Hydraulic power plant operation training system and method based on virtual reality technology

Abstract

The invention relates to the technical field of virtual reality, and discloses a hydraulic power plant operation training system and method based on the virtual reality technology, and the method comprises the steps: firstly obtaining historical operation time series data of a hydraulic power plant, dividing the historical operation time series data into discrete operation segments, and obtaining operation segment time series data; constructing a hydraulic power plant operation knowledge graph based on the operation fragments and the operation fragment time sequence data; mapping the hydraulic power plant operation knowledge graph to a three-dimensional virtual space, and constructing a hydraulic power plant virtual reality scene; and finally, designing a VR operation training course based on the scene, and carrying out immersive VR operation training. According to the invention, hydraulic power plant operation knowledge is extracted by using a knowledge graph technology, and immersive operation training is carried out in combination with a virtual reality technology, so that the intuition, interactivity and immersion of training are greatly improved, and understanding and mastering of students on hydraulic power plant operation regulations and safety precautions are deepened.

Description

Technical Field

[0001] The present invention relates to the field of virtual reality technology, and more specifically, to a hydropower plant operation training system and method based on virtual reality technology. Background Art

[0002] In the field of hydropower plant operation training, virtual reality technology is increasingly being used in the operation training process due to its advantages of immersion and interactivity. However, the current hydropower plant operation training system based on virtual reality technology still has some shortcomings in modeling methods, knowledge representation, etc.

[0003] The Chinese patent application with publication number CN110491232A discloses a hydropower plant simulation training system and method based on virtual reality technology. The system provides an effective training method for hydropower plant maintenance personnel by constructing a simulation system that integrates visualization and digitization and is highly consistent with the actual production process. However, the method mainly focuses on the physical simulation modeling of hydropower plant equipment, and is slightly insufficient in the knowledge mining and representation of equipment operating procedures, safety precautions, etc., which is not conducive to the trainees' comprehensive grasp of the operation essentials of the hydropower plant.

[0004] The Chinese patent application with publication number CN110989842A discloses a training method, system and electronic device based on virtual reality. The method includes the steps of establishing a virtual scene space, loading a virtual model, receiving action information of an interactive object, and generating interactive instructions for controlling the virtual model. The patent proposes a general virtual reality training program, but for large and complex systems such as hydropower plants, how to extract operational knowledge from massive historical data to guide VR scene construction and training course design is not mentioned in the patent.

[0005] In summary, most existing technologies focus on the physical simulation modeling of equipment, while the mining of operating procedures, timing relationships during equipment operation and knowledge representation is relatively weak; there is a lack of valuable operational knowledge extracted from historical operation data to guide the construction of virtual scenes and the design of training courses. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a hydropower plant operation training system and method based on virtual reality technology. With the help of knowledge engineering technology, the explicit or implicit knowledge such as hydropower plant operation experience and procedures are systematically refined and modeled; the refined knowledge is mapped to the VR space method to realize the automatic generation of VR training scenes.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The hydropower plant operation training method based on virtual reality technology includes:

[0009] Obtain the historical operation time series data of the hydropower plant, divide the historical operation time series data into n1 discrete operation segments, and obtain the operation segment time series data; construct the hydropower plant operation knowledge graph based on the operation segments and the operation segment time series data;

[0010] Map the hydropower plant operation knowledge graph to the three-dimensional virtual space, build a hydropower plant virtual reality scene, design VR operation training courses based on the hydropower plant virtual reality scene, and carry out immersive VR operation training based on the VR operation training courses.

[0011] Furthermore, each operation segment corresponds to an operation event; the operation segment timing data includes the start and end timestamps of the operation segment, the device location coordinates involved in the operation segment, the name of the operation event to which the operation segment belongs, and the device operating parameters within the operation segment.

[0012] Furthermore, the construction of the hydropower plant operation knowledge graph includes:

[0013] Each operation fragment is abstracted into an operation node of the hydropower plant operation knowledge graph, and the name of the operation node is the operation event to which the operation fragment belongs;

[0014] Acquire attribute information of an operation node from the operation segment time series data, wherein the attribute information of the operation node includes time attribute, space attribute, semantic attribute and numerical attribute;

[0015] Discover frequent operation patterns from the time series data of operation fragments and construct time series logic rules that reflect the order of operations;

[0016] According to the attribute information of the operation nodes, the correlation between different operation nodes is analyzed; combined with the correlation between the different operation nodes and the temporal logic rules, directed causal logic edges are constructed; the operation nodes and directed causal logic edges are organized into a semantic network to form a knowledge graph of hydropower plant operations.

[0017] Furthermore, the acquiring of the attribute information of the operation node from the operation fragment time series data includes:

[0018] Extract the start and end timestamps of the operation fragment in the operation fragment time series data as the time attribute of the corresponding operation node;

[0019] Extracting the device position coordinates involved in the operation fragment from the operation fragment time series data as the spatial attributes of the corresponding operation node;

[0020] Extracting the name of the operation event to which the operation fragment belongs from the operation fragment time series data as the semantic attribute of the corresponding operation node;

[0021] The statistical characteristics of the equipment operating parameters in the operation fragments in the operation fragment time series data are calculated as the numerical attributes of the corresponding operation nodes.

[0022] Furthermore, the step of discovering frequent operation patterns from the operation fragment time series data and constructing a time series logic rule reflecting the operation sequence includes:

[0023] Perform frequent pattern mining on the operation fragment time series data, summarize n2 operation modes from the operation fragment time series data, and calculate the support of each operation mode;

[0024] Set a support threshold, traverse the support of n2 operation modes, and define the operation mode whose support exceeds the support threshold as a frequent operation mode;

[0025] Perform confidence evaluation on the frequent operation modes, remove the operation modes whose confidence is lower than a preset confidence threshold from the frequent operation modes, and obtain candidate operation modes;

[0026] The candidate operation mode is represented as a sequential logic rule, the left part of the sequential logic rule is the preceding operation, the right part is the subsequent operation, and the arrow in the middle represents the sequential order.

[0027] Further, the analyzing the correlation between different operation nodes according to the attribute information of the operation nodes includes: constructing operation node pairs, calculating the comprehensive correlation coefficients of the operation node pairs according to the attribute information of the operation nodes, and constructing a comprehensive correlation coefficient matrix of all operation node pairs; traversing the comprehensive correlation coefficient matrix, marking the operation node pairs corresponding to the elements in the comprehensive correlation coefficient matrix that are greater than a preset correlation threshold as relevant operation nodes;

[0028] The method for constructing directed causal logic edges is to use the temporal logic rules to label the causal relationships of relevant operation nodes;

[0029] The semantic network takes operation nodes as entities and directed causal logic edges as entity relationships.

[0030] Furthermore, the comprehensive correlation coefficient of the computing operation node pair includes:

[0031] According to the time attributes of the operation nodes, the time correlation coefficient of the operation node pairs is calculated;

[0032] According to the spatial attributes of the operation nodes, the spatial correlation coefficient of the operation node pairs is calculated;

[0033] According to the semantic attributes of the operation nodes, the semantic correlation coefficient of the operation node pairs is calculated;

[0034] According to the numerical attributes of the operation nodes, the numerical correlation coefficient of the operation node pairs is calculated;

[0035] The temporal correlation coefficient, spatial correlation coefficient, semantic correlation coefficient and numerical correlation coefficient of the operation node pair are weighted and fused to obtain the comprehensive correlation coefficient of the operation node pair.

[0036] Further, the time correlation coefficient of the computing operation node pair includes:

[0037] The start timestamp and the end timestamp of the two operation nodes in the operation node pair are extracted, the overlap score and the interval decay coefficient of the operation node pair are calculated, and the overlap score and the interval decay coefficient are multiplied to obtain the time correlation coefficient of the operation node pair.

[0038] Furthermore, the construction of a hydropower plant virtual reality scene includes:

[0039] Map the operation nodes in the hydropower plant operation knowledge graph to virtual equipment objects in the hydropower plant virtual reality scene, and arrange the position coordinates of the virtual equipment objects according to the spatial attributes of the operation nodes;

[0040] Mapping the semantic attributes of the operation nodes in the hydropower plant operation knowledge graph into the operation event scripts of the virtual device objects;

[0041] Map the directed causal logic edges in the hydropower plant operation knowledge graph into interactive logical relationships between virtual device objects;

[0042] The three-dimensional model resources of the hydropower plant are loaded, and the virtual equipment objects, operation event scripts and interactive logical relationships are combined to form a completed virtual reality scene of the hydropower plant.

[0043] A hydropower plant operation training system based on virtual reality technology, which is used to implement the above-mentioned hydropower plant operation training method based on virtual reality technology, comprises:

[0044] Knowledge graph construction module: used to obtain the historical operation time series data of the hydropower plant, divide the historical operation time series data into n1 discrete operation fragments, and obtain the operation fragment time series data; based on the operation fragments and the operation fragment time series data, construct the hydropower plant operation knowledge graph;

[0045] Virtual reality scene construction module: used to map the hydropower plant operation knowledge graph into a three-dimensional virtual space and construct a virtual reality scene of the hydropower plant;

[0046] VR training module: Based on the virtual reality scene of the hydropower plant, design VR operation training courses, and carry out immersive VR operation training based on the VR operation training courses.

[0047] An electronic device comprises a memory, a central processing unit and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the above-mentioned hydropower plant operation training method based on virtual reality technology is implemented.

[0048] A computer-readable storage medium stores a computer program, which, when executed, implements the above-mentioned hydropower plant operation training method based on virtual reality technology.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] By mining historical operation data, the operation knowledge of hydropower plants can be automatically extracted, and the operation rules and causal relationships implicit in the data can be made explicit to form a structured knowledge base, which reduces the cost of manual summarization and improves the comprehensiveness and objectivity of knowledge. The knowledge graph is mapped to a three-dimensional virtual space to build a realistic panoramic model of the hydropower plant. Trainees can observe the equipment structure and layout from a first-person perspective in a virtual environment, understand the process flow, and deepen their intuitive understanding of the overall picture of the power plant. VR equipment is used to provide trainees with an immersive and interactive operation experience. Trainees can personally simulate processes such as valve switching and button operation, and observe the operation effects in real time, which not only improves their interest in learning, but also exercises their practical operation ability. Based on the reasoning function of the knowledge graph, various operating conditions can be intelligently generated, and the standardization and safety of trainees' operations can be evaluated in real time, making the training process more flexible, diverse, and more targeted. The application of virtual reality has greatly reduced the safety risks and economic costs of practical training. Trainees can practice repeatedly in a virtual environment without worrying about equipment damage or personal injury, saving training expenses. The design data and operating experience of the hydropower plant are integrated into the knowledge graph to form a complete and explainable knowledge base, which provides data support and knowledge reference for equipment status diagnosis, production optimization decision-making, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 It is a flow chart of the hydropower plant operation training method based on virtual reality technology in the present invention;

[0053] Figure 2 A flow chart of a method for constructing a knowledge graph of hydropower plant operation in a hydropower plant operation training method based on virtual reality technology of the present invention;

[0054] Figure 3 A flow chart of a method for acquiring attribute information of operation nodes from operation segment time series data in a hydropower plant operation training method based on virtual reality technology of the present invention;

[0055] Figure 4 A method flow chart for constructing a temporal logic rule reflecting the sequence of operations in a hydropower plant operation training method based on virtual reality technology of the present invention;

[0056] Figure 5 A flow chart of a method for calculating a comprehensive correlation coefficient of an operation node pair in a hydropower plant operation training method based on virtual reality technology of the present invention;

[0057] Figure 6 A flow chart of a method for constructing a virtual reality scene of a hydropower plant in a method for training hydropower plant operation based on virtual reality technology of the present invention;

[0058] Figure 7 A method flow chart for designing a VR operation training course in a hydropower plant operation training method based on virtual reality technology of the present invention;

[0059] Figure 8 It is a functional module diagram of the hydropower plant operation training system based on virtual reality technology in the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] See also Figure 1 As shown, this embodiment provides a hydropower plant operation training method based on virtual reality technology, including:

[0063] Step S1000, obtaining historical operation time series data of the hydropower plant, dividing the historical operation time series data into n1 discrete operation segments, and obtaining operation segment time series data; constructing a hydropower plant operation knowledge graph based on the operation segments and the operation segment time series data; n1 is a positive integer;

[0064] Further, step S1000 includes:

[0065] Step S1100, obtaining historical operation time series data of the hydropower plant, dividing the historical operation time series data into n1 discrete operation segments, obtaining operation segment time series data, each operation segment corresponds to an operation event; the operation segment time series data includes the start and end timestamps of the operation segment, the location coordinates of the equipment involved in the operation segment, the name of the operation event to which the operation segment belongs, and the equipment operating parameters in the operation segment;

[0066] Specifically, step S1100 aims to extract information related to the operation of the hydropower plant from massive historical data, and perform structured processing on it to prepare data for the construction of the knowledge graph. The historical operation time series data includes the timestamp of the operation, the equipment involved in the operation, the equipment location, etc., the event name of each operation, and the equipment operating parameters of the operation, such as turbine speed, generator power, head, etc. The equipment operating parameters within a period of time (usually several months to several years) are exported from the database of control systems such as DCS and SCADA. These parameters reflect the operating status of the hydropower unit from different aspects. For example, the turbine speed can characterize the dynamic response characteristics of the speed control system, the generator power is directly related to the economy of the unit and the grid dispatch, and the head affects the power and efficiency of the turbine. By collecting key parameter data, the actual operating conditions of the unit can be more comprehensively portrayed.

[0067] Since these raw data are often mixed with various interferences, such as outliers (erroneous data beyond the normal working range) and missing values ​​(data not collected due to equipment failure), direct use will affect the reliability of subsequent analysis. Therefore, it is necessary to clean the data before feature extraction. Common cleaning methods include: setting the value range, treating the data beyond the range as outliers and removing them; interpolating and correcting the missing values, such as interpolating the data at adjacent moments, or extrapolating the data of the same type of equipment. After data cleaning, the influence of noise can be filtered out and the credibility of the data can be improved.

[0068] After obtaining relatively "clean" data, it is necessary to further extract effective information. Since the original data is usually time-continuous, it is not convenient to use it directly for knowledge representation. Time series segmentation technology can be used to divide the continuous data stream into several discrete operation segments. Each segment corresponds to an operation event, such as "starting the turbine" and "grid-connected power generation". The start and end boundaries of the event can be determined based on the sudden change of the equipment state. For example, the speed of the unit rises from zero and stabilizes after reaching the rated speed, which can be regarded as the completion of the "start" event. Fragmentation makes the time data correspond to the specific operation, which is convenient for mapping the value to a semantic label. The operation segment time series data includes the start and end timestamps of each operation segment, reflecting the time attribute of the operation, the equipment location coordinates involved in each operation segment, reflecting the spatial attribute of the operation, the name of the operation event to which each operation segment belongs, reflecting the semantic attribute of the operation, and the equipment operating parameters in the operation segment, such as turbine speed, generator power, head, etc.

[0069] Step S1100 converts the continuous state data of the equipment operation into discrete, semantically clear operation fragments through a series of data preprocessing operations, which simplifies the representation of time data and highlights the causal relationship between events. Clean and structured data lays the foundation for knowledge extraction. On the one hand, the fragmentation of data is conducive to associating quantitative state data with qualitative operational semantics and realizing the symbolization of numerical information; on the other hand, it also reveals the temporal characteristics of events, such as the order of operations, the duration of operations, etc. These temporal characteristics are clues for mining causal logic. At the same time, the removal of abnormal data and the correction of missing values ​​improve the integrity and consistency of the data, reduce invalid calculations in the knowledge extraction process, and improve the confidence of the mining results. It can be seen that data preprocessing is an important part of the construction of knowledge graphs, providing reliable data support for subsequent association analysis.

[0070] Step S1200, constructing a hydropower plant operation knowledge graph based on the operation fragments and the operation fragment time series data;

[0071] Furthermore, if Figure 2 As shown, step S1200 includes:

[0072] Step S1210, abstracting each operation fragment into an operation node of the hydropower plant operation knowledge graph, and the name of the operation node is the operation event to which the operation fragment belongs;

[0073] Specifically, step S1210 adopts an event recognition algorithm based on pattern matching to detect characteristic changes in equipment status (such as water level, pressure, etc.) and operating parameters (such as valve opening, speed setting, etc.) to identify the start and end boundaries of the operation segment and the event to which it belongs. For example, if the bus voltage rises from zero to the rated value and remains stable for more than 2 minutes, it can be identified as a "unit grid connection" event; if the main valve opening drops from the rated value to zero and remains stable for more than 1 minute, it can be identified as a "unit shutdown" event. The name of the event can be automatically generated in combination with the domain dictionary, and can also be revised and improved by experts. For example, the "guide vane opening increase" event can be named "increase guide vane opening", and the "stator current decrease" event can be named "reduce generator output".

[0074] Mapping operation fragments to operation nodes is conducive to the integration of expert knowledge and data information. Expert experience can guide the definition and identification of events. For example, the "unit start-up" event needs to meet the composite conditions of "main valve opening increase, guide vane opening increase, speed increase" at the same time, while the change of a single feature may be caused by equipment oscillation and other reasons, which is not a valid event. Engineers can interpret and correct the names, attributes, relationships and other contents of the nodes based on their own understanding, and participate in the modeling and application of knowledge graphs. At the same time, a large number of fragment samples inject data blood into the event definition. On the one hand, it can verify the universality of empirical rules and discover new event patterns. On the other hand, it can quantify the statistical indicators of event attributes and provide decision-making references. For example, the "unit start-up" event may have multiple changes under different working conditions. Frequently occurring patterns can be identified as typical standard startups, while occasional abnormal patterns can be classified as non-standard startups, thereby guiding the optimization of standards and the establishment of early warning models.

[0075] Step S1220, acquiring attribute information of the operation node from the operation segment time series data, wherein the attribute information of the operation node includes time attribute, space attribute, semantic attribute and numerical attribute;

[0076] Furthermore, if Figure 3 As shown, step S1220 includes:

[0077] Step S1221, extracting the start and end timestamps of the operation segments in the operation segment timing data as the time attributes of the corresponding operation nodes;

[0078] Step S1222, extracting the device position coordinates involved in the operation fragment in the operation fragment time series data as the spatial attributes of the corresponding operation node;

[0079] Step S1223, extracting the name of the operation event to which the operation fragment belongs from the operation fragment time series data as the semantic attribute of the corresponding operation node;

[0080] Step S1224, calculating the statistical characteristics of the equipment operating parameters in the operation segment in the operation segment time series data as the numerical attributes of the corresponding operation node.

[0081] Specifically, the start and end timestamps of the operation fragments are extracted as the time attributes of the operation nodes. The time attribute reflects the start and end time of the operation and is a basic element for describing the characteristics of the operation sequence. The extraction of the time attribute converts the continuous time value into a discrete time interval, which is convenient for time series comparison and statistical analysis. At the same time, this attribute is also an important clue to track the sequence of events and discover causal laws. The equipment location coordinates involved in the operation fragment are extracted as the spatial attributes of the operation node. The spatial attribute reflects the physical location of the operation in the hydropower plant and is of great significance for describing the spatial association of different operations. For example, the installation position of each unit can be obtained according to the unit layout diagram and expressed in plane coordinates. The spatial coordinates can be used to calculate the distance between different operations and analyze the spatial correlation of operations, such as the principle of "proximity control"; in addition, the spatial attribute can also be used for visualization to generate a spatial distribution map of each operation in the plant, providing an intuitive reference for monitoring decisions.

[0082] The name of the operation event to which the operation fragment belongs is extracted as the semantic attribute of the operation node. The semantic attribute gives the operation a clear business meaning and is a bridge for communicating engineering knowledge and data information. The operation event can be automatically judged according to the change of equipment status, such as "start" when the speed increases and "run" when the output remains the same, etc., or it can be manually labeled in combination with the work order record. The semantic attribute makes the operation node explainable, which is convenient for display and management to business personnel; at the same time, this attribute is also important information for mining business logic and causal dependencies between operations, which can be used to build a knowledge graph that conforms to the laws of physics. For example, it can be found that "on-site startup is required after telemetry lock", thus constraining the logical relationship between operations. The statistical characteristics of the equipment operating parameters in the operation segment are calculated as the numerical attributes of the operation node. The numerical attribute characterizes the input and output of the operation process from a quantitative perspective and is an important indicator for evaluating operation performance. Common statistical features include mean, variance, interval, etc., which can be flexibly selected according to the nature of the parameters and analysis requirements. For example, for the speed parameter, the mean and standard deviation within its segment can be calculated. The former reflects the average level of the speed, and the latter reflects the degree of fluctuation of the speed; for the head parameter, its maximum and minimum values ​​can be given to characterize the range of head variation. These statistical values ​​can be used for operation clustering on the one hand to discover the distribution law of parameters under different operation modes; on the other hand, they can also be used as quantitative indicators to measure the similarity of operation nodes, providing a quantitative basis for constructing numerical associations. Numerical attributes contain rich statistical information and are a powerful tool for reflecting operation differences at the data level.

[0083] The above four attributes describe the characteristics of the operation fragment from multiple angles, from the four dimensions of time, space, semantics, and value. By extracting these attributes from the fragment time series data, the upgrade from data to knowledge is achieved, and the fragment information is further structured and semantic. On the one hand, these attributes give the operation nodes a clear physical meaning, which conforms to people's cognitive habits and is easy to explain to business personnel; on the other hand, the attribute values ​​also provide a quantitative measurement for the association analysis of the operation nodes, so that the mining and evaluation of knowledge have a calculable basis. At the same time, different attributes can complement each other and comprehensively reflect the characteristics of the operation mode. For example, high-output and long-term operations usually correspond to full-load operation, while frequent start-stop and large-scale adjustment operations may be peak-shaving operations. The cross-combination of multi-dimensional attributes can depict more subtle differences in operations.

[0084] In summary, step S1220 realizes the feature representation of the operation node by extracting the time, space, semantic and numerical attributes of the fragment data. This process integrates multiple technologies and methods such as time series analysis, statistical learning, expert knowledge, etc., and reflects the characteristics of multidisciplinary intersection. Attribute extraction enables the information of the operation fragment to be expressed and fully utilized in a standardized manner, which not only facilitates human-computer interaction, but also lays a data foundation for subsequent correlation analysis and causal reasoning. Through structured, semantic and quantitative feature representation, it provides a new idea for extracting knowledge from massive operation data.

[0085] Step S1230, discovering frequent operation patterns from the operation segment time series data, and constructing a time series logic rule reflecting the operation sequence;

[0086] Furthermore, if Figure 4 As shown, step S1220 includes:

[0087] Step S1231, performing frequent pattern mining on the operation fragment time series data, summarizing n2 operation patterns from the operation fragment time series data, and calculating the support of each operation pattern; n2 is a positive integer;

[0088] Step S1232, setting a support threshold, traversing the support of n2 operation modes, and defining an operation mode whose support exceeds the support threshold as a frequent operation mode;

[0089] Step S1233, performing confidence evaluation on the frequent operation modes, removing operation modes with confidences lower than a preset confidence threshold from the frequent operation modes, and obtaining candidate operation modes;

[0090] Step S1234, representing the candidate operation mode as a sequential logic rule, wherein the left part of the sequential logic rule is the preceding operation, the right part is the subsequent operation, and the arrow in the middle represents the sequential order.

[0091] Specifically, step S1230 aims to automatically discover implicit operation rules from a large amount of operation fragment time series data. The operation mode reflects the behavioral habits of personnel controlling equipment, such as "start the turbine first and then connect to the grid for power generation", etc. These operation sequences contain rich prior knowledge. Through the frequent pattern mining algorithm, recurring operation combinations, namely frequent operation modes, can be summarized. The formal definition of frequent patterns is: in a given operation fragment time series data set, a subsequence of operation fragments whose frequency exceeds the minimum support threshold. For example, in 1,000 operation fragments, the pattern of "starting the turbine, adjusting the guide vanes, and connecting to the grid for power generation" appears 200 times. If the support threshold is set to 10%, the pattern is a frequent pattern. The support reflects the universality of the pattern. High support means that the pattern is repeated many times in historical data, representing a stable control method.

[0092] By setting the support threshold, frequent operation patterns are screened out. Due to the complexity and variability of actual working conditions, there may be some noise patterns in the frequent patterns, such as atypical operation sequences caused by equipment failure and human error. In order to exclude these accidental situations, it is necessary to conduct a confidence assessment on the candidate patterns. The credibility of a pattern can be measured by using indicators such as lift and confidence in association rule mining. The larger the lift value, the smaller the probability that the operations in the pattern occur simultaneously by chance, and the stronger the correlation between the operations; the higher the confidence value, the greater the conditional probability that the subsequent operation occurs when the antecedent operation occurs, and the more obvious the causality between the operations.

[0093] After confidence evaluation, unreliable modes are eliminated from the candidate modes to obtain candidate operation modes with high credibility and good stability. To enhance the interpretability of the mode, the candidate mode is represented as a temporal logic rule such as "A'→B'→C'", and the arrows represent the time sequence. The formal definition of the rule is: if operation C' is observed at time (t+1), operation B' is observed at time t, and operation A' is observed at time (t-1), it can be represented as the temporal logic rule "A'→B'→C'". The use of rule representation can intuitively display the timing constraints of the operation, such as "remote control switching→starting the turbine→grid-connected power generation", which reflects an idea of ​​step-by-step operation from remote to local, from the machine side to the grid side.

[0094] The benefits of mining temporal logic rules are:

[0095] Automatically extract control experience. Frequent and reliable operation modes represent the successful practices of front-line employees, and contain rich control strategies and decision-making ideas. By automatically mining historical data, effective operation rules can be discovered, and the implicit knowledge summarized by employees over the years can be transformed into formal rules.

[0096] Guide employees to operate in a standardized manner. The rationality of the operation sequence directly affects the control effect and equipment life. Embedding timing rules into the training system can guide employees to operate in a standard mode, reduce blindness and mistakes, and improve the control level.

[0097] Provide a reference for anomaly detection. If the actual operation sequence deviates significantly from the historical frequent pattern, it often indicates that the equipment condition or control method is abnormal. Using the mined timing rules as a comparison benchmark can automatically identify unconventional operations and provide clues for fault diagnosis and behavior auditing.

[0098] For example, mining found that "Switch to backup system → load reduction control → report main system for repair" is a high-confidence rule. In an actual operation, after a fault alarm occurred in the main system, the employee directly reduced the load without switching to the backup system, which was inconsistent with the rule pattern. Analysis found that the backup system was also under maintenance at the time, resulting in the employee's failure to fully follow the standard process. Based on this, it is inferred that the equipment availability is reduced and the manual operation load is increased, providing a basis for management personnel to make decisions.

[0099] In summary, step S1230 mines frequent operation patterns from time series data and abstracts them into time series logic rules. It regulates employee behavior and optimizes control processes from a data-driven perspective, injects behavioral cognition and strategic thinking into the knowledge graph, and is an important innovation in the application of knowledge engineering in the industrial field.

[0100] Step S1240, analyzing the correlation between different operation nodes according to the attribute information of the operation nodes; constructing directed causal logic edges in combination with the correlation between the different operation nodes and the temporal logic rules; organizing the operation nodes and directed causal logic edges into a semantic network to form a knowledge graph of hydropower plant operations.

[0101] Further, step S1240 includes:

[0102] Step S1241, constructing operation node pairs, calculating the comprehensive correlation coefficients of the operation node pairs according to the attribute information of the operation nodes, and constructing a comprehensive correlation coefficient matrix of all operation node pairs; traversing the comprehensive correlation coefficient matrix, marking the operation node pairs corresponding to the elements in the comprehensive correlation coefficient matrix that are greater than a preset correlation threshold as relevant operation nodes;

[0103] Furthermore, if Figure 5 As shown, step S1241 includes:

[0104] Step S12411, calculating the time correlation coefficient of the operation node pair according to the time attribute of the operation node;

[0105] The time correlation coefficient of the computing operation node pair includes:

[0106] The start timestamp and the end timestamp of the two operation nodes in the operation node pair are extracted, the overlap score and the interval decay coefficient of the operation node pair are calculated, and the overlap score and the interval decay coefficient are multiplied to obtain the time correlation coefficient of the operation node pair.

[0107]

[0108] in:

[0109] R t (A, B) is the time correlation coefficient of the operation node pair (A, B) consisting of operation node A and operation node B;

[0110] A and B represent two operation nodes;

[0111] Indicates the start timestamp of the operation node A;

[0112] Indicates the end timestamp of the operation node A;

[0113] Indicates the start timestamp of operating node B;

[0114] Indicates the end timestamp of the operation node B;

[0115] is the central timestamp of node A;

[0116] is the central timestamp of node B;

[0117] Indicates the duration of node A;

[0118] Indicates the duration of Node B;

[0119] α is the balancing factor of time overlap, which is used to control the influence of overlap on the time correlation coefficient;

[0120] β is a balancing factor of the time interval, which is used to control the influence of the time interval on the time correlation coefficient. α and β can be set based on experience (for example, α=1, β=0.5).

[0121] The formula consists of two parts, which evaluate the temporal correlation of nodes from the perspectives of temporal overlap and time interval.

[0122] Part I: Overlap Measurement - Overlap Score

[0123]

[0124] This part considers the degree of overlap between two operation nodes on the timeline. Specifically:

[0125] and The deviation degrees of the start time of the operation node B relative to the end time of the operation node A and the end time of the operation node B relative to the start time of the operation node A are measured respectively.

[0126] The smaller the offset value, the higher the overlap between the two operation nodes. The total offset level is obtained by taking the square root of the sum of the squares of the two offset values.

[0127] The parameter α can adjust the impact of the offset value. The larger α is, the stronger the penalty effect of the offset is.

[0128] pass The smaller the offset, the closer the value is to 1. Therefore, the higher the overlap of the operation node pair, the higher the time correlation coefficient.

[0129] Part II: Measurement of time intervals - interval decay coefficient

[0130]

[0131] This part takes into account the time interval between two operation nodes:

[0132] T A -T B |The absolute value of the difference between the central timestamps of two operation nodes is calculated, reflecting the time interval of the operation nodes.

[0133] max(D A ,D B ) Take the larger value of the duration of operation node A and operation node B to normalize the time interval and eliminate the influence of different duration of operation nodes on interval evaluation.

[0134] β is used to adjust the decay rate of the interval size on the correlation. The smaller β is, the greater the negative impact of the interval on the time correlation.

[0135] By e -x "The form of a decay coefficient in the interval (0,1] is obtained. The larger the interval, the lower the time correlation coefficient.

[0136] Comprehensive results

[0137] Multiply the overlap fraction and the interval attenuation coefficient to get the final time correlation coefficient R t (A, B); When the overlap of the operation node pair is higher and the interval is smaller, R tThe higher the value of (A, B), the closer the maximum value is to 1; when the overlap is lower and the interval is larger, R t The lower the value of (A,B), the closer it is to 0.

[0138] This formula takes into account both the time overlap and the time interval, and comprehensively describes the time correlation. By processing the overlap and interval through nonlinear transformation, the correlation decreases exponentially when the overlap is low and the interval is large, which is more in line with cognitive intuition. The introduction of adjustable parameters α and β can flexibly adjust the influence of urgency and interval according to different scenarios. The interval is normalized by the duration of the node itself, eliminating the influence of different time scales on time correlation. This formula quantitatively reflects the correlation strength of the operation node pair in the time dimension, and can be effectively used in subsequent operation node correlation analysis.

[0139] Step S12412, calculating the spatial correlation coefficient of the operation node pair according to the spatial attributes of the operation nodes;

[0140] Step S12413, calculating the semantic correlation coefficient of the operation node pair according to the semantic attributes of the operation nodes;

[0141] Step S12414, calculating the numerical correlation coefficient of the operation node pair according to the numerical attributes of the operation nodes;

[0142] The numerical correlation coefficient of the computing operation node pair includes:

[0143]

[0144] in:

[0145] R v (A, B) represents the numerical correlation coefficient of the operation node pair (A, B) consisting of operation node A and operation node B;

[0146] m represents the numerical attribute dimension of the operation node, that is, each operation node has m equipment operation parameters;

[0147] a i represents the statistical characteristic value of the operating parameters of the i-th device of the operating node A;

[0148] b i represents the statistical characteristic value of the operating parameters of the i-th device of the operating node B;

[0149] α' is the steepness parameter of the similarity mapping, which is used to control the sensitivity of similarity changes;

[0150] θ is the similarity threshold parameter, which indicates the tolerance of differences and can be set based on domain knowledge.

[0151] Formula derivation ideas:

[0152] 1. Relative difference calculation

[0153] For each dimension i of the equipment operating parameters, the relative difference between the two operating nodes in this dimension is calculated:

[0154]

[0155] This relative difference measure eliminates the impact of different device operating parameter dimensions, allowing the differences between different dimensions to be compared.

[0156] 2. Similarity Mapping

[0157] The relative difference is mapped to the interval (0,1) through the Sigmoid function to obtain the single-dimensional similarity:

[0158]

[0159] in, For relative difference.

[0160] The Sigmoid function maps any real number to the interval (0,1) and increases monotonically in an S-shaped manner, with the midpoint of the function being 0.5.

[0161] The parameter α' controls the steepness of the function. The larger the α', the more sensitive the similarity changes. The parameter θ determines the difference value when the similarity is 0.5, which can be regarded as the tolerance of the difference.

[0162] 3. Dimensional similarity average

[0163] Take the arithmetic average of the similarities in m dimensions to get the average similarity of the node pair on all numerical attributes:

[0164]

[0165] Mapped to the interval [0,1].

[0166] In order to facilitate the fusion of similarity with other types of correlation coefficients, the range of numerical correlation coefficients should be within [0,1]. However, the average similarity obtained by averaging the dimensional similarities in step 3 has a range of (0,1). In order to map it to [0,1], we use the characteristics of the inverse tangent function arctan to transform it: arctan();

[0167] The inverse tangent function can map (0, +∞) to Multiply by The average similarity in (0,1) can be "stretched" to the range of [0,1] to obtain the final numerical correlation coefficient.

[0168] m,ai ,b i The statistical features extracted from the operation fragment time series data reflect the numerical attributes of the operation node on different device operation parameters. These parameters can be obtained by statistically analyzing the operation fragment time series data, such as mean, variance, interval, etc.

[0169] α',θ are adjustable hyperparameters that need to be set based on expert experience and data distribution. α' controls the sensitivity of the correlation coefficient to parameter differences. The larger α' is, the more dramatic the change of the correlation coefficient with parameter differences. θ represents the tolerable parameter difference. The larger θ is, the greater the parameter difference is, the more the correlation coefficient will decrease significantly.

[0170] The formula takes into account the dimensional differences of different numerical attributes through relative differences and Sigmoid mapping, so that multi-dimensional attributes can be measured uniformly, and the greater the difference, the lower the similarity, which is in line with cognitive common sense. The introduction of adjustable parameters α' and θ can flexibly adjust the sensitivity and tolerance of numerical correlation to parameter differences according to different scenarios, thereby improving the applicability of the formula. The inverse tangent function is used to realize the mapping from similarity to correlation coefficient, so that it satisfies the value range of [0,1], which is convenient for fusion with other correlation metrics to construct a comprehensive correlation evaluation index. The formula quantitatively describes the numerical correlation of operation node pairs from the perspective of multi-parameter differences. The greater the difference, the lower the correlation, and the formula result changes nonlinearly with the parameter difference. The steepness of the change can be controlled by adjusting α' and θ. This formula not only conforms to cognitive intuition, but also has high flexibility and can adapt to different application needs. Therefore, it is of great significance in constructing a semantic network that reflects the association of operation nodes.

[0171] Step S12415, weighted fusion of the temporal correlation coefficient, spatial correlation coefficient, semantic correlation coefficient and numerical correlation coefficient of the operation node pair is performed to obtain a comprehensive correlation coefficient of the operation node pair.

[0172] Specifically, step S1241 comprehensively considers the correlation between operation nodes from multiple dimensions, integrates the four aspects of time, space, semantics, and numerical values, and constructs a correlation measurement matrix that reflects the degree of correlation of operations. Correlation analysis is the core of knowledge graph construction, and its rationality determines the accuracy of the graph structure. Traditional methods often use node labels (such as shared device names) to judge node similarity, ignoring the spatiotemporal characteristics and business attributes of operations in industrial scenarios. In response to this shortcoming, step S1241 systematically proposes the idea of ​​multidimensional correlation analysis:

[0173] First, using the start and end times in the time attribute, the time overlap and interval span of the operation node pair are calculated, and the time correlation coefficient is obtained. Time correlation reflects the closeness of the operations in the timing process. The higher the overlap and the shorter the interval, the stronger the time correlation. For example, "closing the circuit breaker" and "grid-connected power generation" have a high overlap and a short interval, and the two are more closely connected in time, so the time correlation coefficient will be higher. However, "starting the turbine" and "de-energizing the unit" often have a long interval, and the time correlation coefficient is low.

[0174] Next, the physical distance of the operation nodes in the spatial layout of the hydropower plant is calculated using the equipment location coordinates in the spatial attributes, and the spatial correlation coefficient is inferred. Spatial correlation reflects the geographical proximity of different operations. The closer the operation, the higher the correlation from a spatial perspective. For example, the "excitation" and "speed increase" of the same unit are very close in physical location, and the spatial correlation is very strong. However, the operations of different units, such as the "startup" of unit 1 and the "shutdown" of unit 2, although the actions are similar, are separated in space and have weak spatial correlation.

[0175] Furthermore, using the names of operation events in the semantic attributes, with the help of the ontology knowledge base and word embedding model, the similarity of the operation nodes at the semantic level is calculated to obtain the semantic relevance coefficient. Different operations use similar words, such as "grid connection" and "boosting", which often means that they have certain similarities in business functions. Semantic relevance mines the upper and lower relationships and synonymous relationships between operations to reveal implicit business connections. For example, "main valve start" and "guide vane opening" have no intersection in the literal sense, but both belong to the lower level operation of "unit start" in semantics, so they have a high similarity in the semantic space. Using the upper and lower definitions in the knowledge base, such as "main valve" is a lower level concept of "power system" and "guide vane" is a lower level concept of "speed control system", the semantic distance between the two in the knowledge base can be calculated. For example, "decoupling" and "releasing", "putting into" and "switching" respectively constitute synonymous relationships, and the word embedding model can discover the co-occurrence pattern of these words in large-scale corpus, thereby establishing a mathematical mapping and obtaining their similarity in the latent space.

[0176] Finally, the statistical features in the numerical attributes, such as mean and variance, are used to calculate the proximity of the operation nodes at the parameter level to form a numerical correlation coefficient. Numerical correlation mines the input-output relationship of different operations. Operations with consistent parameter change trends often imply internal connections. For example, the "guide vane opening" parameter rises sharply after the "main valve start-up" and rises slowly in the "grid-connected boost", and the two have similar control modes. Through multivariate statistical analysis, the coupling relationship of multiple parameters in the operation process can be found, revealing the causal chain of the operation. For example, in the "starting the steam turbine" operation, as the main steam valve is opened, the steam temperature, pressure, and flow rate rise synergistically, and the speed also increases accordingly. From this, the strong correlation and causal dependence between the physical quantities can be found. For another example, the correlation analysis of "tail water pressure" and "shaft seal water injection volume" shows that the two are significantly negatively correlated. Combined with the physical model, it can be seen that when the tail water pressure is too low, the shaft seal water injection valve needs to be opened to ensure the positive pressure of the steam seal and prevent the steam turbine from extracting steam. The numerical correlation thus confirms the internal cause mechanism between the operation parameters.

[0177] In summary, step S1241 calculates the correlation coefficient of the operation node from four dimensions: time, space, semantics, and value, covering a variety of association modes such as timing constraints, geographic proximity, business associations, and parameter coupling, which can comprehensively and objectively characterize the relationship between operations. On this basis, the four types of correlation coefficients are fused according to certain weights to form a comprehensive correlation measurement matrix. The node pairs with higher comprehensive correlation levels are often adjacent in time and space, semantically connected, and numerically similar. Their combination is more in line with physical laws and cognitive experience, and is more likely to constitute a reasonable operation sequence.

[0178] The advantages of this multidimensional association analysis are:

[0179] Combining soft and hard attributes can reduce misjudgment caused by a single perspective. For example, two operations that are close in time and space are easily judged as highly correlated according to traditional methods. However, after analyzing the semantic attributes, it is found that one operation is "turbine start" and the other is "generator-transformer group protection", and the two are contradictory in business attributes. Introducing the semantic dimension can effectively identify such pseudo-correlations that appear to be consistent but are actually different.

[0180] Introducing qualitative knowledge to make up for the limitations of data samples. In industrial scenarios, the number of samples of special working conditions such as faults is very limited, and it is difficult to discover the correlation patterns by relying solely on data statistics. By using knowledge in the field of fault diagnosis knowledge base and analyzing node correlation at the semantic level, we can supplement the prior knowledge of samples and improve the comprehensiveness of correlation analysis.

[0181] Characterize from multiple angles to enhance the interpretability of the association. When discussing the relationship between two operations, elaborating from different angles such as time, space, semantics, and numerical values ​​can enhance the interpretability of the analysis conclusions, facilitate the transmission of causal logic to business personnel, and apply it to practical guidance. For example, when explaining the relationship between "main valve opening" and "guide vane opening adjustment", the time correlation reveals the order of the two in the timing process, the spatial correlation explains their transmission mechanism in the physical system, the semantic correlation explains the business attributes of the two as "startup" sub-links, and the numerical correlation verifies the control law of the two as cause and effect. Multidimensional analysis makes the association results more intuitive and easier to reason.

[0182] For example, in a certain unit startup operation, the comprehensive correlation strength is calculated for the "main valve opening" node and the "unit grid connection" node: the time attribute shows that the grid connection occurs within 5 minutes after the main valve is opened, with high overlap and short interval, and the time correlation reaches 0.8; the spatial attribute shows that the two belong to the No. 1 unit, the physical distance is very close, and the spatial correlation reaches 0.9; the semantic attribute suggests that "main valve opening" is a necessary link for turbine startup, and "grid connection" is the primary task after startup, the shortest path length in the knowledge base is small, and the semantic correlation reaches 0.7; the correlation analysis of the numerical attribute finds that as the main valve opening gradually increases, the unit power also increases, showing a strong positive correlation trend, and the numerical correlation reaches 0.85. In summary, the correlation strength of this node pair is 0.8125 (assuming that the weights of the four dimensions are the same). It can be seen that multiple dimensions of time, space, semantics, and numerical values ​​jointly confirm the intrinsic connection between the two operations, comprehensively reflecting the similarities of the operations in different feature spaces, making the graph construction closer to engineering reality, and also laying a solid semantic foundation for subsequent reasoning and prediction.

[0183] Step S1242, using temporal logic rules to annotate causal relationships on relevant operation nodes and construct directed causal logic edges;

[0184] Step S1243, taking the operation nodes as entities and the directed causal logic edges as entity relationships, a knowledge graph of hydropower plant operations is formed.

[0185] Specifically, step S1240 aims to mine the intrinsic connections between operation nodes from multidimensional attributes and timing rules and construct causal reasoning logic. The correlation of operation nodes describes the similarity of different operations at the attribute level. Nodes with high similarity often have certain correlations in business attributes. The timing rules reveal the sequence and causal dependency of operations. If two nodes frequently appear in the same rule, they must have some kind of constraint relationship in time and logic. Therefore, by comprehensively considering the correlation of node attributes and the constraints of timing rules, the semantic connection between nodes can be fully characterized.

[0186] Furthermore, when calculating node relevance, it is necessary to measure the similarity of nodes from the time, space, semantics, numerical and other attribute levels, and then weight them together to form a comprehensive relevance. Among them, time similarity can be measured by the overlap and interval length of the operation occurrence time, spatial similarity can be measured by the Euclidean distance of device coordinates, semantic similarity can be measured by the edit distance of the operation name, and numerical similarity can be measured by the correlation coefficient of key parameters. These attribute similarities reflect the association strength of nodes at different semantic levels from different angles.

[0187] By setting a correlation threshold, node pairs with comprehensive correlations higher than the threshold are identified as related operation nodes. This step realizes the transformation from continuous correlation to discrete correlation, highlighting the most significant node associations. For example, when the threshold is set to 0.8, "starting turbine" and "grid-connected power generation" with a comprehensive correlation of 0.9 are marked as related nodes, while "starting turbine" and "maintaining transformer" with a comprehensive correlation of 0.7 do not exceed the threshold and are not related nodes. The selection of the threshold needs to balance the quantity and quality of related operation nodes. If it is too high, valuable weak associations will be missed, and if it is too low, too many invalid associations will be introduced. In practice, it can be adjusted according to the node scale and association distribution.

[0188] After obtaining the relevant operation nodes, it is necessary to further consider the dependency of the operation nodes in temporal logic. Through the temporal rules, the causal order and logical constraints between nodes can be judged, which is of great significance for reasoning and judgment. For example, "remote control start → increase guide vane opening → increase speed → grid-connected power generation" is a high-confidence temporal rule. It can be seen that "remote control start" is the premise of "increase guide vane opening", and "grid-connected power generation" is the result of "increase speed". There is an obvious causal relationship between the nodes. Connect the adjacent node pairs in the temporal rules with directed edges, and the direction of the edge points from the front node to the back node, indicating a unidirectional dependency in time and logic. At the same time, considering the diversity of real operations, it is difficult for a simple linear temporal sequence to cover all modes, and sometimes there are complex structures such as parallelism and loops. Logical attributes can be added to the directed edges, such as AND gates, OR gates, and NOT gates, to construct a causal network with complex temporal semantics.

[0189] The operation nodes and causal logic edges are organized into a semantic network to form a knowledge graph of hydropower plant operations. The semantic network uses nodes to represent conceptual entities and edges to represent the relationship between entities, which is highly consistent with the cognitive model of the human brain. Through association analysis and temporal reasoning, the knowledge graph not only depicts the intrinsic semantics of the operation, but also expresses the decision logic of the control, integrating the originally discrete data into an organic knowledge whole. Compared with the traditional relational model, the knowledge graph has richer semantic expression capabilities and supports complex logical retrieval and analogical reasoning. Organizing knowledge in a graph manner can lay the foundation for applications such as intelligent decision-making, solution optimization, and fault diagnosis.

[0190] Step S1240 mines the intrinsic connections of operation nodes in terms of time, space, semantics, and numerical values ​​through multi-dimensional attribute association analysis, reveals the physical mechanism and business rules behind the operation, and upgrades the node relationship from qualitative description to quantitative measurement, with more accurate semantic expression capabilities. Using temporal rules to construct causal logic edges not only reflects the temporal order of operations, but also depicts the reasoning dependency between nodes. By flexibly combining logical relationships, complex control structures such as parallelism, selection, and loops can be expressed, which helps to analyze and control global decisions. Causal logic enables knowledge graphs to have temporal reasoning capabilities, predict the future, and trace the past. Organizing operation knowledge in a graph format can retrieve and associate information from multiple perspectives and quickly locate related nodes. Users can query and infer the target operation plan through any known information such as operation name, key attributes, and associated nodes. Knowledge graphs are also easy to visualize, providing a friendly knowledge management interface for human-computer interaction. Although the control operations of different hydropower plants differ in equipment parameters, personnel habits, etc., the physical mechanisms and business logic are often similar. Applying knowledge graphs to different scenarios can help discover common rules through analogical reasoning and reduce the cost of knowledge acquisition. Generalized models can be applied to design simulation, virtual training, and other aspects.

[0191] For example, after frequent pattern mining, a hydropower plant found that "switch water diversion system → adjust blade opening → increase turbine speed → grid-connected power generation" and "reduce turbine speed → close blades → stop water diversion → unit decoupling" are two high-confidence time series patterns, corresponding to the key steps of unit start and stop. The operation nodes in these two patterns are mapped to the semantic network, and the correlation between nodes is calculated based on the operation timestamp and equipment parameters. For example, the time overlap between "adjust blade opening" and "increase turbine speed" is as high as 95%, the speed change rate is significantly positively correlated with the blade opening, and the spatial position is also relatively close. A high comprehensive correlation is obtained through weighting, and a strong correlation edge can be constructed. The time series rules are embedded in the semantic network, directed causal edges are added to adjacent nodes, and the parallel logic of blade adjustment and speed increase is considered at the same time, and the AND attribute is added. The final graph can clearly express the step specifications, parameter constraints, and time series logic of the start and stop operations.

[0192] When a new employee is ready to start the unit, the system automatically searches the knowledge graph based on the "start intention" node, quickly locates the operation sequence that is highly related to "unit start", and plans the best decision path for the employee. When an employee mistakenly operates "increasing turbine speed" instead of "adjusting blade opening", the system prompts "blade adjustment is a prerequisite for speed control" based on causal logic, preventing improper operation by the employee.

[0193] In summary, step S1240 constructs an operational knowledge graph that expresses rich semantics and causal logic through attribute association and temporal reasoning, breaks through the bottlenecks of traditional knowledge bases in terms of "structured representation, semantic understanding, and analogical reasoning", and realizes the leap from industrial experience to intelligent models. It is an important innovation in the research of intelligent technologies in the field of hydropower.

[0194] Step S1200 adopts a data-driven knowledge modeling method to realize the functions of semantic abstraction, multi-dimensional description, and association mining of operation fragments, and establishes an operation knowledge graph that combines data foundation and expert wisdom. Through the semantic annotation of nodes in the graph, the strength characterization of edges, and logical causal constraints, coherent business semantics and control methods are extracted from discrete operation fragments. This knowledge automation method that integrates industrial big data with human experience knowledge has bridged the gap between data analysis and business practice, and achieved a leap from perceptual understanding to rational management and control. It has important theoretical value and practical significance for improving the level of hydropower intelligence and ensuring the safe and economical operation of equipment. The graph model is also easy to transplant and reuse, and can be used as a carrier for industry knowledge accumulation and sharing, providing a knowledge engine for intelligent decision-making in different hydropower scenarios, and has promotion value in a wider range of industrial fields.

[0195] Step S2000, mapping the hydropower plant operation knowledge graph to the three-dimensional virtual space, constructing a hydropower plant virtual reality scene, designing a VR operation training course based on the hydropower plant virtual reality scene, and conducting immersive VR operation training based on the VR operation training course.

[0196] Furthermore, step S2000 includes:

[0197] Step S2100, mapping the hydropower plant operation knowledge graph to a three-dimensional virtual space to construct a hydropower plant virtual reality scene;

[0198] Furthermore, if Figure 6 As shown, step S2100 includes:

[0199] Step S2110, mapping the operation nodes in the hydropower plant operation knowledge graph to virtual equipment objects in the hydropower plant virtual reality scene, and arranging the position coordinates of the virtual equipment objects according to the spatial attributes of the operation nodes;

[0200] Step S2120, mapping the semantic attributes of the operation nodes in the hydropower plant operation knowledge graph into the operation event scripts of the virtual device objects;

[0201] Step S2130, mapping the directed causal logic edges in the hydropower plant operation knowledge graph into interactive logical relationships between virtual device objects;

[0202] Step S2140, loading the three-dimensional model resources of the hydropower plant, combining the virtual device objects, operation event scripts and interactive logical relationships, to form a completed virtual reality scene of the hydropower plant.

[0203] Specifically, step S2100 aims to transform the hydropower plant operation knowledge graph into a realistic virtual reality scene, realizing the mapping from symbolic knowledge to visual display. This process uses computer graphics technologies such as 3D modeling, animation scripts, and physics engines to instantiate abstract semantic relationships into specific virtual objects and interactive logic, allowing workers to intuitively perceive and manipulate hydropower equipment in an immersive environment, providing an immersive learning experience.

[0204] First, in step S2110, the operation nodes in the knowledge graph are mapped to virtual device objects. The operation node represents the control event of a certain device in the hydropower plant in a specific time and space, such as "Unit 1 started at 9:30." Through mapping, this node is visualized as a hydroelectric generator set in a virtual environment, located at unit 1 in the plant, and its appearance and structure are highly similar to the real equipment. The spatial attributes of the node are used to determine the coordinate position of the virtual device so that the three-dimensional layout is consistent with the on-site layout. This mapping method intuitively displays the physical semantics of the node, which helps employees to match the abstract symbols in the knowledge graph with specific equipment in the real scene and reduce cognitive load.

[0205] Next, in step S2120, the semantic attributes of the operation node are mapped to the operation event script of the virtual device. The semantic attributes describe the name, type, etc. of the operation, such as "start (event name), turbine (equipment type)". Through mapping, these semantic tags are converted into animation scripts that control the operating status of the virtual device. For example, the "start" attribute triggers the start-up animation of the turbine, accelerating from static to no-load speed; the "grid-connected" attribute triggers the closing animation of the step-up transformer and circuit breaker, connecting the generator set to the virtual power grid. The script-controlled animation process intuitively simulates the state changes of the equipment, making the semantic tags in the knowledge graph "dynamic" and turning them into vivid operation scenes. Employees can intuitively understand the meaning and effects of various operations by observing the operating status of the equipment.

[0206] Then, in step S2130, the directed causal logic edge is mapped to the interaction logic between virtual devices, and the corresponding program code is generated to realize the interaction between virtual device objects. The directed causal logic edge represents the dependency constraint between two operations, such as "turbine start (precondition) -> generator grid connection (result)". Through mapping, this logical relationship is embedded in the interaction code of the virtual device. For example, only after the turbine is successfully started, the control button for the generator grid connection is activated to allow the user to operate; if the turbine is shut down or fails, the grid connection button will be disabled. This interactive logic implements the causal constraint between operations and prevents users from performing illogical and erroneous operations. At the same time, the physical parameters of the equipment, such as the speed and output of the turbine, the voltage and current of the generator, etc., can be calculated in real time, and the numerical attributes in the knowledge graph can be mapped to the operating parameters of the virtual device. When the user changes the control parameters, the state of the virtual device will also change accordingly, reflecting the quantitative effect of the operation. This visual and interactive simulation environment allows employees to intuitively experience the inherent logic of the operating procedures and deepen their understanding of complex systems.

[0207] Finally, in step S2140, the 3D model resources of the hydropower plant are loaded, combined with virtual devices, animation scripts and interactive logic to form a complete virtual reality scene. The 3D model collects the physical environment data of the hydropower plant through laser scanning, panoramic photography and other methods, and then combines CAD drawings and other design materials to restore a digital twin that is highly consistent with the site in the 3D engine. Professional-level rendering effects, such as realistic lighting and dynamic weather, create an immersive visual effect.

[0208] Then, the previously mapped virtual devices are embedded into the corresponding positions in the three-dimensional environment, so that they can be seamlessly connected with the surrounding buildings, terrain, pipelines, etc. Different devices can also be logically associated, such as the outlet pipe of the turbine is connected to the tailwater surge chamber, and the animation of the opening of the baffle interacts with the value of the upstream water level. This integration mechanism reproduces the topological structure and internal connection of the on-site equipment. Finally, a virtual reality system that integrates knowledge, data, and models is presented to the user. Employees can wear VR helmets and interact with virtual devices with handles or somatosensory controllers. For example, grab and turn a valve handwheel to observe the changes in valve opening and pipeline flow; click the button on the screen to activate the corresponding control command and trigger the status response of the device. During the operation, the information in the knowledge graph is called in real time to judge and guide the user's behavior, such as prompting the next step, warning of illegal operations, etc. Employees can practice the operation procedures under various working conditions in a cycle, such as equipment maintenance, emergency shutdown, etc., to improve practical operation ability and emergency response.

[0209] In summary, step S2100 uses virtual reality technology to transform the knowledge graph into an immersive experience space, and constructs a new type of human-computer interaction by concretizing abstract knowledge into visual, perceptible, and controllable virtual objects. This method integrates the principles of cognitive psychology and ergonomics, giving full play to the advantages of human perception and decision-making and the analysis and simulation advantages of computers, and realizes the intuitive transmission and rapid acquisition of knowledge, experience, and skills. The visual and interactive knowledge presentation method not only improves learning efficiency, but also enhances operational skills. It is an important innovation in the training of hydropower talents in the intelligent era. At the same time, the virtual simulation environment can also be used as a testing ground for knowledge application, incubating and verifying various new ideas and methods for optimizing operations, accelerating the transformation and implementation of results, injecting a steady stream of innovative power into hydropower enterprises, and promoting the industry to move towards high-quality development.

[0210] Step S2200, designing a VR operation training course based on the virtual reality scene of the hydropower plant;

[0211] Furthermore, if Figure 7 As shown, step S2200 includes:

[0212] Step S2210, mapping the operation nodes in the hydropower plant operation knowledge graph to the key nodes of the VR operation training course process topology structure, and connecting the key nodes through directed causal logic edges to form the process topology structure of the VR operation training course;

[0213] Step S2220, traversing each key node in the process topology structure of the VR operation training course, matching the corresponding operation event script, and generating a multi-channel interaction event of the key node;

[0214] Step S2230, taking the key node as an anchor point, extracting the operation requirements of its predecessor key node and the operation results of the subsequent key node to form the task constraint conditions of the key node;

[0215] Step S2240, using multi-channel interaction events as input and task constraints as rules, constructing a VR interaction logic decision tree to generate branch interaction scripts of key nodes;

[0216] Step S2250, converting the branch interaction script of the key node into a three-dimensional interaction sequence for the virtual device object to construct an immersive VR operation training unit;

[0217] Step S2260, according to the sequence of key nodes in the process topology structure, the VR operation training units are connected to form a complete VR operation training course.

[0218] Specifically, step S2200 aims to use virtual reality technology to transform hydropower plant operation knowledge into immersive and interactive operation training courses. Taking full advantage of virtual reality, on the one hand, the equipment layout and operating conditions of the hydropower plant can be highly restored in the virtual scene, allowing trainees to experience actual operations in person; on the other hand, it can break the limitations of physical conditions, build a rich and diverse form of human-computer interaction, and improve the fun and initiative of training. The course generation method driven by knowledge graphs can combine expert experience and data analysis to refine typical conditions and key nodes, and optimize training content and training paths.

[0219] Step S2210 plans the overall framework of the VR training course from a macro perspective. Based on the knowledge graph of hydropower plant operations, each operation node is mapped to a key step of the course, such as "starting the turbine" and "grid-connected power generation". The directed edges between nodes represent causal logic, such as "unit decoupling" must be after "reducing output", which constitutes the process constraints of the course. Through graph algorithms such as topological sorting, the nodes and edges are organized into a directed acyclic graph to form the framework structure of the course. This knowledge-driven process planning can reasonably set the operation steps and guide students to learn in depth step by step. At the same time, modularizing the course into knowledge units is also convenient for subsequent differentiated training.

[0220] Take a training on the start and stop operation of a certain unit as an example. According to the knowledge graph, a series of key operation nodes are extracted, such as "start auxiliary equipment → switch to local control mode → guide vane opening control → stator excitation → grid-connected unit → increase active power → monitor operating parameters → reduce active power → unit decoupling → close main valve → stop auxiliary equipment". These nodes are connected in chronological order to form the main process of the course. Each node in the process is a relatively independent training unit, which can be practiced repeatedly according to the trainees' mastery. The process topology clearly shows the dependency constraints between each operation step, providing a framework for the design of VR interaction.

[0221] Step S2220 further designs the human-computer interaction details of each key node. Based on the semantic attributes in the operation knowledge graph, the corresponding operation scripts are matched for the nodes and converted into multi-channel interactions in the VR environment. For example, the "start the turbine" node requires trainees to complete multiple operations such as "start the main valve control loop", "open the main valve", "observe the tachometer", and "report the unit status". Voice recognition technology can be used to capture trainees' verbal instructions for controlling virtual devices; body tracking technology can be used to perceive trainees' gestures for manipulating the dashboard; and eye movement monitoring technology can be used to analyze trainees' line of sight for guidance on equipment parameters. Multi-channel interaction makes full use of the immersiveness of VR, allowing trainees to devote themselves to the operation and achieve the effect of "learning by doing".

[0222] Take "turbine guide vane opening control" as an example. Trainees can use the voice command "set the guide vane opening to 80%" and observe the feedback of the virtual voice assistant to see if it is correctly recognized; they can wear data gloves and adjust the guide vane switch on the virtual control panel to feel the feel and force feedback; they can look at the key parameters of the virtual machine group such as speed and vibration, and the system automatically compares the parameter value with the standard range and marks the deviation with different colors; they can use their hands to gesture the spatial position of key components, and the virtual brush automatically marks the component name. The integrated application of multi-channel interaction enables trainees to experience the real control process in multiple sensory dimensions such as vision, hearing, and touch, deepen muscle memory, and improve practical operation ability. It is worth mentioning that voice interaction breaks the limitations of manual operation, and gesture interaction breaks the constraints of the two-dimensional interface, greatly expanding the imagination space of human-computer interaction.

[0223] Step S2230 refines the task constraints of the key node while interactively designing. These constraints come from the attribute information and timing rules in the operation knowledge graph. For example, the "guide vane opening" operation has a precondition of "switch to local control mode", and manual adjustment is allowed only in local mode; its postcondition is "stator excitation", and the stator can only be powered after the guide vane opening reaches the rated value. Timing constraints are imposed on the front and rear nodes to limit the time window in which the node can be operated. In addition, the numerical attributes of the node itself also form constraints, such as the guide vane opening must be controlled between 60% and 80%, and any deviation from the upper and lower limits is considered unqualified. Spatiotemporal constraints and numerical constraints jointly define the legal space of an operation task and play a judging role in training. Only when the trainee's operation sequence meets all constraints can the task be successfully completed and the next training unit can be entered.

[0224] Let’s go back to the example of guide vane control. In the VR scene, facing a realistic control panel, if the trainee tries to manually adjust the opening before entering the local mode, he will be prompted “Please switch the control mode first!” and the local control button will light up; if the opening is adjusted too quickly or too high, it will warn “The guide vane opening is out of limit! Please control it within the safe range” and mark the danger zone in red; if the training ends after the adjustment is completed, it will remind “Please follow the process and continue stator excitation”. By embedding task constraints, trainees can get targeted guidance and feedback in practice in real time, which helps to cultivate their initiative and awareness of norms. At the same time, task constraints are also an important basis for VR program control logic, which can guide trainees to understand the inherent logic of standard operations while ensuring safety.

[0225] Step S2240 integrates the above multi-channel interactions and task constraints into a decision tree model. Using interaction events as input and task constraints as rules, a branch interaction script of the node is generated. Each node expands multiple branch options according to possible operation actions, such as voice commands, gestures, gaze, etc., to form the middle node of the tree; the legality of the operation is judged in combination with the task constraints. If the constraints are met, it is a legal branch and enters the subsequent process; if the constraints are violated, it is an illegal branch and enters the corresponding prompt scene. In this way, an operation node generates an interactive decision tree, the leaves represent the final operation results, and the path from the root node to the leaf node corresponds to an operation mode of the trainee.

[0226] Let's take the guide vane control as an example. The starting state is "remote control mode" and the target state is "guide vane opening 80%". The actions that the trainee may take include: issuing voice commands, rotating the local control switch, turning the guide vane joystick, observing the tachometer, etc. The decision tree expands branches with these operations as intermediate nodes: if the control switch is rotated first and the action is correct, the local control mode is entered, the pre-constraint is met, and the search can continue along this edge of the tree; if the joystick is turned directly, it will prompt "Please switch the control mode first", and this edge of the tree will be pruned. In the local mode, the trainee continues to turn the joystick. If the guide vane opening is controlled at 60% to 80% and finally stabilizes at 80%, it is a legal path and this leaf node is retained; if the opening is too high or does not reach the target value, it violates the numerical constraint and is also pruned.

[0227] After such iterative decision-making, each node generates multiple acceptable operation sequences, which constitute the interactive script of the node. On the one hand, the script contains the key operations in the knowledge graph, which can guide the implementation of the VR program; on the other hand, different branches reflect different operation strategies, which can be used to distinguish the learning levels of trainees. Those trainees who can make the best decision at one time usually master the essentials of the operation; while those trainees who try repeatedly and take detours need to be focused on and guided. Therefore, the interactive decision tree is a representation model that integrates expert knowledge and system logic, which can provide an explainable and optimizable behavior mode for VR training.

[0228] Step S2250 further converts the interactive script in the form of a decision tree into a three-dimensional interactive sequence executable by the VR engine. With key nodes as anchors and combined with the spatial information in the node attributes, corresponding equipment models such as units and control panels are generated in the virtual scene. Each device is embedded with an operation interface and response logic, which is mapped to the input events in the decision tree. The multi-channel interaction in the script corresponds to the interactive components of the device, such as voice controls, gesture sensing areas, and line of sight tracking points. Through these interfaces, trainees can interact naturally with virtual devices. At the same time, task constraints are attached to the parameter items and components of the equipment model to evaluate the trainees' operation process in real time. For example, the tachometer and vibration meter are displayed on the virtual control panel. Once the parameters are exceeded, the corresponding instrument will light up red.

[0229] After the above steps, a key node is built into an immersive VR operation training unit. Take the "turbine guide vane opening control unit" as an example. The trainee puts on a VR helmet and is placed in a realistic unit plant. In front of him is a 720-degree surround control console covered with various instruments and switches. Through voice or gestures, the trainee can switch the current control mode, and the system gives real-time voice feedback and interface changes. Then, the trainee starts to turn the guide vane joystick, the virtual unit on the screen starts to run, and the tachometer displays the speed change in real time. The trainee needs to control the speed and amplitude of the adjustment, so that the unit can reach the rated speed as soon as possible, and avoid overshoot or under-adjustment. At the same time, the trainee should judge the health of the unit by observing other auxiliary parameters, such as main shaft vibration, bearing temperature, etc. Once an abnormality is found, timely measures should be taken, such as stopping guide vane adjustment and reducing unit load. Throughout the process, the trainee must strictly abide by the operating procedures and operate according to the standard process, otherwise the system will correct it through voice prompts, pop-up warnings, etc. After a period of training, trainees can master the key points of guide vane opening control and can flexibly adjust it according to different working conditions.

[0230] Step S2260 connects each VR training unit into a complete operation training course. Traverse the key nodes in the process topology structure, and determine the order of each unit in combination with the timing dependency of the nodes. For linear processes, just link them in sequence according to the causal logic; for processes containing complex structures such as parallel and loop structures, it is also necessary to establish jump logic and branch conditions. For example, "starting the turbine" and "starting the generator" can be carried out in parallel, and then merged into "grid-connected power generation" after completion; if the "monitoring operating parameters" does not meet the requirements, the previous "guide vane opening control" needs to be executed cyclically. Building a complete course process can prevent students from missing key training steps, and at the same time, it can also teach students in accordance with their aptitude and customize targeted training plans according to the characteristics of different students.

[0231] Applying the above steps to actual cases can generate a complete set of VR operation training courses for hydropower plants. The course covers key operations in various stages such as startup, grid connection, operation, and shutdown. Through rich human-computer interaction methods, trainees can conduct immersive training in simulation scenarios. For example, in the "Starting the Turbine" chapter, the trainee first calls the operation duty by voice to understand the status of the unit; then manually opens the water inlet valve, adjusts the guide vane opening, and visually controls the speed of the unit to rise steadily; at the same time, pay attention to key parameters such as head and vibration to judge the working condition of the unit at any time. Once a fault or limit is exceeded, the trainee must quickly take emergency measures, such as closing the main valve and stopping the guide vane. Through timely human-computer interaction, the VR system can assess whether each step of the operation is standardized and in place, and score the trainee's operation level accordingly. The units are connected in the order of the flowchart, so that the trainee can always grasp the current operation stage. At the same time, multiple branch options are set between different units, and trainees can choose different training paths according to their own situation. For example, after "starting the turbine", you can directly enter "grid-connected power generation" or choose "shutdown operation". Each path corresponds to a typical operating condition, and by teaching students in accordance with their aptitude, the training of each trainee is more targeted.

[0232] The above-mentioned VR course design method driven by knowledge graph is applied to the whole process of hydropower plant operation training. It can integrate multiple technologies such as domain knowledge, data analysis, and human-computer interaction to form a set of system solutions from macro to micro, from modeling to implementation. Macroscopically, with knowledge graph as the outline, the course framework is constructed by summarizing and refining from massive historical data; microscopically, with interactive decision tree as the goal, knowledge constraints are embedded in VR units to achieve immersive operation. In terms of modeling, the integration of multiple knowledge representations such as ontology, graph, and tree not only utilizes the structure of knowledge, but also takes into account the characteristics of VR scenes; in terms of implementation, with three-dimensional interactive sequence as the carrier, abstract symbolic knowledge is mapped to concrete virtual objects, giving knowledge a vivid interactive form. This course generation model can significantly improve the intelligence level of VR training, make the training content closer to actual operation, and make the training process more in line with cognitive laws. On the one hand, the expert experience is solidified into reusable training knowledge, which reduces the workload of manual design and improves the efficiency of course development; on the other hand, historical data is summarized into quantifiable training goals, making the training process traceable and evaluable, thereby improving the quality of training. In addition, relying on the immersive interactive advantages of VR, trainees can fully practice in a simulation environment, strengthen muscle memory, and gain an immersive operating experience, thereby deepening their understanding of theoretical knowledge.

[0233] In short, with knowledge graph as the core, a bridge of virtual-real mapping is built through mechanism modeling and data analysis; VR simulation is used as a means to achieve human-machine collaborative operation training through immersive interaction and experiential learning. On the one hand, this exploration confirms the synergy of knowledge engineering, artificial intelligence, virtual reality and other technologies, and on the other hand, it also provides new ideas for the subsequent development of intelligent training systems. It can be foreseen that with the continuous improvement of knowledge graph construction tools and the increasing maturity of VR software and hardware technology, this integrated and innovative training model will be promoted and applied in more fields, empowering traditional industrial training and exploring new paths for the localization of industrial software.

[0234] Step S2300: Conduct immersive VR operation training based on the VR operation training course.

[0235] Specifically, step S2300 realizes immersive operation training through VR operation training courses. Immersive VR training uses virtual reality technology to build realistic three-dimensional scenes. Trainees wear VR helmets and hold interactive devices to "walk into" the virtual power plant from a first-person perspective to gain an immersive control experience. Compared with traditional training, VR training overcomes the limitations of on-site conditions, and repeated drills can be carried out at any time without worrying about the risk of misoperation, so as to accumulate experience for new employees.

[0236] The VR training system is based on the knowledge graph and integrates knowledge points into virtual scenes. By mapping the operation nodes to interactive tasks in VR, such as virtual buttons and knobs, trainees can simulate various control actions. Through the constraints and prompts of correlation and causal edges in the graph, trainees are guided to follow the recommended path. For example, correlation checks and logical judgments are embedded in the operation chain of "starting → grid connection". Only when the related parameters such as "main valve opening, stator current" meet the requirements can the next step be entered. This training method that integrates knowledge into the context in the form of tasks not only tests practical skills, but also strengthens procedural awareness.

[0237] In addition, the VR system can also present realistic equipment operation screens and environmental special effects. Using game engine rendering technology, it generates microscopic components such as valves and instrument panels, and macroscopic panoramic views of factory buildings and rivers, to construct a virtual and real space that is both real and illusory. At the same time, it simulates interference from environmental factors such as rain, fog, and alarms to enhance students' sense of presence in dealing with emergencies. Especially for some extreme working conditions that endanger personal safety or are difficult to simulate, such as equipment fires and floods, VR can restore a thrilling scene and test students' emergency response. In VR, the negative effects of erroneous operations can be magnified, such as VR special effects of "equipment damage" caused by erroneous operations, to deepen students' sense of responsibility.

[0238] VR training can record the trainee's operation process throughout the whole process, compare it with the recommended operations in the knowledge graph, calculate the similarity score in real time, and generate an assessment report. Through data analysis, weak links can be found, and training can be strengthened in a targeted manner to achieve teaching in accordance with students' aptitude. The timing logic in the graph can also be used to construct standard operating conditions, set various evaluation dimensions such as standardization, timeliness, economy, safety, etc., and evaluate trainees' abilities from multiple angles.

[0239] With the help of VR training platform, cross-scenario and cross-equipment knowledge generalization can be achieved. For different types of units and plant layouts, you only need to replace the 3D model and adjust the knowledge graph parameters to customize a dedicated training system. After completing the training of a scenario, students can move to similar working conditions and understand the common rules. Different companies and colleges can also share high-quality VR resources, spread industry methods, and improve the overall level of practice.

[0240] In short, VR training improves trainees' ability to grasp hydropower operation knowledge from two aspects: perceptual experience and rational cognition. Immersive VR reconstructs the traditional human-computer interaction and teaching mode, integrates situational perception, practical exercises, and evaluation feedback, and gives birth to a new way of immersive learning. In the VR scene, knowledge is transformed into tasks and situational clues, and education is entertaining, abstract symbols are transformed into perceptual experience, and knowledge internalization is deepened. At the same time, VR's low-cost and reusable advantages free training from physical limitations, focusing on core skills, comprehensive assessment, teaching students in accordance with their aptitude, and focusing on the cultivation of transferability. The combination of VR with cutting-edge technologies such as artificial intelligence and big data analysis makes it a powerful tool for empowering employees and shaping organizational learning culture. While improving quality and efficiency, it reshapes the training values ​​and promotes the transformation from "I have to learn" to "I want to learn". It can be foreseen that with the advancement of technology, VR training will be promoted in a wider range of fields and become the mainstream of future training.

[0241] Example 2

[0242] This embodiment provides a hydropower plant operation training system based on virtual reality technology on the basis of embodiment 1. Figure 8 As shown, including:

[0243] Knowledge graph construction module: used to obtain the historical operation time series data of the hydropower plant, divide the historical operation time series data into n1 discrete operation fragments, and obtain the operation fragment time series data; based on the operation fragments and the operation fragment time series data, construct the hydropower plant operation knowledge graph;

[0244] Virtual reality scene construction module: used to map the hydropower plant operation knowledge graph into a three-dimensional virtual space and construct a virtual reality scene of the hydropower plant;

[0245] VR training module: Based on the virtual reality scene of the hydropower plant, design VR operation training courses, and carry out immersive VR operation training based on the VR operation training courses.

[0246] In the knowledge graph construction module, each operation fragment corresponds to an operation event; the operation fragment timing data includes the start and end timestamps of the operation fragment, the device location coordinates involved in the operation fragment, the name of the operation event to which the operation fragment belongs, and the device operation parameters within the operation fragment.

[0247] In the knowledge graph construction module, the construction of the hydropower plant operation knowledge graph based on the operation fragments and the operation fragment time series data includes:

[0248] Step S1210, abstracting each operation fragment into an operation node of the hydropower plant operation knowledge graph, and the name of the operation node is the operation event to which the operation fragment belongs;

[0249] Step S1220, acquiring attribute information of the operation node from the operation segment time series data, wherein the attribute information of the operation node includes time attribute, space attribute, semantic attribute and numerical attribute;

[0250] Step S1230, discovering frequent operation patterns from the operation segment time series data, and constructing a time series logic rule reflecting the operation sequence;

[0251] Step S1240, analyzing the correlation between different operation nodes according to the attribute information of the operation nodes; constructing directed causal logic edges in combination with the correlation between the different operation nodes and the temporal logic rules; organizing the operation nodes and directed causal logic edges into a semantic network to form a knowledge graph of hydropower plant operations.

[0252] The step S1220 includes:

[0253] Step S1221, extracting the start and end timestamps of the operation segments in the operation segment timing data as the time attributes of the corresponding operation nodes;

[0254] Step S1222, extracting the device position coordinates involved in the operation fragment in the operation fragment time series data as the spatial attributes of the corresponding operation node;

[0255] Step S1223, extracting the name of the operation event to which the operation fragment belongs from the operation fragment time series data as the semantic attribute of the corresponding operation node;

[0256] Step S1224, calculating the statistical characteristics of the equipment operating parameters in the operation segment in the operation segment time series data as the numerical attributes of the corresponding operation node.

[0257] The step S1230 includes:

[0258] Step S1231, performing frequent pattern mining on the operation fragment time series data, summarizing n2 operation patterns from the operation fragment time series data, and calculating the support of each operation pattern; n2 is a positive integer;

[0259] Step S1232, setting a support threshold, traversing the support of n2 operation modes, and defining an operation mode whose support exceeds the support threshold as a frequent operation mode;

[0260] Step S1233, performing confidence evaluation on the frequent operation modes, removing operation modes with confidences lower than a preset confidence threshold from the frequent operation modes, and obtaining candidate operation modes;

[0261] Step S1234, representing the candidate operation mode as a sequential logic rule, wherein the left part of the sequential logic rule is the preceding operation, the right part is the subsequent operation, and the arrow in the middle represents the sequential order.

[0262] The step S1240 includes:

[0263] Step S1241, constructing operation node pairs, calculating the comprehensive correlation coefficients of the operation node pairs according to the attribute information of the operation nodes, and constructing a comprehensive correlation coefficient matrix of all operation node pairs; traversing the comprehensive correlation coefficient matrix, marking the operation node pairs corresponding to the elements in the comprehensive correlation coefficient matrix that are greater than a preset correlation threshold as relevant operation nodes;

[0264] Step S1242, using temporal logic rules to annotate causal relationships on relevant operation nodes and construct directed causal logic edges;

[0265] Step S1243, taking the operation nodes as entities and the directed causal logic edges as entity relationships, a knowledge graph of hydropower plant operations is formed.

[0266] The step S1241 includes:

[0267] Step S12411, calculating the time correlation coefficient of the operation node pair according to the time attribute of the operation node;

[0268] Step S12412, calculating the spatial correlation coefficient of the operation node pair according to the spatial attributes of the operation nodes;

[0269] Step S12413, calculating the semantic correlation coefficient of the operation node pair according to the semantic attributes of the operation nodes;

[0270] Step S12414, calculating the numerical correlation coefficient of the operation node pair according to the numerical attributes of the operation nodes;

[0271] Step S12415, weighted fusion of the temporal correlation coefficient, spatial correlation coefficient, semantic correlation coefficient and numerical correlation coefficient of the operation node pair is performed to obtain a comprehensive correlation coefficient of the operation node pair.

[0272] In the virtual reality scene construction module, the construction of the hydropower plant virtual reality scene includes:

[0273] Step S2110, mapping the operation nodes in the hydropower plant operation knowledge graph to virtual equipment objects in the hydropower plant virtual reality scene, and arranging the position coordinates of the virtual equipment objects according to the spatial attributes of the operation nodes;

[0274] Step S2120, mapping the semantic attributes of the operation nodes in the hydropower plant operation knowledge graph into the operation event scripts of the virtual device objects;

[0275] Step S2130, mapping the directed causal logic edges in the hydropower plant operation knowledge graph into interactive logical relationships between virtual device objects;

[0276] Step S2140, loading the three-dimensional model resources of the hydropower plant, combining the virtual device objects, operation event scripts and interactive logical relationships, to form a completed virtual reality scene of the hydropower plant.

[0277] In the VR training module, the VR operation training course includes:

[0278] Step S2210, mapping the operation nodes in the hydropower plant operation knowledge graph to the key nodes of the VR operation training course process topology structure, and connecting the key nodes through directed causal logic edges to form the process topology structure of the VR operation training course;

[0279] Step S2220, traversing each key node in the process topology structure of the VR operation training course, matching the corresponding operation event script, and generating a multi-channel interaction event of the key node;

[0280] Step S2230, taking the key node as an anchor point, extracting the operation requirements of its predecessor key node and the operation results of the subsequent key node to form the task constraint conditions of the key node;

[0281] Step S2240, using multi-channel interaction events as input and task constraints as rules, constructing a VR interaction logic decision tree to generate branch interaction scripts of key nodes;

[0282] Step S2250, converting the branch interaction script of the key node into a three-dimensional interaction sequence for the virtual device object to construct an immersive VR operation training unit;

[0283] Step S2260, according to the sequence of key nodes in the process topology structure, the VR operation training units are connected to form a complete VR operation training course.

[0284] Example 3

[0285] This embodiment discloses an electronic device, which may include one or more processors and one or more memories, wherein the memories store computer-readable codes, and when the computer-readable codes are executed by the one or more processors, the hydropower plant operation training method based on virtual reality technology as described above may be executed.

[0286] The method or system according to the embodiment of the present application can also be implemented with the help of the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in an electronic device, such as a ROM or a hard disk, can store the hydropower plant operation training method based on virtual reality technology provided in this application. The hydropower plant operation training method based on virtual reality technology may, for example, include: obtaining the historical operation time series data of the hydropower plant, dividing the historical operation time series data into n1 discrete operation segments, and obtaining the operation segment time series data; based on the operation segments and the operation segment time series data, constructing a hydropower plant operation knowledge graph; mapping the hydropower plant operation knowledge graph to a three-dimensional virtual space, constructing a hydropower plant virtual reality scene, designing a VR operation training course based on the hydropower plant virtual reality scene, and conducting immersive VR operation training based on the VR operation training course.

[0287] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in the present invention is only exemplary, and when implementing different devices, one or more components in the electronic device disclosed in the present invention may be omitted according to actual needs.

[0288] Example 4

[0289] This embodiment discloses a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the hydropower plant operation training method based on virtual reality technology in the embodiment of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0290] In addition, according to the implementation of the present application, the process described in the above reference flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: obtaining the historical operation time series data of the hydropower plant, dividing the historical operation time series data into n1 discrete operation segments, and obtaining the operation segment time series data; based on the operation segments and the operation segment time series data, constructing the hydropower plant operation knowledge graph; mapping the hydropower plant operation knowledge graph to a three-dimensional virtual space, constructing a hydropower plant virtual reality scene, designing a VR operation training course based on the hydropower plant virtual reality scene, and conducting immersive VR operation training based on the VR operation training course. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0291] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0292] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0293] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A hydropower plant operation training method based on virtual reality technology, characterized in that: The method comprises: Obtain the historical operation time series data of the hydropower plant, divide the historical operation time series data into n1 discrete operation segments, and obtain the operation segment time series data; construct the hydropower plant operation knowledge graph based on the operation segments and the operation segment time series data; n1 is a positive integer; Map the hydropower plant operation knowledge graph to the three-dimensional virtual space, build a hydropower plant virtual reality scene, design VR operation training courses based on the hydropower plant virtual reality scene, and carry out immersive VR operation training based on the VR operation training courses.

2. The hydropower plant operation training method based on virtual reality technology according to claim 1 is characterized in that: Each of the operation segments corresponds to an operation event; the operation segment timing data includes the start and end timestamps of the operation segment, the device location coordinates involved in the operation segment, the name of the operation event to which the operation segment belongs, and the device operating parameters within the operation segment.

3. The hydropower plant operation training method based on virtual reality technology according to claim 2 is characterized in that: The construction of the hydropower plant operation knowledge graph includes: Each operation fragment is abstracted into an operation node of the hydropower plant operation knowledge graph, and the name of the operation node is the operation event to which the operation fragment belongs; Acquire attribute information of an operation node from the operation segment time series data, wherein the attribute information of the operation node includes time attribute, space attribute, semantic attribute and numerical attribute; Discover frequent operation patterns from the time series data of operation fragments and construct time series logic rules that reflect the order of operations; According to the attribute information of the operation nodes, the correlation between different operation nodes is analyzed; combined with the correlation between the different operation nodes and the temporal logic rules, directed causal logic edges are constructed; the operation nodes and directed causal logic edges are organized into a semantic network to form a knowledge graph of hydropower plant operations.

4. The method for hydropower plant operation training based on virtual reality technology according to claim 3 is characterized in that: The step of acquiring attribute information of the operation node from the operation fragment time series data includes: Extract the start and end timestamps of the operation fragment in the operation fragment time series data as the time attribute of the corresponding operation node; Extracting the device position coordinates involved in the operation fragment from the operation fragment time series data as the spatial attributes of the corresponding operation node; Extracting the name of the operation event to which the operation fragment belongs from the operation fragment time series data as the semantic attribute of the corresponding operation node; The statistical characteristics of the equipment operating parameters in the operation fragments in the operation fragment time series data are calculated as the numerical attributes of the corresponding operation nodes.

5. The method for hydropower plant operation training based on virtual reality technology according to claim 3 is characterized in that: The method of discovering frequent operation patterns from the operation fragment time series data and constructing time series logic rules reflecting the operation sequence includes: Perform frequent pattern mining on the operation fragment time series data, summarize n2 operation modes from the operation fragment time series data, and calculate the support of each operation mode; Set a support threshold, traverse the support of n2 operation modes, and define the operation mode whose support exceeds the support threshold as a frequent operation mode; Perform confidence evaluation on the frequent operation modes, remove the operation modes whose confidence is lower than a preset confidence threshold from the frequent operation modes, and obtain candidate operation modes; The candidate operation mode is represented as a sequential logic rule, the left part of the sequential logic rule is the preceding operation, the right part is the subsequent operation, and the arrow in the middle represents the sequential order.

6. The method for hydropower plant operation training based on virtual reality technology according to claim 3 is characterized in that: The analyzing the correlation between different operation nodes according to the attribute information of the operation nodes includes: constructing operation node pairs, calculating the comprehensive correlation coefficients of the operation node pairs according to the attribute information of the operation nodes, and constructing a comprehensive correlation coefficient matrix of all operation node pairs; traversing the comprehensive correlation coefficient matrix, marking the operation node pairs corresponding to the elements in the comprehensive correlation coefficient matrix that are greater than a preset correlation threshold as relevant operation nodes; The method for constructing directed causal logic edges is to use the temporal logic rules to label the causal relationships of relevant operation nodes; The semantic network takes operation nodes as entities and directed causal logic edges as entity relationships.

7. The method for hydropower plant operation training based on virtual reality technology according to claim 6 is characterized in that: The comprehensive correlation coefficient of the computing operation node pair includes: According to the time attributes of the operation nodes, the time correlation coefficient of the operation node pairs is calculated; According to the spatial attributes of the operation nodes, the spatial correlation coefficient of the operation node pairs is calculated; According to the semantic attributes of the operation nodes, the semantic correlation coefficient of the operation node pairs is calculated; According to the numerical attributes of the operation nodes, the numerical correlation coefficient of the operation node pairs is calculated; The temporal correlation coefficient, spatial correlation coefficient, semantic correlation coefficient and numerical correlation coefficient of the operation node pair are weighted and fused to obtain the comprehensive correlation coefficient of the operation node pair.

8. The method for hydropower plant operation training based on virtual reality technology according to claim 7 is characterized in that: The time correlation coefficient of the computing operation node pair includes: The start timestamp and the end timestamp of the two operation nodes in the operation node pair are extracted, the overlap score and the interval decay coefficient of the operation node pair are calculated, and the overlap score and the interval decay coefficient are multiplied to obtain the time correlation coefficient of the operation node pair.

9. The method for hydropower plant operation training based on virtual reality technology according to claim 6 is characterized in that: The construction of the hydropower plant virtual reality scene includes: Map the operation nodes in the hydropower plant operation knowledge graph to virtual equipment objects in the hydropower plant virtual reality scene, and arrange the position coordinates of the virtual equipment objects according to the spatial attributes of the operation nodes; Mapping the semantic attributes of the operation nodes in the hydropower plant operation knowledge graph into the operation event scripts of the virtual device objects; Map the directed causal logic edges in the hydropower plant operation knowledge graph into interactive logical relationships between virtual device objects; The three-dimensional model resources of the hydropower plant are loaded, and the virtual equipment objects, operation event scripts and interactive logical relationships are combined to form a completed virtual reality scene of the hydropower plant.

10. A hydropower plant operation training system based on virtual reality technology, which is used to implement the hydropower plant operation training method based on virtual reality technology as claimed in any one of claims 1 to 9, characterized in that: The system comprises: Knowledge graph construction module: used to obtain the historical operation time series data of the hydropower plant, divide the historical operation time series data into n1 discrete operation fragments, and obtain the operation fragment time series data; based on the operation fragments and the operation fragment time series data, construct the hydropower plant operation knowledge graph; Virtual reality scene construction module: used to map the hydropower plant operation knowledge graph into a three-dimensional virtual space and construct a virtual reality scene of the hydropower plant; VR training module: Based on the virtual reality scene of the hydropower plant, design VR operation training courses, and carry out immersive VR operation training based on the VR operation training courses.

Images