A cloud platform-based power dispatching system

Abstract

The application belongs to the technical field of power dispatching management, and provides a power dispatching system based on a cloud platform, in order to solve the problem of low technical accuracy of power dispatching in traditional technologies, the system provides global and high-credibility decision basis for abnormal identification by fusing five layers of data of physical entities, control execution, model algorithms, business management and system operation and maintenance based on the cloud platform; through the layered diagnosis logic of 'wisdom first, then physics', the reliability of data and models is verified first, ensuring the accuracy of subsequent decision input; after confirming the abnormality of the physical system, the digital twin model is used for hypothesis simulation and abnormal propagation chain construction, realizing accurate positioning from abnormal appearance to physical root cause equipment; finally, alarm information is generated based on the traceability result, which can significantly improve the accuracy and reliability of abnormal monitoring of the power dispatching system, and ensure the accuracy and efficiency, intelligence and automation of power supply dispatching and control.

Description

Technical Field

[0001] This invention relates to the field of power dispatching and management technology, and in particular to a power dispatching system based on a cloud platform. Background Technology

[0002] Power dispatching is the core command and control link of the urban power system. Its core task is to achieve safe, stable and efficient power supply by coordinating the control of facilities such as power plants, substations and power lines according to dynamic power demand.

[0003] In traditional technologies, power dispatch is generally based on automated dispatching that combines Supervisory Control and Data Acquisition (SCADA) systems with fixed rules. This typically includes: deploying sensors at key nodes of power lines to collect physical data such as voltage and current; remotely transmitting the data to the dispatch center for centralized monitoring; and remotely controlling or alarming equipment such as substations and power lines based on preset fixed threshold rules or dispatcher experience.

[0004] However, the inventors discovered the following problems in traditional technologies, leading to low accuracy and efficiency in anomaly monitoring and handling: First, traditional technologies rely entirely on the collected raw data for decision-making. When the collected data is inaccurate, incorrect scheduling decisions will be made based on the wrong data. Second, after an anomaly is identified, traditional technologies often remain at the level of "anomaly alarm" (such as "low voltage in a certain area"). The response relies on the dispatcher's personal experience for manual investigation, and the cause of the anomaly cannot be determined.

[0005] In summary, the core problem with traditional power dispatching technology lies in its passive response, reliance on manual intervention, and low accuracy in monitoring power dispatching anomalies under complex conditions.

[0006] Therefore, improving the technical accuracy of power supply dispatch anomaly monitoring has become an urgent problem to be solved in the field of power supply management and power dispatching systems. Summary of the Invention

[0007] The technical problem solved by this invention is to address the low accuracy of power supply dispatch anomaly monitoring in traditional power management technologies.

[0008] To address the aforementioned technical problems, the present invention provides the following technical solution: a cloud-based power dispatching system, comprising: a first determining module, configured to determine a target power supply line based on a power supply dispatching platform in cloud-based intelligent power management, in response to power dispatching monitoring instructions; a first acquiring module, configured to acquire multi-dimensional data of the target power supply line, the multi-dimensional data including physical entity layer data, control execution layer data, model algorithm layer data, business management layer data, and system operation and maintenance layer data; a first identifying module, configured to identify whether an anomaly exists in the target power supply line based on the multi-dimensional data, using a preset rule engine and / or anomaly detection model; and a first judging module, configured to judge whether the power dispatching cloud platform exists if an anomaly is identified. The power dispatching system is abnormal; the first tracing module is used to trace the source to the inaccurate target data source through fault diagnosis, or to trace the source to the performance degradation target algorithm model through performance monitoring, if the power dispatching system is determined to be abnormal; the second judgment module is used to determine whether there is a physical system abnormality in the target power supply line if the power dispatching system is determined not to be abnormal; the first construction module is used to construct an abnormality propagation chain from the root cause device to the abnormal manifestation by setting an abnormality hypothesis and simulating it in the digital twin power model if the physical system abnormality is determined to be present, so as to locate the physical system root cause that leads to the physical system abnormality; the first alarm module is used to generate and output alarm information based on the tracing result of the power dispatching system abnormality or the physical system root cause.

[0009] As a preferred embodiment of the cloud-based power dispatching system of the present invention, the first construction module includes: a first setting submodule, used to set one or more abnormal assumptions about equipment failure or power line topology changes in the digital twin power model; a first obtaining submodule, used to drive the digital twin power model to perform simulation and obtain the power grid state simulation results under each of the abnormal assumptions; a first calculation submodule, used to calculate the matching degree between the power grid state simulation results and the corresponding actual physical entity layer data; and a first determining submodule, used to determine the abnormal assumption with the highest matching degree as the most likely root cause device causing the physical system abnormality.

[0010] The beneficial effects of this invention are as follows: By constructing a cloud-based power dispatching system that integrates multi-dimensional data perception, intelligent anomaly identification, hierarchical diagnosis, and precise source tracing, a closed-loop intelligent handling of power line anomalies is achieved. Its core improvements lie in the following: First, by integrating five layers of data—physical entities, control execution, model algorithms, business management, and system operation and maintenance—a global and highly reliable decision-making basis is provided for anomaly identification and diagnosis, overcoming the shortcomings of traditional systems with their single data dimension and fragile decision-making foundation. Second, by introducing a hierarchical diagnostic logic of "intelligence first, then physical," the system can first verify the reliability of data and models to rule out anomalies within the power dispatching system itself, ensuring the accuracy of subsequent decision inputs. Then, after confirming physical system anomalies, a digital twin model is used for hypothetical simulation and anomaly propagation chain construction, achieving precise location from anomaly symptoms to the physical root cause device, solving the problems of traditional methods relying on human experience, slow location, and only addressing the symptoms without addressing the root cause. Finally, based on precise source tracing results, alarm information containing specific targets, paths, and handling suggestions is generated, upgrading traditional simple status alarms into directly executable diagnostic reports. Therefore, the traditional passive, open-loop, and manual power management power dispatch and monitoring mode is transformed into a closed-loop intelligent system capable of self-diagnosis, proactive tracing, and precise decision-making. This significantly improves the accuracy, efficiency, and reliability of power dispatch anomaly monitoring and handling in power management, and ensures the accuracy, efficiency, intelligence, and automation of power dispatch and control in power management. Attached Figure Description

[0011] Figure 1 A schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention;

[0012] Figure 2 This is the first sub-schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention;

[0013] Figure 3 This is a second schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0015] This invention provides a cloud-based power dispatching system. The system can be applied to cloud environments, including but not limited to cloud platforms, and can be used for power dispatching in fields including but not limited to power management and power dispatching.

[0016] Example 1, please refer to Figure 1 , Figure 1 This is a schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention. Figure 1 As shown, in this embodiment, the cloud-based power dispatching system 100 includes a first determination module 101, a first acquisition module 102, a first identification module 103, a first judgment module 104, a first tracing module 105, a second judgment module 106, a first construction module 107, and a first alarm module 108. The detailed descriptions of each of these functional modules are as follows:

[0017] The first determining module 101 is used to determine the target power supply line in response to power dispatch monitoring instructions based on the power supply dispatch platform in cloud-based intelligent power management.

[0018] Explained, the power supply dispatching platform in cloud-based smart power management refers to a software system deployed on cloud computing infrastructure that integrates core functions such as IoT data acquisition, power model simulation, big data analysis, and artificial intelligence decision-making to achieve centralized monitoring, intelligent analysis, and optimized dispatching of urban power supply systems.

[0019] A power dispatch monitoring command is a command that triggers the platform to start a power dispatch monitoring and analysis task. This command can be event-driven or scheduled task-driven.

[0020] The target power supply line refers to the logical subset of power supply lines that are designated as the scope of this analysis and diagnosis under a single monitoring command. It is a dynamic area of ​​interest determined based on the command content.

[0021] Based on the above description and settings, in response to power dispatch monitoring instructions, the platform determines the target power supply line according to the following logical steps: Step 1, Instruction parsing and context construction. After receiving the monitoring instruction, the platform first parses its type and content. For example, if the instruction originates from a "user complaint work order (location: X community, problem: low voltage)," the platform will extract the key geographical location information "X community" and call the business management system to obtain the DMA number, voltage zone, and other context information of the community. Step 2, Initial screening of the target range based on GIS and power model. The platform matches the core information in the instruction (such as geographical location and DMA number) with the power supply line geographic information system (GIS) to initially determine a physical range. Then, based on this, the platform calls the digital twin power model. The digital twin power model refers to a virtual dynamic simulation model constructed by computer software that has geometric, physical, behavioral, and rule-consistent characteristics with the physical power supply line system. Related digital twin technologies are referenced from existing technologies and will not be elaborated here. For example, regarding a complaint about Community X, the model will use that point as the center and, through reverse power tracing, simulate and identify all upstream power lines, power plants, and substations that might affect the voltage at that point. These related facilities will be initially screened as potential target power lines. Step three involves multi-factor optimization and final determination. The platform comprehensively considers the required analytical accuracy, real-time computing resource constraints, and ease of operation to optimize the initial screening range, ultimately determining a target power line with clear boundaries that is easy to monitor and analyze. For example, in the case of Community X, if the initial screening range is too large, the platform may prioritize identifying the DMA (Districted Area) directly supplying that community as the target power line to quickly focus on the problem; if the data within that DMA is insufficient for diagnosis, the range will be gradually expanded to the next higher voltage zone.

[0022] Therefore, by identifying the target power supply line through the above process, the scope of analysis is dynamically, accurately, and intelligently defined. This allows vague business issues (such as a power complaint) to be transformed into clear technical analysis objects, laying a solid foundation for subsequent multi-dimensional anomaly identification and precise root cause tracing. It ensures that the entire method is targeted and feasible from the outset, avoids the waste of computing resources in "full network scanning," and greatly improves the efficiency of power dispatch monitoring and analysis.

[0023] The first acquisition module 102 is used to acquire multi-dimensional data of the target power supply line, including physical entity layer data, control execution layer data, model algorithm layer data, business management layer data, and system operation and maintenance layer data.

[0024] Interpretive, multi-dimensional data represents a dataset describing the operational status, control logic, decision-making process, business environment, and system health of a target power supply line from different levels and perspectives. The data at each layer are not isolated but constitute an interconnected and mutually verifying organic whole. Their specific definitions, content, and collaborative relationships are as follows:

[0025] Physical entity layer data represents raw data that directly reflects the physical operating status of power supply lines and electrical equipment. It originates from sensors deployed on physical equipment such as transmission and distribution lines, transformers, switchgear, and generators, including but not limited to voltage, current, frequency, power (active / reactive), power quality indicators (harmonics, voltage sags, flicker, etc.), temperature (equipment temperature rise, joint temperature), partial discharge, equipment vibration, insulation status, and power flow data.

[0026] Control execution layer data represents the interactive data reflecting the system's control intent and actual execution status. This includes control commands (such as target frequency and target opening degree) issued to actuators (e.g., generators, switchgear) and the actual status feedback from the actuators (e.g., actual frequency, actual opening degree, fault signals), used to determine whether the control loop is responding normally. When the control commands and feedback status are inconsistent, it can directly indicate an anomaly in the power dispatching system (e.g., actuator failure). This data, in conjunction with physical entity layer data, is used to verify whether changes in physical state are caused by control commands. For example, is a voltage drop the result of normal dispatch commands (control layer) or caused by a line anomaly (physical layer)?

[0027] Model algorithm layer data represents the data generated by the power dispatching system during the analysis and decision-making process. It reflects the system's analysis and decisions, including but not limited to the inputs (historical data, weather forecasts) and outputs (predicted values) of the electricity consumption prediction model, the generator combination schemes generated by the optimized dispatching algorithm, the simulation results of the digital twin power model, and the performance evaluation indicators of these models (such as prediction error and simulation deviation). It is the core of realizing intelligent dispatching and predictive diagnosis, and is also used to monitor the health of the model itself. Its output (such as predicted values) is the basis for generating control execution layer instructions, and its input and performance data are used to determine whether there are any abnormalities in the power dispatching system itself (algorithm model).

[0028] Business management layer data refers to data that provides the background, environment, and historical context of system operation. It comes from the business operation system in power management and includes, but is not limited to, user complaints and work order information, power supply line asset information (wire type, age), planned maintenance and repair arrangements, and external environmental data (weather, major events). It is used to distinguish between real anomalies and planned operations (such as voltage fluctuations), assist in root cause analysis (such as prioritizing old lines), enrich alarm information (such as determining the scope of affected users), and provide background information for anomaly judgment at the physical layer and model layer, avoid false alarms, and improve the practicality of tracing results.

[0029] System operation and maintenance layer data represents data reflecting the health of the IT infrastructure of the cloud-based smart power management platform itself, including but not limited to server CPU / memory utilization, data communication network latency and packet loss rate, data transmission heartbeat signals, and microservice process status. It is the direct basis for diagnosing power dispatching system anomalies (such as data interruption or computing service unavailability). When physical entity layer data is abnormal, it is necessary to first determine whether it is caused by network interruption or server failure through this layer data, so as to quickly eliminate IT infrastructure problems.

[0030] Based on the above concept and description, acquiring multi-dimensional data of the target power supply line is achieved through a cloud data bus, specifically through the following steps: 1) The platform establishes a unified data interface specification for each layer of data. Data from the physical entity layer and control execution layer is collected in real time from on-site devices such as PLCs and RTUs via IoT gateways and uploaded to the cloud. Data from the model algorithm layer is generated by the platform's internal computing services and published to the data bus. Data from the business management layer and system operation and maintenance layer is extracted periodically or triggered from the corresponding business management systems (such as GIS, work order systems) and cloud platform monitoring tools via API interfaces. 2) All data is aggregated in the cloud data platform. By establishing a unified spatiotemporal data model, multi-source data with "device ID," "timestamp," and "spatial coordinates" as the core are aligned, cleaned, and associated. For example, the "voltage reading (physical layer)" at a certain moment is associated with the "switch equipment opening command (control layer)" and the "maintenance record of the line (business layer)" to form a data entity with a complete context. 3) Data services and supply: The fused data, after being processed, is stored in a real-time database or data lake and encapsulated into standard data service interfaces. When the system performs anomaly identification, diagnosis, or tracing tasks, it can call the required multi-dimensional data from these interfaces in a one-time, correlated manner, rather than querying from multiple isolated systems separately.

[0031] Therefore, by systematically acquiring and integrating data from the above five levels, a leap from "isolated data to contextualized information" can be achieved. This provides a global and context-rich decision-making basis for power supply dispatch anomaly monitoring and root cause tracing in subsequent power management, enabling the system to distinguish whether the cause is "physical system", "power dispatch system" or "external environmental interference". This lays an irreplaceable data foundation for achieving precise and intelligent closed-loop control.

[0032] The first identification module 103 is used to identify whether there is an anomaly in the target power supply line based on the multi-dimensional data and through a preset rule engine and / or anomaly detection model.

[0033] Explained, a pre-configured rule engine refers to a software module that is pre-configured with deterministic logical judgment conditions. Based on expert knowledge and operational standards in the power management field, it transforms abnormal patterns into explicit logical judgment rules. For example, pre-configured rules include, but are not limited to, the following examples: IF (voltage at a voltage monitoring point < 1000V (AC)) AND (this state lasts > 120 seconds) THEN trigger "low voltage abnormality"; IF (generator control command is "start") AND (feedback state is "stop" lasts > 30 seconds) THEN trigger "control execution abnormality".

[0034] An anomaly detection model refers to a statistical model built based on machine learning algorithms that can autonomously learn normal operating patterns from historical data and identify deviations from those patterns. For example, an anomaly detection model could be an isolated forest model, used to identify "outliers" in current or voltage data that behave significantly differently from the majority of data, effectively detecting line anomalies. Alternatively, an anomaly detection model could be an LSTM autoencoder model, which learns the normal variation patterns of parameters such as current and voltage at different times of the day. When the error between the actual data and the model-reconstructed data exceeds a threshold, it is judged as an anomaly, effectively identifying slow-developing abnormal migrations in electricity consumption patterns.

[0035] Based on the above concept and setup, anomaly identification of the target power supply line can be implemented through the following steps: 1) Extract features related to the health status of the target power supply line from the aggregated multi-dimensional data. These features include, but are not limited to: real-time values, moving averages, and trends of physical entity layer data; command-feedback deviation of control execution layer data; and prediction-actual deviation of model algorithm layer data. These features are combined into a comprehensive feature vector to characterize the operating status of the power supply line at the current moment. 2) For the rule engine path, the feature vector is input into the rule engine. The engine traverses all preset rules. For each rule triggered, a rule anomaly score is generated for the anomaly event based on the severity of the rule. For the anomaly detection model path, the same feature vector is input into the trained anomaly detection model. The model outputs a model anomaly probability between 0 and 1, representing the degree to which the current state deviates from the historical normal pattern. 3) Using a weighted average or voting mechanism, the anomaly score of the rule engine and the anomaly probability of the anomaly detection model are fused to obtain a comprehensive anomaly index. When this index exceeds a preset global judgment threshold, the system finally determines that "the target power supply line is abnormal" and activates the subsequent process.

[0036] Thus, by combining a rule engine based on explicit knowledge with a data-driven anomaly detection model, complementary advantages are achieved. The rule engine ensures that known and deterministic fault modes can be captured quickly and reliably, while the anomaly detection model empowers the system to discover unknown, complex, and potential anomalies. The fusion of the two makes anomaly identification both reliable and sensitive, significantly reducing the false alarm and missed alarm rates, and providing accurate and reliable trigger signals for the entire intelligent monitoring system.

[0037] The first judgment module 104 is used to determine whether there is an abnormality in the power dispatching cloud platform if an abnormality is detected.

[0038] Explained, in this embodiment of the invention, "power dispatch system" refers to the collection of software, algorithms, and data links responsible for perception, decision-making, and control in a cloud-based intelligent power management and power supply dispatch platform. Its core functions include data acquisition and transmission, model calculation and prediction, and generation and issuance of control commands. The power dispatch cloud platform carries and runs the overall cloud computing environment and software framework of the "power dispatch system".

[0039] An anomaly in the power dispatching system indicates that the system is malfunctioning or failing. This could be due to unreliable data or model outputs provided to the decision-making process, which includes two core issues: 1) Inaccurate data sources, mainly due to faults in the sensing system, such as sensor drift, communication interruption, or data tampering, leading to distortion of the collected physical entity layer data; 2) Degraded decision-making models, mainly due to faults in the analytical decision-making system, such as performance degradation of predictive models due to data drift or optimization algorithms outputting invalid solutions due to inappropriate parameters.

[0040] Based on the above concept and description, after identifying an anomaly in the target power supply line, the following prior diagnostic logic can be used to determine whether it is an anomaly in the power dispatching system, specifically including but not limited to the following steps: 1) The system utilizes the inherent correlation between multi-dimensional data to cross-validate the reliability and authenticity of the physical entity layer data; 2) Check whether the readings of other sensors near the abnormal data point that have power correlation are normal and conform to power laws. If only one point is abnormal while the surrounding area is normal, it indicates that there is inconsistency between the correlated data, and the sensor at that point is abnormal; 3) Compare the control commands and equipment status feedback in the control execution layer. If the commands and feedback are inconsistent for a long period of time, it is determined that there is an anomaly in the power dispatching system in the execution link (such as PLC, switching equipment). 4) Using the current control commands and some reliable data as boundary conditions, drive the digital twin power model to calculate the theoretical value of the predicted abnormal sensor. If the theoretical value deviates significantly from the measured value, the sensor data is confirmed to be inaccurate. 5) Check the communication link status of the abnormal data stream (such as network latency and packet loss rate). If the communication is abnormal, it is determined to be an abnormality in the power dispatching system. 6) Monitor the model algorithm layer data. For example, calculate the error between the predicted value and the actual value of the key prediction model (such as electricity consumption prediction). If the error continues to exceed the threshold, it is determined that the model performance has degraded and belongs to the abnormality in the power dispatching system. 7) Make a comprehensive judgment. If any of the above verification steps 1) to 6) confirms that the data is unreliable or the model has degraded, it is comprehensively judged that "there is an abnormality in the power dispatching system" and the process jumps to the power dispatching system tracing. Otherwise, it is judged that "there is no abnormality in the power dispatching system", which means that the physical system of the power supply line is very likely to have a real fault, and the process enters the physical system abnormality judgment.

[0041] Therefore, by introducing the "prior self-diagnosis" step, it is ensured that all subsequent analyses and decisions are based on reliable data and models, thereby preventing the system from making erroneous interventions on the physical system of the power supply line due to its own faults, and greatly improving the reliability and intelligence of the system in the entire power management.

[0042] The first tracing module 105 is used to trace the source to the inaccurate target data source through fault diagnosis, or to trace the source to the target algorithm model with degraded performance through performance monitoring, if an anomaly in the power dispatching system is determined.

[0043] Explained, the target data source refers to the specific data source that is predicted and ultimately confirmed as providing distorted data in the anomaly judgment process of the power dispatching system. It usually refers to a specific physical sensor (such as a voltage transmitter, ammeter, etc.) or its data stream, but also includes a link in the data transmission link (such as a specific RTU remote terminal unit), which is the output object of the aforementioned anomaly judgment steps.

[0044] The target algorithm model refers to a specific algorithm that is predicted and ultimately confirmed as having unreliable output results or significantly degraded performance in the anomaly judgment process of the power dispatching system. Common target algorithm models include, but are not limited to: short-term electricity consumption prediction models, optimized dispatching models, and digital twin power models themselves.

[0045] Based on the above concept and description, after determining that there is an anomaly in the power dispatching system, precise location can be achieved through the following methods:

[0046] 1) Trace the source to the inaccurate target data source. For the inaccurate data source problem, locate the specific fault point through a multi-level diagnostic strategy, including but not limited to:

[0047] A) Based on the power topology, identify other sensors that are physically adjacent to or electrically related to the abnormal data point. If the data of these related sensors are consistent and normal, narrow down the abnormal range to the sensor or sensors that initially triggered the abnormality and identify them as the "target data source". The power topology refers to the spatial logical relationship between various devices (such as lines, nodes, generators, switching equipment, and transformers) in the power supply line system based on the physical connection sequence and current direction. Its core lies in the connectivity and power dependence between components.

[0048] B) Using a digital twin power model and other reliable data as input, the theoretical value of the predicted abnormal sensor is calculated through simulation. The theoretical value is continuously compared with the measured value. If the deviation is stable and significant, the sensor can be confirmed as the "target data source" that is out of sync. The digital twin power model refers to a virtual dynamic simulation model constructed by computer software that has geometric, physical, behavioral and rule consistency with the physical power supply line system. It is a complex calculation engine that integrates the power supply line topology, power mechanics principles, equipment operating characteristics and historical data. Its construction can refer to the corresponding digital twin technology or simulation technology, which will not be elaborated here.

[0049] C) Inspect the data transmission path from the sensor to the cloud and analyze system operation and maintenance data, such as signal strength, packet loss rate, and communication latency. If an anomaly is found in this link, the root cause of the problem is confirmed to be in the communication link, and this link is regarded as an inaccurate "target data source".

[0050] 2) Tracing the source to the target algorithm model experiencing performance degradation, and pinpointing the model degradation problem through quantitative evaluation, specifically including but not limited to:

[0051] A) For critical prediction or optimization models, the system continuously calculates their key performance indicators. For example, for electricity consumption prediction models, it continuously calculates the mean absolute percentage error (MAPE) between the predicted and actual values; for optimization scheduling models, it detects the achievement rate between the actual execution effect (such as power consumption) and the expected effect.

[0052] B) The system analyzes the trend of performance indicators within a preset time window (e.g., 24 hours). When an indicator (e.g., MAPE) exceeds a preset threshold for multiple consecutive periods or shows a clear and continuous abnormal trend, the model is judged to have degraded in performance and is identified as the "target algorithm model".

[0053] For example, assuming the system has determined that "there is an anomaly in the power dispatch system of cell X", the system enters the tracing process at this stage: 1) The system has previously predicted the inaccuracy of the DMA inlet ammeter through digital twin simulation. At this time, further correlation positioning is performed to confirm that all voltage data in the DMA are consistent and normal, thus focusing entirely on the voltmeter. At the same time, no anomalies are found in the communication link. Therefore, the system finally traces the source of the inaccuracy to the target data source: voltmeter FT-001 at the DMA inlet. 2) Assuming that when analyzing the same anomaly, the system also finds that the predicted value of the power consumption prediction model for cell X deviates greatly from the actual value, and the MAPE calculated by the performance monitoring module for this model has exceeded 15% continuously in the past 6 hours (the preset threshold is 10%), the system simultaneously traces the source to the target algorithm model with degraded performance: the power consumption prediction model V2.1 for cell X.

[0054] Thus, by accurately locating the target data source or target algorithm model, it provides clear action instructions for subsequent maintenance (such as repairing or calibrating sensors) or system updates (such as retraining AI models). It can transform vague "power dispatch system anomaly" alarms into specific fault points that can be located and operated, realizing the precision and efficiency of system operation and maintenance in smart power management, and completing a key link in the self-diagnosis to self-repair closed loop.

[0055] The second judgment module 106 is used to determine whether there is a physical system abnormality in the target power supply line if it is determined that there is no abnormality in the power dispatching system.

[0056] Explained, in this embodiment of the invention, the physical system refers to the set of all physical devices, lines, etc. in the target power supply line, including but not limited to lines, generators, and switching equipment, which are the power supply service carriers. The target power supply line is a logical subset of the physical system defined in this monitoring and analysis.

[0057] A physical system anomaly indicates that the physical entity's physical state, electrical behavior, or electrical indicators deviate from the normal operating range due to equipment failure, line damage, or other external interference. It is a real fault in the power supply line's physical system, rather than an error in the data or model.

[0058] Based on the above concept and description, assuming there are no anomalies in the power dispatching system (i.e., the data is reliable and the model is healthy), the presence of anomalies in the physical system can be determined through methods including but not limited to the following:

[0059] 1) The system uses verified and reliable physical entity layer data (such as reliable voltage and current of key nodes) as boundary conditions to drive the digital twin power model to perform real-time power simulation of the entire power supply line, calculating the theoretical power state (including voltage and current) of each node and line within the target power supply line. 2) The system compares the full-network simulation results of the digital twin model with actual, global physical entity layer monitoring data, calculating the comprehensive matching degree between the simulated and measured values ​​(e.g., calculating the root mean square error (RMSE) of the simulated voltage and current values ​​at all measurement points). 3) A preset matching degree threshold is set. If the calculated global matching degree is higher than the preset threshold, it indicates that the actual operating state of the physical system highly matches the model's expectation under normal conditions, and it is determined that "no physical system anomaly exists." This alarm may be a brief disturbance or a false alarm. If the matching degree is lower than the preset threshold, it indicates that the actual operation of the physical system has significantly deviated from its normal theoretical mode, and it is determined that "a physical system anomaly exists."

[0060] Therefore, by utilizing reliable data and digital twin models, an objective and quantitative judgment standard for "whether the physical system of the power supply line is healthy" has been established, eliminating the subjectivity of relying on human experience and providing a clear direction for accurately locating the actual equipment or power supply line faults, thus ensuring the efficiency and accuracy of the entire traceability process.

[0061] The first construction module 107 is used to construct an anomaly propagation chain from the root cause device to the abnormal manifestation by setting an anomaly hypothesis in the digital twin power model and performing simulation if a physical system anomaly is determined to exist, so as to locate the physical system root cause that causes the physical system anomaly.

[0062] Interpretive, anomaly assumptions refer to a series of virtual scenarios about equipment failure or changes in the state of power lines that are pre-set in a digital twin power model to simulate physical system failures. Each assumption represents a potential cause of failure.

[0063] The root cause device refers to the single device or line in a physical system that initially fails and directly triggers a series of subsequent abnormal phenomena; it is the starting point of the entire abnormal propagation chain.

[0064] An anomaly propagation chain represents a causal path diagram in which the fault effect (such as voltage drop) is transmitted step by step from the root cause device through the power supply line to each monitoring point. It clearly expresses the logical relationship between "fault point - impact path - abnormal behavior".

[0065] The physical system root cause refers to the fundamental cause of the physical system anomaly that was finally determined through the above analysis. Its specific manifestation is the root cause device and its fault type.

[0066] Based on the above concept and setup, after determining that a physical system anomaly exists, the anomaly propagation chain can be constructed and the root cause of the physical system located through steps including but not limited to the following:

[0067] 1) Based on common failure modes, a set of abnormal assumptions are preset in the digital twin model, including but not limited to: A) Line damage assumption, setting virtual losses in a specific line and setting the loss current; B) Switching device jamming assumption, fixing the opening degree of a specific switching device in a non-command state (such as accidental closing or opening).

[0068] 2) In the digital twin model, the simulation of each abnormal assumption is executed in sequence to obtain the simulated power state of the entire network (such as voltage) under each assumption, and the matching degree of each simulation result is calculated with the actual, global physical entity layer monitoring data.

[0069] 3) Compare the matching degree calculation results of all abnormal hypotheses, and determine the abnormal hypothesis with the highest matching degree as the most likely physical system root cause. The faulty device corresponding to this hypothesis is then located as the root cause device.

[0070] 4) Starting from the identified root cause device, analyze and visualize the propagation path of the fault effect in the digital twin model: A) Identify all downstream power supply lines and equipment affected by it and whose state changes exceed the preset sensitivity threshold; B) Generate a clear causal path diagram, i.e., anomaly propagation chain, based on power correlation and the degree of state change, to intuitively show how the fault affects each monitored abnormal performance point from the root cause device.

[0071] This enables the analysis to move from "perceiving abnormal phenomena" to "locating the root cause of the fault," transforming the traditional, time-consuming, and labor-intensive fault diagnosis that relies on manual experience into an automated and precise analysis based on model simulation. By constructing an anomaly propagation chain, it provides dispatchers with comprehensive situational awareness and precise maintenance decision support, greatly improving operation and maintenance efficiency and power supply safety assurance capabilities.

[0072] The first alarm module 108 is used to generate and output alarm information based on the source tracing results of the power dispatching system anomalies or the root causes of the physical system anomalies.

[0073] Explained, the purpose of this step is to transform the diagnostic conclusions of the preceding steps into operational instructions that can be executed by maintenance personnel. This can be implemented in ways including but not limited to the following: Based on the final tracing results (the tracing results of power dispatching system anomalies or the root cause of physical system anomalies), the system calls the corresponding template from a pre-set structured alarm template library, injects specific diagnostic information, and generates alarm information. This alarm information can be a comprehensive set of content including but not limited to the following core elements: 1) Anomaly target: clearly indicating the anomaly object. For power dispatching system anomalies, this is the inaccurate sensor ID or the name of the performance degradation model; for physical system anomalies, this is the root cause device ID (e.g., "switchgear V101"); 2) Anomaly characterization: clearly describing the nature of the anomaly, such as "sensor data drift," "AI model prediction performance degradation," or "switchgear jamming"; 3) Tracing path / The propagation chain provides key diagnostic evidence. For power dispatch system anomalies, it briefly describes the verification logic (e.g., "the theoretical value deviates significantly after reverse verification by the digital twin model"). For physical system anomalies, it attaches the constructed anomaly propagation chain, explaining how the root cause leads to the observed anomaly. 4) Handling suggestions provide clear maintenance operations, such as "on-site calibration of sensor FT-001", "retraining the power consumption prediction model V2.1", "dispatch the work order to the on-site maintenance of switchgear V101, and perform isolation operations in accordance with the valve shut-off scheme SOP-05". 5) Scope of impact, combined with data from the business management layer, assesses and lists the affected areas or user ranges. In addition, the generated alarm information is visually pushed through the platform's human-machine interface (such as the dispatch dashboard, WEB client), and can also be sent to the terminals of relevant maintenance personnel through message services (such as SMS, mobile application push).

[0074] This completely changes the shortcomings of traditional power dispatch and monitoring systems that "only alarm but do not diagnose," providing alarms in the form of "diagnostic reports." It realizes the final transformation from complex intelligent diagnostic conclusions to concise and operable operation and maintenance instructions, enabling operation and maintenance personnel to accurately and quickly understand the essence of the problem and take precise measures, greatly shortening the fault response and recovery time, and improving the operational efficiency and reliability of the entire power supply system.

[0075] This invention, through the construction of a cloud-based power dispatching system integrating multi-dimensional data perception, intelligent anomaly identification, hierarchical diagnosis, and precise tracing, achieves closed-loop intelligent handling of power line anomalies. Its core improvements lie in two aspects: First, by integrating five layers of data—physical entities, control execution, model algorithms, business management, and system operation and maintenance—it provides a global and highly reliable decision-making basis for anomaly identification and diagnosis, overcoming the shortcomings of traditional systems with their single data dimension and fragile decision-making foundation. Second, by introducing a hierarchical diagnostic logic of "intelligence first, then physical," the system can first verify the reliability of data and models to rule out anomalies within the power dispatching system itself, ensuring the accuracy of subsequent decision inputs. Then, after confirming physical system anomalies, it utilizes a digital twin model... By conducting hypothetical simulations and constructing anomaly propagation chains, the system achieves precise location of the physical root cause device from the anomaly symptoms, solving the problems of traditional methods relying on human experience, slow location, and only addressing the symptoms without addressing the root cause. Finally, based on the accurate source tracing results, alarm information containing specific targets, paths, and handling suggestions is generated, upgrading the traditional simple status alarm into a directly executable diagnostic report. Thus, the traditional passive, open-loop, and manual power supply dispatch and monitoring mode in power management is transformed into a closed-loop intelligent system capable of self-diagnosis, proactive source tracing, and precise decision-making. This significantly improves the accuracy, efficiency, and reliability of power supply dispatch anomaly monitoring and handling in power management, ensuring the accuracy, efficiency, intelligence, and automation of power supply dispatch and control in power management.

[0076] In one embodiment, the first determination module 104 includes:

[0077] The first verification submodule is used to verify the reliability of the physical entity layer data based on the control execution layer data, the system operation and maintenance layer data, and / or the simulation results of the digital twin power model.

[0078] The first determination submodule is used to determine that the power dispatching system is abnormal if the verification result meets the preset untrustworthy condition.

[0079] The second determination submodule is used to determine that the power dispatching system does not have any abnormalities if the verification result does not meet the preset untrustworthy condition.

[0080] Explained, the pre-defined untrustworthy conditions refer to a set of quantifiable logical judgment criteria pre-set for determining the untrustworthiness of data at the physical entity layer. These criteria are composite judgments based on cross-validation of multi-source information. Their core lies in identifying whether the data seriously violates known physical laws, system states, or historical patterns, including but not limited to the following conditions: 1) Data paradox conditions, for example, a voltage sensor reading shows 180V, but according to simulation calculations performed by a digital twin model with surrounding trusted data as boundaries, the theoretical voltage at this point should be 220V, and the deviation continues to exceed the allowable range (e.g., >10V); 2) Control disconnection conditions, for example, control execution layer data shows that a "start" command has been issued to a generator, but its status feedback remains "stopped" throughout the timeout period, and system operation and maintenance layer data shows that the communication link is normal.

[0081] Based on the above concept and setup, the credibility of physical entity layer data can be determined through collaborative verification processes including but not limited to the following:

[0082] 1) Control logic verification: Compare the control commands in the control execution layer data with the equipment status feedback. If the commands and feedback are inconsistent for a preset time and the delay of command switching is excluded, the credibility of the relevant status data of the equipment (such as generator output voltage) is questionable.

[0083] 2) System health verification: Check the system operation and maintenance layer data. If the communication link corresponding to the target data source has high latency, high frequency packet loss or interruption, the data originating from that link is deemed unreliable.

[0084] 3) Physical law verification: using digital twin power models for forward or reverse simulation, including: A) Forward verification: using a few key reliable data (such as power supply outlet voltage) as boundaries, driving model simulation, and performing matching degree analysis between the results and a large number of sensor data to be verified; B) Reverse verification: using the data around the predicted abnormal sensor as boundaries, simulating and calculating the theoretical value of the sensor, comparing it with the measured value, and then, when there is a significant, continuous deviation between the simulation results and the measured data that cannot be explained by power anomalies, the sensor data is determined to be unreliable.

[0085] This invention, through the construction of a self-diagnostic mechanism for a power dispatching system based on multi-source information cross-verification, achieves a proactive guarantee of the reliability of decision-making data. Its core improvement lies in extending the monitoring of power supply dispatch anomalies in power management from a single physical object to an "information-physical" fusion system. Specifically, firstly, by introducing data from the control execution layer and system operation and maintenance layer for logical and health verification, it can quickly identify "command-feedback" loop breaks caused by actuator failures or communication interruptions. This solves the problem of traditional systems misjudging such control link anomalies as physical faults, achieving accurate diagnosis of the failure of the "nerve endings" of the power dispatching system. Secondly, by utilizing digital twins... The power model, a physical law engine, performs simulation-based reverse verification, providing objective and quantitative scientific evidence for judging the authenticity of sensor data. This overcomes the limitations of traditional methods that rely on human experience or simple threshold comparisons, and can effectively identify deeper-level inherent defects in data sources such as sensor drift and distortion. In summary, through the above-mentioned multi-level collaborative verification process, a monitoring mechanism for the health status of the power dispatching system itself is formed, ensuring that any subsequent analysis and decision-making based on physical entity layer data is based on reliable data. This fundamentally avoids the situation of "erroneous data input" leading to "erroneous decision output," and significantly improves the accuracy and effectiveness of anomaly diagnosis in the entire power dispatching system.

[0086] In one embodiment, the second determination module 106 includes:

[0087] The first analysis submodule is used to drive the digital twin power model to perform real-time simulation based on the reliable physical entity layer data, and to perform a matching degree analysis between the simulation results and the actual physical entity layer data.

[0088] The third determination submodule is used to determine that the physical system is abnormal if the matching degree is lower than a preset threshold.

[0089] Interpretive, reliable physical entity layer data refers to a subset of data that has been verified through the aforementioned "power dispatch system anomaly judgment" and determined to truly reflect the state of physical objects.

[0090] Based on the above concept and description, assuming the physical entity layer data is accurate, the existence of a real anomaly in the physical system can be determined by the following steps, including but not limited to:

[0091] 1) Using the verified data of the trusted physical entities (such as substation voltage and key node voltage) from the previous steps as forced boundary conditions, the digital twin power model is driven to perform real-time power grid state simulation. Based on the power supply line topology and power principles, the theoretical voltage of all nodes and all lines in the target power supply line should be calculated under the current boundary conditions.

[0092] 2) Compare the theoretical power state of the entire network (theoretical value) obtained from the simulation of the digital twin power model with the physical entity layer data of the entire network (measured value) actually collected by sensors, and calculate the comprehensive matching degree index (e.g., the root mean square error RMSE or coefficient of determination R² of voltage and current at all comparable measurement points). The comprehensive matching degree index quantifies the overall deviation between the operating state of the real physical system and the theoretical expectation of the ideal lossless model.

[0093] 3) Compare the comprehensive matching index with the corresponding preset threshold. If the comprehensive matching index is lower than the corresponding preset threshold, it indicates that the actual behavior of the physical system has deviated significantly from the theoretical behavior that it should have under normal conditions. Based on this, it is determined that "there is a physical system anomaly". The preset threshold can be obtained based on the statistical analysis of historical normal operation data and is used to distinguish between normal system noise and real abnormal deviation.

[0094] This invention utilizes a digital twin model as a standard baseline to achieve objective and quantitative judgment of physical system anomalies. Its core improvement lies in elevating anomaly judgment from a "superficial perception" relying on isolated data points and fixed thresholds to a global state comparison. Specifically, based on trusted data-driven simulation, it achieves dynamic and personalized expectations of how power lines should operate under given boundary conditions. Through global matching analysis, it obtains the systematic deviation between the actual behavior of the physical system and this expectation. This not only effectively identifies "soft faults" (such as slowly developing line aging) that are difficult to detect using traditional methods and do not have obvious threshold violations but exhibit abnormal overall behavior, but also strictly distinguishes between "power dispatch system distortion" and "real physical system faults," providing accurate input and prerequisites for subsequent physical root cause tracing. Therefore, it significantly improves the accuracy, depth, and reliability of physical system anomaly identification.

[0095] In one embodiment, the first analysis submodule includes:

[0096] The first verification submodule is used to perform credibility verification on the acquired physical entity layer data, and the first verification submodule includes at least one of the following submodules:

[0097] The second verification submodule is used to verify the reasonableness of the range and rate of change of the data values ​​of the physical entity layer data;

[0098] The third verification submodule is used to verify the consistency between the physical entity layer data and other sensor data that have a preset power correlation relationship.

[0099] The fourth verification submodule is used to verify the matching degree between the physical entity layer data and the calculation results of the digital twin power model at the corresponding positions.

[0100] Furthermore, the first analysis submodule further includes: a first identification submodule, used to identify the data that passes the trustworthiness verification as trustworthy physical entity layer data.

[0101] Interpretive rationality verification refers to a rule that rapidly screens the inherent rationality of individual sensor data based on physical laws and device performance, with the data source being the sensor's own readings and preset limit parameters.

[0102] Range checking determines whether data exceeds its possible physical range. For example, the voltage of a power supply line should not be negative and should not be higher than the voltage corresponding to the generator's maximum power.

[0103] Rate of change verification indicates whether the change in data within a unit of time exceeds a reasonable limit. For example, a voltage drop of 1 MPa within 1 second is impossible in a conventional power supply line and may indicate sensor malfunction or communication interference.

[0104] Preset power correlations represent the spatial and logical relationships between power transmission equipment that are predefined based on the power supply line topology and electrical principles. For example, the voltage of adjacent measuring points should maintain power smoothness, i.e., a stable voltage gradient, when there is no significant power consumption or leakage.

[0105] Based on the above concept and setup, a reliable subset of data can be selected from the original physical entity layer data through a three-level progressive verification process, including but not limited to: Level 1: For each original data point, predefined rules are applied for verification: 1) Range check: Confirm that the data value is between [lower limit of range, upper limit of range]; 2) Rate of change check: Calculate the absolute value of the difference between the current value and the previous value, confirming that it is less than the maximum allowable rate of change * sampling interval. Data that fails this level of verification is directly marked as "unreliable". Level 2: For data that passes the rationality verification, based on its geographical location and preset power correlation, its associated sensors are located. The data from the associated sensor group is checked to see if it conforms to the expected power patterns. For example, checking whether the readings of the ammeters and voltmeters upstream and downstream of the power supply line are equal within the error range; checking whether the voltage difference between adjacent voltage points matches the elevation difference and line friction loss. If the collective behavior of a data point significantly contradicts that of the associated points, it is marked as "abnormal". The third level involves further verification of "abnormal" or critical node data. Using all other reliable data points surrounding the data point as boundary conditions, the digital twin power model is driven to simulate and calculate the theoretical value of the point's location. The degree of matching between the theoretical and measured values ​​(e.g., relative error) is then calculated. If the degree of matching is lower than a preset standard, the data is deemed unreliable; if the degree of matching is high, it is re-identified as reliable data. Thus, data that passes these three levels of verification and is ultimately identified as "reliable physical entity layer data" is used for subsequent physical system anomaly detection.

[0106] This invention, through a three-tiered data cleaning and verification process—from shallow to deep, from single points to systems, and from rules to models—lays a solid data foundation for reliable intelligent decision-making. Its core improvements lie in three aspects: First, rationality verification quickly eliminates obvious outliers caused by equipment failures or transient interference, improving data processing efficiency. Second, consistency verification utilizes the inherent power correlation characteristics of power lines to achieve cross-verification of sensor networks, effectively identifying systematic deviations or local failures of individual sensors. Third, matching degree verification leverages digital twins and physical laws to provide a relatively accurate basis for judgment. These three aspects together ensure that the data ultimately flowing into the core analysis engine has a high degree of credibility. Based on this, the correlation between "input distortion leading to decision-making errors" is fundamentally eliminated, significantly improving the robustness of system perception in power management and the accuracy and effectiveness of subsequent anomaly analysis results.

[0107] In one embodiment, please refer to Figure 2 , Figure 2 This is the first sub-schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention. (See diagram below.) Figure 2 As shown, in this embodiment, the first construction module 107 includes:

[0108] The first setting submodule 201 is used to set one or more abnormal assumptions about equipment failure or power line topology changes in the digital twin power model;

[0109] The first acquisition submodule 202 is used to drive the digital twin power model to perform simulation and obtain the power grid state simulation results under each of the above-mentioned abnormal assumptions.

[0110] The first calculation submodule 203 is used to calculate the matching degree between the power grid power state simulation results and the corresponding actual physical entity layer data;

[0111] The first determining submodule 204 is used to determine the anomaly hypothesis with the highest matching degree as the most likely root cause device causing the anomaly of the physical system.

[0112] Explained, the power grid state simulation results represent the global and dynamic power state distribution of the target power supply line under corresponding operating conditions, calculated by solving the power control equations in a digital twin power model based on appropriate boundary conditions and anomaly assumptions. This result is a prediction of the physical equipment's operating state by the digital twin power model, and its data source is the simulation calculation of the digital twin power model, including but not limited to the calculated voltage of each node and the calculated voltage and current of each line in the digital twin power model. For example, when an anomaly assumption of "leakage in a certain line" is set in the digital twin power model, the prEssurE distribution and Flow distribution of the entire power supply line obtained by the digital twin power model simulation are the power grid state simulation results under this leakage assumption.

[0113] Based on the above concepts and descriptions, the root cause of a physical system can be located through a closed-loop process of assumption-simulation-matching, but not limited to:

[0114] 1) Based on common failure modes, one or more abnormal assumptions are preset in the digital twin model, including but not limited to: A) Equipment failure assumptions, such as fixing the opening degree of a specific switch device in a non-command state (such as being stuck in the closed position) or setting the state of a specific generator to failure; B) Power supply line topology change assumptions, such as setting a leakage point in a specific line (equivalent to adding an outgoing line).

[0115] 2) Keeping the credible boundary conditions unchanged, each abnormal hypothesis is injected into the digital twin model in turn, driving the digital twin model to perform parallel power simulation of the entire power supply line. Each simulation outputs the corresponding power grid state simulation result, which characterizes "what state the power supply line should be in if the hypothesis is true".

[0116] 3) Compare the simulation results of the power grid power state (theoretical state) obtained from each simulation with the actual, global physical entity layer data (measured state), calculate the comprehensive matching degree index (such as the coefficient of determination R²), and determine the abnormal hypothesis with the highest matching degree as the most likely root cause of the current physical system abnormality. The equipment or line corresponding to this hypothesis is then located as the "most likely root cause equipment".

[0117] This invention upgrades the digital twin model from a "state reproduction" tool to a "root cause detection" engine, achieving intelligent and precise fault location. The core improvement lies in replacing traditional manual, experience-based, step-by-step troubleshooting with systematic simulation and comparison. Specifically, by pre-setting anomaly hypothesis sets covering major fault modes and using the digital twin model to simulate the global system response under various hypotheses in parallel, multiple possible fault scenarios can be efficiently traversed. Then, by calculating the matching degree between the simulation results of each hypothetical scenario and the corresponding measured data, the fault cause that best matches the actual anomaly can be objectively identified. This achieves a rapid mapping from "abnormal phenomenon" to "fault root cause," not only freeing maintenance personnel from arduous on-site troubleshooting and greatly improving fault location efficiency, but also enabling the discovery of hidden, complexly interconnected systemic faults. This provides an efficient and accurate basis for precise and efficient maintenance decisions, improving the safety and operational efficiency of the power supply system.

[0118] In one embodiment, the first construction module 107 includes:

[0119] The first identification submodule is used to identify, starting from the most likely root cause device, all downstream power supply lines and devices in the digital twin power model that are affected by it and whose state changes exceed a preset sensitivity threshold.

[0120] The first generation submodule is used to generate a causal path diagram from the root cause device to all abnormal performance monitoring points based on the degree of state change and power correlation of the downstream power supply lines and equipment, as the abnormal propagation chain.

[0121] Explained, a preset sensitivity threshold is a quantitative threshold set in advance to determine whether a change in the state of a certain point in a power supply line has practical significance. It is derived from statistical analysis of historical normal operation data (such as multiples of standard deviation) and is used to filter out small, random fluctuations with no diagnostic value, focusing on significant changes triggered by root causes. For example, a voltage change exceeding N times the standard deviation of historical normal fluctuation range can be set as a sensitivity threshold.

[0122] The degree of state change indicates the deviation of the power state (voltage, current) of downstream power supply lines or equipment from normal operating conditions; it is a quantitative indicator. Power correlation refers to the causal relationship between power supply line components based on topological connections and current direction, such as the state of downstream equipment being directly affected by the state of upstream equipment.

[0123] A cause-and-effect path diagram is a graphical representation that clearly shows the complete transmission path and causal relationship of a fault, starting from the root cause device, affecting downstream power lines and equipment in sequence according to power correlation, and finally reaching all abnormal behavior monitoring points. It is a visual representation of the abnormal propagation chain.

[0124] Based on the above concept and description, after locating the most likely root cause device, an anomaly propagation chain can be constructed through, but is not limited to, the following steps:

[0125] 1) Starting with the identified root cause device, in the digital twin power model, based on its simulation results, automatically search and identify all downstream power supply lines and equipment affected by the fault of the device and whose voltage or current changes exceed the preset sensitivity threshold, forming a "set of affected devices".

[0126] 2) Based on the GIS topology and current direction of the power supply line, analyze the power correlation between each device in the affected equipment set, and sort and connect them according to the degree of state change (such as the severity of voltage drop) to generate one or more clear causal path diagrams. The causal path diagrams describe how the fault gradually affects the root cause device and is eventually detected by various sensors.

[0127] For example, assuming the system has identified the most likely root cause device as "leakage in line P205", an anomaly propagation chain can be constructed through, but is not limited to, the following steps:

[0128] 1) In the simulation results of leakage of P205, the system identified that the voltage drop of its downstream power supply lines P206 and P207 exceeded the preset sensitivity threshold, and the voltage drop of nodes J-21 and J-22 also exceeded the threshold. These lines and nodes were included in the "affected equipment set".

[0129] 2) The system generates a cause-effect path diagram based on the topology: "Line P205 leakage → Line P206 current / voltage abnormality → Node J-21 voltage abnormality → Line P207 current / voltage abnormality → Node J-22 voltage abnormality". This cause-effect path diagram clearly shows that all monitored abnormal voltage points (J-21, J-22) can be traced back to the root cause device P205 through this power path.

[0130] This invention, by making implicit power causal relationships explicit into a visualized anomaly propagation chain, achieves precise characterization of the fault's impact range and improves the interpretability of the diagnostic process. Its core improvement lies in providing a complete fault diagnosis chain. Specifically, by setting preset sensitivity thresholds, the system can intelligently focus on significant changes with practical diagnostic significance, avoiding information overload. By comprehensively analyzing the degree of state change and power correlation, it can clearly reveal the fault propagation mechanism and impact boundaries. This not only greatly enhances dispatchers' global understanding of the fault situation and provides direct and reliable decision support for formulating optimal isolation and dispatching schemes (such as which switching devices to shut down with minimal impact), but also makes the logic of the entire AI diagnostic process logically transparent, enhancing the credibility and acceptability of the results, and improving the accuracy and efficiency of power supply dispatching and maintenance in power management.

[0131] In one embodiment, the target data source includes target sensor data; the first tracing module 105 includes:

[0132] The first diagnostic submodule is used to perform fault diagnosis on the target sensor data in the physical entity layer to locate the misaligned sensor, and the first diagnostic submodule includes at least one of the following submodules:

[0133] The fourth determination submodule is used to acquire data from one or more associated sensors that have a preset power correlation with the target sensor data. If the data of the associated sensors conforms to the preset power law, but the target sensor data does not, the target sensor corresponding to the target sensor data is determined to be inaccurate.

[0134] The fifth determination submodule is used to drive the digital twin power model based on other reliable data besides the target sensor data, to calculate the theoretical value of the target sensor monitoring position corresponding to the target sensor data, and if the deviation between the theoretical value and the measured value of the target sensor exceeds the first preset tolerance, the target sensor is determined to be inaccurate.

[0135] The sixth determination submodule is used to determine whether the target sensor data exceeds its measurement range and / or whether the data change rate exceeds the second preset tolerance. If it does, the target sensor corresponding to the target sensor data is determined to be inaccurate.

[0136] The "target sensor data" is determined through the following sub-modules:

[0137] The first screening submodule is used to, in response to the detection of an anomaly in the target power supply line, select one or more suspicious sensor data as the target sensor data from all physical entity layer data based on the power performance and spatial distribution corresponding to the anomaly.

[0138] Explained, the predefined electrical laws represent the predefined physical relationships that sensor data should satisfy under normal conditions, based on the fundamental principles of circuit theory and the topology of the power supply network. These relationships originate from basic physical laws such as Kirchhoff's laws, Ohm's law, and the law of conservation of power. Examples of predefined electrical laws are as follows: 1) Current Conservation Law (based on Kirchhoff's Current Law): At any node in the power grid (such as a substation busbar or distribution box connection point), the sum of the currents flowing into that node should equal the sum of the currents flowing out of that node (considering measurement errors). For example, for the outgoing lines of a distribution transformer, the vector sum of the currents on the low-voltage side should equal the incoming current on the high-voltage side. 2) Voltage Drop Law (based on Kirchhoff's Voltage Law and Ohm's Law): Along the direction of the power supply line, the voltage difference (voltage drop) between adjacent monitoring points should correspond to the impedance and current flowing through that section of the line. Its voltage drop trend should be stable, and the amplitude should conform to the calculation range of ΔU = I × (R cosφ + X sinφ) (where R is resistance, X is reactance, and cosφ is power factor). 3) Power supply characteristic law: The output voltage, current, frequency, active / reactive power and other parameters of the power generation unit (such as generator, inverter) should change in conjunction with the reasonable range defined by its PQ power curve, voltage-reactive power regulation characteristic curve and other equipment performance curves, and match the grid dispatch instructions.

[0139] Based on the above concept and description, the source of the misaligned sensor is traced through, but not limited to, the following steps:

[0140] 1) In response to system anomalies, the system first intelligently filters out one or more suspicious sensor data points most relevant to the anomaly pattern from all physical entity layer data based on the abnormal power characteristics (such as local low voltage, current or voltage fluctuations) and their spatial distribution, and identifies them as "target sensor data". For example, if the anomaly is manifested as an anomaly on a certain line, the current or voltage data of the ammeter or voltmeter upstream and downstream of that line are identified as targets.

[0141] 2) Diagnose the above target sensor data using at least one of the following methods to locate the inaccurate sensor: A) Correlation analysis method: Obtain data from associated sensors that have a preset electrical correlation with the target sensor, and check whether the associated sensor data conforms to a preset electrical law (such as mass conservation). If it does, but the target sensor data does not, then the target sensor is determined to be inaccurate; B) Digital twin verification method: Based on other reliable data besides the target data, drive the digital twin electrical model to calculate the theoretical value of the target sensor's monitoring position. If the deviation between the theoretical value and the measured value of the target sensor exceeds a first preset tolerance, then the target sensor is determined to be inaccurate; C) Inherent characteristic verification method: Directly determine whether the target sensor data exceeds its measurement range and / or whether its data change rate exceeds a second preset tolerance. If it does, then the target sensor is directly determined to be inaccurate.

[0142] This invention, through constructing a complete chain from focusing on "abnormal phenomena" to "suspicious targets," and then using multi-level diagnostic methods for precise verification, achieves rapid and reliable localization of inaccurate sensors. Its core improvement lies in elevating traditional, decentralized, and passive sensor verification to proactive, systematic, and closely linked intelligent diagnosis that is closely integrated with business anomalies. Specifically, by dynamically determining target sensor data based on anomaly characteristics, the diagnostic process becomes highly targeted and efficient. By introducing two advanced diagnostic methods based on physical laws and system simulation—correlation analysis and digital twin verification—it can discover hidden "soft" inaccuracies (such as drift) that cannot be identified by inherent characteristic verification alone, greatly improving the depth and accuracy of the diagnosis. Based on this, it ensures that the "sensory nerves" of the intelligent power dispatching system remain healthy and reliable, providing high-quality data input for all upper-level intelligent applications and serving as the cornerstone for maintaining the accuracy of the entire system's intelligent decision-making.

[0143] In one embodiment, the first filtering submodule includes at least one of the following submodules:

[0144] The second determining submodule is used to determine the sensor data of the monitoring point as the target sensor data if the anomaly is manifested as a serious deviation of the data of a single monitoring point from historical patterns or a preset threshold, while the data of other related points are normal.

[0145] The second filtering submodule is used to filter out the sensor data located at the upstream key node of the power supply line in the region and whose data anomaly pattern is most related to the regional anomaly pattern, based on the topology of the digital twin power model, if the anomaly is manifested as anomalies in the data of multiple monitoring points in a region, as the target sensor data.

[0146] Interpretive, power performance and spatial distribution represent the physical characteristics of anomalies and their geographical location. Power performance includes specific types such as voltage anomalies and current anomalies, as well as their severity and trends. Spatial distribution represents the set of locations of anomaly monitoring points on the GIS map of the power supply line, such as within a DMA, distributed along a power supply line, or concentrated in a certain geometric area.

[0147] Preset thresholds are quantitative standards set based on historical operating data or safety specifications to determine whether data deviates significantly from normal levels. For example, a voltage 0.2 MPa lower than the service voltage is a low-voltage alarm threshold.

[0148] The topology of a digital twin power model represents the set of all lines, nodes, devices and their connections in the model. It accurately reflects the spatial layout and power connectivity of the physical power supply lines, which is the basis for analyzing fault propagation paths and identifying critical nodes.

[0149] Data anomaly patterns indicate specific characteristics of individual sensor data that deviate from normal, such as consistently high values, abrupt changes, or zero values.

[0150] Regional anomaly patterns indicate common characteristics exhibited by data from all monitoring points within the entire anomaly region, such as "voltage decreasing gradually from upstream to downstream".

[0151] Based on the above concept and description, suspicious sensors can be intelligently screened from massive amounts of data using strategies including but not limited to the following: 1) When the power performance shows a severe anomaly in the data of a single monitoring point (such as a sudden drop in voltage to 0), and the data of other adjacent or power-related points in its spatial distribution are normal, the system will identify the sensor data corresponding to this isolated anomaly point as the target sensor data. 2) When the power performance shows a large-scale anomaly in a region (such as low voltage in an entire voltage zone), the system analyzes the power supply path of the region based on the topology of the digital twin power model, locates key nodes upstream of the power supply (such as regional power inlet switchgear, main power supply line connection points), and then filters the data anomaly patterns of sensors on these key nodes (such as abnormal opening feedback of a certain switchgear), and determines whether they are most relevant to the regional anomaly pattern (such as low voltage) in terms of causality and time series, and identifies the key node sensor data most relevant to the root cause of the regional anomaly as the target sensor data.

[0152] This invention, through the introduction of an intelligent screening strategy based on anomaly features, achieves accurate and efficient localization of suspicious sensors. Its core improvement lies in upgrading the traditional "one-size-fits-all" or "manual, step-by-step" sensor inspection mode to a problem-oriented, logic-driven, automated focusing process. Specifically, the isolated point investigation strategy can quickly identify sensors generating significant noise due to their own faults, effectively filtering out local interference; while the regional root cause focusing strategy, by combining power line topology and anomaly pattern analysis, can intelligently infer the most likely source sensor from complex regional anomalies, greatly improving the targeting and efficiency of diagnosis. These two strategies work together to ensure that subsequent in-depth fault diagnosis always revolves around the "most suspicious object," avoiding waste of computing resources and significantly shortening the overall time from "anomaly detection" to "problem sensor localization," providing a crucial guarantee for quickly restoring reliable system perception.

[0153] In one embodiment, please refer to Figure 3 , Figure 3 This is a second sub-schematic block diagram of a cloud-based power dispatching system provided in an embodiment of the present invention. (See diagram below.) Figure 3 As shown, in this embodiment, the target algorithm model includes a target prediction model; the first tracing module 105 includes:

[0154] The third filtering submodule 301 is used to, in response to the identification of an anomaly in the target power supply line, select one or more as the target prediction model from multiple algorithm models based on the correlation between the characteristics of the anomaly and the prediction functions of various models in the model algorithm layer.

[0155] The second calculation submodule 302 is used to continuously calculate the error index between the predicted value and the actual value of the target prediction model within a preset time window.

[0156] The first judgment submodule 303 is used to determine whether the error index continuously exceeds a preset threshold.

[0157] The seventh determination submodule 304 is used to determine, if yes, that the performance of the target prediction model has degraded.

[0158] Explained, performance degradation refers to a sustained, non-accidental decline in the accuracy and reliability of an algorithm model's predictions or decisions due to the failure of internal parameters or changes in the external environment, rendering it unable to meet preset performance requirements. This is typically caused by data drift (the statistical characteristics of the model's input data changing over time) or conceptual drift (changes in the implicit relationship between input variables and the prediction target). For example, a power consumption prediction model may experience a systematic deviation between its historical data and current realities due to changes in urban population structure or the construction of new industrial parks, leading to a continuous increase in prediction errors—this is performance degradation.

[0159] Based on the above concept and description, the source of model performance degradation can be traced through a closed-loop process including but not limited to the following: 1) In response to anomalies identified by the system, analyze the characteristics of the anomaly (such as "systematic deviation between scheduling instructions and actual conditions") and perform correlation analysis with the prediction functions of various models in the model algorithm layer. For example, if the anomaly manifests as persistently inaccurate electricity consumption prediction, lock the electricity consumption prediction model as the target prediction model; if it manifests as poor performance of optimized scheduling schemes, lock the load prediction model or cost model on which it relies as the target. 2) For the locked target prediction model, continuously calculate the error indicators (such as mean absolute percentage error MAPE, root mean square error RMSE) between its predicted and actual values ​​within a preset time window (such as 24 hours) to quantify performance evaluation. 3) Trend-based degradation judgment is not based on a single error, but rather on monitoring whether the error indicator exceeds a preset threshold for multiple consecutive calculation cycles (such as 6 consecutive scheduling cycles). If so, the target prediction model is judged to have degraded performance. This trend-based judgment method effectively avoids misjudgments caused by transient data noise.

[0160] This invention, through the establishment of a data-driven model performance monitoring and degradation judgment mechanism linked to business anomalies, enables the monitoring of software functions in the power dispatching system. Its core improvement lies in transforming model maintenance from passive, periodic monitoring to proactive monitoring linked to system anomalies in real time. Specifically, by intelligently identifying target models based on anomaly characteristics, model performance monitoring becomes highly targeted, avoiding wasted computing power. By continuously calculating error indicators and performing trend judgments, it can efficiently capture the gradual performance degradation of the model. This allows the operations and maintenance team to receive early warnings and promptly retrain or update the model when systemic deviations occur in power dispatching decisions due to model inaccuracies, thus preventing problems before they arise. Based on this, the long-term reliability and adaptability of the core decision engine in smart power management are ensured, which is a crucial guarantee for maintaining the overall intelligent level of the system without degradation.

[0161] In one embodiment, the third screening submodule includes at least one of the following submodules:

[0162] The third determining submodule is used if the anomaly manifests as a persistent systematic deviation between the instructions of the scheduling system and the actual situation, and

[0163] The deviation pattern is directly related to the output of the electricity consumption prediction model, and the electricity consumption prediction model is determined as the target prediction model.

[0164] The fourth determination submodule is used to determine the core model on which the optimized scheduling engine depends as the target prediction model if the anomaly is that the key performance indicators of the optimized scheduling scheme fail to reach the expected target after actual execution.

[0165] The fifth determining submodule is used to determine the digital twin power model itself as the target prediction model if the anomaly is manifested as a persistent large error between the simulation results of the digital twin power model and a large amount of reliable actual measurement data.

[0166] Interpretationally, a deviation pattern represents a regular discrepancy between dispatch system instructions and actual conditions. For example, "the instruction requires maintaining the power supply line voltage at 220V, but the actual voltage fluctuates continuously within the range of 200V," this persistent unidirectional negative deviation constitutes a clear deviation pattern.

[0167] Electricity consumption prediction model refers to an algorithm model used to predict the total electricity consumption of a city or region in the short term (such as the next few hours to 24 hours). Its prediction results are the core basis for formulating dispatch instructions such as the start-up and shutdown of substations or transformers and the production plan of power plants. The improvement of this invention does not lie in the construction and training of the electricity consumption prediction model. Its construction and training can draw on existing related technical means, which will not be elaborated here.

[0168] The optimization scheduling engine is a software system that, based on objective functions (such as minimum energy consumption and minimum cost) and constraints (such as upper and lower voltage limits), solves for optimal scheduling schemes such as transformer group transformer combinations and switching equipment operation. Its core models are the input models on which the optimization scheduling engine relies for calculations. These mainly include power consumption prediction models and power supply line power models. The engine uses the output data of these models (predicted power consumption and power supply line status) to simulate the effects of different schemes and select the best one. Furthermore, the accuracy of the core models directly determines the quality of the optimization scheduling scheme.

[0169] Based on the above concept and setup, this embodiment of the invention mainly establishes a mapping relationship between "abnormal features and model functions" to intelligently select the target prediction model from multiple algorithm models, employing methods including but not limited to the following: 1) Instruction result reverse deduction method: When the anomaly manifests as a persistent systematic deviation between the scheduling instruction and the actual situation, the deviation pattern is analyzed. If the deviation originates from a misjudgment of the supply and demand relationship (e.g., the generator's power generation does not match the actual electricity demand, and the generator's power generation can be adjusted), and this misjudgment is directly related to the output logic of the electricity consumption prediction model, then the electricity consumption prediction model is determined as the target prediction model. 2) Scheme effect tracing method: When the anomaly manifests as the key performance indicators (e.g., unit power consumption) failing to reach the expected target after the optimized scheduling scheme is executed, the generation process of the optimized scheme is analyzed in reverse. The core models (e.g., electricity consumption prediction model, power model) on which the optimized scheduling engine relies when formulating the scheme are determined as the target prediction model for review. 3) Simulation accuracy evaluation method: When the anomaly is that the simulation results of the digital twin power model have a large and continuous error with a large amount of reliable measured data, it indicates that the model itself can no longer accurately describe the physical system, and thus it is determined to be the target prediction model.

[0170] This invention, through the construction of an intelligent diagnostic link that accurately maps macroscopic system anomalies to microscopic problem models, enables rapid localization of model performance issues. Its core improvement lies in placing the model within the entire business loop of the dispatching system for effectiveness evaluation. Specifically, the instruction result reverse reasoning method and the scheme effect tracing method closely link model performance with the final dispatching results, enabling causal reasoning that traces the "problematic model" from the perspective of "poor results." Meanwhile, the simulation accuracy evaluation method ensures the health of the digital twin model itself, which forms the basis for numerous analyses. This problem-oriented screening mechanism transforms model maintenance from a blind, periodic task into targeted analysis and judgment, significantly improving the efficiency of model operation and maintenance and the reliability of the entire power dispatching system's decision-making chain.

[0171] It should be noted that the cloud-based power dispatching system described in the above embodiments can be recombined with the technical features included in different embodiments as needed to obtain a combined implementation scheme, but all of them are within the protection scope claimed by this invention.

[0172] The modules in the aforementioned cloud-based power dispatching system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can invoke and execute the corresponding operations of each module.

[0173] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] The software tools, components, or models not belonging to our company that appear in the embodiments of this invention are merely illustrative examples and do not represent actual use.

[0175] The data collection in this embodiment of the invention complies with the requirements of relevant laws and regulations, such as China's Personal Information Protection Law, GDPR (General Data Protection Regulation of the European Union), or information security standards of other countries and regions.

[0176] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cloud-based power dispatching system, characterized in that, include: The first determination module is used to determine the target power supply line based on the power dispatch cloud platform and in response to power dispatch monitoring instructions; The first acquisition module is used to acquire multi-dimensional data of the target power supply line, including physical entity layer data, control execution layer data, model algorithm layer data, business management layer data, and system operation and maintenance layer data. The first identification module is used to identify whether there is an anomaly in the target power supply line based on the multi-dimensional data and through a preset rule engine and / or anomaly detection model; The first judgment module is used to determine whether there is an anomaly in the power dispatch cloud platform if an anomaly is detected. The first tracing module is used to trace the source to the inaccurate target data source through fault diagnosis, or to trace the source to the target algorithm model with degraded performance through performance monitoring, if an anomaly in the power dispatching system is determined. The second judgment module is used to determine whether there is a physical system abnormality in the target power supply line if it is determined that there is no abnormality in the power dispatching system. The first construction module is used to construct an anomaly propagation chain from the root cause device to the anomaly manifestation by setting an anomaly hypothesis in the digital twin power model and performing simulation if a physical system anomaly is determined to exist, so as to locate the physical system root cause that leads to the physical system anomaly. The first alarm module is used to generate and output alarm information based on the source tracing results of the power dispatching system anomalies or the root causes of the physical system anomalies.

2. The cloud-based power dispatching system as described in claim 1, characterized in that, The first judgment module includes: The first verification submodule is used to verify the reliability of the physical entity layer data based on the control execution layer data, the system operation and maintenance layer data, and / or the simulation results of the digital twin power model. The first determination submodule is used to determine that the power dispatching system is abnormal if the verification result meets the preset untrustworthy condition. The second determination submodule is used to determine that the power dispatching system does not have any abnormalities if the verification result does not meet the preset untrustworthy condition.

3. The cloud-based power dispatching system as described in claim 1, characterized in that, The second judgment module includes: The first analysis submodule is used to drive the digital twin power model to perform real-time simulation based on the reliable physical entity layer data, and to perform a matching degree analysis between the simulation results and the actual physical entity layer data. The third determination submodule is used to determine that the physical system is abnormal if the matching degree is lower than a preset threshold.

4. The cloud-based power dispatching system as described in claim 3, characterized in that, The first analysis submodule includes: The first verification submodule is used to perform credibility verification on the acquired physical entity layer data, and the first verification submodule includes at least one of the following submodules: The second verification submodule is used to verify the reasonableness of the range and rate of change of the data values ​​of the physical entity layer data; The third verification submodule is used to verify the consistency between the physical entity layer data and other sensor data that have a preset power correlation relationship. The fourth verification submodule is used to verify the matching degree between the physical entity layer data and the calculation results of the digital twin power model at the corresponding positions. Furthermore, the first analysis submodule further includes a first identification submodule, used to identify the data that passes the trustworthiness verification as trustworthy physical entity layer data.

5. The cloud-based power dispatching system as described in claim 1, characterized in that, The first building module includes: The first setting submodule is used to set one or more abnormal assumptions about equipment failures or changes in power supply line topology in the digital twin power model; The first acquisition submodule is used to drive the digital twin power model to perform simulation and obtain the power grid state simulation results under each of the above-mentioned abnormal assumptions. The first calculation submodule is used to calculate the matching degree between the power grid power state simulation results and the corresponding actual physical entity layer data; The first determination submodule is used to determine the anomaly hypothesis with the highest matching degree as the most likely root cause device causing the anomaly of the physical system.

6. The cloud-based power dispatching system as described in claim 5, characterized in that, The first building module includes: The first identification submodule is used to identify, starting from the most likely root cause device, all downstream power supply lines and devices in the digital twin power model that are affected by it and whose state changes exceed a preset sensitivity threshold. The first generation submodule is used to generate a causal path diagram from the root cause device to all abnormal performance monitoring points based on the degree of state change and power correlation of the downstream power supply lines and equipment, as the abnormal propagation chain.

7. The cloud-based power dispatching system as described in claim 1, characterized in that, The target data source includes target sensor data; The first tracing module includes: The first diagnostic submodule is used to perform fault diagnosis on the target sensor data in the physical entity layer to locate the misaligned sensor, and the first diagnostic submodule includes at least one of the following submodules: The fourth determination submodule is used to acquire data from one or more associated sensors that have a preset power correlation with the target sensor data. If the data of the associated sensors conforms to the preset power law, but the target sensor data does not, the target sensor corresponding to the target sensor data is determined to be inaccurate. The fifth determination submodule is used to drive the digital twin power model based on other reliable data besides the target sensor data, to calculate the theoretical value of the target sensor monitoring position corresponding to the target sensor data, and if the deviation between the theoretical value and the measured value of the target sensor exceeds the first preset tolerance, the target sensor is determined to be inaccurate. The sixth determination submodule is used to determine whether the target sensor data exceeds its measurement range and / or whether the data change rate exceeds the second preset tolerance. If it does, the target sensor corresponding to the target sensor data is determined to be inaccurate. The "target sensor data" is determined through the following sub-modules: The first screening submodule is used to, in response to the detection of an anomaly in the target power supply line, select one or more suspicious sensor data as the target sensor data from all physical entity layer data based on the power performance and spatial distribution corresponding to the anomaly.

8. The cloud-based power dispatching system as described in claim 7, characterized in that, The first filtering submodule includes at least one of the following submodules: The second determining submodule is used to determine the sensor data of the monitoring point as the target sensor data if the anomaly is manifested as a serious deviation of the data of a single monitoring point from historical patterns or a preset threshold, while the data of other related points are normal. The second filtering submodule is used to filter out the sensor data located at the upstream key node of the power supply line in the region and whose data anomaly pattern is most related to the regional anomaly pattern, based on the topology of the digital twin power model, if the anomaly is manifested as anomalies in the data of multiple monitoring points in a region, as the target sensor data.

9. The cloud-based power dispatching system as described in claim 1, characterized in that, The target algorithm model includes a target prediction model; the first tracing module includes: The third filtering submodule is used to, in response to the identification of an anomaly in the target power supply line, select one or more as the target prediction model from multiple algorithm models based on the correlation between the characteristics of the anomaly and the prediction functions of various models in the model algorithm layer. The second calculation submodule is used to continuously calculate the error index between the predicted value and the actual value of the target prediction model within a preset time window. The first judgment submodule is used to determine whether the error index continuously exceeds a preset threshold. The seventh determination submodule is used to determine if the target prediction model has degraded in performance.

10. The cloud-based power dispatching system as described in claim 9, characterized in that, The third screening submodule includes at least one of the following submodules: The third determining submodule is used to determine the electricity consumption prediction model as the target prediction model if the anomaly is manifested as a continuous systematic deviation between the instructions of the scheduling system and the actual situation, and the deviation pattern is directly related to the output of the electricity consumption prediction model. The fourth determination submodule is used to determine the core model on which the optimized scheduling engine depends as the target prediction model if the anomaly is that the key performance indicators of the optimized scheduling scheme fail to reach the expected target after actual execution. The fifth determining submodule is used to determine the digital twin power model itself as the target prediction model if the anomaly is manifested as a persistent large error between the simulation results of the digital twin power model and a large amount of reliable actual measurement data.

Images