Power distribution station overload risk alarm method and device based on multi-source sensing data fusion

Abstract

The invention discloses a power distribution station overload risk alarm method and device based on multi-source sensing data fusion, and relates to the technical field of intelligent early warning, and the method comprises the following steps: in the operation process of a power distribution station, injecting a micro-amplitude synchronization pulse in a multi-source sensing data collection stage, collecting a pulse feedback result, calculating and collecting clock offset, and constructing a second-level alignment vector; time sequence synchronization among different sensing channels is realized, and a turn-back sensitive baseline is established. According to the method, through multi-source sensing data alignment and turn-back identification, in combination with edge fidelity filtering, heat accumulation calculation and fast and slow variable decoupling, accurate early warning of abnormal load fluctuation is realized, a closed-loop control mechanism of residual feedback is established, and the accuracy and foresight of power distribution station overload risk identification are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent early warning, in particular to a power distribution station overload risk alarm method and device based on multi-source perception data fusion. BACKGROUND

[0002] The "power distribution station overload risk alarm based on multi-source perception data fusion" refers to the process of collecting multiple sources of perception information, such as current, voltage, load power, conductor temperature, equipment status, and environmental temperature and humidity, during the operation of the power distribution station, rather than relying on data from a single monitoring point. These different dimensions of information are processed and analyzed through data fusion technology to form a comprehensive perception of the operating conditions of the power distribution station. On this basis, the system can identify potential risks when the load gradually approaches or exceeds the safety threshold, and issue an overload risk alarm signal in advance by combining trend prediction and dynamic threshold adjustment. This method not only avoids false positives caused by single data anomalies, but also improves the accuracy and real-time performance of overload risk identification, making the operation of the power distribution station more secure and reliable, and providing a basis for early intervention for maintenance personnel.

[0003] The prior art has the following disadvantages: In the prior art, the prediction of power distribution station overload risk mostly relies on dynamic trend modeling methods based on historical load curves. However, when the load experiences severe reverse oscillation in a short period, such as a sharp drop in load in a very short time followed by a rapid rebound, the prediction model in the prior art often treats this abnormal fluctuation as noise or transient disturbance, and then uses smoothing processing to correct the prediction curve. Due to excessive smoothing, the trend curve output by the model shows a low risk or even a stable trend, thus masking the actual continuous rise of the load. Once this problem occurs, the system will not be able to timely trigger an overload risk alarm before the critical state, and eventually the load of the power distribution station may exceed the safety threshold within a few minutes, rapidly amplifying the risk and evolving into an uncontrollable serious accident.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a power distribution station overload risk alarm method and device based on multi-source perception data fusion to solve the problems in the background technology mentioned above.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: a power distribution station overload risk alarm method based on multi-source perception data fusion, comprising the following steps: S001, in the power distribution station operation process, inject a small amplitude synchronous pulse in the multi-source perception data collection stage, collect pulse feedback results, calculate the collection clock offset, construct the second-level alignment vector, realize the timing synchronization between different perception channels, and establish the reentrant sensitive baseline; S002, under the constraint of the reentrant sensitive baseline, according to the multi-source perception data after the second-level alignment, the continuous derivative sequence is calculated by using the sliding window method, the fluctuation mode of first falling and then rising is identified, the fluctuation segment exceeding the set amplitude threshold is screened, and the reentrant window list is constructed; S003, within the constraint range of the reentrant window list, edge fidelity filtering processing is performed, the trend smoothing across the reentrant section is suppressed by dynamically setting the filtering window range, the waveform slope of the rising section is retained, the peak top cutting influence is eliminated, and the enhanced trend data is output; S004, after completing the edge fidelity filtering processing, the freezing operation is performed on the temperature type perception data, the change path decoupling of fast variable and slow variable is realized, the heat accumulation value is generated by using the current square time accumulation calculation method, the heat guardian threshold is constructed, and the overload risk level is improved; S005, after the overload risk level is improved, the weighting coefficients of the multi-source perception channels are redistributed, the weights of current and power type data are increased in the reentrant window, and the weights of temperature and environment type data are reduced, and the pre-warning index is generated as the alarm trigger leading signal; S006, after obtaining the pre-warning index, the residual change between the trend prediction data and the actual measurement data is compared, the threshold parameters, the sliding window width and the perception channel weighting coefficients are dynamically updated, and the updated parameter set is written into the reentrant sensitive baseline, completing the closed-loop risk alarm process from perception, identification, processing to regulation.

[0007] Preferably, step S001 comprises: The preset current sensor, voltage sensor, load detection device, conductor temperature sensor, device running state acquisition device and environment temperature and humidity perception equipment have the ability to receive external small amplitude synchronous pulse and record time stamp; In the initial stage of data collection, the central control unit sends a synchronous pulse signal with a voltage amplitude lower than 100 millivolts and a duration less than 10 milliseconds to each perception channel, collects the receiving time stamp of each channel feedback and performs time difference calculation; In the continuous multiple synchronous pulse periods, the sampling clock offset law of each channel is counted, the second-level alignment vector for unified time reference correction is generated, and the reentrant sensitive baseline based on the unified standard time is constructed, which provides timing consistency guarantee for subsequent abnormal fluctuation identification. According to the second-level alignment vector, the time stamp of all channel sampling data is adjusted, the data time axis is corrected by time translation or interpolation method, and the reentrant sensitive baseline based on the unified standard time is constructed, which provides timing consistency guarantee for subsequent abnormal fluctuation identification.

[0008] Preferably, step S002 comprises: Based on the time-aligned current channel, load power channel and conductor temperature channel, the sliding window is divided in a way of sampling once per second, the window width is set to 30 seconds, and the step is 1 second, and the continuous derivative sequence in each window is calculated; Identify whether there is a fluctuation mode formed by the combination of continuous descending section and continuous rising section in each window, and require that each section change is not less than 3 sampling points, and the amplitude exceeds 8% of the rated current value respectively, and set the time interval between the front and rear sections to be not less than 5 seconds; For the windows that preliminarily meet the conditions, the load power channel and the conductor temperature channel are called to perform consistency verification in the same time interval, to confirm that the power fluctuation direction is consistent and the amplitude exceeds 10% of the rated value, and the temperature channel rises by more than 1.5 degrees Celsius within 10 seconds after the end of the window; The time period that meets the multi-channel verification condition is arranged as a structured turn-back window list, and the window start and end time, channel number, change amplitude and verification result are recorded as the input basis for subsequent trend retention and risk identification.

[0009] Preferably, step S003 comprises: Extract the start time, end time, channel type and fluctuation characteristics of each turn-back window, and intercept the corresponding time period in the original perception data and extend 1 second before and after as a protection area; The sliding window width is dynamically set according to the derivative maximum value and time span of the turn-back section, and the continuous filtering processing is performed in a weighted average way, and the weight gradually decreases from the center of the window to both ends; The slope protection mechanism is enabled for the continuous sampling points in the rising trend section, and when three groups of continuous positive difference values are met and the amplitude is not less than the rated threshold, the filtering is suspended and the original data is retained; The data detected as peak points is not modified, the filtering result is embedded into the original data sequence after completing all turn-back section processing, and linear interpolation is performed in the front and rear 2 second buffer areas to ensure data continuity.

[0010] Preferably, step S004 comprises: Extract the perception data sequence after edge fidelity filtering processing, and obtain the conductor temperature, equipment shell temperature and environment temperature data in the turn-back window corresponding time period, and keep the latest valid temperature value before the start of the window unchanged in the window; Perform square operation on the current perception data sampled per second in the turn-back window, and accumulate all current square values, calculate the current square time integral value as the heat input; The current square time product value is compared with the heat capacity threshold value to divide the overload risk levels into first, second and third levels according to the matching of the equipment model, cable parameters and ventilation conditions and the heat capacity carrying capacity of the equipment. The risk level, start and end time, current peak value and freezing temperature corresponding to each turn-back window are written into a turn-back window list for use in subsequent dynamic weight adjustment and early warning index generation steps.

[0011] Preferably, step S005 comprises: According to the risk level corresponding to the heat value in the turn-back window, the weighting coefficient of the perception channel is dynamically adjusted, the weight of the current channel and the power channel is increased to not less than 1.5, and the weight of the conductor temperature channel and the environment temperature channel is decreased to not more than 0.6; The current, power, conductor temperature and environment temperature data at each sampling time in the turn-back window are multiplied by the corresponding adjusted weight, and the weighted values are summed to generate a continuous weighted risk data sequence; The normalized interval of each perception channel is determined using the historical data within 24 hours, the weighted risk data sequence is standardized to generate a pre-warning index distributed in the interval of 0 to 100; The pre-warning index is compared with the set risk classification threshold value, if the index exceeds the preset threshold value for three consecutive sampling periods, the current turn-back window is marked as an abnormal continuous development state, and the weight configuration and index evolution trajectory are recorded.

[0012] Preferably, step S006 comprises: The residual error between the trend prediction data and the actual perception data in the turn-back window is calculated to generate a residual error change sequence, and the prediction model deviation is evaluated; The alarm parameters are adjusted according to the residual error change result, including lowering the excess alarm threshold of current and power, compressing the sliding window width, increasing the weight of fast variable channel and simultaneously decreasing the weight of slow variable channel; The parameter adjustment result of each time is recorded as an independent parameter set, including the turn-back window number, residual error statistical value, weight before and after updating, window width and alarm threshold change; The independent parameter set is written into the turn-back sensitive baseline according to the time stamp, and is used as the priority configuration basis for the next round of risk judgment and perception preprocessing operation, realizing the closed-loop adaptive control of the alarm process.

[0013] The power distribution station overload risk alarm device based on multi-source perception data fusion comprises a data alignment module, a turn-back identification module, a fidelity filtering module, a variable decoupling and risk modeling module, a weighting adjustment and early warning module, and a closed-loop control module. The data alignment module injects a micro-synchronous pulse in the multi-source perception data collection stage during the operation of the power distribution station, collects pulse feedback results, calculates clock offset, constructs a second-level alignment vector, realizes timing synchronization between different perception channels, and establishes a rollover sensitive baseline; The rollover identification module, under the constraint of the rollover sensitive baseline, calculates a continuous derivative sequence according to the multi-source perception data after the second-level alignment, identifies a fluctuation mode of first falling and then rising, screens fluctuation segments exceeding a set amplitude threshold, and constructs a rollover window list. The fidelity filtering module, within the constraint range of the rollover window list, performs edge fidelity filtering processing, suppresses trend smoothing across rollover segments, retains waveform slope of the rising segment, eliminates peak top cutting influence, and outputs enhanced trend data. The variable decoupling and risk modeling module, after the edge fidelity filtering processing, performs freezing operation on temperature type perception data, realizes decoupling of change paths of fast variables and slow variables, generates heat accumulation value by using current square time accumulation calculation method, constructs heat guardian threshold, and improves overload risk level. The weighted adjustment and early warning module, after the overload risk level is improved, reallocates weighting coefficients of the multi-source perception channels, increases weights of current and power type data in the rollover window, reduces weights of temperature and environment type data, generates a pre-warning index as a warning trigger leading signal. The closed-loop regulation module, after obtaining the pre-warning index, compares residual changes between trend prediction data and actual measurement data, dynamically updates threshold parameters, sliding window width and perception channel weighting coefficients, and writes the updated parameter set into the rollover sensitive baseline, to complete the closed-loop risk warning process from perception, identification, processing to regulation.

[0014] In the above technical solution, the present application has the following technical effects and advantages: The present application accurately captures abnormal patterns of load sudden drop and sudden rise by using timing alignment and rollover feature identification of multi-type perception data, retains key rising trend by edge fidelity filtering, avoids covering risk signals by smoothing processing, introduces heat accumulation and fast-slow variable decoupling mechanism to improve the accuracy of physical risk state determination, further generates pre-warning index and residual feedback correction to establish closed-loop adaptive control logic, and significantly enhances the identification sensitivity and pre-warning foresight of the system to abnormal fluctuations. Compared with the prior art, the present application has the beneficial effects of earlier warning, faster response and more accurate risk identification, and significantly improves the operation safety and risk control ability of the power distribution station under complex dynamic load. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed in the embodiments will be briefly introduced as follows. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained based on these drawings.

[0016] Figure 1 The method flow chart of the power distribution station overload risk alarm method based on multi-source perception data fusion of the present application.

[0017] Figure 2 The module schematic diagram of the power distribution station overload risk alarm device based on multi-source perception data fusion of the present application. DETAILED DESCRIPTION

[0018] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art.

[0019] The present application provides a power distribution station overload risk alarm method based on multi-source perception data fusion as shown in Figure 1 The power distribution station overload risk alarm method based on multi-source perception data fusion includes the following steps: S001, in the process of running the power distribution station, injecting a micro-amplitude synchronous pulse in the multi-source perception data acquisition stage, collecting pulse feedback results, calculating the acquisition clock offset, constructing a second-level alignment vector, realizing the time sequence synchronization between different perception channels, and establishing a turnaround sensitive baseline; In order to realize the high-precision alignment of various perception data in the time dimension in the power distribution station, solve the data fusion offset problem caused by the asynchronous sampling clock of each channel, and ensure the accurate perception of short-period load changes and the accuracy of subsequent overload risk identification, the following specific steps are proposed: In the normal operation process of the power distribution station, for a plurality of sensor channels including current sensors, voltage sensors, load detection devices, conductor temperature sensors, device operating state acquisition devices, and environmental temperature and humidity sensing devices, each data acquisition channel is preset to have hardware circuit capable of receiving external pulse signals and timestamp recording capability. In the initial stage of multi-channel data acquisition, through the unified scheduling of the central control unit, a set of low-amplitude synchronous voltage pulses are sent to all sensing channels at a preset period (for example, 1 second) through physical connection lines or industrial wireless communication protocols (such as RS485, CAN or wireless HART). The voltage amplitude of the pulse is controlled to be lower than the lower limit of the measurement range of various sensors, for example, less than 100 millivolts, and the pulse duration is set to be less than 10 milliseconds to ensure that it does not interfere with the transmission data. This reference pulse signal is injected into the receiving pin of the front end of each channel, triggering the local clock to record the receiving time, and forming a mapping relationship with the local sampling clock through the internal high-precision timer. After receiving the reference pulse signal, the sampling channel immediately generates a timestamp of the current time and feeds it back to the central control unit through the communication network.

[0020] After receiving the timestamps returned by all sensing channels, the central control unit selects the theoretical sending time of each set of reference pulses as the standard time, compares the receiving time points fed back by each channel, and calculates the time difference between each channel and the standard time. To ensure that the time difference reflects the offset behavior of the channel sampling clock itself, rather than accidental network delay or signal propagation jitter, the central control unit will repeat this time offset measurement operation in multiple consecutive reference pulse periods to obtain the stable time offset trend of each channel. The time offset trend reflects the systematic difference between the internal sampling clock of the channel and the standard time reference, such as caused by slight deviation of crystal oscillator frequency, circuit response delay, sampling start delay, etc. Based on the above multi-period time offset measurement results, the central control unit quantifies the actual clock offset value of each channel at the second level, forming a set of alignment vectors representing the difference between the channel and the unified standard time, which is used for subsequent data time reconstruction operations.

[0021] After obtaining the time alignment vector of all channels, the original sampling data from all perception channels is corrected on the time axis. The central control unit adjusts the data timestamp of each channel according to the time offset in the aforementioned alignment vector. The adjustment methods include interpolation adjustment, data translation or resampling processing, wherein for frequency consistent perception data, the original sampling points can be moved forward or backward by a corresponding amount of time by translation; for channels with non-uniform sampling or missing points, new equally time interval sampling points can be generated based on the interpolation of the front and rear sampling values. The adjustment process is strictly based on a unified standard time as a reference, ensuring that each corrected data corresponds to the same actual physical time. After adjustment, the data will be reorganized according to the unified time sequence to form a perception data set that is completely time-aligned. This process is not only suitable for high-frequency perception data such as current and voltage signals, but also covers lower frequency parameters such as conductor temperature, environmental temperature and humidity, thereby forming a cross-channel and cross-dimension synchronous data sequence on a unified time base.

[0022] After the above-mentioned time alignment and data reconstruction are completed, the central control unit will use the adjusted data as an important basis for subsequent identification of short-period dramatic fluctuations and construct a reentry sensitive baseline. The construction of the reentry sensitive baseline is based on the alignment vector and the dynamic characteristics of the aligned data, by analyzing the time synchronization degree and change trend of each channel data within a certain time sliding window, to judge the synchronous disturbance characteristics of the data on a small time scale. Whenever a channel has a significant increase in change rate within a certain time window, and the change time is time-offset from other channels, the synchronous deviation evaluation index of the reentry sensitive baseline is updated, so as to judge whether the change is a real physical phenomenon or a false fluctuation caused by time inconsistency. The role of the baseline is to provide a time reference for the subsequent reentry feature recognition process, and through the continuous dynamic updating of the baseline, the recognition accuracy of short-time abnormal fluctuations can be improved, and the sampling error interference can be prevented. The fluctuation recognition problem caused by the different time of the existing technology is fundamentally solved.

[0023] The role of this step is to achieve high-precision synchronization of data from multiple sensing channels in the power distribution station in time, to ensure that real-time information collected by different types of sensors has a unified time reference, thereby eliminating the problem of data misplacement caused by sampling clock offset. During the operation of the power distribution station, current, voltage, load power, conductor temperature, environmental temperature and humidity, and other parameters are often collected through different devices, different sampling frequencies, and different communication methods. The sampling start point, sampling period, and signal transmission delay between these channels differ, which can easily lead to inconsistent data when fused and analyzed. This step can effectively correct the time difference between channels by injecting a small amplitude synchronization pulse, establishing a time offset model, reconstructing data, and constructing a foldback sensitive baseline, so that all sensing data is aligned on a unified time axis, providing a high-precision timing input basis for subsequent foldback fluctuation identification, trend tracking, and overload risk judgment. This approach avoids the errors caused by relying on a single channel timestamp or static smoothing in existing technologies, improving the accuracy of data fusion and the sensitivity of anomaly identification.

[0024] S002, under the constraint of the foldback sensitive baseline, a sliding window method is used to calculate a continuous derivative sequence based on the second-level aligned multi-source sensing data, to identify a fluctuation pattern that first decreases and then increases, to screen fluctuation segments that exceed a set amplitude threshold, and to construct a foldback window list; To accurately identify possible short-period intense reverse fluctuation behavior in the operation of the power distribution station based on time-aligned multi-source sensing data, and effectively locate abnormal data segments with overload risk, a technical method for constructing a foldback window list is proposed to support subsequent trend correction and risk warning processing. This method is based on second-level synchronized multi-channel sensing data to identify continuous fluctuation intervals that meet the "first decrease and then increase" feature, and to verify their reliability through multi-dimensional verification. The specific implementation is as follows: After completing the time alignment of the sensing data channels of the power distribution station, key sensing indicators are selected for fluctuation identification. The selected sensing channels include current channels (to reflect the actual power transmission intensity), load power channels (to depict user-side energy demand changes), and conductor temperature channels (to track thermal load conduction responses), with a sampling frequency of 1 Hz, i.e., one data point collected per second. These sensing data are divided into sliding windows according to a unified time axis, with each window containing 30 consecutive sampling points, i.e., a window width of 30 seconds and a sliding step of 1 second. Within each window, the difference between adjacent sampling points is calculated in time sequence order to form a derivative sequence containing 29 difference values, which is used to evaluate the change direction and rate of the data in the window. A negative difference value indicates a downward trend, a positive difference value indicates an upward trend, and a transition from consecutive negative values to consecutive positive values may constitute the starting point and endpoint of a foldback behavior.

[0025] In the obtained derivative sequence, it is analyzed whether there is a clear fluctuation structure in the window, that is, the previous part is a continuous downward process, the latter part is a continuous upward process, and the overall change amplitude exceeds the preset threshold. Taking the current channel as an example, the judgment condition is set as follows: at least 3 consecutive negative values are contained in the downward segment, and the maximum downward amplitude is not less than 8% of the current rated current; then at least 3 consecutive positive values must appear, and the maximum upward amplitude is also not less than 8% of the current rated current. In addition, in order to ensure that the identified fluctuation segment does not belong to sensor measurement error or transient disturbance, the time interval between the downward segment and the upward segment must be greater than 5 seconds, that is, it must be a structural continuous reverse change behavior. The window that meets the above judgment condition is preliminarily judged as a suspected turning window, and the starting time, ending time and maximum change amplitude are marked and stored in the internal structure.

[0026] In order to improve the identification accuracy and exclude single channel misjudgment, the time period preliminarily judged as the turning window is verified by multiple channels. Taking the time interval of the window as the basis, the data sequences of the load power channel and the conductor temperature channel in the same time range are called, and the same derivative structure analysis is performed respectively. In the load power channel, it is checked whether there is a turning fluctuation structure in the same direction as the current channel; in the conductor temperature channel, it is checked whether there is a temperature continuous rise phenomenon in the 3 to 10 seconds after the end of the turning window, indicating that there is a real thermal load response. If the fluctuation behavior consistent with the current channel structure is found in the two auxiliary channels respectively, and the change amplitude exceeds the respective preset threshold (for example, the power change exceeds 10% of the rated power, and the temperature rise exceeds 1.5 degrees Celsius), the turning window is confirmed to have physical reasonableness, and is transferred from the candidate list to the formal turning window set.

[0027] All the verified and confirmed turning windows are summarized and arranged to construct a complete turning window list. The list stores the time period information of each turning window in a structured format, including the starting time, the ending time, the channel number, the downward segment amplitude, the upward segment amplitude, the maximum derivative value, the related auxiliary channel matching condition and the like. The list will be used as a boundary reference in subsequent processing to prevent data from being misidentified as transient noise and being smoothed in the smoothing filtering process. Especially before the dynamic trend prediction model processing, the list will be used as an input control parameter to indicate the structural area that the prediction model needs to retain, so as to improve the true accuracy of trend identification, and provide traceable basis support for subsequent risk level assessment and dynamic adjustment of warning threshold.

[0028] The role of this step is to accurately identify the abnormal fluctuation behavior that may occur in the short period of the substation operation from the multi-source perception data that has completed time alignment, and extract the fluctuation segment with actual physical meaning to construct a turnaround window list for subsequent trend correction and risk warning processing. In actual operation, key indicators such as current, load power, etc. may produce violent reverse oscillation in a very short time due to sudden load changes. If such behavior is not identified in time, it is easy to be misjudged as noise by subsequent smoothing processing, resulting in the real risk signal being covered. This step ensures that each turnaround fluctuation extracted has time continuity, significant change amplitude and multi-channel consistency by sliding window division, derivative sequence calculation, fluctuation structure judgment and cross-channel verification, and has a real physical background and operation basis. The constructed turnaround window list can be used as a marker benchmark to retain key fluctuation structures, prevent deviation of system risk situation judgment caused by false smoothing processing, provide high-precision input boundary for dynamic filtering strategy, threshold adjustment and early warning index generation, and significantly improve the real-time and accuracy of substation overload risk identification.

[0029] S003, within the constraint range of the turnaround window list, edge fidelity filtering processing is performed, the trend smoothing across the turnaround section is inhibited by dynamically setting the filtering window range, the waveform slope of the rising section is retained, the peak value is eliminated, and the enhanced trend data is output; To maximize the retention of the real trend characteristics of the data during the turnaround fluctuation, avoid the weakening or covering of key features by traditional smoothing processing, and enhance the authenticity of the trend data and the accuracy of subsequent risk assessment, this step proposes an edge fidelity filtering method based on the turnaround window list for directional filtering and correction of the turnaround section in the load, current, power and other perception channels. This method, aiming to achieve data continuity and readability without destroying the structure, proposes a complete processing flow combining boundary restriction, slope retention and transition compensation. The specific implementation is as follows: On the basis of the turnaround window list constructed in the previous step, the start time, end time, involved channel type, maximum fluctuation amplitude and derivative change rate of each turnaround record are extracted. Taking the current perception channel as an example, assuming that the start time of a turnaround window record is 08:12:30 and the end time is 08:12:58, the time span is 28 seconds. According to this time period, 29 consecutive sampling points from 08:12:30 to 08:12:58 are extracted from the original sampling data in the current channel, with a sampling frequency of 1 Hz, i.e. one data point per second. In order to prevent edge information loss, 1 second is added to both ends of the window as a protection area, finally forming a complete processing section containing 31 points. Each section of data will be independently filtered in the subsequent processing process to ensure that the window features are not covered by the overall trend.

[0030] The filter window parameters are dynamically set according to the duration of each turn-back window and the data fluctuation characteristics. If the fluctuation rate of a certain turn-back section is high (e.g., the maximum derivative is more than 15% of the rated current), the filter window width is set to 5 seconds, i.e., 5 sampling points are processed each time; if the fluctuation rate is low but the time span exceeds 30 seconds, the filter window width is set to more than 10 seconds to improve the overall trend smoothing capability. In actual operation, the filtering process is performed in time series, and the data segment in each window is processed from left to right. In each position of the sliding window, a weighted average processing is performed on 5 sampling points, and the weighting factors are set to decrease from the center to the two ends according to the position of the data in the window, with the center point having the maximum weight and the edge point having the minimum weight, to enhance the response capability to the fluctuation center section and weaken the influence of the edge data.

[0031] In the sliding processing process, a slope protection mechanism is used for continuous data points located on the rising slope of the turn-back section. Under this mechanism, before each sliding calculation, the rising trend of the data in this section is first evaluated, and the judgment standard is that the three consecutive difference values are all positive, and each difference value is not less than a set threshold, for example, the rising amplitude per second is not less than 2% of the rated current. When the above conditions are met, the weighted average processing of this section is suspended, and the original data points are directly retained and used as part of the trend output. This mechanism can effectively prevent the traditional sliding processing method from flattening the slope and avoid weakening the trend rising process. At the same time, when a peak point is detected in the window (i.e., the derivatives of the two preceding data points are positive and the derivatives of the two subsequent data points are negative), the original data is kept unchanged and any processing of the peak point is prohibited, thereby preventing the problem of cutting off the peak of the trend.

[0032] After completing the independent processing of all turn-back windows, each filtered trend data segment is embedded back into the original perception data sequence and accurately replaces the corresponding time period. To eliminate the numerical mutation or discontinuity that may occur at the splicing place of the front and rear data, a transition buffer zone is set at both ends of each turn-back section, with a transition length of 2 seconds, and the connection value of the transition section is calculated by linear interpolation, so that the processed section and the context data are naturally connected and have good continuity. Finally, a set of trend data is formed, which retains the real structural characteristics, suppresses invalid high-frequency jitter, has good smoothness and time sequence continuity, and provides an accurate data basis for subsequent early warning index generation and risk level judgment.

[0033] The purpose of this step is to perform a fidelity filtering process with boundary control capability on the abnormal fluctuation intervals identified and located, to retain the key rising trend slope and peak shape, and prevent the trend information from being weakened or cut off in the data smoothing process. Traditional moving average or median filtering methods often flatten the data edges and lower the peaks when dealing with sharp fluctuations, resulting in the abnormal trend of key indicators such as load and current being hidden, which affects the accuracy of overload risk identification. To solve this problem, this step sets an independent filtering range for each abnormal fluctuation interval based on the time boundary of the turnaround window, and dynamically adjusts the size and weighting of the sliding window. At the same time, a slope protection mechanism and peak retention strategy are introduced to ensure that the data is not misprocessed in the rapid rising stage. The filtered data is seamlessly spliced with the context data through transition compensation, forming a complete perception sequence with enhanced trend. This step provides a structured and clear trend data basis for subsequent risk level enhancement and warning indicator extraction, while maintaining data smoothing and significantly improving the perception ability of abnormal situations.

[0034] S004, after completing the edge fidelity filtering process, a freezing operation is performed on temperature-based perception data to decouple the change paths of fast and slow variables, generate heat accumulation values using current square time accumulation calculation, construct heat guardian thresholds, and enhance the overload risk level; To improve the risk identification accuracy after edge fidelity filtering and avoid misjudgment of the risk level during the turnaround period due to the response lag of temperature data, a fast and slow variable decoupling and dynamic risk enhancement method based on current heat effect is further proposed in the multi-source perception data processing flow of the distribution station. This method freezes the lagging variables, extracts the intensity accumulation characteristics of fast variables, and constructs heat guardian thresholds based on the actual device operating capacity, to accurately and dynamically enhance the potential risk level within the turnaround window. The specific implementation is as follows: After completing the previous edge fidelity filtering process, the corrected perception data sequence output by this step is obtained, and the conductor temperature, device shell temperature, and environmental temperature data within the time period corresponding to the turnaround window are extracted. These temperature data are usually collected by thermistor, platinum resistance, or thermocouple sensors, and their response speed is affected by heat capacity, medium conduction efficiency, and measurement loop lag, generally falling behind the dynamic response of fast variables such as current and voltage. In actual engineering, even if the current has experienced sharp fluctuations, the temperature data often shows a slow rise or no significant change. To prevent this lag phenomenon from interfering with risk identification, a freezing process is performed on temperature-based perception data within the time period corresponding to the turnaround window, i.e., a recent valid temperature value is obtained before the window start time, the temperature value remains unchanged within the window, and the normal collection value is restored at the end of the window, to ensure that the temperature input within this time period remains stable and avoids misjudgment of low risk due to delayed reflection of the true risk by lagging data.

[0035] While the temperature data is frozen, the corresponding current-aware data in the folding window is subjected to thermal intensity calculation processing. In a specific implementation, the sampling value of the current channel per second is selected, and a square operation is performed on each data point to obtain the instantaneous thermal contribution value of the current intensity per second. Subsequently, the current square values in the entire folding window are added item by item to form the current square time product value in the window. Taking an actual power distribution device as an example, if the window time length is 28 seconds, 28 current square values will be generated in the window, and the total value after accumulation represents the approximate heat input of the current to the conductor and the internal heating of the device in this time period. This value is not affected by the lag of temperature changes and can reflect the energy load level that the device bears in real time. This calculation method follows Joule's law, that is, the heat generated by the current passing through the conductor is proportional to the square of the current and time, and is a reliable indicator for evaluating the thermal stress of the device.

[0036] After obtaining the current square time product value, the corresponding thermal capacity carrying capacity is matched by referring to the engineering design parameters such as device type, cable cross-sectional area, conductor material, and ventilation conditions, and a heat guard threshold is constructed. Taking a 10-kilovolt switch cabinet as an example, according to its rated continuous operation capacity and short-time carrying capacity, the upper limit of the heat threshold is set, for example, a certain type of switch cabinet is allowed to accumulate heat input of a certain specific value within 5 minutes. Compare the heat product calculated in the current folding window with the upper limit value, and divide the risk level interval in proportion: when the heat product exceeds 70% but is less than 90% of the allowed upper limit, the risk level is determined as "first-level warning"; when the heat product reaches between 90% and 100%, the risk level is determined as "second-level warning"; when the heat product exceeds 100%, that is, exceeds the allowed heat capacity range of the device, it is determined as "third-level warning", and is considered as a significant overload risk. In this way, the system has a quantitative standard and grading basis for responding to load fluctuations in the thermal effect level, avoiding rough judgment based only on current instantaneous values or temperature lag data.

[0037] After completing the heat product calculation and risk level determination, the corresponding risk level result is recorded in the folding window list, and the start and end time, current peak value, and temperature frozen value of the window are stored together. At the same time, the risk level will be passed to the subsequent weight adjustment and warning index generation steps to increase the reference proportion of current indicators during the folding period and reduce the proportion of temperature data in risk judgment, forming a dynamic weighting mechanism. The system can also combine the risk level sequence of a plurality of consecutive folding windows to determine whether there is a cumulative overload trend. If there are consecutive high-level risks, an alarm signal can be further sent to the upper-layer operation terminal. Through the above process, a complete processing flow from structure trend fidelity, lag data freezing, thermal effect measurement to risk level quantization and promotion is realized, ensuring that potential overload risks can be identified and responded to in advance in the case of severe load fluctuations.

[0038] The purpose of this step is to improve the timeliness and accuracy of overload risk judgment during the return window, and to solve the problem of response lag and early warning delay in traditional temperature threshold judgment. Because temperature perception data usually has strong inertia and heat capacity effect, it cannot reflect the rapid changes of current or load in a short time, which easily leads to misjudgment of low risk when the actual overload is about to occur. Therefore, this step freezes temperature data within the return window to block its lagging influence, and introduces a heat evaluation method based on current square time product, which calculates the total heat load of the device in this time period to form a real-time quantitative evaluation of the thermal stress of the device. Then, combined with the structural parameters and heat capacity of the device, a heat guardian threshold is constructed to classify the risk level and realize the dynamic promotion of risk from "normal" to "warning" to "serious". This mechanism not only realizes the decoupling processing of fast and slow variables, but also introduces physical level risk judgment basis, effectively improving the response speed of the system to the overload trend, and providing a scientific and real-time basis for subsequent early warning decision.

[0039] S005, after the overload risk level is promoted, the weighting coefficients of the multi-source perception channels are redistributed, the weights of current and power data are increased in the return window, and the weights of temperature and environmental data are reduced, to generate a pre-warning index as a pre-alarm trigger signal; To improve the accuracy of overload risk perception during the return window, and to identify the hidden abnormal state that may evolve into a fault in advance, after completing the risk level promotion operation based on heat product, the risk discrimination contribution of each type of perception data needs to be dynamically reconstructed, so as to generate a pre-warning index with predictive and trend response ability. Through the weighted allocation of current, power, temperature and environmental parameters, the response ability of fast variables is enhanced and the lagging interference of slow variables is weakened, which can effectively improve the forward-looking and accuracy of the overall alarm system. The specific implementation is as follows: In the above step, the calculation of the heat accumulation in the return window has been completed, and the heat value is compared with the corresponding device heat capacity threshold to determine the state to which the current risk level belongs. Based on the risk level result, the dynamic weight adjustment of the perception channel involved in the current return window is immediately carried out. Taking a 10kV switch cabinet of a certain power distribution station as an example, its corresponding perception channels include current channel (sampling frequency is 1Hz, unit is ampere), active power channel (unit is kilowatt), conductor temperature channel (unit is Celsius) and environment temperature channel (unit is Celsius). When the risk level is in the first warning state, the weight of the current channel is increased from the base value 1.0 to 1.8, the weight of the power channel is increased from 1.0 to 1.5, the weight of the conductor temperature channel is reduced from 1.0 to 0.6, and the weight of the environment temperature channel is reduced from 1.0 to 0.5; if the risk level is further increased to the second or third warning, the weight of the fast variable is increased by a certain proportion until 2.0, and the proportion of the slow variable channel is reduced at the same time, so as to ensure that the fast response index occupies the dominant position in the fluctuation critical period.

[0040] According to the adjusted channel weight described above, the perception data at each time in the return window is weighted. The operation process is: starting from the starting time point of the window, read the sampling values corresponding to each channel every second; multiply the current value by the adjusted weight of the current channel, multiply the power value by the adjusted weight of the power channel, and multiply the conductor temperature value and the environment temperature value by the reduced weight coefficient respectively. Taking a certain time as an example, if the sampling values are current 220 amperes, power 65 kilowatts, conductor temperature 56 degrees Celsius, and environment temperature 40 degrees Celsius, and the corresponding weights are 2.0, 1.5, 0.6, and 0.5 respectively, then the weighted data is current weighted value 440, power weighted value 97.5, conductor temperature weighted value 33.6, and environment temperature weighted value 20. Superimpose the four weighted values to obtain an instantaneous risk influence total value, which is used to represent the abnormal feature intensity at that time. Apply this processing operation to each sampling time in the entire return window to form a continuous weighted risk data sequence, which completely describes the risk response change process of the adjusted dominant structure of the perception parameter in the return window.

[0041] After obtaining the complete risk data sequence, it is standardized and a pre-warning index is constructed. The standardization step is: first, according to the historical data of each channel in the past 24 hours, the historical maximum and minimum values of current, power, temperature and other data are calculated to construct a normalization interval; then the foregoing weighted data sequence is converted according to the normalization interval and compressed to the index interval of 0 to 100. In this way, the pre-warning index constructed has not only the ability of unified numerical representation, but also can be compatible with the dimensional difference of multiple channels, and improve the actual interpretation availability. In addition, in order to enhance the operability of the index in the warning application, the index interval is further divided into risk areas, for example, 0 to 30 represents stable operation, 30 to 60 represents risk rising trend, 60 to 80 represents entering the warning state, and above 80 is considered as high risk state, which needs to trigger the alarm decision. The pre-warning index is based on the trend expression formed after the risk sensitive channel is strengthened during the turnaround window, which can provide a warning basis before the fault actually occurs.

[0042] The pre-warning index is fed back to the alarm logic of the current processing period and serves as a leading basis for subsequent alarm triggering. In the control process, if the pre-warning index is higher than the preset warning threshold (such as 60) for three consecutive sampling periods and is still within the turnaround window time range, the system automatically marks the current window as an abnormal continuous development state, and records the weight adjustment configuration and pre-index evolution trajectory of the period for subsequent closed-loop adjustment and alarm strategy updating. At the same time, if the current index rises rapidly but the temperature channel has no significant response, further heat accumulation prediction process can be started to confirm whether the current load has reached the upper limit of heat capacity. Throughout the process, the pre-warning index not only provides a warning precursor judgment basis, but also becomes an important result output for the dynamic evolution of the dominant right in the risk modeling process.

[0043] The role of this step is to adjust the participation intensity of different types of sensing channels in risk judgment after the overload risk level is raised during the turnaround window, to enhance the dominant ability of current and power and other fast variables on the risk index by dynamically allocating weights, and to weaken the lagging influence of temperature and environment and other slow variables, so as to construct a pre-warning index with better real-time and predictability. In the operation of power distribution station, the traditional alarm mechanism often relies on fixed threshold and static index, which is difficult to identify abnormal fluctuation trend in short period in time, especially in the case of load surge but temperature has not responded. Through the dynamic reconstruction of the weighting structure of the sensing data in this step, the system can actively improve the response sensitivity to key indicators according to the real-time risk level, and construct a unified quantitative pre-warning index to provide a leading signal for the alarm logic. The index has the characteristics of continuous tracking, hierarchical management and sensitive response, which helps to discover abnormal situation that may evolve into overload in advance, and improves the initiative and reliability of risk identification and warning decision of the whole power distribution station.

[0044] S006, after obtaining the early warning index, the residual change between the trend prediction data and the actual measurement data is compared, the threshold parameter, the sliding window width and the perception channel weighting coefficient are dynamically updated, and the updated parameter set is written into the turn-back sensitive baseline, completing the closed-loop risk alarm process from perception, identification, processing to regulation; To construct an overload risk alarm process with closed-loop regulation capability, after obtaining the early warning index, the deviation between the prediction model output and the real observation result needs to be continuously evaluated in combination with the running track of the index and the actual perception data. By analyzing the residual change in each period and dynamically updating the risk judgment related parameters including the alarm threshold, the sliding window width and the perception channel weighting coefficient according to the analysis result, the update result is finally written into the turn-back sensitive baseline to complete the closed-loop chain from perception, identification, processing to regulation. The specific implementation is as follows: During the turn-back window, the system has generated a corresponding early warning index sequence and recorded its change track with a second-level granularity. The early warning index is obtained by weighting and reconstructing the current, active power, conductor temperature and environmental temperature perception channels in the foregoing steps, and its value reflects the abnormal aggregation strength of the overall perception data in this period. To judge the effectiveness of the early warning index, it is compared with the actual perception data in this period. Taking the current channel as an example, the trend prediction curve of the current turn-back window is extracted, that is, the time-periodic upward trend reference value generated by the sliding window method; the measured current data in the same time period is extracted, the difference between the prediction curve and the measured value is calculated every second, and a residual change sequence is formed. The residual change sequence is used to quantify the deviation between the model expectation and the real operation. If the residual is found to be positive for a long time, it means that the system has underestimated the load rising speed; if the residual change is frequent and volatile, it means that the model response to fluctuations is unstable, and there is a risk of false alarm or missed alarm.

[0045] According to the above residual analysis results, the adaptive updating process of the risk judgment parameter is started. First, the alarm threshold is adjusted: if the residual average value is greater than zero, it means that the trend prediction is generally low, at this time the current alarm threshold is adjusted from the original 10% excess to 8% excess; the power threshold is adjusted from 15% to 12%. Second, the sliding window width is adjusted: if the residual peak value changes at a rate exceeding the set stability threshold, for example, exceeding 20% amplitude change within 5 seconds, it means that the window width is too wide, and the original 30-second sliding window needs to be compressed to 20 seconds to capture the short-period fluctuation characteristics in time. Third, the perception channel weight configuration is adjusted: if the current and power channel residual value proportion exceeds 70% in this time period, and the temperature and environment channel change lags, then the current channel weight is increased from 2.0 to 2.2, the power channel weight is increased from 1.8 to 2.0, and the conductor temperature channel weight is reduced from 0.6 to 0.5, and the environmental temperature weight is reduced from 0.5 to 0.4. Through the above parameter update, the response strength of the model to fast variable driving risks can be improved, and the misjudgment risk caused by slow variable lag can be reduced.

[0046] A parameter set containing all the updated content is constructed and written into the turn sensitive baseline structure. The parameter set includes fields such as the current turn window number, the corresponding start and end time, the alarm threshold correction record, the sliding window adjustment record, the perception channel weight configuration, the residual sequence statistical result, and the maximum value of the pre-warning index triggering this update. The writing method is: in the time sequence structure of the turn sensitive baseline, a new parameter entry is inserted, and the source timestamp of this update, the associated risk level change process and the subsequent effective range are marked in it. The baseline structure is a dynamic expandable data structure, which can form independent parameter records according to the feedback in each window, without affecting the historical judgment and causing confusion in the later model. When the same type of turn structure appears again in the future, this parameter set can be used as the basis for adaptive configuration, improving the prediction accuracy of the model under similar risk evolution conditions.

[0047] Based on the latest turn sensitive baseline after writing, the next round of turn window monitoring and risk judgment process is reinitialized. When starting the next data processing cycle, the system will preferentially call the latest parameter configuration for perception data preprocessing and risk discrimination, ensuring that the model is highly consistent with the current actual running state in structure. If high residuals continue to appear in subsequent operation, the system can repeat the dynamic adjustment according to the current process until the model converges stably. This closed loop process has three characteristics of self-driving, self-feedback and self-adaptation, solving the problem that the model parameters in the prior art depend on manual static configuration and cannot be adjusted according to the running results.

[0048] The step is used for realizing closed-loop self-adaptive regulation and control of the power distribution station overload risk identification mechanism, ensuring that the system can continuously correct and optimize the judgment parameters based on the actual operation data after processing the turning-back fluctuation and generating the pre-warning index. By analyzing the residual error between the trend prediction data and the actual perception data in real time, the deviation degree of the model in risk prediction is identified, and the alarm threshold, the sliding window width and the weighting configuration of the perception channel are dynamically adjusted, so that the model can be more suitable for the current operation condition. The updated parameters are written into the turning-back sensitive baseline, which not only realizes the historical record of the risk identification strategy, but also provides an automatic adaptation reference for subsequent similar fluctuation scenarios. Compared with the traditional static parameter setting method, the method has significant dynamic response capability and continuous optimization capability, which can effectively improve the accuracy and timeliness of risk warning, avoid false alarms, missed alarms or response lag caused by model solidification, and build a risk closed-loop mechanism from data perception to model regulation.

[0049] The application utilizes the time sequence alignment and turning-back feature identification of multi-class perception data to accurately capture the abnormal mode of load sudden drop and sudden rise, and retains the key upward trend through edge fidelity filtering to avoid the risk signal being covered by smoothing processing. Meanwhile, the heat accumulation and fast-slow variable decoupling mechanism is introduced to improve the judgment accuracy of the physical risk state. Further, the pre-warning index generation and residual feedback correction are used to establish a closed-loop adaptive control logic, which significantly enhances the identification sensitivity and pre-warning foresight of the system to abnormal fluctuations. Compared with the prior art, the application has the beneficial effects of earlier warning, faster response and more accurate risk identification, and significantly improves the operation safety and risk control capability of the power distribution station under complex load dynamics.

[0050] The application provides a power distribution station overload risk warning device based on multi-source perception data fusion as shown in Figure 2 The device comprises a data alignment module, a turning-back identification module, a fidelity filtering module, a variable decoupling and risk modeling module, a weighting adjustment and warning module and a closed-loop control module. The data alignment module injects a micro-amplitude synchronization pulse during the multi-source perception data collection stage in the operation process of the power distribution station, collects pulse feedback results, calculates the collection clock offset, constructs a second-level alignment vector, realizes the time sequence synchronization between different perception channels, and establishes a turning-back sensitive baseline. The turning-back identification module identifies the fluctuation mode of first falling and then rising according to the second-level aligned multi-source perception data under the constraint of the turning-back sensitive baseline, calculates the continuous derivative sequence in a sliding window manner, screens the fluctuation segments exceeding the set amplitude threshold, and constructs a turning-back window list. The fidelity filtering module performs edge fidelity filtering processing within the constraint range of the turn window list, suppresses trend smoothing across the turn section, retains the waveform slope of the rising section, eliminates the peak top cutting effect, and outputs enhanced trend data by dynamically setting the filtering window range. The variable decoupling and risk modeling module performs freezing operation on the temperature type perception data after the edge fidelity filtering processing, realizes the change path decoupling of fast variables and slow variables, generates heat accumulation values by using current square time accumulation calculation method, constructs heat guardian threshold, and improves the overload risk level. The weighted adjustment and early warning module reallocates the weighting coefficients of the multi-source perception channels after the overload risk level is improved, improves the weights of the current and power type data in the turn window, reduces the weights of the temperature and environment type data, generates a pre-warning index as a pre-alarm trigger leading signal, and outputs the pre-warning index. The closed-loop regulation module compares the residual change between the trend prediction data and the actual measurement data after obtaining the pre-warning index, dynamically updates the threshold parameters, the sliding window width and the perception channel weighting coefficients, and writes the updated parameter set into the turn sensitive baseline, and completes the closed-loop risk alarm process from perception, identification, processing to regulation.

[0051] The power distribution station overload risk alarm method based on multi-source perception data fusion provided by the embodiment of the application is realized by the power distribution station overload risk alarm device based on multi-source perception data fusion, and the specific method and process of the power distribution station overload risk alarm device based on multi-source perception data fusion are described in the embodiment of the power distribution station overload risk alarm method based on multi-source perception data fusion, and will not be described here.

[0052] The above only describes some exemplary embodiments of the application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the application. Therefore, the above drawings and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the application.

Claims

1. A distribution station overload risk alarm method based on multi-source sensing data fusion is characterized by: The following steps are involved: S001, during the operation of the distribution station, injects micro-synchronous pulses during the multi-source sensing data collection phase, collects pulse feedback results, calculates the acquisition clock offset, constructs second-level alignment vectors, achieves timing synchronization between different sensing channels, and establishes a return-sensitive baseline; S002, under the constraints of the reentry-sensitive baseline, uses a sliding window method to calculate the continuous derivative sequence based on the multi-source perception data after second-level alignment, identifies the fluctuation pattern of first decreasing and then increasing, filters the fluctuation segments that exceed the set amplitude threshold, and constructs the reentry window list; S003, within the constraints of the return window list, perform edge-fidelity filtering processing, and by dynamically setting the filter window range, suppress trend smoothing across the return segment, retain the waveform slope of the rising segment, eliminate the effect of peak clipping, and output enhanced trend data; S004: After edge-fidelity filtering is completed, a freezing operation is performed on the temperature-related sensing data to decouple the change paths of fast and slow variables. A heat accumulation value is generated using the current square time accumulation method to establish a heat protection threshold and increase the overload risk level. S005: When the overload risk level increases, the weight coefficients of the multi-source sensing channels are redistributed. In the return window, the weights of current and power data are increased, and the weights of temperature and environment data are reduced. A pre-warning index is generated as a leading signal for triggering an alarm. S006, after obtaining the advance warning index, compare the residual changes between the trend prediction data and the actual measurement data, dynamically update the threshold parameters, sliding window width and perception channel weighting coefficient, and write the updated parameter set into the return sensitive baseline to complete the closed-loop risk alarm process from perception, identification, processing to regulation.

2. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S001 includes: The preset current sensor, voltage sensor, load detection device, conductor temperature sensor, equipment operation status acquisition device and environmental temperature and humidity sensing device have the ability to receive external micro-amplitude synchronization pulses and record timestamps; In the initial stage of data collection, the central control unit sends a synchronous pulse signal with a voltage amplitude of less than 100 millivolts and a duration of less than 10 milliseconds to each sensing channel, collects the receiving timestamps fed back by each channel and calculates the time difference; The sampling clock offset patterns of each channel are counted over multiple consecutive synchronization pulse cycles to form a trend of time difference changes per second, and a second-level alignment vector is generated for unified time base correction. The timestamps of all channel sampling data are adjusted based on the second-level alignment vector, the data time axis is corrected through time shift or interpolation, and a return-sensitive baseline based on a unified standard time is constructed to provide timing consistency for subsequent abnormal fluctuation identification.

3. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S002 includes: Based on the time-aligned sensing data of the current channel, load power channel, and conductor temperature channel, a sliding window is divided by sampling once per second, with a window width of 30 seconds and a step size of 1 second. The continuous derivative sequence within each window is calculated. Identify whether there is a fluctuation pattern formed by a combination of continuous descending segments and continuous ascending segments in each window. Each segment must have at least three sampling points, with amplitudes exceeding 8% of the rated current value. The time interval between the previous and next segments must be at least 5 seconds. For windows that initially meet the requirements, perform consistency verification on the load power channel and the conductor temperature channel within the same time interval to confirm that the power fluctuation direction is consistent and the amplitude exceeds 10% of the rated value, and that the temperature channel rises by more than 1.5 degrees Celsius within 10 seconds after the window ends; The time periods that meet the multi-channel verification conditions are organized into a structured retracement window list, recording the window start and end time, channel number, change range and verification results as input for subsequent trend retention and risk identification.

4. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S003 includes: Extract the start time, end time, channel type and fluctuation characteristics of each return window, intercept the corresponding time period in the original perception data and expand it by 1 second before and after as a protection area; The sliding window width is dynamically set according to the maximum value of the derivative of the return segment and the time span, and continuous filtering is performed using a weighted average method, with the weight gradually decreasing from the center of the window to the two ends; The slope protection mechanism is enabled for continuous sampling points in the upward trend segment. When three sets of consecutive positive difference values ​​are met and the amplitude is not less than the rated threshold, the filtering is suspended and the original data is retained; The data detected as peak points will not be modified. After all return segments are processed, the filtering results will be embedded into the original data sequence, and a 2-second buffer will be set before and after to perform linear interpolation to ensure data continuity.

5. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S004 includes: Extract the sensor data sequence after edge-fidelity filtering, obtain the conductor temperature, device housing temperature, and ambient temperature data within the corresponding time period of the return window, and keep the most recent valid temperature value before the window start unchanged within the window; Perform a square operation on the current sensing data sampled every second within the return window, accumulate all the current square values, and calculate the current square time product value as the heat input; The device model, cable parameters, and ventilation conditions are combined to match the device's heat capacity carrying capacity. The current-squared time product is compared with the heat capacity threshold to categorize the overload risk levels into primary, secondary, and tertiary levels. The risk level, start and end time, current peak and freezing temperature corresponding to each return window are written into the return window list for use in subsequent dynamic weight adjustment and warning index generation steps.

6. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S005 includes: Dynamically adjust the weighting coefficients of the sensing channels based on the risk level corresponding to the heat accumulation value within the foldback window, increasing the weights of the current channel and power channel to no less than 1.5, and reducing the weights of the conductor temperature channel and ambient temperature channel to no more than 0.6; The current, power, conductor temperature and ambient temperature data at each sampling moment within the foldback window are multiplied by the corresponding adjusted weights, and the weighted values ​​are summed to generate a continuous weighted risk data sequence; The normalized interval of each sensing channel is determined using historical data within 24 hours, and the weighted risk data series is standardized to generate a pre-warning index distributed in the range of 0 to 100; Compare the advance warning index with the set risk classification threshold. If the index exceeds the preset threshold for three consecutive sampling periods, mark the current retracement window as an abnormal continuous development state, and record the weight configuration and index evolution trajectory.

7. The distribution station overload risk alarm method based on multi-source sensing data fusion according to claim 1 is characterized in that: Step S006 includes: Calculate the second-by-second residual between the trend prediction data and the actual perception data within the return window, generate a residual change sequence, and evaluate the degree of deviation from the prediction model; Adjust alarm parameters based on residual change results, including lowering the excess alarm thresholds for current and power, compressing the sliding window width, increasing the weight of fast variable channels, and simultaneously decreasing the weight of slow variable channels; Record each parameter adjustment result as an independent parameter set, including the return window number, residual statistics, weights before and after the update, window width and alarm threshold changes; The independent parameter set is written into the return-sensitive baseline according to the timestamp and used as the priority configuration basis for the next round of risk judgment and perception preprocessing operations to achieve closed-loop adaptive control of the alarm process.

8. A distribution station overload risk alarm device based on multi-source sensing data fusion, used to implement the distribution station overload risk alarm method based on multi-source sensing data fusion as described in any one of claims 1 to 7, characterized in that: It includes data alignment module, return identification module, fidelity filtering module, variable decoupling and risk modeling module, weighted adjustment and early warning module and closed-loop control module: The data alignment module, during the operation of the distribution station, injects micro-synchronization pulses during the multi-source perception data collection phase, collects pulse feedback results, calculates the acquisition clock offset, constructs second-level alignment vectors, achieves timing synchronization between different perception channels, and establishes a return-sensitive baseline; The reentry recognition module, under the constraints of a reentry-sensitive baseline, uses a sliding window approach to calculate a continuous derivative sequence based on second-level aligned multi-source sensor data. It identifies fluctuation patterns that first decrease and then increase, filters out fluctuation segments that exceed a set amplitude threshold, and constructs a reentry window list. The fidelity filtering module performs edge fidelity filtering within the constraints of the return window list. By dynamically setting the filter window range, it suppresses trend smoothing across the return segment, retains the waveform slope of the rising segment, eliminates the effect of peak clipping, and outputs enhanced trend data. The variable decoupling and risk modeling module, after completing edge-fidelity filtering, freezes temperature-related sensor data to decouple the change paths of fast and slow variables. It uses the current-squared time accumulation method to generate heat accumulation values, establish heat protection thresholds, and increase the overload risk level. The weighted adjustment and warning module redistributes the weighting coefficients of the multi-source sensing channels when the overload risk level increases. In the return window, the weights of current and power data are increased, while the weights of temperature and environmental data are decreased. This generates a pre-warning index as a leading signal for alarm triggering. After obtaining the advance warning index, the closed-loop control module compares the residual changes between the trend prediction data and the actual measurement data, dynamically updates the threshold parameters, sliding window width and perception channel weighting coefficient, and writes the updated parameter set into the return sensitive baseline to complete the closed-loop risk alarm process from perception, identification, processing to control.

Images