at different time steps without fixed gating control. This means that the state change of a node not only depends on the state at the previous moment but also takes into account the interaction effects with other nodes.

[0041] The non - linear relationship function mainly addresses the complex interaction relationships between device nodes and no longer relies on fixed weighting parameters (such as the weight matrix in the gating mechanism). Specifically: Each component of the device may produce different responses under different operating states. Simple weighted averaging may not be able to capture such complex interactions, while the non - linear relationship function can adaptively adjust according to the changes in real - time data. During the process of updating the device component states, the non - linear relationship function can dynamically adjust its state update based on the actual working state of the current node. In this way, the changes in the device state are no longer static but can respond to changes in time and working conditions.

[0042] During the process of analyzing the trend of the operating state of device characteristics, by obtaining online data in real - time, in the prediction model, update the first characteristics of some nodes in real - time, and use the connections in the prediction model to update the first characteristics of other nodes.

[0043] Take the updated data as comparison data. While completing the update of the first characteristics of all nodes, analyze the prediction results. If the deviation between the prediction result at the current moment and the comparison data exceeds the preset threshold, then starting from the current moment, re - analyze the trend of the operating state of device characteristics within the time interval. If the deviation between the prediction result at the current moment and the comparison data does not exceed the preset threshold, then only update the current comparison data and retain the analysis results of the trend of the operating state of device characteristics within the time interval.

[0044] During the operation of the device, each node of the device will be affected by different working conditions, and its characteristics (such as temperature, vibration, pressure, etc.) will also change continuously. By obtaining online data in real - time and inputting it into the prediction model to update the first characteristics of some nodes, the current state changes of each component of the device can be reflected in real - time. Further, through the non - linear relationship (i.e., the transfer relationship between nodes) in the prediction model, the characteristics of the updated nodes can be used to calculate and update the state characteristics of other nodes. This real - time update mechanism ensures the instant feedback of the device state, enabling the prediction model to more accurately capture various dynamic changes during the operation of the device.

[0045] After updating the features of each node, the updated node data is used as comparison data, which is combined with the prediction results for comparison to determine the deviation between the current prediction result and the historical comparison data. If the deviation between the prediction result and the comparison data exceeds the preset threshold, it indicates that the state of the current device has changed abnormally or there is a large error in the prediction result, and it is necessary to re - conduct the state trend analysis to ensure the accuracy of the prediction result; if the deviation does not exceed the threshold, it means that the device state change conforms to the expectation of the prediction model, and the current analysis result can be continued to be maintained, and only the current comparison data is updated. The purpose of this design is to avoid frequent re - analysis by setting a reasonable threshold, and at the same time, it can timely identify the abnormalities in the device operation and dynamically adjust the analysis strategy according to the actual situation.

[0046] The state prediction of device features is a continuous and dynamic process. Especially for some devices that require quick response (such as power generation devices), real - time performance is particularly important. Through the comparison data deviation control and dynamic update mechanism in this design, the trend prediction of device features can be adjusted in real time to ensure that the analysis results are both accurate and timely. At the same time, the use of threshold control avoids the ineffective re - analysis process and ensures the analysis efficiency and system stability.

[0047] As Figure 2 shown, the present invention also provides a system for analyzing the running state trend of power station device features, including the following modules: A data acquisition module, which is used to obtain the online data of the first feature in the power station device and obtain standard data through pre - processing; An analysis model construction module, which is used to disassemble and reshape the model of the power station device to obtain an analysis model of digital twin; An analysis module, which is used to obtain the non - linear relationship between different nodes in the analysis model by performing adversarial training on different nodes in the analysis model; A prediction module, which is used to endow the analysis model with the non - linear relationship to obtain a prediction model, and input the standard data into the prediction model to obtain the result of online prediction.

[0048] The present invention also provides an electronic device, which includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory complete mutual communication through the communication bus. The memory is used to store a computer program; The processor is used to execute the computer program stored on the memory to implement the steps of the method for analyzing the running state trend of power station device features as described above.

[0049] The present invention also provides a non-volatile storage medium, in which an executable program is stored. When the executable program is executed by a processor, the method steps for analyzing the characteristic operation state trend of power station equipment as described above are implemented.

[0050] So far, the technical solution of the present invention has been described in combination with the specific experimental process shown in the drawings. However, the protection scope of the present invention is not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for analyzing the trend of the characteristic operating state of power station equipment, characterized in that, It includes the following steps: S1. Obtain the online data of the first feature in the power station equipment, and obtain the standard data through preprocessing; S2. Disassemble and reshape the model of the power station equipment to obtain an analysis model of the digital twin; S3. Through adversarial training of different nodes in the analysis model, obtain the non-linear relationship between different nodes in the analysis model; S4. Assign the non-linear relationship to the analysis model to obtain a prediction model, input the standard data into the prediction model, and obtain the result of online prediction.

2. The method according to claim 1, wherein: In step S1, the first feature includes temperature information and vibration information.

3. The method according to claim 1, wherein: Step S1 further includes the following sub-steps: S11. Obtain the standard deviation, mean, and range in the historical data, combine them with the online data collected by the sensor, and obtain the standardized high-dimensional data. The formula is as follows: In the formula, represents the standard deviation of the first feature, represents the mean value of the first feature, represents the range of the first feature, represents the online data of the first feature, represents the current moment; S12. At the beginning of the equipment operation, continuously record the online data, and at the same time update the real-time standard deviation, mean, and range. Intercept the online data within a time window to obtain the representation of the standard data with respect to the time series. The formula is as follows: In the formula, represents the sequence representation from time t - a to time t, and a represents the time window length, represents the standard data at time t; S13. In the time window, correct the real-time first feature data through a preset verification algorithm to obtain the standard data in each sensor.

4. The method according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S21. Disassemble the equipment by components, take each component as a node, and represent the transfer relationship of the first feature between nodes through the connection between nodes to obtain a dynamic graph network; S22. Through continuously working on the situation of each node, perform the transfer of the first feature in the dynamic graph network, so as to realize the trend analysis of the operation state of the equipment features; S23. During the analysis process, obtain the online data, update the first feature of some nodes in real time, and at the same time update the first feature of all nodes according to the updated first feature. Some nodes include the nodes configured with sensors, and obtain the online data of the first feature through the sensors.

5. The method according to claim 1, characterized in that: In step S3, the adversarial training includes a generator G and a discriminator D, and includes the following sub-steps: S31. Use the output of the generator as the non-linear relationship function of the first feature between two nodes; The generator includes an input: A first feature representing a node And A first feature representing a node And. Output: Indicates the generated non-linear relationship function; The discriminator includes an input: the true relationship function obtained by fitting historical data , and the relationship function generated by the generator ; Output: the probability that the input relationship function is the true relationship; S32. Respectively aiming at minimizing the loss function, update the parameters of the generator G and the discriminator D; Fix the generator G and update the parameters of the discriminator D to make the discrimination result more accurate; Fix the discriminator D and update the parameters of the generator G to make the generated function closer to the real relationship; S33. Repeat the training until the function generated by the generator cannot be distinguished by the discriminator, reaching an adversarial balance; S34. Use the trained generator to generate a non-linear relationship function to describe the interaction relationship of the first feature between real nodes.

6. The method according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S41. Assign the non-linear relationship function to the analysis model, the node represents each component, and the connection represents the non-linear relationship function of the first feature between components; S42. Use the standard data collected by the sensor as the initial data of the first feature, and use the prediction model to obtain the initial value of the first feature of each node. S43. During the continuous operation phase, obtain the nodes H = {h1, h2,..., hn} in the device that generate the first feature. Through a physical and mathematical model, obtain the first feature generated by each node in the set H at the next moment. Among them, H represents the set of nodes that generate the first feature, hn represents the nth node that generates the first feature, and n represents the number of nodes that generate the first feature. S44. Using the prediction model, transfer the first feature generated by each node in the set H among the nodes according to the non-linear relationship function, so as to obtain the first feature of each node at the next moment t1. S45. Integrate the first features of each node to obtain the first feature set of each component of the device, and perform comprehensive analysis through a preset neural network algorithm to obtain the comprehensive evaluation result of the state trend at the next moment t1. S46. According to the first feature set of each component of the device, update the first feature of each component of the device at the next moment t1. On the basis of the next moment t1, obtain the first feature generated by each node in the set H at the next next moment t2, and use the prediction model to complete the prediction of the first feature of each node at the t2 moment; perform comprehensive analysis through a preset neural network algorithm to obtain the comprehensive evaluation result of the state trend at the t2 moment. S47. Recursively update the three steps of the first feature of each component of the device, the prediction of the first feature of each node, and the comprehensive evaluation of the state trend, so as to realize the analysis of the running state trend of the device features within the set time interval. S48. During the process of analyzing the running state trend of the device features, by obtaining online data in real time, update the real-time first feature of some nodes in the prediction model, and update the first feature of other nodes using the connections in the prediction model. S49. Use the updated data as comparison data. While completing the update of the first feature of all nodes, analyze the prediction result: if the deviation between the prediction result at the current moment and the comparison data exceeds the preset threshold, start from the current moment and re-analyze the running state trend of the device features within the time interval. If the deviation between the prediction result at the current moment and the comparison data does not exceed the preset threshold, only update the current comparison data and retain the analysis result of the running state trend of the device features within the time interval.

7. A characteristic operating state trend analysis system for power station equipment, characterized in that, It includes the following modules: A data acquisition module, which is used to obtain the online data of the first feature in the power station equipment and obtain the standard data through preprocessing. An analysis model construction module, which is used to disassemble and reshape the model of the power station equipment to obtain the analysis model of the digital twin. An analysis module, which is used to perform adversarial training on different nodes in the analysis model to obtain the non-linear relationship between different nodes in the analysis model. A prediction module, which is used to endow the analysis model with the non-linear relationship to obtain the prediction model, input the standard data into the prediction model, and obtain the result of online prediction.

8. An electronic device, the electronic device includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The characteristics are as follows: A memory, the memory is used to store a computer program. A processor, which is configured to execute a computer program stored in a memory to implement the method steps of analyzing the trend of the characteristic operating state of power station equipment as described in any one of claims 1-6.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, which, when executed by the processor, implements the method steps of analyzing the trend of the characteristic operating state of power station equipment as described in any one of claims 1-6.

Images