Intelligent power distribution network operation state sensing and predicting method, medium and equipment

Abstract

The invention discloses an intelligent power distribution network operation state sensing and predicting method, a medium and equipment, and relates to the technical field of intelligent power grids, and the specific steps are as follows: data acquisition and preprocessing, data set obtaining, power distribution network holographic digital twinborn model establishment, passive experience learning closed loop execution, and dynamic knowledge base establishment. And executing an active exploration cognitive closed loop, updating the dynamic knowledge base, and sensing and predicting the running state of the power distribution network based on the updated dynamic knowledge base. According to the method, through a double-loop driving mechanism of passive experience learning and active exploration cognition, a heuristic reasoning chain in a dynamic knowledge base is continuously evolved, a digital twinborn model is corrected, and a prediction result is output, so that the accuracy, the foresight and the adaptive capacity of state perception and prediction of the power distribution network are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and in particular to a method, medium, and device for sensing and predicting the operating status of a smart distribution network. Background Technology

[0002] With the widespread integration of new elements such as distributed renewable energy and electric vehicle charging stations, the operating environment of modern power distribution networks is becoming increasingly complex, with power flow direction and load characteristics exhibiting unprecedented time-varying and uncertainties. Therefore, accurate and real-time perception and prediction of the future operating state of the power distribution network has become a crucial prerequisite for ensuring the safe, stable, and economical operation of the power grid. Currently, the technical approaches to achieving power distribution network state prediction mainly rely on physical model-driven or purely data-driven methods. There is also a purely data-driven method based on artificial intelligence, which can also achieve power distribution network state prediction.

[0003] However, the drawback of this approach is:

[0004] (1) Difficulty in achieving dynamic synchronization between the model and reality: Traditional methods based on physical models typically construct simulation models using the topology and equipment parameters of the distribution network, and perform power flow calculations based on boundary conditions such as load forecasting. In the real world, the topology of the power grid may change temporarily due to operation and maintenance, and equipment parameters may drift due to aging. This information often cannot be updated in the model in a timely manner. The load and renewable energy output forecasts that form the basis of the calculations themselves have inherent errors, and the physical model itself lacks the ability to learn from historical forecast deviations and self-correct, making it easy for its forecast results to deviate from the actual situation.

[0005] (2) Over-reliance on historical data: Pure data-driven methods based on artificial intelligence. These methods build predictive models by mining correlation patterns in massive amounts of historical operational data. However, for rare high-risk events that have never appeared in historical data or occur with extremely low frequency, the data-driven model will be unable to make effective predictions due to the lack of training samples. In addition, many advanced data models are like "black boxes," with opaque decision-making processes that are difficult for operators to understand and trust, which is a significant obstacle in power systems that require high safety. Summary of the Invention

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] Firstly, a method for sensing and predicting the operating status of a smart distribution network, comprising:

[0009] Data acquisition and preprocessing yield a dataset;

[0010] Based on the obtained dataset, a holographic digital twin model of the power distribution network is established;

[0011] Based on the established holographic digital twin model of the distribution network, a passive experience learning closed loop is executed to establish a dynamic knowledge base;

[0012] An active exploration and cognitive closed loop is executed, a specific power grid scenario is defined, and the dynamic knowledge base is periodically evaluated to obtain a knowledge uncertainty measure;

[0013] The dynamic knowledge base is updated based on the knowledge uncertainty metric.

[0014] Based on the updated dynamic knowledge base, the operation status of the power distribution network is perceived and predicted.

[0015] As a preferred embodiment of the intelligent distribution network operation status perception and prediction method described in this invention, the method includes: executing a passive experience learning closed loop based on the established holographic digital twin model of the distribution network, comprising:

[0016] Situation prediction is performed based on the aforementioned holographic digital twin model of the distribution network, generating a predicted state vector;

[0017] Obtain the actual state vector, perform the first calculation on the predicted state vector and the actual state vector to obtain the state deviation;

[0018] Set the event trigger threshold;

[0019] When the state deviation exceeds the event trigger threshold, a first-type heuristic reasoning chain is generated;

[0020] Establish a dynamic knowledge base and store the first type of heuristic reasoning chain in the dynamic knowledge base.

[0021] The beneficial effects of this preferred technical solution are as follows: The passive experience-based learning closed loop, acting as a feedback correction pathway, continuously compares the actual operating data of the physical power grid with the model's predictions, extracting experience from real-world deviation events to generate a first-type heuristic reasoning chain. This pathway ensures that the knowledge in the dynamic knowledge base has been tested in the real world, achieving a deepening of understanding and refinement of knowledge regarding known scenarios.

[0022] As a preferred embodiment of the intelligent distribution network operation status perception and prediction method of the present invention, the method includes: executing an active exploratory cognitive closed loop, defining a specific power grid scenario, periodically evaluating the dynamic knowledge base, and obtaining a knowledge uncertainty metric, including:

[0023] Define a specific power grid scenario S j ;

[0024] According to the defined specific power grid scenario S jTraverse all reasoning chains stored in the dynamic knowledge base to obtain a set of reasoning chains.

[0025] According to the set of reasoning chains Perform the second calculation:

[0026]

[0027] Among them, U(S) j ) for a specific power grid scenario S j The measure of knowledge uncertainty is a scalar value; For all applicable scenarios S in the dynamic knowledge base j Heuristic reasoning chain R i The set of; w i Let be the confidence weight of the i-th heuristic reasoning chain in the set;

[0028] Obtain a measure of knowledge uncertainty.

[0029] The beneficial effects of this preferred technical solution are as follows:

[0030] The proactive exploratory cognitive loop serves as a forward-looking supplementary pathway. By proactively assessing the uncertainties of the dynamic knowledge base in specific scenarios, it identifies knowledge gaps and utilizes a holographic digital twin model of the power distribution network for targeted simulation and deduction, thereby generating a second type of heuristic reasoning chain. This pathway enables the dynamic knowledge base to cover high-risk scenarios that are rare or have never occurred in actual operation, achieving cognitive expansion and knowledge pre-filling of unknown domains.

[0031] In a second aspect, a computer device includes a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of a smart distribution network operation status perception and prediction method.

[0032] Thirdly, a computer-readable storage medium storing a computer program thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of a method for sensing and predicting the operating status of a smart power distribution network.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] This invention constructs a holographic digital twin model of a power distribution network and continuously evolves a unified dynamic knowledge base through a dual-loop driving mechanism: on the one hand, through a passive experience-learning closed loop, it extracts experience from measured deviations in the physical power grid to generate a first type of heuristic reasoning chain; on the other hand, through an active exploration-cognitive closed loop, it identifies knowledge gaps and uses the digital twin for simulation and deduction to generate a second type of heuristic reasoning chain. During state prediction, this method retrieves matching reasoning chains from the dynamic knowledge base using the current scene, adjusts the digital twin model with its included corrective actions, and then calculates the final prediction result. This invention integrates real-world experience with virtual exploration, significantly improving the accuracy, foresight, and adaptability of predictions. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is an overall flowchart of a smart distribution network operation status sensing and prediction method according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the construction of a holographic digital twin model of a smart distribution network operation status perception and prediction method according to an embodiment of the present invention;

[0038] Figure 3 This is a flowchart illustrating the passive experience learning closed loop of a smart distribution network operation status perception and prediction method according to an embodiment of the present invention.

[0039] Figure 4 This is a flowchart illustrating the active exploration and cognitive closed loop of a smart distribution network operation status perception and prediction method according to an embodiment of the present invention.

[0040] Figure 5 This is a schematic diagram of the adaptive prediction process of a smart distribution network operation status perception and prediction method according to an embodiment of the present invention. Detailed Implementation

[0041] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0042] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for sensing and predicting the operating status of a smart distribution network is provided, comprising:

[0043] S100: Data acquisition and preprocessing to obtain a dataset;

[0044] S200: Based on the obtained dataset, establish a holographic digital twin model of the power distribution network;

[0045] S300: Based on the established holographic digital twin model of the distribution network, it executes a passive experience learning closed loop to establish a dynamic knowledge base;

[0046] S400: Execute an active exploration and cognition closed loop, define a specific power grid scenario, periodically evaluate the dynamic knowledge base, and obtain a knowledge uncertainty measure;

[0047] S500: Update the dynamic knowledge base based on the knowledge uncertainty measure;

[0048] S600: Based on the updated dynamic knowledge base, it performs distribution network operation status perception and prediction.

[0049] It should be noted that with the widespread integration of new elements such as distributed renewable energy and electric vehicle charging piles, the operating environment of modern power distribution networks is becoming increasingly complex. Existing power distribution network sensing and prediction technologies rely solely on historical data for sensing and prediction, which prevents prediction models from learning from experience to avoid unknown risks. Furthermore, existing technologies cannot achieve dynamic synchronization between models and reality, resulting in low prediction accuracy. Therefore, accurate sensing and prediction of smart power distribution networks is also very important.

[0050] Therefore, to address the aforementioned problems of difficulty in synchronizing the prediction model with reality and low prediction accuracy, steps S100-S600 are employed to collect and preprocess data, establishing a holographic digital twin model of the distribution network; a passive experience learning closed loop is executed to establish a dynamic knowledge base; and an active exploration and cognition closed loop is executed to update the dynamic knowledge base. This achieves dynamic synchronization between the prediction model and reality. Based on the updated dynamic knowledge base, the operating status of the distribution network is perceived and predicted, thus solving the problem of low prediction accuracy in existing technologies.

[0051] Example 2, refer to Figure 2 , Figure 3 , Figure 4 , Figure 5 This is one embodiment of the present invention. Based on the above embodiment, a method for sensing and predicting the operating status of a smart distribution network is provided.

[0052] In this embodiment of the application, the data acquisition and preprocessing in step S100 to obtain a dataset includes the following steps A1-A2:

[0053] A1: Collect data and perform alignment and standardization processing to obtain standardized aligned data;

[0054] A2: The standardized data is cleaned and correlated to obtain a dataset.

[0055] In this embodiment of the application, the specific steps for data acquisition and preprocessing in A1-A2 are as follows:

[0056] A data fusion platform is established to collect data from data systems containing multi-source heterogeneous data. The collected data is then aligned to a spatiotemporal reference and standardized in format to generate standardized aligned data. The standardized aligned data is then cleaned and correlated. The cleaning includes filling in missing values ​​and removing abnormal data. The correlation involves mapping the external environment data and equipment status data to equipment objects or topology nodes within the holographic digital twin model of the power distribution network to generate a fused data stream to be injected.

[0057] It should be noted that the multi-source heterogeneous data includes real-time operating data, equipment status data, external environmental data, and historical load data. Real-time operating data includes voltage, current, active power, and reactive power measurements from the SCADA system, as well as synchronization phasor measurements from the PMU. Equipment status data includes equipment health status records and maintenance history records from the equipment asset management system. External environmental data includes temperature, wind speed, and light intensity information from the meteorological information system.

[0058] In this embodiment of the application, step S200 involves establishing a holographic digital twin model of the power distribution network based on the obtained dataset, including the following steps B1-B4:

[0059] B1: Establish a holographic digital twin model of the power distribution network;

[0060] B2: Input the dataset into the holographic digital twin model of the power distribution network;

[0061] B3: Dynamically update the operating parameters of the holographic digital twin model of the power distribution network based on the dataset.

[0062] In this embodiment of the application, the specific steps for establishing a holographic digital twin model of the distribution network in B1 include:

[0063] First, a refined digital mapping of the physical distribution network structure is performed. This process involves recording the static ledger information and topological connections of all key equipment objects from substations to end users in the distribution network. Equipment objects include feeders, sectionalizing switches, tie switches, transformers, capacitor banks, and also new grid-connected elements such as distributed power sources, energy storage systems, and electric vehicle charging / battery swapping stations. The rated parameters and dynamic electrical parameters of all equipment objects are accurately recorded, thus forming a digital network model that reflects the static structure of the physical power grid.

[0064] In this embodiment of the application, steps B2-B3 involve inputting the dataset into the holographic digital twin model of the distribution network and dynamically updating the operating parameters of the holographic digital twin model of the distribution network, including:

[0065] To drive the digital network model to reflect the dynamic behavior of the physical power grid in real time, this invention fuses and synchronizes multi-source heterogeneous data, establishing a data fusion platform. This platform performs unified data cleaning, format conversion, and timestamp alignment on all types of collected data, forming a standardized data stream. This data stream is continuously input into the digital network model to update the state parameters of each device in the model in real time, thereby ensuring high-precision synchronization between the holographic digital twin model of the distribution network and the physical power grid in terms of operational status.

[0066] It should be noted that by establishing a holographic digital twin model of the distribution network and dynamically updating the operating parameters of the holographic digital twin model of the distribution network based on the dataset, the dynamic synchronization between the physical power grid and the prediction model has been initially achieved. By continuously refreshing the network topology, equipment parameters and operating boundary conditions of the digital twin model through data fusion, the operating status of the physical distribution network is dynamically reflected.

[0067] In an alternative implementation, the establishment of a holographic digital twin model of the distribution network in B1 can also be achieved by using semantic modeling technology based on the IEC 61850 standard to perform standardized semantic modeling of the distribution network equipment through SCL (System Configuration Language). However, this solution has high modeling complexity, requires strict adherence to SCL syntax, involves a large amount of configuration work, and has a long debugging cycle.

[0068] In another alternative implementation, the holographic digital twin model of the distribution network in B1 can also be established by using graph neural network (GNN) to construct a graph structure model of the distribution network topology through dynamic topology inference technology based on graph neural network. However, this scheme has strong data dependence, requires massive labeled data to train the model, and has weak generalization ability for rare fault scenarios.

[0069] In this embodiment of the application, step S300 involves executing a passive experience learning closed loop based on the established holographic digital twin model of the distribution network to build a dynamic knowledge base, including the following steps C1-C5:

[0070] C1: Based on the aforementioned holographic digital twin model of the distribution network, perform situation prediction and generate a predicted state vector;

[0071] C2: Obtain the actual state vector, perform the first calculation on the predicted state vector and the actual state vector to obtain the state deviation;

[0072] C3: Sets the event trigger threshold;

[0073] C4: When the state deviation exceeds the event trigger threshold, generate the first type of heuristic reasoning chain;

[0074] C5: Establish a dynamic knowledge base and store the first type of heuristic reasoning chain in the dynamic knowledge base.

[0075] Specifically, the form in which the predicted state vector is generated in C1 is as follows:

[0076] P pred =[V1,V2,…,V m ,P 12 Q 12 ,…,P ij Q ij ] T

[0077] Among them, P pred V represents the predicted state vector. m P represents the voltage magnitude at the m-th critical node; ij and Q ij These represent the active power flow and reactive power flow on the important branch connecting node i and node j, respectively.

[0078] It should be noted that the specific steps for generating the predicted state vector are as follows:

[0079] First, key variables affecting the future state of the power grid are predicted. A load forecasting module uses historical load data and external environmental data to calculate the predicted active and reactive power values ​​for each load node in the future time frame. Simultaneously, a renewable energy output forecasting module uses external environmental data such as solar irradiance and wind speed to calculate the predicted power output values ​​of distributed generation in the future time frame.

[0080] Secondly, the predicted load power and the predicted renewable energy output are applied as boundary conditions to the holographic digital twin model of the distribution network. Under these boundary conditions, the system performs a power flow calculation on the digital twin model. The result of the power flow calculation is the complete state solution of the distribution network in the future time segment.

[0081] Finally, a set of pre-defined key operational indicators are extracted from the state solutions of the power flow calculation, and these indicators are combined into a structured predicted state vector PpredPpred. The predicted state vector contains at least the voltage magnitude of key nodes and power flow information of important branches in the distribution network.

[0082] Specifically, the first calculation formula in C2 is:

[0083]

[0084] Where ΔP is the final calculated scalar value of the state deviation; P pred P is the predicted state vector; act P is the actual state vector obtained in this step. norm This is the reference vector used for normalization.

[0085] It should be noted that the actual state vector acquisition steps are as follows: the system obtains the predicted state vector P from the data fusion platform of the distribution network holographic digital twin model. pred The target time segment corresponds perfectly to the actual measurement data. The actual measurement data comes from systems such as SCADA and PMU. The system extracts and predicts the state vector P from the actual measurement data. pred A set of key operational metrics that are completely consistent in data structure and physical meaning are combined into an actual state vector P. act .

[0086] It should be noted that P norm The reference vector is used for normalization. It has the same dimension as the state vector, and each element represents the reference or nominal value of the state quantity at the corresponding location. Its function is to unify the deviations of state quantities with different physical units and dimensions onto the same scale for comparison. ||·||2 is the Euclidean norm, used to calculate the magnitude of the vector, aggregating multidimensional state deviation information into a total deviation metric.

[0087] Specifically, the steps for setting the event trigger threshold in C3 are as follows:

[0088] The threshold for triggering deviation events is not a single fixed value, but follows a composite rule that combines dynamic statistical analysis with absolute safety constraints. This composite rule first includes a dynamic statistical threshold T. stat This threshold is adaptively set based on the system's normal prediction error level within recent historical periods. Specifically, the system maintains a sliding time window record of historical state deviation values ​​ΔP and periodically calculates the moving average μ of all deviation values ​​within the window. ΔP and moving standard deviation σ ΔP The dynamic statistical threshold is calculated using the following formula: T stat =μΔP +k·σ ΔP , where k is a preset sensitivity factor.

[0089] This composite rule also includes a set of absolute safety thresholds T. abs This threshold is a fixed set of constraints, set based on inviolable safe operating limits defined by power grid operating procedures, such as the requirement that the voltage amplitude at critical nodes must be maintained within a specific percentage of the nominal value. This threshold is independent of the model's predictive performance.

[0090] The system determines that the state deviation exceeds the preset deviation event trigger threshold when at least one of the following two conditions is met:

[0091] Condition 1: The calculated state deviation scalar value ΔP is greater than the dynamic statistical threshold T. stat ;

[0092] Condition 2: Obtain the actual state vector P act In the middle, at least one component value touches or exceeds its corresponding absolute safety threshold T. abs The defined boundary.

[0093] Specifically, generating the first type of heuristic reasoning chain in C4 includes:

[0094] When the state deviation ΔP exceeds the deviation event trigger threshold, the system determines that a learnable deviation event has occurred and automatically initiates an attribution analysis process to uncover the key combination of preconditions that led to this deviation.

[0095] After identifying the combination of preconditions, the system encapsulates and structures it into a first-class heuristic reasoning chain, which is expressed using rules in the form of "IF THEN".

[0096] It should be noted that the "IF" part of the rule, i.e., the antecedent, contains the identified combination of preconditions. The "THEN" part of the rule, i.e., the consequent, contains one or a set of predefined corrective actions that should be performed for the combination of preconditions. These corrective actions are designed to offset or prevent similar deviations from recurring under the same conditions in the future.

[0097] It should be noted that, in order to quantify the effectiveness and reliability of each heuristic reasoning chain in the dynamic knowledge base, the system assigns an iteratively updatable confidence weight to each newly generated first-type heuristic reasoning chain (and the subsequently generated second-type heuristic reasoning chain). When a reasoning chain is first generated, it is assigned a preset initial confidence weight value.

[0098] The confidence weights are updated periodically to reflect the historical performance of the inference chain in real-world applications. The update of the confidence weights is performed through a weighted summation calculation that incorporates a forgetting factor. The calculation combines the confidence weights of the heuristic inference chain in the previous period with its performance evaluation value in the current period to generate the confidence weights of the heuristic inference chain for the next period.

[0099] Specifically, the update process is implemented using the following formula:

[0100] w i,t+1 =α·w i,t +(1-α)·Φ(R i ,t)

[0101] Among them, w i,t+1 w represents the confidence weight of the i-th heuristic inference chain in the next iteration period t+1; i,t Φ(R) represents the confidence weight of the i-th heuristic reasoning chain in the current iteration period t, α is the forgetting factor, and Φ(R) represents the confidence weight of the i-th heuristic reasoning chain in the current iteration period t. i Let ,t) be the i-th heuristic reasoning chain R. i The performance evaluation function in the current period t.

[0102] It should be noted that the forgetting factor α ranges from 0 to 1. The forgetting factor is used to adjust the weighting of historical performance versus current performance evaluation in the update calculation. A larger α value means that historical performance carries a higher weight, while a smaller α value means that more emphasis is placed on recent performance.

[0103] It should be noted that the performance evaluation function Φ(R) i The specific implementation of the rule (t) involves the system recording the number of times the inference chain is triggered within a cycle, and whether the recommended correction action after each trigger reduces the subsequent state prediction bias. The performance evaluation value is quantified based on its positive contribution to reducing prediction bias; for example, if the prediction accuracy improves after applying the rule, the evaluation value is positive, otherwise it is negative or zero. This iterative update mechanism ensures that the confidence weights dynamically reflect the actual utility of each rule in the constantly changing power grid environment.

[0104] In one alternative implementation, the actual state vector is obtained in C2, and a first calculation is performed on the predicted state vector and the actual state vector to obtain the state deviation. Alternatively, the deviation caused by the time axis offset can be eliminated by using the Dynamic Time Warping (DTW) method to find the optimal matching path between the two sequences. However, the window width constraint of this scheme needs to be set empirically, which affects the accuracy.

[0105] In another alternative implementation, the actual state vector is obtained in C2, and a first calculation is performed on the predicted state vector and the actual state vector to obtain the state deviation. Alternatively, the state vector can be regarded as a probability distribution by a probability distribution metric based on Wasserstein distance, and the difference between the predicted and actual distributions can be quantified by the Earth Mover's Distance. However, this scheme ignores the physical location correlation of the state variables, which can easily lead to information loss.

[0106] In this embodiment of the application, step S400 involves performing an active exploration and cognitive closed loop, defining a specific power grid scenario, periodically evaluating the dynamic knowledge base, and obtaining a knowledge uncertainty measure, including the following steps D1-D3:

[0107] D1: Define a specific power grid scenario S j ;

[0108] D2: Based on the defined specific power grid scenario S j Traverse all reasoning chains stored in the dynamic knowledge base to obtain a set of reasoning chains.

[0109] D3: Based on the set of reasoning chains A second calculation is performed to obtain a measure of knowledge uncertainty.

[0110] Specifically, in D1, the power grid scenario S j It is a combination of a set of preset key operating states or topological features.

[0111] For example, a scenario can be defined as "a combination of high load levels and high distributed generation output," or "a network topology where a specific feeder is disconnected." These scenarios represent typical or extreme operating conditions that require attention in distribution network operation.

[0112] Specifically, the steps for obtaining the inference chain set in D2-D3 and performing the second calculation to obtain the knowledge uncertainty measure are as follows:

[0113] Connect the antecedent (i.e., the IF part) of each inference chain with scenario S. j The defined conditions are logically matched to filter out all heuristic reasoning chains applicable to the scenario, and these filtered reasoning chains are then compiled into a set. set The confidence weights of all heuristic reasoning chains are summed. This sum reflects the current dynamic knowledge base's understanding of scenario S. j To determine the overall knowledge strength, add one to the sum obtained in the previous step, then take its reciprocal to perform the second calculation and obtain the specific power grid scenario S. j Knowledge uncertainty measure U(S)j The second calculation is expressed by the following formula:

[0114]

[0115] Among them, U(S) j ) for a specific power grid scenario S j The measure of knowledge uncertainty is a scalar value; For all applicable scenarios S in the dynamic knowledge base j Heuristic reasoning chain R i The set of; w i Let be the confidence weight of the i-th heuristic reasoning chain in the set.

[0116] It should be noted that this calculation method establishes an inverse relationship. When there are many heuristic reasoning chains applicable to a specific scenario and the sum of their confidence weights is large, the calculated knowledge uncertainty metric is low, indicating that the system has a strong understanding of this scenario. When there are few heuristic reasoning chains applicable to a specific scenario or the sum of their confidence weights is small, the calculated knowledge uncertainty metric is high, indicating that the system's understanding of this scenario is insufficient.

[0117] It should be noted that the traversal operation is performed on the set of heuristic reasoning chains already existing in the dynamic knowledge base at the time this step is executed. In the initial running phase of the method, this set may only contain the first type of heuristic reasoning chains. As the method continues to run, this set will contain both the first type and subsequently generated second type of heuristic reasoning chains. This step's periodic evaluation of the current state of the dynamic knowledge base provides a basis for triggering judgments in subsequent steps.

[0118] In one alternative implementation, D3 is based on the inference chain set. A second calculation is performed to obtain a measure of knowledge uncertainty. Furthermore, the information entropy method can be used to refine the dynamic knowledge base to fit scenario S. j All inference chains are implemented as a probability distribution system, but this approach only relies on numerical distribution and does not distinguish the logical correlation between rules.

[0119] In another alternative implementation, D3 is based on the inference chain set. A second calculation is performed to obtain a measure of knowledge uncertainty. Furthermore, each inference chain can be treated as a piece of evidence using a DS-based evidence theory method, and the confidence weight can be transformed into the relationship between that piece of evidence and scenario S. j This method implements basic probability allocation, but the BPA allocation relies on manual setting or heuristic transformation, and has high computational complexity.

[0120] In this embodiment of the application, step S500, which updates the dynamic knowledge base based on the knowledge uncertainty measure, includes the following steps E1-E3:

[0121] E1: Establish an active exploration trigger threshold;

[0122] E2: When the uncertainty measure exceeds the preset active exploration trigger threshold, a second type of heuristic reasoning chain is generated;

[0123] E3: Update the dynamic knowledge base based on the second type of heuristic reasoning chain.

[0124] Specifically, the steps for establishing the active exploration trigger threshold in E1 are as follows:

[0125] For a specific power grid scenario S j Criticality is classified. The classification is based on the severity of the potential consequences for the power grid if the scenario were to occur.

[0126] For example, it can be divided into:

[0127] Level 1 (highest criticality): Scenarios that may lead to widespread power outages or damage to critical equipment;

[0128] Level 2 (higher criticality) involves scenarios that may lead to a deterioration in local power quality;

[0129] Level 3 (General Criticality): Scenarios involving other conventional operating methods.

[0130] Secondly, based on the criticality level, the active exploration trigger threshold T is set. explore The approach is to reverse the settings. That is, the higher the criticality level of a scenario, the lower the corresponding threshold for triggering proactive exploration. The underlying logic is that the system has extremely low tolerance for cognitive uncertainty in high-risk scenarios; even if the knowledge uncertainty metric is relatively low, proactive exploration should still be triggered to ensure knowledge completeness. For general scenarios, exploration is only triggered when knowledge is extremely scarce.

[0131] It should be noted that the active exploration trigger threshold is not a globally uniform fixed value, but follows a principle related to the criticality of the power grid scenario, and is configured differently according to the potential impact of different scenarios on the safe operation of the power grid.

[0132] Specifically, the steps for generating the second type of heuristic reasoning chain in E2 are as follows:

[0133] Determine scenario S j The criticality level is determined, and its corresponding trigger threshold T is obtained. explore,level_n When the knowledge uncertainty measure U(S) jWhen the threshold for triggering a level is exceeded (i.e., when U(S) is satisfied), the threshold for triggering a level is exceeded. j )>T explore,level_n The system determines that the current dynamic knowledge base is relevant to scenario S. j The understanding is insufficient, and an active exploratory deduction task is automatically generated for this scenario.

[0134] It should be noted that the proactive exploratory inference task is a "What-if" type simulation task. The task construction process is as follows: the system constructs a high-uncertainty scenario S. j The defined conditions are transformed into a set of specific simulation boundary conditions and event sequences. For example, if scenario S j If defined as "the load in a certain area exceeds 120% of the rated value and an N-1 fault occurs at the same time", then the generated simulation task is to set the load parameters of the corresponding area to the value in the holographic digital twin model of the distribution network and simulate the occurrence of the N-1 fault.

[0135] The simulation boundary conditions defined in the simulation task are applied to the digital twin model, causing the model to enter the specific state to be explored. Then, a pre-set sequence of events is run in the model, and one or more simulation calculations are performed, such as power flow analysis or dynamic simulation.

[0136] It should be noted that the data from the entire simulation calculation process were recorded, forming a dataset describing the complete response behavior of the power grid under this exploratory scenario. This dataset, because it is not derived from direct measurements of the physical power grid, is defined as "synthetic experience."

[0137] The system inputs the synthesized experience into an inference chain generation module, which analyzes the experience to extract the logical relationship between simulation conditions and grid response results. This process identifies key precondition combinations that lead to specific system states (e.g., voltage exceedance, line overload) under specific virtual operating conditions. After identifying the logical relationship, the system encapsulates and structures it into a second-type heuristic inference chain. Similar to the first-type heuristic inference chain, the second-type heuristic inference chain also uses "IF THEN" rules for expression.

[0138] It should be noted that the "IF" part of a rule in the form of "IF THEN," i.e., the antecedent, contains a combination of conditions for a specific power grid scenario simulated by the simulation task. The "THEN" part of the rule, i.e., the consequent, contains a set of control strategies or corrective actions determined based on the simulation results, which can prevent or mitigate adverse consequences in that scenario.

[0139] It should be noted that this newly generated second-type heuristic reasoning chain, along with its preset initial confidence weights, is updated in the dynamic knowledge base. Its confidence weights are also subsequently optimized following the iterative update mechanism described in the passive experience learning loop, thus forming a unified knowledge foundation for this invention together with the first-type heuristic reasoning chain.

[0140] In this embodiment of the application, step S600, based on the updated dynamic knowledge base, performs distribution network operation status perception and prediction, including the following steps F1-F2:

[0141] F1: Continuously execute passive experience learning loops and active exploration and cognition loops to update the dynamic knowledge base;

[0142] F2: Based on the updated dynamic knowledge base, extract the heuristic reasoning chain from the dynamic knowledge base, correct the holographic digital twin model of the power distribution network, and output the prediction results.

[0143] In this embodiment of the application, the specific steps for correcting the holographic digital twin model of the distribution network and outputting the prediction results in F2 are as follows:

[0144] When a state prediction task for a future time segment is triggered, the system first performs an initial power flow calculation on the holographic digital twin model of the distribution network according to the method in step S300, that is, based on boundary conditions such as load prediction and new energy output prediction, to generate a preliminary predicted state vector.

[0145] While generating the initial predicted state vector, the system extracts the complete set of input conditions used to generate the prediction. This set of input conditions constitutes the current power grid scenario to be predicted. The system uses this scenario as query conditions to perform a matching retrieval in the dynamic knowledge base. This retrieval process logically compares the scenario conditions with the "IF" antecedent parts of all first- and second-type heuristic reasoning chains in the dynamic knowledge base.

[0146] If one or more heuristic inference chains that match the current scenario are found, the system will retrieve these matching inference chains. When multiple matching inference chains exist, the system will make a decision based on the confidence weight of each inference chain. For example, it may select the inference chain with the highest confidence weight, or combine the correction actions of multiple inference chains according to a preset aggregation algorithm.

[0147] The system applies the correction actions defined in the "THEN" consequent of the selected inference chain to the holographic digital twin model of the distribution network. The correction action is a one-time adjustment of the model parameters or boundary conditions, such as increasing or decreasing the load forecast value of a specific node in the model by a specific percentage, or adjusting the reactive power output setpoint of a power generation device.

[0148] After the state of the holographic digital twin model of the distribution network is adjusted, the system re-executes the power flow calculation based on the adjusted model state. The result of this calculation is output as the final, knowledge-corrected prediction result.

[0149] It should be noted that, because the generation process of this vector incorporates knowledge gained from historical experience and forward-looking deductions, its description of the future state of the real world is more accurate than that of the initial predicted state vector.

[0150] In one alternative implementation, F2 can extract heuristic reasoning chains from the updated dynamic knowledge base, correct the holographic digital twin model of the distribution network, and output prediction results. Alternatively, it can use static knowledge injection technology based on rule engines, employing enterprise-level rule engines (such as Drools and Jess) to predefine the power grid operation rule base, and trigger the rule correction model by matching scenarios in real time. However, this solution has fixed knowledge and the rules rely on manual summarization, making it difficult to adapt to new topologies or rare operating conditions.

[0151] In one alternative implementation, F2 can extract heuristic reasoning chains from the updated dynamic knowledge base, correct the holographic digital twin model of the distribution network, and output prediction results. Alternatively, the digital twin model can be used as a simulation environment to train a reinforcement learning (RL) agent to learn and correct the policy through dynamic policy optimization technology based on reinforcement learning. However, this approach has high training costs, requires tens of thousands of simulation iterations, and consumes a lot of computational resources.

[0152] In summary, this invention establishes a holographic digital twin model of the distribution network, builds a dynamic knowledge base, and executes a passive experience-based learning loop and an active exploration-based cognitive loop. This ensures that the knowledge in the dynamic knowledge base has been tested in the real world, achieving a deepening of understanding and refinement of knowledge for known scenarios. Furthermore, the dynamic knowledge base can cover high-risk scenarios that are rare or have never occurred in actual operation, achieving cognitive expansion and knowledge pre-filling for unknown areas. Through this dual-loop drive, the dynamic knowledge base achieves continuous and iterative self-improvement. Finally, based on the updated dynamic knowledge base, the distribution network operating status is perceived and predicted. This invention significantly improves the accuracy, foresight, and adaptability of distribution network status perception and prediction.

[0153] Example 3: This example provides a computer device, including a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement the steps of a smart distribution network operation status perception and prediction method.

[0154] This embodiment proposes a computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of a smart distribution network operation status perception and prediction method.

[0155] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0156] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 method for smart power grid operation state perception and prediction, characterized in that, The method comprises the following steps: data acquisition and preprocessing to obtain a data set; establishing a holographic digital twin model of the power distribution network according to the obtained data set; performing a passive experience learning closed loop according to the established holographic digital twin model of the power distribution network to establish a dynamic knowledge base; performing an active exploration and cognition closed loop, defining a specific power grid scenario, periodically evaluating the dynamic knowledge base, and obtaining a knowledge uncertainty measure; updating the dynamic knowledge base according to the knowledge uncertainty measure; based on the updated dynamic knowledge base, performing power distribution network operation state perception and prediction.

2. The smart grid operational state awareness and prediction method of claim 1, wherein, The data acquisition and preprocessing to obtain a data set comprises: collecting data and aligning and standardizing the data to obtain standardized aligned data; cleaning and correlating the standardized aligned data to obtain a data set.

3. The smart grid operational state awareness and prediction method of claim 2, wherein, The method of establishing a holographic digital twin model of the power distribution network according to the obtained data set comprises: establishing a holographic digital twin model of the power distribution network; inputting the data set into the holographic digital twin model of the power distribution network; dynamically updating the operating parameters of the holographic digital twin model of the power distribution network according to the data set.

4. The smart grid operational state awareness and prediction method of claim 3, wherein, The method of performing a passive experience learning closed loop according to the established holographic digital twin model of the power distribution network to establish a dynamic knowledge base comprises: performing situation prediction based on the holographic digital twin model of the power distribution network to generate a predicted state vector; obtaining an actual state vector, performing a first calculation on the predicted state vector and the actual state vector, and obtaining a state deviation; setting an event trigger threshold; when the state deviation exceeds the event trigger threshold, generating a first heuristic reasoning chain; establishing a dynamic knowledge base and storing the first heuristic reasoning chain in the dynamic knowledge base.

5. The smart grid operational state awareness and prediction method of claim 4, wherein, The method of performing an active exploration and cognition closed loop, defining a specific power grid scenario, periodically evaluating the dynamic knowledge base, and obtaining a knowledge uncertainty measure comprises: Defining a specific power grid scenario S j ; According to the defined specific grid scenario S j , traverse all inference chains stored in the dynamic knowledge base to obtain an inference chain set According to the set of inference chains A second calculation is made: where U(S j ) is the knowledge uncertainty measure for a specific grid scenario S j , which is a scalar value; is the set of all heuristic inference chains R j applicable to scenario S i in the dynamic knowledge base; w i is the confidence weight of the i-th heuristic inference chain in the set. obtaining a knowledge uncertainty measure.

6. The smart grid operational state awareness and prediction method of claim 5, wherein, The method of updating the dynamic knowledge base according to the knowledge uncertainty measure comprises: establishing an active exploration trigger threshold; when the uncertainty measure exceeds the preset active exploration trigger threshold, generating a second heuristic reasoning chain; updating the dynamic knowledge base according to the second heuristic reasoning chain.

7. The smart grid operational state awareness and prediction method of claim 6, wherein, The method of performing power distribution network operation state perception and prediction based on the updated dynamic knowledge base comprises: continuously performing a passive experience learning closed loop and an active exploration and cognition closed loop to update the dynamic knowledge base; based on the updated dynamic knowledge base, extracting heuristic reasoning chains in the dynamic knowledge base to obtain correction factors; according to the correction factors, correcting the holographic digital twin model of the power distribution network and outputting a prediction result.

8. The smart grid operational state awareness and prediction method of claim 7, wherein, The first and second heuristic reasoning chains comprise: when the first and second heuristic reasoning chains are established, they are assigned an iteratively updated confidence weight; the confidence weight is periodically updated, and the updating process is realized by the following formula: w i,t+1 = a - w i,t + (1 - a) - Φ(R i , t) where w i,t+1 represents the confidence weight of the ith heuristic inference chain in the next iteration period t+1; w i,t represents the confidence weight of the ith heuristic inference chain in the current iteration period t, and a is a forgetting factor. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the intelligent power distribution network operation state perception and prediction method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the intelligent power distribution network operation state perception and prediction method of any one of claims 1 to 7.

Images