Intelligent operation and maintenance processing method and system for thermoelectric unit

Abstract

The invention provides an intelligent operation and maintenance processing method and system for a thermoelectric unit, and the method comprises the steps: obtaining a real-time operation data set of the thermoelectric unit, constructing an equipment association graph to display the operation association relation between equipment, calling a pre-trained graph neural network model to carry out the feature learning of the equipment association graph, and carrying out the feature learning of the equipment association graph; the method comprises the following steps: generating global association features reflecting interactive influence among equipment states, identifying an equipment abnormal state based on the global association features, generating an identification result containing an abnormal equipment identifier and type, generating an operation and maintenance optimization strategy according to the identification result, and feeding back the operation and maintenance optimization strategy to a thermoelectric unit operation and maintenance management system to trigger maintenance operation. Therefore, intelligence and automation of operation and maintenance management of the thermoelectric unit are realized, the operation and maintenance efficiency and accuracy are improved, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of thermal power unit operation and maintenance technology, and more specifically, to an intelligent operation and maintenance method and system for thermal power units. Background Technology

[0002] In the field of operation and maintenance management of cogeneration units, traditional methods mainly rely on manual inspections and periodic maintenance. This approach is not only inefficient but also makes it difficult to grasp the real-time and accurate operating status of each piece of equipment in the cogeneration unit. With the continuous expansion and increasing complexity of cogeneration units, the operational relationships between equipment are becoming increasingly complex. Changes in the status of a single piece of equipment can have a cascading effect on other equipment, leading to a decrease in the overall operating efficiency of the unit or even malfunctions. While existing operation and maintenance technologies can collect real-time operating data, they lack in-depth mining and analysis of the relationships between these data, failing to effectively identify the interactive effects between equipment, thus making it difficult to formulate scientific and reasonable operation and maintenance optimization strategies. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an intelligent operation and maintenance method for thermal power units, the method comprising: Acquire a real-time operating data set of the thermal power unit, wherein the real-time operating data set contains a sequence of state parameters of multiple devices at consecutive time nodes; The real-time operation data set is processed to construct a graph structure to generate an equipment association graph. In the equipment association graph, nodes represent individual equipment of the thermal power unit, and edges represent the operation association relationships between equipment. The pre-trained graph neural network model is invoked to perform feature learning processing on the device association graph, generating global association features that reflect the interaction and influence between device states; Based on the global correlation features, the abnormal device status identification process is performed to generate an identification result containing the abnormal device identifier and the abnormality type. Based on the identification results, an operation and maintenance optimization strategy for the thermal power unit is generated, and the operation and maintenance optimization strategy is fed back to the thermal power unit operation and maintenance management system to trigger maintenance operations.

[0004] In another aspect, embodiments of the present invention also provide an intelligent operation and maintenance system for thermal power units, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0005] Based on the above, this embodiment of the invention acquires real-time operating data sets of thermal power units and constructs equipment association diagrams, which can intuitively display the operational relationships between equipment. By calling a pre-trained graph neural network model to perform feature learning processing on the equipment association diagram, it can deeply explore the interactive influence between equipment states, generate information reflecting global association features, and perform equipment abnormal state identification processing based on global association features. This can accurately and timely identify abnormal equipment and its types. The operation and maintenance optimization strategies generated based on the identification results can specifically solve equipment abnormality problems, improve the operating efficiency and stability of thermal power units, thereby realizing the intelligent and automated operation and maintenance management of thermal power units, significantly improving operation and maintenance efficiency and accuracy, reducing operation and maintenance costs, and thus contributing to the safe and stable operation of thermal power units. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the intelligent operation and maintenance method for thermal power units provided in the embodiments of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the intelligent operation and maintenance system for thermal power units provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an intelligent operation and maintenance method for thermal power units according to an embodiment of the present invention. The following is a detailed description of the intelligent operation and maintenance method for thermal power units.

[0009] Step S110: Obtain the real-time operation data set of the thermal power unit, wherein the real-time operation data set contains a sequence of state parameters of multiple devices at continuous time nodes.

[0010] In actual operation scenarios of thermal power units, these units contain various functional devices such as boilers, steam turbines, generators, and condensers. To obtain operational status information for these devices, appropriate sensors need to be installed on each unit. Different types of sensors are used to collect different status parameters; for example, temperature sensors collect the temperature of the equipment, pressure sensors collect the pressure, and speed sensors collect the rotational speed.

[0011] The sensors collect data at consecutive time points. Assume there are three devices in the thermal power unit, labeled Device A, Device B, and Device C. At a series of consecutive time points, such as time point 1, time point 2, time point 3, etc., the sensors of each device will collect corresponding state parameters. For Device A, the state parameters collected at time point 1 are recorded as Device A's state parameters at time point 1, the state parameters collected at time point 2 are recorded as Device A's state parameters at time point 2, and so on, forming a state parameter sequence for Device A. Similarly, Device B and Device C will also form their own state parameter sequences. Combining the state parameter sequences of these three devices constitutes the real-time operating data set of the thermal power unit.

[0012] It is important to note that the status parameters of each device are not single values, but rather contain information from multiple dimensions. For example, the status parameters of device A at a certain point in time may simultaneously include values ​​for temperature, pressure, rotational speed, and other aspects. This multi-dimensional data can more comprehensively reflect the operating status of the device.

[0013] During data acquisition, the collected data may contain errors due to factors such as environmental interference and the sensor's own accuracy. Therefore, preprocessing of the acquired data is necessary, including noise removal and data calibration. Noise removal involves identifying and eliminating abnormal data caused by interference to improve data accuracy; data calibration, on the other hand, involves comparing and adjusting the data with standard data to ensure that the acquired data accurately reflects the actual operating status of the equipment.

[0014] Step S120: Perform graph structure construction processing on the real-time operation data set to generate a device association graph. In the device association graph, nodes represent individual devices of the thermal power unit, and edges represent the operation association relationships between devices.

[0015] After obtaining the real-time operating data set of the thermal power unit, it is necessary to perform graph structure construction on this data set to generate an equipment association diagram. An equipment association diagram is a graphical structure used to represent the relationships between equipment, displaying the operational relationships between equipment through nodes and edges. The specific steps are as follows: Step S121: Perform device-dimensional partitioning on the state parameter sequence in the real-time running data set to obtain the state parameter subsequence corresponding to each device.

[0016] This step involves dividing the real-time operational data set according to the devices. Taking devices A, B, and C as examples, the real-time operational data in the set is separated by device, resulting in state parameter subsequences for device A, device B, and device C. Each state parameter subsequence contains the state parameters of that device at various time points and can independently reflect the changes in the device's operational status.

[0017] Step S122: Assign a unique node identifier to each device's corresponding state parameter subsequence, bind and map the node identifier to the device's physical identifier, and generate a node set of the device association graph.

[0018] After obtaining the state parameter subsequence for each device, a unique node identifier needs to be assigned to each state parameter subsequence. This node identifier is used to uniquely represent a device in the device association graph. For example, node identifier A is assigned to the state parameter subsequence of device A, node identifier B is assigned to the state parameter subsequence of device B, and node identifier C is assigned to the state parameter subsequence of device C.

[0019] Then, the node identifiers are bound and mapped to the physical identifiers of the devices. The physical identifier of a device is its actual number or name within the thermal power unit. This binding mapping establishes a correspondence between node identifiers and actual devices. All node identifiers are then combined to generate a node set for the device association graph, where each node represents a device within the thermal power unit.

[0020] Step S123: Extract the time-series correlation index between different device status parameter subsequences in the real-time operation data set. The time-series correlation index is obtained by calculating the consistency of the change trend of any two device status parameter subsequences at the same time node.

[0021] To determine the operational relationships between devices, it is necessary to extract a time-series correlation index between different device state parameter subsequences. This time-series correlation index reflects the degree of consistency in the trend of state parameter changes between any two devices at the same time point. The specific steps are as follows: Step S1231: Perform time alignment processing on any two device status parameter subsequences so that the two device status parameter subsequences contain the same number of time nodes.

[0022] In actual data acquisition, the state parameter subsequences of different devices may differ at different time points. To accurately calculate the time series correlation index, it is necessary to perform time alignment processing on any two device state parameter subsequences. For example, the state parameter subsequence of device A may contain more time points, while the state parameter subsequence of device B may contain fewer time points. In this case, interpolation or deletion of redundant data points can be used to make the two state parameter subsequences have the same number of time points.

[0023] Step S1232: Calculate the matching degree of the change direction of the two device state parameters at each time node. When the matching degree of the change direction is 1, it means that the two device state parameters rise or fall at the same time node. When it is 0, it means that the change direction is opposite.

[0024] After completing the time alignment process, for each time point, the matching degree of the change direction of the state parameters of the two devices is calculated. Specifically, this can be achieved by comparing the changes in the state parameters of the two devices at adjacent time points. If the state parameters of the two devices simultaneously increase or decrease relative to the previous time point, the matching degree of the change direction is 1; if the state parameter of one device increases while the state parameter of the other device decreases, the matching degree of the change direction is 0.

[0025] Step S1233: Calculate the rate of difference in the change of the two device status parameters at each time point. The rate of difference in the change is the ratio of the absolute change of the two device status parameters at that time point.

[0026] In addition to the matching degree of change direction, it is also necessary to consider the rate of difference in the change magnitude of the two device status parameters at each time node. The calculation method is to first calculate the absolute change of the two devices at the current time node relative to the previous time node, and then divide the two absolute changes to obtain the rate of difference in change magnitude. This rate of difference in change magnitude reflects the relative magnitude of the change in the status parameters of the two devices at that time node.

[0027] Step S1234: Perform standardized weighted processing on the change direction matching degree and the change magnitude difference rate to generate a time-series correlation score for each time node.

[0028] After obtaining the direction-of-change matching degree and magnitude-of-change difference rate at each time point, these two values ​​need to be standardized and weighted to generate a time-series association score for each time point. Assume the weight of the direction-of-change matching degree is w1, the weight of the magnitude-of-change difference rate is w2, and w1 + w2 = 1. For each time point, multiply the standardized direction-of-change matching degree by w1 and the standardized magnitude-of-change difference rate by w2, then add the two results to obtain the time-series association score for that time point.

[0029] Step S1235: Calculate the mean of the time series correlation scores at all time points to generate a time series correlation index that reflects the consistency of the overall change trend of the two device state parameter subsequences.

[0030] In this embodiment, the temporal correlation scores at all time points are averaged to obtain a temporal correlation index that reflects the consistency of the overall changing trends of the state parameter subsequences of the two devices. Specifically, the temporal correlation scores at all time points are summed, and then divided by the number of time points to obtain the temporal correlation index. This temporal correlation index can quantitatively describe the degree of operational correlation between the two devices.

[0031] Step S124: Determine the strength of the operational correlation between devices based on the time-series correlation index. When the strength of the operational correlation exceeds a preset correlation threshold, establish an edge connection between the corresponding two device nodes.

[0032] After obtaining the temporal correlation index between different device state parameter subsequences, the strength of the operational association between devices is determined based on this index. The strength of the operational association can be directly represented by the temporal correlation index. A preset association threshold can be set. If the temporal correlation index between two devices exceeds this threshold, it indicates a strong operational association between the two devices, and an edge is established between the corresponding device nodes in the device association graph. For example, if the temporal correlation index between device A and device B exceeds the preset association threshold, an edge is established between node A and node B.

[0033] Step S125: Assign an edge weight value to each edge that is positively correlated with the strength of the operational association relationship, and complete the construction process of the device association graph.

[0034] After establishing edge connections, each edge needs to be assigned a weight value. The edge weight value is positively correlated with the strength of the operational correlation between devices; that is, the stronger the operational correlation, the larger the edge weight value. The temporal correlation index can be directly used as the edge weight value. For example, for the edge connecting node A and node B, its edge weight value is equal to the temporal correlation index between device A and device B. Thus, through the above steps, the construction of the device correlation graph is completed.

[0035] Step S130: Call the pre-trained graph neural network model to perform feature learning processing on the device association graph to generate global association features that reflect the interaction and influence between device states.

[0036] After generating the device association graph, a pre-trained graph neural network model needs to be invoked to perform feature learning processing on the graph, in order to generate global association features reflecting the interaction and influence between device states. A graph neural network model is a deep learning model used to process graph-structured data, capable of effectively mining potential association information between devices. The specific steps are as follows: Step S131: Input the set of nodes of the device association graph into the node feature encoding layer of the graph neural network model, perform time series feature extraction processing on the state parameter subsequence of each node, and generate local node features containing time dependence.

[0037] First, the set of nodes in the device association graph is input into the node feature encoding layer of the graph neural network model. The main function of this layer is to perform time-series feature extraction on the state parameter subsequence of each node. Taking node A (representing device A) as an example, its state parameter subsequence contains the state parameters of device A at various time nodes. The node feature encoding layer processes this state parameter subsequence, using models such as recurrent neural networks or long short-term memory networks to capture the temporal change patterns in the state parameter subsequence, generating local node features containing time dependencies. Similarly, corresponding local node features are generated for nodes B and C.

[0038] Step S132: Input the set of edges of the device association graph into the edge feature encoding layer of the graph neural network model, perform association feature extraction processing on the edge weight value of each edge and the corresponding operational association strength, and generate edge local features that reflect the interaction mode between devices.

[0039] Next, the set of edges in the device association graph is input into the edge feature encoding layer of the graph neural network model. This layer's task is to extract association features from the edge weights and corresponding operational association strengths of each edge. Taking the edge connecting node A and node B as an example, its edge weight represents the strength of the operational association between device A and device B. The edge feature encoding layer processes this edge weight value, mapping it to a high-dimensional feature space through neural network structures such as fully connected layers or convolutional layers, generating local edge features that reflect the interaction patterns between devices. The same processing is performed on other edges.

[0040] Step S133: Through the message passing layer of the graph neural network model, neighborhood information aggregation processing is performed based on the local features of the nodes and the local features of the edges. The local features of the neighboring nodes and the local features of the corresponding edges of each node are weighted and fused to generate the node neighborhood aggregation features.

[0041] Step S1331: Traverse all adjacent nodes of each node, and extract the local features of the current node, the local features of the adjacent nodes, and the local features of the edges connecting them.

[0042] In the message passing layer of a graph neural network model, it is necessary to traverse all neighboring nodes of each node. Taking node A as an example, we traverse all its neighboring nodes, and for each neighboring node, we extract the local features of node A, the local features of the neighboring node, and the local features of the edge connecting node A and the neighboring node.

[0043] Step S1332: Generate a similarity score through the feature similarity calculation module, which reflects the degree of association between the local features of the current node, the local features of adjacent nodes, and the local features of edges.

[0044] After extracting relevant features, the feature similarity calculation module calculates the correlation between the local features of the current node, the local features of adjacent nodes, and the local features of edges, obtaining a similarity score. For example, this feature similarity calculation module can use methods such as cosine similarity to calculate the similarity between features.

[0045] Step S1333: Based on the similarity score, perform weighted processing on the local node features and edge features of the adjacent nodes to generate weighted features of the adjacency information.

[0046] Based on the calculated similarity score, the local features of adjacent nodes and edges are weighted. A higher similarity score indicates a greater influence of the adjacent node and edge on the current node, and thus a higher feature weight. This weighted processing generates weighted features for adjacency information.

[0047] Step S1334: Perform weighted feature fusion processing on all adjacency information of the current node to generate the initial neighborhood aggregation feature of the current node.

[0048] The weighted features of all adjacency information of the current node are fused together, for example, by summing or concatenating, to generate the initial neighborhood aggregation feature of the current node. This initial neighborhood aggregation feature integrates the information of the current node's adjacent nodes and edges.

[0049] Step S1335: The initial features of the neighborhood aggregation are concatenated with the local features of the current node to generate a node neighborhood aggregation feature that includes the node's own information and neighborhood interaction information.

[0050] Finally, the initial neighborhood aggregation feature is concatenated with the local features of the current node to obtain the node's neighborhood aggregation feature, which includes information about the node itself and its interactions with neighboring nodes. This node neighborhood aggregation feature can more comprehensively reflect the information of the node and its interaction relationships with neighboring nodes.

[0051] Step S134: Through the global pooling layer of the graph neural network model, the neighborhood aggregation features of all nodes are integrated and processed globally to generate global aggregation features containing information on the interaction and influence between device states.

[0052] After obtaining the neighborhood aggregation features of each node, the neighborhood aggregation features of all nodes are input into the global pooling layer of the graph neural network model. The role of the global pooling layer is to perform global information integration processing on the neighborhood aggregation features of all nodes. Methods such as average pooling or max pooling can be used to integrate the neighborhood aggregation features of all nodes to generate a global aggregation feature that includes information on the interaction and influence between equipment states. This global aggregation feature can reflect the interaction and influence between equipment in the entire thermal power unit.

[0053] Step S135: Input the global aggregated features into the feature enhancement layer of the graph neural network model, and generate global correlation features that reflect the interaction between device states through feature dimension expansion and nonlinear transformation processing.

[0054] Finally, the global aggregated features are input into the feature enhancement layer of the graph neural network model. The feature enhancement layer performs feature dimension expansion and nonlinear transformation on the global aggregated features. Feature dimension expansion increases the expressive power of the features, while nonlinear transformation introduces nonlinear factors, improving the learning ability of the graph neural network model. Through the above processing, a global correlation feature reflecting the interactive influence between equipment states is generated. This global correlation feature can more accurately reflect the interactive influence relationship between equipment states in a thermal power unit.

[0055] Step S140: Based on the global association features, perform device abnormal status identification processing to generate an identification result containing the abnormal device identifier and the abnormal type.

[0056] After obtaining the global correlation features reflecting the interaction and influence between equipment states, it is necessary to perform equipment abnormality identification processing based on these features to determine whether there are abnormal devices in the thermal power unit and the type of abnormality. The specific steps are as follows: Step S141: Input the global correlation features into a pre-trained anomaly detection classifier. The anomaly detection classifier generates anomaly confidence by calculating the feature distance between the global correlation features and the normal state feature template.

[0057] First, the globally correlated features are input into a pre-trained anomaly detection classifier. The anomaly detection classifier is a trained model whose function is to calculate the feature distance between the globally correlated features and the normal-state feature template. The normal-state feature template is a set of features extracted during the normal operation of the thermal power unit, representing its normal operating state. The anomaly confidence score is obtained by calculating the feature distance between the globally correlated features and the normal-state feature template. A higher anomaly confidence score indicates that the current state of the thermal power unit deviates more from its normal state.

[0058] Step S142: When the anomaly confidence level exceeds the preset anomaly threshold, the anomaly location processing flow is triggered.

[0059] For example, a preset anomaly threshold can be set. If the calculated anomaly confidence level exceeds this threshold, it indicates that the thermal power unit may be experiencing an anomaly, triggering the anomaly location processing procedure. The purpose of the anomaly location processing procedure is to identify the specific abnormal equipment and the type of anomaly.

[0060] Step S143: In the anomaly localization process, extract the feature contribution degree corresponding to each device node in the global correlation features. The feature contribution degree represents the degree of influence of the device node on the anomaly confidence.

[0061] In the anomaly localization process, it is necessary to extract the feature contribution of each device node from the global correlation features. The feature contribution reflects the degree of influence of each device node on the anomaly confidence score. By analyzing the generation process of the global correlation features, the weight of each device node's features in the anomaly confidence score calculation can be determined, thereby obtaining the feature contribution of each device node.

[0062] Step S144: Based on the feature contribution, filter out device nodes whose contribution exceeds the preset contribution threshold, and use the node identifier of the device node as the abnormal device identifier.

[0063] For example, a preset contribution threshold can be set, and device nodes whose contribution exceeds the preset threshold can be selected based on the calculated feature contribution. These device nodes are considered to be nodes that contribute significantly to the abnormal situation, and their node identifiers are used as abnormal device identifiers.

[0064] Step S145: Perform pattern matching processing on the state parameter subsequence of the device node corresponding to the abnormal device identifier, and take the matched preset abnormal pattern type as the abnormal type.

[0065] Step S1451: Standardize the state parameter subsequence of the device node corresponding to the abnormal device identifier, and extract the statistical feature set of the standardized state parameter subsequence. The statistical feature set includes mean, variance, maximum value, minimum value and rate of change.

[0066] For the device nodes corresponding to the identified abnormal device identifiers, their state parameter subsequences are first standardized. Standardization makes the state parameters of different devices comparable and eliminates the influence of dimensions. Then, a set of statistical features of the standardized state parameter subsequences is extracted, including mean, variance, maximum value, minimum value, and rate of change. These statistical features reflect the distribution and changes of the device state parameters.

[0067] Step S1452: Extract an abnormal pattern template set corresponding to the device type from a preset abnormal pattern database. The abnormal pattern template set contains statistical feature threshold ranges corresponding to different abnormal types.

[0068] Extract a set of anomaly pattern templates corresponding to the anomaly device type from a pre-defined anomaly pattern database. This database stores the statistical characteristic threshold ranges for various devices under different anomaly conditions. Each anomaly pattern template corresponds to an anomaly type and includes the statistical characteristic threshold ranges for the device status parameters under that anomaly type.

[0069] Step S1453: Perform a matching degree calculation on the statistical feature set and the statistical feature threshold range of each abnormal pattern template. The matching degree is obtained by calculating the proportion of statistical feature values ​​falling within the threshold range.

[0070] When calculating the matching degree between the extracted statistical feature set and the statistical feature threshold range of each abnormal pattern template, the specific operation can be to check whether each statistical feature value in the statistical feature set falls within the statistical feature threshold range corresponding to the abnormal pattern template. For the mean feature in the statistical feature set, check whether its value is within the threshold range of the mean in the abnormal pattern template; for the variance feature, determine whether its value is within the corresponding variance threshold range; for features such as maximum value, minimum value, and rate of change, the same judgment is performed. Then, calculate the proportion of statistical feature values ​​falling within the threshold range as the matching degree.

[0071] For example, a statistical feature set contains multiple statistical features. Upon inspection, it is found that some of these statistical feature values ​​fall within the threshold range of a certain abnormal pattern template. By calculating the ratio of the number of statistical features falling within this range to the total number of statistical features in the statistical feature set, the matching degree between the abnormal pattern template and the statistical feature set is obtained. By performing the above matching degree calculation on all abnormal pattern templates, the matching degree between each abnormal pattern template and the statistical feature set is obtained.

[0072] Step S1454: Select the anomaly type corresponding to the anomaly pattern template with the highest matching degree as the anomaly type of the anomaly device.

[0073] After obtaining the matching degree between each anomaly pattern template and the statistical feature set, these matching degrees are compared. The anomaly pattern template with the highest matching degree is identified, and the anomaly type corresponding to this anomaly pattern template is determined as the anomaly type of the abnormal device, thus completing the identification of the anomaly type of the abnormal device.

[0074] Step S146: Associate and bind the abnormal device identifier and the abnormal type to generate an identification result containing the abnormal device identifier and the abnormal type.

[0075] After identifying the abnormal equipment identifier and the abnormality type, they are associated and bound together. A pre-defined association and binding method can be used, such as establishing a correspondence to match the abnormal equipment identifier and the abnormality type one-to-one. This generates an identification result containing the abnormal equipment identifier and the abnormality type, clearly indicating which equipment in the thermal power unit has malfunctioned and the specific type of malfunction.

[0076] Step S150: Generate an operation and maintenance optimization strategy for the thermal power unit based on the identification results, and feed the operation and maintenance optimization strategy back to the thermal power unit operation and maintenance management system to trigger maintenance operations.

[0077] After obtaining the identification results, which include the abnormal equipment identifier and the abnormality type, it is necessary to generate an operation and maintenance optimization strategy for the thermal power unit based on the identification results, and then feed this optimization strategy back to the thermal power unit operation and maintenance management system to trigger corresponding maintenance operations. The specific steps are as follows: Step S151: Parse the abnormal device identifier and abnormal type in the identification result, and extract the device type information and installation location information corresponding to the abnormal device.

[0078] First, the identification results are analyzed to separate the abnormal equipment identifier and the abnormality type. Then, based on the abnormal equipment identifier, the corresponding equipment type information and installation location information are extracted from the thermal power unit's equipment information database. The equipment type information helps determine the equipment's function and characteristics, while the installation location information helps to quickly locate the abnormal equipment during maintenance operations.

[0079] Step S152: Query the set of basic maintenance operation instructions associated with the device type, anomaly type and installation location information from the preset operation and maintenance strategy rule base.

[0080] The pre-defined operation and maintenance policy rule base stores basic maintenance operation instructions for various devices under different anomaly types and installation locations. Based on the extracted device type, anomaly type, and installation location information, a query is performed in the operation and maintenance policy rule base. By matching the device type, anomaly type, and installation location information, the associated set of basic maintenance operation instructions is found. This set of basic maintenance operation instructions contains a series of basic maintenance operation instructions for the device with the anomaly.

[0081] Step S153: Prioritize the set of basic maintenance operation instructions to obtain a priority ranking result. The priority ranking result is determined by evaluating the impact of different maintenance operations on the overall operational stability of the thermal power unit.

[0082] After obtaining the basic maintenance operation instruction set, these instructions need to be prioritized. Assessing the impact of different maintenance operations on the overall operational stability of the thermal power unit is crucial for determining priorities. Maintenance operations with a significant impact on the overall operational stability of the thermal power unit are given higher priorities, while those with a smaller impact are given relatively lower priorities.

[0083] When assessing the degree of impact, multiple factors can be considered. For example, some maintenance operations may directly affect the power generation efficiency of a thermal power unit. If these operations are not performed in a timely manner, the power generation efficiency may drop significantly, affecting the overall economic benefits and power supply stability of the thermal power unit. In such cases, the priority of these maintenance operations should be higher. Conversely, some maintenance operations, while not directly affecting power generation efficiency, may have a significant impact on the lifespan of the equipment. If these operations are not performed, the equipment may fail prematurely, increasing maintenance costs and downtime. The priority of these maintenance operations also needs to be appropriately increased. By comprehensively considering the above factors, the priority of each maintenance operation instruction in the basic maintenance operation instruction set is determined, and then the basic maintenance operation instruction set is sorted from highest to lowest priority.

[0084] Step S154: Extract the state parameter subsequence of the abnormal device in the real-time operation data set before the abnormality occurs, analyze the changing trend of the state parameter subsequence, and generate abnormal development prediction information.

[0085] A subsequence of state parameters of the malfunctioning device prior to the occurrence of the malfunction was extracted from the real-time operational data set. This subsequence records the changes in the operating state of the malfunctioning device over a period of time before the malfunction occurred. The subsequence was analyzed to observe the trends of each state parameter over time.

[0086] Various analytical methods can be used to analyze the changing trends of state parameter subsequences. For example, by plotting curves showing the state parameters changing over time, one can visually observe whether the parameters are gradually increasing, gradually decreasing, or fluctuating. The rate of change of state parameters can also be calculated to understand the speed of their change. By analyzing these trends and rates of change, the development of anomalies can be predicted. For instance, if a state parameter shows a continuous upward trend with an accelerating rate of increase, it can be predicted that the anomaly may worsen further; if the trend of the state parameter tends to stabilize, the anomaly may not develop significantly in the short term. Based on the above analytical results, anomaly development prediction information is generated.

[0087] Step S155: Adjust the time dimension of the basic maintenance operation instruction set according to the abnormal development prediction information to generate dynamic maintenance operation instructions containing operation time windows.

[0088] Step S1551: Perform time series prediction processing on the state parameter subsequence of the abnormal device before the abnormality occurs to generate a state parameter prediction sequence for a future preset time period.

[0089] When performing time series forecasting on the state parameter subsequence of an abnormal device prior to the occurrence of the abnormality, an appropriate time series forecasting method should be adopted. For example, a forecasting model based on historical data can be used, such as analyzing past change patterns of the state parameter subsequence and using these patterns to predict future state parameter values.

[0090] When making predictions, the characteristics and variation patterns of the state parameter subsequences must be considered. If the state parameter subsequences exhibit periodic changes, future state parameters can be predicted based on this periodicity. If the changes in the state parameter subsequences are influenced by multiple factors, it may be necessary to consider these factors comprehensively for a more accurate prediction. Through the above time series prediction processing, a state parameter prediction sequence for a preset future time period is generated. This state parameter prediction sequence illustrates the possible changes in the state parameters of the abnormal device over a future period.

[0091] Step S1552: Calculate the anomaly severity change curve over time based on the state parameter prediction sequence, wherein the anomaly severity is determined by the degree of deviation of the state parameter value from the normal range.

[0092] Based on the generated state parameter prediction sequence, the curve of anomaly severity over time is calculated. First, the normal range of the equipment state parameters must be determined, which can be based on the design standards and historical operating data of the thermal power unit. Then, for each state parameter value in the state parameter prediction sequence, its deviation from the normal range is calculated.

[0093] For example, for a certain state parameter, if its normal range is an interval, the distance between the state parameter value and the upper and lower limits of the interval is calculated, and the degree of deviation is determined based on the magnitude of the distance. If the state parameter value exceeds the normal range significantly, it indicates a large deviation and a high degree of anomaly severity; if the state parameter value is close to the normal range, the deviation is small and the anomaly severity is low. By performing the above calculation on the state parameter value at each time point in the state parameter prediction sequence, the anomaly severity at different time points is obtained. Connecting these anomaly severity values ​​in chronological order yields a curve showing the change in anomaly severity over time.

[0094] Step S1553: Identify the key time points on the abnormality severity change curve, including the time points when the abnormality reaches the mild impact threshold, the time points when it reaches the moderate impact threshold, and the time points when it reaches the severe impact threshold.

[0095] After obtaining the curve showing the change in the severity of the anomaly over time, the curve is analyzed to identify key time points. Pre-set thresholds for mild, moderate, and severe impacts are established, representing the boundaries at which the severity of the anomaly affects the thermal power unit to varying degrees.

[0096] On the anomaly severity change curve, find the time point when the anomaly severity reaches the mild impact threshold. This time point marks the beginning of the anomaly's impact on the thermal power unit. Find the time point when the anomaly severity reaches the moderate impact threshold. At this point, the impact of the anomaly on the thermal power unit further increases. Find the time point when the anomaly severity reaches the severe impact threshold. This means that the anomaly has already caused serious impact on the thermal power unit. These time points are the critical time nodes.

[0097] Step S1554: Assign a corresponding impact mitigation capability value to each maintenance operation instruction in the basic maintenance operation instruction set, wherein the impact mitigation capability value represents the effect of the maintenance operation on reducing the severity of the anomaly.

[0098] For each maintenance operation instruction in the basic maintenance operation instruction set, a corresponding impact mitigation capability value needs to be assigned to it. This impact mitigation capability value reflects the effectiveness of the maintenance operation in reducing the severity of the anomaly.

[0099] Impact mitigation capability values ​​can be determined by analyzing and evaluating historical maintenance data. This involves examining the reduction in anomaly severity after each past maintenance operation. If a maintenance operation significantly reduces anomaly severity, it indicates a good mitigation effect, and its impact mitigation capability value is high; conversely, if the reduction is not significant, its impact mitigation capability value is low. Based on this analysis and evaluation, an appropriate impact mitigation capability value can be assigned to each maintenance operation instruction.

[0100] Step S1555: Based on the key time nodes and the impact mitigation capability value, calculate the optimal execution time window for each maintenance operation instruction. The optimal execution time window is from the start time point when the severity of the anomaly reaches the point where the maintenance operation can effectively mitigate it to the end time point when the severity of the anomaly exceeds the maximum mitigation capability of the maintenance operation.

[0101] Based on the identified critical time points and the impact mitigation capability value assigned to each maintenance operation instruction, the optimal execution time window for each maintenance operation instruction is calculated. For each maintenance operation instruction, the range within which the maintenance operation can effectively mitigate the severity of the anomaly is determined based on its impact mitigation capability value.

[0102] By combining key time points on the anomaly severity change curve, we find the starting time point when the anomaly severity reaches a level that the maintenance operation can effectively mitigate; this time point is the start time of the optimal execution time window. Simultaneously, we find the ending time point when the anomaly severity exceeds the maximum mitigation capacity of the maintenance operation; this time point is the end time of the optimal execution time window. Through these calculations, the optimal execution time window for each maintenance operation instruction is obtained.

[0103] Step S1556: Associate and bind the optimal execution time window with the corresponding maintenance operation instruction to generate a dynamic maintenance operation instruction containing the operation time window.

[0104] After obtaining the optimal execution time window for each maintenance operation instruction, the optimal execution time window is associated and bound with the corresponding maintenance operation instruction. For example, a correspondence can be established to map each maintenance operation instruction to its optimal execution time window. This generates dynamic maintenance operation instructions that include operation time windows. These dynamic maintenance operation instructions not only specify the maintenance operation to be performed but also define the optimal execution time range for each maintenance operation.

[0105] Step S156: Combine the priority ranking results and the dynamic maintenance operation instructions to generate an operation and maintenance optimization strategy for the thermal power unit.

[0106] The system integrates a set of basic maintenance operation instructions prioritized by order, with dynamic maintenance operation instructions including operation time windows. During this integration, both the priority of maintenance operations and the operation time windows must be considered. Higher-priority maintenance operation instructions are scheduled for execution within appropriate time windows; lower-priority instructions are scheduled for execution based on remaining time resources and the operating status of the thermal power unit.

[0107] For example, a detailed operation and maintenance plan can be developed, listing maintenance operation instructions in priority order and clearly indicating the operation time window for each instruction. Through this integration, a comprehensive and reasonable operation and maintenance optimization strategy for thermal power units can be generated, taking into account both the importance of maintenance operations and the rationality of their execution time.

[0108] Step S157: After converting the operation and maintenance optimization strategy into an instruction format compatible with the thermal power unit operation and maintenance management system interface, send it to the operation and maintenance management system to trigger maintenance operations.

[0109] Finally, the generated operation and maintenance optimization strategy needs to be converted into a compatible format. The thermal power unit operation and maintenance management system has its own defined interface and command format requirements, which necessitate converting the operation and maintenance optimization strategy into a compatible command format.

[0110] By writing corresponding programs or using data conversion tools, information such as maintenance operation instructions and operation time windows in the operation and maintenance optimization strategy can be converted into a format that meets the interface requirements of the operation and maintenance management system. After conversion, the converted instructions are sent to the thermal power unit operation and maintenance management system. Upon receiving these instructions, the operation and maintenance management system can trigger corresponding maintenance operations based on the content of the instructions, thereby realizing intelligent operation and maintenance processing of the thermal power unit.

[0111] For example, in one possible implementation, the pre-trained graph neural network model is obtained through the following steps.

[0112] Step S211: Collect a sample operating data set of the thermal power unit, which includes a sequence of equipment status parameters during normal operation and abnormal operation.

[0113] To train the graph neural network model, the first step is to collect a sample operating data set of the thermal power unit. This sample operating data set should include the sequence of equipment state parameters during the normal operation and abnormal operation phases of the thermal power unit.

[0114] When collecting the sample operation data set, sensors are installed on various devices of the thermal power unit to continuously collect the status parameters of the devices. During the normal operation period of the thermal power unit, the status parameters of each device are recorded to form a sequence of device status parameters for the normal operation phase; similarly, during the period when the thermal power unit experiences an anomaly, the status parameters of the devices are recorded to form a sequence of device status parameters for the abnormal operation phase. These sequence of device status parameters for the normal operation phase and the abnormal operation phase are combined to form this operation data set.

[0115] Step S212: Perform graph structure construction processing on the sample running data set to generate a training device association graph set.

[0116] The process of constructing a graph structure for the collected sample running data set is similar to the process of constructing a graph structure for the real-time running data set mentioned earlier.

[0117] First, the state parameter sequences in the sample runtime dataset are partitioned by device dimension to obtain a subsequence of state parameters for each device. Then, a unique node identifier is assigned to each subsequence of state parameters, and these identifiers are bound and mapped to the device physical identifiers to generate a node set for the training device association graph. Next, a temporal correlation index is extracted between different device state parameter subsequences. Based on this index, the strength of the operational association between devices is determined. When the operational association strength exceeds a preset association threshold, an edge connection is established between the corresponding two device nodes, and each edge is assigned a weight value positively correlated with the operational association strength. Through these steps, a graph structure is constructed for each set of data in the sample runtime dataset, generating a training device association graph set.

[0118] Step S213: Assign a corresponding anomaly label to each training device association icon. The anomaly label includes an anomaly device identifier and anomaly type.

[0119] For each training device association graph in the generated set of training device association graphs, a corresponding anomaly label needs to be added. If the sample running data corresponding to the training device association graph was collected during the abnormal running phase, the abnormal device and the anomaly type are identified, and the identifier of the abnormal device and the anomaly type are used as anomaly labels to mark the training device association graph; if the sample running data corresponding to the training device association graph was collected during the normal running phase, a normal running label is added.

[0120] Step S214: Input the set of training device association graphs into the initial graph neural network model, and generate predicted association features through forward propagation.

[0121] The set of device association graphs used for training is input into the initial graph neural network model. The initial graph neural network model includes modules such as node feature encoding layer, edge feature encoding layer, message passing layer, global pooling layer, and feature enhancement layer.

[0122] In the model, the set of nodes in the device association graph used for training is input into the node feature encoding layer. Time-series feature extraction is performed on the state parameter subsequences of each node to generate local node features containing time dependencies. The set of edges is input into the edge feature encoding layer. Association feature extraction is performed on the edge weight value and corresponding operational association strength of each edge to generate local edge features reflecting the interaction patterns between devices. Then, the node local features and edge local features are aggregated using a message passing layer to generate node neighborhood aggregated features. Next, a global pooling layer integrates the neighborhood aggregated features of all nodes to generate global aggregated features containing information about the interaction effects between device states. Finally, the global aggregated features are input into the feature enhancement layer, where feature dimension expansion and nonlinear transformation are applied to generate predicted association features.

[0123] Step S215: Perform abnormal state prediction processing based on the predicted association features to generate predicted abnormal labels.

[0124] Based on the generated predicted association features, abnormal state prediction processing is performed in the model. The model uses the feature information of the predicted association features to determine whether the operating status of the thermal power unit corresponding to the training equipment association diagram is normal or abnormal. If it is determined to be abnormal, the abnormal equipment and abnormal type are further identified, and a predicted abnormal label is generated; if it is determined to be normal, a predicted label for normal operation is generated.

[0125] Step S216: Calculate the loss value between the predicted anomaly label and the labeled anomaly label. The loss value is obtained by calculating the cross-entropy loss function.

[0126] The generated predicted anomaly labels are compared with the anomaly labels previously assigned to the training device association icons, and the loss value between the two is calculated. The cross-entropy loss function is used to calculate the loss value, which measures the degree of difference between the predicted result and the true label. The loss value is obtained by calculating the cross-entropy between the predicted anomaly label and the labeled anomaly label.

[0127] Step S217: Use the backpropagation algorithm to adjust the network parameters of the initial graph neural network model according to the loss value until the loss value converges to a preset threshold, thus completing the training of the initial graph neural network model.

[0128] Based on the calculated loss value, the backpropagation algorithm is used to adjust the network parameters of the initial graph neural network model. The backpropagation algorithm calculates the gradient of each parameter in the network according to the magnitude and direction of the loss value, and then updates the parameters based on the gradient.

[0129] By continuously performing forward propagation, calculating the loss value, and adjusting the parameters through backpropagation, the loss value gradually decreases. This training process continues until the loss value converges to a preset threshold. When the loss value reaches the preset threshold, it indicates that the model can generate accurate predicted association features and predicted anomaly labels based on the input training device association graph. At this point, the initial graph neural network model training is complete, resulting in a pre-trained graph neural network model that can be used for feature learning processing of real-time device association graphs.

[0130] Figure 2 The illustration shows exemplary hardware and software components of an intelligent operation and maintenance processing system 100 for thermal power units, which can implement the ideas of this application, according to some embodiments of this application. For example, processor 120 can be used in the intelligent operation and maintenance processing system 100 for thermal power units and to perform the functions in this application.

[0131] The intelligent operation and maintenance system 100 for thermal power units can be a general-purpose server or a special-purpose server; both can be used to implement the intelligent operation and maintenance method for thermal power units described in this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0132] For example, the intelligent operation and maintenance processing system 100 for thermal power units may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the intelligent operation and maintenance processing system 100 for thermal power units may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The intelligent operation and maintenance processing system 100 for thermal power units also includes an I / O interface 150 between the computer and other input / output devices.

[0133] For ease of explanation, only one processor is described in the intelligent operation and maintenance processing system 100 for thermal power units. However, it should be noted that the intelligent operation and maintenance processing system 100 for thermal power units in this application may also include multiple processors. Therefore, the steps performed by one processor described in this application may also be performed jointly by multiple processors or individually. For example, if the processor of the intelligent operation and maintenance processing system 100 for thermal power units performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0134] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned intelligent operation and maintenance processing method for thermal power units is implemented.

[0135] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An intelligent operation and maintenance method for thermal power units, characterized in that, The method includes: Acquire a real-time operating data set of the thermal power unit, wherein the real-time operating data set contains a sequence of state parameters of multiple devices at consecutive time nodes; The real-time operation data set is processed to construct a graph structure to generate an equipment association graph. In the equipment association graph, nodes represent individual equipment of the thermal power unit, and edges represent the operation association relationships between equipment. The pre-trained graph neural network model is invoked to perform feature learning processing on the device association graph, generating global association features that reflect the interaction and influence between device states; Based on the global correlation features, the abnormal device status identification process is performed to generate an identification result containing the abnormal device identifier and the abnormality type. Based on the identification results, an operation and maintenance optimization strategy for the thermal power unit is generated, and the operation and maintenance optimization strategy is fed back to the thermal power unit operation and maintenance management system to trigger maintenance operations.

2. The intelligent operation and maintenance method for thermal power units according to claim 1, characterized in that, The step of performing graph structure construction processing on the real-time operating data set to generate a device association graph includes: The status parameter sequence in the real-time running data set is divided into device dimensions to obtain the status parameter subsequence corresponding to each device. A unique node identifier is assigned to each device's corresponding state parameter subsequence, and the node identifier is bound and mapped to the device's physical identifier to generate a set of nodes in the device association graph. Extract the temporal correlation index between different device status parameter subsequences in the real-time operation data set. The temporal correlation index is obtained by calculating the consistency of the changing trends of any two device status parameter subsequences at the same time node. The strength of the operational correlation between devices is determined based on the time-series correlation index. When the strength of the operational correlation exceeds a preset correlation threshold, an edge connection is established between the corresponding two device nodes. Assign an edge weight value to each edge that is positively correlated with the strength of the operational association relationship, and complete the construction process of the device association graph.

3. The intelligent operation and maintenance method for thermal power units according to claim 2, characterized in that, The extraction of time-series correlation indicators between different device status parameter subsequences in the real-time operating data set includes: Perform time alignment processing on any two device status parameter subsequences so that the two device status parameter subsequences contain the same number of time nodes; Calculate the matching degree of the change direction of two device status parameters at each time node. When the matching degree of the change direction is 1, it means that the two device status parameters rise or fall at the same time node. When it is 0, it means that the change direction is opposite. Calculate the rate of difference in the change of two device status parameters at each time point, where the rate of difference in change is the ratio of the absolute change of the two device status parameters at that time point. The matching degree of the change direction and the difference rate of the change magnitude are standardized and weighted to generate a time-series correlation score for each time node; The mean value of the time series correlation scores at all time points is calculated to generate a time series correlation index that reflects the consistency of the overall change trend of the two device status parameter subsequences.

4. The intelligent operation and maintenance method for thermal power units according to claim 1, characterized in that, The step of calling a pre-trained graph neural network model to perform feature learning processing on the device association graph, generating global association features reflecting the interaction and influence between device states, includes: The set of nodes in the device association graph is input into the node feature encoding layer of the graph neural network model, and time series feature extraction processing is performed on the state parameter subsequence of each node to generate local node features containing time dependencies. The set of edges of the device association graph is input into the edge feature encoding layer of the graph neural network model. The edge weight value of each edge and the corresponding operational association strength are processed for association feature extraction to generate local edge features that reflect the interaction mode between devices. Through the message passing layer of the graph neural network model, neighborhood information aggregation processing is performed based on the local features of the nodes and the local features of the edges. The local features of the neighboring nodes and the local features of the corresponding edges of each node are weighted and fused to generate node neighborhood aggregation features. The global pooling layer of the graph neural network model is used to integrate the neighborhood aggregation features of all nodes to generate global aggregation features that include information on the interaction and influence between device states. The global aggregated features are input into the feature enhancement layer of the graph neural network model, and through feature dimension expansion and nonlinear transformation processing, global correlation features reflecting the interaction and influence between device states are generated.

5. The intelligent operation and maintenance method for thermal power units according to claim 4, characterized in that, The message passing layer of the graph neural network model performs neighborhood information aggregation processing based on the local features of nodes and the local features of edges. It weights and fuses the local features of adjacent nodes and the corresponding local features of edges for each node to generate node neighborhood aggregation features, including: For each node, traverse all its neighboring nodes and extract the local features of the current node, the local features of its neighboring nodes, and the local features of the edges connecting them. The feature similarity calculation module generates a similarity score that reflects the degree of association between the local features of the current node, the local features of adjacent nodes, and the local features of edges. The local features of the neighboring nodes and the local features of the edges are weighted based on the similarity scores to generate weighted features of the adjacency information. The weighted features of all adjacency information of the current node are fused to generate the initial features of the current node's neighborhood aggregation. The initial features of the neighborhood aggregation are concatenated with the local features of the current node to generate a node neighborhood aggregation feature that includes the node's own information and neighborhood interaction information.

6. The intelligent operation and maintenance method for thermal power units according to claim 1, characterized in that, The process of identifying abnormal device states based on the global correlation features generates an identification result containing an abnormal device identifier and an abnormality type, including: The global correlation features are input into a pre-trained anomaly detection classifier, which generates anomaly confidence by calculating the feature distance between the global correlation features and the normal state feature template. When the anomaly confidence level exceeds the preset anomaly threshold, the anomaly localization process is triggered. In the anomaly localization process, the feature contribution of each device node in the global correlation features is extracted, and the feature contribution represents the degree of influence of the device node on the anomaly confidence. Based on the feature contribution, device nodes whose contribution exceeds a preset contribution threshold are selected, and the node identifier of the device node is used as an abnormal device identifier. The status parameter subsequence of the device node corresponding to the abnormal device identifier is subjected to pattern matching processing, and the matched preset abnormal pattern type is taken as the abnormal type; The abnormal device identifier and the abnormal type are associated and bound together to generate an identification result containing the abnormal device identifier and the abnormal type.

7. The intelligent operation and maintenance method for thermal power units according to claim 6, characterized in that, The step of performing pattern matching processing on the state parameter subsequence of the device node corresponding to the abnormal device identifier, and taking the matched preset abnormal pattern type as the abnormal type, includes: The status parameter subsequence of the device node corresponding to the abnormal device identifier is standardized, and the statistical feature set of the standardized status parameter subsequence is extracted. The statistical feature set includes mean, variance, maximum value, minimum value and rate of change. Extract an anomaly pattern template set corresponding to the device type from a preset anomaly pattern database. The anomaly pattern template set contains statistical feature threshold ranges corresponding to different anomaly types. The matching degree is calculated by comparing the statistical feature set with the statistical feature threshold range of each abnormal pattern template. The matching degree is obtained by calculating the proportion of statistical feature values ​​falling within the threshold range. The anomaly type corresponding to the anomaly pattern template with the highest matching degree is selected as the anomaly type of the anomaly device.

8. The intelligent operation and maintenance method for thermal power units according to claim 1, characterized in that, The step of generating an operation and maintenance optimization strategy for the thermal power unit based on the identification result, and feeding the operation and maintenance optimization strategy back to the thermal power unit operation and maintenance management system to trigger maintenance operations, includes: The abnormal device identifier and abnormal type in the identification result are analyzed, and the device type information and installation location information corresponding to the abnormal device are extracted. Query the set of basic maintenance operation instructions associated with the equipment type information, anomaly type and installation location information from the preset operation and maintenance strategy rule base; The set of basic maintenance operation instructions is sorted by priority to obtain a priority sorting result. The priority sorting result is determined by evaluating the degree of impact of different maintenance operations on the overall operational stability of the thermal power unit. Extract the state parameter subsequence of the abnormal device in the real-time operation data set before the abnormality occurs, analyze the changing trend of the state parameter subsequence, and generate abnormal development prediction information; Based on the abnormal development prediction information, the basic maintenance operation instruction set is adjusted in terms of time dimension to generate dynamic maintenance operation instructions containing operation time windows; By integrating the priority ranking results and the dynamic maintenance operation instructions, an operation and maintenance optimization strategy for the thermal power unit is generated. The operation and maintenance optimization strategy is converted into an instruction format compatible with the thermal power unit operation and maintenance management system interface and then sent to the operation and maintenance management system to trigger maintenance operations.

9. The intelligent operation and maintenance method for thermal power units according to claim 8, characterized in that, The step of adjusting the basic maintenance operation instruction set according to the abnormal development prediction information in terms of time dimension to generate dynamic maintenance operation instructions containing operation time windows includes: Perform time series prediction processing on the subsequence of state parameters of the abnormal device before the occurrence of the abnormality to generate a predicted sequence of state parameters for a future preset time period; The anomaly severity variation curve over time is calculated based on the state parameter prediction sequence, wherein the anomaly severity is determined by the degree of deviation of the state parameter values ​​from the normal range; Identify key time points on the anomaly severity change curve, including the time points when the anomaly severity reaches the mild impact threshold, the time points when it reaches the moderate impact threshold, and the time points when it reaches the severe impact threshold; Assign a corresponding impact mitigation capability value to each maintenance operation instruction in the basic maintenance operation instruction set, wherein the impact mitigation capability value represents the effect of the maintenance operation on reducing the severity of the anomaly; Based on the key time nodes and the impact mitigation capability value, calculate the optimal execution time window for each maintenance operation instruction. The optimal execution time window is from the start time point when the severity of the anomaly reaches the point where the maintenance operation can effectively mitigate it to the end time point when the severity of the anomaly exceeds the maximum mitigation capability of the maintenance operation. The optimal execution time window is associated and bound with the corresponding maintenance operation instruction to generate a dynamic maintenance operation instruction containing the operation time window.

10. An intelligent operation and maintenance system for thermal power units, characterized in that, The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the intelligent operation and maintenance method for thermal power units as described in any one of claims 1-9.

Images