An intelligent substation automation management system

Abstract

The application relates to the technical field of power system automation, and particularly discloses an intelligent substation automation management system. The system comprises a digital twin synchronous management module, a multi-engine collaborative deduction module, an online strategy optimization and simulation module, and an instruction safety checking and feedback module. The system builds a digital twin synchronized with a physical station as a unified analysis environment; multiple professional analysis engines interacting through a shared cognitive state library are run in the twin, interwoven evaluation and joint deduction of equipment state, protection logic and network reliability are realized; based on the deduction result, online simulation iteration is used to search for a multi-step optimization control plan in the twin environment; and after safety checking of the plan instruction, the instruction is executed, and feedback data is collected to drive the closed-loop evolution of the model and the strategy. The application realizes multi-source data fusion, dynamic collaboration of cross-professional analysis and adaptive optimization of control strategies, and improves the integrity and autonomy of intelligent substation management.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and more specifically, to an intelligent substation automation management system. Background Technology

[0002] With the deepening of smart grid construction, the automation and intelligence level of substations has become crucial to ensuring the safe, reliable, and efficient operation of the power system. Smart substations are equipped with a massive number of sensors and intelligent devices, generating multi-dimensional operational data covering primary equipment status, secondary system behavior, and network communication performance. Effectively integrating and analyzing this multi-source, heterogeneous data, and on this basis, achieving intelligent management from passive response to proactive early warning, and from single-point control to collaborative optimization, is a requirement for improving substation operation and maintenance efficiency.

[0003] Existing intelligent substation automation systems often have relatively independent functional modules. For example, systems for equipment status monitoring, protection setting management, and network performance analysis are typically separate, creating "information silos." This results in fragmented assessments of substation operating status, lacking in-depth integration and collaborative analysis of cross-disciplinary and multi-dimensional data. For instance, it's difficult to accurately map the deterioration trend of equipment status to the impact on the coordination logic of protection systems in real time; fluctuations in network performance are also difficult to quantitatively incorporate into the safety margins of control strategies. Furthermore, the system's decision-making mechanisms often rely on pre-set fixed rules or simple thresholds, lacking adaptive adjustment and forward-looking optimization capabilities when facing complex and changing operating conditions and potential risks. Existing technologies struggle to achieve closed-loop deduction and self-evolution based on global real-time situational awareness, hindering the progress of substation automation management towards higher levels of intelligence.

[0004] Therefore, an intelligent substation automation management system is proposed to address the above problems. The issues to be solved are: how to break down professional barriers and achieve deep integration and integrated analysis of multi-source data; how to construct an analysis framework that can reflect the dynamic coupling relationship between various systems to generate collaborative and forward-looking situation assessments; and how to establish a decision-making mechanism with online learning and adaptive capabilities to generate and execute dynamically optimized control strategies. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent substation automation management system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent substation automation management system, comprising: The digital twin synchronization management module is used to build and maintain a digital twin of the substation that operates synchronously with the physical substation. This twin is driven by mapping the substation's equipment parameters, topology connections, and secondary system logic, and injecting real-time data streams, thereby providing a unified, dynamic, and computable high-fidelity virtual environment for the comprehensive analysis, simulation, and decision-making of the entire substation. The multi-engine collaborative simulation module is connected to the digital twin synchronization management module. It is used to run multiple professional analysis engines that interact through a shared cognitive state library in the digital twin. The cognitive state library serves as a common medium for information exchange and event triggering among the engines, enabling the originally isolated processes such as equipment status prediction, protection logic adaptive analysis, and network service reliability assessment to be intertwined, evaluated, and iteratively solved based on a unified data source. This generates operational status simulation results and risk quantification indicators that comprehensively reflect the multi-dimensional coupling relationship of the system. The online strategy optimization and simulation module connects the digital twin synchronization management module and the multi-engine collaborative inference module. Based on the inference results and indicators, it performs online iterative search in the simulation environment composed of the digital twin by simulating the execution of different control action sequences and evaluating their long-term comprehensive benefits. This generates a multi-step optimized control plan that can proactively adapt to future changes in the system, thereby transforming static rule-based decision-making into scenario-adaptive optimization based on dynamic simulation. The instruction security verification and feedback module is connected to the online policy optimization and simulation module. It is used to perform multi-level security verification on the immediate execution instructions in the contingency plan, covering semantics, topology, timing and permissions. This multi-dimensional verification aims to ensure that any automatic control instruction complies with security procedures and is conflict-free. After the verification is passed, it is converted into standard protocol instructions and sent to physical devices for execution. At the same time, the response data of the devices and system status changes are collected. The instruction security verification and feedback module feeds back the collected response data to the digital twin synchronization management module and the multi-engine collaborative inference module, thus forming a closed loop. This closed loop enables the digital twin model to be corrected online, the strategy effect to be evaluated, and the overall decision-making capability of the system to be iteratively evolved based on actual interaction data.

[0007] Furthermore, the professional analysis engine performs collaborative computation through the shared cognitive state library in the following ways: The cognitive state library stores standardized intermediate results and event labels generated by each engine. This standardized design eliminates semantic barriers between different professional data. The first analysis engine publishes the device's predictive state range and confidence information to the cognitive state database, which characterizes the device's possible future health status. The second analysis engine subscribes to the state interval and confidence information, dynamically calculates the coordination logic and adjustable margin of the protection settings under the current and predicted equipment operating conditions, and publishes them to the cognitive state library. This process realizes real-time correlation analysis between equipment status and protection setting logic. The third analysis engine integrates the adjustable margin and real-time network status to assess the expected reliability indicators of key control service links and publishes them to the cognitive state database, thereby incorporating the impact of network performance into service availability analysis. In this process, any key information update event published by any engine triggers other engines that have subscribed to the event to start related calculations. This event-based triggering mechanism ensures that any local state change can trigger a chain of globally related analyses and rapid responses, realizing dynamic coupling and self-coordination of the analysis process.

[0008] Furthermore, the online strategy optimization and simulation module generates multi-step optimization control plans in the following ways: The digital twin is used as a simulation environment, and the operational situation simulation results are encoded as an environmental state vector, which provides a quantitative initial state input for policy search. In the simulation environment, based on the preset system optimization goals and safety constraints, different control action sequences are simulated and executed in the forward direction. Through simulation, the system state trajectory triggered by each action sequence in the future can be predicted. By comparing the long-term comprehensive benefits generated by each control action sequence during simulation, the action sequence that satisfies all safety constraints is iteratively selected as the output plan. This method based on long-term benefit comparison guides the strategy to converge toward the global optimum or suboptimal direction.

[0009] Furthermore, the online policy optimization and simulation module also includes a policy robustness testing unit, which performs the following: In a simulation environment, construct two or more different future disturbance scenarios for candidate control action sequences. These scenarios are used to simulate uncertainties such as load fluctuations or random equipment failures. The candidate control action sequences are executed in parallel for multi-step look-ahead simulations under different perturbation scenarios to observe the performance differences of the same strategy under different potential futures. The stability of key performance indicators of each candidate control action sequence under different disturbance scenarios is quantitatively evaluated. A strategy with high stability means that it is not sensitive to environmental changes and has stronger adaptability. The strategy optimization engine selects the control action sequence as the output plan based on the quantitative evaluation results, and prioritizes robust strategies that can maintain stable performance under various disturbances, thereby improving the reliability of automatic control in real complex environments.

[0010] Furthermore, the digital twin synchronization management module dynamically maintains the substation digital twin in the following ways: It continuously receives real-time data streams from physical substations and synchronously maps them to the corresponding models in the digital twin, ensuring that the basic data of the virtual environment and the physical world are synchronized. Calculate the residual between the digital twin simulation data and the physical station measured data; this residual reflects the model prediction accuracy. When the statistical characteristics of the residuals exceed the adaptive threshold, it indicates that there is a significant deviation between the model and reality. The model parameter calibration routine is then initiated, and the simulation parameters of the corresponding model in the digital twin are adjusted based on the residuals. Through this online calibration mechanism, the digital twin can adaptively update as physical devices age, their characteristics drift, or their topology changes, thus maintaining its fidelity.

[0011] Furthermore, it also includes a closed-loop learning and evolution management module, which: The simulation process data, the sequence of executed strategy instructions and the corresponding physical feedback results are serialized into structured experience data and stored, thereby accumulating historical experience in system operation and interaction; Data is sampled from the historical experience database to retrain and optimize the parameters of the internal evaluation model of the professional analysis engine and the decision logic parameters of the strategy optimization engine. Historical data is used to optimize the internal algorithm and improve the accuracy of future analysis and the quality of decision-making. By controlling the incremental model update process and switching the validated model parameters and decision logic to an online state, this controlled update method ensures that the improvement of system capabilities is smooth, safe, and traceable, thus realizing the autonomous and continuous evolution of the system's intelligence level.

[0012] Furthermore, the multi-engine collaborative inference module also includes a collaborative solution scheduling unit, which: Manage the activation cycles of the first, second, and third analysis engines and coordinate the allocation of their computing resources; The core scheduler that listens to the event tags in the cognitive state database and schedules the correlation analysis engine to start the calculation according to the event type and predefined rules is the core scheduler that drives the collaborative work of multiple engines. Coordinate the timing of data exchange between various analysis engines to ensure the consistency and integrity of data access during concurrent computation and prevent errors in analysis results caused by data competition.

[0013] Furthermore, the security compliance verification performed by the instruction security verification and feedback module includes: The semantics of the instructions are parsed and compliance is compared to ensure that the intent of the instructions conforms to the substation operating procedures and logic. By combining real-time power grid topology for connectivity verification and load calculation, misoperation can be prevented from the perspective of electrical connection and capacity. By comparing the instruction sequence, conflict detection is performed between the occupancy status of the operation object and the timing logic to prevent operation overlap or sequence errors; Verify the operational permissions of the entity issuing the instruction, and confirm its identity and authorization. In this process, instructions are allowed to be converted and issued only after passing all verifications. This multi-layered defensive verification constitutes a solid defense for the secure execution of automated instructions.

[0014] Furthermore, it also includes a unified data service module, which: Multi-source heterogeneous real-time data is collected from station control layer, bay layer and process layer equipment, serving as the sole and authoritative data source for all upper-layer applications of the system; The collected data undergoes protocol parsing, timestamp synchronization, and formatting to generate a standardized data stream that conforms to a unified information model, thus resolving the inconsistency between the data "interface" and "syntax". The standardized data stream is distributed to modules that subscribe to the data, and feedback data is received and stored. It plays the role of the data bus and governance core of the entire system, ensuring the consistency, timeliness and traceability of data throughout its entire lifecycle.

[0015] Furthermore, the closed-loop learning and evolution management module controls the incremental model update process, including: Validate the new model offline in a test environment using historical or simulation data, evaluate its performance baseline, and ensure that the update does not introduce performance regression. The new and old models are run online in parallel using the shadow mode, which means that the new and old models process real-time data at the same time, but only the output of the old model is used to control the physical equipment. At the same time, the output differences between the new and old models are compared. During this stage, the behavior data of the new model in the real environment can be collected without affecting production safety. Once the preset conditions are met, the new model that meets the performance standards will be switched to the online primary model. This phased and verified switching strategy ensures the smoothness and security of the system upgrade process to the greatest extent.

[0016] The technical effects and advantages of this invention are as follows: Compared to existing technologies, this invention constructs a dynamic digital twin that operates synchronously with the physical substation, serving as a unified quantitative environment for all analyses and simulations. This system continuously drives and calibrates the twin based on standardized data streams, enabling the analysis of equipment status, power flow, protection logic, and network performance on a unified, accurate, and reproducible simulation platform. This approach overcomes the shortcomings of traditional multi-system data model inconsistencies and asynchronous analysis benchmarks, providing a highly consistent and reliable digital foundation for subsequent collaborative analysis and strategy optimization, thereby improving the consistency and accuracy of overall situational awareness.

[0017] Compared to existing technologies, this invention deploys multiple specialized analysis engines that interact through a shared cognitive state database and designs an event-based collaborative solution mechanism. Each engine (such as equipment status prediction, protection logic adaptability analysis, and service reliability assessment) publishes intermediate results to the public state database. Updates to any key information automatically trigger related engines to start a new round of analysis. This essentially establishes a dynamically coupled analysis network, enabling previously isolated technical indicators such as equipment health, protection setting margin, and network reliability to mutually verify and iteratively converge in real time. This achieves deep interweaving and joint deduction of cross-disciplinary information, effectively solving the problems of single analytical perspective and weak collaboration in traditional systems.

[0018] Compared to existing technologies, this invention generates control plans by integrating an online simulation-based iterative strategy optimization engine into a digital twin. This engine transforms multi-dimensional risk assessment results into optimization objectives and constraints, performs forward simulation and benefit evaluation on a massive number of possible control action sequences in the twin environment, and introduces multi-perturbation scenario testing to screen for highly robust solutions. This method abandons the static decision-making model that relies on a fixed rule base, instead employing dynamic optimization based on global simulation search. This allows the generated strategy to adapt to specific, changing system states and future risk trends, thereby improving the situational adaptability and optimization capability of control decisions. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall system workflow of the present invention.

[0020] Figure 2 This is a flowchart of the multi-engine collaborative simulation process of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1 As attached Figures 1 to 2 The diagram shows an intelligent substation automation management system. The system architecture consists of a digital twin synchronization management module, a multi-engine collaborative simulation module, an online strategy optimization and simulation module, and an instruction security verification and feedback module.

[0023] The digital twin synchronization management module constructs and dynamically maintains a high-fidelity virtual image that operates synchronously with the physical substation, based on the substation's physical equipment parameters, network topology, and secondary system logic configuration, combined with real-time data streams. This image serves as a unified analysis and simulation environment for the entire system. The multi-engine collaborative simulation module deploys multiple specialized analysis engines within this virtual environment. These engines interact and are event-driven through a shared cognitive state library based on a publish-subscribe model, enabling the fusion analysis and joint simulation of multi-dimensional information such as equipment status, protection logic, and network performance, outputting quantitative comprehensive operational status and risk indicators.

[0024] The online strategy optimization and simulation module receives the aforementioned situation indicators, encodes them into machine-readable state representations, and generates a sequence of optimized control plans for future multi-period changes through parallel simulation and iterative search in a digital twin environment.

[0025] The instruction security verification and feedback module performs multi-level security compliance verification on the executable instructions in the contingency plan. After the verification is passed, the instructions are converted into standard communication protocol instructions and sent to physical devices. At the same time, the module collects device response and system status change data after the instructions are executed.

[0026] The collected feedback data is sent back to the digital twin and analysis engine to drive online calibration of model parameters, post-evaluation of strategy effectiveness, and iterative updates of the system decision model.

[0027] The further specific implementation method is as follows: In the specific implementation of the multi-engine collaborative inference module, its collaborative solution mechanism is implemented through a standardized data exchange protocol and event triggering logic. The shared cognitive state library can be technically instantiated as a distributed message middleware, defining a series of standardized message topics and data formats.

[0028] The device status prediction engine integrates time series prediction models (such as those based on long short-term memory networks), periodically processes device monitoring data, and publishes messages to the topic "Device.Health.Prediction". The message content includes information about the device's status in the future. (For example, within the next 12 hours) at different health levels Probability distribution of (e.g., normal, attention, abnormal, danger) and the confidence level of the prediction. The protection logic analysis engine subscribes to this topic. Upon receiving a new health prediction message, the engine's internal setpoint verification algorithm is activated.

[0029] The algorithm combines real-time collected power flow vectors from the power grid. (Including contributions from each branch road) No merit Current (etc.) and network topology connection matrix The simulation calculates the current settings of relevant protection devices under various possible health degradation scenarios of the equipment. Whether the requirements for selectivity, speed, and sensitivity can still be met, and the safety margin required to maintain protection performance should be calculated. Or a suggested adjustment of the setpoint, the result of which is published to the topic "Protection.Margin".

[0030] The network reliability assessment engine subscribes to both of these topics. It combines real-time performance data obtained from process-layer network probes, such as end-to-end transmission latency of critical packets. (Unit: ms) and packet loss rate Assess the probability of successful execution of critical protection trip commands or automatic control instructions from issuance to reliable execution under a given equipment risk state and protection configuration. .

[0031] Important state updates released by any engine (e.g.) When a danger exceeds a threshold, it is written as an event flag in the message header. The cognitive state library's event listener then automatically notifies all engines subscribed to the relevant topic to initiate a new round of correlation calculations. This mechanism ensures that potential risks to device status are transmitted in real time and automatically to the adaptive analysis of the protection system and the reliability assessment of network services.

[0032] In the online strategy optimization and simulation module, the generation of control plans is modeled as a sequential decision optimization problem in a simulation environment. This module first processes various risk quantification indicators (such as equipment risk values) output by the multi-engine collaborative simulation module. Trend exceeding limits probability Protecting mismatch (etc.) are mapped to a comprehensive environment state vector through a feature encoding network.

[0033] The core of the strategy optimization engine is a model-based planner. It uses the current state as its starting point. As the root node, in the environment model simulated by the digital twin, containing the complete differential-algebraic equations of the power system. In this context, we unfold a policy search tree. Each node in the tree represents a system state, and each edge represents an atomic control action. (e.g., “Circuit breaker CB101”, “Capacitor bank C12”, “Set the protection setting of line L1 to 1.2 times the rated current”).

[0034] The exploration is carried out through four iterative steps: selection, expansion, simulation, and backtracking. In the simulation phase, for a candidate action sequence starting from the root node... ,in Indicates the number of steps (or time window length) in the forward extrapolation model. Used to calculate the predicted state trajectory of the system after executing the sequence. .

[0035] Meanwhile, a pre-defined multi-objective reward function The state-action pair is evaluated at each time step. The function is as follows: ,in This represents the reduction in the total system risk. For operational costs (such as the number of switching actions). For voltage quality scoring, These are the weighting coefficients.

[0036] The long-term cumulative discounted return of this action sequence is Discount factor It determines the present value of future earnings.

[0037] Through extensive parallel simulations, the planner evaluates and compares different action sequences. The value will eventually be output as a single value. The value is the highest and in all simulation steps All of them meet the requirements. The sequence is used as an optimized control plan, in which... This represents all electrical safety constraints (such as equipment current). Bus voltage This process involves a set of rules (systems in transient stable states). This process represents a shift from static rule-based matching to dynamic, forward-looking optimization decision-making based on high-fidelity simulation.

[0038] To enhance the practicality of the generated strategies, an embedded strategy robustness testing unit is included in the online strategy optimization and simulation module. This unit tests the initial high-return scenario set output by the planner. After startup, stress testing and scenario verification are performed. Based on historical operational data statistics or expert knowledge, this unit predefines a set of differentiated perturbation scenarios within the digital twin environment. .

[0039] Each scene This represents a source of uncertainty, such as: the load power being superimposed on the predicted value by a factor with a mean of zero and a standard deviation of [missing value]. Random fluctuations; a non-core auxiliary device is randomly set to experience a momentary failure during the simulation; the delay of a specific communication link is increased by a random value. .

[0040] For each candidate plan The robustness test unit places it under each disturbance scenario. Next, re-execute the complete process. Step-by-step simulation. After the simulation, extract the contingency plan. In each scene The final performance index value (For example, the system's final overall risk value).

[0041] Subsequently, the calculation plan was prepared. In all Performance metrics in various scenarios sample mean and standard deviation The robustness score of a contingency plan can be defined as follows: ,in This is the risk aversion coefficient, used to penalize performance volatility.

[0042] When making a final decision, the system does not solely rely on the reward under the original, undisturbed scenario. Instead, it is a comprehensive consideration. and They tend to choose contingency plans that are stable and reliable under multiple possible "futures," thereby enhancing the adaptability of automatic control strategies to the uncertainties of the real world.

[0043] The digital twin synchronization management module continuously receives measured data vectors from the physical substation monitoring system. Examples include transformer oil temperature, line active power, and bus voltage.

[0044] Meanwhile, the digital twin operates under the same boundary conditions, generating corresponding simulation data vectors, among which... This represents the set of parameters to be calibrated in the model (such as transformer thermal resistance and line impedance parameters).

[0045] The system calculates the residual vector in real time.

[0046] An adaptive monitoring algorithm continuously analyzes the residual sequence. For example, calculating the sliding time window. Root mean square error within ,when continuous The number of cycles exceeded the threshold dynamically calculated based on historical noise levels. At that time, it was determined that the sub-model related to a specific measurement had a bias.

[0047] Subsequently, the parameter identification routine is triggered. For linear or linearizable model relationships, online calibration is performed using recursive least squares with a forgetting factor. Its recursive form facilitates real-time data stream processing: Among them, The parameter estimates updated at each time step; The regression vector is composed of other known state variables or inputs in the model; The parameter estimation error covariance matrix; Here is the gain matrix; forgetting factor. (Typically a value of 0.98) gives new data a higher weight, enabling the algorithm to track the slow time-varying parameters.

[0048] By continuously running this calibration loop, the model parameters of the digital twin are... It can adaptively adjust to ensure that the simulation output matches the physical measurements. Maintaining consistency over the long term provides a reliable basis for all higher-level analyses.

[0049] The system's continuous evolution capability is achieved through a closed-loop learning and evolution management module. This module records each complete "decision-execution" cycle of the system as an experience sample. .in, This is the state before a decision is made. This is the actual sequence of actions performed. It is an instant reward calculated based on actual feedback (such as the actual degree of risk reduction and operating costs). This represents the new state after execution. All experience samples are stored in a distributed experience replay buffer. middle.

[0050] The system periodically (e.g., during periods of low daily load) starts offline learning tasks. These tasks begin from... A batch of empirical data is randomly sampled to update the parameters of the trainable models in the system. For example, the online policy optimization module might use a deep Q-network to approximate the optimal action-value function. ,in These are the network parameters. The learning process updates them by minimizing the temporal difference error. The loss function is: ; in, The parameters are the output of the target network. Periodically from main network parameters Replicate the model to stabilize the training. The parameters of the new model obtained after training are... Before deployment, it must undergo rigorous verification: firstly, its performance must be evaluated on an offline test set to ensure it is no less than the current online model. ; It then enters "shadow mode," where the new model processes real-time data in parallel in the production environment, but the strategies it generates are only used for simulation and recording and are not used for actual control output.

[0051] The system compares the decision-making differences between the shadow model and the online model, and monitors the performance of the shadow model in virtual simulations. Only after running in shadow mode for a sufficient period (e.g., one week) and meeting all monitoring indicators can the new model be promoted to the online primary model through hot-swapping technology. This process ensures the safe, continuous, and automated improvement of the system's intelligence level without human intervention.

[0052] The efficient operation of the multi-engine collaborative inference module relies on its internal collaborative solution scheduling unit. This unit, acting as an intelligent orchestrator, manages the lifecycle and execution logic of each analysis engine. It monitors the event stream in a shared cognitive state library and dynamically schedules events according to a pre-configured rule engine. Rules define the event types and the chains of analysis tasks to be triggered.

[0053] For example, the rule "ONEvent.Device.Health.AlertSEVERITY>'HIGH'DO" The `Schedule(Task.ProtectionRecheck, Priority=HIGH);Schedule(Task.NetworkAssessment, Priority=MEDIUM);` function means that when a high-severity device health alarm event is received, the protection recheck task is immediately scheduled with high priority, followed by the network assessment task scheduled with medium priority. The scheduling unit ensures that tasks are executed in the order of their dependencies and manages their concurrent access to shared resources to prevent data races. This dynamic orchestration mechanism guarantees that critical risk events can trigger a timely, orderly, and complete cross-disciplinary analysis process.

[0054] The instruction security verification and feedback module implements a layered defensive verification strategy. Instructions first undergo syntax and semantic parsing to ensure compliance with the IEC61850 ACSI service model or other established operational specifications.

[0055] Then it enters the topology security verification layer: the verifier is based on the latest real-time topology graph. The command is simulated and executed, and graph theory algorithms are used to analyze whether unexpected electrical islanding or illegal loop paths between different power sources will occur. Power flow calculations are also used to predict whether any equipment will be overloaded. ).

[0056] The third layer is for logic and timing verification: This layer maintains a dynamic list of "electronic operation tickets", checks whether the operation object of the current instruction has been locked by the previous instruction, and verifies whether its operation sequence conforms to logic (to prevent the opening and closing of disconnecting switches under load, and to prevent the accidental opening and closing of circuit breakers, etc.).

[0057] The final layer is for identity and authorization verification, which verifies the digital certificate and role permissions of the initiator of the command, and requires secondary dynamic password confirmation for high-risk operations. Commands that pass all verifications are encapsulated into standard protocol frames (such as IEC60870-5-104ASDU) and sent to the target device via a forward-type physical isolation device. After the command is issued, the module synchronously collects feedback information such as remote signaling changes and electrical quantity changes from relevant devices within milliseconds, forming a precise "command-response" closed-loop record.

[0058] The unified data service module serves as the system's data infrastructure, providing end-to-end data processing capabilities from data acquisition to service. This module deploys multiple protocol adapters (such as IEC61850MMS / GOOSE / SV, IEC104, ModbusTCP) to achieve seamless integration with Layer 3 devices within the station. The acquired raw data enters a stream processing engine, where it is processed sequentially: (1) High-precision time synchronization, aligning the data of each device to a unified PTP or IRIG-B time reference; (2) Data quality assessment, marking each data point with a status code of "valid", "invalid" or "suspicious"; (3) Standardize the format and convert and label it according to a predefined unified data model (such as an extended SCL model).

[0059] The processed, standardized data stream is persisted to a time-series database and simultaneously published to a high-speed data bus.

[0060] Consumers such as the digital twin synchronization management module and the multi-engine collaborative simulation module obtain the required data in real time by subscribing to specific topics on the bus. This module also receives and stores execution result data from the instruction security verification and feedback module, thus forming a unified view and management capability covering the entire data lifecycle.

[0061] The progressive model update process provides strict operational guidelines for system evolution. The newly trained model... First, benchmarking is conducted on a validation platform isolated from the production environment.

[0062] This test uses a historical dataset containing various typical and edge scenarios to comprehensively evaluate... Its performance in key indicators such as accuracy, security, and efficiency will ensure that its overall performance is no less than that of the current online model. .

[0063] After passing the benchmark test Deployed to the production system in "shadow" mode. In this mode, production traffic is copied to the input. However, its output is only used for recording and... The system compares and analyzes the outputs, but does not exert any actual control. It monitors metrics such as decision consistency and strategy returns in a virtual environment.

[0064] After shadow mode runs for a complete business cycle (e.g., covering several days with different load levels), if If performance remains stable and meets preset standards, the system will enter the controlled switchover phase. The switchover strategy involves initially redirecting a small portion (e.g., 5%) of non-core business decision-making traffic. Closely monitor the system status; if no abnormalities are found, gradually increase the traffic ratio until completion. right A complete replacement. Even after the switchover, it remains a hot backup for rapid rollback in case of any unforeseen issues. This process ensures a high degree of controllability and business continuity during the intelligent system upgrade process.

[0065] Based on the detailed description of the system architecture and module functions above, the following section uses a typical operating scenario to illustrate the complete implementation process of the intelligent substation automation management system from situational awareness to decision execution. This process takes the event of "an abnormal rise in oil temperature of a critical 220kV transformer accompanied by an enhanced partial discharge signal" as an example.

[0066] S100: Data Acquisition and Preliminary Alarm. The unified data service module collects the top-level oil temperature of the transformer in real time through oil temperature sensors, partial discharge online monitoring devices, and process-level merging units deployed on the transformer. The temperature rose continuously from 75°C to 82°C (exceeding the warning value of 80°C), while the partial discharge amplitude increased. The data throughput increased from 500pC to 1500pC. After the module performs timestamp synchronization and quality verification on the data, it generates standardized data points and publishes them in real time through the data bus.

[0067] S101: Digital Twin Synchronization and State Injection. The digital twin synchronization management module subscribes to and receives standardized data from S100. It immediately maps this data to the corresponding "Transformer T1" model instance in the digital twin, driving its thermal model and insulation degradation model to perform simulation calculations. Simulation results show that under the current load... and ambient temperature Oil temperature predicted by the model under ℃ conditions It should be 78℃, which is different from the measured value. ℃ exists The residual at ℃.

[0068] The residual monitoring algorithm determines that the persistent deviation exceeds the adaptive threshold. At ℃, an online calibration routine for the thermal model parameters of the transformer is triggered, and the heat dissipation coefficient is fine-tuned using the recursive least squares method. These parameters make the twin model more closely resemble the actual state of the current physical equipment.

[0069] S102: Multi-engine collaborative simulation initiated. The equipment status prediction engine in the multi-engine collaborative simulation module detected anomalies in the transformer status data within the twin in real time. This engine invoked its internal LSTM prediction model to calculate future... The probability of transformer failure within one hour It rose sharply from 0.05 to 0.35, that is... Prediction confidence The prediction result As a key event, the "Device.Health.Alert" topic was published by the Device Status Prediction Engine to the shared cognitive status library.

[0070] S103: Protection Logic Correlation Analysis. The protection logic analysis engine subscribes to the "Device.Health.Early Warning" topic. Upon receiving an event indicating an increased risk for transformer T1, the engine is immediately activated by the collaborative solution scheduling unit. The engine obtains the current power grid topology connection matrix. With trend vector And combined with the received risk probability Dynamic verification is performed.

[0071] The verification analysis suggests that if the transformer experiences a sudden internal fault under heavy load, its differential protection setting will be... Although it can operate, the setting of the adjacent 110kV side backup overcurrent protection is incorrect. The matching sensitivity margin The value will drop below the critical value of 0.1, posing a risk of protection exceeding its limit. The analysis concludes that "backup protection needs to be..." Temporary reduction of fixed value "To maintain sufficient margin" along with the recommended margin adjustment It was posted under the topic "Protection. Dynamic Adjustment".

[0072] S104: Network Reliability Assessment Linkage. The network reliability assessment engine subscribes to the "Protection. Dynamic Adjustment" topic. Regarding the soft modification instructions for protection settings proposed in S103, the engine assesses the criticality of their execution.

[0073] It queries the process layer network status and finds that the current load rate of the core switching link carrying the GOOSE messages of the protection device is... Average latency The engine calculates, using a queuing-based transmission model, the expected success rate of issuing the fixed-value modification command and ensuring its reliable reception and execution under this network condition. Meeting the reliability threshold Requirements. The assessment results were published.

[0074] S105: Online policy optimization generates contingency plans. The integrated situational awareness interface of the online policy optimization and simulation module aggregates high-risk device states from S102, weak protection coordination analysis from S103, and reliable network command channel information from S104, fusing and encoding these into an environmental state vector. .

[0075] The strategy optimization engine is activated to... Starting with a calibrated digital twin, multi-step simulations were performed. Various control schemes were simulated. The future performance of each scheme was calculated. Long-term cumulative discount rewards within hours (in , reward function Taking into account risk reduction, power supply reliability, and operational costs, the engine selects an optimal sequence. Step 1 (A1): Immediately switch the load on transformer T1 from [unclear - possibly a device or procedure] via remote control. Transfer To adjacent bus (target load) The second step (A2) involves simultaneously issuing instructions to adjust the settings of the relevant 110kV side backup protection. Depend on Temporarily downgraded to (i.e., downgrade) The contingency plan was submitted to the strategy robustness testing unit.

[0076] S106: Strategy robustness testing. The strategy robustness testing unit tests the contingency plan. Stress testing was conducted. Three perturbation scenarios were constructed within the digital twin. .

[0077] After the simulation is completed, the contingency plan is calculated. Final system risk value in all scenarios mean with standard deviation Its robustness score (Pick If the value exceeds the threshold, the contingency plan is ultimately adopted.

[0078] S107: Command Security Verification and Issuance. Command Security Verification and Feedback Module Receive Contingency Plan. The verification process is as follows: 1. Semantic Validation: Parsing A1 as "Opening circuit breaker CB101, closing circuit breaker CB102", and A2 as "Setting the protection device PR123 setting value Point.setVal is..." ", conforms to the standard model.

[0079] 2. Topology error prevention verification: After simulating and executing A1, verify that the power grid has no islanding and that the target bus load is within acceptable limits. ; Verify the new value of A2 It is greater than the minimum operating current.

[0080] 3. Timing conflict check: Confirm that no other instructions are locking CB101, CB102 and PR123.

[0081] 4. Permission and Password Verification: Verifies that the automation policy engine has "emergency control" permissions. After all verifications pass, the module converts the command into a standard protocol frame and distributes it through the security isolation gateway.

[0082] S108: Physical Execution and Feedback Acquisition. The measurement, control, and protection devices of the physical substation execute commands. The command safety verification and feedback module synchronously acquires feedback: CB101 and CB102 are in the correct position; the real-time load current of transformer T1... Down to The PR123 protection device's calibration settings have been modified. This data is packaged into feedback data packets in real time. .

[0083] S109: Feedback Loop and Model Evolution. Feedback data packets are sent back to the unified data service module for storage and simultaneously pushed to relevant modules. The digital twin uses measured load and temperature data to verify the model's prediction accuracy.

[0084] The closed-loop learning and evolution management module will generate experience tuples from the data of this entire process. Store in the experience replay pool Among them, the actual reward This data is calculated based on the actual degree of risk reduction. This empirical data will be used for subsequent offline training to update the parameters of the device prediction model and the policy optimization model, driving the continuous evolution of the system's decision-making intelligence.

[0085] The steps S100 to S109 above constitute a complete closed loop of automated management, demonstrating that the system starts from raw data and, through digital twin synchronization, multi-engine collaborative simulation, online strategy optimization and robustness testing, and multi-level security verification, ultimately achieves safe, optimized, and adaptive control, and completes the complete logical chain of experience accumulation and self-evolution.

[0086] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent substation automation management system, characterized in that, include: The digital twin synchronization management module is used to build and maintain a digital twin of the substation that operates synchronously with the physical substation. The multi-engine collaborative simulation module is connected to the digital twin synchronization management module and is used to run multiple professional analysis engines in the digital twin that interact through a shared cognitive state library to generate operational situation simulation results and risk quantification indicators. The online strategy optimization and simulation module connects the digital twin synchronization management module and the multi-engine collaborative simulation module, and is used to generate a multi-step optimization control plan through online simulation iteration based on the simulation results and indicators. The instruction security verification and feedback module is connected to the online strategy optimization and simulation module. It is used to perform security verification on the immediate execution instructions in the contingency plan, and after passing the verification, the instructions are issued for execution and response data is collected. The instruction security verification and feedback module feeds back the collected response data to the digital twin synchronization management module and the multi-engine collaborative simulation module.

2. The intelligent substation automation management system according to claim 1, characterized in that, The professional analysis engine performs collaborative computation through the shared cognitive state library in the following ways: The cognitive state library stores standardized intermediate results and event labels generated by each engine; The first analysis engine publishes the device's predictive state range and confidence information to the cognitive state database; The second analysis engine subscribes to the state interval and confidence information, dynamically calculates the coordination logic and adjustable margin of the protection setpoint, and publishes it to the cognitive state library. The third analysis engine integrates the adjustable margin and real-time network status to assess the expected reliability indicators of key control service links and publishes them to the cognitive state library. Among them, a key information update event published by any engine triggers other engines that have subscribed to the event to start associated calculations.

3. The intelligent substation automation management system according to claim 1, characterized in that, The online strategy optimization and simulation module generates multi-step optimization control plans in the following ways: The digital twin is used as a simulation environment, and the operational situation simulation results are encoded as an environmental state vector. In the simulation environment, different control action sequences are simulated and executed forward according to the preset system optimization objectives and safety constraints; By comparing the long-term comprehensive benefits generated by each control action sequence during simulation, the action sequence that satisfies all safety constraints is iteratively selected as the output plan.

4. The intelligent substation automation management system according to claim 3, characterized in that, The online policy optimization and simulation module also includes a policy robustness testing unit, which performs the following: Construct two or more different future perturbation scenarios for candidate control action sequences in a simulation environment; Parallel execution of multi-step look-ahead simulations of each candidate control action sequence under different disturbance scenarios; Quantitatively evaluate the stability of key performance indicators of each candidate control action sequence under different disturbance scenarios; The strategy optimization engine selects a sequence of control actions as the output plan based on the quantitative evaluation results.

5. The intelligent substation automation management system according to claim 1, characterized in that, The digital twin synchronization management module dynamically maintains the substation digital twin in the following ways: It continuously receives real-time data streams from physical substations and synchronously maps them to the corresponding models in the digital twin; Calculate the residual between the digital twin simulation data and the physical station measured data; When the statistical characteristics of the residuals exceed the adaptive threshold, the model parameter calibration routine is initiated to adjust the simulation parameters of the corresponding model in the digital twin based on the residuals.

6. The intelligent substation automation management system according to claim 1, characterized in that, It also includes a closed-loop learning and evolution management module, which: The simulation process data, the sequence of executed strategy instructions, and the corresponding physical feedback results are serialized into structured empirical data and stored. Data is sampled from the historical experience database to retrain and optimize the parameters of the internal evaluation model of the professional analysis engine and the decision logic parameters of the strategy optimization engine. Control the incremental model update process and switch the validated model parameters and decision logic to an online state.

7. The intelligent substation automation management system according to claim 2, characterized in that, The multi-engine collaborative simulation module also includes a collaborative solution scheduling unit, which: Manage the activation cycles of the first analysis engine, the second analysis engine, and the third analysis engine; Listen to the event tags in the cognitive state database, and schedule the correlation analysis engine to start the calculation according to the event type and predefined rules; Coordinate the timing of data exchange between various analysis engines.

8. The intelligent substation automation management system according to claim 1, characterized in that, The security compliance verification performed by the instruction security verification and feedback module includes: Perform semantic analysis and compliance verification of instructions; Connectivity verification and load calculation are performed by combining the real-time power grid topology; By comparing the instruction sequence, conflict detection is performed between the occupancy status of the operation object and the timing logic; Verify the operational permissions of the entity that issued the instruction; Once the instruction passes all verifications, it is allowed to be converted and sent.

9. The intelligent substation automation management system according to claim 1, characterized in that, It also includes a unified data service module, which: Collect multi-source heterogeneous real-time data from equipment at the station control layer, bay layer, and process layer; The collected data undergoes protocol parsing, timestamp synchronization, and formatting to generate a standardized data stream. The standardized data stream is distributed to modules that subscribe to the data, and feedback data is received and stored.

10. The intelligent substation automation management system according to claim 6, characterized in that, The closed-loop learning and evolution management module controls the incremental model update process, including: Validate the new model offline in a test environment using historical or simulation data; The old and new models were run online in parallel using shadow mode to compare the differences in their outputs. Once the preset conditions are met, the new model that meets the performance standards will be switched to the online primary model.

Images