Multi-dimensional data collaborative analysis and early warning method based on smart operation and maintenance of power transmission network

Abstract

The application discloses a multi-dimensional data collaborative analysis and early warning method based on smart operation and maintenance of a power transmission network, relates to the technical field of power transmission network early warning, and comprises the following steps: collecting target data of a target device, and simultaneously acquiring mutation data of power load in a system; in combination with the mutation data of the power load, the instantaneous characteristics of electrical quantities and the dynamic changes of the power grid structure are correlated and analyzed to obtain correlated factors that can predict faults; and according to the mutation amplitude of the power load, the mutual influence mode between the correlated factors is determined. The application solves the problem of inaccurate and untimely fault early warning caused by insufficient data collaboration and delayed early warning response in the operation and maintenance of the power transmission network, and achieves the effect of improving the early warning precision and response speed of the smart operation and maintenance of the power transmission network.

Description

Technical Field

[0001] This invention belongs to the field of early warning technology for power transmission networks, specifically a method for early warning based on multi-dimensional data collaborative analysis of intelligent operation and maintenance of power transmission networks. Background Technology

[0002] In practical applications, the multi-dimensional data collaborative analysis and early warning method based on intelligent operation and maintenance of power transmission networks often faces the problem of insufficient data collaborative application.

[0003] Taking fault early warning in the daily operation and maintenance of power transmission networks as an example, after the system collects data on target equipment and power load mutations, it fails to fully integrate the power load mutation data into the correlation analysis between the instantaneous characteristics of electrical quantities and the dynamic changes in the power grid structure. This makes it difficult to accurately obtain the correlation factors that can predict faults. Furthermore, when determining the mutual influence of each correlation factor, the lack of targeted analysis based on the magnitude of power load mutations results in inaccurate analysis results. In addition, the invocation of early warning model combinations does not fully consider the real-time values ​​and influence methods of correlation factors, information exchange between models is not timely, and when the deviation exceeds the set value, the adjustment of the data acquisition frequency and load control parameters fails to effectively combine with the model judgment results, leading to insufficient real-time performance and effectiveness of the early warning scheme.

[0004] This situation not only affects the efficiency of power grid operation and maintenance, but may also pose potential risks to the stable operation of the power grid due to untimely or inaccurate early warnings, especially when the power grid load fluctuates greatly. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional data collaborative analysis and early warning method based on intelligent operation and maintenance of power transmission networks, which solves the technical problems of inaccurate and untimely fault warnings caused by insufficient data collaboration and delayed early warning response in the operation and maintenance of power transmission networks.

[0006] A multi-dimensional data collaborative analysis and early warning method based on intelligent operation and maintenance of power transmission networks includes:

[0007] Collect target data from the target device, and simultaneously acquire data on sudden changes in the power load of the system;

[0008] By combining the sudden change data of power load, we can conduct correlation analysis on the instantaneous characteristics of electrical quantities and the dynamic changes of power grid structure to obtain the correlation factors that can predict faults.

[0009] Based on the magnitude of sudden changes in power load, determine the interaction patterns among various related factors;

[0010] Based on the real-time values ​​and impact patterns of related factors, the corresponding combination of early warning models is invoked, and the models exchange judgment information in a timely manner as the factors change.

[0011] When the deviation exceeds the set value, the system takes the linkage response of data acquisition frequency and load control parameters as the core, combines the judgment results of various models, dynamically adjusts the adaptation relationship between the two, and generates a real-time early warning scheme.

[0012] Furthermore, a correlation analysis is conducted between the instantaneous characteristics of electrical quantities and the dynamic changes in the power grid structure, including the following steps:

[0013] S1: Convert the instantaneous characteristics of electrical quantities into corresponding energy signals, and then map them to the relevant nodes of the power grid structure;

[0014] S2: After receiving this signal, the node generates an initial state change;

[0015] S3: Track the chain transmission path formed by this state change as the power grid structure changes. Each node in the path will provide real-time feedback on the effect of its own state change on the upstream signal.

[0016] S4: Apply the effect in reverse to the original corresponding node to form the first round of interaction between features and structure;

[0017] S5: Use the result of this loop as a new starting point and repeat steps S1 to S4 above;

[0018] S6: Obtain the cumulative deviation of node status in multiple cycles, extract the related factors, and introduce the real-time load fluctuation of the power grid as a dynamic adjustment parameter in each cycle, so that the factor extraction process is consistent with the actual operation of the power grid.

[0019] Furthermore, when determining the ways in which the various related factors interact:

[0020] Based on the correspondence between electrical quantity characteristics and power grid structural nodes, an initial causal relationship sequence is established according to the order in which the factors appear.

[0021] Based on the obtained power load change data, the effectiveness of the triggering of each link in the sequence is verified by real-time data. Invalid links are replaced and supplemented. During the supplementation process, relevant factors that may affect the information exchange of subsequent model judgments need to be included.

[0022] Based on the extraction of the related factors, the response delay differences and amplitude fluctuation ranges of other related factors under different power grid load states after a change in a single related factor are recorded, and the effect of the recorded results on the subsequent deviation judgment process is predicted simultaneously.

[0023] Based on the above records, a dynamic, updatable table of multi-scenario impact relationships is created, and the contents of the table are continuously corrected as new data is added.

[0024] Furthermore, when collecting target data, this includes:

[0025] First, the filtering threshold is dynamically adjusted based on the real-time load fluctuation of the power grid, and then the algorithm is used to distinguish between interference information and effective features in the data.

[0026] The filtered valid data is bound to key node parameters of power grid structure changes and time-series marked, so that when the archived data enters the correlation analysis stage, the corresponding factor transmission path can be automatically activated according to the marking.

[0027] Furthermore, when distinguishing between interference information and valid features using dynamic filtering algorithms:

[0028] First, capture the reference signal generated by the current load fluctuation of the power grid. Then, perform differential analysis between the electrical quantity data and the reference signal to extract abnormal segments whose difference exceeds the dynamically generated threshold.

[0029] For abnormal segments, the source is traced by combining the real-time trajectory of changes in the power grid structure. If it can be linked to the state change of a specific node, it is marked as a valid feature; otherwise, it is judged as interference information.

[0030] Furthermore, when invoking a combination of early warning models, the following are included:

[0031] First, the mutual influence intensity of the related factors is converted into a dynamic adjustment coefficient. The higher the influence intensity, the larger the coefficient value.

[0032] The computing power weights are assigned to the corresponding models according to the coefficient values, and after each round of analysis, the coefficient values ​​are reverse-calibrated based on the newly generated influence relationship table, so that the computing power allocation can adapt to the correlation patterns of factors in real time.

[0033] Furthermore, the judgment information shared between models includes factor feature matching results, trend change predictions, and risk probability assessments:

[0034] When the real-time value of the related factors changes by more than a set proportion, the relevant model will send updated judgment information synchronously, and the receiving model will cross-check the information with its own analysis results.

[0035] Content that matches the verification results will be used as the basis for a comprehensive judgment; content that does not match will be subject to secondary analysis.

[0036] Furthermore, when the deviation exceeds the set value, it also includes:

[0037] Taking the linkage response of data acquisition frequency and load control parameters as the core, the judgment results of each model are summarized to obtain a consistent conclusion about the cause of the deviation.

[0038] Based on the conclusions, adjust the relationship between the data acquisition frequency and the load control parameters. The greater the load control range, the higher the acquisition frequency should be.

[0039] Based on the adjusted coordination relationship and the results of the model supplementary analysis, a real-time early warning plan is generated, which includes risk level, scope of impact and control recommendations.

[0040] Furthermore, when the power grid experiences a sudden load surge, it also includes:

[0041] The routine factor correlation analysis process was suspended, and the characteristics of electrical quantity abrupt changes before the impact were extracted;

[0042] The above features are quickly compared with typical features in the historical impact case database, and reference solutions are generated after being sorted by matching degree.

[0043] Meanwhile, each model performs parallel calculations on the current power grid structure's tolerance capabilities, and combines the calculation results with the reference scheme to form a temporary control strategy.

[0044] After the impact subsides, routine analysis will be conducted and the temporary strategy will be calibrated.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] This invention combines power load mutation data for multidimensional correlation analysis, determines the mutual influence of related factors based on the mutation magnitude, calls early warning model combinations and realizes information exchange between models, and generates early warning schemes by dynamically adjusting the adaptation relationship between data acquisition frequency and load control parameters. This solves the problems of inaccurate and untimely fault early warning caused by insufficient data coordination and delayed early warning response in power transmission network operation and maintenance, and achieves the effect of improving the accuracy and response speed of intelligent operation and maintenance early warning in power transmission networks. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the method framework structure of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see Figure 1 This application provides a multi-dimensional data collaborative analysis and early warning method based on intelligent operation and maintenance of power transmission networks, including:

[0050] Collect target data from the target device, and simultaneously acquire data on sudden changes in the power load of the system;

[0051] It should be noted that target data refers to various types of data related to the operating status of target equipment in the power transmission network. It can be obtained through real-time acquisition by sensors, uploading by the equipment's built-in monitoring module, or manual periodic entry. It is mainly used to reflect the basic operating status of the target equipment and provide basic information for subsequent analysis.

[0052] By combining the sudden change data of power load, we can conduct correlation analysis on the instantaneous characteristics of electrical quantities and the dynamic changes of power grid structure to obtain the correlation factors that can predict faults. It should be noted that the sudden change data of power load is the data that indicates that the load in the power system has changed significantly in a short period of time. It can be achieved by real-time recording by the power monitoring system, collection by the load management terminal, or by comparing and identifying historical load data through the data analysis platform.

[0053] Furthermore, predictable fault correlation factors refer to various factors that are potentially related to power grid faults and are obtained through correlation analysis. These factors can be identified through methods such as big data association rule mining, machine learning classification algorithms, or expert experience screening. For example, the correspondence between instantaneous fluctuations in electrical quantities and weak links in the power grid structure is mainly used to detect potential hazards that may lead to faults in advance and provide a basis for early warning.

[0054] Based on the magnitude of sudden changes in power load, determine the mutual influence mode between various related factors. It should be understood that the mutual influence mode between related factors refers to the relationship between different related factors in the process of affecting the operation of the power grid. It can be realized by means of causal relationship analysis, correlation calculation or system dynamics model, etc. For example, a positive correlation relationship in which the enhancement of one related factor will lead to a significant change in another related factor. It is mainly used to clarify the mechanism of action between various factors and improve the accuracy of early warning analysis.

[0055] Based on the real-time values ​​and influence patterns of the related factors, the corresponding early warning model combination is invoked, and the models exchange judgment information in a timely manner as the factors change. It should be noted that the early warning model combination refers to a collection of multiple models selected to achieve early warning of transmission network faults, such as fault prediction models based on neural networks, risk assessment models based on decision trees, etc., which correspond to early warning strategies (such as early warning thresholds, response levels, and processing procedures) under different combinations of related factors.

[0056] When the deviation exceeds the set value, the system dynamically adjusts the adaptation relationship between the data acquisition frequency and load control parameters, based on the judgment results of various models, and generates a real-time early warning scheme, with the data acquisition frequency and load control parameters acting as the core. For example, the linkage response between the data acquisition frequency and load control parameters refers to the process of mutual coordination and synergy between the data acquisition time interval and various parameters used to regulate the power load. This can be achieved using preset linkage rules, dynamic response algorithms, or intelligent control strategies.

[0057] Specifically, this invention combines power load mutation data for multidimensional correlation analysis, determines the mutual influence of related factors based on the mutation magnitude, calls early warning model combinations and achieves information exchange between models, and generates early warning schemes by dynamically adjusting the data acquisition frequency and load control parameters. This solves the problems of inaccurate and untimely fault early warning caused by insufficient data coordination and delayed early warning response in power transmission network operation and maintenance, and achieves the effect of improving the accuracy and response speed of intelligent operation and maintenance early warning in power transmission networks.

[0058] It should be understood that, in some examples of this embodiment, the intelligent operation and maintenance system of the power transmission network may include one or more data acquisition devices, such as smart meters for collecting electrical quantities, and a data processing center, such as a high-performance server cluster.

[0059] Specifically, the data processing center receives target data and power load mutation data from data acquisition devices, analyzes correlation factors, and determines mutual influence patterns. The data processing center internally stores preset early warning model combinations and processing strategies corresponding to different correlation factors. When a real-time numerical deviation of a correlation factor exceeds a set value, the data processing center invokes the corresponding early warning model combination. The models exchange judgment information in real time and dynamically adjust the data acquisition frequency and load control parameters based on their judgment results. The system generates a real-time early warning plan based on the adjusted adaptation. For example, when the load mutation amplitude is large and the correlation factor indicates high risk, the data acquisition frequency is increased to the highest level, and load control parameters are adjusted to limit the load growth rate. The data processing center sends the generated early warning plan to the operation and maintenance terminal, reminding staff to take appropriate measures. During intelligent operation and maintenance of the power transmission network, through multi-dimensional data collaborative analysis and dynamic parameter adjustment, timely and accurate fault warnings can be issued. This solves the problem of poor early warning effects caused by isolated data and rigid response in existing technologies, reducing the probability of power transmission network faults. It has improved the stability and reliability of power transmission network operation, reduced operation and maintenance costs, and enhanced the level of intelligent management of the power grid.

[0060] In one embodiment, a correlation analysis is performed between the instantaneous characteristics of electrical quantities and the dynamic changes in the power grid structure, including the following steps:

[0061] S1: Convert the instantaneous characteristics of electrical quantities into corresponding energy signals, and then map them to the relevant nodes of the power grid structure. It should be noted that the energy signal refers to the quantifiable energy representation obtained by mapping the instantaneous characteristics (such as peak value and change rate) of electrical quantities such as voltage and current through energy conversion algorithms (such as Fourier transform and wavelet analysis). It can be converted through the signal processing module of the power grid monitoring terminal.

[0062] S2: After receiving this signal, the node generates an initial state change;

[0063] S3: Track the chain transmission path formed by the change of this state as the power grid structure changes. Each node in the path will provide real-time feedback on the effect of its own state change on the upstream signal. It should be noted that the chain transmission path refers to the path formed by the sequential transmission of the state change of nodes in the power grid structure according to the topological connection relationship. It can be dynamically tracked and determined based on the power grid GIS topology map and real-time status monitoring data.

[0064] S4: Apply the effect in reverse to the original corresponding node to form the first round of interaction between features and structure;

[0065] S5: Use the result of this loop as a new starting point and repeat steps S1 to S4 above;

[0066] S6: Obtain the cumulative deviation of node states across multiple cycles, extract related factors, and incorporate real-time grid load fluctuations as dynamic adjustment parameters in each cycle to ensure the factor extraction process closely aligns with the actual operation of the power grid. It should be noted that the cumulative deviation refers to the sum of the deviations between the actual state and the theoretical stable state of each node across multiple cycles, which can be obtained by continuously recording node state data and calculating the deviation for each cycle. Dynamic adjustment parameters refer to quantitative indicators reflecting the degree of real-time load fluctuations in the power grid (such as load volatility and load mutation amplitude), which can be collected in real-time by the load management system and transmitted to the correlation analysis module.

[0067] For example, in the prior art, if only electrical quantity characteristics or power grid structure are analyzed separately, it is difficult to capture the potential fault correlation under the dynamic interaction of the two. Changes in electrical quantity may be caused by adjustments to the power grid structure, and changes in the structural state may also have a reaction effect on electrical quantity characteristics. Single-dimensional analysis is prone to ignoring this mutual influence.

[0068] This embodiment constructs a cyclical analysis method that interacts with the structure by converting the instantaneous characteristics of electrical quantities into energy signals and associating them with grid nodes. Furthermore, by introducing energy signal conversion, abstract electrical quantity characteristics can be transformed into input quantities that grid nodes can perceive, thus establishing a bridge between the two.

[0069] Furthermore, tracing the chain-like transmission path can clearly present the transmission process of state changes, avoiding the omission of correlations caused by partial analysis; finally, through multiple rounds of interactive cycles, the cumulative effect of subtle correlations can be amplified, making the originally hidden fault correlation factors gradually emerge.

[0070] Furthermore, by introducing real-time load fluctuations as an adjustment parameter in each cycle, the analysis process consistently aligns with the actual operating conditions of the power grid. Specifically, load fluctuations are a core variable affecting the relationship between electrical quantities and the power grid structure; ignoring their impact would lead to unrealistic correlation analysis. This multi-cycle approach, linking characteristics to structure and then to load, allows the extracted correlation factors to more closely reflect the true causes of faults, resolving the distortion of correlation factors caused by static or single-dimensional analysis, and providing a more reliable foundation for subsequent early warning.

[0071] For ease of understanding, the specific implementation method provided in this embodiment is as follows: Assuming the target device is an outgoing switch of a 110kV substation, it is necessary to analyze the related factors of its short-circuit fault. When a peak current (instantaneous electrical quantity characteristic) occurs at the switch, the system first converts the peak current into a corresponding energy signal and maps it to the bus node where the switch is located (S1). After receiving the signal, the bus node experiences an initial state change of voltage dip (S2). This change is sequentially propagated along the outgoing topology to the line tower node and the user transformer node, forming a chain transmission path. Each node provides real-time feedback on the attenuation effect of its own impedance change on the upstream voltage signal (S3). The system superimposes these attenuation effects back onto the bus node, completing the first round of interactive loop (S4). Starting from the state of the bus node after this loop, the above conversion, transmission, and feedback process is repeated (S5). After 3 rounds of loops, the system calculates the cumulative deviation of the state of each node. For example, if the cumulative impedance deviation of the line tower node reaches 5%, a real-time load fluctuation parameter (such as a sudden 20% load increase during a certain period) is introduced into each round of loops to adjust the loop weight. Ultimately, the factors that can predict short-circuit faults were extracted as "bus voltage sag, cumulative line impedance deviation, and node response delay during load abrupt changes". These factors are consistent with the actual logic that load abrupt changes cause changes in line impedance during actual power grid operation, which in turn aggravates the impact of peak switching current.

[0072] In one embodiment, when determining the mutual influence of the various related factors:

[0073] Based on the correspondence between electrical quantity characteristics and power grid structural nodes, an initial causal relationship sequence is established according to the order in which the factors appear. It should be noted that the initial causal relationship sequence refers to a chain relationship model constructed according to the time sequence and logical association of the related factors, which can be generated through timestamp comparison and topological association algorithms.

[0074] Based on the acquired power load change data, the effectiveness of the triggering of each link in the sequence is verified by real-time data. Invalid links are replaced and supplemented. During the supplementation process, relevant factors that may affect the information exchange of subsequent model judgments need to be included. Among them, invalid links refer to causal relationship nodes that cannot be verified by actual operating data in real-time data verification. Their determination can be achieved by setting a trigger threshold (such as factor B not producing the expected fluctuation within a preset time window after factor A changes).

[0075] Based on the extraction of related factors, the response delay differences and amplitude fluctuation ranges of other related factors under different power grid load states after a change in a single related factor are recorded, and the potential impact of the recorded results on the subsequent deviation judgment process is predicted simultaneously. It should be understood that the response delay difference refers to the difference in the time interval between corresponding changes in other related factors under different load states after a change in a certain factor, which can be calculated through timestamp marking and time series analysis tools.

[0076] Based on the above records, a dynamically updatable multi-scenario impact relationship table is formed, and the contents of the table are continuously corrected as new data is added. It should be understood that the multi-scenario impact relationship table refers to a dynamic data table that integrates the interaction patterns of various factors under different load levels and power grid topologies. Its storage format can be a relational database or a time-series database, and it supports dynamic updates according to the frequency of new data collection.

[0077] For example, the present invention adds a process for verifying and dynamically correcting power load mutation data on the basis of establishing an initial causal relationship sequence.

[0078] It should be understood that simply constructing a sequence based on the order in which factors appear may not accurately reflect the true impact relationships between related factors. However, by introducing real-time verification of power load fluctuation data, the system can determine whether there are actual triggering relationships between each link in the sequence, thereby preliminarily verifying the rationality of the impact pattern.

[0079] Furthermore, by replacing and completing invalid links and recording response characteristics under different loads, the system can eliminate false or partial correlations, ensuring that the acquired mutual influence patterns conform to the actual operating logic of the power grid. This ensures that only after these steps are completed can a dynamically updatable multi-scenario influence relationship table be finally formed. This approach, combining initial sequence construction, load data verification and correction, and dynamic updates, makes the determination of the mutual influence patterns of related factors more accurate and realistic.

[0080] Specifically, the initial causal sequence is used to build the basic correlation framework, while the verification and correction of power load mutation data serves as a key calibration method to ensure that the impact pattern is consistent with the actual operation of the power grid. This solves the problem of correlation distortion that may be caused by relying solely on the order of factors, avoids subsequent early warning models from making misjudgments based on incorrect impact relationships, and improves the reliability of intelligent operation and maintenance early warning for the power transmission network.

[0081] To facilitate understanding of the above technical solution, we illustrate it with an example: Assume a 220kV transmission network where the related factors are line voltage fluctuations, transformer operating temperature, and sudden changes in regional power load. First, establish an initial sequence in order of occurrence: sudden change in regional power load - line voltage fluctuation - transformer operating temperature. Then, by examining the power load change data, it is found that when the load change amplitude is less than 5%, the link between line voltage fluctuations and transformer operating temperature is ineffective. Therefore, line current changes are added, and the sequence is adjusted to sudden change in regional power load - line voltage fluctuation - line current change - transformer operating temperature. Record the responses under different loads: Under light load (<30%), voltage fluctuation is delayed by 1.2 seconds with an amplitude of ±2%, and current change is delayed by 1.5 seconds with an amplitude of ±3%; under heavy load (>70%), voltage fluctuation is delayed by 0.8 seconds with an amplitude of ±5%, and current change is delayed by 1.0 second with an amplitude of ±6%. It is predicted that these data will affect the load adaptability of the subsequent deviation judgment threshold. Finally, a multi-scenario impact relationship table was formed. After collecting new rainstorm weather data, features such as an increase of ±1% in voltage fluctuation amplitude under the same load change were added to continuously improve the content of the table.

[0082] In one embodiment, collecting target data includes:

[0083] First, the filtering threshold is dynamically adjusted based on the real-time load fluctuations of the power grid. Then, an algorithm is used to distinguish between interference information and valid features in the data. It should be noted that dynamically adjusting the filtering threshold means changing the critical value for data filtering in real time according to the amplitude and frequency of the real-time load fluctuations of the power grid. This can be achieved by establishing a mapping relationship model between the load fluctuation amplitude and the threshold. Interference information refers to signals in the data that are unrelated to the operating status of power grid equipment and fault warnings, such as random noise from measuring equipment. This can be identified by comparing it with baseline data during normal operation.

[0084] The filtered valid data is time-stamped by binding key node parameters of power grid structure changes. This allows the archived data to automatically activate the corresponding factor transmission paths based on the stamps when entering the correlation analysis stage. It should be noted that key node parameters refer to core parameters reflecting changes in the power grid structure, such as node connection relationships and impedance values, which can be retrieved in real time from the power grid topology database. Time-stamping refers to adding timestamps and corresponding power grid operating time information to the valid data; the accuracy of the stamping can be ensured through high-precision clock synchronization technology.

[0085] Referring to the foregoing, in another embodiment of this application, when distinguishing interference information from valid features using a dynamic filtering algorithm:

[0086] First, a reference signal generated by the current load fluctuations of the power grid is captured. Then, differential analysis is performed between the electrical quantity data and this reference signal to extract abnormal segments where the difference exceeds the dynamic generation threshold. It should be noted that the reference signal refers to the characteristic electrical quantity signal (such as the average voltage value and the effective current value) that reflects the stable operating state of the current power grid load. It can be generated by smoothing load data from five consecutive sampling periods. Abnormal segments refer to continuous data segments where the difference between the electrical quantity data and the reference signal exceeds the dynamic threshold, and their duration must meet at least two sampling periods to eliminate transient noise.

[0087] For abnormal segments, the source is traced by combining the real-time trajectory of changes in the power grid structure. If it can be linked to the state change of a specific node, it is marked as a valid feature; otherwise, it is judged as interference information.

[0088] For example, this embodiment constructs a complete mechanism from data acquisition to effective feature extraction through a progressive process of load fluctuation benchmark, dynamic threshold adjustment, differential analysis, and source tracing verification.

[0089] It should be understood that using a fixed filtering threshold alone is difficult to adapt to the dynamic changes in the power grid load. Under light load, a fixed threshold may misjudge minor anomalies as interference, while under heavy load, it may miss significant anomalies, resulting in the loss of effective data or the introduction of interference information.

[0090] This embodiment adjusts the threshold based on real-time load fluctuations, allowing the filtering process to adapt synchronously with the power grid's operating status. For example, when the load fluctuation increases from 5% to 10%, the filtering threshold is adjusted from 2V to 4V accordingly, avoiding a disconnect between the threshold and actual operating conditions.

[0091] Next, by introducing differential analysis of the reference signal, abnormal data deviating from normal operating conditions can be accurately identified. Combined with tracing the source of power grid structural changes, the core issue of whether abnormal signals truly reflect the power grid state is resolved. Further explanation is needed: some anomalies may be caused by sensor malfunctions or electromagnetic interference (such as electromagnetic pulses during thunderstorms), which cannot be distinguished by numerical differences alone. Their validity can only be confirmed by associating them with changes in the state of specific nodes (such as switch opening / closing or tower tilting). Binding valid data to node parameters and time-series labeling further establishes a connection from data to nodes and finally to the path for subsequent correlation analysis, avoiding repeated retrieval of the factor transmission path corresponding to the data during correlation analysis and improving analysis efficiency.

[0092] The above process ensures the accuracy of effective feature extraction and provides precise guidance for the flow of data to the association analysis stage. It solves the problems of missing effective information or misjudging interference information caused by traditional fixed filtering, making the basic data entering the association analysis more reliable and more efficient to call.

[0093] To further illustrate this technical solution, a specific example is used: Suppose that line voltage data of a 35kV distribution network is collected and analyzed, and the real-time load fluctuation is 8%. The system first dynamically adjusts the filtering threshold to 10% (8% × 1.25). At the same time, it captures the voltage reference signal (e.g., 10kV ± 0.2kV) when the load is running stably. The real-time voltage data is compared with this reference signal. If it is found that the voltage drops to 9.5kV for three consecutive sampling cycles during a certain period, the difference of 0.5kV exceeds the dynamic threshold, then this segment is extracted as an abnormal segment.

[0094] Subsequently, by tracing the power grid structure change trajectory, it was found that the corresponding mid-section tower of the line tilted due to an external impact (change in node state) during that period. Therefore, this abnormal segment was marked as a valid feature. If no node state change was found during tracing, it was determined to be interference information (such as signal drift caused by loose sensor wiring). Finally, key parameters such as the tower ID "G35-012" and the time of structural change "14:32:15" were bound to the valid data, and time-series tags were added before archiving. When this data enters the correlation analysis stage, the system automatically activates the factor transmission path of "tower node line voltage load distribution" based on the tags, without the need to rematch the correlation relationship.

[0095] Furthermore, in one embodiment, invoking the early warning model combination includes:

[0096] First, the mutual influence intensity of the related factors is converted into a dynamic adjustment coefficient. The higher the influence intensity, the larger the coefficient value. It should be understood that the dynamic adjustment coefficient is a numerical parameter used to adjust the distribution of computing power in the model by mapping the mutual influence intensity between related factors. Its value range can be set to 0-1.

[0097] It should be understood that the intensity of the influence is positively correlated with the coefficient value.

[0098] Computational power weights are assigned to corresponding models based on coefficient values. After each round of analysis, the coefficient values ​​are reverse-calibrated according to the newly generated influence relationship table, ensuring that the computational power allocation adapts to the correlation patterns of factors in real time. It should be understood that the computational power weight refers to the proportion of total computational power resources allocated to each early warning model, and their sum is 1. It can be determined by dynamically adjusting coefficients proportionally.

[0099] It should be understood that distributing computing power equally across all models can lead to insufficient resources for models corresponding to critical factors, while models for non-critical factors consume excessive resources. For example, in fault scenarios dominated by load mutations, the model responsible for load correlation analysis should receive more computing power; otherwise, insufficient computation may result in delayed early warnings.

[0100] In this embodiment, the influence intensity is converted into an adjustment coefficient, which allows the computing power to be tilted toward the model with a more significant influence. For example, when the influence intensity of line current and temperature is 0.8, the corresponding model obtains a higher computing power weight than the environmental humidity and insulation performance model with an influence intensity of 0.3.

[0101] It should be further understood that the reverse calibration mechanism after each round of analysis solves the coefficient mismatch problem caused by changes in the correlation patterns of factors. The intensity of the influence of factors in power grid operation changes with conditions such as load and weather. For example, the influence of ambient temperature and equipment temperature increases in hot weather. If the coefficients are not adjusted in time, the computing power allocation will lag behind actual demand. Calibrating the coefficients based on the newly generated influence relationship table ensures that the adjustment coefficients are always synchronized with the current correlation patterns of factors. For example, when the relationship table shows that the influence of load mutations on voltage fluctuations increases from 0.6 to 0.9, the corresponding adjustment coefficient of the model is adjusted upwards, and the computing power weight increases accordingly.

[0102] This method, which converts influence intensity into adjustment coefficients and then allocates computing power based on these coefficients, combined with dynamic calibration, enables the early warning model combination to flexibly allocate computing resources according to real-time changes in factor correlations. This ensures the computational efficiency of key models while avoiding resource waste, solving the problem of insufficient adaptability of fixed computing power allocation in scenarios with dynamically changing factor correlations, and improving the real-time performance and accuracy of early warning analysis. For example, suppose an early warning model combination for a 110kV transmission network involves three models: a line overload early warning model, an equipment temperature early warning model, and a voltage stability early warning model. Analysis shows that the interaction intensity between line overload and equipment temperature is 0.8, and the interaction intensity between line overload and voltage stability is 0.5. The system converts these intensities into dynamic adjustment coefficients of 0.8 and 0.5, respectively.

[0103] Furthermore, based on the aforementioned coefficient values ​​for allocating computing power weights, the line overload early warning model and the equipment temperature early warning model receive a higher proportion of computing power, while the voltage stability early warning model receives a relatively lower proportion. After one round of analysis, the newly generated influence relationship table shows that the mutual influence strength between line overload and voltage stability has become 0.7. Based on this, the system performs reverse calibration of the dynamic adjustment coefficient, adjusting the corresponding coefficient to 0.7, and then reallocates the computing power weights, allowing the voltage stability early warning model to receive more computing power to adapt to the new factor correlation patterns.

[0104] Furthermore, in another embodiment of this application, when the judgment information shared between models includes factor feature matching results, trend change predictions, and risk probability assessments, the following steps are included: It should be understood that factor feature matching results refer to the degree of consistency determination obtained by different models after comparing the features of related factors with a preset feature library, which can be calculated through feature vector similarity. Trend change predictions refer to the model's prediction of the future development trend of related factors based on the real-time values ​​and historical change patterns of related factors, which can be achieved using time series prediction algorithms. Risk probability assessments refer to the model's estimation of the probability of possible failure risks based on the current state and mutual influence relationships of related factors, usually presented as a percentage.

[0105] When the real-time numerical fluctuation of a related factor exceeds a set proportion, the relevant models will synchronously send updated judgment information. The receiving models will cross-check their own analysis results; those that match will be used as the basis for comprehensive judgment, while those that do not will undergo secondary analysis. It should be understood that the set proportion refers to a pre-set threshold for the magnitude of numerical fluctuations of related factors that triggers the synchronous transmission of information between models. This threshold can be adjusted according to different power grid operating scenarios and the importance of the related factors. Secondary analysis refers to the process of re-analyzing and verifying the discrepancies by introducing more relevant data or employing more sophisticated algorithms when the cross-check results between models are inconsistent.

[0106] This embodiment ensures the accuracy and reliability of the early warning model combination analysis by clearly defining the judgment information types and information interaction and verification methods between models.

[0107] It should be understood that when the real-time value of related factors changes by more than a set proportion, updated judgment information is sent synchronously. This ensures that each model receives the latest analytical basis in a timely manner, avoiding judgments based on outdated information. The receiving models cross-check their own analysis results with the information. Those that match are used as the basis for comprehensive judgment, enhancing the credibility of the judgment results. When there are inconsistencies, secondary analysis is performed to help identify the reasons for the differences and reduce misjudgments.

[0108] This method of information exchange and verification between models can solve the problem of analysis result deviation caused by the inability of models to process data collaboratively, enabling the combination of early warning models to fully exert its synergistic effect and improve the accuracy of multi-dimensional data collaborative analysis and early warning in intelligent operation and maintenance of power transmission networks. Simply put, suppose that the early warning model combination of a certain 110kV power transmission network includes a line status analysis model, a load forecasting model, and a fault diagnosis model. When the real-time value change of the associated factor, line current, exceeds a set proportion of 10%, the line status analysis model simultaneously sends its factor characteristic matching results (85% matching degree with line overload characteristics), trend change prediction (potentially rising by 15% within the next 5 minutes), and risk probability assessment (60% probability of line overload fault) to the load forecasting model and the fault diagnosis model.

[0109] After receiving the information, the load forecasting model cross-checks it with its own load trend analysis. It finds that both models agree on the trend of line current changes, and this information is used as the basis for the overall judgment. Similarly, after receiving the information, the fault diagnosis model compares it with its own analysis of line insulation status. It finds that the assessment of fault risk probability is inconsistent (its own assessment is 40%). Therefore, it initiates a secondary analysis, incorporating more relevant data such as line temperature and humidity to recalculate, ultimately yielding a more accurate risk probability assessment.

[0110] In one embodiment, when the deviation exceeds a set value, the method further includes:

[0111] With the linkage response of data acquisition frequency and load control parameters as the core, the judgment results of each model are summarized to obtain a consistent conclusion about the cause of the deviation. It should be noted that the data acquisition frequency refers to the number of times relevant data of the transmission network are collected per unit time. Its frequency directly affects the real-time performance and completeness of the data and can be dynamically set according to actual needs.

[0112] It needs to be clarified again that load control parameters refer to various parameters used to adjust the distribution and magnitude of power grid load, such as transformer tap positions and reactive power compensation device switching capacity. Interlocking response refers to the response mode in which data acquisition frequency and load control parameters are correlated and change in a coordinated manner, ensuring that both can work together according to power grid deviations.

[0113] It should be further explained that a consensus conclusion refers to the same or similar judgments reached by each model after analyzing the causes of the deviations, which is an important basis for subsequent adjustments and decisions.

[0114] Based on the conclusions, adjust the relationship between the data acquisition frequency and the load control parameters. The greater the load control range, the higher the acquisition frequency should be.

[0115] Based on the adjusted coordination relationships and the results of supplementary model analysis, a real-time early warning plan is generated, including risk level, impact range, and control recommendations. It should be noted that the risk level refers to the classification of potential risks based on factors such as the severity and development trend of the deviation, typically categorized as minor, moderate, severe, or urgent. The impact range refers to the area or equipment range where the deviation may affect the transmission network. Control recommendations are specific operational suggestions proposed based on the analysis and judgment of the deviation to resolve the deviation and reduce risk.

[0116] For example, when dealing with deviations, if only one parameter is adjusted, the deviation may not be properly handled due to an inaccurate understanding of the power grid status.

[0117] Therefore, by summarizing the judgment results of various models to obtain a consistent conclusion on the cause of the deviation, we can avoid the one-sidedness of analysis based on a single model and provide a reliable basis for subsequent adjustments. Based on this conclusion, we can adjust the coordination between the two models; for example, if the load regulation range is large, we can increase the data acquisition frequency. This allows us to capture the changes in the power grid during large-scale regulation in more detail, providing more sufficient data for analysis and decision-making.

[0118] Finally, by combining the adjusted coordination relationships and supplementary model analysis results, a real-time early warning scheme is generated. This makes the scheme more consistent with the actual situation of the power grid, improving the accuracy and practicality of the early warning, unlike traditional schemes which may have poor results due to poor parameter coordination or incomplete analysis. Specifically, suppose that in a certain 220kV transmission network, the voltage deviation exceeds the set value by 5%. The system takes the linkage response of data acquisition frequency and load control parameters as its core, summarizes the judgment results of the line status analysis model, load prediction model, and voltage stability model, and draws a consistent conclusion that the deviation is caused by a sudden increase in load in a certain area.

[0119] Based on the above conclusions, the system adjusts the coordination between the data acquisition frequency and the load control parameters. Since the load control range for this area needs to be set to 20%, which is a relatively large range control, the data acquisition frequency is increased from the original 2 times per second to 5 times per second.

[0120] Subsequently, the system combines the adjusted coordination relationship with the results of supplementary analysis of each model to generate a real-time early warning plan. The risk level is determined to be general, the affected area is the power grid equipment within 10km of the region, and the control suggestion is to appropriately activate reactive power compensation devices to stabilize the voltage.

[0121] Furthermore, in one embodiment, when a sudden load surge occurs in the power grid:

[0122] The routine factor correlation analysis process is suspended, and the characteristics of electrical quantity mutations before the impact are extracted. The above characteristics are quickly compared with typical characteristics in the historical impact case library, and reference schemes are generated after being sorted by matching degree. At the same time, the parallel calculation of the current power grid structure tolerance of each model is performed, and the calculation results are combined with the reference scheme to form a temporary control strategy. After the impact is relieved, routine analysis is performed and the temporary strategy is calibrated.

[0123] It should be noted that methods for extracting electrical quantity abrupt changes before an impact include, for example, Hankel matrix singular value decomposition. This algorithm detects abrupt changes in the recorded waveform signal and compares the absolute values ​​of elements in the component signals with a set threshold value to determine the timing of the electrical quantity abrupt change. Another method is wavelet transform-based, which uses multi-scale analysis to extract transient features of the current waveform, such as spikes and abrupt changes. Additionally, catastrophe theory is employed to extract multiple features characterizing arc current abrupt changes, such as time-domain integral values ​​and difference-root mean square features. Then, catastrophe theory evaluation methods are used to comprehensively calculate the membership value of the line current abrupt change, and so on.

[0124] Sudden load surges refer to a significant and rapid increase or decrease in grid load within a short period of time, which may be caused by sudden factors such as extreme weather or the start-up and shutdown of large equipment. Abnormal changes in electrical quantities refer to the unusual changes in electrical quantities such as voltage, current, and power before a load surge occurs, such as a sudden drop in voltage or a sudden increase in current.

[0125] The historical impact case library is built using a relational database (such as MySQL). Each case includes the time of impact, the magnitude of load change, electrical quantity characteristic parameters corresponding to the extracted features, the power grid topology, the handling measures, and the implementation effect.

[0126] The structural withstand capability of a power grid refers to the maximum pressure that the current power grid's equipment, lines, and other structures can withstand under load impacts, reflecting the power grid's ability to resist impacts.

[0127] The temporary control strategy can be generated in, but is not limited to, JSON format, and includes feedback verification indicators such as the control object ID, operation amount, execution timing accurate to milliseconds, and allowable voltage fluctuation range.

[0128] It should be further explained that parallel computing is implemented based on a distributed computing architecture (such as the Apache Spark framework), which means that multiple models perform calculations on the tolerance of different structural parts of the power grid at the same time.

[0129] Furthermore, when the power grid encounters a sudden load surge, by suspending the regular process and prioritizing the extraction of electrical quantity change characteristics before the surge, the key to the problem can be quickly identified.

[0130] Furthermore, the extracted features are quickly compared with typical features in a historical impact case database, and reference solutions are generated based on the matching degree. This is equivalent to drawing on past successful experiences, enabling the development of targeted response methods in a short period of time. Because historical cases contain handling methods for various impact scenarios, the development of solutions can be accelerated.

[0131] It should be understood that the various models should be able to calculate the current grid structure's resilience in parallel. Specifically, this involves allocating independent calculation threads to different models. The line resilience model focuses on calculating the maximum current carrying capacity and temperature tolerance limits of each line segment under the current load impact; the transformer resilience model focuses on the load factor limits and insulation tolerance of each transformer; and the switchgear model calculates the switching capacity and withstand voltage of various switches. Each model calls its corresponding data module, such as the line parameter library, transformer nameplate data, and switchgear performance parameters, and the calculations proceed synchronously without interference. This allows for a comprehensive acquisition of the resilience limits of key grid structures in a short time, accurately grasping the grid's current capacity and avoiding a disconnect between reference schemes and actual conditions. Combining the calculation results with the reference scheme to form a temporary control strategy, which references historical experience while adhering to the current grid reality, makes the strategy more feasible and secure.

[0132] It should be noted that the execution of the temporary control strategy adopts a closed-loop control mode: after the control command is issued, the change curves of target parameters such as node voltage are monitored in real time and compared with the expected curve preset by the strategy. If the deviation exceeds the range, a secondary fine adjustment is immediately triggered.

[0133] Once the impact subsides, the system will activate the routine analysis module, use cause-effect graph analysis to trace the root cause of the impact, and adjust the parameters of the temporary strategy based on the actual data. The adjusted values ​​will be automatically updated to the historical case library, forming a closed loop for strategy optimization.

[0134] To better illustrate this scheme: Suppose that a 110kV power grid experiences a sudden load surge due to the sudden start-up of high-power equipment at a nearby factory, with the load increasing by 30% within 10 seconds.

[0135] The system immediately paused the routine factor correlation analysis process, extracted the electrical quantity change characteristics before the impact, and found that the voltage dropped from 110kV to 102kV within 2 seconds before the impact, and the current suddenly increased from 500A to 680A.

[0136] By comparing these characteristics with a historical impact case database, it was found that the characteristics matched 85% with those of a load impact caused by the startup of equipment in a factory three years ago. Based on this, a reference solution was generated, including temporarily reducing non-critical loads in the surrounding area and adjusting transformer taps.

[0137] Simultaneously, parallel calculations were initiated for the line withstand capability model and the transformer load-bearing model. The line withstand capability model, utilizing parameters such as material, cross-sectional area, and installation environment of each section of the 110kV power grid, calculated the maximum current that the current line could withstand to be 700A. The transformer load-bearing model, based on data such as the rated capacity, service life, and cooling method of each transformer within the jurisdiction, determined that the maximum transformer load rate was 90%. Combined with the reference scheme, a temporary control strategy was formulated, namely, reducing non-critical loads in the surrounding area by 10% and adjusting the transformer taps to appropriate positions.

[0138] After 15 minutes, the load surge subsided, and the system performed routine analysis. It was found that the temporary strategy effectively stabilized the power grid, but could be further optimized. Therefore, the temporary strategy was calibrated, such as adjusting the reduction of non-critical loads to 8%, to make the strategy more reasonable.

[0139] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-dimensional data collaborative analysis and early warning method for intelligent operation and maintenance of power transmission networks, characterized in that: include: Collect target data from the target device, and simultaneously acquire data on sudden changes in the power load of the system; By combining the sudden change data of power load, we can conduct correlation analysis on the instantaneous characteristics of electrical quantities and the dynamic changes of power grid structure to obtain the correlation factors that can predict faults. Based on the magnitude of sudden changes in power load, determine the interaction patterns among various related factors; Based on the real-time values ​​and impact patterns of related factors, the corresponding combination of early warning models is invoked, and the models exchange judgment information in a timely manner as the factors change. When the real-time numerical deviation of the related factors exceeds the set value, the system takes the linkage response of data acquisition frequency and load control parameters as the core, combines the judgment results of each model, dynamically adjusts the adaptation relationship of data acquisition frequency and load control parameters, and generates a real-time early warning scheme. The correlation analysis between the instantaneous characteristics of electrical quantities and the dynamic changes in the power grid structure includes the following steps: S1: Convert the instantaneous characteristics of electrical quantities into corresponding energy signals, and then map them to the relevant nodes of the power grid structure; S2: After receiving this signal, the node generates an initial state change; S3: Track the chain transmission path formed by this state change as the power grid structure changes. Each node in the path will provide real-time feedback on the effect of its own state change on the upstream signal. S4: Apply the effect in reverse to the original corresponding node to form the first round of interaction between features and structure; S5: Use the result of this loop as a new starting point and repeat steps S1 to S4 above; S6: Obtain the cumulative deviation of node status in multiple cycles, extract the related factors, and introduce the real-time load fluctuation of the power grid as a dynamic adjustment parameter in each cycle, so that the factor extraction process is consistent with the actual operation of the power grid. When determining how the various related factors influence each other: Based on the correspondence between electrical quantity characteristics and power grid structural nodes, an initial causal relationship sequence is established according to the order in which the factors appear. Based on the obtained power load change data, the effectiveness of the triggering of each link in the sequence is verified by real-time data. Invalid links are replaced and supplemented. During the supplementation process, relevant factors that may affect the information exchange of subsequent model judgments need to be included. Based on the extraction of the related factors, the response delay differences and amplitude fluctuation ranges of other related factors under different power grid load states after a change in a single related factor are recorded, and the effect of the recorded results on the subsequent deviation judgment process is predicted simultaneously. Based on the above records, a multi-scenario impact relationship table is formed that can be dynamically updated. The contents of the table are continuously corrected as new data is added. When calling a combination of early warning models, the following are included: First, the mutual influence intensity of the related factors is converted into a dynamic adjustment coefficient. The higher the influence intensity, the larger the coefficient value. The computing power weights are assigned to the corresponding models according to the coefficient values, and after each round of analysis, the coefficient values ​​are reverse-calibrated according to the newly generated influence relationship table, so that the computing power allocation can adapt to the correlation patterns of factors in real time. The information shared between models includes factor feature matching results, trend change predictions, and risk probability assessments: When the real-time value of the related factors changes by more than a set proportion, the relevant model will send updated judgment information synchronously, and the receiving model will cross-check the information with its own analysis results. Content that matches the verification results will be used as the basis for a comprehensive judgment; content that does not match will be subject to secondary analysis.

2. The method for collaborative analysis and early warning of multi-dimensional data based on intelligent operation and maintenance of power transmission networks according to claim 1, characterized in that, When collecting target data, this includes: First, the filtering threshold is dynamically adjusted based on the real-time load fluctuation of the power grid, and then the algorithm is used to distinguish between interference information and effective features in the data. The filtered valid data is bound to key node parameters of power grid structure changes and time-series marked, so that when the archived data enters the correlation analysis stage, the corresponding factor transmission path can be automatically activated according to the marking.

3. The method for collaborative analysis and early warning of multi-dimensional data based on intelligent operation and maintenance of power transmission networks according to claim 2, characterized in that, When distinguishing between interference information and valid features using dynamic filtering algorithms: First, capture the reference signal generated by the current load fluctuation of the power grid. Then, perform differential analysis between the electrical quantity data and the reference signal to extract abnormal segments whose difference exceeds the dynamically generated threshold. For abnormal segments, the source is traced by combining the real-time trajectory of changes in the power grid structure. If it can be linked to the state change of a specific node, it is marked as a valid feature; otherwise, it is judged as interference information.

4. The method for collaborative analysis and early warning of multi-dimensional data based on intelligent operation and maintenance of power transmission networks according to claim 1, characterized in that, When the deviation exceeds the set value, it also includes: Taking the linkage response of data acquisition frequency and load control parameters as the core, the judgment results of each model are summarized to obtain a consistent conclusion about the cause of the deviation. Based on the conclusions, adjust the relationship between the data acquisition frequency and the load control parameters. The greater the load control range, the higher the acquisition frequency should be. Based on the adjusted coordination relationship and the results of the model supplementary analysis, a real-time early warning plan is generated, which includes risk level, scope of impact and control recommendations.

5. The method for multi-dimensional data collaborative analysis and early warning based on intelligent operation and maintenance of power transmission networks according to claim 1, characterized in that, When the power grid experiences a sudden load surge, it also includes: The routine factor correlation analysis process was suspended, and the characteristics of electrical quantity abrupt changes before the impact were extracted; The above features are quickly compared with typical features in the historical impact case database, and reference solutions are generated after being sorted by matching degree. Meanwhile, each model performs parallel calculations on the current power grid structure's tolerance capabilities, and combines the calculation results with the reference scheme to form a temporary control strategy. After the impact subsides, routine analysis will be conducted and the temporary strategy will be calibrated.

Images