Power grid dispatching anti-error method and system based on topology analysis

Abstract

The invention discloses a power grid dispatching anti-error method and system based on topology analysis, and relates to the technical field of power grid dispatching anti-error, and the method comprises the following steps: building a remote signaling credibility evaluation model based on remote signaling evaluation data, and obtaining high-credibility remote signaling data; performing sequential logic verification based on high-credibility remote signaling data and multi-source data fusion to generate a power grid topology model; carrying out partition division on the power grid topology model based on spectral clustering and a state-driven edge weight method, and carrying out boundary equivalence processing to obtain a plurality of topology modules; and performing simulation scheduling operation on the plurality of topology modules based on a scheduling instruction dispatching order objectification analysis result, performing path traversal based on a depth-first search method, and performing anti-error analysis on the traversed path in combination with a reinforcement learning method. According to the power grid dispatching anti-error method combining remote signaling credibility evaluation, dynamic topology partition and reinforcement learning algorithm, the anti-error verification capability and the intelligent level of a power grid dispatching system are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching error prevention, and more specifically, to a power grid dispatching error prevention method and system based on topology analysis. Background Art

[0002] The grid dispatching system is the core support for the safe and stable operation of the power system. The correctness of its dispatching operation is directly related to the reliability of power supply and the safety of system operation. Traditional grid dispatching operations rely on manual experience and static grid topology information. Especially in terms of preventing misoperation, it mainly relies on rule bases, five-defense systems, and process control to prevent and pursue responsibility. This type of method shows obvious lag and limited adaptability when faced with complex and changeable grid operation environments, frequently adjusted wiring methods, and uncertainty in telesignaling data.

[0003] For example, the invention patent with announcement number CN119599285A announces a power grid analysis system and analysis method based on a topological graph, including: collecting power grid equipment information and topological connection relationship data, building an equipment status library and a power grid topology model; obtaining several node importances and several edge connection densities according to the power grid topology model, and partitioning the power grid using a multi-scale strongly connected region finite partitioning method to obtain several power grid partitions; obtaining comprehensive risk indexes of several power grid partitions according to several node importances and several edge connection densities, and screening out power grid partitions that exceed a preset risk threshold to obtain risk partitions; collecting operating data of risk partitions, inputting the operating data into a fault prediction model built based on deep learning, and obtaining the fault time and fault equipment predicted by the risk partitions. The present invention relates to the technical field of power grid risk analysis, and solves the technical problems of low analysis efficiency and insufficient pertinence in existing power grid analysis methods.

[0004] For example, the invention patent with announcement number: CN103413044B announces a method for estimating the local topology of a power system based on substation measurement information, including logical analysis of the telemetry and telesignaling data of the power system collected by the SCADA system, identifying suspicious telemetry and telesignaling data, and locating the substation to which the suspicious data belongs, using the power and current measurement information collected in real time in a single or multiple adjacent substations, establishing a linear programming model, and estimating the local topology of the transmission network in real time, and correcting errors in the data. The method proposed by the present invention can accurately identify topological errors in substations through optimization principles, and provide reliable input data for the state estimation program. This reduces the interference of suspicious measurements and prevents local errors from affecting global results.

[0005] The above disclosed technical solutions have at least the following technical problems: The existing technology for processing telesignaling data has the problems of low credibility and inability to dynamically verify. The current system usually assumes that telesignaling data is authentic and reliable, and lacks a mechanism for quantitative evaluation and multi-source verification of telesignaling signals, which can easily cause topology identification errors and lead to scheduling decision errors. For example, phenomena such as telesignaling jitter, signal delay, or equipment status misreporting are often not recognized and corrected in time in the existing system; Most traditional topology modeling methods are based on static graph models or manual input methods, which are difficult to adapt to the characteristics of frequent changes in power grid structure and dynamic evolution of operating status. This method cannot reflect the conduction status and operating logic of power grid equipment in real time, which seriously restricts the intelligence level and error prevention ability of dispatching decisions. In addition, traditional path verification is mainly based on rule matching, which cannot effectively discover potential linkage risks in complex logical paths or possible misoperations under boundary conditions, further reducing the effectiveness of anti-error verification of dispatching instructions.

[0006] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a power grid dispatching error prevention method and system based on topology analysis. By combining remote signal credibility evaluation, dynamic topology partitioning, graph neural search and reinforcement learning algorithm, the power grid dispatching error prevention method can realize dynamic simulation of the entire process of dispatching instructions, error prevention verification and intelligent path identification, so as to comprehensively improve the error prevention verification capability and intelligence level of the power grid dispatching system.

[0008] To achieve the above object, the present invention provides the following technical solutions: The power grid dispatch error prevention method based on topological analysis includes the following steps: constructing a telesignal credibility evaluation model based on telesignal evaluation data to obtain highly reliable telesignal data; performing temporal logic verification based on highly reliable telesignal data and multi-source data fusion to generate a power grid topology model; partitioning the power grid topology model based on spectral clustering and state-driven edge weight method, and performing boundary equivalence processing to obtain several topological modules; simulating dispatch operations on several topological modules based on the object-oriented parsing results of dispatch instructions, performing path traversal based on the depth-first search method, and performing error prevention analysis on the traversed path in combination with the reinforcement learning method.

[0009] In a preferred embodiment, a sequential logic verification is performed based on high-reliability telesignaling data and multi-source data fusion to generate a power grid topology model, specifically: primary equipment is acquired, and an initial structure diagram is established based on the power connection relationship and physical location; the sequential logic verification is performed on the high-reliability telesignaling data based on SOE and waveform data; the equipment operation status is acquired based on the verified telesignaling data, and the equipment operation status is obtained by parsing the telesignaling data based on the mapping relationship between the verified telesignaling data and the primary equipment; the initial structure diagram is dynamically updated according to the equipment operation status and the graph traversal algorithm, and the updated initial structure diagram is subjected to connected component analysis to identify the main connection form of the substation and generate a power grid topology model.

[0010] In a preferred embodiment, the highly reliable telesignaling data is subjected to a sequential logic verification based on the SOE and the waveform recording data, specifically: the telesignaling change time and the telesignaling state change point in the telesignaling data are obtained; the telesignaling change time is matched and compared with the event timing registered in the SOE, and it is determined whether the telesignaling change is triggered by a real operation; the telesignaling state change point is corresponded to the digital quantity change trajectory in the waveform recording data, and the waveform slope mutation point within a preset time window is extracted to determine whether there is a corresponding change; and the telesignaling data that the telesignaling change is a real operation after verification and there is a corresponding change is obtained.

[0011] In a preferred embodiment, the power grid topology model is partitioned based on spectral clustering and state-driven edge weight method, specifically: based on the normalized Laplace matrix, the first several eigenvectors of the Laplace matrix are obtained to form a spectral embedding space, and the normalized Laplace matrix is ​​constructed based on a weighted adjacency matrix, and the weighted adjacency matrix includes state-aware edge weights, logical linkage strength weights, and historical misoperation statistics; spectral clustering is used to embed and map the power grid topology based on the weighted adjacency matrix, and k-Means is introduced to cluster the nodes after the nodes are projected into the spectral embedding space; the partition category after each node clustering is obtained and the boundary of the power grid topology model is divided to obtain several subgraphs.

[0012] In a preferred embodiment, boundary equivalence processing is performed to obtain several topological modules, specifically: the edges of the several subgraphs are traversed to identify boundary nodes and cross-zone edges; the cross-zone edges are classified according to the switch status, the cross-zone edges that need to be subjected to impedance equivalence processing are retained, and the operation data of the cross-zone edges are obtained; equivalent nodes are introduced into several subgraphs, and the initial voltage values ​​of the equivalent nodes are set according to the operation data; the boundary nodes and equivalent nodes in the subgraphs are connected through impedance elements to form virtual branches; the boundary line status is detected, and the voltage value of each boundary node is obtained in real time, and the initial voltage value of the equivalent node is dynamically corrected; each subgraph after the boundary equivalence processing is stored as a topological module, and voltage and power injection simulation is performed based on the virtual branch and the boundary node, and comparative verification is performed to obtain several topological modules.

[0013] In a preferred embodiment, a scheduling operation is simulated for several topological modules based on the object-oriented parsing results of the scheduling instructions, and a path traversal is performed based on a depth-first search method, and an error-prevention analysis is performed on the traversed path in combination with a reinforcement learning method, specifically: a scheduling instruction is written based on a graphic proposed scheduling order and a text proposed scheduling order to obtain a scheduling instruction; the scheduling instruction is object-oriented and parsed into standardized digital tasks, and parsed into steps with a single device operation as the smallest granularity based on error-prevention rules; the object-oriented scheduling instruction is simulated and bound to the corresponding topological module, and simulated execution is performed based on the smallest granularity steps, and the topological model is updated in real time; an operation graph structure is obtained based on the updated topological model, and the operation graph structure is constructed by connecting each device node and the logical conduction relationship in the topological model; the conduction path of the device nodes in the operation graph structure is traversed by a depth-first search method, and an error-prevention analysis is performed on the conduction path based on a reinforcement learning method.

[0014] In a preferred embodiment, the conductive paths of the device nodes in the operation graph structure are traversed by a depth-first search algorithm, and the conductive paths are analyzed for error prevention based on a reinforcement learning method, specifically: the device state data at the starting point of the depth-first search algorithm is obtained as the state in reinforcement learning; the set of all scheduling instructions that can be executed at the current starting point is used as the action in reinforcement learning; the reinforcement learning strategy Agent is trained based on the probability of selecting the corresponding action under the current state; a path screening reward function is constructed; a reinforcement learning state-action model is constructed based on the state, action reinforcement learning strategy Agent and the path screening reward function; based on the reinforcement learning state-action model, the path screening value is predicted for all conductive paths traversed by the depth-first search algorithm; based on the path screening value, path judgment is performed, and potential error prevention risks are marked.

[0015] In a preferred embodiment, the time jitter entropy data is specifically obtained in the following manner: Assume that the telesignal point changes position several times within a preset time window, and obtain the time interval vector by calculating the sequence of time intervals between adjacent changes; map the time interval vector to several intervals of fixed width for binning, count the number of intervals contained in each interval, and normalize it to probability distribution to obtain probability density; calculate Shannon entropy based on probability density to obtain time jitter entropy data.

[0016] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. By introducing a remote signal credibility assessment model, integrating multi-source data indicators such as time jitter entropy, state mutation rate, and multi-channel signal consistency rate, and combining SOE and recording data for timing logic verification, the dynamic credibility judgment and abnormal elimination of remote signal data are realized, and the accuracy and stability of power grid equipment status identification are improved from the source. This method effectively avoids the risk of interference from remote signal false alarms, missed alarms, or glitch signals on topology modeling and dispatching operations, ensures that the data foundation for the construction of the dynamic topology structure is true and reliable, and the logical link closure is rigorous, significantly enhancing the anti-error locking mechanism of the dispatching operation and the system safety protection capability, and has high engineering practical value and foresight.

[0017] 2. By introducing the state-driven edge weight definition and spectral clustering algorithm to realize the dynamic partitioning of the power grid topology model, it can accurately identify the logical disconnection and connectivity adjustment caused by the change of the switch device state, and avoid the risk of misidentification caused by the static processing of topology partitioning; at the same time, combined with the boundary equivalence processing mechanism, the electrical effect of the cross-zone edge is retained after the subgraph is divided, ensuring that each topology module has electrical integrity and boundary response capabilities in error prevention analysis. This method significantly improves the real-time, accuracy and robustness of partition modeling, supports the rapidity and localized response capabilities of error prevention analysis in power grid dispatching, takes into account global consistency and local sensitivity, and has extremely high practical application value.

[0018] 3. Through object-oriented parsing of dispatching orders and reinforcement learning-assisted depth-first search path analysis, standardized mapping and intelligent verification of dispatching orders from text / graphic input to device-level operation sequences are achieved, and an anti-error analysis model integrating operation sequences, device topology status, and high-risk node perception is constructed. Compared with the traditional static rule verification method, this method can not only dynamically simulate the dispatching order execution process and update the topology status in real time, but also use the reinforcement learning mechanism to intelligently screen high-risk paths in the search space, identify illegal connections and blocking logic conflicts, significantly improving the comprehensiveness, real-time and intelligent level of anti-error identification, and strongly supporting the automatic verification of dispatching orders and the prediction of operational risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1A schematic diagram of the flow of a method for preventing errors in power grid scheduling based on topology analysis provided in an embodiment of the present application.

[0020] Figure 2 A schematic diagram of the structure of a power grid dispatching error prevention system based on topology analysis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Embodiment 1, Figure 1 The schematic diagram of the flow chart of the power grid dispatching error prevention method based on topology analysis provided in the embodiment of the present application includes the following steps: S1, build a remote signaling credibility assessment model based on the remote signaling assessment data to obtain highly reliable remote signaling data.

[0023] The dynamic evaluation mechanism of telesignal credibility is to ensure that each device state entering the topology identification link is logically credible and data stable. The essential goal of this step is to solve the common telesignal anomalies, state errors, loss or false signal problems in actual projects, and ensure that the device state based on which the dynamic topology map is constructed is real, credible and stable, thereby avoiding the risk of incorrect topology identification or misoperation caused by telesignal anomalies.

[0024] The remote signaling credibility evaluation model is constructed based on the remote signaling evaluation data to obtain highly reliable remote signaling data, specifically: Acquire remote signaling evaluation data and construct a remote signaling credibility evaluation model, wherein the remote signaling evaluation data includes time jitter entropy data, state mutation data, and multi-channel signal consistency rate data; The telesignaling data is obtained, and the credibility of each telesignaling point of the telesignaling data is evaluated based on the telesignaling credibility evaluation model. If the output of the telesignaling credibility evaluation model is greater than the preset credibility, the telesignaling point at this time is highly credible, and the telesignaling data with high credibility is selected.

[0025] Furthermore, the remote signaling evaluation data includes time jitter entropy data, state mutation data, and multi-channel signal consistency rate data.

[0026] Among them, the time jitter entropy data is used to measure the uncertainty of the time interval of the telesignal state change, that is, whether the change is regular. If the telesignal change interval is very random and the entropy value is high, it means that the signal stability is poor and there may be jitter or interference. By obtaining the time jitter entropy data to analyze the credibility of the telesignal data, it has the following benefits for the construction of the power grid topology model and subsequent error prevention: Finely characterize the time series characteristics of the telesignaling state: By constructing a telesignaling position change interval sequence and calculating its entropy value, the following situations can be identified: strong regularity and low entropy value: mostly caused by manual operation, normal protection / control; chaotic intervals and high entropy value: may be caused by abnormal factors such as communication jitter, electromagnetic interference, and contact burrs. This enables the topology recognition algorithm to achieve: signal credibility weighted modeling based on entropy value; dynamically determine the impact range of telesignaling anomalies, and improve the accuracy of topology recognition from the source.

[0027] As the intelligent dispatching platform evolves towards spatiotemporal coordination, state adaptation, and trusted fusion, the traditional static telesignaling judgment method is gradually exposing its limitations. The introduction of time jitter entropy provides an effective means to characterize the stability of telesignaling from the perspective of "time series statistics".

[0028] The specific method for obtaining the time jitter entropy data is as follows: Assume that the telesignal point changes position several times within the preset time window, and obtain the time interval vector by calculating the sequence of adjacent time intervals of changes; Map the time interval vector into several fixed-width intervals for binning, count the number of intervals contained in each interval, and normalize it to a probability distribution to obtain a probability density; Shannon entropy is calculated based on the probability density to obtain time jitter entropy data.

[0029] The specific calculation formula of the adjacent displacement time interval sequence is as follows:

[0030] The specific calculation formula of the probability distribution is as follows:

[0031] The specific calculation formula of the time jitter entropy data is as follows:

[0032] In the formula, is the time interval between two adjacent shifts, For the The interval between the events of the telesignal change. For the The interval between telesignal changes. For the The probability density of an interval is is the total number of telesignal changes within the preset time window, To fall into The number of time intervals in an interval, is the time jitter entropy data, is the number of bins.

[0033] Among them, the state mutation data measures the rapid mutation trend of the state of the telesignal signal within a time window, which is essentially to detect whether the telesignal signal has abnormal frequent jitter. By obtaining the state mutation data to analyze the credibility of the telesignal data, it has the following benefits for the construction of the power grid topology model and subsequent error prevention: Improve the credibility screening mechanism of telesignaling data: Through the state mutation frequency index, a dynamic telesignaling quality assessment model can be established to automatically identify telesignaling points with abnormal and frequent changes, thereby achieving: automatic elimination or downgrading of abnormal telesignaling signals; improving the robustness of status judgment of specific equipment (such as disconnectors, earthing switches, etc.); reducing the "false closing, false separation" phenomenon caused by burr data in topology identification.

[0034] Enhance the steady-state recognition capability of the topology model: Since topology modeling relies on the remote signal status to determine the status of the circuit breaker and the switch, frequent jitter signals can easily lead to pseudo-states in the modeling logic. After the introduction of state mutation data: it can assist in identifying state jumps caused by "non-real operations" and filter out "non-valid operations"; Improve the ability to intercept misjudgments: When the remote signal has glitches and misjudges the equipment to be in the open / closed state, the anti-misjudgment blocking logic may be bypassed. After introducing the state mutation indicator: set logical delay and redundant confirmation for the jitter signal; avoid the remote signal signal misleading the operation ticket blocking condition judgment, and improve the ability to prevent misoperation.

[0035] The specific method for obtaining the state mutation data is as follows: Count the number of state mutations of the device in the past several periods of time to form a time distribution function; The mutation rate is calculated based on the preset mutation rate formula by calculating the standard deviation in the distribution function and the mean of the mutation time; The absolute value of the mutation rate is compared and analyzed based on the preset maximum mutation rate to quantify the state mutation data.

[0036] The mutation rate is specifically calculated as follows:

[0037] The specific calculation formula for the state mutation data is as follows:

[0038] In the formula, is the mutation rate, is the total number of state mutations during the statistical period, is the switching function, is the mean mutation time, is the standard deviation, is the state mutation data, is the maximum mutation rate.

[0039] Among them, the multi-channel signal consistency rate data is for telesignaling points with multiple telesignaling sources or redundant acquisition (such as busbar disconnector dual-end acquisition, main and standby communication links), and calculates the time consistency rate of different telesignaling channel states. By obtaining multi-channel signal consistency rate data to analyze the credibility of telesignaling data, it has the following benefits for the construction of power grid topology model and subsequent error prevention: Enhance the robustness of telesignal source fusion: In devices with multiple telesignal sources (such as dual-side position telesignaling of disconnectors, control box + station-side concurrent telesignaling): the consistency rate indicator can quantify the timing and state consistency of different signal channels; when the consistency rate is too low, the "signal conflict" handling mechanism can be triggered to prevent the wrong topology state from being established.

[0040] Improve the redundant judgment capability of topological state modeling: Topological modeling often relies on the joint judgment of "switching device state + adjacent contact state", and the multi-channel consistency rate can be used as an auxiliary input for the credibility of the state judgment signal.

[0041] Improve the credibility of the judgment of the locking conditions: In the locking judgment of the operation ticket, if a telesignal point has multiple sources, the consistency rate can be used for: Dynamic locking acceptance selection: give priority to channel information with high consistency rate; if the consistency rate is lower than the threshold, the system can automatically determine that the telesignal is "unreliable" and trigger the anti-mistake interlocking reinforcement; avoid locking judgment errors caused by abnormalities in a certain signal channel.

[0042] The specific method for obtaining the multi-channel signal consistency rate data is as follows: Obtain the telesignal status values ​​of several acquisition channels of the telesignal point; Under the unified time axis, the multi-channel signal consistency rate data is calculated based on the preset multi-channel signal consistency rate formula.

[0043] The multi-channel signal consistency rate formula, the specific calculation formula is as follows:

[0044] In the formula, is the multi-channel signal consistency data, To detect the length of the time axis, To detect the start time of the time axis, For the The acquisition channel is in the remote signal state value, is an indicator function, which is 1 if the condition is met and 0 otherwise.

[0045] A remote signal credibility assessment model is constructed based on LSTM (long short-term memory network) according to time jitter entropy data, state mutation data and multi-channel signal consistency rate data.

[0046] The specific calculation formula of the remote signal credibility evaluation model is as follows:

[0047] In the formula, is the remote signal credibility, is the multi-channel signal consistency data, is the state mutation data, is the time jitter entropy data, , , are weights respectively.

[0048] S2, based on high-reliability telesignaling data and multi-source data fusion, performs sequential logic verification and generates a power grid topology model.

[0049] In this embodiment, by performing logic verification on the timing dependency between multiple signals, signal anomalies (such as over-tripping and remote signal jitter) can be identified, which helps to build a dynamic evolution topology model and realize state-driven structural modeling. In addition, through timing logic analysis, it is possible to: verify the logic consistency between multiple types of remote signals (such as closed position, open position, abnormality, and remote control flag) of the same device; combine SOE and recording to verify the integrity of the timing chain of "action → remote signal → protection response"; and provide high-precision event confirmation support for anti-mislocking and auxiliary decision-making systems.

[0050] The high-reliability telesignaling data and multi-source data fusion are used to perform sequential logic verification to generate a power grid topology model. The multi-source data includes SOE and waveform data, specifically: Obtaining power grid equipment data, and parsing the power grid equipment data to obtain primary equipment in the power grid, the power grid equipment data including a GIS map, a system map and a real-time database, the primary equipment including switchgear, busbar, disconnector, voltage transformer, and current transformer; Based on the modeling method of edges and nodes in the graph theory method, an initial structure diagram is established for the primary equipment based on the power connection relationship and physical location, and the modeling method includes node-branch or node-equipment-node; Perform sequential logic verification on high-reliability telesignaling data based on SOE and waveform data; According to the mapping relationship between the verified telesignal data and the primary equipment, the telesignal data is parsed into the equipment operation status; The initial structure diagram is dynamically updated based on the graph traversal algorithm according to the equipment operation status, and the updated initial structure diagram is subjected to connected component analysis to identify the main connection form of the substation and generate a power grid topology model. The main connection form includes single bus, double bus, and double bus with bypass.

[0051] Furthermore, in the process of building the power grid topology model, the change event of the telesignaling state is compared with the physical event records in the auxiliary data sources such as SOE (sequential event record data), wave recording, PMU, etc., which plays a key role in judging the authenticity of the telesignaling data and improving the reliability of the state, so as to verify whether the telesignaling state truly reflects the changes in the equipment operation or physical state. The power grid topology modeling is highly dependent on the telesignaling state, such as whether the circuit breaker is closed or the knife switch is disconnected; if there is a false alarm in the telesignaling, it will directly lead to topological structure misjudgment problems such as "the circuit breaker is actually disconnected but modeled as closed". This step can effectively identify the abnormality of the telesignaling, eliminate unreliable signals, and improve the correctness of the topology model.

[0052] The timing logic verification of the high-reliability remote signaling data based on SOE and recorded wave data is specifically as follows: Obtaining the telesignaling change time and telesignaling state change point in the telesignaling data; The remote signal change time is matched and compared with the event sequence registered in the SOE. If the time difference is within the preset standard time difference and the event type matches, it is determined that the remote signal change is triggered by a real operation. The event type includes the device type and the operation type. The operation type includes closing, tripping and remote control, etc.; The telesignal state change point is mapped to the digital quantity position change track in the wave recording data, and the waveform slope mutation point within the preset time window is extracted to determine whether there is a corresponding position change. If there is a corresponding position change, it means that there is a digital quantity jump event in the wave recording, which supports that the telesignal position change is a real physical position change. After obtaining the verification, the telesignal change operation is real and there is telesignal data corresponding to the change position.

[0053] The specific calculation formula of the waveform slope mutation point is as follows:

[0054] If in the window Memory in: Make , indicating that there is a corresponding strain position; In the formula, is the waveform slope mutation factor, is the time series value of the analog quantity in the recording system, is the detected analog mutation time, is the timing matching tolerance window, is the remote signal change time, is the slope mutation threshold.

[0055] S3, based on spectral clustering and state-driven edge weight method, the power grid topology model is partitioned and the boundary equivalence processing is performed to obtain several topology modules.

[0056] Fixed topology or fixed edge weight only considers physical connection relationships, ignores changes in device status, and cannot reflect changes in device status, resulting in disconnected devices and locked devices being mistakenly placed in the same area when partitioning, and unable to capture logical disconnections or misconnections caused by status. By introducing state-driven edge weights, edge weights can dynamically reflect the current operating status (such as whether it is on or locked).

[0057] The grid topology model is partitioned based on spectral clustering and state-driven edge weight method, specifically: Construct a weighted adjacency matrix and dynamically adjust the weighted adjacency matrix according to the operation status of the equipment, wherein the weighted adjacency matrix includes state-aware edge weights, logic linkage strength weights, and historical misoperation statistics; Construct a degree matrix based on the weighted adjacency matrix, and construct a normalized Laplace matrix in combination with the weighted adjacency matrix; The first several eigenvectors of the Laplace matrix are obtained to form a spectral embedding space, and spectral clustering is used to embed and map the power grid topology based on the weighted adjacency matrix. After the nodes are projected into the spectral embedding space, k-Means is introduced to cluster the nodes. The partition category to which each node belongs is obtained and the boundary of the power grid topology model is divided to obtain several subgraphs.

[0058] It should be noted that the state-aware edge weight reflects the real-time availability and physical state of the connection, avoiding misjudgment caused by false trips / refusals to operate of remote signals; the logical linkage strength weight indicates the strength of the logical operational coupling or protection linkage relationship between nodes, making the topological division more in line with scheduling cognition; the historical misoperation statistics indicate the frequency / impact of abnormalities or accidents caused by misoperation between equipment pairs in the process of scheduling and operation and maintenance in history, which can be used as a reference for adjusting the division boundaries, thereby enhancing the division's ability to identify high-risk areas.

[0059] The weighted adjacency matrix is ​​specifically:

[0060] The degree matrix is ​​specifically:

[0061] The Laplace matrix is ​​specifically:

[0062] In the formula, is the weighted adjacency matrix, For equipment and equipment The edge weights between For equipment Total strength with other equipment, is the total number of devices, For equipment and equipment The connection edge weights, is the Laplace matrix, is the identity matrix, is the inverse square root of the degree matrix, is the weighted adjacency matrix Normalization of .

[0063] It should be noted that the weighted adjacency matrix is ​​an n×n matrix, which is used to represent the strength of the connection between devices in the power grid topology; the equipment operation status includes two devices being closed at the same time, two devices being one open and one open, or when a lock exists, both devices are disconnected; the logical linkage strength between combined devices refers to the node device being a combined device (such as a knife switch-circuit breaker, grounding switch-isolating switch), and a high coupling value is assigned.

[0064] Furthermore, the boundary equivalence method is used on several subgraphs obtained after the partitioning to ensure that the error prevention analysis of the entire network after the topological partitioning is still complete. After the automatic partitioning of the power grid and the boundary equivalence, a small-scale calculation model is formed, and each partition is calculated independently, which reduces the overall calculation scale, improves the calculation efficiency, and meets the real-time error prevention needs.

[0065] The boundary equalization processing is performed to obtain several topological modules, specifically: Traversing the edges of the plurality of subgraphs to identify boundary nodes and cross-region edges; Classify the cross-zone edges according to the switch status, and retain the cross-zone edges that need to be processed with impedance equivalence. The classification processing includes that when the switch is closed, impedance equivalence processing is required and its electrical connection effect is retained; when the switch is open, the connection is ignored and its electrical effect is not modeled in the module; Based on the reserved cross-zone edges, the operation data of the cross-zone edges are obtained, wherein the operation data includes real-time equivalent impedance, voltage amplitude and phase of boundary nodes, and power flow of boundary branches; Introducing equivalent nodes into several subgraphs, and setting initial voltage values ​​of the equivalent nodes according to the operation data, wherein the voltage values ​​are obtained by comparing the voltage amplitudes of the boundary nodes in the operation data with the preset voltages; Connect the boundary nodes and equivalent nodes in the subgraph through impedance elements to form a virtual branch; Real-time detection of boundary line status, acquisition of voltage value of each boundary node, and dynamic correction of initial voltage value of equivalent node; Each subgraph after boundary equalization processing is stored as a topological module, and voltage and power injection simulation is performed based on virtual branches and boundary nodes, and comparative verification is performed to obtain several topological modules.

[0066] It should be noted that the comparison verification is performed by performing voltage and power injection simulation to compare the global model results before modularization to ensure that the relative errors of voltage and power meet the preset thresholds.

[0067] S4, based on the object-oriented parsing results of the scheduling instructions, simulate the scheduling operations on several topological modules, and traverse the path based on the depth-first search method, and combine the reinforcement learning method to perform error-proof analysis on the traversed path, specifically: The dispatching instructions are compiled based on the graphic and textual dispatching instructions. The graphic dispatching instructions are based on the whole network diagram data. Users can quickly generate dispatching instructions by graphical clicking, which are suitable for single device operation, main transformer state switching, reverse mother operation, side task and other types. The textual dispatching instructions are based on natural language processing to identify the tasks of the text input dispatching instructions, and map the identified tasks to the specified graphics, so as to facilitate the subsequent object-oriented steps based on the text. The dispatch instructions are analyzed by objectification, decomposed into standardized digital tasks, and further analyzed into steps with a single device operation as the smallest granularity based on error prevention rules, specifically: Objectification of switch state conversion: the device state conversion of the switch is parsed into the device state switching sequence of the switch, the knife switches on both sides and the related ground switches; objectification of busbar state conversion: the state conversion of the bus is parsed into the state switching sequence of the busbar and its affiliated equipment; objectification of main transformer state conversion: the state conversion of the main transformer is parsed into the state switching sequence of the main transformer and its affiliated equipment; objectification of line state conversion: the line state conversion is parsed into the device state conversion sequence of the line local and the related ground switches; objectification of bypass state conversion: divided into two types: bypass line and bypass switch. The bypass task state is parsed into the state switching sequence of the switch, knife switch and 4-knife switch bypassed to the specified equipment (switches, lines); objectification of reverse bus conversion: intelligently distinguish between hot reverse and cold reverse, and objectify the task into the operation sequence of the related knife switches and bus coupling intervals according to the bus conduction rules; objectification of reverse main transformer: the main transformer is subjected to the state conversion sequence of the three-side switches and the bus couplings and segmented intervals on each side according to the state conversion principle; The dispatching instructions after objectification are matched with corresponding topological modules in several topological modules based on the device connection relationship, and simulated binding is performed. Each objectified minimum granularity step is simulated and executed in turn to update the topological model in real time. An operation graph structure based on the updated topology model, wherein the operation graph structure uses device nodes in the topology model as graph vertices and conduction relationships as edges; The conduction paths of device nodes in the operation graph structure are traversed by the depth-first search method, and the conduction paths are analyzed to prevent errors based on the reinforcement learning method.

[0068] It should be noted that the device is changed from the disconnected state to the on state due to a certain operation in the dispatching order (such as closing the switch or closing the knife switch). The state change of these devices will directly affect the connection path in the electrical topology and is the object that must be focused on in the error prevention logic judgment.

[0069] Furthermore, the traditional DFS (depth-first search algorithm) exhaustively explores all paths, and its path selection is not intelligent. In large-scale topologies, DFS will generate a large number of redundant paths, resulting in low efficiency and lack of focus. By introducing reinforcement learning to provide a strategy selection mechanism, a high-risk verification path is intelligently selected from multiple possible scheduling paths to improve the efficiency and quality of error-prevention analysis, and assist dispatchers in quickly determining whether instructions have potential risks.

[0070] The conduction paths of the device nodes in the operation graph structure are traversed by the depth-first search algorithm, and error-proof analysis of the conduction paths is performed based on the reinforcement learning method, specifically: Acquire device status data at the starting point of the depth-first search algorithm as a state in reinforcement learning, wherein the device status data includes operation authority, locking state, and adjacent node state; All the scheduling instruction sets that can be executed at the current starting point are used as actions in reinforcement learning, wherein all the scheduling instruction sets include closing the knife switch, pulling the knife switch, throwing protection, and exiting the remote control lock; The reinforcement learning strategy Agent is trained based on the probability of selecting the corresponding action in the current state; Construct a path screening reward function, which includes that all operations in the path comply with the anti-error logic, there are high-risk erroneous operations in the path (such as closing the switch under the condition of locking failure), the remote signal credibility of the equipment in the path is abnormal, and the operation violates the anti-error locking logic; Construct a reinforcement learning state-action model based on the state and action reinforcement learning strategy Agent and the path screening reward function; Based on the reinforcement learning state-action model, the path screening value is predicted for all the conductive paths traversed by the depth-first search algorithm; Based on the path screening value, a path judgment is performed. When the path judgment shows that the path screening value is lower than a preset path screening threshold, the action of the corresponding scheduling instruction is marked as a potential error prevention risk.

[0071] The specific calculation formula of the path screening reward function is as follows:

[0072] In the formula, Filter values ​​for paths, To comply with the error-proof logic reward value, is the penalty value for high-risk misoperation. is the device remote signal abnormal penalty value, To violate the penalty value of the anti-error locking logic, To comply with the error-proof logic reward value coefficient, is the penalty coefficient for high-risk misoperation, is the equipment remote signal abnormal value penalty coefficient, It is the penalty coefficient for violating the anti-false locking logic.

[0073] in, .

[0074] It should be noted that the operation graph structure needs to be encoded in vector form, such as adjacency matrix embedding encoding, the operation sequence can be encoded by RNN, and the state is combined into a high-dimensional feature vector for RL network input. Reinforcement learning agent refers to an intelligent agent that makes decisions, interacts with the environment, and continuously optimizes its strategy based on reward feedback in the reinforcement learning framework. Its core task is to learn a strategy that maximizes the long-term rewards accumulated in a given environment.

[0075] Embodiment 2, Figure 2 The schematic diagram of the structure of the power grid dispatching error prevention system based on topology analysis provided in the embodiment of the present application includes a remote signaling credibility evaluation module, a topology model construction module, a topology model processing module and an error prevention analysis module, and there are connections between the modules: The remote signaling credibility evaluation module is used to build a remote signaling credibility evaluation model based on the remote signaling evaluation data to obtain highly reliable remote signaling data; Topology model building module, used to perform sequential logic verification based on high-reliability telesignaling data and multi-source data fusion to generate power grid topology model; A topology model processing module is used to partition the power grid topology model based on spectral clustering and state-driven edge weight method, and perform boundary equivalence processing to obtain several topology modules; The error-prevention analysis module is used to simulate the scheduling operation of several topological modules based on the object-oriented analysis results of the scheduling instructions, and to traverse the path based on the depth-first search method, and to conduct error-prevention analysis on the traversed path in combination with the reinforcement learning method.

[0076] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0077] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0078] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0079] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0080] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0081] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for preventing power grid dispatch errors based on topology analysis, characterized in that: The steps include: Construct a remote signaling credibility assessment model based on remote signaling assessment data to obtain highly reliable remote signaling data; Based on high-reliability telesignaling data and multi-source data fusion, sequential logic verification is performed to generate a power grid topology model; Based on spectral clustering and state-driven edge weight method, the power grid topology model is partitioned and boundary equivalence processing is performed to obtain several topological modules. Based on the object-oriented parsing results of scheduling instructions, several topological modules are simulated for scheduling operations, and the path traversal is performed based on the depth-first search method. The traversed path is then analyzed for error prevention using the reinforcement learning method.

2. The method for preventing power grid dispatch errors based on topology analysis according to claim 1, characterized in that: The high-reliability remote signaling data and multi-source data fusion are used to perform sequential logic verification and generate a power grid topology model, specifically: Obtain primary equipment and establish an initial structure diagram based on power connection relationship and physical location; Perform sequential logic verification on high-reliability telesignaling data based on SOE and waveform data; According to the verified telesignaling data, the equipment operation status is obtained by parsing the telesignaling data based on the mapping relationship between the verified telesignaling data and the primary equipment; The initial structure diagram is dynamically updated according to the equipment operation status and graph traversal algorithm, and the updated initial structure diagram is subjected to connected component analysis to identify the main wiring form of the substation and generate a power grid topology model.

3. The method for preventing power grid dispatch errors based on topology analysis according to claim 2 is characterized in that: The high-reliability telesignal data is subjected to sequential logic validation based on SOE and wave recording data, including: judging the authenticity of the telesignal change operation and judging the change of the telesignal state change point; The authenticity of the remote signal change operation is determined by comparing the remote signal change time in the remote signal data with the event sequence registered in the SOE; The change point of the telesignal state corresponds to the change position judgment, which is obtained by obtaining the digital change position trajectory corresponding to the telesignal state change point in the telesignal data and the recorded data, and extracting the waveform slope mutation point within the preset time window.

4. The method for preventing power grid dispatch errors based on topology analysis according to claim 1, characterized in that: The grid topology model is partitioned based on spectral clustering and state-driven edge weight method, specifically: Based on a normalized Laplace matrix, the first several eigenvectors of the Laplace matrix are obtained to form a spectral embedding space, wherein the normalized Laplace matrix is ​​constructed based on a weighted adjacency matrix, and the weighted adjacency matrix includes state-aware edge weights, logic linkage strength weights, and historical misoperation statistics; Spectral clustering is used to embed and map the power grid topology based on the weighted adjacency matrix. After the nodes are projected into the spectral embedding space, k-Means is introduced to cluster the nodes. The partition category after clustering each node is obtained and the boundary of the power grid topology model is divided to obtain several subgraphs.

5. The method for preventing power grid dispatch errors based on topology analysis according to claim 4 is characterized in that: The boundary equalization processing is performed to obtain several topological modules, specifically: Traverse the edges of several subgraphs to identify boundary nodes and cross-region edges; Classify the cross-zone edges according to the switch status, retain the cross-zone edges that need impedance equalization processing, and obtain the operation data of the cross-zone edges; Introduce equivalent nodes into several subgraphs, and set the initial voltage values ​​of the equivalent nodes according to the operating data; Connect the boundary nodes and equivalent nodes in the subgraph through impedance elements to form a virtual branch; Detect the boundary line status, obtain the voltage value of each boundary node in real time, and dynamically correct the initial voltage value of the equivalent node; Each subgraph after boundary equalization processing is stored as a topological module, and voltage and power injection simulation is performed based on virtual branches and boundary nodes, and comparative verification is performed to obtain several topological modules.

6. The method for preventing power grid dispatch errors based on topology analysis according to claim 1, characterized in that: The method simulates the scheduling operation of several topological modules based on the object-oriented parsing result of the scheduling instruction, performs path traversal based on the depth-first search method, and performs error-proof analysis on the traversed path in combination with the reinforcement learning method, specifically: The dispatch order is compiled based on the graphic and textual proposed dispatch orders to obtain the dispatch instruction; Perform object-oriented analysis on dispatch instructions, decompose them into standardized digital tasks, and parse them into steps with a single device operation as the smallest granularity based on error prevention rules; The objectified dispatching instructions are simulated and bound to the corresponding topology modules, and simulated and executed based on the smallest granularity steps to update the topology model in real time. Obtaining an operation graph structure based on the updated topology model, wherein the operation graph structure is formed by connecting various device nodes and logical conduction relationships in the topology model; The conduction paths of device nodes in the operation graph structure are traversed by the depth-first search method, and the conduction paths are analyzed to prevent errors based on the reinforcement learning method.

7. The method for preventing power grid dispatch errors based on topology analysis according to claim 6, characterized in that: The conduction paths of the device nodes in the operation graph structure are traversed by the depth-first search algorithm, and error-proof analysis of the conduction paths is performed based on the reinforcement learning method, specifically: Obtain the device state data at the starting point of the depth-first search algorithm as the state in reinforcement learning; Take all the scheduling instructions that can be executed at the current starting point as actions in reinforcement learning and train the reinforcement learning strategy Agent; Construct a path screening reward function, and combine the state, action, and reinforcement learning strategy Agent to build a reinforcement learning state-action model; Through all the conductive paths traversed by the depth-first search algorithm, the path screening value is predicted based on the reinforcement learning state-action model, the path judgment is performed, and the potential error prevention risks are marked.

8. The method for preventing power grid dispatch errors based on topology analysis according to claim 2, characterized in that: The remote signaling evaluation data includes time jitter entropy data, and the specific acquisition method is as follows: Assume that the telesignal point changes position several times within the preset time window, and obtain the time interval vector by calculating the sequence of adjacent time intervals of changes; Map the time interval vector into several fixed-width intervals for binning, count the number of intervals contained in each interval, and normalize it to a probability distribution to obtain a probability density; Shannon entropy is calculated based on the probability density to obtain time jitter entropy data.

9. A system using the power grid dispatch error prevention method based on topology analysis as claimed in any one of claims 1 to 8, characterized in that: It includes remote signal credibility assessment module, topology model construction module, topology model processing module and error prevention analysis module. There are connections between the modules: The remote signaling credibility evaluation module is used to build a remote signaling credibility evaluation model based on the remote signaling evaluation data to obtain highly reliable remote signaling data; Topology model building module, used to perform sequential logic verification based on high-reliability telesignaling data and multi-source data fusion to generate power grid topology model; A topology model processing module is used to partition the power grid topology model based on spectral clustering and state-driven edge weight method, and perform boundary equivalence processing to obtain several topology modules; The error-prevention analysis module is used to simulate the scheduling operation of several topological modules based on the object-oriented analysis results of the scheduling instructions, and to traverse the path based on the depth-first search method, and to conduct error-prevention analysis on the traversed path in combination with the reinforcement learning method.

Images