Centralized management method and management system for power supply system

Abstract

The invention relates to the technical field of power supply system management, and discloses a centralized management method and management system for a power supply system. The method comprises the following steps: monitoring a current fluctuation sequence and voltage offset data of each node of a power supply network in a continuous operation period, and collecting an equipment state log text and load change time sequence information; identifying a time point of abnormal sudden change of current fluctuation, and extracting a characteristic parameter set of a power supply quality reduction event in combination with a voltage offset data change amplitude; matching a historical abnormal event library, positioning key operation record fragments strongly related to load change in the log, and generating a mapping relation table of equipment operation behaviors and power supply quality fluctuation through semantic analysis and time sequence alignment; and correcting the parameters of the load prediction model, outputting and executing a power supply priority dynamic adjustment strategy, and triggering automatic calibration of equipment operation parameters. According to the method, comprehensive control of the power supply system is realized, abnormity can be identified and processed in time, association between operation and quality is clear, and stability and flexibility of operation of the power supply system are improved.

Description

Technical Field

[0001] This invention relates to the field of power supply system management technology, specifically to a centralized management method and management system for power supply systems. Background Technology

[0002] With the continuous expansion of power grid scale, the structure of power supply systems is becoming increasingly complex, and the number of various node devices is growing exponentially, leading to ever-increasing demands for power supply stability and reliability. Traditional power supply management methods largely rely on decentralized monitoring, where the collection and analysis of operational data from each node device are relatively independent, making it difficult to achieve coordinated scheduling across the entire network. In actual operation, phenomena such as current fluctuations and voltage deviations frequently occur. These anomalies are often caused by a combination of factors, including equipment aging, sudden load changes, and operational errors. Under the decentralized management model, data transmission delays result in insufficient timeliness in the identification and handling of abnormal events. Existing management systems often only superficially analyze power supply equipment status logs, failing to delve into the key information related to load changes within the log text. This results in an unclear correlation between equipment operation behavior and power quality fluctuations. When power quality degradation occurs, managers struggle to quickly pinpoint the root cause, often requiring extensive manual investigation. This not only increases workload but can also lead to wider power outages due to delayed responses. Furthermore, load changes are highly dynamic and random. Traditional predictive models with fixed parameters struggle to adapt to real-time load variations, resulting in inflexible power priority adjustment strategies that fail to fully meet the electricity needs of different regions and users, ultimately impacting the overall operational efficiency of the power supply system. Summary of the Invention

[0003] The purpose of this invention is to provide a centralized management method and management system for power supply systems to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a centralized management method for a power supply system, the method comprising: Monitor the current fluctuation sequence and voltage offset data of each node in the power supply network during continuous operation cycle, and collect the status log text information and load change time sequence information of the power supply equipment; Identify the start and end times of abnormal abrupt changes in the current fluctuation sequence, and extract the set of characteristic parameters of the power quality degradation event by combining the change amplitude of voltage offset data within the corresponding time window. Based on the set of feature parameters, match the associated event patterns in the historical abnormal event database to locate key operation record fragments in the power supply equipment status log text that are strongly correlated with the timing of load changes. Semantic parsing and operation timing alignment are performed on the key operation record segments to generate a mapping table between device operation behavior and power quality fluctuations; Based on the mapping table, the prediction model parameters of the real-time load change sequence are corrected, and a dynamic power supply priority adjustment strategy is output. The power supply priority dynamic adjustment strategy is executed, triggering an automatic calibration command for the operating parameters of the power supply equipment.

[0005] Preferably, the set of feature parameters for extracting power quality degradation events includes: Within the time interval from the start time to the end time of the abnormal mutation, multiple continuous detection sub-windows are divided; Calculate the product of the current fluctuation range and the average voltage deviation in each sub-window, and then calculate the degree of deviation of the product from the preset safety threshold. Detect the overlap between the load change rate in each sub-window and the timing of switching actions in the device status log; The deviation degree and the temporal overlap degree are combined to generate a sub-window anomaly index; Aggregate all sub-window anomaly indices to form a set of feature parameters.

[0006] Preferably, the key operation record segments in the status log text of the positioning power supply equipment that are strongly correlated with the timing of load changes include: Iterate through the timestamps of all operation instructions recorded in the power supply equipment status log text; Filter operation instruction texts whose timestamps fall within the time interval corresponding to the feature parameter set; Extract the device number, operation type code, and operation duration from the filtered operation instruction text; The duration of the operation is matched with the duration of the sub-window anomaly index in the feature parameter set based on similarity. Retain operation instruction texts with similarity exceeding the matching threshold as key operation record fragments.

[0007] Preferably, the mapping table between the generated device operation behavior and power quality fluctuations includes: Perform topology analysis on the device numbers in key operation record segments to determine the hierarchical position of the devices in the power supply network; The frequency of concurrent operations on devices at the same level is statistically analyzed based on the operation type code. Establish the correspondence between the frequency of concurrent operations and the numerical trend of the sub-window anomaly index in the feature parameter set; The device hierarchy location, concurrent operation frequency, and abnormal index change trends are integrated into a mapping table.

[0008] Preferably, the prediction model parameters for correcting the real-time load change time series based on the mapping table include: Retrieve the historical load baseline data corresponding to the device level position in the mapping table; Calculate the difference vector between the current real-time load change time series and the historical load baseline; Based on the correlation between the difference vector and the abnormal index change trend in the mapping table, model parameter correction coefficients are generated; The correction coefficients are injected into the time-series feature extraction layer of the load prediction model for parameter updates.

[0009] Preferably, the dynamic adjustment strategy for output power supply priority includes: Obtain the load distribution forecast values ​​for the next three periods from the updated load forecasting model; The power supply emergency level is determined based on the maximum load percentage of each node in the load distribution forecast. Based on the association rules between device hierarchical location and operation frequency in the mapping table, a hierarchical power supply weight allocation scheme is generated; A dynamic adjustment strategy is generated by integrating the emergency power supply level and the hierarchical power supply weight allocation scheme.

[0010] Preferably, the automatic calibration command for triggering the operating parameters of the power supply equipment includes: Analysis of the target power supply level and emergency level threshold in the dynamic adjustment strategy; Match the real-time operating parameter dataset of all power supply devices within the target power supply level; Compare the deviations between the current stability index and the emergency level threshold in the operating parameter dataset; Generate a set of calibration instructions for devices whose deviation values ​​exceed the tolerance range.

[0011] Preferably, the method further includes: Collect timing data of the device's operational response after executing calibration commands; Extract the feature values ​​of voltage recovery rate and current fluctuation convergence time from the response time series data; The feature values ​​are fed back to the feature parameter set of the power quality degradation event for incremental updates.

[0012] Preferably, the method further includes: Establish a version association index between the power supply node monitoring dataset and the device status log; When a new power supply node is connected to the network, the matching dimensions of the historical abnormal event database are synchronously expanded based on the version association index; When dynamically adjusting the output power supply priority strategy, the expanded matching dimension is loaded to verify the strategy.

[0013] Preferably, the present invention further includes a centralized management system for a power supply system, the system comprising: The power quality monitoring module is used to collect real-time current fluctuation sequences, voltage offset data, and equipment status log text of power supply nodes; An anomaly feature extraction module, connected to the power quality monitoring module, is used to identify the feature parameter set of power quality degradation events and generate a sub-window anomaly index; The operation behavior analysis module is connected to the abnormal feature extraction module and is used to locate key operation record segments and analyze the device level position. The mapping relationship construction module, connected to the operation behavior analysis module, is used to generate a mapping relationship table between device operation behavior and power supply quality fluctuations; The prediction model update module, connected to the mapping relationship construction module, is used to correct the load prediction model parameters; The strategy generation module is connected to the prediction model update module and is used to output a dynamic adjustment strategy for power supply priority. The instruction execution module, connected to the strategy generation module, is used to trigger automatic calibration instructions for the operating parameters of the power supply equipment; The feedback optimization module connects the instruction execution module and the abnormal feature extraction module, and is used to incrementally update the feature parameter set. The data version management module connects all modules and is used to maintain the version association index of power supply node monitoring data.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By monitoring the current fluctuation sequence and voltage offset data of each node in the power supply network during continuous operation cycles, and collecting status log text information and load change timing information of the power supply equipment, a comprehensive understanding of the power supply system's operating status is achieved. This comprehensive data acquisition method can promptly capture any anomalies that may occur in the system, providing rich foundational data for subsequent analysis and processing. In identifying power quality degradation events, by accurately locating the start and end times of abnormal abrupt changes in the current fluctuation sequence and combining this with the magnitude of voltage offset data changes within the corresponding time window, a set of characteristic parameters is extracted, making the characteristics of the abnormal events clearer and more explicit. Matching this set of characteristic parameters with associated event patterns in a historical abnormal event database helps to quickly identify the type and potential impact range of the current abnormal event, reducing the blindness in judging abnormal events. Semantic parsing and timing alignment of key operation record segments strongly correlated with load change timing in power supply equipment status log text generate a mapping table between equipment operation behavior and power quality fluctuations. This reveals the intrinsic connection between equipment operation and power quality. This clear correlation allows managers to quickly trace relevant equipment operations when facing power quality issues, providing direct clues for problem investigation and resolution. Based on a mapping table, the parameters of the prediction model are corrected according to the real-time load change sequence, and a dynamic power supply priority adjustment strategy is output. This allows the prediction model to better adapt to real-time load changes, and the adjustment strategy to be more targeted and flexible. This dynamic adjustment mechanism can rationally allocate power supply resources according to different load changes and power quality conditions, ensuring that the power supply system maintains stable operation in complex and ever-changing environments. By implementing a dynamic power supply priority adjustment strategy, automatic calibration commands for the operating parameters of power supply equipment are triggered, enabling intelligent management of the equipment. The timely triggering of these automatic calibration commands can quickly correct deviations in equipment operating parameters, reducing manual intervention, improving the response speed and processing efficiency of the power supply system, helping to reduce the incidence of power supply failures, and ensuring the continuity and stability of power supply. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the centralized management method for the power supply system described in this invention.

[0016] Figure 2 A flowchart for extracting the set of characteristic parameters of power quality degradation events.

[0017] Figure 3 A flowchart for locating key operation recording segments.

[0018] Figure 4 A flowchart for correcting the parameters of the prediction model.

[0019] Figure 5 The flowchart for triggering the automatic calibration command. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 This invention provides a centralized management method for a power supply system, the method comprising: Sensors deployed in the power supply network collect current fluctuation sequences and voltage offset data, while textual information and load change timing information are extracted from the status logs of power supply equipment. Abnormal abrupt changes in the current fluctuation sequence are identified by time window analysis to pinpoint the start and end times, and a set of characteristic parameters for power quality degradation events is extracted by combining the magnitude of voltage offset data changes.

[0022] The historical anomaly database stores past anomaly patterns. By matching the set of feature parameters with historical patterns, key operation record segments in the status log that are strongly correlated with load changes are located. After semantic parsing and operation timing alignment, these segments generate a mapping table between device operation behavior and power quality fluctuations. This table is used to correct the parameters of the real-time load prediction model, dynamically adjust the power supply priority strategy, and trigger automatic calibration commands for power supply equipment to ensure the stable operation of the power supply network.

[0023] Example 1: See Figure 2 In the process of extracting the characteristic parameter set of power quality degradation events, the time interval of abnormal changes is used as the core analysis unit. This time interval begins at the first sampling point in the current fluctuation sequence that exceeds the preset fluctuation threshold and ends at the last sampling point where the fluctuation value returns to a stable range. This interval is divided into multiple continuous detection sub-windows, and the length of the sub-windows is dynamically adjusted according to the total duration of the change: if the total duration does not exceed 60 seconds, the number of sub-windows is set to 5; if the total duration is between 60 and 300 seconds, the number of sub-windows is 10; and if the total duration exceeds 300 seconds, the number of sub-windows is fixed at 15. Each sub-window completely covers the abnormal period without gaps, and the window boundaries are divided at equal time intervals.

[0024] The data analysis within each sub-window comprises two independent calculation dimensions. The first dimension calculates the current fluctuation range by extracting the instantaneous maximum and minimum values ​​of the current fluctuation sequence. Simultaneously, it collects values ​​from all voltage sampling points within the same time period and calculates the arithmetic mean of the voltage offset data as the voltage offset mean. The product of the current fluctuation range and the voltage offset mean is used as the basic indicator for the sub-window. A preset safety threshold is set based on the equipment's rated parameters, serving as the upper limit for the basic indicator. Deviation is calculated using the relative deviation method, which measures the percentage of the basic indicator exceeding the safety threshold. When the basic indicator is below the safety threshold, the deviation is recorded as 0.

[0025] The second dimension analyzes the correlation between device status log text and load changes. Text records containing key actions such as on / off operations, mode switching, and load adjustment are extracted from the device status logs, and entries with clearly marked operation timestamps are identified. Within a sub-window time range, all operation records with timestamps falling within this interval are filtered. Simultaneously, power change data recorded by the load monitoring system within the same sub-window is acquired, and the average absolute value of load change per minute is calculated as the load change rate. Timing overlap is determined through a dual-track time axis comparison: a set of operation timestamps and a set of moments with significant load rate changes are established. If the number of overlapping operation points in the two sets exceeds 50% of the total operation records, the timing overlap is determined to be 1; an overlap between 30% and 50% is recorded as 0.5; and less than 30% is recorded as 0.

[0026] The sub-window anomaly index is generated by combining the two dimensions mentioned above. The deviation degree is linearly normalized and mapped to the range of 0 to 1, while the temporal overlap degree is directly used as a discrete value in the 0 to 1 interval. The anomaly index uses a weighted synthesis formula, with the deviation degree accounting for 70% and the temporal overlap degree accounting for 30%, and the two are added together to obtain the final index value. The anomaly index values ​​of each sub-window form an ordered sequence, forming the basic structure of the feature parameter set.

[0027] The feature parameter set contains multidimensional metadata: the start and end times of each sub-window, the anomaly index value, the original value of the deviation degree, and the time series overlap judgment flag. The set as a whole is labeled with the anomaly mutation type code, which is divided into three categories according to the current fluctuation waveform characteristics: instantaneous pulse type, continuous oscillation type, and step change type. Global statistical values ​​are appended at the end of the set, including the maximum, minimum, and average values ​​of the anomaly index of all sub-windows, as well as the time distribution variance value.

[0028] A dual verification mechanism is employed during data processing. After the first round of sub-window division, the system automatically detects abnormal index differences between adjacent sub-windows. If the index change of three consecutive sub-windows exceeds 40%, a window re-division procedure is triggered: new window boundaries are added on both sides of the index abrupt change point, and the index calculation is re-executed. The final generated feature parameter set is stored in a JSON structured format, containing timestamp indexes, data type identifiers, and checksums, for direct parsing by downstream modules.

[0029] The sub-window partitioning process is compatible with non-uniform sampling scenarios. When the power supply network has multi-band mixed sampling, the system automatically aligns the data acquisition timescales of each node. For cases where current fluctuation sequences are sampled at 1kHz and voltage offset data at 100Hz, the lower sampling rate data is reconstructed by interpolation based on the higher sampling rate, ensuring time consistency of all data points within the sub-window. Load change timing information is unified to the same time resolution using a sliding window averaging method.

[0030] The structure of the feature parameter set supports dynamic expansion. When a new abnormal fluctuation pattern is detected, the system automatically appends newly defined extended fields to the end of the set. For example, when a composite anomaly of negative voltage offset superimposed with high-frequency oscillation occurs for the first time, new fields are added to record the percentage of negative voltage duration and the integral value of high-frequency oscillation energy. This design ensures that the feature parameter set can adapt to the new anomaly pattern recognition requirements brought about by power supply network equipment upgrades.

[0031] Example 2: See Figure 3 The process of filtering key operation record fragments begins with traversing all entries in the power supply equipment status log text. Each operation instruction record contains a millisecond-level timestamp, an operation device identifier, and an operation description text. The system employs a time interval matching algorithm, using the start and end times of abnormal events marked by a set of feature parameters as the main timeline. All original text records whose timestamps fall within this interval are extracted from the log database, forming an initial candidate set. The text parsing engine processes each record in the candidate set, removing redundant descriptive information and extracting structured triples: the device number is converted to a 12-bit string using standard address encoding rules; the operation type code is mapped to a 3-digit numeric code according to the International Electrotechnical Commission (IEC) standard classification library; and the operation duration is automatically calculated in seconds from the start and end time fields in the parsed text.

[0032] The association between operation duration and feature parameter set employs a multi-layer time alignment mechanism. Each sub-window's anomaly index is labeled with its own duration, forming a discrete time-segment sequence. The system creates a bidirectional matching channel between the operation record duration sequence and the anomaly index duration sequence. A dynamic time warping algorithm establishes the optimal alignment path between the two sequences, calculating the cumulative value of the shortest distance between sequence points. A waveform similarity factor is introduced into the matching score calculation, comparing the morphological consistency between the operation record duration curve and the anomaly index duration curve. A similarity threshold of 0.75 is set; operation records exceeding this value are marked as strongly correlated objects. The system retains the top 20% of matching results as key operation record segments and labels the matching score as a confidence index.

[0033] The hierarchical topology resolution is based on a power supply network node connection graph database. After inputting a device number, it queries information about its superior hub nodes and subordinate branch nodes, generating metadata containing five layers of topological relationships: regional substation level, trunk line level, distribution cabinet level, terminal equipment level, and user access level. Each device is labeled with its physical coordinates and logical location index. The resolution process simultaneously calculates the topology weight value of each device, which depends on the area of ​​the power supply range affected when the device is disconnected. The weight values ​​are normalized to a range of 0 to 100 and embedded in the attribute fields of key operation record fragments.

[0034] Concurrent operation frequency statistics are performed at a unified topology level. First, key operation record segments are categorized by level, and within the same level, they are aggregated into 10-second time buckets. Three key indicators are calculated within each time bucket: the number of operating devices, the frequency of the same operation type, and the proportional distribution of different operation types. Frequency calculations are weighted by topology weight values, with the operation count of high-weight devices multiplied by a coefficient of 1.5. The system generates a time-sorted hierarchical operation frequency curve, with a sampling frequency of 2 data points per second. This curve is synchronized with the sub-window anomaly index curve in the feature parameter set on the time axis, and a bivariate correlation model is established using sliding window correlation coefficient analysis.

[0035] The mapping table is constructed using a multi-column data structure for physical storage. The primary key field is a timestamp sequence with millisecond precision. The core fields contain five types of information: device level code using a 6-digit alphanumeric identifier; concurrent operation frequency value recording the number of operations after topology weighting; anomaly index change trend storing differentially encoded data, i.e., the change in the anomaly index between adjacent time points; operation type distribution vector recording the proportion of each type of operation in the current period; and topology weight value archiving the normalized result of the final calculation. Each row of data corresponds to aggregated information within a 250-millisecond time unit. The table structure supports dynamic column expansion; when a new operation type is added, a new column field is automatically added to store the new type of data.

[0036] The relationship establishment process involves multiple verification steps. The initially constructed mapping table undergoes a trend consistency check, verifying that the time-series offset between the peak operation frequency and the peak anomaly index is within a 200-millisecond tolerance range. Data rows exceeding the offset limit trigger an automatic compensation mechanism: dynamically calibrating the operation timestamps based on a device response latency characteristic library. The final generated mapping table is appended with a version control tag, recording the creation time and data source hash value, and achieves efficient read and write operations through the partitioned storage mechanism of a relational database. A global snapshot of the table data is generated hourly and persistently stored in a distributed file system for historical traceability and analysis.

[0037] The semantic parsing module processes natural language descriptions from key operation record fragments in real time. It employs an electrical engineering dictionary for entity recognition, extracting verb sets for operational actions and noun sets for operational objects. The verb sets include eight categories, such as switching, regulating, and changing, while the noun sets distinguish between equipment parts and system components. The parsing results are converted into standardized operational semantic codes and embedded into the semantic feature columns of a mapping table. The reconstruction of the operation sequence generates operation chains based on timestamp sorting, and the chain structure is stored in a separate graph database to maintain the dependencies between equipment operations. This design preserves the complete context of the operational behavior, providing supplementary semantic explanations for the mapping relationships.

[0038] The data processing workflow includes an anomaly protection mechanism. When there are gaps in the timestamps of the log text, the system calls the device clock to synchronize the records and perform interpolation to repair them. An alarm mechanism is triggered for abnormal operation durations, and records exceeding the preset maximum duration require manual review and confirmation. All intermediate result data are tagged with data quality labels, marking completeness, accuracy, and consistency scores for subsequent module decision-making reference. The final generated mapping table provides query services through an API interface, supporting time range retrieval and hierarchical dimensional slicing analysis.

[0039] Example 3: See Figure 4 The process of correcting the parameters of the real-time load prediction model begins with data preprocessing of the mapping table. The system extracts the device-level location code and the corresponding historical load baseline data from the table. The baseline data is stored in a distributed time-series database, containing load statistics for the same period over the past 365 days. The current real-time load change time series is collected through a sliding window with a window length of 15 minutes, a sliding step size of 1 minute, and a sampling frequency of 4 times per second for data points within the window. The difference vector calculation between the real-time time series and the historical baseline uses a multi-scale analysis method, unfolding in both the time and frequency domains. The time-domain difference calculation calculates the absolute deviation between the load value and the baseline value at each sampling point, while the frequency-domain difference is extracted using a fast Fourier transform to extract the amplitude difference of the main fluctuation components.

[0040] The correlation analysis between the difference vector and the trend of anomaly index changes employs an improved mutual information algorithm. Let the difference vector be... The abnormal index change trend sequence is Then the correction factor The calculation is as follows:

[0041] in: Indicates the difference value The function undergoing nonlinear transformation is activated using hyperbolic tangent activation. Indicates the value of abnormal index change The weighted processing function has weights that depend on the importance factor of the device's hierarchical location. This represents the total number of data points within the sliding window. Index the data points. Calculation results. The range is between -1 and 1, with positive values ​​indicating a positive correlation between the difference vector and the abnormal trend.

[0042] Model parameter updates are performed in the temporal feature extraction layer of the load prediction neural network. This network employs a long short-term memory architecture and contains 128 hidden units. Correction coefficients are injected into the network via a gating mechanism: the input gate receives... As an auxiliary feature, the forgetting gate is based on The symbol is adjusted to determine the proportion of historical memory retained. Network parameter updates employ a gradient descent method with momentum, and the learning rate increases accordingly. The absolute value of the parameters is dynamically adjusted. After each parameter update, the system records the change in the Frobenius norm of the weight matrix before and after the correction, which serves as an indicator for monitoring model stability.

[0043] The load distribution forecast for the next three cycles employs a multi-step rolling forecasting strategy. In the first cycle, the updated model is directly called for a single forward calculation; in the second and third cycles, an iterative forecasting method is used, with the output of the previous cycle serving as the input for the next. The forecast results include the expected load value for each power supply node, the upper and lower bounds of the 95% confidence interval, and a fluctuation sensitivity score. The maximum load percentage of a node is calculated using the moving percentile method, taking the 99th percentile value within a 15-minute window as the representative value.

[0044] The emergency power supply classification is based on a dynamic clustering algorithm. Inputting the maximum load percentage data of all nodes, the number of cluster centers is automatically determined through density peak detection. Class boundaries are set based on the natural dividing points of the clustering results, typically generating 3 to 5 discrete classes. Each class is labeled with a color code and a response priority value, forming a class mapping dictionary. The hierarchical power supply weight allocation scheme is calculated using the equipment hierarchy data in the mapping table. The baseline weight for the backbone line level is set to 2.0, for the distribution cabinet level to 1.5, and for the terminal equipment level to remain unchanged at 1.0. The weight values ​​are then exponentially smoothed based on operation frequency data to generate the final allocation coefficients.

[0045] The dynamic adjustment strategy is generated using a multi-objective optimization framework. The optimization objectives include three dimensions: nodes with higher power supply emergency levels receive more resource allocation; devices with larger hierarchical weight coefficients maintain higher power supply stability; and overall network losses are controlled within a reasonable range. Constraints include physical limitations such as transformer capacity limits and line transmission limits. The solution process employs a genetic algorithm with elite retention, with a population size of 100 and 50 generations. The final output strategy is represented as a set of instructions in JSON format, including the target voltage setting, maximum allowable load rate, and backup power switching conditions for each power supply node.

[0046] The strategy verification process is performed in a virtual simulation environment. The system loads the equivalent circuit model of the current network topology, injects strategy instructions, and runs a real-time digital simulation. The simulation step size is set to 1 millisecond, simulating an operation process for 15 minutes. Quality indicators such as voltage deviation rate and current harmonic distortion rate of each node are monitored during the simulation. When safety limits are exceeded, a strategy rollback mechanism is triggered. Validated strategies are marked as executable and pushed to the control center's standby queue. The average execution time of the entire correction and prediction process is 8 seconds, meeting the timeliness requirements of real-time power supply system control.

[0047] The data processing pipeline incorporates multiple fault-tolerant designs. During the difference vector calculation phase, linear interpolation is automatically used to fill in gaps in the real-time data stream. During model parameter updates, if gradient explosion is detected, a more conservative learning rate strategy is immediately switched. Prediction results undergo a range reasonableness check before publication; predictions exceeding 200% of the equipment's rated capacity are automatically discarded and recalculated. All intermediate calculation results are accompanied by data quality labels, recording metadata such as the proportion of missing values ​​and the number of outliers, for reference and decision-making in subsequent stages.

[0048] A complete audit log for system maintenance and correction operations is maintained. Each parameter update records the following information: trigger timestamp, version of the mapping table used, correction coefficient value, name of the affected neural network layer, and hash value of the prediction result. The log data is compressed and stored in a blockchain structure, forming an immutable historical record of operations. The audit interface supports querying correction records by time range and reproducing the model parameter state at any given time.

[0049] Example 4: See Figure 5 The implementation process of triggering automatic calibration commands for power supply equipment operating parameters is illustrated using the real-time control of a regional power distribution network as an example. This region includes one 110kV substation, eight 10kV feeders, and 53 distribution transformers. A dynamic adjustment strategy has been generated and simulated for verification. The system first parses the target level setting in the strategy document. In this example, the "main line level" is identified as the current key control target, with the emergency level threshold set to level three (corresponding to a load rate range of 85%-92%). The control center retrieves the operating parameter dataset of 12 key devices at this level from the real-time database, including 18 monitoring indicators such as current, voltage, and power factor, with a sampling interval of 200 milliseconds.

[0050] Equipment selection was based on the current stability index, which is calculated as follows: the standard deviation of the effective current value within a 10-second sliding window is divided by the mean value of the window to obtain the relative fluctuation coefficient. Table 1 shows the comparison of real-time parameters and threshold values ​​for some equipment.

[0051] Table 1: Comparison of real-time parameters and thresholds for some devices.

[0052] Equipment Number Current load rate (%) Current stability Voltage deviation (%) Power factor Deviation value determination DL-4108 89.2 0.15 +3.1 0.93 Exceeding limits DL-4112 86.7 0.09 -1.8 0.95 normal DL-4115 91.4 0.21 +4.5 0.88 Exceeding limits DL-4120 83.9 0.07 -0.6 0.96 normal

[0053] The deviation value judgment column in the table is generated according to the tolerance rules of the Level 3 emergency level: the current stability threshold is set to 0.12, the voltage deviation limit is ±3%, and the power factor requirement is greater than 0.9. The DL-4108 device is marked as a priority device due to excessive current stability and positive voltage deviation. The system generates a customized calibration instruction set for each device exceeding the limits. For example, the instructions for the DL-4108 device include three parallel operations: adjusting the on-load tap changer to the -2 position, activating two sets of reactive power compensation capacitors, and limiting the load on the downstream branch 3 to no more than 80% of the rated value.

[0054] The command execution process adopts a hierarchical progressive control mode. The master station system first issues a preparatory command to put the target equipment into a control-ready state. After the field intelligent terminal returns a readiness confirmation signal, the master station sends an execution command containing specific parameter adjustment values. Taking voltage regulation as an example, the system does not directly adjust the voltage setpoint from the current 10.5kV to the target 10.2kV, but adopts a step-by-step approach: the first adjustment is to 10.4kV, and after monitoring for two sampling cycles (400 milliseconds) to confirm that there are no abnormalities, the remaining 0.2kV adjustment is then completed. This segmented regulation effectively avoids malfunctions of equipment protection devices.

[0055] Response data acquisition begins immediately after command execution. For the DL-4115 equipment control case, the system recorded the following typical response sequence: 300 milliseconds after the reactive power compensation capacitor was connected, the power factor increased from 0.88 to 0.92; 800 milliseconds after voltage regulation was completed, the current stability decreased from 0.21 to 0.13. The voltage recovery rate was calculated using a ramp detection method to identify the time span from the 90% regulation starting point to entering the ±1% target range of the voltage curve. The current fluctuation convergence time was defined as the duration from the completion of the last parameter adjustment until the fluctuation coefficient of five consecutive sampling points was below the threshold of 0.12.

[0056] The eigenvalue feedback mechanism employs an incremental update strategy. The characteristic parameters of the DL-4108 equipment's control case (voltage recovery rate 1.2kV / s, current convergence time 2.4s) are converted to a standardized format: the voltage recovery rate is quantized in units of 0.5kV / s, and the current convergence time is discretized in 0.5s intervals. These values ​​are appended to the "Equipment Response Characteristics" field of the characteristic parameter set, while simultaneously updating the typical response mode library for this equipment model. The system automatically establishes an association index between the current control parameters and historical records; for example, when the same model of equipment experiences a similar out-of-limit situation again, the system prioritizes recommending the parameter combination that has been validated in this case.

[0057] The anomaly handling module monitors the entire control process. When an abnormal harmonic amplification phenomenon is detected in the DL-4115 device during voltage regulation (total harmonic distortion suddenly increases to 8.7%), the system immediately suspends the original regulation plan and switches to the backup strategy: withdrawing the reactive power compensation command and gradually reducing the transformer load rate. This anomaly is recorded as a special case, triggering an emergency expansion process for the characteristic parameter set, adding a "harmonic sensitivity" monitoring dimension to predict harmonic impacts in similar scenarios in the future.

[0058] The version control system maintains a complete change history of equipment parameters. After each calibration command is executed, a version record is generated containing the following elements: a snapshot of the parameters before regulation, a detailed list of the command content, the parameter status after regulation, and the feature value extraction results. These records are stored using blockchain technology, forming an immutable chain of regulatory evidence. The query interface supports searching historical operations by multiple dimensions such as equipment number, time range, and regulation type. For example, it can trace the detailed parameter change curves of all voltage regulation operations of a transformer over the past 30 days.

[0059] Data quality control is implemented throughout the entire process. Raw monitoring data undergoes three checks: sampling time continuity (timestamp interval error not exceeding ±10 milliseconds), numerical range reasonableness verification (current value not exceeding 120% of the equipment nameplate value), and data integrity check (each frame of data contains all 18 categories of indicators). Intermediate calculation results, such as derived indicators like current stability, require reverse verification against the raw monitoring values ​​to ensure that the calculation logic does not introduce abnormal deviations. The final output calibration command set must pass a semantic consistency check to confirm that the mathematical relationship between the command parameters and the target adjustment amount conforms to the physical characteristics of the equipment.

[0060] The collaboration between field devices and the master station system employs an adaptive communication mechanism. For remote sites with high network latency, command data packets are automatically compressed and transmission frequency is reduced; critical backbone nodes maintain millisecond-level real-time communication. Each command transmission includes a data checksum, and the receiving end unlocks device operation permissions only after verification. Emergency handling plans for communication interruptions include: locally caching the most recently valid commands, automatically switching to a preset security mode, and triggering the device's autonomous steady-state maintenance mechanism to ensure the system continues to operate safely during control interruptions.

[0061] The feedback optimization process forms a closed-loop control. By injecting characteristic parameters from successful control cases of the DL-4108 equipment into the real-time analysis model, the system's sensitivity to identifying similar abnormal patterns improves by approximately 40% in the next control cycle. The dynamic expansion mechanism of the characteristic parameter set continuously incorporates new control experiences, such as adding a "composite abnormal response mode" category to record the optimal control strategy when voltage deviation and harmonic exceedances coexist. This continuously evolving knowledge base enables the system to adapt to new abnormal operating conditions.

[0062] Example 5: When establishing a version-associative index between the power supply node monitoring dataset and the equipment status log, a distributed database architecture is used to design the index table structure. The index table includes primary key fields: the network topology version number is a 12-bit code composed of year, month, day, and serial number; the data acquisition timestamp is accurate to the millisecond level; and the equipment type code is extended to a 5-digit alphanumeric identifier based on international classification standards. Index records generate unique identifiers using a hash algorithm, establishing a bidirectional mapping relationship with the historical anomaly event database. The timeline alignment mechanism uses a GPS clock synchronization signal as a reference; a time calibration procedure is automatically triggered when the topology version changes, with the deviation controlled within 50 milliseconds.

[0063] The immediate processing flow for adding a new power supply node to the network comprises four operational stages. When the node activation command is issued, the registration service first obtains the node's device model and technical specifications. A new entry is created in the indexing system, assigning an initial version number that is the current network topology version plus a ".1" suffix. The monitoring data acquisition module initiates a specific configuration program: the current monitoring channel is set to 16-bit resolution mode, and the voltage sampling frequency is increased from the standard 100Hz to 1kHz to capture initial fluctuation characteristics. The logging system creates a dedicated text storage area for this node, and the operation record timestamps and monitoring data timestamps are synchronized using an atomic clock.

[0064] The historical anomaly event database matching dimension expansion process employs a layered injection strategy. The device model of a newly added node activates a preset dimension expansion template, which includes three types of dimension slots: electrical characteristic slots inject current fluctuation pattern and voltage offset range fields; structural characteristic slots add physical connection parameters and impedance matching coefficient fields; and operational characteristic slots expand with typical operational response curve slots. The dimension expansion operation uses a dual-thread parallel mechanism: the main thread loads the template framework, while the auxiliary thread populates the first batch of 72 hours of operational data collected from the new node in real time. Feature extraction is performed during the population process; for example, before entering the current fluctuation pattern field, the matching degree of 10 standard waveform patterns needs to be calculated, retaining waveform features with a matching degree higher than 60%.

[0065] During the policy verification phase, extended dimensions are loaded for compatibility checks. The verification environment constructs a digital twin model of the current network topology, injecting the extended dimension data of the newly added nodes into the corresponding node instances during initialization. Boundary condition settings include two typical scenarios: a peak load scenario where the load rate of the new node is preset to 85% of its rated capacity, and a fault scenario simulating a short-term disconnection of its upstream equipment. The dynamic policy parser preloads the dimension data structure of the newly added nodes into the memory cache before processing policy instructions. During policy instruction execution, three special verification points are accessed: checking whether the newly added fluctuation mode feature is invoked during node load allocation coefficient calculation; checking whether the typical response curve is read during device operation sequence generation; and verifying whether the impedance matching coefficient is referenced during power supply priority sorting.

[0066] A dynamic monitoring protocol is implemented during the verification process. During each policy calculation, the frequency of access to the newly added dimensions and the amount of data used by the core judgment logic are recorded. Unused dimension fields are marked with a yellow warning icon, and frequently accessed dimension fields are marked with a special monitoring flag. During boundary scenario testing, a video stream of the electrical response of the new node under the policy is recorded simultaneously. The video stream includes visual analysis data such as voltage waveforms and current harmonic spectra. After the policy terminates, a verification report is generated, which includes statistical charts of the usage rate of the newly added dimensions, a list of inactive dimensions, and a summary index of the boundary scenario test videos.

[0067] The distributed database maintenance mechanism includes an automated cleanup procedure. Index defragmentation is performed every 24 hours, merging duplicate operation records on the same node. When the volume of a single node's associated index exceeds 500MB, a dimension data compression procedure is triggered: retaining detailed runtime data from the most recent 30 days, while older data is converted to feature summary format for storage. Index backup employs a differential replication strategy, transmitting only incremental data after version changes. Historical rollback functionality is supported during network reconstruction, allowing restoration of dimension configuration states at any given time by specifying a historical version number.

[0068] The newly added node data transmission pipeline is equipped with multiple filtering mechanisms. The raw monitoring data stream passes through a signal quality filter to eliminate spikes caused by sampling anomalies; the log text stream passes through a sensitive word filter to block non-technical descriptive content. Data verification checkpoints are set up during the dimension population process: the current fluctuation pattern field must pass a steady-state retention test before entry to ensure that the fluctuation characteristics remain effective during 12 hours of continuous operation; curve smoothness is checked during typical response curve slot injection to remove invalid records containing high-frequency noise. These technical measures ensure that the data quality of the expanded dimensions meets the policy verification requirements.

[0069] The exception event database employs a time-driven, two-tiered archiving structure. Currently valid dimensions are stored on a solid-state drive array to ensure read / write efficiency, while historical versions of dimensions are transferred to a tape library for long-term storage. The archiving process records the complete operation chain: the entire lifecycle from new node access application to dimension freezing, including timestamps for each version change, operator identification, and a summary of the modified content. The query service supports cross-version retrieval; for example, it can compare the voltage offset range variation trend of newly added nodes in different topology versions, and the search results are output as a time-stamped sequence of dimension data.

[0070] The version-related index service interface provides dimensional traceability functionality. Inputting any policy execution timestamp, the interface returns the complete dimensional data structure effective at that moment. When analyzing historical policy execution records, this feature ensures accurate reconstruction of the computational environment at that time. The version evolution tree of the dimensional data is visualized, clearly showing the dimensional expansion path caused by each network change. This design maintains the centralized management method of the power supply system's continuous adaptability to new device access scenarios, achieving dynamic consistency of management policies.

[0071] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A centralized management method for a power supply system, characterized in that, Includes the following steps: The system monitors the current fluctuation sequence and voltage offset data of each node in the power supply network during continuous operation cycles, and collects the status log text information and load change timing information of the power supply equipment. It identifies the start and end times of abnormal changes in the current fluctuation sequence, and extracts a set of characteristic parameters for power quality degradation events by combining the voltage offset data with the magnitude of changes within the corresponding time window. Based on the set of characteristic parameters, it matches related event patterns in the historical abnormal event database to locate key operation record segments in the power supply equipment status log text that are strongly correlated with the load change timing. It then performs semantic parsing and operation timing alignment processing on these key operation record segments to generate a mapping table between equipment operation behavior and power quality fluctuations. Based on the mapping table, the prediction model parameters for real-time load change timing are corrected, and a dynamic power supply priority adjustment strategy is output. The power supply priority dynamic adjustment strategy is executed, triggering an automatic calibration command for the operating parameters of the power supply equipment.

2. The centralized management method for a power supply system according to claim 1, characterized in that, The extraction of the feature parameter set for power quality degradation events includes: dividing the time interval from the start time to the end time of the abnormal change into multiple continuous detection sub-windows; calculating the product of the current fluctuation range and the average voltage deviation in each sub-window, and calculating the deviation of the product value from a preset safety threshold; detecting the time sequence overlap between the load change rate in each sub-window and the switching actions in the device status log; merging the deviation and the time sequence overlap to generate a sub-window anomaly index; and aggregating all sub-window anomaly indices to form a feature parameter set.

3. The centralized management method for a power supply system according to claim 2, characterized in that, The key operation record segments in the power supply equipment status log text that are strongly correlated with the load change time sequence include: traversing the timestamps of all operation instruction records in the power supply equipment status log text; filtering operation instruction texts whose timestamps fall within the time interval corresponding to the feature parameter set; extracting the device number, operation type code, and operation duration from the filtered operation instruction texts; performing similarity matching between the operation duration and the duration of the sub-window anomaly index in the feature parameter set; and retaining operation instruction texts with similarity exceeding the matching threshold as key operation record segments.

4. The centralized management method for a power supply system according to claim 3, characterized in that, The mapping table between generated device operation behavior and power quality fluctuations includes: performing topological relationship analysis on device numbers in key operation record segments to determine the hierarchical position of the device in the power supply network; classifying and statistically analyzing the concurrent operation frequency of devices at the same level according to operation type codes; establishing a correspondence between the concurrent operation frequency and the numerical change trend of the sub-window anomaly index in the feature parameter set; and integrating the device hierarchical position, concurrent operation frequency, and anomaly index change trend into a mapping table.

5. The centralized management method for a power supply system according to claim 4, characterized in that, The steps for correcting the prediction model parameters based on the mapping table include: calling the historical load baseline data corresponding to the device level position in the mapping table; calculating the difference vector between the current real-time load change time series and the historical load baseline; generating model parameter correction coefficients based on the correlation between the difference vector and the abnormal index change trend in the mapping table; and injecting the correction coefficients into the time series feature extraction layer of the load prediction model for parameter updates.

6. The centralized management method for a power supply system according to claim 5, characterized in that, The dynamic adjustment strategy for output power supply priority includes: obtaining the load distribution prediction values ​​for the next three periods output by the updated load prediction model; classifying the power supply emergency level according to the maximum load ratio of each node in the load distribution prediction value; generating a hierarchical power supply weight allocation scheme by combining the association rules between device hierarchical position and operation frequency in the mapping relationship table; and generating a dynamic adjustment strategy by integrating the power supply emergency level and the hierarchical power supply weight allocation scheme.

7. The centralized management method for a power supply system according to claim 6, characterized in that, The automatic calibration instructions for triggering the operating parameters of the power supply equipment include: parsing the target power supply level and emergency level threshold in the dynamic adjustment strategy; matching the real-time operating parameter dataset of all power supply equipment within the target power supply level; comparing the deviation between the current stability index in the operating parameter dataset and the emergency level threshold; and generating a set of calibration instructions for equipment whose deviation exceeds the tolerance range.

8. The centralized management method for a power supply system according to claim 7, characterized in that, The method further includes: collecting device operation response timing data after executing calibration instructions; extracting feature values ​​of voltage recovery rate and current fluctuation convergence time from the response timing data; and feeding the feature values ​​back to the feature parameter set of power quality degradation events for incremental updates.

9. The centralized management method for a power supply system according to claim 1, characterized in that, The method further includes: establishing a version association index between the power supply node monitoring dataset and the device status log; when a new power supply node is added to the network, synchronously expanding the matching dimensions of the historical abnormal event database based on the version association index; and loading the expanded matching dimensions to verify the strategy when outputting the dynamic adjustment strategy for power supply priority.

10. A centralized management system for a power supply system, characterized in that, include: The power quality monitoring module is used to collect real-time current fluctuation sequences, voltage offset data, and equipment status log text of power supply nodes; An anomaly feature extraction module, connected to the power quality monitoring module, is used to identify the feature parameter set of power quality degradation events and generate a sub-window anomaly index; an operation behavior analysis module, connected to the anomaly feature extraction module, is used to locate key operation record segments and analyze the device level position; a mapping relationship construction module, connected to the operation behavior analysis module, is used to generate a mapping relationship table between device operation behavior and power quality fluctuations; a prediction model update module, connected to the mapping relationship construction module, is used to correct the load prediction model parameters; and a strategy generation module, connected to the prediction model update module, is used to output a dynamic adjustment strategy for power supply priority. The instruction execution module, connected to the strategy generation module, is used to trigger automatic calibration instructions for the operating parameters of the power supply equipment; the feedback optimization module, connected to the instruction execution module and the abnormal feature extraction module, is used to incrementally update the feature parameter set; and the data version management module, connected to all modules, is used to maintain the version association index of the power supply node monitoring data.

Images