An intelligent operation and maintenance operation management system based on artificial intelligence

Abstract

The application discloses an intelligent operation and maintenance management system based on artificial intelligence, which comprises a multi-source data collection module, a data optimization module, a power plant equipment maintenance prediction model establishing module and an intelligent operation and maintenance management module. The application relates to the technical field of data processing, and particularly relates to an intelligent operation and maintenance management system based on artificial intelligence. The application innovatively proposes that equipment is taken as a network node, equipment process dependence is taken as a directed edge, an equipment maintenance network graph structure is constructed, the risk propagation chain among different equipment can be effectively identified, and power plant operation and maintenance management is changed from local prediction to global correlation modeling; a direct and indirect double-layer space dependence joint modeling mechanism is adopted, direct space dependence models the explicit process relationship among adjacent equipment through a graph attention network, indirect space dependence models the implicit correlation of non-adjacent equipment through a space attention mechanism, the prediction precision is improved, and the accuracy of intelligent operation and maintenance operation decision is improved.

Description

An intelligent operation and maintenance management system based on artificial intelligence Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an intelligent operation and maintenance management system based on artificial intelligence. Background Technology

[0002] An intelligent operation and maintenance management system based on artificial intelligence is a system that uses artificial intelligence technology to intelligently manage the operation and maintenance of power plant equipment. Through real-time data collection and analysis, the system can predict possible equipment failures, performance degradation or anomalies, and then take intelligent operation and maintenance actions in advance. This can effectively avoid sudden equipment failures and downtime, ensure that the equipment is always in the best operating condition, reduce the operation and maintenance costs of the power plant, and realize the intelligent operation and maintenance management of the power plant.

[0003] However, traditional intelligent operation and maintenance management systems are limited to micro-data analysis of single devices and fail to consider the technical issues of parallel operation characteristics of multiple devices and cross-device process interaction relationships. This results in the system's inability to build a global interconnected network for the power plant and makes it difficult to capture the global chain reaction of faults caused by local device anomalies. Consequently, power plant operation and maintenance management decisions are lagging behind. Existing predictive models for power plant equipment maintenance generally only consider the direct adjacency relationship between devices and ignore the indirect process coupling of non-adjacent devices, resulting in insufficient prediction accuracy and weak responsiveness to the global process status, which affects the reliability and foresight of operation and maintenance decisions. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent operation and maintenance management system based on artificial intelligence. Traditional intelligent operation and maintenance management systems are limited to micro-level data analysis of single devices, failing to consider the parallel operation characteristics of multiple devices and cross-device process interactions. This results in the system's inability to construct a global interconnected network for the power plant, making it difficult to capture the global cascading failure propagation effect caused by local device anomalies, thus leading to delayed power plant operation and maintenance management decisions. This solution innovatively proposes constructing a device maintenance network graph structure with devices as network nodes and device process dependencies as directed edges. This effectively identifies risk propagation chains between different devices, improves the ability to identify cross-device associated risks, enhances the global coordination and dynamic responsiveness of operation and maintenance operations, and improves the comprehensiveness and real-time performance of predictions. It breaks through the limitations of traditional single-device prediction, transforming power plant operation and maintenance management from local prediction to global interconnected modeling, significantly improving plant-wide maintenance strategies. This solution enhances collaboration and decision-making efficiency, promoting the scientific allocation of operation and maintenance resources. Addressing the technical problem that existing power plant equipment maintenance prediction models generally only consider direct adjacency relationships between equipment and neglect indirect process coupling between non-adjacent equipment, resulting in insufficient prediction accuracy and weak responsiveness to global process states, thus affecting the reliability and foresight of operation and maintenance decisions, this solution innovatively adopts a joint modeling mechanism of direct and indirect spatial dependencies. Direct spatial dependencies model explicit process relationships between adjacent equipment through graph attention networks, while indirect spatial dependencies model implicit associations between non-adjacent equipment through spatial attention mechanisms. Finally, the two types of spatial features are fused and aggregated to achieve comprehensive capture of multi-level spatial dependencies. This enables the identification of potential risk propagation paths and implicit dependencies between equipment, significantly improving the model's global cognitive ability and prediction accuracy. It also enhances the system's interpretability and robustness in the collaborative operation of multiple equipment, thereby improving the accuracy and scientific nature of intelligent operation and maintenance decision-making.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent operation and maintenance management system based on artificial intelligence, including a multi-source data collection module, a data optimization module, a power plant equipment maintenance prediction model establishment module, and an intelligent operation and maintenance management module;

[0006] The multi-source data collection module specifically obtains raw data of power plant operation and maintenance through data acquisition operations.

[0007] The data optimization module specifically obtains optimized power plant operation and maintenance data through data cleaning, data standardization, and data encoding.

[0008] The module for establishing a power plant equipment maintenance prediction model specifically involves constructing an equipment maintenance network graph structure by introducing a process topology weighting mechanism, then extracting maintenance time-series features, and then extracting the direct spatial features of explicit process relationships between adjacent equipment and the global indirect spatial features of implicit relationships between non-adjacent equipment based on a graph attention network. Finally, the multi-layer features are globally aggregated to output the power plant equipment operation and maintenance demand prediction results. The model is then trained based on historical operation and maintenance data to obtain the trained power plant equipment maintenance prediction model.

[0009] The intelligent operation and maintenance management module is used to realize intelligent operation and maintenance management of power plant equipment. Specifically, it inputs real-time operation and maintenance data into the trained model to obtain real-time operation and maintenance results of the power plant, and realizes intelligent operation and maintenance management of power plant equipment based on these results.

[0010] Furthermore, the multi-source data collection module specifically obtains raw power plant operation and maintenance data through data acquisition operations; the raw power plant operation and maintenance data includes historical operation and maintenance data and real-time operation and maintenance data; both the historical operation and maintenance data and the real-time operation and maintenance data include power plant equipment operation data, fault event data and equipment maintenance data; the historical operation and maintenance data also includes operation and maintenance records.

[0011] Furthermore, the data optimization module specifically includes the following steps:

[0012] Data cleaning specifically involves imputing missing values, removing outliers, and normalizing fields in the original data.

[0013] Data standardization is specifically performed by mapping all continuous variables to the [0,1] interval using the min-max normalization method.

[0014] The data encoding process specifically involves using one-hot encoding to encode the category fields in the original data.

[0015] Furthermore, the module for establishing a power plant equipment maintenance prediction model specifically includes the following steps:

[0016] The equipment maintenance network graph structure is constructed by introducing a process topology weighting mechanism to define graph network nodes, graph network edges, and embed the parallel operation relationship of multiple devices, thereby obtaining the equipment maintenance network graph structure, the equipment process dependency adjacency matrix, and the set of multi-device graph network nodes.

[0017] The graph network node definition is specifically as follows: power plant operation and maintenance equipment is defined as graph network nodes based on power plant equipment operation data, and each node is uniquely associated with the operation and maintenance data of the corresponding equipment.

[0018] The graph network edge definition specifically involves defining graph network edges based on the equipment process dependencies between power plant equipment, and generating an adjacency matrix based on these dependencies. If device c has a process dependency on device h, then the adjacency matrix elements ,otherwise , where n is the total number of maintenance equipment;

[0019] The equipment process dependency relationship is specifically defined as follows: if the operating status of equipment c directly affects the maintenance requirements of equipment h, then a directed edge is established between the node corresponding to equipment c and the node corresponding to equipment h, forming a direct equipment process dependency relationship; specifically, it includes three types of equipment process dependencies: energy transfer dependency, material supply dependency, and control signal dependency.

[0020] The embedding of the multi-device parallel operation relationship specifically adopts time window slicing. The operation and maintenance data of multiple devices are sliced ​​through time windows, and the operation and maintenance data of multiple devices within the same time window are mapped to the initial features of the corresponding nodes to obtain a set of nodes in the multi-device graph network.

[0021] The maintenance feature temporal dependency extraction is specifically carried out by extracting the graph network node features of each device node for T consecutive time windows from the multi-device graph network node set, constructing the temporal features of the corresponding device nodes, and then using a temporal attention mechanism model to calculate the original attention weights between different time windows to obtain the original attention weights. Subsequently, the original attention weights are normalized to obtain normalized temporal attention weights, which are then multiplied by the temporal features to obtain the maintenance temporal features.

[0022] The direct spatial feature extraction of adjacent devices is specifically carried out by locating the set of adjacent devices for each device node based on the device process dependency adjacency matrix, inputting the maintenance time-series features into the graph attention network, performing dimensional mapping on the features of the current node and adjacent nodes, calculating the attention weights of the node and adjacent nodes, and using a multi-head attention mechanism to aggregate multiple sets of adjacency graph node features and update the current node features to obtain the direct spatial features.

[0023] Global indirect spatial feature extraction for equipment is used to capture global indirect process dependency information between equipment nodes in a graph network; specifically, it includes the following steps:

[0024] Multi-hop adjacency relationship localization specifically involves generating a multi-hop adjacency matrix based on the equipment process dependency adjacency matrix through matrix exponentiation. Among them, the elements in the multi-hop adjacency matrix =1 indicates the device and There exists a v-hop indirect process dependency, where v represents the hop count. By integrating all v-hop adjacency matrices, a global indirect adjacency set is constructed. ;

[0025] The indirect process dependency refers to an indirect association relationship formed through one or more intermediate devices, where there is no direct process dependency between equipment nodes.

[0026] The calculation of global indirect attention weights involves inputting the maintained temporal features into the graph attention network, performing a high-dimensional mapping of the features of the current device and indirect devices using a feature transformation matrix, and then calculating the global indirect attention weights between the device and the indirect devices. The formula used is as follows:

[0027] ;

[0028] In the formula, This represents the normalized global attention weight between the current device node c and the device node u hop from v. This represents the transpose of the global indirect attention coefficient vector. This represents the feature transformation matrix used in the calculation of global indirect attention weights. This represents the maintenance timing characteristics of device node u in the i-th time window. Indicates feature concatenation operation;

[0029] Global feature aggregation specifically involves introducing a multi-hop distance decay factor, and first aggregating the temporal features of v-hop indirect devices and global indirect attention weights according to the number of hops v. Then, the aggregation results of different hop counts are weighted and summed. Simultaneously, combined with a multi-head attention mechanism, multiple sets of aggregated features are concatenated to update the features of the current device node, resulting in global indirect spatial features. The formula used is as follows:

[0030] ;

[0031] In the formula, This represents the global indirect space characteristics of device node c. This represents the number of global attention heads, and V represents the maximum number of hops. Indicates the weight of the number of hops. This represents the multi-hop distance decay factor, which indicates the number of hops v. The range of values ​​is between, This represents the set of devices in the v-hop indirection of device c. Let represent the normalized attention weight of the v-hop indirect device u in the g-th attention head with respect to the current device c. This represents the global feature transformation matrix of the g-th attention head. This indicates a multi-head feature splicing operation;

[0032] The power plant equipment operation and maintenance demand prediction output specifically involves cross-dimensional fusion of direct spatial features and global indirect spatial features to generate spatiotemporal fusion features. These spatiotemporal fusion features are then input into a fully connected layer. After nonlinear feature mapping and dimension adaptation are completed through the fully connected layer, the probability distribution of different operation and maintenance operation categories corresponding to each device is quantized and output through the Softmax activation function. The category with the highest probability is selected as the power plant equipment operation and maintenance demand prediction result.

[0033] The model is constructed and trained by building a network graph structure for equipment maintenance, extracting the temporal dependency of maintenance features, extracting the direct spatial features of adjacent equipment, extracting the global indirect spatial features of equipment, and predicting the operation and maintenance needs of power plant equipment. The model is then trained using historical operation and maintenance data to obtain the final trained power plant equipment maintenance prediction model.

[0034] Furthermore, the intelligent operation and maintenance management module specifically inputs real-time operation and maintenance data into the trained power plant equipment maintenance prediction model to obtain real-time operation and maintenance results of the power plant, and generates corresponding operation and maintenance decisions and execution instructions based on the real-time operation and maintenance results of the power plant, executes intelligent operation and maintenance operations, automatically issues early warnings, generates maintenance tasks and dispatches equipment maintenance personnel, thereby realizing intelligent operation and maintenance management of power plant equipment.

[0035] The beneficial effects achieved by the present invention using the above solution are as follows:

[0036] (1) In view of the technical problem that traditional intelligent operation and maintenance management systems are limited to the micro-data analysis of a single device and fail to consider the parallel operation characteristics of multiple devices and the process interaction relationship between devices, the system cannot build a global network of power plants and it is difficult to capture the global chain failure propagation effect caused by local device anomalies, thus making the decision-making of power plant operation and maintenance management lagging behind. This solution innovatively proposes to build a device maintenance network graph structure with devices as network nodes and device process dependencies as directed edges. It can effectively identify the risk propagation chain between different devices, improve the ability to identify cross-device related risks, enhance the global coordination and dynamic responsiveness of operation and maintenance, improve the comprehensiveness and real-time performance of prediction, break through the limitations of traditional single device prediction, and transform power plant operation and maintenance management from local prediction to global correlation modeling, significantly improve the coordination and decision-making efficiency of the whole plant-level maintenance strategy, and promote the scientific allocation of operation and maintenance resources.

[0037] (2) In view of the technical problem that existing prediction models for power plant equipment maintenance generally only consider the direct adjacency relationship between equipment and ignore the indirect process coupling of non-adjacent equipment, resulting in insufficient prediction accuracy and weak response to the global process status, which affects the reliability and foresight of operation and maintenance decisions, this solution innovatively adopts a joint modeling mechanism of direct and indirect two-layer spatial dependencies. Direct spatial dependencies model the explicit process relationship between adjacent equipment through graph attention network, while indirect spatial dependencies model the implicit association of non-adjacent equipment through spatial attention mechanism. Finally, the two types of spatial features are fused and aggregated to achieve a comprehensive capture of multi-level spatial dependencies. It can identify potential risk propagation paths and implicit dependencies between equipment, significantly improve the global cognitive ability and prediction accuracy of the model, enhance the interpretability and robustness of the system for the collaborative operation status of multiple equipment, and thus improve the accuracy and scientific nature of intelligent operation and maintenance decision-making. Attached Figure Description

[0038] Figure 1 is a schematic diagram of the modules of an intelligent operation and maintenance management system based on artificial intelligence provided by the present invention;

[0039] Figure 2 is a flowchart illustrating the process of establishing a power plant equipment maintenance prediction model module;

[0040] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0043] Example 1, referring to Figure 1, the present invention provides an intelligent operation and maintenance management system based on artificial intelligence, including a multi-source data collection module, a data optimization module, a power plant equipment maintenance prediction model establishment module, and an intelligent operation and maintenance management module;

[0044] The multi-source data collection module specifically obtains raw data of power plant operation and maintenance through data acquisition operations and sends the data to the data optimization module.

[0045] The data optimization module receives data sent by the multi-source data collection module to improve the quality of the original data and the adaptability of the model. Specifically, it obtains power plant operation and maintenance optimization data through data cleaning, data standardization and data encoding, and sends the data to the power plant equipment maintenance prediction model module and the intelligent operation and maintenance management module.

[0046] The module for establishing a power plant equipment maintenance prediction model receives data sent by the data optimization module. Specifically, it constructs an equipment maintenance network graph structure by introducing a process topology weighting mechanism, then extracts maintenance time-series features, and then extracts the direct spatial features of explicit process relationships between adjacent equipment and the global indirect spatial features of implicit relationships between non-adjacent equipment based on a graph attention network. Finally, it performs global aggregation of multi-layer features, outputs the prediction results of power plant equipment operation and maintenance needs, and trains the model based on historical operation and maintenance data to obtain the trained power plant equipment maintenance prediction model. The data is then sent to the intelligent operation and maintenance management module.

[0047] The intelligent operation and maintenance management module receives data from the data optimization module and the power plant equipment maintenance prediction model module to realize intelligent operation and maintenance management of power plant equipment. Specifically, it inputs real-time operation and maintenance data into the trained model to obtain real-time operation and maintenance results of the power plant, and realizes intelligent operation and maintenance management of power plant equipment based on these results.

[0048] Example 2, referring to Figure 1, is based on the above example. The multi-source data collection module is used to collect multi-dimensional data of power plant equipment during operation and maintenance management. Specifically, it collects equipment operating parameter data through various types of sensors, data acquisition terminals, and industrial IoT nodes deployed on key power plant equipment to obtain raw power plant operation and maintenance data. The raw power plant operation and maintenance data includes historical operation and maintenance data and real-time operation and maintenance data. Both historical and real-time operation and maintenance data include power plant equipment operation data, fault event data, and equipment maintenance data. The historical operation and maintenance data also includes operation and maintenance records.

[0049] The power plant equipment operation data includes equipment name, equipment power output, load rate, temperature, pressure, speed, current, voltage, and vibration frequency; the fault event data includes historical fault types, fault occurrence time, downtime, fault location, and fault cause analysis; the equipment maintenance data includes equipment maintenance date, maintenance cycle, maintenance content, maintenance method, maintenance time, and information on replaced parts.

[0050] Example 3, referring to Figure 1, is based on the above examples. The data optimization module is used to improve the integrity, consistency, and effectiveness of the original data. Specifically, it performs data cleaning, data standardization, and data encoding on the original data to obtain optimized power plant operation and maintenance data. The steps include:

[0051] Data cleaning is used to improve the quality and consistency of raw data, specifically by imputing missing values, removing outliers, and normalizing fields in the raw data.

[0052] The missing value imputation specifically involves filling in missing values ​​using the mean imputation method; the outlier removal specifically involves detecting and removing extreme values ​​and logical outliers in the original data using the Z-Score algorithm; and the field normalization process specifically involves standardizing data of different formats using a unified unit conversion rule to ensure the consistency of model input.

[0053] Data standardization is used to unify the dimensions of structured numerical data. Specifically, it maps all continuous variables to the [0,1] interval using the min-max normalization method.

[0054] Data encoding processing is used to convert non-numerical data into structured data that can be computed by the model. Specifically, it uses one-hot encoding to encode the category fields in the original data and converts discrete text and label variables into sparse numerical vectors.

[0055] Example 4, referring to Figures 1 and 2, is based on the above examples. The module for establishing a power plant equipment maintenance prediction model specifically includes the following steps:

[0056] A device maintenance network graph structure is constructed to transform scattered device operation and maintenance data into a structured graph network based on the operational relationships between power plant devices and the parallel characteristics of multiple devices. Specifically, a process topology weighting mechanism is introduced to define graph network nodes, graph network edges, and embed the parallel operation relationships of multiple devices, resulting in the device maintenance network graph structure, the device process dependency adjacency matrix, and the set of multi-device graph network nodes.

[0057] The graph network node definition is specifically as follows: power plant operation and maintenance equipment is defined as graph network nodes based on power plant equipment operation data, and each node is uniquely associated with the operation and maintenance data of the corresponding equipment.

[0058] The graph network edge definition specifically involves defining graph network edges based on the equipment process dependencies between power plant equipment, and generating an adjacency matrix based on these dependencies. If device c has a process dependency on device h, then the adjacency matrix elements ,otherwise , where n is the total number of maintenance equipment;

[0059] The equipment process dependency relationship is specifically defined as follows: if the operating status of equipment c directly affects the maintenance requirements of equipment h, then a directed edge is established between the node corresponding to equipment c and the node corresponding to equipment h, forming a direct equipment process dependency relationship; specifically, it includes three types of equipment process dependencies: energy transfer dependency, material supply dependency, and control signal dependency.

[0060] The energy transfer dependency specifically refers to the fact that device c provides core energy to device h, and the energy output status of device c directly affects the operational stability and maintenance requirements of device h.

[0061] The material supply dependence specifically refers to the fact that equipment c provides key production materials to equipment h, and the material supply status of equipment c directly affects the operational stability and maintenance requirements of equipment h.

[0062] The control signal dependence specifically refers to the fact that device c provides control command signals to device h, and the signal output state of device c directly affects the operational stability and maintenance requirements of device h.

[0063] The embedding of the multi-device parallel operation relationship specifically adopts time window slicing, with the window duration set to 1 hour. The operation and maintenance data of multiple devices are sliced ​​through the time window, and the operation and maintenance data of multiple devices within the same time window are mapped to the initial features of the corresponding nodes to obtain a set of multi-device graph network nodes.

[0064] The parallel operation of multiple devices specifically refers to the operation and maintenance related events that are executed synchronously by different power plant devices around the same production target or process link within the same time window.

[0065] Maintenance feature temporal dependency extraction is used to capture the temporal dynamics of power plant equipment maintenance features. Specifically, it extracts graph network node features for each equipment node over T consecutive time windows from a multi-equipment graph network node set, constructing the temporal features of the corresponding equipment nodes. Then, a temporal attention mechanism model is used to calculate the original attention weights between different time windows, obtaining the original attention weights. Subsequently, the original attention weights are normalized using the Softmax activation function to obtain normalized temporal attention weights. These normalized temporal attention weights are then multiplied by the temporal features to obtain the maintenance temporal features. The formula used is as follows:

[0066] ;

[0067] ;

[0068] In the formula, The elements in the original attention weight matrix represent the influence strength of the temporal features of the device node in the j-th time window on the temporal features of the device node in the i-th time window. This represents a learnable weight vector. , and This represents a learnable parameter matrix used for dimensional mapping and interaction of maintenance features. Indicates the first The temporal characteristics of each time window Indicates the first The temporal characteristics of each time window Learnable bias terms are used to adjust the baseline level of the original attention weights. Indicates the first Maintenance timing characteristics of each time window, Represents the normalized temporal attention weights;

[0069] Direct spatial feature extraction of adjacent devices is used to aggregate explicit process dependency information of adjacent devices in the graph network, capturing the impact of direct process associations between devices on maintenance features. Specifically, based on the device process dependency adjacency matrix, the set of adjacent devices for each device node is located, and the maintenance time-series features are input into the graph attention network. Dimensional mapping is performed on the features of the current node and its adjacent nodes, and the attention weights between the node and its adjacent nodes are calculated. A multi-head attention mechanism is used to aggregate multiple sets of adjacency graph node features and update the current node's features to obtain the direct spatial features. The formula used is as follows:

[0070] ;

[0071] ;

[0072] In the formula, This represents the normalized attention weights between the current device node c and its adjacent device node h, used to quantify the strength of the direct process dependency of h on c. This represents the transpose of the attention coefficient vector, used to calculate the interaction score of the concatenated features. Represents the feature transformation matrix, This represents the maintenance timing characteristics of device node c in the i-th time window. This represents the maintenance timing characteristics of device node h in the i-th time window. This indicates a feature concatenation operation. This represents the direct spatial characteristics of device node c. This represents a multi-head feature concatenation operation, which concatenates the output features of K attention heads along a dimension, where K represents the number of attention heads. This represents the feature transformation matrix of the k-th attention head. This represents the set of adjacent devices of the current device node c. This represents the normalized attention weights between the current node c and its neighboring node h in the k-th attention head.

[0073] Global indirect spatial feature extraction for equipment is used to capture global indirect process dependency information between equipment nodes in a graph network; specifically, it includes the following steps:

[0074] Multi-hop adjacency relationship localization specifically involves generating a multi-hop adjacency matrix based on the equipment process dependency adjacency matrix through matrix exponentiation. Among them, the elements in the multi-hop adjacency matrix =1 indicates the device and There exists a v-hop indirect process dependency, where v represents the hop count. By integrating all v-hop adjacency matrices, a global indirect adjacency set is constructed. ;

[0075] The indirect process dependency refers to an indirect association relationship formed through one or more intermediate devices, where there is no direct process dependency between equipment nodes.

[0076] The calculation of global indirect attention weights involves inputting the maintained temporal features into the graph attention network, performing a high-dimensional mapping of the features of the current device and indirect devices using a feature transformation matrix, and then calculating the global indirect attention weights between the device and the indirect devices. The formula used is as follows:

[0077] ;

[0078] In the formula, This represents the normalized global attention weight between the current device node c and the device node u hop from v. This represents the transpose of the global indirect attention coefficient vector. This represents the feature transformation matrix used in the calculation of global indirect attention weights. This represents the maintenance timing characteristics of device node u in the i-th time window;

[0079] Global feature aggregation specifically involves introducing a multi-hop distance decay factor, and first aggregating the temporal features of v-hop indirect devices and global indirect attention weights according to the number of hops v. Then, the aggregation results of different hop counts are weighted and summed. Simultaneously, combined with a multi-head attention mechanism, multiple sets of aggregated features are concatenated to update the features of the current device node, resulting in global indirect spatial features. The formula used is as follows:

[0080] ;

[0081] In the formula, This represents the global indirect space characteristics of device node c. This represents the number of global attention heads, and V represents the maximum number of hops. Indicates the weight of the number of hops. This represents the multi-hop distance decay factor, which indicates the number of hops v. The range of values ​​is between, This represents the set of devices in the v-hop indirection of device c. Let represent the normalized attention weight of the v-hop indirect device u in the g-th attention head with respect to the current device c. This represents the global feature transformation matrix of the g-th attention head. This indicates a multi-head feature splicing operation;

[0082] The power plant equipment operation and maintenance demand prediction output specifically involves cross-dimensional fusion of direct spatial features and global indirect spatial features to generate spatiotemporal fusion features. These spatiotemporal fusion features are then input into a fully connected layer. After nonlinear feature mapping and dimension adaptation are completed through the fully connected layer, the probability distribution of different operation and maintenance operation categories corresponding to each device is quantized and output through the Softmax activation function. The category with the highest probability is selected as the power plant equipment operation and maintenance demand prediction result.

[0083] The equipment operation and maintenance categories include emergency shutdown for maintenance, planned shutdown for repair, online component replacement, enhanced inspection, and routine monitoring;

[0084] The model is constructed and trained by building a network graph structure for equipment maintenance, extracting the temporal dependency of maintenance features, extracting the direct spatial features of adjacent equipment, extracting the global indirect spatial features of equipment, and predicting the operation and maintenance needs of power plant equipment. The model is then trained using historical operation and maintenance data to obtain the final trained power plant equipment maintenance prediction model.

[0085] By performing the above operations, this solution addresses the technical problem of traditional intelligent operation and maintenance management systems, which are limited to micro-data analysis of single devices and fail to consider the parallel operation characteristics of multiple devices and cross-device process interactions. This results in the system's inability to construct a global interconnected network for the power plant, making it difficult to capture the global cascading failure propagation effect caused by local device anomalies, thus leading to lagging power plant operation and maintenance management decisions. This solution innovatively proposes constructing a device maintenance network graph structure with devices as network nodes and device process dependencies as directed edges. This effectively identifies risk propagation chains between different devices, improves the ability to identify cross-device associated risks, enhances the global coordination and dynamic responsiveness of operation and maintenance operations, and improves the comprehensiveness and real-time nature of predictions. It breaks through the limitations of traditional single-device prediction, transforming power plant operation and maintenance management from local prediction to global interconnected modeling, significantly improving the coordination and decision-making efficiency of plant-wide maintenance strategies, and promoting the scientific management of operation and maintenance resources. To address the technical problem that existing power plant equipment maintenance prediction models generally only consider direct adjacency relationships between equipment and ignore indirect process coupling between non-adjacent equipment, resulting in insufficient prediction accuracy and weak responsiveness to global process states, thus affecting the reliability and foresight of operation and maintenance decisions, this solution innovatively adopts a joint modeling mechanism of direct and indirect spatial dependencies. Direct spatial dependencies model explicit process relationships between adjacent equipment through graph attention networks, while indirect spatial dependencies model implicit associations between non-adjacent equipment through spatial attention mechanisms. Finally, the two types of spatial features are fused and aggregated to achieve a comprehensive capture of multi-level spatial dependencies. This enables the identification of potential risk propagation paths and implicit dependencies between equipment, significantly improving the model's global cognitive ability and prediction accuracy, enhancing the system's interpretability and robustness to the collaborative operation status of multiple equipment, and thus improving the accuracy and scientific nature of intelligent operation and maintenance decision-making.

[0086] Example 5, referring to Figure 1, is based on the above examples. The intelligent operation and maintenance management module specifically inputs real-time operation and maintenance data into the trained power plant equipment maintenance prediction model to obtain the real-time operation and maintenance results of the power plant. Based on the real-time operation and maintenance results, it generates corresponding operation and maintenance decisions and execution instructions, executes intelligent operation and maintenance operations, automatically issues warnings, generates maintenance tasks, and dispatches equipment maintenance personnel to realize intelligent operation and maintenance management of power plant equipment.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0089] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent operation and maintenance management system based on artificial intelligence, characterized in that: The system includes a multi-source data collection module, a data optimization module, a power plant equipment maintenance prediction model building module, and an intelligent operation and management module. Specifically, the multi-source data collection module obtains raw power plant operation and maintenance data through data acquisition. The data optimization module obtains optimized operation and maintenance data through data cleaning, standardization, and encoding. The power plant equipment maintenance prediction model building module constructs an equipment maintenance network graph structure by introducing a process topology weighting mechanism, then extracts maintenance time-series features, and then extracts the direct spatial features of explicit process relationships between adjacent equipment and the global indirect spatial features of implicit relationships between non-adjacent equipment based on a graph attention network. Finally, it performs a full analysis of the multi-layer features. The system aggregates and outputs predictions of power plant equipment operation and maintenance needs. Based on historical operation and maintenance data, it trains the model to obtain a trained power plant equipment maintenance prediction model. Specifically, this includes the following steps: constructing an equipment maintenance network graph structure; extracting the temporal dependency of maintenance features, specifically extracting the graph network node features of each equipment node for T consecutive time windows from a multi-equipment graph network node set, constructing the temporal features of the corresponding equipment nodes, then using a temporal attention mechanism model to calculate the original attention weights between different time windows, obtaining the original attention weights, then normalizing the original attention weights to obtain normalized temporal attention weights, multiplying them by the temporal features to obtain the maintenance temporal features; and identifying direct spatial features of adjacent equipment. The extraction process involves locating the set of neighboring devices for each device node based on the device's process dependency adjacency matrix, inputting maintenance time-series features into a graph attention network, performing dimensional mapping on the features of the current node and its neighboring nodes, calculating the attention weights between the node and its neighboring nodes, and using a multi-head attention mechanism to aggregate multiple sets of adjacency graph node features and update the current node's features to obtain direct spatial features. Global indirect spatial features of the equipment are also extracted. The power plant equipment operation and maintenance demand prediction output involves cross-dimensional fusion of direct spatial features and global indirect spatial features to generate spatiotemporal fusion features. These spatiotemporal fusion features are then input into a fully connected layer. After nonlinear feature mapping and dimensional adaptation are completed through the fully connected layer, each device is quantized and output using a Softmax activation function. Based on the probability distribution of different operation and maintenance (O&M) operation categories, the category with the highest probability is selected as the prediction result for power plant equipment O&M demand. A model is constructed and trained, specifically by building an equipment maintenance network graph structure, extracting the temporal dependency of maintenance features, extracting direct spatial features of adjacent devices, extracting global indirect spatial features of devices, and generating a power plant equipment O&M demand prediction output. This constructs a power plant equipment maintenance prediction model, which is then trained using historical O&M operation data. The intelligent O&M operation management module specifically inputs real-time O&M operation data into the trained model to obtain real-time O&M operation results for the power plant. Based on these results, intelligent O&M management of the power plant equipment is achieved.

2. The intelligent operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that: The construction of the equipment maintenance network graph structure specifically involves introducing a process topology weighting mechanism to define graph network nodes, graph network edges, and embedding multi-device parallel operation relationships, resulting in an equipment maintenance network graph structure, an equipment process dependency adjacency matrix, and a set of multi-device graph network nodes. Specifically, the graph network node definition involves defining power plant maintenance equipment as graph network nodes based on power plant equipment operation data, with each node uniquely associated with the maintenance operation data of the corresponding equipment. The graph network edge definition involves defining graph network edges based on the equipment process dependencies between power plant equipment, and generating an adjacency matrix based on these dependencies. If device c has a process dependency on device h, then the adjacency matrix elements ,otherwise Where n is the total number of maintenance equipment; the equipment process dependency relationship is specifically defined as follows: if the operating status of equipment c directly affects the maintenance requirements of equipment h, then a directed edge is established between the node corresponding to equipment c and the node corresponding to equipment h, forming a direct equipment process dependency relationship; specifically including three types of equipment process dependencies: energy transfer dependency, material supply dependency, and control signal dependency; the embedding of the multi-equipment parallel operation relationship is specifically achieved by using time window slicing, which slices the multi-equipment operation data through time windows, and maps the operation data of multiple devices within the same time window to the initial features of the corresponding nodes, thereby obtaining a set of nodes in the multi-equipment graph network.

3. The intelligent operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that: The extraction of global indirect spatial features of the device specifically includes the following steps: multi-hop adjacency relationship localization; global indirect attention weight calculation, specifically by inputting the maintenance temporal features into a graph attention network, performing high-dimensional mapping of the features of the current device and indirect devices through a feature transformation matrix, and calculating the global indirect attention weights between the device and indirect devices; the formula used is as follows: In the formula, This represents the normalized global indirect attention weight between the current device node c and the device node u hop from v. This represents the transpose of the global indirect attention coefficient vector. This represents the feature transformation matrix used in the calculation of global indirect attention weights. This represents the maintenance timing characteristics of device node c in the i-th time window. This represents the maintenance timing characteristics of device node u in the i-th time window. This represents the feature concatenation operation; global feature aggregation specifically involves introducing a multi-hop distance decay factor, and first aggregating the temporal features of v-hop indirect devices and the global indirect attention weights according to the number of hops v. Then, the aggregation results of different hop counts are weighted and summed. Simultaneously, combined with a multi-head attention mechanism, multiple sets of aggregated features are concatenated to update the features of the current device node, resulting in the global indirect spatial features. The formula used is as follows: In the formula, This represents the global indirect space characteristics of device node c. This represents the number of global attention heads, and V represents the maximum number of hops. Indicates the weight of the number of hops. This represents the multi-hop distance decay factor, which indicates the number of hops v. The range of values ​​is between, This represents the set of devices in the v-hop indirection of device c. Let represent the normalized global indirect attention weight of the v-hop indirect device u in the g-th attention head relative to the current device c. This represents the global feature transformation matrix of the g-th attention head. This indicates a multi-head feature splicing operation.

4. The intelligent operation and maintenance management system based on artificial intelligence according to claim 3, characterized in that: The multi-hop adjacency relationship localization is specifically based on the equipment process-dependent adjacency matrix, and generates a multi-hop adjacency matrix through matrix exponentiation. Among them, the elements in the multi-hop adjacency matrix =1 indicates the device and There exists a v-hop indirect process dependency, where v represents the hop count. By integrating all v-hop adjacency matrices, a global indirect adjacency set is constructed. The indirect process dependency refers to an indirect relationship formed through one or more intermediate devices, where there is no direct process dependency between equipment nodes.

5. The intelligent operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that: The multi-source data collection module specifically obtains raw power plant operation and maintenance data through data acquisition operations. The raw power plant operation and maintenance data includes historical operation and maintenance data and real-time operation and maintenance data. Both the historical operation and maintenance data and the real-time operation and maintenance data include power plant equipment operation data, fault event data, and equipment maintenance data. The historical operation and maintenance data also includes operation and maintenance records.

6. The intelligent operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that: The data optimization module specifically includes the following steps: data cleaning, specifically imputing missing values, removing outliers, and normalizing fields in the original data; data standardization, specifically mapping all continuous variables to the [0,1] interval using the min-max normalization method; and data encoding, specifically encoding the category fields in the original data using the one-hot encoding method.

7. The intelligent operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that: The intelligent operation and maintenance management module specifically inputs real-time operation and maintenance data into the trained power plant equipment maintenance prediction model to obtain real-time operation and maintenance results of the power plant. Based on the real-time operation and maintenance results of the power plant, it generates corresponding operation and maintenance decisions and execution instructions, executes intelligent operation and maintenance operations, automatically issues early warnings, generates maintenance tasks, and dispatches equipment maintenance personnel to realize intelligent operation and maintenance management of power plant equipment.

Images