Intelligent power plant fault early warning method and system applying artificial intelligence

Abstract

The application provides a smart power plant fault early warning method and system applying artificial intelligence, which comprehensively reveals the internal relationship among the generator set, the sensor network and the control system by constructing a multi-dimensional associated knowledge graph, can accurately locate the key link and potential risk point in the operation of the power plant by identifying the key cluster of the composite knowledge network, can not only discover the known fault mode in time, but also can early warn the new and unknown fault mode by deeply analyzing the key cluster members and matching with the historical fault mode library, thereby greatly improving the intelligent level of the power plant fault early warning. The risk level division based on the fault mode and the fault symptom makes the early warning information more refined and hierarchical, and facilitates the power plant managers to develop reasonable countermeasures according to the risk level, and ensures the safe and efficient operation of the power plant. Therefore, the accuracy and intelligent level of the power plant fault early warning are significantly improved.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of power operation and maintenance platform, in particular to a smart power plant fault early warning method and system applying artificial intelligence. BACKGROUND

[0002] With the continuous progress of science and technology, smart power plants have become an important development direction of the power industry. Smart power plants integrate advanced information technology and control technology to achieve efficient, safe and environmentally friendly operation of power plants. However, with the increasing complexity of power plant equipment and the increasing amount of operation data, how to effectively monitor the operation status of power plant equipment and timely discover and warn potential faults has become a problem to be solved in the development of smart power plants.

[0003] Traditional power plant fault early warning methods mainly rely on periodic equipment inspection, manual monitoring and simple threshold alarms. These methods can discover equipment faults to some extent, but have many limitations. For example, periodic inspection may not be able to discover sudden faults in time; manual monitoring is limited by the experience and skills of personnel and is easily affected by subjective factors; and simple threshold alarms may produce false alarms or missed alarms due to dynamic changes in equipment operation. SUMMARY

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the embodiment of the present application is to provide a smart power plant fault early warning method and system applying artificial intelligence, which comprehensively analyzes the operation status and potential faults of power plant equipment by constructing a multi-dimensional associated knowledge graph, accurately locates potential risk points by identifying key clusters in a complex knowledge network, determines fault patterns and symptoms by matching with a historical fault pattern library, and finally generates early warning information according to the risk level, thereby not only improving the accuracy and intelligent level of fault early warning, but also providing a strong guarantee for the safe operation of power plants.

[0005] In a first aspect, the embodiment of the present application provides a smart power plant fault early warning method applying artificial intelligence, which comprises:

[0006] analyzing the operation log data of the smart power plant by using an artificial intelligence network model to obtain a unit associated knowledge graph between each generator unit, a sensor associated knowledge graph between each sensor network, a control system associated knowledge graph between each control system, a first associated knowledge graph between each sensor network and each generator unit, and a second associated knowledge graph between each sensor network and each control system;

[0007] identifying key clusters in a complex knowledge network formed by the unit associated knowledge graph, the sensor associated knowledge graph, the control system associated knowledge graph, the first associated knowledge graph and the second associated knowledge graph;

[0008] analyzing the identified key cluster, determining key cluster members of the key cluster in the generator set knowledge hierarchy, and determining participation relationship features of the key cluster members in the generator set;

[0009] based on the participation relationship features of the key cluster members in the generator set, matching with a historical fault mode library to determine fault modes and fault symptoms existing in the key cluster members, the historical fault mode library being a fault feature mode library between key components, sensor networks and control systems of the generator set established according to historical fault data and prior fault features;

[0010] based on the fault modes and fault symptoms existing in the key cluster members, dividing the key cluster members into different risk levels, and generating corresponding early warning information.

[0011] In a possible implementation of the first aspect, the identification of the key cluster from the composite knowledge network formed by the unit associated knowledge graph, the sensor associated knowledge graph, the control system associated knowledge graph, the first associated knowledge graph and the second associated knowledge graph comprises:

[0012] constructing a knowledge member selection vector of a knowledge hierarchy x; wherein x = 0, 1, 2, the knowledge member selection vector of the knowledge hierarchy x is a description content formed by a plurality of knowledge member indication features, each knowledge member indication feature being used to reflect whether a knowledge member corresponding to the knowledge hierarchy x is selected as a key cluster member in the key cluster;

[0013] generating a target model according to the knowledge member selection vectors of the composite knowledge network in each knowledge hierarchy, the relationship array M1 corresponding to the unit associated knowledge graph, the relationship array M2 corresponding to the sensor associated knowledge graph, the relationship array M3 corresponding to the control system associated knowledge graph, the relationship array M4 corresponding to the first associated knowledge graph and the relationship array M5 corresponding to the second associated knowledge graph;

[0014] analyzing the knowledge member selection vectors of the target model in each knowledge hierarchy to identify the key cluster.

[0015] In a possible implementation of the first aspect, the generating of the target model according to the knowledge member selection vectors of the composite knowledge network in each knowledge hierarchy, the relationship array M1 corresponding to the unit associated knowledge graph, the relationship array M2 corresponding to the sensor associated knowledge graph, the relationship array M3 corresponding to the control system associated knowledge graph, the relationship array M4 corresponding to the first associated knowledge graph and the relationship array M5 corresponding to the second associated knowledge graph comprises:

[0016] generating an intra-knowledge-level model according to the knowledge member selection vector of each knowledge level of the composite knowledge network, the relation array M1, the relation array M2 and the relation array M3;

[0017] generating an inter-knowledge-level model according to the knowledge member selection vector of each knowledge level of the composite knowledge network, the relation array M4 and the relation array M5;

[0018] performing a minimization strategy after adding the intra-knowledge-level model and the inter-knowledge-level model to generate the target model.

[0019] In a possible implementation of the first aspect, the generating an intra-knowledge-level model according to the knowledge member selection vector of each knowledge level of the composite knowledge network, the relation array M1, the relation array M2 and the relation array M3 comprises:

[0020] generating array feature data F1 of the knowledge level x according to a first weight vector corresponding to the knowledge level x, a base array and a relation array corresponding to the knowledge level x; wherein the relation array corresponding to the knowledge level 0 is the relation array M1, the relation array corresponding to the knowledge level 1 is the relation array M2, the relation array corresponding to the knowledge level 2 is the relation array M3, the first weight vector is a vector with all elements being 1, and the base array is a vector array with only diagonal elements being 1 and other elements being 0;

[0021] generating array feature data F2 of the knowledge level x according to the array feature data F1 of the knowledge level x, a first preset model parameter and the knowledge member selection vector of the knowledge level x;

[0022] generating array feature data F3 of the knowledge level x according to the knowledge member selection vector of the knowledge level x and the relation array corresponding to the knowledge level x;

[0023] generating array feature data F4 of the knowledge level x according to the array feature data F2 and the array feature data F3 of the knowledge level x;

[0024] generating the intra-knowledge-level model according to the array feature data F4 of each knowledge level of the composite knowledge network.

[0025] In a possible implementation of the first aspect, the generating array feature data F1 of the knowledge level x according to a first weight vector corresponding to the knowledge level x, a base array and a relation array corresponding to the knowledge level x comprises:

[0026] determining a fusion calculation result of the first weight vector corresponding to the knowledge level x and an inverse array of the first weight vector, to generate array feature data F5 of the knowledge level x;

[0027] determining an array difference between the array feature data F5 of the knowledge level x and a base array of the knowledge level x, and an array difference between the relationship array corresponding to the knowledge level x, to generate the array feature data F1 of the knowledge level x;

[0028] The generating of the array feature data F2 of the knowledge level x according to the array feature data F1 of the knowledge level x, the first preset model parameter and the knowledge member selection vector of the knowledge level x comprises:

[0029] determining a fusion calculation result of an inverse array of the knowledge member selection vector of the knowledge level x and the array feature data F1, and a fusion calculation result of the knowledge member selection vector of the knowledge level x, to generate the array feature data F6 of the knowledge level x;

[0030] determining a fusion calculation result of the first preset model parameter and the array feature data F6 of the knowledge level x, to generate the array feature data F2 of the knowledge level x;

[0031] The generating of the array feature data F3 of the knowledge level x according to the knowledge member selection vector of the knowledge level x and the relationship array corresponding to the knowledge level x comprises:

[0032] determining a fusion calculation result of an inverse array of the knowledge member selection vector of the knowledge level x and the relationship array of the knowledge level x, and a fusion calculation result of the knowledge member selection vector of the knowledge level x, to generate the array feature data F3 of the knowledge level x;

[0033] The generating of the array feature data F4 of the knowledge level x according to the array feature data F2 and the array feature data F3 of the knowledge level x comprises:

[0034] determining a difference between the array feature data F2 of the knowledge level x and the array feature data F3 of the knowledge level x, to generate the array feature data F4 of the knowledge level x.

[0035] In a possible implementation of the first aspect, the generating of the intra-knowledge-level model according to the array feature data F4 of each knowledge level of the composite knowledge network comprises:

[0036] summing the array feature data F4 of each knowledge level of the composite knowledge network to generate the intra-knowledge-level model.

[0037] In one possible implementation of the first aspect, generating an inter-knowledge-level model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M4, and the relation array M5 includes:

[0038] Based on the knowledge member selection vector of knowledge level x, the knowledge member selection vector of knowledge level y, the first weight vector corresponding to knowledge level x, and the first weight vector of knowledge level y, array feature data F7 between knowledge level x and knowledge level y is generated; where y = 0, 1, 2, y ≠ x;

[0039] Based on the knowledge member selection vector of knowledge level x and the knowledge member selection vector of knowledge level y, generate array feature data F8 between knowledge level x and knowledge level y;

[0040] Based on the relation array between knowledge level x and knowledge level y, the second preset model parameters of the relation array between knowledge level x and knowledge level y, the array feature data F7 and array feature data F8 between knowledge level x and knowledge level y, the array feature data F9 between knowledge level x and knowledge level y is generated; wherein, when x=0 and y=1, the relation array between knowledge level 0 and knowledge level 1 is the relation array M4, when x=1 and y=2, the relation array between knowledge level 1 and knowledge level 2 is the relation array M5, and when x=0 and y=2, the relation array between knowledge level 0 and knowledge level 2 is defaulted to an array where all element vector values ​​are 0;

[0041] Based on the array feature data F9 between every two knowledge levels of the composite knowledge network, the inter-knowledge level model is generated.

[0042] In one possible implementation of the first aspect, generating array feature data F7 between knowledge level x and knowledge level y based on the knowledge member selection vector of knowledge level x, the knowledge member selection vector of knowledge level y, the first weight vector corresponding to knowledge level x, and the first weight vector of knowledge level y includes:

[0043] The difference between the first weight vector corresponding to the knowledge level x and the knowledge member selection vector of the knowledge level x is determined, and the array feature data F10 of the knowledge level x is generated.

[0044] The difference between the first weight vector corresponding to the knowledge level y and the knowledge member selection vector of the knowledge level y is determined, and the array feature data F11 of the knowledge level y is generated.

[0045] The fusion calculation result of the array feature data F10 of knowledge level x and the reverse array feature data F11 of knowledge level y is determined, and the array feature data F7 between knowledge level x and knowledge level y is generated.

[0046] The step of generating array feature data F8 between knowledge level x and knowledge level y based on the knowledge member selection vector of knowledge level x and the knowledge member selection vector of knowledge level y includes:

[0047] The fusion calculation result of the knowledge member selection vector of knowledge level x and the reverse array of knowledge member selection vector of knowledge level y is determined, and array feature data F8 between knowledge level x and knowledge level y is generated.

[0048] The step of generating array feature data F9 between knowledge level x and knowledge level y based on the corresponding relationship array between knowledge level x and knowledge level y, the second preset model parameters of the corresponding relationship array between knowledge level x and knowledge level y, array feature data F7 between knowledge level x and knowledge level y, and array feature data F8 includes:

[0049] The difference between the relation array corresponding to knowledge level x and knowledge level y and the array feature data F8 between knowledge level x and knowledge level y is determined, and then the difference between the array feature data F7 between knowledge level x and knowledge level y is used to generate the array feature data F12 between knowledge level x and knowledge level y.

[0050] The element-wise product of the relation array corresponding to the knowledge level x and the knowledge level y and the array feature data F12 between the knowledge level x and the knowledge level y is determined to generate the array feature data F13 between the knowledge level x and the knowledge level y.

[0051] Determine the sum of squares of each element in the array of array feature data F13 between knowledge level x and knowledge level y;

[0052] The sum of squares of each element in the array is determined as the result of the fusion calculation of the second preset model parameters between the knowledge level x and the knowledge level y, and array feature data F9 between the knowledge level x and the knowledge level y is generated.

[0053] The step of generating the inter-knowledge-level model based on the array feature data F9 between every two knowledge levels of the composite knowledge network includes:

[0054] The array feature data F9 between every two knowledge levels of the composite knowledge network are summed to generate the inter-knowledge level model.

[0055] In one possible implementation of the first aspect, parsing the knowledge member selection vectors of the target model at each knowledge level to identify the key clusters includes:

[0056] The knowledge member selection vectors of the target model at each knowledge level are softened, and the target model is transformed to generate a transformed target model.

[0057] The knowledge member selection sequences of all knowledge levels of the composite knowledge network are concatenated to generate a concatenated sequence.

[0058] The transformation target model is applied to the cascaded sequence using a stepwise decreasing method to generate array feature data F14;

[0059] The key cluster is identified based on the array feature data F14;

[0060] Wherein, the array feature data F14 is a description content with the same dimension as the concatenated sequence, and the array members in the array feature data F14 correspond one-to-one with the knowledge member indicator features in the concatenated sequence;

[0061] The step of identifying the key cluster based on the array feature data F14 includes:

[0062] For any array member in the array feature data F14, if the array member is greater than a threshold value, then the knowledge member corresponding to the array member is determined to be selected and loaded into the critical cluster; if the array member is not greater than the threshold value, then the knowledge member corresponding to the array member is determined not to be selected and loaded into the critical cluster.

[0063] Secondly, embodiments of this application also provide a smart power plant fault early warning system applying artificial intelligence. The smart power plant fault early warning system applying artificial intelligence includes a processor and a machine-readable storage medium. The machine-readable storage medium stores machine-executable instructions, which are loaded and executed by the processor to implement the smart power plant fault early warning method applying artificial intelligence in any possible implementation of the first aspect.

[0064] In all aspects described above, by constructing a multi-dimensional relational knowledge graph, the intrinsic connections between generator units, sensor networks, and control systems are comprehensively revealed, providing a rich data foundation and in-depth analytical perspective for subsequent fault early warning. By identifying key clusters within the composite knowledge network, critical links and potential risk points in power plant operation can be accurately located, improving the targeting and accuracy of fault early warning. Through in-depth analysis of key cluster members and matching with a historical fault mode database, not only can known fault modes be discovered in a timely manner, but also new and unknown fault modes can be warned, thus greatly enhancing the intelligence level of power plant fault early warning. Risk level classification based on fault modes and symptoms makes early warning information more refined and hierarchical, facilitating power plant managers to formulate reasonable response measures according to risk levels, ensuring the safe and efficient operation of the power plant. Therefore, the accuracy and intelligence level of power plant fault early warning are significantly improved. Attached Figure Description

[0065] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be extracted in conjunction with these drawings without creative effort.

[0066] Figure 1 A flowchart illustrating the intelligent power plant fault early warning method using artificial intelligence provided in an embodiment of the present invention;

[0067] Figure 2 This is a schematic block diagram of a smart power plant fault early warning system using artificial intelligence, provided in an embodiment of the present invention, for implementing the above-described smart power plant fault early warning method using artificial intelligence. Detailed Implementation

[0068] The following description is provided to enable those skilled in the art to practice and incorporate the invention, and is given in a specific application scenario and with specific requirements. It will be apparent to those skilled in the art that various modifications can be made to the disclosed embodiments, and that the general principles defined herein can be applied to other embodiments and application scenarios without departing from the principles and scope of the invention. Therefore, the invention is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0069] Figure 1 This is a flowchart illustrating a smart power plant fault early warning method using artificial intelligence, provided in one embodiment of the present invention. The following is a detailed description of this smart power plant fault early warning method using artificial intelligence.

[0070] Step S110: Analyze the smart power plant operation log data using an artificial intelligence network model to obtain the knowledge graph of unit association between generator units, the knowledge graph of sensor association between sensor networks, the knowledge graph of control system association between control systems, the first knowledge graph of association between sensor networks and generator units, and the second knowledge graph of association between sensor networks and control systems.

[0071] In detail, the aforementioned artificial intelligence network model typically refers to a data analysis model built using artificial intelligence technology, capable of learning and recognizing patterns in data. Specifically, this artificial intelligence network model can be used to extract useful information and knowledge from large amounts of operational log data. For example, a deep learning model (such as a convolutional neural network (CNN) or a recurrent neural network (RNN)) can be trained to recognize specific patterns in the operational log data of a smart power plant, which may represent a certain behavior or state of the generator sets, sensors, or control systems.

[0072] The smart power plant operation log data refers to various data records generated by the power plant in its daily operation, including but not limited to the operating status of generator sets, sensor operating data, and control system operation records. It can exist in the form of a time series, recording the behavior and performance of various components of the power plant over time.

[0073] A knowledge graph is a graphical data structure used to represent relationships between entities. The unit association knowledge graph specifically refers to a knowledge graph depicting the relationships between various generator units, showing how they influence and depend on each other. For example, in a power plant, if there are multiple generator units, they may be connected to the power grid to jointly supply power. The unit association knowledge graph can show which units operate in parallel, which units serve as backups under specific circumstances, and other related information.

[0074] The sensor association knowledge graph refers to a knowledge graph that describes the relationships between various sensors in a power plant. Sensor networks are a key component of power plant monitoring systems, and the relationships between them reflect the interrelationships between different parts and systems within the power plant. For example, a pressure sensor and a temperature sensor may simultaneously monitor the status of a boiler. When the pressure increases, the temperature may also increase accordingly; this relationship can be represented in the sensor association knowledge graph.

[0075] The control system association knowledge graph refers to a knowledge graph that describes the relationships between various control systems in a power plant. Control systems are responsible for monitoring and regulating various operating parameters of the power plant to ensure its safe and efficient operation. For example, one control system might be responsible for regulating fuel supply, while another might be responsible for monitoring emissions; these two systems are interconnected because changes in fuel supply directly affect emissions.

[0076] The first association knowledge graph refers to a knowledge graph describing the relationships between sensor networks and generator sets, showing how sensors monitor the operating status of generator sets and how the status of generator sets affects sensor readings. For example, a sensor network monitoring the temperature of a generator set directly reflects the operating temperature of the generator set. When the generator set overheats, the sensor readings will rise, and this relationship is reflected in the first association knowledge graph.

[0077] The second association knowledge graph describes the relationships between sensor networks and control systems, showing how sensors provide data to the control system and how the control system adjusts the power plant's operation based on this data. For example, a control system might adjust the cooling system's operation based on data from a temperature sensor network to ensure the generator set doesn't overheat; this adjustment relationship is reflected in the second association knowledge graph.

[0078] In this embodiment, the server receives operation log data from the smart power plant. This log data includes the operating status of the generator sets, the operation data of the sensor network, and the operation records of the control system. The server uses a pre-trained artificial intelligence network model, such as a deep learning model, to analyze this log data. First, it identifies the relationships between the generator sets, such as which sets operate in parallel, which operate in series, and the energy flow relationships between them, thereby constructing a generator set relationship knowledge graph. Next, it analyzes the data in the sensor network to identify the spatial and temporal relationships between sensors, constructing a sensor relationship knowledge graph. Similarly, it analyzes the operation records in the control system to identify the collaborative relationships between different control systems, constructing a control system relationship knowledge graph. Finally, it analyzes the data interaction relationships between the sensor network and the generator sets, and between the sensor network and the control system, constructing a first relationship knowledge graph and a second relationship knowledge graph, respectively.

[0079] Step S120: Identify key clusters in the composite knowledge network formed by the unit-related knowledge graph, the sensor-related knowledge graph, the control system-related knowledge graph, the first related knowledge graph, and the second related knowledge graph.

[0080] In detail, the composite knowledge network refers to a more complex network structure formed by integrating multiple knowledge graphs, showcasing the comprehensive relationships between various components and systems of a power plant. For example, in a composite knowledge network, you can see how the state of a generator unit affects the readings of connected sensors, and how these readings are used by the control system to adjust the operation of the power plant.

[0081] A critical cluster refers to a group of interconnected nodes (such as generator sets, sensors, or control systems) identified within a complex knowledge network that are crucial to the operation of a power plant. The state and behavior of these nodes have a significant impact on the overall performance and stability of the power plant. For example, critical cluster identification may reveal that a particular generator set, along with its associated sensors and control systems, forms a critical cluster. If this generator set malfunctions, abnormal readings from its associated sensors will occur, affecting the control system's decisions and ultimately leading to instability in the power plant's operation. Therefore, this generator set and its associated sensors and control systems constitute a critical cluster requiring special attention and maintenance.

[0082] In other words, in this embodiment, the server integrates the five knowledge graphs constructed in step S110 into a composite knowledge network, which includes the relationships between the generator set, sensor network, and control system. To identify key clusters within this composite knowledge network, the server employs a graph theory-based algorithm. First, it constructs a selection vector for each knowledge level (generator set, sensor network, control system), indicating whether each member in that knowledge level is selected as part of a key cluster. Then, based on the structure of the composite knowledge network and the various selection vectors, the server constructs a target model that reflects the relationships and importance between members at different knowledge levels.

[0083] Step S130: Analyze the identified key clusters, determine the key cluster members of the key clusters in the generator set knowledge level, and determine the participation relationship characteristics of the key cluster members in the generator set.

[0084] In detail, the critical cluster members refer to specific components such as generator sets, sensors, or control systems within the identified critical clusters. These critical cluster members are the basic units constituting the critical clusters, and their performance and status are crucial to the operation of the entire cluster and even the power plant. For example, in a critical cluster consisting of specific generator sets, related temperature and pressure sensors, and control systems, these specific generator sets, sensors, and control systems are considered critical cluster members.

[0085] The participation relationship characteristics refer to the features of the interaction and dependence relationships between key cluster members during the operation of generator sets. These characteristics may include the data flow transmission method, the response speed of control signals, and the collaborative working mode among members. For example, in a key cluster, generator sets may rely on specific sensors to provide real-time operating data, while the control system adjusts the operating state of the generator sets based on this data. This data transmission, dependence, and control relationship constitutes the participation relationship characteristics.

[0086] In other words, in this embodiment, the server performs an in-depth analysis of the key clusters identified in step S120. It first determines which members of the key clusters belong to the generator set knowledge level, i.e., the key cluster members. Then, the server analyzes the specific roles of these key cluster members in the generator set, such as whether they are core components responsible for energy conversion or key sensors monitoring important parameters. By analyzing the participation relationship characteristics of these key cluster members, the server can more accurately understand their importance in the operation of the generator set.

[0087] Step S140: Based on the participation relationship characteristics of the key cluster members in the generator set, match them with the historical fault mode library to determine the fault modes and fault symptoms of the key cluster members. The historical fault mode library is a fault feature mode library between key components of the generator set, sensor network and control system established based on historical fault data and prior fault characteristics.

[0088] In detail, a failure mode refers to the typical manifestation or behavioral pattern when a generator set, sensor, or control system malfunctions. Failure symptoms, on the other hand, are the specific manifestations of these failure modes in actual operation, such as abnormal readings, performance degradation, or system errors. For example, a common failure mode could be sensor malfunction, with symptoms including unstable readings, deviations from normal ranges, or discrepancies with actual conditions.

[0089] In this embodiment, the server matches the participation relationship characteristics of the key cluster members determined in step S130 with a historical fault mode database. This database is constructed based on past fault data and expert experience and includes various fault modes and symptoms that may occur between key components of the generator set, sensor networks, and control systems. Through matching, the server can determine the possible fault modes and existing fault symptoms of the current key cluster members.

[0090] Step S150: Based on the fault modes and fault symptoms of the key cluster members, the key cluster members are divided into different risk levels and corresponding early warning information is generated.

[0091] In detail, risk levels are categorized based on the severity of the failure modes and symptoms present in critical cluster members. Different risk levels reflect the varying degrees of impact that member failures may have on power plant operations. For example, a slight deviation in sensor readings might be classified as low-risk, while a failure that directly leads to generator shutdown might be classified as high-risk.

[0092] The early warning information is generated based on the risk level of key cluster members and is used to inform operations and maintenance personnel in advance of potential risks and problems so that they can take appropriate preventive and response measures. For example, when the reading of a key sensor begins to deviate from the normal range, an early warning message can be generated to prompt operations and maintenance personnel to check the sensor and take appropriate measures to prevent potential failures.

[0093] In other words, in this embodiment, based on the fault modes and symptoms determined in step S140, the server performs a risk assessment on critical cluster members. For example, critical cluster members are classified into different risk levels according to the severity and frequency of the fault. For instance, if certain critical sensors malfunction, it may cause the entire generator unit to shut down, and therefore they are classified as high-risk. Finally, the server generates corresponding early warning information based on the risk assessment results. This early warning information includes the probability of fault occurrence, potential impact, and suggested preventive measures. This early warning information will be promptly sent to power plant management personnel so that they can take appropriate actions to ensure the safe and stable operation of the power plant.

[0094] Based on the above steps, by constructing a multi-dimensional interconnected knowledge graph, the intrinsic relationships between generator units, sensor networks, and control systems are comprehensively revealed, providing a rich data foundation and in-depth analytical perspective for subsequent fault early warning. By identifying key clusters in the composite knowledge network, critical links and potential risk points in power plant operation can be accurately located, improving the targeting and accuracy of fault early warning. In-depth analysis of key cluster members, combined with matching with a historical fault mode database, not only can known fault modes be discovered in a timely manner, but also new and unknown fault modes can be warned, thus greatly improving the intelligence level of power plant fault early warning. Risk level classification based on fault modes and symptoms makes early warning information more refined and hierarchical, facilitating power plant managers to formulate reasonable response measures according to risk levels, ensuring the safe and efficient operation of the power plant. Therefore, the accuracy and intelligence level of power plant fault early warning are significantly improved.

[0095] In one possible implementation, step S120 includes:

[0096] Step S121: Construct a knowledge member selection vector for knowledge level x. Where x = 0, 1, 2, the knowledge member selection vector for knowledge level x is a description formed by multiple knowledge member indicator features. Each knowledge member indicator feature reflects whether the knowledge member corresponding to knowledge level x has been selected as a key cluster member in the key cluster.

[0097] In this embodiment, the server first constructs selection vectors for different knowledge levels. These knowledge levels reflect the abstraction levels of different components in the composite knowledge network, where x=0 represents the generator set level, x=1 represents the sensor network level, and x=2 represents the control system level.

[0098] Generator set level (x=0):

[0099] The server traverses the knowledge graph of generator set relationships to identify all generator set nodes.

[0100] For each generator set, the server creates a knowledge member indicator feature, which is a binary value indicating whether the generator set has been selected as a critical cluster member. For example, if the server initially determines that a generator set is in a core position in the power grid, its corresponding indicator feature may be set to 1 (selected); otherwise, it is 0 (not selected). The indicator features of all generator sets are combined to form the knowledge member selection vector at the generator set level.

[0101] Sensor network hierarchy (x=1):

[0102] Similarly, the server identifies all sensor nodes through a sensor association knowledge graph.

[0103] For each sensor, the server assigns a knowledge member indicator feature based on factors such as its importance and historical failure rate, indicating whether it has been selected as part of a critical cluster.

[0104] The indication features of all sensors are combined to form a knowledge member selection vector for the sensor network hierarchy.

[0105] Control system hierarchy (x=2):

[0106] The server identifies all control system nodes by associating them with a knowledge graph of control systems. For each control system, considering its criticality and complexity, a knowledge member indicator feature is assigned. The indicator features of all control systems constitute the knowledge member selection vector at the control system hierarchy level.

[0107] Step S122: Generate a target model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1 corresponding to the unit-related knowledge graph, the relation array M2 corresponding to the sensor-related knowledge graph, the relation array M3 corresponding to the control system-related knowledge graph, the relation array M4 corresponding to the first-related knowledge graph, and the relation array M5 corresponding to the second-related knowledge graph.

[0108] In this embodiment, the server will then use the previously constructed knowledge member selection vector and the relation arrays corresponding to each knowledge graph to generate the target model.

[0109] The server first retrieves the relationship array M1 from the generator set association knowledge graph, which describes the connections and dependencies between generator sets. Next, it retrieves the relationship array M2 from the sensor association knowledge graph, reflecting the data interaction and correlation between sensors. Then, it retrieves the relationship array M3 from the control system association knowledge graph, revealing the logical and control relationships between control systems. Finally, it retrieves the relationship array M4 from the first association knowledge graph, demonstrating the monitoring and control relationships between the sensor network and the generator sets. Finally, it retrieves the relationship array M5 from the second association knowledge graph, illustrating the information transmission and feedback relationships between the sensor network and the control system.

[0110] The server combines these relational arrays with selection vectors at various knowledge levels, and through complex mathematical operations and model building techniques (such as matrix operations and graph theory analysis), generates a comprehensive target model. This target model not only includes the direct relationships between generator sets, sensors, and control systems, but also incorporates their indirect influences and dependencies, forming a complex network model that comprehensively reflects the operating status of the power plant.

[0111] Step S123: Analyze the knowledge member selection vectors of the target model at each knowledge level to identify the key clusters.

[0112] In this embodiment, the server now performs deep analysis on the generated target model to identify key clusters in the power plant operation.

[0113] The server first analyzes the selection vectors at each knowledge level to identify generator sets, sensors, and control systems that are marked as critical (i.e., those with an indicator feature of 1).

[0114] By analyzing the connections and interactions among these key knowledge members in the target model, the server can identify those closely connected and interdependent clusters, namely critical clusters. These critical clusters are the parts that need special attention and optimization in the operation of the power plant, because their status and performance are directly related to the stability and efficiency of the entire power plant.

[0115] Through the above steps, the server can accurately identify the critical clusters in the power plant operation, providing an important basis for subsequent fault prevention, performance optimization, and other tasks.

[0116] In one possible implementation, step S122 includes:

[0117] Step S1221: Based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1, the relation array M2, and the relation array M3, generate a model within the knowledge level.

[0118] In this embodiment, the server first constructs a knowledge hierarchy model, which mainly focuses on the relationships and interactions within the respective levels of the generator set, sensors, and control system.

[0119] In detail, the server has extracted knowledge member selection vectors for each knowledge level from the composite knowledge network. These vectors reflect which generator sets, sensors, and control systems are considered critical. Simultaneously, the server has also acquired the relationship arrays M1, M2, and M3 of the generator set association knowledge graph, sensor association knowledge graph, and control system association knowledge graph. These arrays detail the connections and dependencies between elements within each knowledge level.

[0120] For the generator set level, the server uses the relation array M1 of the generator set association knowledge graph and the corresponding knowledge member selection vector to quantify the mutual influence and correlation between generator sets through mathematical operations (such as matrix multiplication, weighted summation, etc.), thereby constructing a model within the generator set level.

[0121] Similarly, the server uses the relation array M2 of the sensor association knowledge graph and the relation array M3 of the control system association knowledge graph, and combines them with the knowledge member selection vectors of their respective levels to construct the sensor-level intra-level model and the control system-level intra-level model.

[0122] During the construction process, the server may use some optimization algorithms (such as gradient descent, least squares, etc.) to adjust the model parameters to ensure that the model can more accurately reflect the relationships and dynamics within the actual hierarchy.

[0123] Step S1222: Based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M4, and the relation array M5, generate a knowledge level model.

[0124] Next, the server will build a knowledge hierarchy model, which focuses on the interactions and influences between different knowledge levels (such as generator sets, sensors, and control systems).

[0125] The server utilizes relation array M4 corresponding to the first association knowledge graph and relation array M5 corresponding to the second association knowledge graph. These two arrays detail the cross-level relationships between the generator set, sensors, and control system. Simultaneously, the server also considers knowledge member selection vectors at each knowledge level to determine which cross-level interactions are critical.

[0126] The server can combine relation arrays M4 and M5 with knowledge member selection vectors at each knowledge level through techniques such as tensor decomposition and network analysis. This allows the server to quantify the mutual influence and dependencies between different levels, enabling it to construct a comprehensive inter-knowledge level model that reveals how generator sets, sensors, and control systems interact and how these interactions affect the overall operation of the power plant.

[0127] Step S1223: After summing the models within the knowledge hierarchy and the models between knowledge hierarchies, a minimization strategy is implemented to generate the target model.

[0128] Finally, the server will sum the models within and between knowledge levels and implement a minimization strategy to generate the final target model.

[0129] In detail, the server first adds the models within each knowledge level (generator set, sensor, control system) with the models between knowledge levels to form a comprehensive initial model.

[0130] To optimize the performance and accuracy of the model, the server can implement a minimization strategy, such as least squares or gradient descent optimization, to adjust the model's parameters and weights. This process aims to ensure that the model can better fit the actual data and accurately reflect the complex relationships and dynamic behaviors between the various components in the power plant.

[0131] After optimization, the server obtained the final target model, which not only includes the relationships within the generator set, sensor, and control system hierarchy, but also considers the interactions and dependencies between these knowledge levels. The target model provides powerful analytical tools for subsequent fault prediction, performance optimization, and decision support.

[0132] In one possible implementation, generating a knowledge-level intra-model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1, the relation array M2, and the relation array M3 includes:

[0133] Step S1221-1: Based on the first weight vector, the basic array, and the relation array corresponding to knowledge level x of the composite knowledge network, generate array feature data F1 for knowledge level x. Wherein, the relation array corresponding to knowledge level 0 is relation array M1, the relation array corresponding to knowledge level 1 is relation array M2, the relation array corresponding to knowledge level 2 is relation array M3, the first weight vector is a vector where all elements are 1, and the basic array is a vector array where only the diagonal elements are 1 and the remaining elements are zero.

[0134] In this embodiment, the server begins processing data from the composite knowledge network at each knowledge level. For knowledge level x, the server first obtains the corresponding first weight vector, which is a vector where all elements are 1, representing that all elements within this knowledge level have the same initial weight. Next, the server obtains the basic array, which is characterized by having only diagonal elements that are 1 and the remaining elements that are 0. This basic array represents the basic relationship structure of each element within the level.

[0135] Subsequently, the server selects the corresponding relation array based on the knowledge level number: if x is 0, relation array M1 is selected, which represents the internal association of the generator set level; if x is 1, relation array M2 is selected, which reflects the relationship of the sensor network level; if x is 2, relation array M3 is selected, which reveals the internal connection of the control system level.

[0136] With this data, the server begins to calculate the array feature data F1. This is done by performing specific mathematical operations (such as element-wise multiplication or matrix multiplication) on the first weight vector and the base array, and then further operations (such as matrix addition or multiplication) on the selected relation array. The purpose of this series of operations is to integrate the basic structure within the hierarchy and the relationships between elements, ultimately generating the array feature data F1 for knowledge level x.

[0137] Step S1221-2: Based on the array feature data F1 of the knowledge level x, the first preset model parameters and the knowledge member selection vector of the knowledge level x, generate the array feature data F2 of the knowledge level x.

[0138] After obtaining the array feature data F1, the server then processes the knowledge member selection vector at knowledge level x, which is a vector identifying which elements have been selected as key cluster members. The server combines this selection vector with first preset model parameters, which can be pre-set based on historical data or expert knowledge, to adjust the model's focus on different elements.

[0139] The server then uses this numerical value, which combines the selection vector and preset parameters, to further transform the array feature data F1. This may involve complex mathematical operations, such as matrix transformations or element-wise nonlinear operations. Through these operations, the server obtains the array feature data F2 for knowledge level x, which not only contains the basic structure and relationships within the level but also incorporates the selection information of key cluster members.

[0140] Step S1221-3: Based on the knowledge member selection vector of the knowledge level x and the relation array corresponding to the knowledge level x, generate the array feature data F3 of the knowledge level x.

[0141] To generate array feature data F3, the server again considers the knowledge member selection vector of knowledge level x. This time, it directly performs operations on this selection vector with the corresponding relation array. The operations can be matrix multiplication, element-wise multiplication, or other forms of mathematical operations, aiming to reveal the special status and influence of the selected key cluster members in the hierarchical relationships.

[0142] After the calculation process, the server obtained the array feature data F3 of knowledge level x, which reflects the specific role and importance of key cluster members in the hierarchical relationship.

[0143] Step S1221-4: Based on the array feature data F2 and array feature data F3 of the knowledge level x, generate the array feature data F4 of the knowledge level x.

[0144] After generating array feature data F2 and F3, the server performs a final calculation to obtain array feature data F4. For example, F2 and F3 are combined through specific mathematical operations, such as element-wise addition, multiplication, or other more complex operations. The purpose of this step is to integrate the information in F2 and F3 to comprehensively reflect the integrated features and relationships within knowledge level x.

[0145] The final generated array feature data F4 is a highly comprehensive representation that integrates the basic structure, relationships, and selection information of key cluster members within the hierarchy.

[0146] Step S1221-5: Generate the model within the knowledge level based on the array feature data F4 of each knowledge level of the composite knowledge network.

[0147] After the server completes the calculation of the array feature data F4 for each knowledge level, it begins to build the model within each knowledge level. First, the array feature data F4 for all levels is summarized and integrated. This may require preprocessing steps such as data normalization or feature dimensionality reduction to improve the stability and efficiency of the model.

[0148] Next, the server uses this integrated feature data to build a statistical model that captures the deep relationships and patterns within each knowledge level, providing a powerful tool for subsequent tasks such as decision support and performance optimization. Through this process, the server successfully generates a knowledge-level intra-level model, which can deeply understand and predict the behavior and trends of the composite knowledge network within each knowledge level.

[0149] In one possible implementation, step S1221-1 includes:

[0150] Step S1221-11: Determine the fusion calculation result of the first weight vector and the reverse array of the first weight vector corresponding to the knowledge level x, and generate the array feature data F5 of the knowledge level x.

[0151] Step S1221-12: Determine the array difference between the array feature data F5 of the knowledge level x and the basic array of the knowledge level x, and then the array difference between the basic array of the knowledge level x and the relation array corresponding to the knowledge level x, to generate the array feature data F1 of the knowledge level x.

[0152] Step S1221-2 includes:

[0153] Steps S1221-21: Determine the fusion calculation result of the reverse array of the knowledge member selection vector of the knowledge level x and the array feature data F1, and then combine it with the fusion calculation result of the knowledge member selection vector of the knowledge level x to generate the array feature data F6 of the knowledge level x.

[0154] Steps S1221-22: Determine the fusion calculation result of the first preset model parameters and the array feature data F6 of the knowledge level x, and generate the array feature data F2 of the knowledge level x.

[0155] Step S1221-3 includes:

[0156] Steps S1221-31: Determine the fusion calculation result of the reverse array of the knowledge member selection vector of the knowledge level x and the relationship array of the knowledge level x, and then combine it with the fusion calculation result of the knowledge member selection vector of the knowledge level x to generate the array feature data F3 of the knowledge level x.

[0157] Step S1221-4 includes:

[0158] Steps S1221-41: Determine the difference between the array feature data F2 and the array feature data F3 of the knowledge level x, and generate the array feature data F4 of the knowledge level x.

[0159] In this embodiment, the server first begins to process the relevant data of knowledge level x, which has its specific first weight vector, basic array, and relation array, all of which are important components of the composite knowledge network.

[0160] The server first determines the first weight vector corresponding to knowledge level x, which reflects the importance of different elements in that level. Next, the server calculates the inverse array of this first weight vector, which typically involves transforming each element of the first weight vector using a specific algorithm. After this step, the server fuses the original first weight vector with its inverse array, combining information from both arrays through weighted averaging, dot product, or other mathematical operations. Finally, the result of this fusion calculation is defined as the array feature data F5 of knowledge level x.

[0161] Next, the server can retrieve the base array for knowledge level x, which represents the basic structure or initial state of that level. Then, the server calculates the array difference between the array feature data F5 and the base array, which typically involves calculating the differences between corresponding elements. Similarly, the server uses the relation array corresponding to knowledge level x, which describes the relationships or interactions between elements within the level. The server then calculates the array difference between this relation array and the previously obtained array difference. Finally, this result of the double array difference is defined as the array feature data F1 for knowledge level x.

[0162] Before proceeding to the next step, the server can consider the knowledge member selection vector for knowledge level x, which indicates which knowledge members are active or important in the current level. The server first computes the inverse array of this selection vector and fuses it with the array feature data F1. Then, this result is combined with the direct fusion result of the knowledge member selection vector to generate the array feature data F6.

[0163] Subsequently, the server can utilize first preset model parameters, which may be set based on previous experience or training data, to adjust or optimize the feature data. The server then fuses the first preset model parameters with the array feature data F6 to generate array feature data F2 for knowledge level x.

[0164] In this step, the server again utilizes the knowledge member selection vector of knowledge level x, but this time it fuses its inverse array with the relation array. This calculation reveals how the relations within the level change when a specific knowledge member is selected. The server then combines this result with the direct fusion calculation result of the knowledge member selection vector to generate array feature data F3.

[0165] Finally, the server generates array feature data F4 by calculating the difference between array feature data F2 and F3. This difference can represent the net change in hierarchical features after considering the selection of knowledge members and changes in relationships within the hierarchy. This array feature data F4 can provide important information for subsequent tasks such as knowledge reasoning, decision support, or pattern recognition.

[0166] In one possible implementation, generating the model within the knowledge level based on the array feature data F4 of each knowledge level of the composite knowledge network includes:

[0167] The array feature data F4 of each knowledge level of the composite knowledge network are summed to generate the model within the knowledge level.

[0168] In this embodiment, the server begins to process the array feature data F4 of each knowledge level in the composite knowledge network. In order to construct a unified intra-level model, the server needs to sum these array feature data F4.

[0169] Specifically, the server traverses each knowledge level in the composite knowledge network and obtains the array feature data F4 for each knowledge level. Then, the server performs a summation operation on these array feature data F4, adding up the array feature data F4 of all levels together. This process can be seen as fusing the internal features of each knowledge level to form a holistic knowledge level intra-model. This knowledge level intra-model can capture the common features and patterns of each knowledge level in the composite knowledge network.

[0170] In one possible implementation, step S1222 includes:

[0171] Step S1222-1: Based on the knowledge member selection vector of knowledge level x, the knowledge member selection vector of knowledge level y, the first weight vector corresponding to knowledge level x, and the first weight vector of knowledge level y, generate array feature data F7 between knowledge level x and knowledge level y. Where y = 0, 1, 2, y ≠ x.

[0172] Step S1222-2: Based on the knowledge member selection vector of knowledge level x and the knowledge member selection vector of knowledge level y, generate array feature data F8 between knowledge level x and knowledge level y.

[0173] Step S1222-3: Based on the relation array corresponding to the knowledge level x and knowledge level y, the second preset model parameters of the relation array corresponding to the knowledge level x and knowledge level y, the array feature data F7 and array feature data F8 between the knowledge level x and knowledge level y, generate the array feature data F9 between the knowledge level x and knowledge level y; wherein, when x=0 and y=1, the relation array between knowledge level 0 and knowledge level 1 is the relation array M4, when x=1 and y=2, the relation array between knowledge level 1 and knowledge level 2 is the relation array M5, and when x=0 and y=2, the relation array between knowledge level 0 and knowledge level 2 is defaulted to an array where all element vector values ​​are 0.

[0174] Step S1222-4: Generate the inter-knowledge level model based on the array feature data F9 between every two knowledge levels of the composite knowledge network.

[0175] When generating models between knowledge levels, the server needs to consider the relationships and mutual influences between different knowledge levels, a process involving multiple steps and complex calculations.

[0176] First, the server selects two different knowledge levels in the composite knowledge network, denoted as knowledge level x and knowledge level y. For these two knowledge levels, the server needs to obtain their knowledge member selection vectors, first weight vectors, and corresponding relation arrays.

[0177] Next, the server uses this data to generate array feature data F7 between knowledge level x and knowledge level y. This step may involve complex operations on selection vectors and weight vectors, as well as considering the array of relationships between the two knowledge levels. The F7 data reflects the interaction between the two knowledge levels in terms of knowledge member selection and weight allocation.

[0178] Then, the server generates array feature data F8 based on the knowledge member selection vectors of knowledge level x and knowledge level y. This data mainly focuses on the similarity or difference between the two knowledge levels in terms of knowledge member selection.

[0179] Next, the server uses the relation array, the second preset model parameters, and array feature data F7 and F8 to generate array feature data F9. In this step, the server may need to perform complex mathematical operations and data processing to capture the deep relationships and mutual influences between the two knowledge levels. Specifically, when x=0 and y=1, the server can use relation array M4 for the relationship between knowledge level 0 and knowledge level 1; when x=1 and y=2, the server can use relation array M5 for the relationship between knowledge level 1 and knowledge level 2; and when x=0 and y=2, since there is no direct connection between knowledge level 0 and knowledge level 2, a default array with all element vector values ​​of 0 is used to represent their relationship.

[0180] Finally, the server generates an inter-knowledge level model based on the array feature data F9 between every two knowledge levels in the composite knowledge network. This inter-knowledge level model can comprehensively reflect the complex relationships and interactions between different knowledge levels, providing strong support for subsequent knowledge reasoning, decision support and other tasks.

[0181] In one possible implementation, step S1222-1 includes:

[0182] The difference between the first weight vector corresponding to the knowledge level x and the knowledge member selection vector of the knowledge level x is determined, and the array feature data F10 of the knowledge level x is generated.

[0183] The difference between the first weight vector corresponding to the knowledge level y and the knowledge member selection vector of the knowledge level y is determined, and the array feature data F11 of the knowledge level y is generated.

[0184] The fusion calculation result of the array feature data F10 of knowledge level x and the reverse array feature data F11 of knowledge level y is determined, and the array feature data F7 between knowledge level x and knowledge level y is generated.

[0185] In this embodiment, the server first obtains the first weight vector and the knowledge member selection vector corresponding to knowledge level x. The first weight vector reflects the importance of each element in this level, while the knowledge member selection vector indicates which knowledge members are active or important. The server calculates the difference between these two vectors, i.e., subtracts the corresponding elements, to obtain the array feature data F10 of knowledge level x. This data reflects the difference between weight allocation and knowledge member selection in knowledge level x.

[0186] Next, the server processes knowledge level y in the same way. The server obtains the first weight vector and knowledge member selection vector corresponding to knowledge level y, calculates their difference, and generates array feature data F11 for knowledge level y, which captures the degree of mismatch between weight allocation and knowledge member selection in knowledge level y.

[0187] Finally, to generate the array feature data F7 between knowledge level x and knowledge level y, the server performs a fusion calculation on the inverse array of the array feature data F10 of knowledge level x and the array feature data F11 of knowledge level y. The inverse array is obtained by performing a specific transformation on each element in the original array, and the fusion calculation may involve mathematical operations such as weighted averaging and dot product. The calculation result F7 integrates the differences in weight allocation and knowledge member selection between knowledge level x and knowledge level y, providing an important basis for subsequent analysis of the relationship between the two knowledge levels.

[0188] Step S1222-2 includes:

[0189] The fusion calculation result of the knowledge member selection vector of knowledge level x and the reverse array of knowledge member selection vector of knowledge level y is determined, and array feature data F8 between knowledge level x and knowledge level y is generated.

[0190] When generating array feature data F8, the server focuses on the relationship between knowledge level x and knowledge level y in the selection of knowledge members.

[0191] The server first obtains the knowledge member selection vectors for knowledge level x and knowledge level y. Then, the server calculates the fusion result of the inverse arrays of the knowledge member selection vectors for knowledge level x and knowledge level y. This calculation may involve complex mathematical operations, such as dot product and weighted average, aiming to capture the similarity or difference between the two knowledge levels in terms of knowledge member selection.

[0192] The final generated array feature data F8 reflects the relationship between knowledge level x and knowledge level y in the selection of knowledge members. This data is of great significance for understanding the knowledge flow, sharing and complementarity between the two knowledge levels, and can provide valuable information for subsequent knowledge management, decision support and other tasks.

[0193] Step S1222-3 includes:

[0194] The difference between the relation array corresponding to knowledge level x and knowledge level y and the array feature data F8 between knowledge level x and knowledge level y is determined, and then the difference between the array feature data F7 between knowledge level x and knowledge level y is used to generate the array feature data F12 between knowledge level x and knowledge level y.

[0195] The element-wise product of the relation array corresponding to knowledge level x and knowledge level y and the array feature data F12 between knowledge level x and knowledge level y is determined to generate the array feature data F13 between knowledge level x and knowledge level y.

[0196] Determine the sum of squares of each element in the array of array feature data F13 between knowledge level x and knowledge level y.

[0197] The sum of squares of each element in the array is determined as the result of the fusion calculation of the second preset model parameters between the knowledge level x and the knowledge level y, and array feature data F9 between the knowledge level x and the knowledge level y is generated.

[0198] In this embodiment, the server begins processing the relationship between knowledge level x and knowledge level y to generate array feature data F9. First, the server needs to determine the difference between the corresponding relationship array between knowledge level x and knowledge level y and the array feature data F8. This reflects the difference between the actual relationship between the two knowledge levels and the knowledge member selection vector relationship. Next, the server calculates the difference between this difference and the array feature data F7 (which reflects the combined difference in weight allocation and knowledge member selection), obtaining array feature data F12. This data F12 captures the combined differences between the two knowledge levels in terms of relationship, weight allocation, and knowledge member selection.

[0199] Next, the server performs element-level product operations, that is, it calculates the product of the relation array and each corresponding element of the array feature data F12 to generate the array feature data F13. This step further strengthens the differences between hierarchical relationships and takes into account the interaction between actual relationships and various differences.

[0200] The server then calculates the sum of squares for each element in the array feature data F13. This operation aggregates the effects of all elements into a single value, reflecting the overall differences and complex relationships within the entire array.

[0201] Finally, the server fuses this sum of squares with a second pre-defined model parameter between knowledge level x and knowledge level y. This model parameter can be a constant used to adjust or standardize the results, or a weight value set based on prior experience. The result of the fusion calculation generates array feature data F9, which integrates all the important features and differences in the relationships between levels, providing crucial information for subsequent analysis and model building.

[0202] Step S1222-4 includes:

[0203] The array feature data F9 between every two knowledge levels of the composite knowledge network are summed to generate the inter-knowledge level model.

[0204] When generating inter-knowledge hierarchy models, the server aims to integrate the array feature data F9 between all knowledge hierarchy pairs.

[0205] The server first traverses the array feature data F9 between every two knowledge levels in the composite knowledge network. This data captures the complex relationships and differences between different level pairs.

[0206] Next, the server performs a summation operation on these array feature data F9. This process integrates the relationship features between all hierarchical pairs into a unified model, forming a comprehensive knowledge hierarchy model. This model can reflect the interaction and influence between different levels in the composite knowledge network, providing a solid foundation for subsequent knowledge reasoning, prediction, and decision support.

[0207] In one possible implementation, step S123 may include:

[0208] Step S1231: Soften the knowledge member selection vectors of the target model at each knowledge level, and transform the target model to generate a transformed target model.

[0209] Step S1232: Concatenate the knowledge member selection sequences of all knowledge levels of the composite knowledge network to generate a concatenated sequence.

[0210] Step S1233: Apply a stepwise decreasing method to the transformation target model with respect to the cascaded sequence to generate array feature data F14.

[0211] Step S1234: Identify the key cluster based on the array feature data F14.

[0212] The array feature data F14 is a description with the same dimension as the concatenated sequence, and the array members in the array feature data F14 correspond one-to-one with the knowledge member indicator features in the concatenated sequence.

[0213] Step S1234 includes: for any array member in the array feature data F14, if the array member is greater than a threshold value, then it is determined that the knowledge member corresponding to the array member is selected and loaded into the critical cluster. If the array member is not greater than the threshold value, then it is determined that the knowledge member corresponding to the array member is not selected and loaded into the critical cluster.

[0214] In this embodiment, the server first obtains the knowledge member selection vectors of the target model at each knowledge level. These vectors reflect which knowledge members are selected or considered important at each knowledge level. For more in-depth analysis, the server needs to perform a softening operation on these vectors. Softening operations may include smoothing, normalization, or applying some form of weight adjustment to reduce noise in the data and highlight important features.

[0215] After the softening operation is completed, the server transforms the target model. This transformation may involve adjusting model parameters, recalculating weights, or applying other mathematical transformations to better suit subsequent analysis steps. The transformed model is called the transformed target model.

[0216] Next, the server needs to process the knowledge member selection sequences across all knowledge levels of the composite knowledge network. These sequences represent the knowledge members selected at different levels. The server concatenates these sequences into a continuous sequence, called a concatenated sequence. This process is similar to merging multiple lists into a long list, where each list represents a selection at a knowledge level.

[0217] The server now uses a progressive reduction method to process the target model and the cascaded sequence. The progressive reduction method may involve a series of complex mathematical operations aimed at extracting features from the model that correspond to the cascaded sequence. This process generates a new data array called the array feature data F14, which has the same dimension as the cascaded sequence, and each member of the array corresponds to a knowledge member indicative feature in the cascaded sequence.

[0218] Finally, the server uses the array feature data F14 to identify critical clusters. The server can iterate through each array member in F14 and compare it with a preset threshold. If the value of an array member is greater than the threshold, then its corresponding knowledge member is considered critical and selected to be loaded into the critical cluster. Conversely, if the value of an array member is not greater than the threshold, then the corresponding knowledge member will not be added to the critical cluster.

[0219] This process will eventually generate a critical cluster containing all key knowledge members, which may be of great significance for subsequent tasks such as knowledge analysis, decision making, or resource optimization.

[0220] Figure 2 The following is a schematic diagram of the hardware structure of a smart power plant fault early warning system 100 based on artificial intelligence, provided by an embodiment of the present invention, for implementing the above-described smart power plant fault early warning method based on artificial intelligence. Figure 2 As shown, the intelligent power plant fault early warning system 100 applying artificial intelligence may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0221] In one exemplary design concept, the AI-based smart power plant fault early warning system 100 can be a single AI-based smart power plant fault early warning system or a server group. The server group can be centralized or distributed (e.g., the AI-based smart power plant fault early warning system 100 can be a distributed system). In another exemplary design concept, the AI-based smart power plant fault early warning system 100 can be local or remote. For example, the AI-based smart power plant fault early warning system 100 can access data and / or data stored in machine-readable storage medium 120 via a network. Alternatively, the AI-based smart power plant fault early warning system 100 can directly connect to machine-readable storage medium 120 to access stored data and / or data. In yet another exemplary design concept, the AI-based smart power plant fault early warning system 100 can be implemented on a cloud platform. As an example only, the cloud platform can include private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-layered cloud, etc., or any combination thereof.

[0222] Machine-readable storage medium 120 can store data and / or instructions. In one exemplary design, machine-readable storage medium 120 can store multi-dimensional monitoring data acquired from an external terminal. In another exemplary design, machine-readable storage medium 120 can store multi-dimensional monitoring data and / or instructions used by a smart power plant fault early warning system 100 applying artificial intelligence to perform or use the exemplary methods described in this invention. In one exemplary design, machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write storage, read-only storage (ROM), etc., or any combination thereof. Exemplary mass storage may include disks, optical disks, solid-state drives, etc. Exemplary removable storage may include flash drives, floppy disks, optical disks, memory cards, compact disks, magnetic tapes, etc. Exemplary volatile read-write storage may include random access memory (RAM). Exemplary RAM may include active random access memory (DRAM), double data rate synchronous active random access memory (DDR SDRAM), passive random access memory (SRAM), thyristor random access memory (T-RAM), and zero-capacitance random access memory (Z-RAM), etc. Exemplary read-only memories may include mask read-only memories (MROMs), programmable read-only memories (PROMs), erasable programmable read-only memories (PEROMs), electrically erasable programmable read-only memories (EEPROMs), optical disc read-only memories (CD-ROMs), and digital multifunction disk read-only memories, etc. In one exemplary design, the machine-readable storage medium 120 can be implemented on a cloud platform. By way of example only, the cloud platform may include private clouds, public clouds, hybrid clouds, community clouds, distributed clouds, internal clouds, multi-tiered clouds, etc., or any combination thereof.

[0223] In the specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the smart power plant fault early warning method with artificial intelligence as described in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130. The processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0224] The specific implementation process of processor 110 can be found in the various method embodiments executed by the smart power plant fault early warning system 100 using artificial intelligence described above. The implementation principle and technical effect are similar, and will not be repeated here.

[0225] 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 smart power plant fault early warning method applying artificial intelligence is implemented.

[0226] Similarly, it should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof.

Claims

1. A smart power plant fault early warning method applying artificial intelligence, characterized in that, The method includes: Artificial intelligence network models are used to analyze the operation log data of smart power plants to obtain knowledge graphs of unit associations between generator units, sensor associations between sensor networks, control system associations between control systems, first association knowledge graphs between sensor networks and generator units, and second association knowledge graphs between sensor networks and control systems. Key clusters are identified in the composite knowledge network formed by the unit-related knowledge graph, the sensor-related knowledge graph, the control system-related knowledge graph, the first related knowledge graph, and the second related knowledge graph. The identified key clusters are analyzed to determine the key cluster members in the generator set knowledge level, and the participation relationship characteristics of the key cluster members in the generator set are determined. Based on the participation relationship characteristics of the key cluster members in the generator set, the fault modes and fault symptoms of the key cluster members are determined by matching with the historical fault mode library. The historical fault mode library is a fault feature mode library between key components of the generator set, sensor network and control system established based on historical fault data and prior fault characteristics. Based on the failure modes and symptoms of the key cluster members, the key cluster members are divided into different risk levels, and corresponding early warning information is generated.

2. The smart power plant fault early warning method using artificial intelligence according to claim 1, characterized in that, The identification of key clusters in the composite knowledge network formed by the unit-related knowledge graph, the sensor-related knowledge graph, the control system-related knowledge graph, the first related knowledge graph, and the second related knowledge graph includes: Construct a knowledge member selection vector for knowledge level x; where x = 0, 1, 2, the knowledge member selection vector for knowledge level x is a description formed by multiple knowledge member indicator features, each knowledge member indicator feature is used to reflect whether the knowledge member corresponding to knowledge level x is selected as a key cluster member in the key cluster; Based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1 corresponding to the unit-related knowledge graph, the relation array M2 corresponding to the sensor-related knowledge graph, the relation array M3 corresponding to the control system-related knowledge graph, the relation array M4 corresponding to the first related knowledge graph, and the relation array M5 corresponding to the second related knowledge graph, a target model is generated. The knowledge member selection vectors of the target model at each knowledge level are parsed to identify the key clusters.

3. The smart power plant fault early warning method using artificial intelligence according to claim 2, characterized in that, The step of generating a target model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1 corresponding to the unit-related knowledge graph, the relation array M2 corresponding to the sensor-related knowledge graph, the relation array M3 corresponding to the control system-related knowledge graph, the relation array M4 corresponding to the first related knowledge graph, and the relation array M5 corresponding to the second related knowledge graph includes: Based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1, the relation array M2, and the relation array M3, a model within the knowledge level is generated; Based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M4, and the relation array M5, a knowledge level model is generated. The target model is generated by summing the models within the knowledge hierarchy and the models between the knowledge hierarchies and then implementing a minimization strategy.

4. The smart power plant fault early warning method using artificial intelligence according to claim 3, characterized in that, The step of generating a knowledge-level model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M1, relation array M2, and relation array M3 includes: Based on the first weight vector, the basic array, and the relation array corresponding to knowledge level x of the composite knowledge network, array feature data F1 of knowledge level x is generated; wherein, the relation array corresponding to knowledge level 0 is the relation array M1, the relation array corresponding to knowledge level 1 is the relation array M2, the relation array corresponding to knowledge level 2 is the relation array M3, the first weight vector is a vector in which all elements are 1, and the basic array is a vector array in which only the diagonal elements are 1 and the rest are zero; Based on the array feature data F1 of the knowledge level x, the first preset model parameters, and the knowledge member selection vector of the knowledge level x, the array feature data F2 of the knowledge level x is generated. Based on the knowledge member selection vector of the knowledge level x and the relation array corresponding to the knowledge level x, the array feature data F3 of the knowledge level x is generated; Based on the array feature data F2 and array feature data F3 of the knowledge level x, the array feature data F4 of the knowledge level x is generated; Based on the array feature data F4 of each knowledge level of the composite knowledge network, a model within the knowledge level is generated.

5. The smart power plant fault early warning method using artificial intelligence according to claim 4, characterized in that, The step of generating array feature data F1 for knowledge level x based on the first weight vector, the basic array, and the relation array corresponding to knowledge level x of the composite knowledge network includes: The fusion calculation result of the first weight vector and the reverse array of the first weight vector corresponding to the knowledge level x is determined, and the array feature data F5 of the knowledge level x is generated. The array feature data F5 of the knowledge level x is determined by the array difference between the basic array of the knowledge level x and the array difference between the basic array of the knowledge level x and the relation array corresponding to the knowledge level x, and the array feature data F1 of the knowledge level x is generated. The step of generating array feature data F2 of knowledge level x based on array feature data F1 of knowledge level x, first preset model parameters, and knowledge member selection vector of knowledge level x includes: The fusion calculation result of the reverse array of the knowledge member selection vector of the knowledge level x and the array feature data F1 is determined, and then fused with the fusion calculation result of the knowledge member selection vector of the knowledge level x to generate the array feature data F6 of the knowledge level x. The fusion calculation result of the first preset model parameters and the array feature data F6 of the knowledge level x is determined, and the array feature data F2 of the knowledge level x is generated. The step of generating array feature data F3 for knowledge level x based on the knowledge member selection vector of knowledge level x and the relation array corresponding to knowledge level x includes: The fusion calculation result of the reverse array of the knowledge member selection vector of the knowledge level x and the relation array of the knowledge level x is determined, and then fused with the fusion calculation result of the knowledge member selection vector of the knowledge level x to generate the array feature data F3 of the knowledge level x. The step of generating array feature data F4 for knowledge level x based on array feature data F2 and array feature data F3 includes: The difference between the array feature data F2 and the array feature data F3 of the knowledge level x is determined, and the array feature data F4 of the knowledge level x is generated.

6. The smart power plant fault early warning method using artificial intelligence according to claim 4, characterized in that, The step of generating the model within each knowledge level based on the array feature data F4 of each knowledge level of the composite knowledge network includes: The array feature data F4 of each knowledge level of the composite knowledge network are summed to generate the model within the knowledge level.

7. The smart power plant fault early warning method using artificial intelligence according to claim 3, characterized in that, The step of generating an inter-knowledge level model based on the knowledge member selection vectors of the composite knowledge network at each knowledge level, the relation array M4, and the relation array M5 includes: Based on the knowledge member selection vector of knowledge level x, the knowledge member selection vector of knowledge level y, the first weight vector corresponding to knowledge level x, and the first weight vector of knowledge level y, array feature data F7 between knowledge level x and knowledge level y is generated; where y = 0, 1, 2, y ≠ x; Based on the knowledge member selection vector of knowledge level x and the knowledge member selection vector of knowledge level y, generate array feature data F8 between knowledge level x and knowledge level y; Based on the relation array between knowledge level x and knowledge level y, the second preset model parameters of the relation array between knowledge level x and knowledge level y, the array feature data F7 and array feature data F8 between knowledge level x and knowledge level y, the array feature data F9 between knowledge level x and knowledge level y is generated; wherein, when x=0 and y=1, the relation array between knowledge level 0 and knowledge level 1 is the relation array M4, when x=1 and y=2, the relation array between knowledge level 1 and knowledge level 2 is the relation array M5, and when x=0 and y=2, the relation array between knowledge level 0 and knowledge level 2 is defaulted to an array where all element vector values ​​are 0; Based on the array feature data F9 between every two knowledge levels of the composite knowledge network, the inter-knowledge level model is generated.

8. The smart power plant fault early warning method using artificial intelligence according to claim 7, characterized in that, The step of generating array feature data F7 between knowledge level x and knowledge level y based on the knowledge member selection vector of knowledge level x, the knowledge member selection vector of knowledge level y, the first weight vector corresponding to knowledge level x, and the first weight vector of knowledge level y includes: The difference between the first weight vector corresponding to the knowledge level x and the knowledge member selection vector of the knowledge level x is determined, and the array feature data F10 of the knowledge level x is generated. The difference between the first weight vector corresponding to the knowledge level y and the knowledge member selection vector of the knowledge level y is determined, and the array feature data F11 of the knowledge level y is generated. The fusion calculation result of the array feature data F10 of knowledge level x and the reverse array feature data F11 of knowledge level y is determined, and the array feature data F7 between knowledge level x and knowledge level y is generated. The step of generating array feature data F8 between knowledge level x and knowledge level y based on the knowledge member selection vector of knowledge level x and the knowledge member selection vector of knowledge level y includes: The fusion calculation result of the knowledge member selection vector of knowledge level x and the reverse array of knowledge member selection vector of knowledge level y is determined, and array feature data F8 between knowledge level x and knowledge level y is generated. The step of generating array feature data F9 between knowledge level x and knowledge level y based on the corresponding relationship array between knowledge level x and knowledge level y, the second preset model parameters of the corresponding relationship array between knowledge level x and knowledge level y, array feature data F7 between knowledge level x and knowledge level y, and array feature data F8 includes: The difference between the relation array corresponding to knowledge level x and knowledge level y and the array feature data F8 between knowledge level x and knowledge level y is determined, and then the difference between the array feature data F7 between knowledge level x and knowledge level y is used to generate the array feature data F12 between knowledge level x and knowledge level y. The element-wise product of the relation array corresponding to the knowledge level x and the knowledge level y and the array feature data F12 between the knowledge level x and the knowledge level y is determined to generate the array feature data F13 between the knowledge level x and the knowledge level y. Determine the sum of squares of each element in the array of array feature data F13 between knowledge level x and knowledge level y; The sum of squares of each element in the array is determined as the result of the fusion calculation of the second preset model parameters between the knowledge level x and the knowledge level y, and array feature data F9 between the knowledge level x and the knowledge level y is generated. The step of generating the inter-knowledge-level model based on the array feature data F9 between every two knowledge levels of the composite knowledge network includes: The array feature data F9 between every two knowledge levels of the composite knowledge network are summed to generate the inter-knowledge level model.

9. The smart power plant fault early warning method using artificial intelligence according to any one of claims 2-8, characterized in that, The step of parsing the knowledge member selection vectors of the target model at each knowledge level to identify the key clusters includes: The knowledge member selection vectors of the target model at each knowledge level are softened, and the target model is transformed to generate a transformed target model. The knowledge member selection sequences of all knowledge levels of the composite knowledge network are concatenated to generate a concatenated sequence. The transformation target model is applied to the cascaded sequence using a stepwise decreasing method to generate array feature data F14; The key cluster is identified based on the array feature data F14; Wherein, the array feature data F14 is a description content with the same dimension as the concatenated sequence, and the array members in the array feature data F14 correspond one-to-one with the knowledge member indicator features in the concatenated sequence; The step of identifying the key cluster based on the array feature data F14 includes: For any array member in the array feature data F14, if the array member is greater than a threshold value, then the knowledge member corresponding to the array member is determined to be selected and loaded into the critical cluster; if the array member is not greater than the threshold value, then the knowledge member corresponding to the array member is determined not to be selected and loaded into the critical cluster.

10. A smart power plant fault early warning system applying artificial intelligence, characterized in that, The intelligent power plant fault early warning system applying artificial intelligence includes a processor and a machine-readable storage medium. The machine-readable storage medium stores machine-executable instructions, which are loaded and executed by the processor to implement the intelligent power plant fault early warning method applying artificial intelligence according to any one of claims 1-9.

Images