A Fault Early Warning Method and System for Power Supply and Distribution Systems Based on Multi-parameter Trend Analysis

Abstract

This application provides a fault early warning method and system for power supply and distribution systems based on multi-parameter trend analysis, relating to the field of power monitoring. The method includes: collecting multi-parameter time-series data of the power supply and distribution system and classifying it by operating environment to obtain a classified dataset. A sliding window algorithm is used to extract the time-series features of the classified dataset, and linear regression is used to calculate the slope of the parameter change trend within each window to obtain trend feature data. Based on the trend feature data, abnormal trends are identified and abnormal indicators and associated parameter sets are generated. Fault location analysis and temperature rise coefficient calculation are performed on the abnormal indicators and associated parameter sets, and fault state identification is performed in conjunction with short-term environmental data to generate fault location analysis results. Based on the fault location analysis results, an early warning signal is generated. This method solves the technical problems of delayed fault early warning and low location accuracy caused by the lack of analysis of multi-parameter change trends and the failure to exclude environmental interference.

Description

Technical Field

[0001] This application relates to the field of power monitoring, and in particular to a method and system for early warning of faults in power supply and distribution systems based on multi-parameter trend analysis. Background Technology

[0002] In the current era of rapid development in the power industry, power supply stability has become a core foundation for ensuring industrial production, commercial operations, and residential electricity use, directly impacting the normal operation and safe use of various electronic and electrical equipment. Currently, the mainstream technical means for maintaining power supply stability in the industry fall into two categories: one is through a switching mechanism between main and backup power supplies, enabling emergency switching when power supply is interrupted or fluctuations exceed thresholds to ensure power continuity; the other is through periodic maintenance and inspection, using manual or simple monitoring tools to periodically inspect power distribution equipment and identify potential faults. However, in actual use, early warning detection often focuses on judging the instantaneous value of a single parameter, failing to consider the temporal changes of multiple parameters such as current, voltage, and temperature. This makes it difficult to identify potential gradual faults in their early stages, resulting in overall delayed early warnings. Furthermore, when judging based on detection data, it fails to consider differences in operating conditions and environmental interference, as well as the significant differences in equipment operating characteristics under different load levels, time periods, and ambient temperatures. Additionally, it fails to contextualize the data, making fault judgment susceptible to interference and leading to a high rate of misjudgment and missed judgment. Summary of the Invention

[0003] This application provides a fault early warning method and system for power supply and distribution systems based on multi-parameter trend analysis, which solves the technical problems of existing technologies that lack analysis of the changing trends of multiple parameters and fail to exclude interference from environmental factors, resulting in delayed fault early warning and low positioning accuracy.

[0004] To achieve the above objectives, this application adopts the following technical solution:

[0005] Firstly, a fault early warning method for power supply and distribution systems based on multi-parameter trend analysis includes: collecting multi-parameter time-series data of the power supply and distribution system, classifying it by operating environment, and obtaining a classified dataset. A sliding window algorithm is used to extract the time-series features of the classified dataset, and linear regression is used to calculate the slope of the parameter change trend within each window to obtain trend feature data. Based on the trend feature data, abnormal trends are identified and abnormal indicators and associated parameter sets are generated. Fault location analysis and temperature rise coefficient calculation are performed on the abnormal indicators and associated parameter sets, and fault state identification is performed in conjunction with short-term environmental data to generate fault location analysis results. Based on the fault location analysis results, an early warning signal is generated.

[0006] In conjunction with the first aspect mentioned above, one possible implementation involves collecting multi-parameter time-series data of the power supply and distribution system, classifying it by operating environment bins, and obtaining a classified dataset. Specifically, this includes: collecting multi-parameter time-series data of the power supply and distribution system, and dividing the data into multiple operating environment bins according to preset operating environment bin classification rules. Each operating environment bin is uniquely defined by a combination of three dimensions: load level, time period, and ambient temperature. The load level is classified based on the effective current value into low load, medium load, high load, and impact load. The divided multi-parameter time-series data is then stored according to the corresponding operating environment bins, forming a classified dataset. Each dataset contains parameter sequences for all time points within the same bin.

[0007] In conjunction with the first aspect mentioned above, one possible implementation involves calculating the slope of the parameter change trend within each window using linear regression to obtain trend characteristic data. Specifically, this includes: pre-setting the size and step size of the sliding window based on the data acquisition frequency and trend analysis accuracy requirements, with the step size being less than or equal to the window size; extracting multi-parameter time-series data points within each sliding window from the dataset, where each data point includes a timestamp and its corresponding parameter value; performing linear regression fitting on the parameter values ​​and timestamps within each sliding window, and calculating the slope of the regression line using the least squares method.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the process of determining abnormal trends and generating abnormal indicators and associated parameter sets based on trend feature data specifically includes: setting an abnormal trend judgment threshold based on historical normal operation data; filtering for trends where the absolute value of the slope is greater than the abnormal trend judgment threshold and multiple consecutive sliding windows meet the condition, defining them as abnormal trends and generating abnormal indicators; extracting the parameter set associated with the abnormal trend, and encapsulating the abnormal indicator and associated parameter set into structured data output, wherein the associated parameter set includes the average current, average voltage, average temperature, and current-temperature correspondence sequence within the abnormal window.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the process of fault location analysis and temperature rise coefficient calculation based on the anomaly flags and associated parameter sets specifically includes: locating the target electrical equipment corresponding to the anomaly trend based on the average current, average voltage, and average temperature values ​​in the associated parameter set, combined with the power supply and distribution system topology diagram; obtaining the temperature and current sequences of the main power components of the target electrical equipment within the anomaly window; calculating the temperature rise per unit time by linearly fitting the slope of the temperature sequence relative to time; and calculating the temperature rise coefficient based on the temperature rise and current sequence.

[0010] In conjunction with the first aspect mentioned above, one possible implementation involves identifying fault states and generating fault location analysis results by combining short-term environmental data. Specifically, this includes: acquiring the ambient temperature sequence within a preset short-term time window and calculating the slope of the ambient temperature change over time through linear fitting to obtain an ambient temperature rise compensation value; using the ambient temperature rise compensation value to calculate the temperature rise coefficient, obtaining the compensated temperature rise coefficient; filtering data where the compensated temperature rise coefficient is greater than a preset fault threshold, recording them as fault states, and generating a fault state flag; and based on the fault state flag and the power supply and distribution system topology diagram, generating fault location analysis results, which include fault equipment identification, fault type, and warning level.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the process of generating an early warning signal based on the fault location analysis results specifically includes: matching response strategies from a predefined early warning response strategy library according to the early warning level and fault type in the fault location analysis results; generating multi-level early warning signals based on the matched response strategies, including local visual early warning signals, local auditory early warning signals, and remote notification signals; and adjusting the output frequency and duration of the early warning signals based on the early warning level and temperature rise coefficient compensation value.

[0012] Secondly, a fault early warning system for power supply and distribution systems based on multi-parameter trend analysis is provided, including: a data acquisition and classification module, which collects multi-parameter time-series data of the power supply and distribution system, classifies it by operating environment, and obtains a classified dataset; a trend feature extraction module, which uses a sliding window algorithm to extract the time-series features of the classified dataset, calculates the slope of the parameter change trend within each window through linear regression, and obtains trend feature data; a fault probability prediction module, which judges abnormal trends based on the trend feature data and generates abnormal signs and associated parameter sets; fault location analysis and temperature rise coefficient calculation are performed on the abnormal signs and associated parameter sets, and fault state identification is performed in conjunction with short-term environmental data to generate fault location analysis results; and an early warning module, which generates early warning signals based on the fault location analysis results.

[0013] In conjunction with the second aspect mentioned above, in one possible implementation, the trend feature extraction module specifically includes: a window data partitioning unit, which presets the size and sliding step of the sliding window based on the data collection frequency and trend analysis accuracy requirements, wherein the sliding step is less than or equal to the window size; extracting multi-parameter time-series data points within each sliding window from the dataset, wherein the data points include timestamps and corresponding parameter values; and a linear regression fitting unit, which performs linear regression fitting on the parameter values ​​and timestamps within each sliding window and calculates the slope of the regression line using the least squares method.

[0014] In conjunction with the second aspect mentioned above, in one possible implementation, the fault probability prediction module specifically includes: a suspect equipment location unit, which locates the target electrical equipment corresponding to the abnormal trend based on the average current, average voltage, and average temperature values ​​in the associated parameter set, combined with the power supply and distribution system topology map; a data acquisition unit, which acquires the temperature sequence and current sequence of the main power components of the target electrical equipment within the abnormal window; a parameter calculation unit, which calculates the temperature rise per unit time by linearly fitting the slope of the temperature sequence relative to time, and calculates the temperature rise coefficient based on the temperature rise and current sequence; an ambient temperature sequence within a preset short-term time window, and calculates the slope of the ambient temperature change over time by linear fitting to obtain the ambient temperature rise compensation value; a compensation calculation of the temperature rise coefficient using the ambient temperature rise compensation value to obtain the compensated temperature rise coefficient; a screening and analysis unit, which filters data whose compensated temperature rise coefficient is greater than a preset fault threshold, records them as fault states, and generates a fault state flag; and a fault location analysis result based on the fault state flag and the power supply and distribution system topology map, which includes the fault equipment identifier, fault type, and warning level.

[0015] This application provides a fault early warning method and system for power supply and distribution systems based on multi-parameter trend analysis. It can classify multi-parameter time-series data into structures by categorizing operating conditions according to three dimensions: load level, time period, and ambient temperature, thus solving the problem of analysis distortion caused by mixed data from different operating conditions. Simultaneously, it employs a sliding window algorithm combined with linear regression to calculate the slope of parameter changes, accurately capturing the trend characteristics of time-series data and overcoming the difficulty of quantifying the rate of parameter change in traditional methods, enabling early identification of abnormal trends. Furthermore, it filters continuous abnormal windows using historical thresholds and extracts associated parameter sets, providing clear data support for fault analysis and reducing ineffective troubleshooting. Subsequently, it locates target equipment using the system topology map, calculates the temperature rise coefficient, and introduces an environmental compensation mechanism to eliminate environmental interference, solving the problems of inaccurate fault location and easy misjudgment, and accurately identifying faulty equipment and types. Finally, it generates multi-level early warning signals and dynamically adjusts output parameters to ensure timely transmission of early warning information, solving the pain point of delayed response in traditional passive maintenance, improving operation and maintenance efficiency, and reducing power outage losses caused by faults.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0018] Figure 2 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0019] Figure 3 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0020] Figure 4 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0021] Figure 5 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0022] Figure 6 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0023] Figure 7 A flowchart illustrating the fault early warning method for power supply and distribution systems based on multi-parameter trend analysis provided in this application embodiment;

[0024] Figure 8 The system architecture diagram of the power supply and distribution system fault early warning system based on multi-parameter trend analysis provided in the embodiments of this application is shown. Detailed Implementation

[0025] like Figure 1As shown in the embodiments of this application, the power supply and distribution system fault early warning method based on multi-parameter trend analysis includes:

[0026] Step 101: Collect multi-parameter time-series data of the power supply and distribution system, classify the operating conditions and environment, and obtain the classified dataset.

[0027] Among them, the operating environment classification refers to dividing the collected multi-parameter time-series data into different predefined categories based on various operating conditions such as load level, time period, and ambient temperature during the operation of the power supply and distribution system, thereby realizing contextualized data management and analysis. Multi-parameter time-series data includes electrical parameters such as current, voltage, and temperature that are continuously collected over time.

[0028] In some implementations, various sensors deployed in the power supply and distribution network collect real-time time-series data on multiple parameters such as current, voltage, and temperature, which can then be stored chronologically. Based on pre-defined operating condition classification rules, the data collected at each moment is categorized into corresponding operating condition categories, resulting in multiple classified datasets. Each dataset corresponds to a specific operating condition, providing a structured data foundation for subsequent trend feature extraction and anomaly analysis under the same operating conditions.

[0029] The pre-defined operating environment classification rules include: based on historical operating data and expert experience, such as classifying load levels into low, medium, high, and shock loads; dividing time periods into nighttime and daytime; and classifying ambient temperatures into low temperature, normal temperature, and high temperature. These dimensions are combined to form multiple specific operating environment bins, such as a "low load - nighttime - low temperature" bin. This allows the system to automatically match the data to the corresponding operating environment bin category based on real-time collected load values, time information, and temperature measurements, and then categorize the data for that specific time into the dataset corresponding to that bin.

[0030] It should be noted that the division of the power supply system based on the working environment can be adjusted and optimized according to the actual application scenario. For example, in the power supply system of an industrial park, factors such as production shifts and seasonal load changes can be further considered.

[0031] For example, in a supermarket power supply system, current, voltage, and ambient temperature data are continuously collected from the transformer output side. At 2:00 PM on a certain day, the system measures a load rate of 85% and an ambient temperature of 32°C. This period falls within daytime operating hours, and according to preset rules, this data will be categorized into the "High Load - Daytime - High Temperature" sub-database. As time progresses, all data collected under similar operating conditions will be grouped into the same dataset. This allows for analysis of current or voltage trends within this dataset, unaffected by data from low-load or low-temperature nighttime conditions, thus more accurately identifying potential abnormal trends under specific operating conditions.

[0032] Step 102: Use the sliding window algorithm to extract the temporal features of the classified dataset, and use linear regression to calculate the slope of the parameter change trend within each window to obtain trend feature data.

[0033] The sliding window algorithm is a commonly used technique for processing time series data. It defines a fixed-size window and slides it across the time series at preset steps to sequentially extract subsequences of data within the window for analysis.

[0034] In some implementations, the size of the sliding window (e.g., 60 data points, representing a 1-minute time span) and the sliding step size (e.g., 10 data points, representing a window that slides once every 10 seconds, with the step size being smaller than the window size to ensure data overlap and continuity) are preset based on the data acquisition frequency (e.g., once per second) and the desired trend analysis accuracy (e.g., to detect trend changes at the minute level).

[0035] Then, from the dataset categorized by operating conditions, for each specified operating condition bin (e.g., "High Load - Daytime - High Temperature"), extract the timestamps and their corresponding parameter values ​​(e.g., RMS current values) stored within that bin. Starting from the beginning of the sequence, extract all data points (timestamps and corresponding parameter values) within the first window. Apply a linear regression model to these data points, using timestamps as independent variables and parameter values ​​as dependent variables. Calculate the slope and intercept of the optimal fitted line using the least squares method, where the slope represents the trend of parameter change within that window.

[0036] After the calculation is completed, the window moves forward according to a preset sliding step size, repeating the above data extraction and linear regression calculation process until the entire time series under the binning condition has been traversed. This allows the slope values ​​calculated from all windows to be organized in chronological order, forming a trend characteristic data sequence of the parameter under that operating condition, reflecting the rate of change of the parameter in different consecutive time periods.

[0037] For example, in a power supply system for a supermarket operating under "high load - daytime - high temperature" conditions, the system stores 5 consecutive minutes of RMS current data from a certain afternoon, collected once per second, totaling 300 data points. With a preset sliding window size of 60 (representing 1 minute) and a sliding step of 10 (representing 10 seconds), the window covers the data from the 1st to the 60th second. A linear regression is performed on these 60 data points (timestamp, current value) to calculate the first slope value, for example, 0.05 amperes / second, indicating a slow upward trend in current during that minute. Then, the window slides for 10 seconds, covering the data from the 11th to the 70th second, and the same regression calculation is performed to obtain the second slope value. This process is repeated until the window covers the data from the 241st to the 300th second. Finally, a series of slope values ​​arranged chronologically (e.g., 25 slope values) are obtained. This sequence represents the trend characteristics of the current parameters during that period, clearly showing the fluctuations in the rate of current change.

[0038] Step 103: Based on trend feature data, identify abnormal trends and generate abnormal flags and associated parameter sets.

[0039] In some implementations, the trend characteristic data (i.e., a series of continuous window slope values) calculated in real time under the current operating condition bin is compared with the preset threshold of the corresponding bin. Sequences where the absolute value of the slope is greater than the threshold (indicating an excessively fast or slow rate of change, deviating from the normal range) and where multiple consecutive sliding windows (e.g., three consecutive windows) meet this condition are selected. Once such a continuous abnormal window sequence is confirmed, it is marked as an abnormal trend event, and a corresponding abnormal flag (such as a flag set to "true" or a specific event ID) is generated. Subsequently, the detailed parameter set associated with this abnormal trend event is extracted, and the generated abnormal flag and the extracted associated parameter set are encapsulated into a structured data object (such as JSON format or a specific data structure) for output to the fault location analysis and temperature rise coefficient calculation modules.

[0040] For example, assuming a substation operates under "high load - daytime - high temperature" conditions, the threshold for judging abnormal current slope based on historical data is set to an absolute value of 0.1 amperes / second. During real-time monitoring, a continuous window slope sequence is calculated as: [0.05, 0.12, 0.15, 0.18, 0.04] amperes / second. Analysis reveals that the absolute values ​​of the slopes (0.12, 0.15, 0.18) in the 2nd, 3rd, and 4th windows are all greater than the threshold of 0.1, and these three consecutive windows meet the condition. Therefore, it is determined that there is an abnormal upward trend in current during the time period from the 2nd to the 4th window, and an anomaly flag (such as Event_001) is generated. Next, the associated parameters within these three abnormal windows are extracted from the original data: the average current is calculated to be 225 amperes, the average voltage to be 380 volts, and the average temperature to be 85 degrees Celsius. The current value and corresponding temperature value for each second within these three window time periods are extracted to form a current-temperature correspondence sequence. Finally, the exception flags are encapsulated and output along with these associated parameters.

[0041] Step 104: Perform fault location analysis and temperature rise coefficient calculation on the abnormal signs and associated parameter set, and combine them with short-term environmental data to identify the fault status and generate fault location analysis results.

[0042] The temperature rise coefficient is a parameter used to quantify the relationship between the rate of temperature increase of equipment or components and the current flowing through them. It helps to distinguish between temperature rise caused by normal load and abnormal temperature rise caused by faults. Short-term environmental data refers to the series of ambient temperature data around the system collected in the recent past (such as within the last few hours or day).

[0043] In some implementations, anomaly flags and associated parameter sets are collected to activate fault location analysis. This allows for direct comparison of anomalous parameters (such as abnormal current or voltage on a specific line) with the relationships between devices in the topology diagram, identifying the target electrical equipment most likely causing the anomalous trend. For example, if the current on a certain busbar segment rises abnormally and the temperature of a downstream switchgear is also abnormal, the faulty switchgear may be identified. After locating the target equipment, detailed temperature sequences and current sequences flowing through the main power components (such as switch contacts, transformer windings, and cable conductors) of that equipment are obtained within the anomaly window (this data can be obtained from the associated parameter set or by directly querying historical data from device-level sensors).

[0044] Then, by linearly fitting the slope of the temperature series relative to time, the temperature rise per unit time (e.g., how many degrees Celsius per minute) is calculated. Then, based on the calculated temperature rise and the corresponding current series (usually its effective value or average value), the temperature rise coefficient is calculated.

[0045] Next, acquire ambient temperature sequence data (usually from environmental sensors) within a preset short-term time window (e.g., 1 hour before the anomaly occurs), and calculate the slope of the ambient temperature change over time during this period using linear fitting to obtain the ambient temperature rise compensation value (i.e., the rate of change of the ambient temperature itself). Use this ambient temperature rise compensation value to compensate for the previously calculated equipment temperature rise coefficient, and compare the compensated temperature rise coefficient with the system's preset fault threshold (this threshold is set based on equipment type, rated parameters, and historical normal operation data, and is used to distinguish between normal temperature rise and fault temperature rise). Select data points or time periods where the compensated temperature rise coefficient is greater than the preset fault threshold, mark them as fault states, and generate corresponding fault state flags.

[0046] Finally, based on this fault status indicator, the fault location and impact assessment are performed again in conjunction with the power supply and distribution system topology diagram to generate the final fault location analysis results. These results specifically include: fault equipment identification (such as equipment ID and name), inferred fault type (such as poor contact, overload, insulation aging, etc., derived from an experience database matching abnormal parameter patterns and temperature rise characteristics), and warning level (determined comprehensively based on the degree to which the temperature rise coefficient exceeds the threshold, the degree of current abnormality, etc., such as general warning, severe warning, emergency warning, etc.).

[0047] For example, a flag and associated parameter set regarding an abnormal rise in current on a feeder are received, where the average current is significantly higher than normal, and the average temperature of a switchgear node is also higher than normal. Based on the topology map, the target device is located as switchgear numbered "SW-05". The temperature sequence of the main contacts of this switchgear within the abnormal window (assumed to be 5 minutes) is obtained: rising from 80°C to 95°C, with an average current sequence of 300A during the same period. The temperature rise of the contacts is calculated to be 3°C per minute. The temperature rise coefficient is calculated, assuming the temperature rise is divided by the average current, resulting in a coefficient of 0.01 (°C / min) / A. Simultaneously, short-term environmental data is obtained, and the slope of the ambient temperature rise during this time period is calculated to be 0.1°C per minute (ambient temperature rise compensation value). To compensate for the equipment temperature rise coefficient, for example, subtracting the ambient temperature rise contribution of 0.1°C / min from the equipment temperature rise of 3°C / min yields a net equipment temperature rise of 2.9°C / min. Dividing this by the current gives a compensated temperature rise coefficient of approximately 0.00967 (°C / min) / A. The preset fault threshold for this type of switchgear contact is 0.008 (°C / min) / A. Since the compensated temperature rise coefficient of 0.00967 is greater than the threshold of 0.008, the system determines that the switchgear contact is faulty and generates a fault flag. Combining this with the topology diagram, the final fault location analysis result is generated: the faulty equipment is identified as "SW-05 switchgear main contact", the fault type is "possible abnormal increase in contact resistance", and the warning level is "serious warning".

[0048] Step 105: Generate an early warning signal based on the fault location analysis results.

[0049] In some implementations, a predefined warning response strategy library is directly queried based on the received warning level and fault type. Strategy matching is then performed, and specific multi-level warning signals are generated based on the matched response strategy. These typically include: local visual warning signals (such as popping up an alarm window on the local monitoring screen of the faulty device or illuminating the device's warning light), local auditory warning signals (such as triggering a buzzer or voice broadcast near the device or in the power distribution room), and remote notification signals (such as sending alarm information to remote maintenance personnel or the dispatch center via SMS, application push, email, or a work order system). Next, the system dynamically adjusts the output parameters of the generated warning signals based on the warning level and temperature rise coefficient compensation value (which reflects the severity of the fault's overheating) from the fault location analysis results.

[0050] The early warning response strategy library is pre-analyzed based on historical failure cases, equipment importance, safety procedures, and operational experience. It maps different combinations of "early warning level - failure type" to corresponding "response strategies" in the form of tables or a rule engine. The response strategy defines in detail which early warning signals should be generated for that situation, the initial parameters of the signals (such as frequency and duration), and the sending target.

[0051] The logic for adjusting warning signals typically includes the following: the higher the warning level and the greater the temperature rise coefficient compensation value, the higher the output frequency of the warning signal (e.g., the buzzer sounds more urgently, the screen flashes faster), and the longer the duration (e.g., continuous alarm until manual confirmation), to enhance the urgency and continuity of the warning. Finally, the system executes the signal output, sending the adjusted warning signal through the corresponding hardware interface (e.g., relay, communication module) or software interface (e.g., message queue, API call).

[0052] For example, upon receiving the fault location analysis results: the faulty equipment is "#2 transformer low-voltage winding", the fault type is "overheating", the warning level is "emergency", and the temperature rise coefficient compensation value is 0.012 (°C / min) / A. The system will query the warning response strategy library based on the "emergency" level and "overheating" type, and generate the following matching strategy requirements: the local control cabinet's red warning light will remain constantly on, the local buzzer will sound intermittently, and an emergency work order will be sent to the maintenance team and the dispatch center. The system generates these signal instructions. Then, based on the "emergency" level and the high compensation value of 0.012, it decides to increase the warning intensity: the buzzer's sounding frequency is adjusted to 2 times per second (high frequency), and all warning signals are set to be continuously output until the remote maintenance personnel perform a "confirmation" operation in the system. Finally, the red warning light on the #2 transformer control cabinet illuminates, the buzzer begins to sound at a high frequency, and an emergency work order notification containing detailed fault information immediately pops up on the terminals of the maintenance personnel and the dispatch center.

[0053] Based on the above technical solution, by classifying operating conditions into bins according to three dimensions—load level, time period, and ambient temperature—multi-parameter time-series data can be structurally categorized, solving the problem of analysis distortion caused by the mixing of data from different operating conditions. Simultaneously, a sliding window algorithm combined with linear regression is used to calculate the slope of parameter changes, accurately capturing the trend characteristics of time-series data and overcoming the difficulty of quantifying the rate of parameter change in traditional methods, enabling early identification of abnormal trends. Furthermore, continuous abnormal windows are filtered using historical thresholds, and associated parameter sets are extracted, providing clear data support for fault analysis and reducing ineffective troubleshooting. Subsequently, the target equipment is located using the system topology map, the temperature rise coefficient is calculated, and an environmental compensation mechanism is introduced to eliminate environmental interference, solving the problems of inaccurate fault location and easy misjudgment, and accurately identifying faulty equipment and types. Finally, multi-level early warning signals are generated, and output parameters are dynamically adjusted to ensure timely transmission of early warning information, solving the pain point of delayed response in traditional passive maintenance, improving operation and maintenance efficiency, and reducing power outage losses caused by faults.

[0054] In another possible implementation of the embodiments of this application, combined with Figure 1-2 As shown, the process of collecting multi-parameter time-series data of the power supply and distribution system, classifying it by operating condition and environment, and obtaining the classified dataset can be achieved through the following steps 201 to 202, which are explained in detail below:

[0055] Step 201: Collect multi-parameter time-series data of the power supply and distribution system, and divide the multi-parameter time-series data into multiple operating environment sub-boxes according to the preset operating environment sub-boxing rules. Each operating environment sub-box is uniquely defined by a combination of three dimensions: load level, time period, and ambient temperature. The load level is divided into low load, medium load, high load, and impact load based on the effective value of the current.

[0056] In some implementations, current, voltage, and temperature sensors deployed in the power supply and distribution system collect real-time time-series data of multiple parameters, including current, voltage, and temperature, and record the timestamps corresponding to each data point. This allows for automatic classification of the data collected at each moment according to preset operating condition classification rules. Classification begins by determining the load level range (low load, medium load, high load, or impact load) based on the effective current value at that moment. Then, the time period range is determined based on the time period (e.g., nighttime, daytime based on clock information). Simultaneously, the ambient temperature range (e.g., low temperature, normal temperature, high temperature) is determined based on the monitored ambient temperature value at that moment. These three classification results are combined to form a unique operating condition classification code. Finally, all parameter data (including current, voltage, and temperature) and their timestamps are stored in the dataset corresponding to this classification code, resulting in multiple classified datasets. Each dataset contains parameter sequences for all time points under the same operating condition.

[0057] It should be noted that the preset rules for dividing the working environment into boxes should be based on historical operating data, equipment characteristics and expert experience. For example, the threshold for dividing the load level can be set according to the percentage of the rated current of the transformer or line, the division of time periods can take into account the peak and valley patterns of electricity consumption, and the division of ambient temperature can be determined according to local climate characteristics and the operating temperature range of the equipment.

[0058] For example, suppose in a power supply system of an industrial park, the preset load level classification thresholds are: current effective value below 30% of the rated value is low load, 30%-70% is medium load, 70%-100% is high load, and above 100% or short-term drastic fluctuations are impact load; time periods are divided into 00:00-08:00 as nighttime and 08:00-20:00 as daytime; ambient temperature is divided into below 10℃ as low temperature, 10℃-30℃ as normal temperature, and above 30℃ as high temperature. If at 14:05 on a certain day, the effective value of the current of a certain line is 85% of the rated value and the ambient temperature is 35℃, then the data at that moment will be automatically classified into the "high load - daytime - high temperature" operating environment bin, and the current, voltage, temperature and other parameters at that moment will be stored in the corresponding dataset to avoid interference from low temperature or low load data at night.

[0059] Step 202: Store the divided multi-parameter time series data in bins according to the corresponding working conditions to form a classified dataset. Each dataset contains the parameter sequence of all time points in the same bin.

[0060] In some implementations, after binning and classifying the multi-parameter time-series data by operating environment, the next stage is data storage and organization. This involves directly creating an independent data storage structure in memory for each predefined and actually occurring operating environment bin. Thus, for each newly acquired and classified data point, it is written into the corresponding storage structure according to its binning code. Simultaneously, the write operation must ensure that at least a timestamp and key parameters such as the current, voltage, and temperature values ​​acquired at that moment are included, and stored in time-stamp order to form a parameter sequence. As data is continuously acquired, new data points are continuously added to the storage structure corresponding to each bin, dynamically growing to form a complete, time-ordered multi-parameter time-series dataset for that bin.

[0061] For example, if a sub-database is preset to "High Load - Daytime - High Temperature," whenever data meeting these conditions is collected (e.g., current RMS value between 70% and 100% of rated value, time between 08:00 and 20:00, ambient temperature above 30℃), the data point (including timestamp, current value, voltage value, and temperature value) is appended and stored in a database labeled "High Load - Daytime - High Temperature." After a period of operation, this database forms a continuous historical sequence of multiple parameters containing only the "High Load - Daytime - High Temperature" operating conditions. Therefore, when it is necessary to analyze the long-term trend of current under this operating condition, the analysis module can directly retrieve this dataset for processing, without mixing in data from low-load or low-temperature conditions at night, thus ensuring the accuracy and relevance of the trend analysis.

[0062] Based on the above technical solution, the operating environment is uniquely defined by a combination of three dimensions: load level, time period, and ambient temperature. Load levels are further subdivided into four categories based on the effective current value. Data is then stored in each bin to form a dedicated parameter sequence. This allows the three-dimensional bin structure to accurately anchor the data to its operating condition, resolving the problem of data mixing across different loads, times, and temperatures, and avoiding distortion caused by cross-condition analysis. Simultaneously, the load subdivision based on the effective current value adapts to the differences in equipment operating characteristics under different load intensities, solving the problem of inaccurate operating condition matching caused by the coarse load division in traditional methods. Finally, the dedicated parameter sequence formed by bin storage enables structured data management, solving the problem of disorganized and difficult-to-analyze raw time-series data. This allows subsequent trend feature extraction and anomaly detection to be based on data from the same source, significantly improving the accuracy of early data processing and providing high-quality data support for power distribution system fault early warning, reducing the risk of misjudgment and missed detection from the source.

[0063] In another possible implementation of the embodiments of this application, combined with Figure 1-3As shown, the process of obtaining trend feature data by calculating the slope of the parameter change trend within each window through linear regression can be achieved through the following steps 301 to 303, which are explained in detail below:

[0064] Step 301: Based on the data acquisition frequency and trend analysis accuracy requirements, preset the size and sliding step of the sliding window, with the sliding step being less than or equal to the window size.

[0065] Among them, the trend analysis accuracy requirement refers to the minimum time dimension or slope sensitivity of the trend change that you want to identify from time series data, which affects the time span setting of the sliding window.

[0066] In some implementations, the data acquisition frequency is determined based on the specific monitoring needs and data acquisition conditions of the power supply and distribution system. For example, if the system requires monitoring trend changes at the minute level and the sensor supports data acquisition once per second, the acquisition frequency can be set to 1Hz. The time span of the sliding window is then determined based on the required trend analysis accuracy. For instance, if it is desired to identify the trend changes of parameters within 5 minutes, and the acquisition frequency is 1Hz, the sliding window size should be set to 300 data points. To balance computational efficiency and trend continuity, a sliding step size needs to be set. Typically, the step size is less than or equal to the window size to ensure data overlap between windows. For example, the step size can be set to one-tenth of the window size, i.e., 30 data points.

[0067] Based on these preset values, a sliding window analysis model can be directly constructed. The model is initialized with the window size and step size parameters loaded. During actual operation, data points are read sequentially from the classified dataset, and data point sequences of consecutive window sizes are extracted at step size intervals as the current analysis window until the entire dataset is traversed, thus providing input data for the linear regression calculation within each window.

[0068] It should be noted that setting the sliding step size to be less than or equal to the window size ensures that there is at least some data overlap between adjacent analysis windows. This helps to capture the continuity of parameter changes and avoid missing short but important trend segments due to excessive window jumps. At the same time, the overlapping design can also enhance the smoothness and robustness of trend analysis.

[0069] For example, in a power supply and distribution system monitoring scenario, the data acquisition frequency is set to once per second, meaning that parameters such as current and voltage are recorded once per second. The trend analysis accuracy requirement is to identify significant trend changes within a 10-minute timeframe. Therefore, the sliding window size can be set to 600 data points (corresponding to a 10-minute time window). To maintain the continuity of trend tracking and control the computational load, the sliding step size is set to 60 data points (corresponding to 1 minute). In actual operation, the analysis model reads data from the dataset of the "high load - daytime - high temperature" operating condition sub-group. First, it takes the data points from the 1st second to the 600th second as the first analysis window and calculates the trend slope of the current value within this window. Then, the window slides 60 points, taking the data from the 61st second to the 660th second as the second analysis window for calculation; this process is repeated until all data is processed. In this way, data segments of 10 minutes are continuously analyzed, with a 9-minute overlap between adjacent analysis segments, ensuring continuous monitoring of trend changes.

[0070] Step 302: Extract multi-parameter time series data points from the dataset for each sliding window. The data points include timestamps and corresponding parameter values.

[0071] In some implementations, the start and end positions of each sliding window in the time series are determined based on a pre-set sliding window size and step size. Then, for the dataset corresponding to the bins of the current working environment to be analyzed, the stored data is read in timestamp order. This allows for direct filtering of all data points whose timestamps fall within the specified time interval from the dataset, based on the start and end timestamps of the current sliding window.

[0072] For each selected data point, its timestamp field and one or more parameter value fields to be analyzed, such as current or temperature values, are extracted. These extracted timestamps and parameter values ​​are then combined to form a subset of data points within the current sliding window. Finally, this subset of data points is sorted in ascending order of timestamp to form a complete and ordered multi-parameter time-series data sequence within the window.

[0073] For example, when analyzing the "high load - daytime - high temperature" operating condition binning, assume the current sliding window size corresponds to 600 seconds, and the sliding step size is 60 seconds. The start time of the first window is T1, and the end time is T1+600 seconds. From this binning dataset, read all data rows with timestamps within the interval [T1, T1+600 seconds]. Each data row contains fields such as timestamp (e.g., "2023-10-27 14:05:01"), current value (e.g., 225A), and voltage value (e.g., 380V). The program extracts the timestamps and current values ​​(if the current trend is being analyzed) from these rows to form a set of data points, such as [(T1, 225.1), (T1+1, 225.3), ..., (T1+600, 228.5)]. This set is the multi-parameter time series data point within the first sliding window (here, the current parameter), used for the next step of calculating the trend slope. After the window is slid, the program will repeat the above extraction process based on the new time interval.

[0074] Step 303: Perform linear regression fitting on the parameter values ​​and timestamps within each sliding window, and calculate the slope of the regression line using the least squares method.

[0075] In some implementations, after extracting multi-parameter time-series data points within a specified sliding window from the dataset, for the current window, the timestamp data is converted into a format suitable for numerical computation. For example, each timestamp is converted into seconds relative to the start time of the window or a sampling point number as the independent variable x. Simultaneously, the corresponding parameter values ​​are used as the dependent variable y. Then, the least squares formula is applied. Calculate the slope β of the regression line, where This represents the number of data points within the current window. This represents summing over all data points from 1 to n. and These are the independent variable (e.g., time sequence number) and dependent variable (e.g., parameter value) for the i-th data point, respectively.

[0076] Each sliding window performs the above calculation independently, and the resulting slope value represents the rate of change of the parameter within the time span of that window. Therefore, the calculated slope value can be stored or output as the trend characteristic data of that window for subsequent anomaly trend identification.

[0077] For example, a sliding window covers current data from timestamp T0 to T0+10 minutes, totaling 600 data points (one per second). The timestamps are converted to serial numbers (x) from 0 to 599, with the corresponding current values ​​(y) in amperes (A). The calculation yields... =179700, =135000A, =40455000A·, , =600. Substitute into the slope formula. ≈0.05A / point. Since the time interval is 1 second / point, this slope can be interpreted as the current increasing at an average rate of 0.05 amperes per second within this 10-minute window. This value is the trend slope of the current parameter for this window.

[0078] Based on the above technical solution, by combining the data acquisition frequency and trend analysis accuracy to preset the sliding window size and step size, with the step size not exceeding the window size, it can adapt to different power supply and distribution monitoring scenarios. This ensures data overlap between adjacent windows to maintain trend continuity while avoiding redundant calculations, solving the problems of poor adaptability and efficiency-accuracy imbalance associated with traditional fixed windows. In practical applications, it can flexibly match the monitoring needs of different scenarios such as supermarkets and industrial parks. Furthermore, data points containing timestamps and parameter values ​​are extracted from the window, completely preserving time-series correlation information. This solves the problem of incomplete trend analysis due to missing data dimensions, ensuring the reliability of the analysis basis. The slope can then be calculated using linear regression fitting and least squares method, quantifying parameter change trends into intuitive indicators. This solves the problem of traditional methods struggling to accurately capture parameter change rates and identify gradual faults early. The generated slope data can directly provide core features for power supply and distribution system fault early warning, assisting in early anomaly identification and improving the timeliness and targeting of operation and maintenance responses.

[0079] In another possible implementation of the embodiments of this application, combined with Figure 1-4 As shown, the process of identifying abnormal trends and generating abnormal flags and associated parameter sets based on trend feature data can be achieved through steps 401 to 403, which are explained in detail below:

[0080] Step 401: Set an abnormal trend judgment threshold based on historical normal operation data.

[0081] In some implementations, time-series data of all normal operating periods stored in each predefined bin is extracted from the historical dataset classified by operating environment bins. For example, all data points marked as normal operating status in the past three months under the "high load - daytime - high temperature" bin are extracted.

[0082] For the data within this sub-bin, the same sliding window algorithm and linear regression method as in real-time analysis are used to calculate the slope value of the changing trend of each parameter (such as current) within each historical sliding window, thereby obtaining a complete trend slope value sequence during the historical normal operation of this sub-bin. Statistical analysis is then performed based on this sequence to calculate its statistical distribution characteristics, such as calculating the mean μ and standard deviation σ of all historical slope values, and determining a threshold boundary based on a preset confidence level (e.g., 99.7% corresponds to μ±3σ) or through the percentile method (e.g., taking the 95th percentile of the absolute value).

[0083] In another implementation, expert experience and safety margin can be combined to adjust the above statistical values ​​and finally set a specific abnormal trend judgment threshold for each parameter of each working condition sub-box. For example, the abnormal current slope judgment threshold under the "high load - daytime - high temperature" sub-box is set to an absolute value of 0.1 amperes / second.

[0084] It should be noted that the process of constructing the abnormal trend judgment threshold needs to be reviewed and updated regularly (e.g., quarterly or annually) using updated historical data to adapt to feature drift caused by equipment aging or slow changes in operating mode.

[0085] For example, in a supermarket power supply system operating under "medium load - daytime - normal temperature" conditions, historical normal operating current data from the past six months were retrieved. After sliding window analysis (e.g., window size 1 minute, step size 10 seconds) and linear regression calculations, approximately 26,000 historical current slope values ​​were obtained. Statistical analysis revealed that these slope values ​​roughly follow a normal distribution with a mean of 0.01 amperes / second and a standard deviation of 0.02 amperes / second. To ensure high anomaly detection sensitivity and control false alarms, the anomaly trend judgment threshold was set to the absolute value |0.01+2.5σ|, based on the principle of μ±2.5σ. 0.02 ≈ 0.06 amperes / second. This means that in future real-time monitoring, for current data under the same operating condition in different sub-slots, if the absolute value of the calculated slope of the continuous sliding window exceeds 0.06 amperes / second, it will be preliminarily judged as having an abnormal trend.

[0086] Step 402: If the absolute value of the slope is greater than the abnormal trend judgment threshold and multiple consecutive sliding windows meet the condition, it is set as an abnormal trend and an abnormal flag is generated.

[0087] In some implementations, a series of sliding window slope values ​​calculated under the current working condition bin are first read from the trend feature data sequence in chronological order. Then, the absolute value of the slope of each window is compared with a pre-set abnormal trend judgment threshold for that bin and its parameters. This allows direct checking for the existence of several consecutive (e.g., three consecutive, adjustable based on sensitivity requirements for trend persistence) sliding windows; for example, requiring two, three, or more consecutive windows to meet the condition to balance detection timeliness and noise resistance. When a window sequence that meets both the conditions of "absolute slope value greater than the threshold" and "multiple consecutive windows" is identified, an abnormal trend event is determined to exist within that time period. A unique anomaly flag is then generated for this abnormal trend event, such as an incrementing event number or a composite identifier containing a timestamp and bin information. This flag is used to associate and track the abnormal event in subsequent processes. Finally, the generated anomaly flag is associated with and stored or output with the specific time period corresponding to the identified abnormal trend, the parameters involved, and the calculated slope value, thus completing the entire process from data judgment to event identification.

[0088] For example, assuming a substation operates under "high load - daytime - high temperature" conditions, the threshold for judging abnormal current slope trends is set to 0.1 amperes / second, and an abnormal trend is only determined if three consecutive sliding windows meet the condition. The real-time calculated continuous window slope sequence is: [0.05, 0.12, 0.15, 0.18, 0.04] amperes / second. The absolute value of each slope is then compared to the threshold of 0.1: the first window (0.05) is not greater than the threshold; the second window (0.12) is greater than the threshold; the third window (0.15) is greater than the threshold; the fourth window (0.18) is greater than the threshold; and the fifth window (0.04) is not greater than the threshold. The inspection reveals that from the second to the fourth window, the absolute values ​​of the slopes (0.12, 0.15, 0.18) are all greater than the threshold of 0.1, satisfying the "three consecutive sliding windows" condition. Therefore, an abnormal upward trend in current is determined to exist within the time period corresponding to the second to fourth windows, and an abnormality flag is generated accordingly.

[0089] Step 403: Extract the parameter set associated with the abnormal trend, and encapsulate the abnormal flag and the associated parameter set into structured data output, wherein the associated parameter set includes the average current, average voltage, average temperature and current-temperature correspondence sequence within the abnormal window.

[0090] The associated parameter set refers to the collection of relevant electrical and temperature parameters extracted and calculated from the original multi-parameter time series data for specific time periods (i.e., abnormal window sequences) corresponding to identified abnormal trend events. This provides specific and condensed data support for subsequent fault location and status analysis. The current-temperature correspondence sequence refers to the paired current and temperature values ​​arranged chronologically within the same abnormal time period, corresponding to each sampling moment or sub-time period. This sequence preserves the detailed correlation of parameter changes over time and is used to analyze the dynamic relationship between current changes and temperature response.

[0091] In some implementations, after identifying an abnormal trend and generating an anomaly flag, the specific start and end timestamps on the timeline are determined based on the anomaly window sequence associated with the flag. Then, the process traces back to the original dataset after binning and classifying the operating conditions, locating all original sampled data points within the same bin whose timestamps fall within the anomaly time period. From these original data points, the raw value sequences of current, voltage, and temperature parameters are extracted. The statistical characteristics of these parameters within the anomaly time period are then calculated: the arithmetic mean of the current value sequence is calculated to obtain the average current within the anomaly window; the arithmetic mean of the voltage value sequence is calculated to obtain the average voltage within the anomaly window; and the arithmetic mean of the temperature value sequence is calculated to obtain the average temperature within the anomaly window. Simultaneously, the current and temperature values ​​corresponding to each sampling moment (or each sub-time period in sliding window units) are paired and arranged chronologically to form a sequence containing multiple (current, temperature) data pairs, constructing a current-temperature correspondence sequence.

[0092] The calculated average current, average voltage, and average temperature, along with the constructed current-temperature correspondence sequence, are correlated with the previously generated anomaly flags to form a complete data packet. Finally, the system encapsulates the data packet according to a preset structured data format and outputs it to a designated data bus, message queue, or storage interface for reception and use by the fault location analysis and temperature rise coefficient calculation modules.

[0093] For example, an abnormal trend is identified, marked with "Alert_001," and the corresponding abnormal window covers a 30-second time period from 14:05:00 to 14:05:30 (assuming a sliding window size of 10 seconds, and three consecutive windows are judged as abnormal). The current, voltage, and temperature data sampled per second within this 30-second period can then be extracted from the original dataset of the "High Load - Daytime - High Temperature" bin. The average value of these 30 current sampling points is calculated to be 228 amperes (average current), the average value of these 30 voltage sampling points is 381 volts (average voltage), and the average value of these 30 temperature sampling points is 87 degrees Celsius (average temperature). Simultaneously, the current and temperature values ​​per second within these 30 seconds are paired and arranged chronologically to form a current-temperature correspondence sequence, for example: [(225,85),(226,85.5),(228,86),...,(230,89)]. Finally, this data is encapsulated into a structured JSON object for output.

[0094] Based on the above technical solution, anomaly trend judgment thresholds are set according to historical normal operation data. By tailoring its structure to the actual operating conditions of the power supply and distribution system, it can accurately match parameter change patterns in different operating scenarios, solving the problems of poor adaptability and disconnect from actual operating conditions associated with traditional fixed thresholds, effectively improving the scientific rigor and relevance of the thresholds. Furthermore, a rigorous anomaly trend identification logic is constructed through a dual judgment rule that filters for both absolute slope values ​​exceeding thresholds and conditions met by multiple consecutive sliding windows. This avoids false alarms caused by single parameter fluctuations, addressing the pain point of high fault misjudgment rates and ensuring that only truly existing gradual anomalies are captured. Finally, the average current, voltage, and temperature values ​​and the corresponding current-temperature sequences within the anomaly window are extracted and encapsulated into structured data output. This not only fully preserves the key parameter information associated with the anomaly but also solves the problems of fragmented data and difficulty in traceability analysis after an anomaly occurs. It provides comprehensive and standardized data support for subsequent fault location and cause investigation, progressively improving the accuracy and practical value of fault warnings.

[0095] In another possible implementation of the embodiments of this application, combined with Figure 1-5 As shown, the process of fault location analysis and temperature rise coefficient calculation for anomaly flags and associated parameter sets can be achieved through the following steps 501 to 503, which are explained in detail below:

[0096] Step 501: Based on the average current, average voltage, and average temperature in the associated parameter set, and combined with the power supply and distribution system topology map, locate the target electrical equipment corresponding to the abnormal trend.

[0097] A power supply and distribution system topology diagram is a graphical representation of the electrical connections and physical layout of various electrical devices (such as transformers, circuit breakers, switchgear, busbars, and cables) within a power supply and distribution system. It clarifies the hierarchical relationships of the devices, upstream and downstream paths, and the locations of monitoring points. Target electrical equipment refers to specific equipment units within the power supply and distribution system that are initially identified as potentially causing abnormal trends, such as a transformer, a switchgear, or a section of cable.

[0098] In some implementations, after receiving a set of associated parameters containing average current, average voltage, and average temperature, the parameters are parsed to extract the corresponding measurement point identifiers or line numbers. This allows direct access to a pre-stored power supply and distribution system topology database (nodes represent electrical equipment and include equipment ID, type, rated parameters, and associated monitoring point information; edges represent electrical connections and are labeled with the physical parameters of the connecting lines). The average current, voltage, and temperature values ​​in the associated parameter set are then compared with the historical normal value ranges of each monitoring point or equipment node in the topology map. Nodes whose parameter values ​​significantly deviate from their historical normal operating ranges (e.g., current exceeding a specific percentage of the rated value, temperature exceeding the typical value of similar equipment under the same operating conditions) are selected, forming an initial list of suspected equipment.

[0099] By combining the time period of the abnormal trend and the electrical path in the topology diagram, the propagation path of the abnormal parameters is analyzed: for example, if the average current of a certain bus increases abnormally, the downstream equipment of the bus is traced, and the equipment nodes with abnormal average temperature associated with it are checked. The equipment that has both abnormal current and temperature and is located in a critical position in the electrical path is marked as a high-suspicion target.

[0100] In another possible implementation, a pre-defined suspicion determination rule base (built based on the correspondence between equipment failure modes and abnormal parameter characteristics) can be invoked. For example, "an increase in current accompanied by a significant increase in temperature may indicate an increase in contact resistance." These rules are then applied to score and rank high-suspicion targets. Finally, the device with the highest score is identified as the target electrical device corresponding to the abnormal trend, and its device identifier and location basis are output, completing the mapping and location from parameter abnormalities to specific devices.

[0101] For example, suppose a power distribution system in an industrial park receives a set of associated parameters showing that the average current of a 10kV feeder line is 520A (historically normal range 450-500A), the average voltage is 9.8kV (slightly lower than the rated 10kV), and the average temperature of a switchgear node is 95℃ (similar equipment under the same load typically has a temperature of 70-80℃). This feeder line can be directly located in the topology map, with 5 switchgear units connected downstream. Comparison reveals that the average temperature of switchgear numbered SW-12 is abnormally high, and this unit is located at the end of the feeder line, with a slight increase in current in its upstream equipment along the electrical path. Therefore, the rule base is invoked, matching the rule "significantly increased temperature of end equipment accompanied by a slight increase in line current may indicate aging of the equipment's contacts," and this switchgear is given a high-weight score. Meanwhile, the temperatures of other switchgear units are all within the normal range. Therefore, switchgear SW-12 is identified as the target electrical equipment, the location result is output, and maintenance personnel are prompted to focus on checking the contact connection status of this switchgear.

[0102] Step 502: Obtain the temperature and current sequences of the main power components of the target electrical equipment within the abnormal window.

[0103] Among them, the main power component refers to the physical component in the target electrical equipment that undertakes the main functions of power transmission, conversion, or switching and thus generates significant heat (Joule heating), such as the contacts of switchgear or circuit breaker contacts, transformer windings, cable conductors, contactor contacts, etc. An abnormal window refers to a specific time period covered by one or more consecutive sliding windows that have been identified as exhibiting abnormal trends in previous steps.

[0104] In some implementations, after locating the target electrical equipment, a pre-configured device-sensor mapping table is queried based on the equipment's type and identifier. Then, a query request is initiated to the data storage layer based on the start and end timestamps of the anomaly window, retrieving all raw sampling data collected by the specific sensor within that time period. Since sensor data is typically stored in a time-series database, the corresponding temperature and current sampling values ​​are extracted and aligned according to their timestamps, forming two independent but time-synchronized sequences. These two sequences are then arranged in ascending order by timestamp and temporarily stored in memory, forming a structured data pair that can be directly used for subsequent calculations.

[0105] To ensure the integrity and accuracy of the two generated sequence data, the system performs simple data quality checks, such as removing outliers that are clearly beyond the physical possibilities (e.g., temperature values ​​that are lower than the ambient temperature or higher than the material's melting point), and filling in missing points caused by brief communication interruptions using linear interpolation of data from the preceding and following time points.

[0106] The device-sensor mapping table is manually entered during system initialization based on the on-site sensor installation locations and equipment drawings. It clearly records the temperature sensor ID and current sensor ID corresponding to the main power component of each device. During actual use, if the main power component has multiple temperature monitoring points (such as the high-voltage and low-voltage windings of a transformer), the monitoring point with the highest correlation to the anomaly can be directly selected according to the mapping table, or the average value of multiple monitoring points can be used as the representative temperature sequence.

[0107] For example, suppose the target electrical equipment is located as a dry-type transformer numbered "TR-01", whose main power component is the low-voltage side winding. A lookup of the mapping table reveals that the temperature of this winding is monitored by a fiber optic temperature sensor "TT-TR01-L" installed at the winding's hot spot, and the current flowing through the winding is monitored by a current transformer "CT-TR01-L" in the low-voltage side outgoing cabinet. The abnormal window is determined to be a 30-second time period from 14:05:00 to 14:05:30 on the current day. The system then queries the database for the data sampled per second by these two sensors during this time period, obtaining the temperature sequence (e.g., [85.2, 85.5, 86.1, ..., 89.8]℃) and the current sequence (e.g., [1023, 1025, 1028, ..., 1045]A). The system checks that the data is complete and without missing data, and then extracts these two sequences, each 30 seconds long, for use in the next step of temperature rise calculation.

[0108] Step 503: Calculate the temperature rise per unit time by linearly fitting the slope of the temperature sequence relative to time, and calculate the temperature rise coefficient based on the temperature rise and the current sequence.

[0109] In some implementations, temperature and time series data are obtained as input data, and the least squares method is used to perform linear regression fitting on these two sets of data to construct a model T=β. Let t+α, where T represents temperature, t represents relative time, α is the intercept of the fitted straight line (representing the estimated initial temperature), and β is the slope of the temperature change relative to time. The arithmetic mean (or root mean square, i.e., effective value) of the current sequence within the same anomaly window is then calculated as the typical current characteristic I_avg flowing through the main power component during that period. The temperature rise coefficient K can then be calculated using K=β / I_avg based on the given temperature rise and current sequence.

[0110] For example, suppose the main power component of the target device has the following temperature sequence (one point per second) within a 30-second anomaly window: [85.0, 85.3, 85.9, 86.6, 87.4, 88.3, ​​89.3, 90.4, 91.6, 92.9, 94.3, 95.8, 97.4, 99.1, 100.9, 102.8, 104.8, 106.9, 109.1, 111.4, 113.8, 116.3, 118.9, 121.6, 124.4, 127.3, 130.3, 133.4, 136.6, 139.9]℃, corresponding to a relative time sequence value of [0, 1, 2, ..., 29] seconds. Simultaneously, the average current sequence value is I_avg = 1020A. First, a linear fit can be performed on 30 (t, T) data points. The slope β is calculated to be approximately 1.83℃ / s, which is the temperature rise per unit time (second). Then, the temperature rise coefficient K is calculated as K = β / I_avg ≈ 1.83 / 1020 ≈ 0.001794 (℃ / s) / A.

[0111] Based on the above technical solution, by relying on the average values ​​of current, voltage, and temperature from the associated parameter set and combining them with the power supply and distribution system topology map to locate target electrical equipment, a deep coupling between parameter data and the physical connection relationship of the system is achieved. This solves the problems of traditional fault location relying on manual experience and having an ambiguous scope, thereby shortening the time for investigating suspected equipment and improving fault location efficiency in practical applications. Simultaneously, focusing on the main power components of the target equipment, temperature and current sequences within the abnormal window are collected, effectively locking the time-series data of core fault-prone components, avoiding redundant interference from comprehensive monitoring, and solving the problem of inaccurate capture of the status of key components, providing targeted data support for subsequent analysis. Finally, the temperature rise is calculated by linearly fitting the slope of the temperature sequence, and the temperature rise coefficient is obtained by combining it with the current sequence to quantify the correlation between temperature rise and current. This solves the technical problem of not being able to distinguish between normal load temperature rise and fault temperature rise, and can effectively quantify fault risk indicators.

[0112] In another possible implementation of the embodiments of this application, combined with Figure 1-6 As shown, the process of identifying fault status and generating fault location analysis results by combining short-term environmental data can be achieved through the following steps 601 to 604, which are explained in detail below:

[0113] Step 601: Obtain the ambient temperature sequence within a preset short-term time window, and calculate the slope of the ambient temperature change over time through linear fitting to obtain the ambient temperature rise compensation value.

[0114] In some implementations, the duration of the "short-term time window" is determined according to preset rules, such as 120 minutes before the start of the abnormal trend. All original sampling points collected by the specified ambient temperature sensor within this time period are queried from the time series database and arranged in ascending order by timestamp to form an "ambient temperature sequence". At this time, each data point in the sequence contains the collection time and the corresponding temperature value.

[0115] Then, using the same linear regression method as the previous analysis of equipment temperature trends, with time as the independent variable and ambient temperature as the dependent variable, the least squares method is used to fit the sequence, and the slope of the fitted line is calculated. This slope is the "ambient temperature rise compensation value".

[0116] The preset window duration and fitting method are determined during system deployment based on typical environmental temperature change cycles (such as daily cycles) and sensor accuracy, and are verified through testing with historical environmental data to ensure stable capture of representative environmental change trends. Furthermore, no additional model training is required; the system is implemented directly based on statistical calculations.

[0117] For example, suppose that in the power distribution system of an industrial park, an abnormal current trend is detected at 14:05 on a certain day. The system then sets a short-term time window to 2 hours before the onset of the abnormality (i.e., from 12:05 to 14:05) and queries the records of the ambient temperature sensor in the power distribution room during this period to obtain an ambient temperature sequence of one sampling point per minute, for example, gradually fluctuating from 25°C at 12:05 to 28°C at 14:05, for a total of 120 data points. Linear fitting is performed on this sequence to calculate an ambient temperature rise compensation value of approximately 0.025°C / minute, indicating that in the 2 hours before the abnormality occurred, the ambient temperature slowly rose at an average rate of 0.025 degrees Celsius per minute. This compensation value will be used in subsequent steps to subtract the environmental contribution from the original temperature rise calculated from the main power components of the target electrical equipment. For example, if the equipment temperature rise is 3°C per minute, the net equipment temperature rise after compensation is approximately 2.975°C / minute, thereby more accurately determining whether the abnormal temperature rise is caused by a fault.

[0118] Step 602: Use the ambient temperature rise compensation value to calculate the temperature rise coefficient to obtain the compensated temperature rise coefficient.

[0119] The temperature rise coefficient is a parameter calculated by linearly fitting the slope of the temperature sequence of the main power component of the target electrical equipment relative to time within an abnormal window, and dividing it by the average (or effective) value of the current sequence within the same window. Its physical meaning is the rate of temperature change of the equipment component caused by a unit current. It is used to quantify the correlation between current and temperature rise and is a key indicator for distinguishing between normal load temperature rise and fault temperature rise. The compensated temperature rise coefficient is a new coefficient obtained by correcting the original temperature rise coefficient using environmental temperature rise compensation values. Its purpose is to eliminate the influence of natural changes in ambient temperature from the total temperature rise of the equipment, thereby more purely reflecting the heating rate of the equipment itself due to electrical conditions (such as increased contact resistance, insulation aging, etc.), and improving the accuracy and reliability of fault condition judgment.

[0120] In some implementations, the environmental temperature rise compensation value and the temperature rise coefficient are obtained. Since the core logic of the compensation calculation is to separate the environmental temperature rise from the total equipment temperature rise, and considering that the environmental temperature rise compensation value represents the rate of temperature change purely caused by the environment, while the original equipment temperature rise is the sum of the equipment's own heat generation and the environmental impact, the compensated net temperature rise rate can be obtained by subtracting the original equipment temperature rise from the temperature rise compensation value. Then, the temperature rise coefficient, i.e., the compensated temperature rise coefficient, is recalculated using this net temperature rise rate.

[0121] In practice, the calculation process is encapsulated as a deterministic arithmetic operation module, which does not require model training. Its preset rules are based on the principle of physical heat transfer: the temperature change of the device monitoring point is the result of the combined effect of its own heat generation and heat dissipation and the heat exchange with the environment. If the ambient temperature changes linearly within a short time window, its impact can be linearly deducted.

[0122] For example, assuming the initial temperature change slope β of a switchgear's main contacts within an abnormal window is 3°C / min, and the average current I_avg within that window is 300A, then the initial temperature rise coefficient K = 3 / 300 = 0.01 (°C / min) / A. Meanwhile, the environmental temperature rise compensation value β_env calculated from short-term environmental data is 0.2°C / min, indicating that the ambient temperature rises by 0.2°C per minute. Therefore, when performing the compensation calculation, first calculate the net temperature rise rate of the equipment β_dev = 3 - 0.2 = 2.8°C / min. Then calculate the compensated temperature rise coefficient K_comp = 2.8 / 300 ≈ 0.00933 (°C / min) / A. Compared to the original coefficient of 0.01, the coefficient decreases after compensation, which reflects that some of the temperature rise comes from environmental warming rather than equipment failure. If the preset fault threshold is 0.009 (°C / minute) / A, the compensated coefficient of 0.00933 is still slightly higher than the threshold, and the system may still judge it as a fault state. However, if no compensation is made, using 0.01 for judgment will overestimate the fault risk or lead to misjudgment, which shows the importance of environmental compensation for accurate judgment.

[0123] Step 603: Filter data whose temperature rise coefficient after compensation is greater than the preset fault threshold, record it as a fault state, and generate a fault state flag.

[0124] In some implementations, the temperature rise coefficient after compensation is obtained (denoted as K_comp), and a preset fault threshold (denoted as Th_fault) matching the current target electrical equipment type and its main power components is called. This allows direct determination of whether K_comp is greater than Th_fault. If the condition is met (K_comp > Th_fault), the equipment is determined to be in a fault state within the current anomaly trend window. A fault status flag is then generated. This flag is typically a structured data object containing at least the following fields: flag type (e.g., "fault"), associated unique equipment identifier (from the power supply and distribution system topology), corresponding anomaly time period (start and end timestamps), calculated K_comp value, used Th_fault value, and trigger time. Finally, this fault status flag is output to the downstream processing flow, providing the core basis for generating fault location analysis results.

[0125] The preset fault threshold is constructed based on historical data and engineering specifications. That is, when the system is initialized or the equipment file is established, the historical temperature rise coefficient data of the equipment during long-term normal operation under various working conditions is collected, statistical analysis is performed (such as calculating the 95th percentile or mean of the distribution plus several times the standard deviation), and the preset threshold is then combined with the rated temperature rise limit provided by the equipment manufacturer, industry safety standards, and expert experience for comprehensive calibration.

[0126] It should be noted that the preset fault threshold may vary depending on the equipment model, service life, or even specific operating conditions. The system needs to maintain an equipment type-component-threshold mapping table to achieve accurate matching.

[0127] For example, suppose the preset fault threshold Th_fault for the low-voltage winding of a certain type of dry-type transformer is set to 0.0015 (°C / min) / A. In one analysis, the compensated temperature rise coefficient K_comp for this winding within a certain abnormal window is calculated to be 0.0018 (°C / min) / A. The system performs a comparison: 0.0018 > 0.0015, the condition is met. Therefore, the system determines that the transformer winding is in a "fault state" during this time period. Subsequently, the system generates a fault state flag, the content of which is, for example: {Flag type: "overheat fault", Device ID: "TR-01-LV-Winding", Time window: "2023-10-27 14:05:00 to 14:05:30", K_comp: 0.0018, Th_fault: 0.0015, Generation time: "2023-10-27 14:05:35"}. This flag indicates that a specific fault condition has been identified, and subsequent steps will generate a complete analysis result based on this flag, including the faulty device identifier, type, and warning level.

[0128] Step 604: Based on the fault status flags and combined with the power supply and distribution system topology diagram, generate fault location analysis results. The fault location analysis results include fault equipment identification, fault type, and warning level.

[0129] In some implementations, a fault status flag is received, and a pre-stored power supply and distribution system topology map is loaded. This allows for precise matching and location of the fault in the topology map based on the device identifier in the fault status flag, thus identifying the specific physical device experiencing the fault and forming a fault device identifier. Then, relying on a preset fault mode-feature mapping table, key features in the current fault status flag (mainly the compensated temperature rise coefficient K_comp, combined with auxiliary information such as whether the average current value is significantly higher or lower than the average voltage value in the associated parameter set) are compared and matched with features in the mapping table to find the most suitable fault type, thereby determining the warning level. Finally, the determined fault device identifier, fault type, and warning level are integrated and encapsulated to form a structured fault location analysis result output.

[0130] The Fault Mode-Feature Mapping Table is built before system deployment based on equipment failure mechanisms, historical failure case analysis, and expert experience. It establishes a correlation between different types of faults (such as poor contact, overload, and insulation aging) and observable parameter anomaly patterns (especially abnormal temperature rise coefficients within a specific range, which may be combined with abnormal average values ​​of current and voltage in the associated parameter set).

[0131] Determining the warning level relies on a preset warning level judgment rule, which is also based on engineering practice and is usually a piecewise function or decision table. Its inputs include: the percentage or absolute difference (K_comp-Th_fault) / Th_fault of the compensated temperature rise coefficient K_comp exceeding the preset fault threshold Th_fault. The system calculates or looks up the specific warning level based on these input parameters, including the criticality level of the faulty equipment in the topology diagram (such as whether it is a critical bus, main transformer, etc., which is predefined in the node attributes of the topology diagram), and the potential hazards of the inferred fault type itself.

[0132] For example, the system receives a fault status flag indicating that device ID "CB-203A" (which, according to the topology diagram, is a feeder circuit breaker) is in a fault state. Its compensated temperature rise coefficient K_comp is 0.012 (°C / min) / A, the preset fault threshold Th_fault is 0.008 (°C / min) / A, the associated average current is 10% higher than normal, and the average voltage is normal. The faulty device can then be located in the topology diagram and confirmed to be "Circuit breaker CB-203A for feeder 5". Next, the fault mode-feature mapping table is queried: the features "significantly high temperature rise coefficient (>50% threshold), high current, normal voltage" match the fault type "abnormally increased main contact resistance". Then, the warning level was determined: the exceedance rate was calculated as (0.012-0.008) / 0.008=50%. Querying the topology map attributes revealed that the circuit breaker was a secondary critical device. Combined with the fault type "abnormally increased contact resistance," which carries a medium risk of causing a fire, and based on the preset rule table, the warning level was comprehensively determined to be "Severe Warning." Finally, the generated fault location analysis result was: {Fault Equipment Identifier: "Circuit Breaker CB-203A of Feeder No. 5", Fault Type: "Abnormally Increased Main Contact Resistance", Warning Level: "Severe Warning"}. This result clearly indicates where the problem occurred, what the possible problem is, and the severity of the problem.

[0133] Based on the above technical solution, by acquiring short-term ambient temperature sequences and linearly fitting the slope to obtain ambient temperature rise compensation values, the interference of natural ambient temperature rise can be accurately separated by relying on time-series data and linear fitting algorithms. This solves the problem of misjudging equipment temperature rise caused by environmental factors, ensuring that subsequent analysis focuses on the equipment's own state. The compensation value is then used to correct the temperature rise coefficient, and the accuracy of core indicators is optimized through numerical calibration. This overcomes the technical challenge of traditional methods failing to distinguish between environmental and equipment-specific heating, making the temperature rise data more consistent with the actual operating conditions of the equipment. By combining data with compensated temperature rise coefficients exceeding preset fault thresholds, a quantitative standard for fault indicators can be generated to clearly define the fault status, solving the problems of ambiguous fault identification and lack of unified judgment criteria, providing clear guidance for fault handling. Finally, by combining the power supply and distribution system topology map, a location result containing fault equipment identification, type, and warning level is generated, achieving deep correlation between data and system physical layout. This solves the problems of inaccurate fault location and inefficient operation and maintenance, effectively improving the targeting and efficiency of fault handling in practical applications and reducing the risk of power outages.

[0134] In another possible implementation of the embodiments of this application, combined with Figure 1-7 As shown, the process of generating an early warning signal based on the fault location analysis results can be achieved through the following steps 701 to 703, which are explained in detail below:

[0135] Step 701: Based on the warning level and fault type in the fault location analysis results, match the response strategy in the predefined warning response strategy library.

[0136] In some implementations, during the deployment or initialization phase, a large number of historical fault events are analyzed to summarize the response measures that should be taken for different fault types (such as poor contact or overload) at different warning levels (such as severe or emergency). For example, for a "severe warning" level fault of "abnormally increased contact resistance," the strategy may require triggering both local audible and visual alarms and remote work order notifications simultaneously. The correspondence between these warning levels, fault types, and response measures is then structured and stored in the form of tables, configuration files, or rule scripts, thus forming a warning response strategy library.

[0137] In actual operation, after the system generates fault location analysis results containing specific warning levels and fault types, the warning level and fault type fields can be directly used as joint query keys to perform precise matching and searching in the warning response strategy library. Once a match is successful, the response strategy is locked, and its content is loaded into memory as the direct basis for generating specific warning signals subsequently.

[0138] For example, suppose that in a substation power distribution system, the fault location analysis result outputs: the warning level is "Severe Warning", and the fault type is "Abnormally increased contact resistance of the main contact of the switchgear". Then, this step is executed: using "Severe Warning" and "Abnormally increased contact resistance" as a combination key, the predefined warning response strategy library is queried. If a corresponding strategy is pre-stored in the strategy library, its content is defined as: "Generate a local visual warning signal (the red warning light of the control cabinet flashes), a local auditory warning signal (the buzzer on site sounds at a frequency of 1 time / second), and a remote notification signal (push alarm information and equipment location to the mobile terminal of the maintenance team). This allows the system to lock the strategy content after successfully matching it and output it to the warning signal generation module of the next stage as a direct operation instruction to generate the above three specific warning signals.

[0139] Step 702: Based on the matching response strategy, generate multi-level early warning signals, including local visual early warning signals, local auditory early warning signals, and remote notification signals.

[0140] In some implementations, a list of signal types to be generated is extracted based on the matching response strategy. For example, it may be required to generate local visual warning signals, local auditory warning signals, and remote notification signals simultaneously.

[0141] For local visual warning signals, the control software will call preset hardware driver instructions according to the modes that may be specified in the strategy (such as constant light, flashing, specific color). These instructions will be sent to the target device (such as the red warning light on the fault switch cabinet or the field monitoring screen) through the industrial bus (such as RS485, CAN) or IO module, driving it to work according to the specified mode.

[0142] For local auditory warning signals, corresponding control commands are generated according to the strategy (such as specifying the sounding frequency and tone), and the buzzer or voice broadcaster on site is driven to emit sound through the audio output interface or relay control circuit.

[0143] For remote notification signals, the fault location analysis results (including device identification, fault type, warning level, time, etc.) are filled into the template according to the receiving object (such as a specific mobile phone number, APP user group, email address or work order system API address) and message template defined in the strategy, generating a structured alarm message. Then, the message is sent to the specified remote terminal or system through the integrated SMS gateway, push server or network API interface.

[0144] Step 703: Based on the warning level and the temperature rise coefficient compensation value, adjust the output frequency and duration of the warning signal.

[0145] In some implementations, upon receiving a basic warning signal (such as a light with a specific flashing pattern or a buzzer at a fixed frequency), a dynamic signal parameter adjustment module is activated.

[0146] Based on the warning level and calculated temperature rise coefficient compensation value from the fault location analysis, the system uses a preset signal strength adjustment mapping table or a set of adjustment rule functions. By parsing the current warning level (e.g., "Severe Warning") and the temperature rise coefficient compensation value (e.g., 0.00933 (°C / min) / A), the system calculates the required adjustment frequency and duration coefficients for the specific signal (e.g., buzzer sounding) according to the preset mapping table or rule functions. These calculated coefficients are then applied to the original output parameters of the basic warning signal. For example, multiplying the basic buzzer sounding frequency (e.g., 1 time / second) by a frequency coefficient (e.g., 1.6) yields the adjusted output frequency (1.6 times / second), and multiplying the basic duration (e.g., 30 seconds) by a duration coefficient (e.g., 2) yields the adjusted duration (60 seconds). This allows the system to reconfigure and drive the corresponding hardware (e.g., buzzer, warning light) or modify the repetition strategy of remote notifications according to the adjusted new parameters (higher output frequency and longer duration), thus enabling the urgency and warning intensity of the warning signal to more accurately match the actual severity and risk level of the current fault.

[0147] The construction of signal strength adjustment mapping tables or a set of adjustment rule functions is based on engineering experience and safety guidelines. They are usually set by operation and maintenance experts during system deployment according to the equipment risk level and actual operation and maintenance needs. For example, the rule may be defined as: for each increase in the warning level (such as from "normal" to "severe"), the output frequency increases by 50% and the duration is extended by 100%; at the same time, for every 20% increase in the proportion of the temperature rise coefficient compensation value exceeding the preset fault threshold, the output frequency increases by an additional 10% and the duration is extended by an additional 20%.

[0148] For example, suppose a fault warning level is "emergency," and the temperature rise coefficient compensation value is 0.012 (°C / minute) / A, exceeding the fault threshold by 50%. The preset rules are: for the warning level "emergency," the base frequency coefficient is 1.8, and the base duration coefficient is 2.5; for every 10% exceeding the threshold, the frequency coefficient increases by an additional 0.1, and the duration coefficient increases by an additional 0.2. The calculated frequency coefficient is: 1.8 + (50% / 10%). 0.1 = 2.3; Duration coefficient = 2.5 + (50% / 10%) 0.2 = 3.5. If the base buzzer frequency is 1 time per second and the base duration is 30 seconds, after adjustment, the buzzer will sound at a high frequency of 2.3 times per second for 105 seconds (30 seconds). 3.5). At the same time, the flashing frequency and duration of the red warning lights on site are also increased proportionally according to the same rules, thereby significantly improving the warning intensity and matching the high urgency and high risk of the fault.

[0149] Based on the above technical solution, a pre-set response strategy library is first matched according to the warning level and fault type to provide customized handling solutions for precise fault scenarios. This addresses the problems of traditional warning responses lacking specificity and having chaotic strategies, giving maintenance personnel a clear basis for action. Next, multi-layered warning signals are generated, including local visual, auditory, and remote notifications, constructing a comprehensive information transmission channel covering both local and remote areas. This overcomes the pain points of single warning methods being prone to omissions and untimely transmission, ensuring that maintenance personnel are quickly informed of the fault situation. Finally, the signal output frequency and duration are dynamically adjusted based on the warning level and temperature rise coefficient compensation value, ensuring a precise match between warning intensity and fault severity. This addresses the disconnect between the sense of urgency in warnings and actual risks, guiding maintenance personnel to prioritize high-risk faults. This three-pronged approach, progressively enhancing response accuracy, transmission effectiveness, and intensity adaptability, comprehensively improves the practical value of fault warnings, assists in efficient maintenance, and reduces losses caused by power supply and distribution system faults.

[0150] This application embodiment can divide the power supply and distribution system fault early warning system based on multi-parameter trend analysis into functional units according to the above method example. For example, each function can be divided into different functional units, or two or more functions can be integrated into the same processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0151] When using integrated units, Figure 8 The power supply and distribution system fault early warning system based on multi-parameter trend analysis described in the above embodiments includes: a data acquisition and classification module, which collects multi-parameter time-series data of the power supply and distribution system, classifies it by operating environment, and obtains a classified dataset; a trend feature extraction module, which uses a sliding window algorithm to extract the time-series features of the classified dataset, and calculates the slope of the parameter change trend within each window through linear regression to obtain trend feature data; a fault probability prediction module, which judges abnormal trends based on the trend feature data and generates abnormal signs and associated parameter sets; a fault location analysis and temperature rise coefficient calculation are performed on the abnormal signs and associated parameter sets, and fault state identification is performed in conjunction with short-term environmental data to generate fault location analysis results; and an early warning module, which generates an early warning signal based on the fault location analysis results.

[0152] In another possible implementation, the trend feature extraction module specifically includes: a window data partitioning unit, which presets the size and step size of the sliding window based on the data collection frequency and trend analysis accuracy requirements, with the step size being less than or equal to the window size; extracting multi-parameter time-series data points within each sliding window from the dataset, where each data point includes a timestamp and its corresponding parameter value; and a linear regression fitting unit, which performs linear regression fitting on the parameter values ​​and timestamps within each sliding window and calculates the slope of the regression line using the least squares method.

[0153] In another possible implementation, the fault probability prediction module specifically includes: a suspect equipment location unit, which locates the target electrical equipment corresponding to the abnormal trend based on the average current, average voltage, and average temperature values ​​in the associated parameter set, combined with the power supply and distribution system topology map; a data acquisition unit, which acquires the temperature and current sequences of the main power components of the target electrical equipment within the abnormal window; a parameter calculation unit, which calculates the temperature rise per unit time by linearly fitting the slope of the temperature sequence relative to time, and calculates the temperature rise coefficient based on the temperature rise and current sequence; an ambient temperature sequence within a preset short-term time window, and calculates the slope of the ambient temperature change over time by linear fitting to obtain the ambient temperature rise compensation value; a compensation calculation of the temperature rise coefficient using the ambient temperature rise compensation value to obtain the compensated temperature rise coefficient; a screening and analysis unit, which filters data with a compensated temperature rise coefficient greater than a preset fault threshold, records them as fault states, and generates a fault state flag; and a fault state flag, combined with the power supply and distribution system topology map, generates fault location analysis results, which include fault equipment identification, fault type, and warning level.

[0154] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A fault early warning method for power supply and distribution systems based on multi-parameter trend analysis, characterized in that, include: Collect multi-parameter time-series data of the power supply and distribution system, classify the operating conditions and environment, and obtain the classified dataset. The sliding window algorithm is used to extract the temporal features of the classified dataset, and the slope of the parameter change trend within each window is calculated by linear regression to obtain trend feature data. Based on the trend feature data, abnormal trends are identified and abnormal flags and associated parameter sets are generated. The fault location analysis and temperature rise coefficient are performed on the abnormal flags and associated parameter set, and the fault status is identified by combining short-term environmental data to generate fault location analysis results. Based on the fault location analysis results, an early warning signal is generated.

2. The fault early warning method for power supply and distribution systems based on multi-parameter trend analysis according to claim 1, characterized in that, The process of collecting multi-parameter time-series data from the power supply and distribution system, classifying it by operating environment, and obtaining the classified dataset specifically includes: Collect multi-parameter time-series data of the power supply and distribution system, and divide the multi-parameter time-series data into multiple operating environment sub-bins according to the preset operating environment sub-bin rules. Each operating environment sub-bin is uniquely defined by a combination of three dimensions: load level, time period, and ambient temperature. The load level is divided into low load, medium load, high load, and impact load based on the effective value of the current. The divided multi-parameter time series data are stored in bins according to the corresponding working conditions to form a classified dataset. Each dataset contains the parameter sequence of all time points in the same bin.

3. The fault early warning method for power supply and distribution systems based on multi-parameter trend analysis according to claim 2, characterized in that, The process of calculating the slope of the changing trend of parameters within each window through linear regression to obtain trend feature data specifically includes: The size and sliding step of the sliding window are preset based on the data acquisition frequency and trend analysis accuracy requirements, wherein the sliding step is less than or equal to the window size; Extract multi-parameter time-series data points within each sliding window from the dataset. The data points include timestamps and corresponding parameter values. Linear regression fitting is performed on the parameter values ​​and timestamps within each sliding window, and the slope of the regression line is calculated using the least squares method.

4. The power supply and distribution system fault early warning method based on multi-parameter trend analysis according to claim 3, characterized in that, The process of identifying abnormal trends and generating abnormal flags and associated parameter sets based on the aforementioned trend feature data specifically includes: Set thresholds for judging abnormal trends based on historical normal operation data; If the absolute value of the slope is greater than the abnormal trend judgment threshold and multiple consecutive sliding windows meet the condition, it is set as an abnormal trend and an abnormal flag is generated. Extract the parameter set associated with the abnormal trend, and encapsulate the abnormal flag and the associated parameter set into structured data output, wherein the associated parameter set includes the average current, average voltage, average temperature and current-temperature correspondence sequence within the abnormal window.

5. The power supply and distribution system fault early warning method based on multi-parameter trend analysis according to claim 4, characterized in that, The process of performing fault location analysis and temperature rise coefficient calculation on the aforementioned anomaly flags and associated parameter set specifically includes: Based on the average current, average voltage, and average temperature in the aforementioned set of related parameters, and in conjunction with the power supply and distribution system topology, the target electrical equipment corresponding to the abnormal trend is located. Obtain the temperature and current sequences of the main power components of the target electrical equipment within the abnormal window; The temperature rise per unit time is calculated by linearly fitting the slope of the temperature sequence relative to time, and the temperature rise coefficient is calculated based on the temperature rise and the current sequence.

6. The power supply and distribution system fault early warning method based on multi-parameter trend analysis according to claim 5, characterized in that, The process of combining short-term environmental data to identify fault states and generate fault location analysis results specifically includes: The ambient temperature sequence within a preset short time window is obtained, and the slope of the change in ambient temperature over time is calculated by linear fitting to obtain the ambient temperature rise compensation value. The temperature rise coefficient is calculated by using the environmental temperature rise compensation value to obtain the compensated temperature rise coefficient. Data whose compensated temperature rise coefficient is greater than a preset fault threshold are selected, recorded as fault states, and fault state flags are generated. Based on the fault status indicators and combined with the power supply and distribution system topology diagram, fault location analysis results are generated, which include fault equipment identification, fault type, and warning level.

7. The power supply and distribution system fault early warning method based on multi-parameter trend analysis according to claim 6, characterized in that, The process of generating an early warning signal based on the fault location analysis results specifically includes: Based on the warning level and fault type in the fault location analysis results, match the response strategy in the predefined warning response strategy library; Based on the matching response strategy, a multi-level early warning signal is generated, which includes a local visual early warning signal, a local auditory early warning signal, and a remote notification signal. Based on the warning level and the temperature rise coefficient compensation value, the output frequency and duration of the warning signal are adjusted.

8. A power supply and distribution system fault early warning system based on multi-parameter trend analysis, characterized in that, The fault early warning system for power supply and distribution systems based on multi-parameter trend analysis, as described in any one of claims 1-7, specifically includes: The data acquisition and classification module collects multi-parameter time-series data of the power supply and distribution system, classifies it by operating condition and environment, and obtains the classified dataset. The trend feature extraction module uses a sliding window algorithm to extract the temporal features of the classified dataset, and calculates the slope of the parameter change trend within each window through linear regression to obtain trend feature data. The fault probability prediction module, based on the trend feature data, judges abnormal trends and generates abnormal signs and a set of associated parameters; performs fault location analysis and calculates the temperature rise coefficient on the abnormal signs and the set of associated parameters, and combines short-term environmental data to identify the fault state and generate fault location analysis results. The early warning module generates an early warning signal based on the fault location analysis results.

9. The power supply and distribution system fault early warning system based on multi-parameter trend analysis according to claim 8, characterized in that, The trend feature extraction module specifically includes: The window data partitioning unit presets the size and sliding step of the sliding window based on the data acquisition frequency and trend analysis accuracy requirements. The sliding step is less than or equal to the window size. Multi-parameter time-series data points are extracted from the dataset within each sliding window. The data points include timestamps and corresponding parameter values. A linear regression unit is fitted, and linear regression is performed on the parameter values ​​and timestamps within each sliding window. The slope of the regression line is calculated using the least squares method.

10. The power supply and distribution system fault early warning system based on multi-parameter trend analysis according to claim 9, characterized in that, The fault probability prediction module specifically includes: The suspected device location unit locates the target electrical equipment corresponding to the abnormal trend based on the average current, average voltage, and average temperature in the associated parameter set, combined with the power supply and distribution system topology map. The data acquisition unit acquires the temperature and current sequences of the main power components of the target electrical equipment within the abnormal window; The parameter calculation unit calculates the temperature rise per unit time by linearly fitting the slope of the temperature sequence relative to time, and calculates the temperature rise coefficient based on the temperature rise and current sequence; it obtains the ambient temperature sequence within a preset short time window, and calculates the slope of the ambient temperature change with time by linear fitting to obtain the ambient temperature rise compensation value; it uses the ambient temperature rise compensation value to compensate the temperature rise coefficient to obtain the compensated temperature rise coefficient. The filtering and analysis unit filters data whose compensated temperature rise coefficient is greater than a preset fault threshold, records them as fault states, and generates a fault state flag. Based on the fault state flag and combined with the power supply and distribution system topology diagram, it generates fault location analysis results, which include fault equipment identification, fault type, and warning level.

Images