Power system digital twin model generation method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a power system digital twin model generation method and device, computer equipment, a readable storage medium and a program product. The method comprises the steps of obtaining multi-source heterogeneous data of a transformer substation, preprocessing the multi-source heterogeneous data to obtain preprocessed data, performing fuzzification processing based on the preprocessed data, extracting a fuzzy logic rule, constructing a to-be-trained adversarial network model, and training the to-be-trained adversarial network model based on the fuzzy logic rule. And when it is determined that the model verification conditions are met, training is stopped, and the digital twin model of the power system is obtained. In the modeling process, a model combining fuzzy logic and a generative adversarial network is introduced, and through dynamic rule base management and self-adaptive updating, it is ensured that the system can adapt to the complex operation environment of a power system in real time, so that the accuracy of automatic generation of the substation operation order is greatly improved, and the automation degree of the substation operation order is improved. And meanwhile, the rule compliance and the compatibility of the power system standard are ensured.

Description

Technical Field

[0001] The present application relates to the technical field of substation automation, and particularly to a method, device, computer device, readable storage medium and program product for generating a digital twin model of a power system. Background Technique

[0002] In recent years, with the continuous development of the power system and the increasing expansion of the power grid scale, the power system substation, as the core node of the power grid, undertakes key functions such as voltage conversion, power distribution and system protection. With the rapid development of the smart grid, the substation automation system needs to monitor the device status in real time, dynamically adjust the operation strategy, and quickly respond to power grid faults, which puts higher requirements on the accuracy and real-time performance of the modeling technology.

[0003] Traditional substation modeling methods mainly rely on manual experience and static rule libraries, and construct system models by manually configuring Substation Configuration Description (SCD) files or monitoring information tables. However, manual modeling is inefficient and difficult to meet the rapid deployment requirements of large-scale complex devices. Moreover, the static rule library cannot effectively handle the uncertainty of device status, resulting in a large deviation between the model and the actual operation, affecting the system control accuracy and reliability. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, computer device, computer-readable storage medium and computer program product for generating a digital twin model of a power system that can efficiently model for the above technical problems.

[0005] In a first aspect, the present application provides a method for generating a digital twin model of a power system, the method comprising:

[0006] Obtain multi-source heterogeneous data of the substation, and preprocess the multi-source heterogeneous data to obtain preprocessed data; the multi-source heterogeneous data includes structured data, unstructured data and time-series data;

[0007] Perform fuzzification processing based on the preprocessed data, and extract fuzzy logic rules;

[0008] Construct an adversarial network model to be trained, and train the adversarial network model to be trained based on the fuzzy logic rules;

[0009] When it is determined that the model verification condition is met, stop training to obtain a digital twin model of the power system, and the digital twin model is used for health monitoring and dynamic strategy adjustment of power system devices.

[0010] In one embodiment, the preprocessing of the multi-source heterogeneous data to obtain preprocessed data includes: performing format standardization processing on the multi-source heterogeneous data to construct a multi-dimensional data set with spatio-temporal alignment; performing data cleaning and outlier processing on the multi-dimensional data set to obtain a multi-dimensional data set after cleaning; performing data annotation on the multi-dimensional data set after cleaning based on predefined annotation rules; performing semantic parsing and fusion on the annotated multi-dimensional data set to obtain corresponding semantic information, and using the semantic information as the preprocessed data.

[0011] In one embodiment, after the preprocessing of the multi-source heterogeneous data to obtain preprocessed data, the method further includes: verifying the data quality of the preprocessed data through preset quantization metrics to obtain verified data, where the quantization metrics include data integrity metrics, consistency metrics, timeliness metrics, and correctness metrics.

[0012] In one embodiment, the fuzzification processing based on the preprocessed data to extract fuzzy logic rules includes: obtaining fuzzy descriptions in the preprocessed data, and defining membership functions of key parameters in the fuzzy descriptions based on power system operation standards; using dependency syntactic analysis and semantic role annotation techniques to identify condition-action pairs from the fuzzy descriptions to extract fuzzy logic rules.

[0013] In one embodiment, the construction of the adversarial network model to be trained includes: constructing an adversarial network model to be trained based on a generator and a discriminator; the generator adopts an encoder-decoder structure and is used to generate device behavior data by embedding the fuzzy logic rule constraints; the discriminator is used to perform multi-modal verification on the generated device behavior data.

[0014] In one embodiment, the training of the adversarial network model to be trained based on the fuzzy logic rules includes: based on forward propagation, inputting a fuzzy condition vector to the generator to enable the generator to generate device behavior data; evaluating the adversarial loss between the generated device behavior data and the real data based on the discriminator, and determining the rule satisfaction loss of the device behavior data through a fuzzy rule engine; if the device behavior data triggers a rule conflict, extracting conflict rule features and updating the rule constraint weight matrix of the generator; dynamically adjusting the learning rates of the generator and the discriminator according to the convergence of the rule satisfaction loss and the adversarial loss to train the adversarial network model to be trained.

[0015] In a second aspect, the present application further provides a device for generating a digital twin model of a power system, and the device includes:

[0016] A data acquisition module, configured to acquire multi-source heterogeneous data of a substation, preprocess the multi-source heterogeneous data to obtain preprocessed data; the multi-source heterogeneous data includes structured data, unstructured data, and time-series data;

[0017] A fuzzy processing module, configured to perform fuzzy processing based on the preprocessed data and extract fuzzy logic rules;

[0018] A model training module, configured to construct an adversarial network model to be trained and train the adversarial network model to be trained based on the fuzzy logic rules;

[0019] A model generation module, configured to stop training when it is determined that the model verification condition is satisfied, and obtain a digital twin model of the power system, where the digital twin model is used for health monitoring and dynamic policy adjustment of power system equipment.

[0020] In a third aspect, the present application further provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0021] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0022] In a fifth aspect, the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0023] The above method, device, computer device, computer-readable storage medium, and computer program product for generating a digital twin model of a power system obtain multi-source heterogeneous data of a substation, preprocess the multi-source heterogeneous data to obtain preprocessed data, perform fuzzy processing based on the preprocessed data, extract fuzzy logic rules, construct an adversarial network model to be trained, train the adversarial network model to be trained based on the fuzzy logic rules, and stop training when it is determined that the model verification condition is satisfied to obtain a digital twin model of the power system. In the modeling process, a model combining fuzzy logic and a generative adversarial network is introduced, and through dynamic rule library management and adaptive update, it is ensured that the system can adapt to the complex environment of power system operation in real time, thereby greatly improving the accuracy of automatic generation of substation operation tickets, while ensuring rule compliance and compatibility with power system standards. An automatic model generation technology strategy based on artificial intelligence fuzzy recognition is realized, providing an efficient and intelligent solution for the substation automation system. Description of the Drawings

[0024] To more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or the related art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0025] Figure 1 It is a schematic flowchart of a method for generating a digital twin model of a power system in an embodiment;

[0026] Figure 2 It is a schematic flowchart of a data preprocessing step in an embodiment;

[0027] Figure 3 It is a schematic flowchart of a fuzzification processing step in an embodiment;

[0028] Figure 4 It is a schematic flowchart of a method for generating a digital twin model of a power system in another embodiment;

[0029] Figure 5 It is a structural block diagram of a device for generating a digital twin model of a power system in an embodiment;

[0030] Figure 6 It is an internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0031] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the following further details the present application in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0032] Due to the low efficiency of traditional manual modeling, it is difficult to meet the rapid deployment requirements of large-scale complex equipment. And the static rule base cannot effectively handle the uncertainty of equipment states, such as voltage fluctuations and fuzzy descriptions of loads, which will lead to a large deviation between the model and the actual operation. Also, due to the high proportion of renewable energy access and the increase in power electronic devices, the complexity of system dynamic behavior has increased sharply, and traditional modeling techniques are difficult to meet the requirements of real-time simulation and prediction, affecting the fault response speed and system stability.

[0033] In recent years, digital twin technology has provided a new solution for substation automation modeling, which realizes real-time monitoring of equipment status and behavior prediction through virtual-real mapping. Traditional research is mostly based on physical mechanism modeling or data-driven methods, but there are still limitations. Physical mechanism models rely on accurate mathematical equations and are difficult to describe fuzzy logic, such as "voltage is too high" and "load is critical"; while pure data-driven methods are limited by the scarcity of high-quality labeled data, and the generated models lack compatibility with power system standards, such as meeting the IEC 61850 standard. In addition, traditional digital twin modeling tools usually adopt a centralized architecture and rely on manual configuration of the rule base, resulting in lagging model updates and inability to adapt to the dynamic changes of grid equipment and the requirements of edge scenarios.

[0034] To address the above problems, some research has attempted to combine fuzzy logic with generative adversarial networks to improve the model's ability to handle uncertainty. For example, the generation process is constrained by a fuzzy rule base, or generative adversarial networks are used to generate simulated data to expand the training samples. However, traditional methods still have the following deficiencies: First, the fuzzy rule base is mostly statically designed and lacks an adaptive update mechanism, making it difficult to cope with equipment upgrades or regulation revisions; Second, the compatibility between the generated model and power system standards is insufficient, and the generated SCD files or operation tickets need to be manually adjusted, resulting in limited automation; Third, there is a lack of a multi-modal verification mechanism during the model training process, and the generated results may have logical conflicts or security hazards.

[0035] Based on this, this application provides a method for generating a digital twin model of a power system, which deeply integrates fuzzy logic and adversarial generation technology. Through the construction of a dynamic fuzzy rule base, the design of a multi-modal discriminator, and the encapsulation of a standardized interface, the end-to-end automation generation from multi-source heterogeneous data to a standardized digital twin model is realized. This method not only significantly improves the modeling efficiency and accuracy, but also ensures that the generated model strictly complies with power industry standards, providing reliable technical support for the real-time monitoring and decision-making optimization of intelligent substations.

[0036] In one embodiment, as Figure 1 shown, a method for generating a digital twin model of a power system is provided, and the method includes the following steps:

[0037] Step 102, obtain multi-source heterogeneous data of the substation, and preprocess the multi-source heterogeneous data to obtain preprocessed data.

[0038] Among them, multi-source heterogeneous data includes structured data, unstructured data, and time-series data. Structured data covers typical monitoring information tables (including device real-time status, voltage, current, power parameters), sequence control operation tickets (such as standardized operation steps and device control instructions), and SCD files (such as substation configuration description files based on IEC 61850 standard). Unstructured data includes power system regulations (such as "Power System Safety Guidelines"), device maintenance logs, and fault record texts. Time-series data includes historical monitoring data streams (such as sampling frequency not less than 1Hz) and device alarm event sequences.

[0039] Preprocessing includes but is not limited to data collection and integration, cleaning and outlier handling, domain knowledge-driven annotation, Chinese semantic parsing, and knowledge graph construction, etc., to ensure the high quality, standardization, and security of the preprocessed data, and provide reliable input for subsequent model training and optimization.

[0040] In this embodiment, by obtaining multi-source heterogeneous data of the substation and preprocessing the multi-source heterogeneous data, preprocessed data is obtained, and then modeling is carried out based on subsequent steps. The standardized processing and efficient utilization of multi-source heterogeneous data are realized.

[0041] Step 104, perform fuzzification processing on the preprocessed data and extract fuzzy logic rules.

[0042] Among them, in the fuzzification processing stage, membership functions can be defined, and fuzzy C-means (FCM) can be used to quantify the semantic description of the preprocessed data. Furthermore, dependency syntax analysis and semantic role annotation techniques are used to identify condition-action pairs from unstructured texts, and the rule priorities are optimized based on the weight assignment strategy to extract fuzzy logic rules. Conflicts are resolved through weighted voting, and the rule adaptability is improved by combining reinforcement learning to optimize the rules. By introducing online learning and adaptive update mechanisms, and using blockchain technology to record the version evolution of the rule library for rule dynamic management, the traceability and reliability of the rules are ensured.

[0043] Step 106, construct an adversarial network model to be trained, and train the adversarial network model to be trained based on the fuzzy logic rules.

[0044] Among them, the adversarial network model to be trained includes a generator and a discriminator, through a fuzzy-adversarial joint training strategy and power standard compatibility design. The generator adopts an encoder-decoder structure and embeds fuzzy rule constraints to ensure that the generated results conform to the logic of the power system. The discriminator includes data distribution discrimination and rule compliance discrimination to enhance the credibility of the model. During the training process, the rule constraint weights and learning rates are dynamically adjusted to optimize the generation quality. So that the model output conforms to the IEC 61850 standard, ensuring its direct application in the power system.

[0045] Step 108, when it is determined that the model verification conditions are met, stop the training to obtain the digital twin model of the power system.

[0046] Among them, model verification includes logical compliance check, dynamic simulation test and closed-loop optimization mechanism to ensure the accuracy, reliability and applicability of the model in the power system. The digital twin model can be used for real-time perception, dynamic simulation and decision support of power system equipment. In the power system, the digital twin model can not only be used for health monitoring and condition assessment of equipment, but also provide real-time perception and deduction of grid equipment through multi-dimensional and multi-time scale dynamic virtual simulation.

[0047] In the above method for generating a digital twin model of a power system, by obtaining multi-source heterogeneous data of a substation, preprocessing the multi-source heterogeneous data to obtain preprocessed data, performing fuzzification processing on the preprocessed data, extracting fuzzy logic rules, constructing an adversarial network model to be trained, training the adversarial network model to be trained based on the fuzzy logic rules, and when it is determined that the model verification conditions are met, stopping the training to obtain the digital twin model of the power system. Its model automation generation technology strategy based on artificial intelligence fuzzy recognition provides an efficient and intelligent solution for the substation automation system. In the modeling process, a model combining fuzzy logic and generative adversarial network is introduced. Through dynamic rule base management and adaptive update, it is ensured that the system can adapt to the complex environment of power system operation in real time, thereby greatly improving the accuracy of automatic generation of substation operation tickets, while ensuring rule compliance and power system standard compatibility. In addition, the present invention constructs a perfect digital twin system, performs static verification, logical compliance check and real-time simulation test on the automatically generated model to ensure its applicability to various grid operation scenarios, and continuously optimizes through a closed-loop feedback mechanism to improve the robustness and applicability of the system. Generally speaking, it provides an innovative reference solution for power system automation, showing significant application value in improving modeling efficiency, reducing human errors and optimizing power dispatching, and having broad promotion and application prospects.

[0048] In an exemplary embodiment, such as Figure 2As shown in the figure, in step 102, preprocess the multi-source heterogeneous data to obtain the preprocessed data, which may specifically include:

[0049] Step 202, perform format standardization processing on the multi-source heterogeneous data to construct a multi-dimensional data set with spatio-temporal alignment.

[0050] Among them, format standardization processing is a key step in data preprocessing, aiming to convert data from different sources and different formats into a unified format for subsequent analysis and modeling. In this embodiment, the format of the multi-source heterogeneous data can be standardized through an Extract-Transform-Load (ETL) tool, and the timestamp format can be unified (such as adopting the ISO 8601 standard), and the device encoding follows the DL / T 860.74 specification, so as to construct a multi-dimensional data set with spatio-temporal alignment and provide high-quality input for subsequent model training.

[0051] Step 204, perform data cleaning and outlier processing on the multi-dimensional data set to obtain the multi-dimensional data set after cleaning.

[0052] In the data cleaning stage, the sliding window mean filtering algorithm (such as the window length can be configured as 5-10 sampling points) can be used to eliminate instantaneous noise, and the isolation forest algorithm can be used to detect and remove abnormal monitoring values (such as sudden current, abnormal temperature rise, etc.). For the problem of missing device sensor data, the time series interpolation method (such as linear interpolation or spline interpolation, etc.) is applied to supplement the data, and the interpolation interval is marked for distinction during model training. For unstructured text data (such as operation tickets, regulations), redundant symbols (such as garbled characters, special characters, etc.) are removed by regular expression matching, and the term expressions are unified (such as "circuit breaker" and "switch" can be standardized into the same term) to ensure the consistency and usability of the data.

[0053] Step 206, perform data annotation on the multi-dimensional data set after cleaning based on predefined annotation rules.

[0054] Among them, the annotation rules can be defined according to the power industry standards (such as Q / GDW 11612-2016). For example, the device status labels (normal / alarm / fault) in the monitoring information table and the operation step priorities (urgent / routine) in the operation tickets can be annotated. The device logical node attributes can also be extracted from the SCD file and their function types can be annotated (such as XCBR is the circuit breaker logical node). To improve the annotation efficiency, a semi-automated annotation tool based on a rule engine can also be used for data annotation, which supports regular expression matching (such as marking "tripping" as a fault event) and combines an expert knowledge base (such as the "Typical Substation Operation Ticket Library") for semantic verification to ensure the accuracy and consistency of the annotation results.

[0055] Step 208: Semantically parse and fuse the labeled multi-dimensional dataset to obtain corresponding semantic information, and use the semantic information as the preprocessed data.

[0056] Specifically, in the Chinese semantic parsing stage, an improved Jieba word segmentation algorithm can be adopted, a custom dictionary in the power field (including equipment models and operation terms) can be loaded, and a bidirectional LSTM-CRF model can be combined to identify entities in the text (such as equipment names and operation actions). Based on dependency syntactic analysis, condition-action pairs in the operation ticket are extracted (such as "when the bus voltage > 110%, perform a voltage reduction operation"), and fuzzy logic rule triples are constructed, usually represented as (subject - predicate - object). Among them: Subject: represents the condition or state. Predicate: represents the logical relationship, such as "if", "when", etc. Object: represents the action or result. Then, according to the above action pairs, the following triples can be constructed: Subject: bus voltage > 110%; Predicate: if (or "when"); Object: perform a voltage reduction operation. Then the fuzzy logic rule triple is: (bus voltage > 110%, if, perform a voltage reduction operation).

[0057] By preprocessing multi-source heterogeneous data in the above manner, the quality, standardization, and security of the preprocessed data can be ensured, providing reliable input for subsequent model training and optimization.

[0058] Furthermore, the semantic relationship between structured data and unstructured text can be fused to generate a substation equipment topology map, where nodes represent equipment entities, edges represent electrical connections or logical dependencies, and it supports storage and query in a graph database (such as Neo4j), laying a foundation for the subsequent construction of the fuzzy rule base.

[0059] In one embodiment, in step 102, after preprocessing the multi-source heterogeneous data to obtain the preprocessed data, the above method may further include: verifying the data quality of the preprocessed data through preset quantization metrics to obtain the verified data. Among them, the quantization metrics include but are not limited to data integrity metrics, consistency metrics, timeliness metrics, and correctness metrics.

[0060] Specifically, after data preprocessing, the data quality can also be evaluated through preset quantization metrics, including data integrity metrics (such as missing rate < 2%), consistency metrics (such as field conflict rate < 0.5%), and timeliness metrics (such as data latency < 1s). At the same time, sampling and auditing of the above data annotation results can be carried out to ensure the correctness metric (such as error rate below 1%). In addition, the above data preprocessing process strictly follows the "Network Security Law" and the "Regulations on the Security Protection of Power Monitoring Systems", adopts encrypted transmission (such as AES-256) and anonymizes sensitive information (such as equipment geographical location), thereby ensuring data security and compliance.

[0061] In an exemplary embodiment, as Figure 3 shown, in step 104, based on the preprocessed data, a fuzzification process is performed to extract fuzzy logic rules, which may specifically include:

[0062] Step 302, obtaining the fuzzy descriptions in the preprocessed data, and defining the membership functions of the key parameters in the fuzzy descriptions based on the operation standards of the power system.

[0063] Specifically, the fuzzy descriptions in the preprocessed data can be obtained, and the membership functions of the key parameters in the fuzzy descriptions are defined based on the operation standards of the power system. For example, for the fuzzy descriptions in substation monitoring information (such as "voltage is too high", "load is critical"), they can be transformed into mathematically processable fuzzy sets through a domain knowledge-driven method. Based on the operation standards of the power system (such as DL / T 1080-2016), the fuzzy membership functions of the key parameters can be defined. For example, for continuous variables (such as voltage, current), a Gaussian membership function can be used, and its mathematical expression can be:

[0064]

[0065] where is the continuous variable to be evaluated, is the mean of the fuzzy set (such as the nominal voltage value), is the standard deviation. The core interval can be set based on expert experience (such as the "normal voltage" range is ±5% of the nominal value), and the transition interval can be dynamically expanded (such as "voltage is too high" is +5%~+10% of the nominal value).

[0066] For discrete states (such as "minor alarm", "severe fault" of equipment), a piecewise membership function can be used, and the state thresholds are quantified by combining historical fault data (such as when the temperature > 80°C, the membership degree , when the temperature > 100°C ). In addition, the fuzzy C-means clustering (FCM) algorithm can be used to semantically quantify the fuzzy descriptions in unstructured text (such as "quickly cut off the fault" in the operation ticket), and its objective function can be:

[0067]

[0068] where represents the membership degree of text segment i belonging to fuzzy label j, is the clustering center, is the fuzzy factor, and standardized fuzzy labels are generated through iterative optimization, thus providing a basis for subsequent rule extraction.

[0069] Step 304: Use dependency syntactic analysis and semantic role labeling techniques to identify condition-action pairs from the fuzzy descriptions and extract fuzzy logic rules.

[0070] In this embodiment, fuzzy logic rules can be extracted from sequence control operation tickets, regulations, and historical operation records. Specifically, based on dependency syntactic analysis and semantic role labeling techniques, condition-action pairs in unstructured texts can be identified. For example, from the statement "If the bus voltage continues to exceed the limit, then start the standby power supply", the fuzzy logic rule can be extracted as: IF bus voltage > threshold AND duration > T THEN execute switching of the standby power supply.

[0071] To eliminate rule conflicts, a rule priority matrix can also be constructed. According to the urgency of the operation ticket and the mandatory level of the regulations (such as "Guide for Power System Security and Stability"), weights are assigned to the rules , and its calculation method can be:

[0072]

[0073] where, and are adjustment coefficients (determined by fitting historical operation data). When multiple rules conflict, a weighted voting mechanism can be used to select the optimal action, such as:

[0074]

[0075] where, is the set of rules supporting action . Further, by simulating typical substation scenarios (such as short-circuit faults, load mutations, etc.) through a reinforcement learning framework, the effectiveness of the rules can be verified, redundant rules can be removed, and edge scenario rules (such as equipment protection strategies under extreme weather conditions) can be supplemented to ensure the completeness and robustness of the rule base.

[0076] Furthermore, to achieve the dynamic optimization of the rule base, an online learning module, an incremental update strategy, and a version control mechanism can also be adopted for the dynamic optimization of the rule base. Specifically, the online learning module can receive substation monitoring data and operation feedback in real time, and evaluate the rule execution effect through a fuzzy inference engine (such as the Mamdani model). If the system state does not meet the expectation after the rule is triggered (such as the alarm is not eliminated), the rule will be automatically marked and the manual review process will be triggered. When new equipment is added or regulations are revised, the existing rule base can also be mapped to the new scenario based on transfer learning technology, thereby reducing the cost of repeated annotation. For example, when adding a new photovoltaic inverter device, the rule logic of the original "overvoltage protection" can be reused, and only the voltage threshold parameter needs to be adjusted. Thus, the existing rules can be automatically migrated to the new scenario, reducing manual intervention and adapting to the complex and changeable operation requirements of the substation. The rule base can use blockchain technology to store the version history, thereby ensuring that the modification records are traceable and supporting a quick rollback to a stable version to ensure the reliability and maintainability of the rule base.

[0077] In an exemplary embodiment, in step 106, constructing the adversarial network model to be trained may specifically include: constructing the adversarial network model to be trained based on a generator and a discriminator.

[0078] Among them, the generator can adopt an encoder-decoder structure based on a deep neural network. The input is a fuzzy condition vector and real-time monitoring data (such as equipment status parameters, environmental variables), and the output is a standardized digital twin model (such as equipment behavior data such as equipment dynamic behavior prediction, operation ticket sequence, etc.). To ensure that the generated results conform to the power system logic, a fuzzy rule constraint module can be embedded inside the generator, and its core is the rule satisfaction calculation layer. Specifically, the fuzzy rules are incorporated into the forward propagation process of the generator in the form of a weight matrix, and the rule satisfaction loss function can be defined as:

[0079]

[0080] Among them, represents the membership degree of the generated result to the i-th rule, and N is the total number of rules. For example, when generating a sequence control operation ticket, if the rule requires that "the reclosing must be blocked before isolating the fault", the operation sequence output by the generator must meet this condition, otherwise the network parameters are adjusted through backpropagation.

[0081] The discriminator can be designed as a multi-task learning architecture, including a data distribution discrimination branch and a rule compliance discrimination branch. Specifically, the data distribution discrimination branch extracts the features of historical monitoring data through a convolutional neural network (CNN), calculates the Jensen-Shannon divergence between the generated data and the real data, and the adversarial loss function can be expressed as:

[0082]

[0083] Among them, x represents the real data sample, represents the probability distribution of the real data, which is usually obtained by statistical analysis of historical data sets (such as substation monitoring information tables, sequence control operation tickets, etc.). D(x) represents the output of the discriminator for the real data sample x, z represents the random noise vector, represents the prior distribution of the random noise vector, and G(z) represents the output of the generator for the random noise vector z.

[0084] The rule compliance discrimination branch integrates a fuzzy inference engine to perform logical verification on the generation results of the generator. For example, the discriminator checks whether the operation ticket generated by the generator violates the operation sequence in the "Guide for Power System Security and Stability". Its output is the probability of rule violation. The final total loss function of the discriminator is:

[0085]

[0086] Among them, is the adversarial loss, is the balance coefficient, is the rule satisfaction loss.

[0087] In this embodiment, by integrating fuzzy logic and generative adversarial networks, and dynamically constraining the generation process through a fuzzy rule base, the problem of insufficient description of uncertain states (such as "high voltage" and "critical load") by traditional modeling methods is solved. Combining with the efficient data generation ability of generative adversarial networks, the physical rationality and logical accuracy of the digital twin model can be significantly improved.

[0088] Furthermore, to meet the power industry standards (such as IEC 61850), the output layer of the generator can be designed as a standardized interface conversion module to map the generated digital twin model into the SCD file format, supporting the automatic encapsulation of logical nodes and service interfaces. For example, the generated circuit breaker behavior model needs to conform to the XCBR logical node definition, and its data attributes (such as Pos.stVal representing the switch position) are fully compatible with the standard. At the same time, the IEC 61850 semantic library can be pre-loaded in the discriminator to ensure seamless integration of the generated model with the existing monitoring system at the semantic layer, thus realizing the compatibility design of the model with the power system standard.

[0089] In an exemplary embodiment, in step 106, the adversarial network model to be trained is trained based on fuzzy logic rules. Specifically, a fuzzy-adversarial joint training strategy can be adopted, that is, a strategy of alternately optimizing the generator and the discriminator is adopted, and at the same time, a dynamic update mechanism of the fuzzy rule base is introduced. Each round of training may include the following steps: (1) Generator forward propagation: Input a fuzzy condition vector (such as device state "overload", environmental temperature "high"), and generate device behavior data such as a candidate model or an operation sequence; (2) Discriminator multimodal verification: Evaluate the distribution consistency between the generated data and the real data, and check the logical compliance through a fuzzy rule engine; (3) Rule feedback optimization: If the generated result triggers a rule conflict, extract the conflict rule features and update the rule constraint weight matrix of the generator; (4) Adaptive learning rate adjustment: Dynamically adjust the learning rates of the generator and the discriminator according to the rule satisfaction degree and the convergence of the adversarial loss to optimize the generation quality and avoid mode collapse, thereby improving the credibility of the model.

[0090] Further, based on the trained fuzzy generative adversarial network model, by inputting real-time monitoring data (such as device state parameters, environmental variables, and grid load information), a standardized digital twin model can be generated. The generation process includes: First, map the input data to a low-dimensional feature vector through an encoder network, and combine the conditional constraints in the fuzzy rule base (such as device overload threshold, operation priority) to construct the initial state space of the generator. The output layer of the generator is designed as a multi-branch structure, which respectively generates a device dynamic behavior model (such as the action time sequence of a circuit breaker, the temperature rise curve of a transformer, etc.), a sequence control operation ticket (compliant with the DL / T634.5104 standard), and an SCD file (defined based on the IEC 61850 logical node). Then, map the generated model to the standard format of the power system through an interface conversion module. For example, encapsulate the circuit breaker behavior model as an XCBR logical node, and its data attributes (such as Pos.stVal representing the switch position) are fully compatible with the SCD file, so as to ensure seamless docking with the existing monitoring system and avoid manual secondary adjustment.

[0091] In one scenario, static verification of the generated model can also be performed to ensure its compliance with the power system specifications. Specifically, it can be checked whether the logical node naming, data type, and service interface of the SCD file comply with the IEC61850-6 standard. For example, the error tolerance rate needs to be lower than 0.1%. Secondly, the fuzzy inference engine can be used to calculate the membership degree of the generated model to the rule base, and the satisfaction degree of key rules (such as "lock the reclosing before isolating the fault") . It is also possible to parse the generated operation ticket text through natural language processing technology to verify whether its action sequence is consistent with the standard process in the "Substation Typical Operation Ticket Library". For example, the semantic matching rate needs to be higher than 98%. Thus, static verification of the generated model and logical compliance check are realized.

[0092] In an exemplary embodiment, the operation scenario of the substation can also be simulated on the digital twin platform to dynamically verify the real-time performance and accuracy of the generation model. First, scenarios such as short-circuit faults and load mutations can be simulated to test the model response time (e.g., the alarm trigger delay needs to be less than 100 ms) and the correctness of the action sequence (e.g., the fault isolation steps comply with the Q / GDW11612-2016 standard). Second, for extreme working conditions (such as cascaded faults of multiple devices and voltage sags), the robustness of the model is verified. For example, it is required that the availability rate of the generated operation ticket is not less than 99.5%. Finally, the real-time data of the generated model and the physical device are synchronized through the OPC UA protocol, and the root mean square error (RMSE) between the model prediction value and the actual value is calculated. For example, it can be required that the RMSE is lower than 2% of the nominal value. Thus, the dynamic simulation and real-time performance test of the generation model are realized.

[0093] In an exemplary embodiment, based on the closed-loop verification mechanism, the test results can also be fed back to the training process to achieve continuous optimization of the model. Specifically, if rule conflicts or logical errors (such as reversed operation sequences) are found during verification, the error sample features can be automatically extracted, the violated rule entries can be marked, and they can be added to the training set for incremental learning. Then, the rule constraint weight matrix of the generator is dynamically updated through the reinforcement learning framework. For example, for high-frequency error rules (such as "no reclosing lockout"), its balance coefficient can be increased to strengthen the constraint. The blockchain technology can also be used to record the iteration history of the model version to ensure that each update is traceable, and the new model is gradually deployed to the substation automation system through the gray release strategy, thereby minimizing the operation risk to the greatest extent.

[0094] This embodiment is based on end-to-end automated modeling and closed-loop optimization, from the preprocessing of multi-source heterogeneous data to the full process automation of standardized model generation. Combining reinforcement learning with the error feedback mechanism, the model can be iteratively updated monthly, the rule coverage rate can be increased to 99.8%, and the false alarm rate can be reduced to less than 1%, which can significantly improve the operation safety and response efficiency of the substation.

[0095] In an exemplary embodiment, as Figure 4 shown, the above method for generating a digital twin model of a power system is further described below, which specifically may include the following steps:

[0096] Step 402, input the initial data.

[0097] Specifically, the initial data includes multi-source heterogeneous data of a substation, namely, structured data (typical monitoring information tables, sequence control operation tickets, and SCD files), unstructured data (power system regulations and procedures), and time-series data (historical monitoring data streams and device alarm event sequences). Clean and handle outliers for the multi-source heterogeneous data, and label the data to perform Chinese semantic parsing and knowledge graph construction. Specifically, reference can be made to the embodiments such as Figure 2 shown in the embodiment, and this embodiment will not elaborate on this again.

[0098] Step 404, fuzzy rule construction.

[0099] Specifically, it includes quality assessment of the previously constructed Chinese semantics, fuzzy processing, and membership function design to extract fuzzy logic rules. If new data needs to be added, dynamic rule library management and adaptive update need to be executed. Specifically, reference can be made to the embodiments such as Figure 3 shown in the embodiment, and this embodiment will not elaborate on this again.

[0100] Step 406, design of a fuzzy generative adversarial network model.

[0101] Specifically, it includes a generator network architecture and fuzzy rule embedding, a multi-modal verification mechanism for the discriminator network, and a compatibility design of the model with power system standards. For details, reference can be made to the detailed description of the above embodiments, and this embodiment will not elaborate on this again

[0102] Step 408, train the constructed fuzzy generative adversarial network model.

[0103] Step 410, determine whether the automated model verification condition is met.

[0104] Specifically, if the automated model verification condition is not met, return to step 408 to continue training. If it is met, execute step 412.

[0105] Step 412, stop training and output the trained model.

[0106] Specifically, if the automated model verification condition is met, stop training and output the trained model.

[0107] This embodiment proposes an automated generation technology strategy for the digital twin model of the power system based on artificial intelligence fuzzy recognition, providing an efficient and intelligent solution for the substation automation system. This method covers key steps such as data collection, cleaning, annotation, semantic parsing, knowledge graph construction, and fuzzy logic rule extraction, realizing the standardized processing and efficient utilization of multi-source heterogeneous data. And in the modeling process, a model combining fuzzy logic and generative adversarial network is introduced. Through dynamic rule base management and adaptive update, it ensures that the system can adapt to the complex environment of power system operation in real time. Compared with traditional modeling methods, this embodiment greatly improves the accuracy of automated generation of substation operation tickets, while ensuring rule compliance and compatibility with power system standards. In addition, this embodiment constructs a complete digital twin system, conducts static verification, logical compliance check, and real-time simulation test on the automatically generated model to ensure its applicability to various power grid operation scenarios, and continuously optimizes through a closed-loop feedback mechanism, improving the robustness and applicability of the system. Generally speaking, this embodiment provides an innovative reference solution for power system automation, showing significant application value in improving modeling efficiency, reducing human errors, and optimizing power dispatching, and having broad promotion and application prospects.

[0108] It should be understood that although the steps in the flowcharts involved in the above embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0109] Based on the same inventive concept, the embodiments of the present application also provide a device for generating a digital twin model of a power system for implementing the above-mentioned method for generating a digital twin model of a power system. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the device for generating a digital twin model of a power system provided below can refer to the limitations on the method for generating a digital twin model of a power system in the above text, and will not be repeated here.

[0110] In an exemplary embodiment, as Figure 5As shown, a device for generating a digital twin model of a power system is provided, including: a data acquisition module 502, a fuzzy processing module 504, a model training module 506, and a model generation module 508, where:

[0111] The data acquisition module 502 is configured to acquire multi-source heterogeneous data of a substation, preprocess the multi-source heterogeneous data to obtain preprocessed data; the multi-source heterogeneous data includes structured data, unstructured data, and time-series data;

[0112] The fuzzy processing module 504 is configured to perform fuzzy processing based on the preprocessed data and extract fuzzy logic rules;

[0113] The model training module 506 is configured to construct an adversarial network model to be trained and train the adversarial network model to be trained based on the fuzzy logic rules;

[0114] The model generation module 508 is configured to stop training when it is determined that the model verification condition is met, and obtain a digital twin model of the power system, where the digital twin model is used for health monitoring and dynamic policy adjustment of power system equipment.

[0115] In an exemplary embodiment, the data acquisition module is further configured to: perform format standardization processing on the multi-source heterogeneous data to construct a multi-dimensional data set with spatio-temporal alignment; perform data cleaning and outlier processing on the multi-dimensional data set to obtain a multi-dimensional data set after cleaning processing; perform data annotation on the multi-dimensional data set after cleaning processing based on predefined annotation rules; perform semantic parsing and fusion on the annotated multi-dimensional data set to obtain corresponding semantic information, and use the semantic information as the preprocessed data.

[0116] In an exemplary embodiment, the device further includes a data verification module, configured to: verify the data quality of the preprocessed data through preset quantization metrics to obtain verified data, where the quantization metrics include data integrity metrics, consistency metrics, timeliness metrics, and correctness metrics.

[0117] In an exemplary embodiment, the fuzzy processing module is further configured to: obtain fuzzy descriptions in the preprocessed data, and define membership functions of key parameters in the fuzzy descriptions based on power system operation standards; adopt dependency syntax analysis and semantic role annotation technologies to identify condition-action pairs from the fuzzy descriptions to extract fuzzy logic rules.

[0118] In an exemplary embodiment, the model training module is further configured to: construct an adversarial network model to be trained based on a generator and a discriminator; the generator adopts an encoder-decoder structure and is used to generate device behavior data by embedding the fuzzy logic rule constraints; the discriminator is used to perform multi-modal verification on the generated device behavior data.

[0119] In an exemplary embodiment, the model training module is further configured to: based on forward propagation, input a fuzzy condition vector to the generator to enable the generator to generate device behavior data; evaluate the adversarial loss between the generated device behavior data and the real data based on the discriminator, and determine the rule satisfaction loss of the device behavior data through a fuzzy rule engine; if the device behavior data triggers a rule conflict, extract the conflict rule features and update the rule constraint weight matrix of the generator; dynamically adjust the learning rates of the generator and the discriminator according to the convergence of the rule satisfaction loss and the adversarial loss to train the adversarial network model to be trained.

[0120] Each module in the above power system digital twin model generation device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.

[0121] In an exemplary embodiment, a computer device is provided, and its internal structure diagram can be as Figure 6As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for generating a digital twin model of a power system. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0122] Those skilled in the art can understand that Figure 6 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0123] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0124] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0125] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0126] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0127] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., and are not limited thereto.

[0128] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.

[0129] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application should be subject to the appended claims.

Claims

1. A method for generating a digital twin model of a power system, characterized in that, The method includes: Obtaining multi-source heterogeneous data of a substation, preprocessing the multi-source heterogeneous data to obtain preprocessed data; the multi-source heterogeneous data includes structured data, unstructured data, and time-series data; Performing fuzzification processing based on the preprocessed data to extract fuzzy logic rules; Constructing an adversarial network model to be trained, and training the adversarial network model to be trained based on the fuzzy logic rules; When it is determined that the model verification conditions are met, stop training to obtain a digital twin model of the power system, and the digital twin model is used for health monitoring and dynamic policy adjustment of power system equipment.

2. The method according to claim 1, wherein The preprocessing of the multi-source heterogeneous data to obtain preprocessed data includes: Performing format standardization processing on the multi-source heterogeneous data to construct a multi-dimensional data set with spatio-temporal alignment; Performing data cleaning and outlier processing on the multi-dimensional data set to obtain a cleaned multi-dimensional data set; Performing data annotation on the cleaned multi-dimensional data set based on predefined annotation rules; Performing semantic parsing and fusion on the annotated multi-dimensional data set to obtain corresponding semantic information, and using the semantic information as the preprocessed data.

3. The method according to claim 1 or 2, characterized in that, After the preprocessing of the multi-source heterogeneous data to obtain preprocessed data, the method further includes: Verifying the data quality of the preprocessed data through preset quantization metrics to obtain verified data, and the quantization metrics include data integrity metrics, consistency metrics, timeliness metrics, and correctness metrics.

4. The method according to claim 3, characterized in that The fuzzification processing based on the preprocessed data to extract fuzzy logic rules includes: Obtaining fuzzy descriptions in the preprocessed data, and defining membership functions of key parameters in the fuzzy descriptions based on the operation standards of the power system; Adopting dependency syntax analysis and semantic role annotation techniques to identify condition-action pairs from the fuzzy descriptions to extract fuzzy logic rules.

5. The method according to claim 1, characterized in that The construction of the adversarial network model to be trained includes: Constructing an adversarial network model to be trained based on a generator and a discriminator; the generator adopts an encoder-decoder structure and is used to generate device behavior data by embedding the fuzzy logic rule constraints; the discriminator is used to perform multi-modal verification on the generated device behavior data.

6. The method according to claim 5, wherein The training of the adversarial network model to be trained based on the fuzzy logic rules includes: Based on forward propagation, inputting a fuzzy condition vector to the generator to enable the generator to generate device behavior data; Evaluating the adversarial loss between the generated device behavior data and the real data based on the discriminator, and determining the rule satisfaction loss of the device behavior data through a fuzzy rule engine; If the device behavior data triggers a rule conflict, extract the conflict rule features and update the rule constraint weight matrix of the generator; Dynamically adjust the learning rates of the generator and the discriminator according to the convergence of the rule satisfaction loss and the adversarial loss to train the adversarial network model to be trained.

7. A digital twin model generation device for a power system, characterized in that The device includes: A data acquisition module, configured to acquire multi-source heterogeneous data of a substation, preprocess the multi-source heterogeneous data to obtain preprocessed data; the multi-source heterogeneous data includes structured data, unstructured data, and time-series data; A fuzzy processing module, configured to perform fuzzy processing based on the preprocessed data and extract fuzzy logic rules; A model training module, configured to construct an adversarial network model to be trained and train the adversarial network model to be trained based on the fuzzy logic rules; A model generation module, configured to stop training when it is determined that the model verification condition is satisfied, and obtain a digital twin model of the power system, where the digital twin model is used for health monitoring and dynamic policy adjustment of power system equipment.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 6.

Images