Knowledge graph generation method and system based on remote centralized control event

Abstract

The embodiment of the invention provides a knowledge graph generation method and system based on a remote centralized control event, and relates to the technical field of power systems. The method comprises the following steps: acquiring log data about the hydroelectric centralized control system, and extracting at least one group of to-be-processed data from the log data according to a timestamp; performing entity identification on the operation content in each to-be-processed data to obtain one or more entities, and determining a centralized control event indicated by two entities having an association relationship; and forming at least one hydrological node based on the hydrological data of the centralized control event under the timestamp to which the centralized control event belongs, so that the hydrological nodes are mutually connected to form a graph evolution axis, and establishing connection between the event node for representing the centralized control event and the hydrological node by taking the timestamp as an index so as to create the knowledge graph. According to the embodiment of the invention, the problem that an existing knowledge graph construction method cannot overcome the influence of non-system factors in a hydropower centralized control scene is solved, and the accuracy of hydropower centralized control is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular, the present application relates to a knowledge graph generation method and system based on remote control events. BACKGROUND

[0002] Water power remote control refers to a technology and system for monitoring and controlling water power systems through remote control and centralized management. The water power control system connects sensors, instruments, actuators and other devices with computer networks, and realizes remote monitoring, remote control and centralized management of water power devices and networks by transmitting and processing various data and information. However, due to the large number of devices involved in the water power control system, the complex network, there are a large amount of device information, network topology, operation instructions and other data, and traditional databases and search engines cannot meet the analysis and mining needs of the internal correlation of the system. Knowledge graph can form semantic knowledge expression by establishing global link relationships between entities, realize in-depth understanding and analysis of internal information of the system, and provide reliable support for system optimization and fault troubleshooting.

[0003] At present, the construction of knowledge graph is divided into rule construction and automatic construction. Rule construction refers to analyzing source data and constructing corresponding knowledge graph by artificial or artificial rules. This method depends on the user's professional knowledge in this field. However, in the water power remote control scene, the characteristics of control events in different hydrological scenes are different, and the control link is complex. From the aspects of manpower cost and time cost, it is not advisable to construct knowledge graph by using rule construction method. Secondly, by using automatic construction method, although the entity in the source data can be recognized by NLP model, the existing knowledge graph automatic construction method is only suitable for the situation that non-system factors do not interfere with the system time, and the non-system factors in the water power scene will affect the characteristics of the system event, so that the existing method is difficult to cope with.

[0004] Therefore, the existing knowledge graph construction method cannot overcome the influence of hydrological factors and other non-system factors in the water power control scene, and cannot be applied to the water power control system, which greatly affects the accuracy of water power control. SUMMARY

[0005] The embodiment of the present application provides a knowledge graph generation method and system based on remote control events, which is used to solve the technical problem that the existing knowledge graph construction method cannot overcome the influence of hydrological factors and other non-system factors in the water power control scene, and can be applied to the water power control system, thereby improving the accuracy of water power control.

[0006] According to one aspect of the embodiment of the present application, a knowledge graph generation method based on remote control events is provided, comprising:

[0007] Obtain log data about the hydropower centralized control system, and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data.

[0008] Entity identification is performed on the operation content in each of the data to be processed to obtain one or more entities, and the centralized control event indicated by two entities with an association relationship is determined;

[0009] Based on the hydrological data of the centralized control event under its corresponding timestamp, at least one hydrological node is formed, and the hydrological nodes are connected to each other to form a graph evolution axis. The event node used to represent the centralized control event is connected to the hydrological node to create a knowledge graph.

[0010] In one possible implementation, the step of acquiring log data about the hydropower centralized control system and extracting at least one set of data to be processed from the log data by timestamp includes:

[0011] The log data is divided according to timestamps to obtain at least one set of log data to be processed;

[0012] Using a keyword template to indicate the identification of at least one keyword, one or more target keywords are identified in each of the log data to be processed, and the start and end positions of each target keyword are determined.

[0013] The starting point of any extraction interval is taken as the end position of any target keyword, and the ending point of the extraction interval is taken as the starting position of the next target keyword, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

[0014] In one possible implementation, the step of performing entity recognition on the operation content in each of the data to be processed to obtain one or more entities includes:

[0015] The operation content is divided into at least one sentence, and each sentence is encoded into a tensor with a preset encoding dimension;

[0016] The tensor is input into a pre-trained decoder for entity recognition, and at least one entity is output.

[0017] Identifiers are marked on the entity data used to characterize the entity in the operation content.

[0018] In one possible implementation, the decoder is trained through the following steps:

[0019] The set of historical tensors constructed from pre-acquired historical data is used as the training data;

[0020] For each training data, Gaussian noise is introduced, and after executing a specified number of steps, multiple training data to be identified are obtained, and each training data to be identified is associated with at least one real entity;

[0021] The training data to be identified is used as the input to a pre-established decoder. The entity corresponding to the training data to be identified is determined based on the output of the decoder, so as to train the decoder.

[0022] The decoder includes a loss function, which is constructed based on the correspondence between real entities and candidate entities generated by the decoder. The correspondence includes a one-to-one mapping relationship or a one-to-many mapping relationship between real entities and candidate entities.

[0023] In one possible implementation, the method further includes:

[0024] The training data to be identified is determined by the following formula (1):

[0025]

[0026] Where, x t The training data to be identified is denoted as , where 'a' is the noise factor, '∈' is Gaussian noise, 't' is the number of steps, and 'x0' is the training data.

[0027] The loss function is determined by the following formula (2):

[0028]

[0029] Where L is the loss function, z is the number of candidate entities, σ is an intermediate variable, r is the right boundary, l is the left boundary, and c is a specific identifier. Let ξ be the classification probability of the i-th candidate entity. σ (i) represents the classification of the i-th candidate entity.

[0030] In one possible implementation, the central control event is determined through the following steps:

[0031] Extract at least one non-entity word from the data other than the entity data marked with the identifier in the data to be processed;

[0032] When a subject-verb-object relationship is detected between any two entity data marked with the identifier and the non-entity words located between them, it is determined that the two entities have an association relationship in order to obtain the centralized control event indicated by them.

[0033] In one possible implementation, the method further includes:

[0034] Determine the system state of the entities included in the centralized control event at their respective timestamps, and generate parameter nodes to characterize the determined system state;

[0035] In the knowledge graph, connections are created between each event node and its associated parameter nodes.

[0036] In one possible implementation, based on the hydrological data of the centralized control event at its corresponding timestamp, at least one hydrological node is formed, such that the hydrological nodes are interconnected to form a graph evolution axis, and the event node representing the centralized control event is used as an index to establish a connection with the hydrological node to create a knowledge graph, including:

[0037] In the hydrological data, hydrological parameter information associated with at least one specified hydrological keyword is obtained to generate at least one hydrological node, each hydrological node corresponding to one of the hydrological keywords;

[0038] The hydrological nodes are connected according to their associated hydrological parameter information in a specified trend to form a core axis and a spectral evolution axis containing multiple core axes.

[0039] Multiple hydrological nodes at the same timestamp are interconnected in the graph evolution axis.

[0040] According to another aspect of the embodiments of this application, a knowledge graph generation system based on remote centralized control events is provided, comprising:

[0041] The log acquisition module is used to acquire log data about the hydropower centralized control system and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data.

[0042] An entity recognition module is used to perform entity recognition on the operation content in each of the data to be processed, so as to obtain one or more entities and determine the centralized control event indicated by two entities with an association relationship;

[0043] The knowledge graph creation module is used to form at least one hydrological node based on the hydrological data of the centralized control event under its corresponding timestamp, so that the hydrological nodes are connected to each other to form a graph evolution axis, and the event node representing the centralized control event is used as an index to establish a connection with the hydrological node to create a knowledge graph.

[0044] In one possible implementation, the log acquisition module includes:

[0045] A log partitioning unit is used to partition the log data according to timestamps to obtain at least one set of log data to be processed;

[0046] The keyword extraction unit is used to determine one or more target keywords in each of the log data to be processed by using a keyword template for indicating the identification of at least one keyword, and to determine the start and end positions of each target keyword;

[0047] The data extraction unit is used to take the end position of any target keyword as the starting point of any extraction interval and the starting position of the next target keyword as the ending point of the extraction interval, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

[0048] The beneficial effects of the technical solutions provided in this application are:

[0049] The knowledge graph generation method based on remote centralized control events provided in this application embodiment acquires log data about a hydropower centralized control system and extracts at least one set of data to be processed from the log data according to timestamps. The data to be processed includes operation content, system status, and hydrological data. Then, entity recognition is performed on the operation content in each set of data to obtain one or more entities, and centralized control events indicated by two related entities are identified. Based on the hydrological data of the centralized control events under their respective timestamps, at least one hydrological node is formed, and the hydrological nodes are connected to each other to form a knowledge graph evolution axis. The timestamp is used as an index to establish connections between the event nodes representing the centralized control events and the hydrological nodes to create a knowledge graph. In this way, by constructing centralized control events and integrating hydrological data, connections are established between different hydrological scenarios and centralized control events. Taking into account the influence of non-system factors such as hydrological factors in hydropower scenarios, the knowledge graph constructed in this way is applicable to complex and ever-changing hydropower control scenarios. It can achieve flexible and accurate control of hydropower centralized control systems, thereby solving the technical problem that existing knowledge graph construction methods cannot overcome the influence of non-system factors such as hydrological factors in hydropower centralized control scenarios, and improving the accuracy and stability of hydropower centralized control. Attached Figure Description

[0050] Figure 1 A flowchart illustrating a knowledge graph generation method based on remote centralized control events, provided in an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of the structure of a knowledge graph generation system based on remote centralized control events, provided in an embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] Example 1

[0054] Figure 1 This is a flowchart illustrating a knowledge graph generation method based on remote centralized control events, provided in an embodiment of this application. The knowledge graph generation method based on remote centralized control events includes steps S101 to S103.

[0055] S101. Obtain log data about the hydropower centralized control system, and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data.

[0056] In this application, the log data records relevant operational data of the hydropower centralized control system, such as facility operation status data, fault and anomaly data, operation records, system operation status data, and operation logs. Facility operation status includes real-time and historical data such as the on / off status, operating mode, water level, flow rate, and temperature of the hydropower facilities; fault and anomaly data records faults, abnormal situations, and alarm information of the hydropower facilities; operation records include operations performed on the hydropower facilities (such as starting / stopping, adjusting parameters, etc.) and operator information; system operation status data includes the operation status of the hydropower centralized control system, network connection status, and data transmission status; operation logs record operations performed on the hydropower centralized control system (such as logging in, logging out, and configuration changes, etc.) and operator information. Subsequently, the log data is divided according to the timestamps it carries, and multiple sets of data to be processed, Data(D1, D2, ..., D...), are extracted. n (), where n is the number of data sets to be processed. Each data set Di contains data parameters related to the operation content, system status, and hydrological data. Specifically, the operation content includes the operation object, operation type, and operation parameters; the system status includes the operating status parameters of the hydropower centralized control system at the corresponding timestamp; and the hydrological data includes water level data, flow data, water quality data, and precipitation data at the corresponding timestamp. This application does not limit the content related to the operation content, system status, and hydrological data.

[0057] In one optional embodiment, the data to be processed is stored in a structured form, that is, the operation content, system status and hydrological data are used as indexes to store the relevant data parameters, so that the data can be quickly retrieved and obtained in subsequent processing, which facilitates data management.

[0058] S102. Perform entity identification on the operation content in each of the data to be processed to obtain one or more entities, and determine the centralized control event indicated by two entities with an association relationship.

[0059] S103. Based on the hydrological data of the centralized control event under its corresponding timestamp, at least one hydrological node is formed, such that the hydrological nodes are connected to each other to form a graph evolution axis, and the event node used to represent the centralized control event is connected to the hydrological node to create a knowledge graph.

[0060] In this application, the control logic of centralized control is identified by judging data such as the operation object, operation type, and operation parameters in the control content, thereby obtaining multiple entities. An event jointly indicated by two related entities is considered as a centralized control event. Subsequently, an event node (Node) corresponding to the centralized control event is constructed. B It also acquires the hydrological data for each centralized control event at its corresponding timestamp, thereby generating one or more hydrological nodes related to the hydrological data. This represents the i-th hydrological node, meaning each event node corresponds to at least one hydrological node. Furthermore, multiple hydrological nodes are interconnected to form a graph evolution axis, which characterizes the evolution process of the knowledge graph. Simultaneously, event nodes are connected to hydrological nodes using timestamps as indices, thereby obtaining the knowledge graph.

[0061] Therefore, this application establishes a link between each centralized control event and its corresponding hydrological data. Considering the correlation between centralized control events and hydrological factors, it can more accurately predict and compensate for the impact of non-system factors such as hydrological factors, enabling the hydropower centralized control system to maintain stable operation when facing complex and diverse hydrological factors, meet actual operation needs, and facilitate quick and accurate adjustment to cope with different application scenarios, thereby improving the control accuracy of the hydropower centralized control system.

[0062] In this application, because the knowledge graph generated by the above method contains the correlation between centralized control events and hydrological conditions, the knowledge graph can support the operation of the hydropower centralized control system under different hydropower scenarios. Specifically, hydropower centralized control can be to monitor the operation data of various equipment in the hydropower station in real time (monitoring content includes the operation data of equipment such as generating units, switchyard equipment, and public equipment, such as the flow rate, water level, head, generating unit status, and power generation of the hydropower station) to ensure its normal operation. It can also be to remotely control and regulate the hydropower station according to its operation mode and control strategy. The control functions include generating unit start-up and shutdown, circuit breaker operation, and load regulation, while the regulation functions include the adjustment of parameters such as water level and flow rate. It can also be to analyze the operating data of the hydropower station to provide early warning and diagnosis of possible faults and provide corresponding handling suggestions and solutions.

[0063] The knowledge graph generation method based on remote centralized control events provided in this embodiment obtains log data about a hydropower centralized control system and extracts at least one set of data to be processed from the log data according to timestamps. The data to be processed includes operation content, system status, and hydrological data. Then, entity recognition is performed on the operation content in each set of data to obtain one or more entities, and centralized control events indicated by two related entities are identified. Based on the hydrological data of the centralized control events under their respective timestamps, at least one hydrological node is formed, and the hydrological nodes are connected to each other to form a knowledge graph evolution axis. Interstamps are used as indexes to establish connections between event nodes representing the centralized control events and hydrological nodes to create a knowledge graph. In this way, by constructing centralized control events and integrating hydrological data, connections are established between different hydrological scenarios and centralized control events. Taking into account the influence of non-system factors such as hydrological factors in hydropower scenarios, the knowledge graph constructed in this way is applicable to complex and ever-changing hydropower control scenarios, enabling flexible and accurate control of hydropower centralized control systems. This solves the technical problem that existing knowledge graph construction methods cannot overcome the influence of non-system factors such as hydrological factors in hydropower centralized control scenarios, and improves the accuracy and stability of hydropower centralized control.

[0064] In some embodiments, acquiring log data about the hydropower centralized control system and extracting at least one set of data to be processed from the log data by timestamp includes:

[0065] The log data is divided according to timestamps to obtain at least one set of log data to be processed;

[0066] Using a keyword template to indicate the identification of at least one keyword, one or more target keywords are identified in each of the log data to be processed, and the start and end positions of each target keyword are determined.

[0067] The starting point of any extraction interval is taken as the end position of any target keyword, and the ending point of the extraction interval is taken as the starting position of the next target keyword, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

[0068] In this embodiment, keyword templates are used to obtain the operation content, system status, and hydrological data from each set of log data to be processed. Specifically, keywords corresponding to multiple keywords recorded in the keyword template are located in the log data to be processed and used as target keywords, meaning at least one target keyword is extracted from each set of log data to be processed. Then, for each set of log data to be processed, every two adjacent target keywords are used as the starting and ending points of an extraction interval, meaning each set of log data to be processed contains at least one extraction interval. Data within all extraction intervals is then combined into a set of data to be processed. Therefore, this embodiment utilizes keyword templates to divide the data, enabling rapid extraction of data to be processed and improving data processing efficiency.

[0069] In some embodiments, the step of performing entity recognition on the operation content in each of the data to be processed to obtain one or more entities includes:

[0070] The operation content is divided into at least one sentence, and each sentence is encoded into a tensor with a preset encoding dimension;

[0071] The tensor is input into a pre-trained decoder for entity recognition, and at least one entity is output.

[0072] Identifiers are marked on the entity data used to characterize the entity in the operation content.

[0073] In this embodiment, the operation content is extracted from multiple sets of data to be processed, and the operation content is divided into at least one sentence S(s1, s2, ..., s) according to a specified sentence segmentation method. m ), where m is the number of sentences, and Sm is the m-th sentence. For example, the sentence segmentation method can be based on specific keywords, symbols (such as Chinese and English periods), etc. Optionally, the sentences are input into an encoding model for encoding, and the output is a tensor with a preset encoding dimension, such as R×M×1. This encoding model is a neural network model composed of a BERT model and a bidirectional LSTM model. Further, the output of the encoding model is used to perform entity recognition, obtain multiple entities, and insert the identifier [CLS] at the corresponding entity data of each entity.

[0074] Therefore, this embodiment improves the accuracy of entity recognition by encoding and decoding after dividing the sentences into multiple parts, which helps to reduce the possibility of misidentifying other irrelevant words or phrases. At the same time, the granularity of sentence division can be adjusted according to the needs of centralized control logic to better adapt to different centralized control tasks and improve the flexibility of entity extraction.

[0075] Based on the above embodiments, in some embodiments, the decoder is trained by the following steps:

[0076] The set of historical tensors constructed from pre-acquired historical data is used as the training data;

[0077] For each training data, Gaussian noise is introduced, and after executing a specified number of steps, multiple training data to be identified are obtained, and each training data to be identified is associated with at least one real entity;

[0078] The training data to be identified is used as the input to a pre-established decoder. The entity corresponding to the training data to be identified is determined based on the output of the decoder, so as to train the decoder.

[0079] The decoder includes a loss function, which is constructed based on the correspondence between real entities and candidate entities generated by the decoder. The correspondence includes a one-to-one mapping relationship or a one-to-many mapping relationship between real entities and candidate entities.

[0080] In this embodiment, the historical tensor set H constructed from historical data s (h1, h2, ..., h S ) represents the training data, s represents the number of tensors in the historical tensor set, and h represents the training data. s Let h be the s-th tensor. It should be noted that, since the hydropower centralized control system involves multiple hydropower devices, and different devices represent the same thing differently, the diffusion principle is used during training for each tensor h. s K-step Gaussian noise is introduced to achieve entity disambiguation, simplifying the disambiguation method. In the noise processing step, a new data point is generated at each step; therefore, a tensor h after K-step noise processing... s K new data points are generated, resulting in K+1 data points (i.e., training data to be identified). Each training data point contains at least one real entity, which is represented by its official terminology. For example, the official term for an apple is "apple," but in a certain region it might be called "ping'er." Furthermore, due to the introduction of Gaussian noise, the multilayer perceptron configured in the decoder generates Z candidate entities. The relationship between real entities and candidate entities can be a one-to-one mapping or a one-to-many mapping. This embodiment uses the Hungarian algorithm to represent this correspondence and constructs a loss function based on it to measure the similarity or difference between real entities and candidate entities, thereby reducing the gap between the decoder's output and the actual results.

[0081] In some embodiments, the method further includes:

[0082] The training data to be identified is determined by the following formula (1):

[0083]

[0084] Where, x t The training data to be identified is denoted as , where 'a' is the noise factor, '∈' is Gaussian noise, 't' is the number of steps, and 'x0' is the training data.

[0085] The loss function is determined by the following formula (2):

[0086]

[0087] Where L is the loss function, z is the number of candidate entities, σ is an intermediate variable, r is the right boundary, l is the left boundary, and c is a specific identifier. Let ξ be the classification probability of the i-th candidate entity. σ (i) represents the classification of the i-th candidate entity. Specifically, ξ r (i) represents the right boundary of the i-th candidate entity, ξ l (i) represents the left boundary of the i-th candidate entity, ξ σ (i) represents the real entity to which the i-th candidate entity belongs. Let be the classification probability that the i-th candidate entity is the right boundary. Let be the classification probability that the i-th candidate entity is the left boundary. Let be the classification probability that the i-th candidate entity is a real entity.

[0088] In some embodiments, the central control event is determined by the following steps:

[0089] Extract at least one non-entity word from the data other than the entity data marked with the identifier in the data to be processed;

[0090] When a subject-verb-object relationship is detected between any two entity data marked with the identifier and the non-entity words located between them, it is determined that the two entities have an association relationship in order to obtain the centralized control event indicated by them.

[0091] In this embodiment, data other than entity data in the data to be processed is segmented to extract non-entity words such as verbs, prepositions, and nouns. Then, based on the non-entity words between two entities, it is determined whether a subject-verb-object relationship exists between them. If so, the two entities are considered to have a certain correlation, i.e., an association relationship. The events indicated by the two entities determined to have an association relationship are then designated as centralized control events. Therefore, this embodiment generates a knowledge graph by constructing centralized control events about two related entities, specifying and refining them to smaller control operations, thus improving the accuracy and detail of the knowledge graph. It also facilitates the connection between centralized control events related to two entities and hydrological data, and considering the correlation between centralized control events and hydrological factors, improves the control precision of the hydropower centralized control system.

[0092] In some embodiments, the method further includes:

[0093] Determine the system state of the entities included in the centralized control event at their respective timestamps, and generate parameter nodes to characterize the determined system state;

[0094] In the knowledge graph, connections are created between each event node and its associated parameter nodes.

[0095] In this embodiment, the system state under the corresponding timestamp is extracted based on the entity contained in the centralized control event, that is, the running state parameters of the corresponding entity, in order to construct a parameter node. A Secondly, construct the event nodes corresponding to the centralized control events. B Therefore, based on the timestamp, the event nodes associated with any centralized control event are... B With parameter node Node A Connections are established, and the edges between event nodes and parameter nodes in the knowledge graph represent the correlation between them. Therefore, this embodiment visually displays and marks the relationship between events and system states by establishing connections between event nodes and parameter nodes related to centralized control events in the knowledge graph.

[0096] In some embodiments, the process of forming at least one hydrological node based on the hydrological data of the centralized control event at its respective timestamp, connecting the hydrological nodes to form a graph evolution axis, and using the timestamp as an index to establish connections between event nodes representing the centralized control event and the hydrological nodes to create a knowledge graph, includes:

[0097] In the hydrological data, hydrological parameter information associated with at least one specified hydrological keyword is obtained to generate at least one hydrological node, each hydrological node corresponding to one of the hydrological keywords;

[0098] The hydrological nodes are connected according to their associated hydrological parameter information in a specified trend to form a core axis and a spectral evolution axis containing multiple core axes.

[0099] Multiple hydrological nodes at the same timestamp are interconnected in the graph evolution axis.

[0100] In this embodiment, hydrological data for each centralized control event at its corresponding timestamp is obtained. This can be achieved by matching keyword templates to retrieve hydrological parameter information under different hydrological keywords for each hydrological node. A hydrological keyword is represented. Then, hydrological nodes are connected according to their associated hydrological parameters with a specified trend, forming axis cores. For example, to represent precipitation, hydrological nodes are connected with a monotonically increasing or decreasing trend. It should be noted that the knowledge graph includes a graph evolution axis with multiple axis cores. Simultaneously, hydrological nodes corresponding to different hydrological keywords at the same time stamp are connected to each other along the graph evolution axis, using the time stamp as the index to establish connections between hydrological nodes and event nodes, thus enabling the establishment of relationships between hydrological nodes, event nodes, and parameter nodes guided by the time stamp. Furthermore, the graph construction module also supports the generation of specified node edges.

[0101] Therefore, this embodiment establishes connections between hydrological nodes, event nodes, and parameter nodes in the knowledge graph. Considering the influence of non-system factors such as hydrological factors on system state, control operations, and other events in hydropower scenarios, the knowledge graph constructed in this way is applicable to complex and ever-changing hydropower control scenarios, enabling flexible and accurate control of hydropower centralized control systems.

[0102] Example 2

[0103] Figure 2 This application provides a schematic diagram of the structure of a knowledge graph generation system based on remote centralized control events. The knowledge graph generation system 200 based on remote centralized control events includes:

[0104] Log acquisition module 201 is used to acquire log data about the hydropower centralized control system and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data.

[0105] The entity recognition module 202 is used to perform entity recognition on the operation content in each of the data to be processed, so as to obtain one or more entities and determine the centralized control event indicated by two entities with an association relationship;

[0106] The knowledge graph creation module 203 is used to form at least one hydrological node based on the hydrological data of the centralized control event under its respective timestamp, so that the hydrological nodes are connected to each other to form a graph evolution axis, and the event node used to represent the centralized control event is connected to the hydrological node to create a knowledge graph.

[0107] In some embodiments, the log acquisition module 201 includes:

[0108] A log partitioning unit is used to partition the log data according to timestamps to obtain at least one set of log data to be processed;

[0109] The keyword extraction unit is used to determine one or more target keywords in each of the log data to be processed by using a keyword template for indicating the identification of at least one keyword, and to determine the start and end positions of each target keyword;

[0110] The data extraction unit is used to take the end position of any target keyword as the starting point of any extraction interval and the starting position of the next target keyword as the ending point of the extraction interval, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

[0111] In some embodiments, the entity recognition module 202 includes:

[0112] The encoding unit is used to divide the operation content into at least one sentence and encode each sentence into a tensor with a preset encoding dimension;

[0113] A decoding unit is used to input the tensor into a pre-trained decoder to perform entity recognition and output at least one entity.

[0114] A marking unit is used to mark an identifier on entity data used to characterize an entity in the operation content.

[0115] In some embodiments, the decoding unit includes:

[0116] The training data acquisition unit is used to determine the set of historical tensors constructed from pre-acquired historical data as training data;

[0117] The noise processing unit is used to introduce Gaussian noise into each training data and obtain multiple training data to be identified after executing a specified number of steps, and each training data to be identified is associated with at least one real entity.

[0118] A decoder training unit is used to use the training data to be identified as the input of a pre-established decoder, and to determine the entity corresponding to the training data to be identified based on the output of the decoder, so as to train the decoder.

[0119] The decoder includes a loss function, which is constructed based on the correspondence between real entities and candidate entities generated by the decoder. The correspondence includes a one-to-one mapping relationship or a one-to-many mapping relationship between real entities and candidate entities.

[0120] In some embodiments, the decoder training unit includes:

[0121] The training data calculation unit is used to determine the training data to be identified using the following formula (1):

[0122]

[0123] Where, x t The training data to be identified is denoted as , where 'a' is the noise factor, '∈' is Gaussian noise, 't' is the number of steps, and 'x0' is the training data.

[0124] The loss function construction unit is used to determine the loss function by the following formula (2):

[0125]

[0126] Where L is the loss function, z is the number of candidate entities, σ is an intermediate variable, r is the right boundary, l is the left boundary, and c is a specific identifier. Let ξ be the classification probability of the i-th candidate entity. σ (i) represents the classification of the i-th candidate entity.

[0127] In some embodiments, the entity recognition module 202 further includes:

[0128] A non-entity word extraction unit is used to extract at least one non-entity word from the data to be processed, excluding entity data marked with the identifier.

[0129] The centralized control event acquisition unit is used to determine that the two entities have an association relationship when it is detected that there is a subject-verb-object relationship between any two entity data marked with the identifier and the non-entity words located between them, so as to obtain the centralized control event indicated by it.

[0130] In some embodiments, the knowledge graph creation module 203 includes:

[0131] The parameter node generation unit is used to determine the system state of the entities included in the centralized control event under their respective timestamps, and to generate parameter nodes to characterize the determined system state.

[0132] The node connection unit is used to create connections between each event node and its associated parameter nodes in the knowledge graph.

[0133] In some embodiments, the knowledge graph creation module 203 further includes:

[0134] A hydrological node generation unit is used to obtain hydrological parameter information associated with at least one specified hydrological keyword from the hydrological data, so as to generate at least one hydrological node, each hydrological node corresponding to one of the hydrological keywords;

[0135] The map evolution axis construction unit is used to connect the hydrological nodes according to their associated hydrological parameter information in a specified trend to form a core axis and a map evolution axis containing multiple core axes.

[0136] A hydrological node connection unit is used to connect multiple hydrological nodes at the same time stamp to each other in the graph evolution axis.

[0137] The system of this application embodiment can execute the method provided in this application embodiment. The implementation principle is similar. The actions performed by each module in the system of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the system, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0138] In the description of embodiments of the present invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0139] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0140] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A method for generating a knowledge graph based on remote centralized control events, characterized in that, include: Obtain log data about the hydropower centralized control system, and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data. Entity identification is performed on the operation content in each of the data to be processed to obtain one or more entities, and the centralized control event indicated by two entities with an association relationship is determined; Based on the hydrological data of the centralized control event under its corresponding timestamp, at least one hydrological node is formed, and the hydrological nodes are connected to each other to form a graph evolution axis. The event node used to represent the centralized control event is connected to the hydrological node to create a knowledge graph.

2. The knowledge graph generation method based on remote centralized control events according to claim 1, characterized in that, The process of acquiring log data about the hydropower centralized control system and extracting at least one set of data to be processed from the log data by timestamp includes: The log data is divided according to timestamps to obtain at least one set of log data to be processed; Using a keyword template to indicate the identification of at least one keyword, one or more target keywords are identified in each of the log data to be processed, and the start and end positions of each target keyword are determined. The starting point of any extraction interval is taken as the end position of any target keyword, and the ending point of the extraction interval is taken as the starting position of the next target keyword, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

3. The knowledge graph generation method based on remote centralized control events according to claim 2, characterized in that, The step of performing entity recognition on the operation content in each of the data to be processed to obtain one or more entities includes: The operation content is divided into at least one sentence, and each sentence is encoded into a tensor with a preset encoding dimension; The tensor is input into a pre-trained decoder for entity recognition, and at least one entity is output. Identifiers are marked on the entity data used to characterize the entity in the operation content.

4. The knowledge graph generation method based on remote centralized control events according to claim 3, characterized in that, The decoder is trained using the following steps: The set of historical tensors constructed from pre-acquired historical data is used as the training data; For each training data, Gaussian noise is introduced, and after executing a specified number of steps, multiple training data to be identified are obtained, and each training data to be identified is associated with at least one real entity; The training data to be identified is used as the input to a pre-established decoder. The entity corresponding to the training data to be identified is determined based on the output of the decoder, so as to train the decoder. The decoder includes a loss function, which is constructed based on the correspondence between real entities and candidate entities generated by the decoder. The correspondence includes a one-to-one mapping relationship or a one-to-many mapping relationship between real entities and candidate entities.

5. The knowledge graph generation method based on remote centralized control events according to claim 4, characterized in that, The method further includes: The training data to be identified is determined by the following formula (1): Where, x t The training data to be identified is denoted as , where 'a' is the noise factor, '∈' is Gaussian noise, 't' is the number of steps, and 'x0' is the training data. The loss function is determined by the following formula (2): Where L is the loss function, Z is the number of candidate entities, σ is an intermediate variable, r is the right boundary, l is the left boundary, and c is a specific identifier. Let ξ be the classification probability of the i-th candidate entity. σ (i) represents the classification of the i-th candidate entity.

6. The knowledge graph generation method based on remote centralized control events according to claim 4 or 5, characterized in that, The following steps are used to identify centralized control events, including: Extract at least one non-entity word from the data other than the entity data marked with the identifier in the data to be processed; When a subject-verb-object relationship is detected between any two entity data marked with the identifier and the non-entity words located between them, it is determined that the two entities have an association relationship in order to obtain the centralized control event indicated by them.

7. The knowledge graph generation method based on remote centralized control events according to claim 6, characterized in that, The method further includes: Determine the system state of the entities included in the centralized control event at their respective timestamps, and generate parameter nodes to characterize the determined system state; In the knowledge graph, connections are created between each event node and its associated parameter nodes.

8. The knowledge graph generation method based on remote centralized control events according to claim 7, characterized in that, The process involves forming at least one hydrological node based on the hydrological data of the centralized control event at its corresponding timestamp, connecting these hydrological nodes to form a graph evolution axis, and using the timestamp as an index to establish connections between event nodes representing the centralized control event and the hydrological nodes to create a knowledge graph, including: In the hydrological data, hydrological parameter information associated with at least one specified hydrological keyword is obtained to generate at least one hydrological node, each hydrological node corresponding to one of the hydrological keywords; The hydrological nodes are connected according to their associated hydrological parameter information in a specified trend to form a core axis and a spectral evolution axis containing multiple core axes. Multiple hydrological nodes at the same timestamp are interconnected in the graph evolution axis.

9. A knowledge graph generation system based on remote centralized control events, characterized in that, include: The log acquisition module is used to acquire log data about the hydropower centralized control system and extract at least one set of data to be processed from the log data according to the timestamp. The data to be processed includes operation content, system status and hydrological data. An entity recognition module is used to perform entity recognition on the operation content in each of the data to be processed, so as to obtain one or more entities and determine the centralized control event indicated by two entities with an association relationship; The knowledge graph creation module is used to form at least one hydrological node based on the hydrological data of the centralized control event under its corresponding timestamp, so that the hydrological nodes are connected to each other to form a graph evolution axis, and the event node representing the centralized control event is used as an index to establish a connection with the hydrological node to create a knowledge graph.

10. The knowledge graph generation system based on remote centralized control events according to claim 9, characterized in that, The log acquisition module includes: A log partitioning unit is used to partition the log data according to timestamps to obtain at least one set of log data to be processed; The keyword extraction unit is used to determine one or more target keywords in each of the log data to be processed by using a keyword template for indicating the identification of at least one keyword, and to determine the start and end positions of each target keyword; The data extraction unit is used to take the end position of any target keyword as the starting point of any extraction interval and the starting position of the next target keyword as the ending point of the extraction interval, so that the log data to be processed in at least one extraction interval is determined as a set of data to be processed.

Images