Power data processing method and device based on uniform coding, equipment, storage medium and program product

Abstract

The application relates to a power data processing method and device based on unified coding, equipment, a storage medium and a program product, and relates to the technical field of a power system. The method comprises the following steps: collecting multi-dimensional electrical quantity time series data of a user side; detecting a target event trigger point based on the electrical quantity time series data; constructing a multi-time scale window according to the target event trigger point; performing cross-scale consistency processing on the multi-time scale window; extracting multi-scale features corresponding to the electrical quantity time series data based on the processed multi-time scale window; fusing and coding the multi-scale features to obtain a unified coding result; and using the unified coding result for operation state analysis of a power device, wherein the operation state analysis at least comprises one of device identification, state perception, abnormality detection and operation deduction. The method can fuse and code multi-scale features into a unified representation, can completely represent power event whole-process information, can be flexibly adapted to various downstream analysis tasks, and has good reusability.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a power data processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product based on unified coding. Background Technology

[0002] With the large-scale integration of distributed power sources, energy storage devices, electric vehicles, and various flexible loads into urban power distribution networks, user-side power generation and consumption behavior exhibits stronger volatility, randomness, and non-stationarity. Compared to traditional load analysis methods that primarily rely on steady-state statistics, user-side power events often simultaneously encompass millisecond-level transient impacts, second-level dynamic transition processes, and minute-level steady-state persistence results. There are clear correlations between different time scales, making it difficult to fully characterize the true features of the same power event by relying solely on data from a single scale.

[0003] In traditional technologies, the analysis of power generation and consumption events on the user side typically employs methods involving fixed time windows and single-scale feature extraction. Some solutions focus on current surges, harmonic disturbances, or transient waveform changes at the moment an event occurs to identify equipment startup, shutdown, or state switching; others focus on power change trends at the second or minute level to characterize equipment operation phases or load change features.

[0004] Therefore, traditional methods often only model a single stage within a transient or steady-state state, failing to simultaneously express the continuous relationship between different stages before and after an event, resulting in an incomplete description of the same event. Furthermore, some solutions heavily rely on a specific recognition model or classification task, making their feature construction methods difficult to reuse in different downstream scenarios. Summary of the Invention

[0005] Therefore, it is necessary to provide a power data processing method, device, computer equipment, computer-readable storage medium, and computer program product based on unified coding to address the above-mentioned technical problems. By constructing multi-timescale windows and implementing cross-scale consistency processing, multi-scale features are fused and encoded into a unified representation, which can completely represent the information of the entire process of power events and flexibly adapt to various downstream analysis tasks such as equipment identification, status perception, anomaly detection, and operation simulation, and has good reusability.

[0006] In a first aspect, this application provides a power data processing method based on unified coding, comprising:

[0007] Collect multi-dimensional electrical quantity time-series data from the user side;

[0008] Based on electrical quantity timing data, detect the trigger point of the target event;

[0009] Construct a multi-timescale window based on the target event trigger point;

[0010] Perform cross-scale consistency processing on multi-timescale windows;

[0011] Based on the processed multi-timescale windows, multi-scale features corresponding to electrical quantity time series data are extracted.

[0012] Multi-scale features are fused and encoded to obtain a unified coding result; the unified coding result is used for the operation status analysis of power equipment, which includes at least one of equipment identification, status perception, anomaly detection and operation simulation.

[0013] In one embodiment, the electrical quantity timing data includes at least current timing data, active power timing data, and harmonic component timing data; based on the electrical quantity timing data, detecting the target event trigger point includes:

[0014] Based on the current time series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points.

[0015] Based on the active power time series data, determine the first-order power difference component of the current sampling point and the maximum absolute value of the first-order power difference component of all sampling points.

[0016] Based on the harmonic component time series data, determine the harmonic disturbance intensity at the current sampling point and the maximum value of the harmonic disturbance intensity at all sampling points;

[0017] Based on the comprehensive mutation rate scoring function, the comprehensive mutation rate score of the current sampling point is determined according to the first-order difference component of current, the first-order difference component of power, the harmonic disturbance intensity of the current sampling point, and the maximum absolute value of the first-order difference component of current, the maximum absolute value of the first-order difference component of power, and the maximum harmonic disturbance intensity of all sampling points.

[0018] If the overall mutation score of the current sampling point is greater than the preset threshold, the current sampling point will be identified as a candidate event trigger point.

[0019] If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is less than the preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and adjacent candidate event trigger points will be determined as the target event trigger point.

[0020] If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is not less than a preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0021] In one embodiment, cross-scale consistency processing for multi-timescale windows includes:

[0022] Phase alignment, amplitude normalization, and timing registration are performed on the multi-timescale windows respectively.

[0023] In one embodiment, the multi-timescale window includes a millisecond-level window, a second-level window, and a minute-level window.

[0024] In one embodiment, based on the processed multi-timescale window, multi-scale features corresponding to the electrical quantity time-series data are extracted, including:

[0025] Based on the processed millisecond-level window, peak impulse, transient energy, and spectral disturbance characteristics are extracted from electrical quantity time-series data;

[0026] Based on the processed second-level window, the change slope, fluctuation intensity, and recovery time are extracted from the electrical quantity time series data;

[0027] Based on the processed minute-level window, the steady-state mean, steady-state offset, and steady-state volatility are extracted from the electrical quantity time-series data.

[0028] In one embodiment, the fusion of multi-scale features includes:

[0029] Based on the linear mapping functions corresponding to the multi-timescale windows, the multi-scale features are mapped, and the resulting multi-scale mapping values ​​are concatenated to obtain the fused features; or,

[0030] Based on the linear mapping functions corresponding to the multi-timescale windows, the multi-scale features are mapped, and the resulting multi-scale mapping values ​​are weighted and fused to obtain the fused features; or,

[0031] Multi-scale features are mapped using a feature rearrangement mapping function, and a two-dimensional feature map is constructed based on the multi-scale mapping values ​​obtained from the mapping to obtain fused features.

[0032] Secondly, this application also provides a power data processing device based on unified coding, comprising:

[0033] The data acquisition module is used to collect multi-dimensional electrical quantity time-series data from the user side;

[0034] The event detection module is used to detect the trigger point of the target event based on electrical quantity timing data;

[0035] The feature extraction module is used to construct a multi-timescale window based on the target event trigger point; perform cross-scale consistency processing on the multi-timescale window; and extract multi-scale features corresponding to the electrical quantity time series data based on the processed multi-timescale window.

[0036] The encoding module is used to fuse and encode multi-scale features to obtain a unified encoding result. The unified encoding result is used for the operation status analysis of power equipment, which includes at least one of equipment identification, status perception, anomaly detection and operation simulation.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect above.

[0038] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0039] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0040] The aforementioned power data processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product based on unified coding collect multi-dimensional electrical quantity time-series data from the user side; detect target event trigger points based on the electrical quantity time-series data; construct multi-time-scale windows according to the target event trigger points; perform cross-scale consistency processing on the multi-time-scale windows; extract multi-scale features corresponding to the electrical quantity time-series data based on the processed multi-time-scale windows; fuse and encode the multi-scale features to obtain unified coding results; the unified coding results are used for the operation status analysis of power equipment, and the operation status analysis includes at least one of equipment identification, status perception, anomaly detection, and operation inference. Through this method, by constructing multi-time-scale windows, the electrical quantity time-series data of the same power event can be simultaneously captured and retained at multiple time scales, thereby ensuring that subsequent analysis can obtain complete information about the event at different time resolutions, avoiding the problem of incomplete information coverage under a single-scale window. By performing cross-scale consistency processing on the multi-time-scale windows, the comparability and coordination of data at different scales are ensured in time, thus providing a well-aligned data foundation for the subsequent collaborative extraction of multi-scale features. The extracted multi-scale features are fused and encoded to generate a unified encoding result. This unified encoding result integrates feature information from multiple time scales, enabling a more complete representation of the entire process attributes of the same event. Since this unified encoding result is a general representation oriented towards the event itself, its generation process is independent of specific downstream analysis tasks. It can be directly used in various operational status analysis scenarios such as device identification, status awareness, anomaly detection, and operational simulation, without requiring the redesign of feature extraction and encoding processes for each task, thus exhibiting good reusability. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating a power data processing method based on unified coding in one embodiment;

[0043] Figure 2 This is a schematic diagram of event trigger point detection and multi-timescale window construction in one embodiment;

[0044] Figure 3 This is a schematic diagram of cross-timescale consistency processing in one embodiment;

[0045] Figure 4 This is a schematic diagram illustrating the mapping of multi-timescale features to a unified coding space in one embodiment;

[0046] Figure 5 This is a schematic diagram illustrating the unified encoding result output format in one embodiment;

[0047] Figure 6 This is a structural block diagram of a power data processing device based on unified coding in one embodiment;

[0048] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0051] To facilitate understanding of the embodiments provided in this application, some terms that appear in the detailed description are introduced below:

[0052] A power event refers to a state switching process that causes significant changes in electrical quantities such as voltage, current, and power on the user side within a short period of time, such as equipment startup, shutdown, mode switching, energy storage charging and discharging conversion, and sudden changes in photovoltaic output.

[0053] Transient refers to the rapid change phase at the instant of an electrical event, which is usually manifested as millisecond-level waveform impulses, harmonic disturbances, phase shifts, or short-term overshoots.

[0054] The transition state refers to the intermediate stage in the evolution of a power event from its initial change to a new equilibrium state, which is usually manifested as a dynamic recovery, fluctuation decay, or power adjustment process on the order of seconds.

[0055] Steady state refers to the stage in which the system enters a relatively stable operation after a power event ends, which is usually characterized by minute-level statistics tending to stabilize.

[0056] The event trigger point refers to the time position used to determine the start of an electrical event, denoted as . .

[0057] A unified coding space refers to a common representation space formed by mapping features at different time scales according to unified rules. This space can be a two-dimensional feature map space, a low-dimensional vector embedding space, or a symbolic semantic space, etc.

[0058] Understandably, traditional methods typically extract event features from a single time scale or simply concatenate features from different time scales. This results in a lack of unified correlation expression between the three stages of the same power event: the millisecond-level transient state, the second-level transitional state, and the minute-level steady state. It is difficult to form a general semantic representation that can be directly called upon by subsequent equipment for identification, state perception, anomaly detection, or operational simulation.

[0059] This application embodiment focuses on the triggering time of the same power event, uniformly extracting, aligning, normalizing, and encoding high-frequency and low-frequency electrical quantity data. It maps transient, transitional, and steady-state characteristics reflecting the same event at different time scales to the same representation space, ultimately generating a unified event encoding result. Since this unified encoding result is a general representation oriented towards the event itself, its generation process is independent of specific downstream analysis tasks. It can be directly used in various operational state analysis scenarios such as equipment identification, state perception, anomaly detection, and operational simulation, without requiring redesign of feature extraction and encoding processes for each task, thus exhibiting good reusability.

[0060] In one exemplary embodiment, such as Figure 1As shown, a power data processing method based on unified coding is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0061] Step 102: Collect multi-dimensional electrical quantity time-series data from the user side.

[0062] The user side refers to the power system portion from the electricity meter back end to the user's electrical equipment, including the user's internal power distribution lines, various electrical loads, and distributed power sources, energy storage devices, and other equipment. This embodiment collects time-series electrical quantity data during the user's power generation and consumption process, including at least one or more of the following types: voltage time-series data. Current timing data Active power time series data Reactive power time series data Power factor timing data Harmonic component time series data , Indicates the sampling time. Indicates the harmonic order.

[0063] In one alternative implementation, current and voltage are sampled at a higher rate to preserve transient impact details; power and statistics can be sampled at a lower rate to characterize steady-state evolution. The discretely sampled signal is denoted as:

[0064] ;

[0065] in, This represents the total number of sampling points.

[0066] Step 104: Detect the target event trigger point based on electrical quantity timing data.

[0067] Among them, the target event trigger point This refers to the time position at which a power event on the user side begins to appear in the time-series data. This position marks the critical moment when the electrical quantity undergoes a substantial transition from a normal operating state to an event response state. Based on one or more signals in the electrical quantity time-series data, it detects changes in the electrical quantity that exceed a preset detection threshold within a short period of time.

[0068] For example, any one or a combination of thresholding, sliding windowing, statistical mutation, or learned detection methods can be used.

[0069] Threshold methods set fixed amplitude thresholds for electrical quantity data in a certain dimension. When the instantaneous value or short-term average value of the electrical quantity exceeds or falls below the threshold, an event is determined to be triggered. For example, in a unidirectional threshold detection implementation, a sudden increase threshold and a sudden decrease threshold can be set for current time series data. When the current value exceeds the increase threshold, it is determined to be a positive sudden event, such as equipment startup or load connection; when the current value falls below the decrease threshold, it is determined to be a negative sudden event, such as equipment shutdown or load disconnection. For active power sequences and reactive power sequences, corresponding amplitude thresholds can also be set for bidirectional detection.

[0070] To avoid the problem of insufficient adaptability of fixed thresholds under different operating conditions, a relative threshold detection method can be used. Using the electrical quantity statistics during a normal operating period before the event as a benchmark, the change in the current value relative to the benchmark value is calculated. When this change exceeds a set proportional threshold or a multiple of the standard deviation threshold, the event is considered triggered.

[0071] The sliding window method detects event trigger points by comparing the differences in electrical quantity statistical characteristics within two sliding windows before and after the current moment. In real-time detection scenarios, the sliding window continuously slides and updates as new sampling points arrive, calculating a detection statistic at each new sampling point to achieve continuous online monitoring of event trigger points. For example, in the implementation of mean-based sliding window detection, the electrical quantity sampling values ​​within the forward and backward windows at the current moment are taken, and their respective means are calculated. The forward window represents the background level before the event occurs, and the backward window represents the current level after the event occurs. The difference between the means of the forward and backward windows is calculated; when this difference exceeds a preset threshold, the current moment is determined to be a candidate event trigger point.

[0072] For example, in an implementation of variance sliding window detection, for certain events where the current amplitude changes insignificantly but the degree of fluctuation changes significantly, such as when equipment enters an unstable operating state or a harmonic source is connected, the variance within the preceding and following windows can be calculated separately. When the ratio or difference between the variance of the following window and the variance of the preceding window exceeds a set threshold, the event is determined to be triggered.

[0073] The statistical catastrophe method models the detection of event trigger points as a problem of detecting changes in statistical characteristics in time-series data. It identifies event trigger points by recognizing moments when data distribution or statistical features undergo significant changes. For example, in an implementation of the cumulative sum detection method, the persistent shift in the mean is detected by accumulating the deviation between the current value and the target value for normal operation. Positive and negative cumulative statistics are established separately. Under normal operating conditions, the positive and negative deviations caused by random fluctuations cancel each other out, keeping the cumulative statistics at a low level. When an event occurs that causes a persistent positive or negative shift in the electrical quantity mean, the corresponding cumulative statistics will continuously increase. When the cumulative amount exceeds a preset decision threshold, the event is determined to be triggered, and the trigger point is located at the starting moment when the cumulative statistics begin to rise significantly.

[0074] For example, in the implementation of Bayesian change point detection, event triggering is considered as a parameter change occurring at a certain moment during the data generation process. Under certain model assumptions, this method calculates the posterior probability that each possible historical moment is a change point. It updates the estimated probability of the current moment being a change point upon the arrival of each new sampling point. When this probability exceeds a preset threshold, an event is determined to be triggered. Bayesian change point detection does not require a pre-set fixed amplitude threshold and has a certain degree of adaptive detection capability for events of different types and amplitudes. However, it involves relatively large computational loads and is suitable for main station analysis platforms with sufficient computing resources.

[0075] Learning-based detection methods utilize machine learning or deep learning models to automatically learn the pattern features of event occurrences from labeled historical event data, thereby enabling real-time discrimination of input sequences and outputting the location of event trigger points. This approach is suitable for scenarios with abundant historical labeled data, comprehensively utilizing the coordinated changes in multi-dimensional electrical quantities for discrimination, and continuously optimizing detection performance through learning. For example, in a trigger detection implementation based on a convolutional neural network, a one-dimensional convolutional neural network model is constructed using collected multi-dimensional electrical quantity time-series data as input. The input data consists of short time-series segments containing multiple channels, each segment having a fixed length and corresponding to a time span of several milliseconds to several seconds. After training, the model outputs a binary classification result indicating whether the segment contains an event trigger point, and simultaneously outputs the positional offset of the trigger point within the segment when a trigger is detected. In one embodiment, the convolutional neural network can employ a residual connection structure to deepen the network layers, improving the ability to recognize complex transient patterns.

[0076] For example, in a trigger detection implementation based on Long Short-Term Memory (LSTM) networks, the ability of LTM networks to model long-term dependencies in time-series data is utilized. Using continuously sampled multi-dimensional time-series data as input, the network outputs the probability value of that moment being an event trigger point at each time step. When this probability exceeds a preset threshold, the event trigger point is output. Compared to convolutional neural networks, LTM networks are more suitable for handling event detection tasks where the timing of events is not fixed and the duration is inconsistent.

[0077] For example, convolutional neural networks can be combined with long short-term memory networks. Convolutional neural networks can be used to extract local transient patterns, while long short-term memory networks can be used to capture temporal context information, forming a convolutional long short-term memory network model to further improve detection performance.

[0078] To further improve the accuracy and robustness of event trigger point detection, a combination strategy of multiple detection methods can be adopted. For example, in a serial combined detection implementation, a coarse screening is first performed using the threshold method or sliding window method with minimal computational cost to quickly eliminate long periods of data where no events have occurred. When the coarse screening results indicate the possible presence of an event, a more computationally intensive statistical catastrophe method or learning-based detection method is then initiated for precise location. This combined strategy effectively reduces overall computational overhead while ensuring detection performance.

[0079] For example, in the implementation of parallel combined detection and comprehensive judgment, multiple detection methods are used simultaneously to independently detect the same input data, with each method outputting candidate trigger points. The final event trigger point is then determined through a decision fusion mechanism. Decision fusion can employ methods such as voting, weighted averaging, or a comprehensive scoring method based on credibility. When the candidate trigger points output by multiple methods are consistent or close in time, the confidence level of the trigger point is high. When there are significant discrepancies in the output results of different methods, a comprehensive judgment can be made by considering the applicable priority of each method under different signal characteristics.

[0080] The final confirmed event trigger point is output as a timestamp, which provides a unified time reference for the subsequent construction of multi-timescale windows.

[0081] Step 106: Construct a multi-timescale window based on the target event trigger point.

[0082] In one exemplary embodiment, the multi-timescale window includes a millisecond-level window, a second-level window, and a minute-level window.

[0083] From the occurrence to the end of a power event, the electrical quantity signals exhibit different stages of change on the time axis: the event is accompanied by millisecond-level current surges, voltage drops, or harmonic abrupt changes; this is followed by second-level power oscillations, current declines, or dynamic adjustment processes; and finally, it transitions to a new steady-state operating state on the minute level. These three stages are sequentially connected in time, together constituting the complete evolution process of the same event.

[0084] In order to selectively extract data from the three stages mentioned above, this embodiment takes the event trigger point as the origin and extends forward and backward along the time axis by a certain length to construct three analysis windows with different time spans, corresponding to the analysis needs of the three stages mentioned above.

[0085] Among them, constructing a millisecond-level window Used to capture transient features such as instantaneous surges, harmonic abrupt changes, and phase shifts; constructs second-level windows. Used to characterize the dynamic recovery and power transition process after an event; constructs minute-level windows. Used to characterize the new steady-state level, persistent offset, and running results. Among them:

[0086] ;

[0087] ;

[0088] ;

[0089] in: These represent the lengths of the window before and after the millisecond-level window; These represent the lengths of the window before and after each second; These represent the lengths before and after the minute-level window, which generally satisfy the following: .

[0090] Reference Figure 2 Using the same power event trigger point as a baseline, millisecond-level, second-level, and minute-level time windows are constructed respectively. Different time scales correspond to transient, transitional, and steady-state characteristics. The millisecond-level window is an interval centered on the trigger point, extending a certain length of milliseconds before and after it. This window has the shortest time span and is used to cover signal changes in the instant of the event triggering and the very short period before and after. The second-level window is an interval centered on the trigger point, extending a certain length of seconds before and after it. This window typically has a time span of several seconds to tens of seconds and is used to cover the signal evolution process over a longer period from before the event triggering to after the event. The minute-level window is defined as an interval centered on the trigger point, extending a certain length of minutes before and after it. This window typically has a time span of several minutes to tens of minutes and is used to cover the electrical quantity change trends over a longer period before and after the event.

[0091] Step 108: Perform cross-scale consistency processing on multi-timescale windows.

[0092] While the constructed multi-timescale windows share the same event trigger point as the time base, each window exhibits inherent differences in data characteristics. Regarding sampling frequency, voltage and current data within millisecond-level windows typically use higher sampling rates to preserve transient impact details, while power and statistical data within minute-level windows usually use lower equivalent sampling rates. These different sampling rates result in inconsistent data point densities on the time axis for each window. In terms of amplitude range, different electrical quantity types have different physical dimensions and numerical ranges. Without normalization, electrical quantities with larger values ​​will dominate the encoding results during subsequent fusion coding. Regarding phase relationships, voltage and current in power systems are both power frequency AC signals, and their instantaneous values ​​change periodically with the power frequency cycle. The same event will exhibit different sampling value distributions when occurring at different power frequency phases.

[0093] To address the aforementioned issues, in an exemplary embodiment, step 108 includes: performing phase alignment, amplitude normalization, and timing registration processing on the multi-timescale windows respectively. (Refer to...) Figure 3 Phase alignment, amplitude normalization, and time registration are performed on electrical quantity time series data in different time scale windows to establish a consistent correspondence across scales.

[0094] Phase alignment is used to eliminate the differences in the distribution of sampled values ​​in power frequency AC signals caused by the different phase positions of the event occurrence time relative to the power frequency cycle, so that the same type of event can present a phase-independent and consistent characterization when it occurs under different power frequency phases.

[0095] For alternating current signals, phase alignment is performed using the power frequency reference. Let the reference phase point detected by zero-crossing detection or phase-locked loop within a certain window be denoted as . The reference phase point can be the positive zero-crossing point, the negative zero-crossing point, or any specified power frequency phase position, and the signal is shifted accordingly.

[0096] ;

[0097] ;

[0098] in, At the moment the event is triggered, The time of the detected reference phase point.

[0099] The signal within the window is shifted along the time axis so that the event trigger point is aligned with the power frequency phase position corresponding to the reference phase point, achieving a unified correction of the phase at the event trigger time. The shifted signal is represented as a translation of the original signal along the time axis, making the data of different events comparable within a unified framework with the power frequency phase as a reference.

[0100] In an alternative implementation, a zero-crossing alignment method can also be used. For a voltage sequence, all its zero-crossing points are detected to form a set of zero-crossing points. From this set, the zero-crossing point closest to the event trigger point on the time axis is selected as the reference point, i.e., the zero-crossing point with the smallest time distance from the trigger point is selected. This reference point is used as the phase reference, and the signals within each window are uniformly mapped to this phase reference. For example, if The set of zero cross points is Then choose the closest one. zero crossover point For reference:

[0101] ;

[0102] in, This is the event trigger point. All window signals are uniformly mapped to this phase reference.

[0103] Amplitude normalization is used to eliminate the differences in dimensions and numerical ranges between different electrical quantity types, as well as the amplitude differences caused by different rated capacities of different equipment. It maps data from each dimension to a unified amplitude range, avoiding the dominance of large electrical quantities in the coding results during subsequent feature extraction and coding processes.

[0104] In one optional implementation, a maximum-minimum normalization method is used. For the i-th sampling point in a sequence of electrical quantities, its normalized value is calculated by subtracting the minimum value of that electrical quantity within the window from the original value, and then dividing by the difference between the maximum and minimum values. The calculation method is as follows:

[0105] ;

[0106] To prevent division by zero errors, a very small positive number is added to the denominator. As a safeguard, max-min normalization linearly maps all data to the range of zero to one, preserving the relative magnitudes of the original data.

[0107] In one optional implementation, a standardization approach is used. For a given electrical quantity sequence, the mean and standard deviation within a window are calculated. Then, each sampling point is standardized by subtracting the mean from the original value and dividing by the standard deviation. The calculation method is as follows:

[0108] ;

[0109] in, The mean, The standard deviation is 1. Standardization transforms data into a distribution with a mean of zero and a standard deviation of 1, providing a good balance for multi-dimensional electrical quantity data with different amplitude scales. Similarly, to prevent division by zero errors, a very small positive number is added to the denominator. As a protected item.

[0110] For example, the choice between maximum and minimum normalization or standardization can be made based on the actual data distribution characteristics and application requirements. Different normalization methods can also be used for different types of electrical quantities. For instance, maximum and minimum normalization can be used for signals with relatively stable amplitude distributions, such as current and voltage, while standardization can be used for signals that may exhibit trend changes, such as power.

[0111] Temporal registration addresses the issue of inconsistent sampling frequencies across different windows by mapping signals within each window onto a standardized time axis of the same length. This ensures a one-to-one positional relationship between features at different scales during subsequent fusion. Since millisecond-level windows use high sampling rates, second-level windows use medium sampling rates, and minute-level windows use low sampling rates, the total number of sampling points and sampling intervals within each window differ, making direct cross-scale point-by-point comparisons or feature fusion impossible. Temporal registration solves this problem through unified time coordinate mapping and resampling operations.

[0112] Set the timeline of the original window To define a uniform time coordinate system from the start to the end of the window, we use dimensionless standardized coordinates. For any data within the window, we construct a mapping from the original time axis to the standardized coordinates:

[0113] ;

[0114] in, Signals at different time scales are all resampled to a uniform length:

[0115] ;

[0116] in, To standardize the number of sampling points.

[0117] After time-series registration, data from different time scales have a unified representation in the time dimension, and the values ​​of features at different scales at the same standardized time position are comparable, providing a position-aligned input structure for subsequent multi-scale feature fusion encoding.

[0118] By performing phase alignment, amplitude normalization, and temporal registration under a unified triggering benchmark, a correspondence between data at different time scales was established, overcoming the fragmentation problem caused by simple splicing of traditional multi-scale features, and making the representation results more consistent.

[0119] Step 110: Based on the processed multi-timescale window, extract the multi-scale features corresponding to the electrical quantity time series data.

[0120] After cross-scale consistency processing, the electrical quantity time-series data within millisecond, second, and minute windows have achieved uniformity in phase reference, amplitude range, and standardized time axis. However, the data within each window still exists in the form of raw sampled value sequences, which are high-dimensional and contain a large amount of redundant information. Directly using them as input for subsequent encoding would lead to excessive computation and overfitting. Therefore, it is necessary to extract compact feature representations from the data in each window that can reflect the essential characteristics of electrical quantities at that time scale.

[0121] Feature extraction can employ time-domain features, frequency-domain features, time-frequency features, or combinations thereof. Optionally, depending on the differences in the physical characteristics of the power signal at different time scales, the appropriate feature types for each window will vary. Millisecond-level windows focus on describing the transient behavior of transient impacts and are suitable for extracting time-domain and time-frequency features that reflect waveform abrupt changes, impact intensity, and harmonic components; second-level windows focus on describing dynamic transition processes and are suitable for extracting time-domain statistical features and frequency-domain features that reflect changing trends, oscillation decay, and regulation characteristics; minute-level windows focus on describing steady-state operating conditions and are suitable for extracting time-domain features that reflect steady-state levels, long-term offsets, and statistical distributions.

[0122] Step 112: The multi-scale features are fused and encoded to obtain a unified coding result; the unified coding result is used for the operation status analysis of power equipment, and the operation status analysis includes at least one of equipment identification, status perception, anomaly detection and operation simulation.

[0123] Multi-scale feature fusion is the process of merging millisecond-level, second-level, and minute-level features into a single fused feature vector. It can employ any combination of one or more of the following methods: concatenation, weighted summation, projection mapping, or attention weighting.

[0124] Concatenation and fusion involves linking feature vectors from different scales together in a predetermined order to form a new fused feature vector. For example, millisecond-level, second-level, and minute-level features are concatenated sequentially, and the dimension of the fused feature is equal to the sum of the dimensions of the three.

[0125] Weighted summation fusion involves assigning different weight coefficients to feature vectors at each scale, calculating their weighted sum, and obtaining a fused feature vector.

[0126] Projective mapping fusion maps features at various scales to a shared low-dimensional feature space through linear or nonlinear projection transformations, achieving feature fusion within this space. Projective mapping can employ linear dimensionality reduction methods such as principal component analysis and linear discriminant analysis, or nonlinear mapping methods such as autoencoders or deep neural networks.

[0127] Attention-weighted fusion utilizes an attention mechanism to automatically learn the importance weights of features at each scale to the current representation task. Importance scores for each scale feature are calculated, and these scores are converted into weight coefficients through normalization. The fused features are then obtained by weighting and summing the features according to their respective weight coefficients. The attention weights can be adaptively adjusted based on the input data, enabling the fusion process to dynamically allocate weights according to different event types and information richness at different scales.

[0128] Unified coding is the process of mapping the fused feature vectors to a unified coding space. For example, refer to... Figure 4 The millisecond-level features, second-level features, and minute-level features are uniformly organized to form a single fusion representation result. The unified coding space can be a vector space, graph space, tensor space, or symbolic space, namely a two-dimensional feature space, vector embedding space, or symbolic semantic space.

[0129] Vector space encoding maps fused features to a fixed-dimensional low-dimensional real-valued vector, resulting in a single numerical vector. This encoding method maps fused features from the original space to the target vector space using linear projection or a nonlinear neural network.

[0130] Graph spatial coding maps fused features into a graph-structured data object. In graph spatial coding, the encoded result is expressed as a graph consisting of nodes and edges, where nodes represent feature primitives or event stages at different time scales, and edges represent temporal relationships, causal associations, or similarity relationships between feature primitives.

[0131] Tensor space coding maps fused features into a multidimensional tensor structure. Tensor space coding preserves the multidimensional structural information of the original features and is suitable for scenarios that require simultaneous preservation of multidimensional information such as time, scale, and electrical quantities.

[0132] Symbolic space coding maps fused features to discrete symbols or sequences of symbols, such as quantizing continuous feature values ​​into a finite number of discrete levels, or mapping features to discrete prototype labels through clustering.

[0133] For example, the unified encoding result is a fixed-dimensional numerical vector. Regardless of the encoding space used, the encoding process does not depend on specific downstream task labels. The information in the encoding result comes from the multi-scale features of the event itself, making the encoding result a general semantic representation oriented towards the event itself. It can be output as a standardized interface to downstream analysis modules for direct use by different tasks such as device identification, status awareness, anomaly detection, and operational simulation.

[0134] For example, different analysis tasks can share the same coding results. Only the corresponding analysis model or judgment criteria need to be configured at the application layer according to the specific goals of each task. There is no need to re-collect data, detect events and extract features for each task, which can flexibly adapt to different types of operational status analysis needs.

[0135] Among them, reference Figure 5 The unified coding result can be represented as a fused feature map, a unified feature vector, or an event semantic label, and can be further adapted to application scenarios such as device identification, state perception, anomaly detection, or operation simulation. Device identification refers to determining the type of electrical equipment that triggered the power event based on the electrical quantity change pattern corresponding to the power event. Different devices have different electrical characteristics during startup, operation, and shutdown. Optionally, device identification is achieved through a classifier. The unified coding result is input into the classifier, which outputs a predicted label for the device type after discrimination. The label can include motors, power electronic equipment, resistive loads, distributed power sources, energy storage converters, and hybrid loads, etc. The specific category can be defined and expanded according to the device type in the actual application scenario. The classifier can use traditional machine learning models such as support vector machines, random forests, and gradient boosting trees, or it can use deep neural network models. The classifier takes the unified coding result as input and the device category label as output. It is trained on labeled historical data in the offline training phase and classifies and discriminates real-time input in the online application phase.

[0136] For example, the input classifier :

[0137] ;

[0138] in, To predict the device category.

[0139] Anomaly detection refers to determining whether a current power event exhibits abnormal behavior deviating from its normal operating state. When the operating state of power equipment deviates from its normal mode, it may indicate abnormalities such as equipment failure, performance degradation, parameter drift, or external disturbances. Optionally, anomaly detection is achieved by comparing it with a normal operating template vector. In the offline phase, the unified coding results of multiple events occurring under normal operating conditions are collected, their mean vector is calculated, and a corresponding normal operating template vector for the equipment is constructed. During online detection, the distance between the unified coding result of the current event and this template vector is calculated. When this distance exceeds a preset anomaly threshold, the current event is determined to be abnormal; when the distance does not exceed the threshold, it is determined to be normal. In the distance calculation, Euclidean distance can be used to measure the degree of difference between the coding result and the template vector. Euclidean distance reflects the deviation of the overall multi-scale characteristics of the current event from the normal operating mode; a larger distance indicates a higher degree of deviation. The anomaly threshold can be set based on the distance statistical distribution of historical normal operating data, or it can be adjusted according to the trade-off requirements of false alarm rate and false negative rate. For example, in order to improve the reliability of anomaly detection, multiple normal template vector libraries for different types of devices and different operating conditions can be pre-built. During detection, the device type of the current event is first determined by device identification, and then the corresponding normal template vector is called for comparison to eliminate the impact of device type differences on distance measurement.

[0140] For example, define a normal template vector. Distance:

[0141] ;

[0142] when When this occurs, it is determined to be an abnormal event. This is the abnormal threshold.

[0143] Operational simulation refers to predicting or extrapolating the future operational state of a system based on the multi-scale encoding results of the current event and the current system operational state. Operational simulation can be used to predict the operational state of equipment in the next stage after the completion of the current event, providing a basis for subsequent scheduling control, alarm prediction, or maintenance decisions. Optionally, the unified encoding result can be used as one of the input features of the simulation model, combined with the current operational state, to output the predicted state for the next moment. The simulation model can be a time-series prediction model built based on Long Short-Term Memory networks or gated recurrent units. It utilizes the event type information and transition process information encoded in the unified encoding result, combined with current state variables (such as current power level, current operating mode, current load rate, etc.), to extrapolate the operational state the system will enter after the event ends. For example, the simulation model can adopt an encoder-decoder framework, where the encoder encodes the unified encoding result and the current state, and the decoder outputs the predicted operational state. During the offline training phase, the simulation model uses the multi-scale encoding results of historical events and their subsequent actual operational states as training samples, establishing a mapping relationship between the two through supervised learning.

[0144] For example, the unified encoding result is used as the input feature for subsequent inference modules:

[0145] ;

[0146] in, Indicates the current running status. Indicates the deduction model, This indicates the predicted state.

[0147] State awareness refers to using unified coding results to reflect the current operating condition and status of power equipment. Unified coding results integrate electrical quantity information of the same event across multiple time scales, including transient impact intensity, transient process characteristics, and steady-state changes, enabling the reflection of equipment behavior patterns in the current event from multiple dimensions. By mapping unified coding results to a predefined state space, the current operating stage of the equipment can be identified, such as the startup stage, normal operation stage, shutdown stage, or fault operation stage. State awareness can provide operators with an intuitive judgment of the current operating condition of the equipment and can also serve as the basis for operational simulation and anomaly detection. For example, state awareness can be achieved by performing similarity matching between the coding results and historical coding templates of various types of equipment under typical operating states.

[0148] The aforementioned power data processing method based on unified coding involves: collecting multi-dimensional electrical quantity time-series data from the user side; detecting target event trigger points based on the electrical quantity time-series data; constructing multi-time-scale windows based on the target event trigger points; performing cross-scale consistency processing on the multi-time-scale windows; extracting multi-scale features corresponding to the electrical quantity time-series data based on the processed multi-time-scale windows; fusing and encoding the multi-scale features to obtain unified coding results; and using the unified coding results for power equipment operation status analysis, which includes at least one of equipment identification, status perception, anomaly detection, and operation simulation. By constructing multi-time-scale windows, the electrical quantity time-series data of the same power event can be simultaneously captured and retained at multiple time scales, ensuring that subsequent analysis can obtain complete information about the event at different time resolutions and avoiding the problem of incomplete information coverage under a single-scale window. By performing cross-scale consistency processing on the multi-time-scale windows, the comparability and coordination of data at different scales are ensured in time, thus providing a well-aligned data foundation for the subsequent collaborative extraction of multi-scale features. The extracted multi-scale features are fused and encoded to generate a unified encoding result. This unified encoding result integrates feature information from multiple time scales, enabling a more complete representation of the entire process attributes of the same event. Since this unified encoding result is a general representation oriented towards the event itself, its generation process is independent of specific downstream analysis tasks. It can be directly used in various operational status analysis scenarios such as device identification, status awareness, anomaly detection, and operational simulation, without requiring the redesign of feature extraction and encoding processes for each task, thus exhibiting good reusability.

[0149] In an exemplary embodiment, the electrical quantity time-series data includes at least current time-series data, active power time-series data, and harmonic component time-series data; step 104 includes: determining, based on the current time-series data, the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points; determining, based on the active power time-series data, the first-order difference component of the power at the current sampling point, and the maximum absolute value of the first-order difference component of the power at all sampling points; determining, based on the harmonic component time-series data, the harmonic disturbance intensity at the current sampling point, and the maximum harmonic disturbance intensity at all sampling points; and, based on the comprehensive abrupt change scoring function, determining, based on the first-order difference component of the current, the first-order difference component of the power, and the harmonic disturbance intensity at the current sampling point, using... The comprehensive mutation score of the current sampling point is determined by taking the maximum absolute values ​​of the first-order difference components of current, power, and harmonic disturbance intensity of all sampling points. If the comprehensive mutation score of the current sampling point is greater than a preset threshold, the current sampling point is determined as a candidate event trigger point. If the time interval between a candidate event trigger point and its adjacent candidate event trigger points is less than a preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and its adjacent candidate event trigger points is determined as the target event trigger point. If the time interval between a candidate event trigger point and its adjacent candidate event trigger points is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0150] This step involves locating the start time of the power event within the continuously sampled data. This ensures that subsequent timescale windows all revolve around the same physical event. Change indicators can be constructed based on signals such as current, voltage, and power. The first-order difference component is defined as follows:

[0151] ;

[0152] ;

[0153] Define the rate of change of the slip:

[0154] ;

[0155] in, This represents the length of the sliding window.

[0156] Define the comprehensive mutation rate scoring function:

[0157] ;

[0158] in, Indicates the intensity of harmonic disturbance. These are the weighting coefficients; To prevent extremely small constants with a denominator of zero.

[0159] When satisfied At that time, the judgment of the first The time corresponding to each sampling point is the candidate event trigger point, where This is a preset threshold.

[0160] To prevent the same event from being triggered repeatedly within a short period of time, a minimum trigger interval can be set, i.e., a preset trigger interval. If the interval between two candidate trigger points is less than... If the score is higher, then the scorer with the higher score will be retained.

[0161] Finally, the event trigger time is obtained:

[0162] ;

[0163] in, This represents the set of candidate trigger points.

[0164] In an exemplary embodiment, step 110 includes: extracting peak impulse, transient energy, and spectral disturbance characteristics from electrical quantity time series data based on a processed millisecond-level window; extracting change slope, fluctuation intensity, and recovery time from electrical quantity time series data based on a processed second-level window; and extracting steady-state mean, steady-state offset, and steady-state volatility from electrical quantity time series data based on a processed minute-level window.

[0165] Among them, based on the processed millisecond-level window, millisecond-level transient features are extracted. The millisecond-level window mainly extracts rapid impact features, including peak impact amount, transient energy, and spectral perturbation features.

[0166] The peak impact value is extracted using the following formula:

[0167] ;

[0168] In the formula, Indicates the reference current before triggering. This indicates a window in milliseconds.

[0169] Transient energy is extracted using the following formula:

[0170] ;

[0171] The discrete form can be written as:

[0172] ;

[0173] In the formula, This indicates a window in milliseconds.

[0174] The process of extracting spectral perturbation features includes: performing a short-time Fourier transform on the window signal.

[0175] ;

[0176] In the formula, For window functions, For time displacement, For frequency. Further extract indicators such as the dominant frequency component, harmonic amplitude distribution, and spectral centroid.

[0177] Based on the processed second-level window, second-level transition state features are extracted. The second-level window mainly extracts features that characterize the dynamic recovery process, including the change slope, fluctuation intensity, and recovery time.

[0178] The slope of change can be extracted using the following formula:

[0179] ;

[0180] In the formula, This indicates the backward extension length of the second-level window after the event trigger point.

[0181] The fluctuation intensity is extracted using the following formula:

[0182] ;

[0183] In the formula, Indicates a window in seconds. This represents the average power within a second-level window.

[0184] The process of extracting recovery time includes: defining the system's entry into the stable band. The earliest moment :

[0185] ;

[0186] The recovery time is defined as:

[0187] ;

[0188] In the formula, This is the steady-state reference value.

[0189] Based on the processed minute-level window, minute-level steady-state features are extracted. The minute-level window mainly extracts features that characterize the new steady-state results, including steady-state mean, steady-state offset, and steady-state volatility.

[0190] The steady-state mean can be extracted using the following formula:

[0191] ;

[0192] The steady-state offset is extracted using the following formula:

[0193] ;

[0194] In the formula, This is the steady-state mean after the event; This represents the baseline mean before the event.

[0195] Steady-state volatility can be extracted using the following formula:

[0196] ;

[0197] In the formula, This represents the standard deviation within the steady-state window.

[0198] The features at the three scales are represented as follows:

[0199] ;

[0200] ;

[0201] ;

[0202] in, denoted as the feature dimension at each scale.

[0203] In this embodiment, the analysis is not limited to a single stage of the millisecond-level transient state, the second-level transition state, or the minute-level steady state. Instead, information from the three scales is incorporated into the same characterization process around the event trigger point, which can more completely reflect the full-process characteristics of the power event.

[0204] In an exemplary embodiment, fusing multi-scale features includes: mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and concatenating the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and weighting and fusing the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on feature rearrangement mapping functions, and constructing a two-dimensional feature map based on the multi-scale mapping values ​​obtained by mapping to obtain fused features.

[0205] Optionally, the unified fusion method in this embodiment includes linear mapping followed by concatenation, which is expressed as:

[0206] ;

[0207] in, For mapping functions, This indicates a serial connection operation.

[0208] This can be further expressed as:

[0209] ;

[0210] ;

[0211] ;

[0212] in, Let be the projection matrix. This is a bias term.

[0213] Optionally, to reflect the differences in importance of the same event at different scales, the unified fusion method in this embodiment includes weighted fusion, which is expressed as:

[0214] ;

[0215] in, Let be the weighting coefficients for each scale, and we have: .

[0216] Optionally, the unified fusion method in this embodiment includes constructing a unified two-dimensional feature map. :

[0217] ;

[0218] in, It represents the feature rearrangement mapping function, and the rows of the graph represent the time scale and the columns represent the feature dimensions, which is convenient for subsequent image-type models or visualization modules.

[0219] It is understandable that the final unified encoding result can be represented as:

[0220] ;

[0221] in, To unify the encoding dimensions, this encoding serves as a semantic representation of the same power event across multiple time scales. This approach enhances the coupled representational capability of multi-stage features of the same power event at the representational level, enabling subsequent models to conduct analysis based on more complete and standardized inputs. This improves recognition accuracy, anomaly detection capabilities, and adaptability to complex scenarios.

[0222] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0223] Based on the same inventive concept, this application also provides a unified coding-based power data processing apparatus for implementing the aforementioned unified coding-based power data processing method. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the unified coding-based power data processing apparatus provided below can be found in the limitations of the unified coding-based power data processing method described above, and will not be repeated here.

[0224] In one exemplary embodiment, such as Figure 6 As shown, a power data processing device based on unified coding is provided, comprising:

[0225] The acquisition module 602 is used to acquire multi-dimensional electrical quantity time-series data from the user side;

[0226] Event detection module 604 is used to detect the trigger point of a target event based on electrical quantity timing data;

[0227] The feature extraction module 606 is used to construct a multi-timescale window based on the target event trigger point; perform cross-scale consistency processing on the multi-timescale window; and extract multi-scale features corresponding to the electrical quantity time series data based on the processed multi-timescale window.

[0228] The encoding module 608 is used to fuse and encode multi-scale features to obtain a unified encoding result. The unified encoding result is used for the operation status analysis of power equipment. The operation status analysis includes at least one of equipment identification, status perception, anomaly detection and operation simulation.

[0229] In the aforementioned power data processing device based on unified coding, by constructing multi-timescale windows, the electrical quantity time-series data of the same power event can be simultaneously captured and retained at multiple time scales. This ensures that subsequent analysis can obtain complete information about the event at different time resolutions, avoiding the problem of incomplete information coverage under a single-scale window. By performing cross-scale consistency processing on the multi-timescale windows, the comparability and coordination of data at different scales are ensured in time, thus providing a well-aligned data foundation for the collaborative extraction of subsequent multi-scale features. The extracted multi-scale features are fused and encoded to generate a unified coding result. This unified coding result integrates feature information from multiple time scales and can more completely represent the full-process attributes of the same event. Since this unified coding result is a general representation oriented towards the event itself, its generation process is independent of specific downstream analysis tasks. It can be directly used in various operational status analysis scenarios such as equipment identification, status perception, anomaly detection, and operational simulation, without the need to redesign the feature extraction and coding process for each task, thus exhibiting good reusability.

[0230] In an exemplary embodiment, the electrical quantity time-series data includes at least current time-series data, active power time-series data, and harmonic component time-series data; the event detection module 604 is further configured to, based on the current time-series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points; based on the active power time-series data, determine the first-order difference component of the power at the current sampling point, and the maximum absolute value of the first-order difference component of the power at all sampling points; based on the harmonic component time-series data, determine the harmonic disturbance intensity at the current sampling point, and the maximum harmonic disturbance intensity at all sampling points; and based on the comprehensive mutation scoring function, determine the first-order difference component of the current, the first-order difference component of the power, and the maximum absolute value of the harmonic disturbance intensity at the current sampling point. The comprehensive mutation score of the current sampling point is determined by the maximum absolute values ​​of the first-order difference components of current, power, and harmonic disturbance intensity of all sampling points. If the comprehensive mutation score of the current sampling point is greater than a preset threshold, the current sampling point is determined as a candidate event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is less than a preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and adjacent candidate event trigger points is determined as the target event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0231] In an exemplary embodiment, the feature extraction module 606 is further configured to perform phase alignment, amplitude normalization, and temporal registration processing on the multi-timescale windows respectively.

[0232] In one exemplary embodiment, the multi-timescale window includes a millisecond-level window, a second-level window, and a minute-level window.

[0233] In an exemplary embodiment, the feature extraction module 606 is further configured to extract peak impulse, transient energy, and spectral disturbance features from electrical quantity time series data based on a processed millisecond-level window; extract change slope, fluctuation intensity, and recovery time from electrical quantity time series data based on a processed second-level window; and extract steady-state mean, steady-state offset, and steady-state volatility from electrical quantity time series data based on a processed minute-level window.

[0234] In an exemplary embodiment, the encoding module 608 is further configured to map multi-scale features based on linear mapping functions corresponding to multi-timescale windows respectively, and concatenate the multi-scale mapping values ​​obtained by mapping to obtain fused features; or, map multi-scale features based on linear mapping functions corresponding to multi-timescale windows respectively, and perform weighted fusion of the multi-scale mapping values ​​obtained by mapping to obtain fused features; or, map multi-scale features based on feature rearrangement mapping functions, and construct a two-dimensional feature map based on the multi-scale mapping values ​​obtained by mapping to obtain fused features.

[0235] The modules in the aforementioned power data processing device based on unified coding can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0236] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power data processing method based on unified coding.

[0237] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0238] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: collecting multi-dimensional electrical quantity time-series data from the user side; detecting target event trigger points based on the electrical quantity time-series data; constructing multi-time-scale windows based on the target event trigger points; performing cross-scale consistency processing on the multi-time-scale windows; extracting multi-scale features corresponding to the electrical quantity time-series data based on the processed multi-time-scale windows; fusing and encoding the multi-scale features to obtain a unified encoding result; and using the unified encoding result for power equipment operation status analysis, wherein the operation status analysis includes at least one of equipment identification, status perception, anomaly detection, and operation simulation.

[0239] In one embodiment, when the processor executes the computer program, it further performs the following steps: Based on current time-series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points; based on active power time-series data, determine the first-order difference component of the power at the current sampling point, and the maximum absolute value of the first-order difference component of the power at all sampling points; based on harmonic component time-series data, determine the harmonic disturbance intensity at the current sampling point, and the maximum harmonic disturbance intensity at all sampling points; based on the comprehensive mutation scoring function, determine the first-order difference component of the current, the first-order difference component of the power, the harmonic disturbance intensity, and the maximum absolute value of the first-order difference component of the current at the current sampling point. The maximum absolute value, the maximum absolute value of the first-order power difference component, and the maximum harmonic disturbance intensity are used to determine the comprehensive mutation score of the current sampling point. If the comprehensive mutation score of the current sampling point is greater than a preset threshold, the current sampling point is determined as a candidate event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is less than a preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and adjacent candidate event trigger points is determined as the target event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0240] In one embodiment, when the processor executes the computer program, it further performs the following steps: phase alignment, amplitude normalization, and timing registration processing on the multi-timescale windows respectively.

[0241] In one embodiment, when the processor executes the computer program, it further performs the following steps: extracting peak impulse, transient energy, and spectral disturbance characteristics from electrical quantity time series data based on a processed millisecond-level window; extracting change slope, fluctuation intensity, and recovery time from electrical quantity time series data based on a processed second-level window; and extracting steady-state mean, steady-state offset, and steady-state volatility from electrical quantity time series data based on a processed minute-level window.

[0242] In one embodiment, when the processor executes the computer program, it further performs the following steps: mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and concatenating the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and weighting and fusing the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on feature rearrangement mapping functions, and constructing a two-dimensional feature map based on the multi-scale mapping values ​​obtained by mapping to obtain fused features.

[0243] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: collecting multi-dimensional electrical quantity time-series data from the user side; detecting target event trigger points based on the electrical quantity time-series data; constructing multi-time-scale windows based on the target event trigger points; performing cross-scale consistency processing on the multi-time-scale windows; extracting multi-scale features corresponding to the electrical quantity time-series data based on the processed multi-time-scale windows; fusing and encoding the multi-scale features to obtain a unified encoding result; and using the unified encoding result for power equipment operation status analysis, wherein the operation status analysis includes at least one of equipment identification, status perception, anomaly detection, and operation simulation.

[0244] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: Based on current time-series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points; based on active power time-series data, determine the first-order difference component of the power at the current sampling point, and the maximum absolute value of the first-order difference component of the power at all sampling points; based on harmonic component time-series data, determine the harmonic disturbance intensity at the current sampling point, and the maximum harmonic disturbance intensity at all sampling points; based on the comprehensive mutation scoring function, determine the first-order difference component of the current, the first-order difference component of the power, the harmonic disturbance intensity, and the maximum absolute value of the first-order difference component of the current at all sampling points. The maximum absolute value, the maximum absolute value of the first-order power difference component, and the maximum harmonic disturbance intensity are used to determine the comprehensive mutation score of the current sampling point. If the comprehensive mutation score of the current sampling point is greater than a preset threshold, the current sampling point is determined as a candidate event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is less than a preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and adjacent candidate event trigger points is determined as the target event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0245] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: performing phase alignment, amplitude normalization, and timing registration on the multi-timescale windows respectively.

[0246] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: extracting peak impulse, transient energy, and spectral disturbance characteristics from electrical quantity time series data based on a processed millisecond-level window; extracting change slope, fluctuation intensity, and recovery time from electrical quantity time series data based on a processed second-level window; and extracting steady-state mean, steady-state offset, and steady-state volatility from electrical quantity time series data based on a processed minute-level window.

[0247] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and concatenating the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and weighting and fusing the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on feature rearrangement mapping functions, and constructing a two-dimensional feature map based on the multi-scale mapping values ​​obtained by mapping to obtain fused features.

[0248] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: collecting multi-dimensional electrical quantity time-series data from the user side; detecting target event trigger points based on the electrical quantity time-series data; constructing multi-time-scale windows based on the target event trigger points; performing cross-scale consistency processing on the multi-time-scale windows; extracting multi-scale features corresponding to the electrical quantity time-series data based on the processed multi-time-scale windows; fusing and encoding the multi-scale features to obtain a unified encoding result; and using the unified encoding result for power equipment operation status analysis, wherein the operation status analysis includes at least one of equipment identification, status perception, anomaly detection, and operation simulation.

[0249] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: Based on current time-series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points; based on active power time-series data, determine the first-order difference component of the power at the current sampling point, and the maximum absolute value of the first-order difference component of the power at all sampling points; based on harmonic component time-series data, determine the harmonic disturbance intensity at the current sampling point, and the maximum harmonic disturbance intensity at all sampling points; based on the comprehensive mutation scoring function, determine the first-order difference component of the current, the first-order difference component of the power, the harmonic disturbance intensity, and the maximum absolute value of the first-order difference component of the current at all sampling points. The maximum absolute value, the maximum absolute value of the first-order power difference component, and the maximum harmonic disturbance intensity are used to determine the comprehensive mutation score of the current sampling point. If the comprehensive mutation score of the current sampling point is greater than a preset threshold, the current sampling point is determined as a candidate event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is less than a preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and adjacent candidate event trigger points is determined as the target event trigger point. If the time interval between a candidate event trigger point and an adjacent candidate event trigger point is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

[0250] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: performing phase alignment, amplitude normalization, and timing registration on the multi-timescale windows respectively.

[0251] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: extracting peak impulse, transient energy, and spectral disturbance characteristics from electrical quantity time series data based on a processed millisecond-level window; extracting change slope, fluctuation intensity, and recovery time from electrical quantity time series data based on a processed second-level window; and extracting steady-state mean, steady-state offset, and steady-state volatility from electrical quantity time series data based on a processed minute-level window.

[0252] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and concatenating the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on linear mapping functions corresponding to multi-time-scale windows respectively, and weighting and fusing the multi-scale mapping values ​​obtained by mapping to obtain fused features; or mapping multi-scale features based on feature rearrangement mapping functions, and constructing a two-dimensional feature map based on the multi-scale mapping values ​​obtained by mapping to obtain fused features.

[0253] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0254] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0255] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0256] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A power data processing method based on unified coding, characterized in that, The method includes: Collect multi-dimensional electrical quantity time-series data from the user side; Based on the electrical quantity timing data, detect the target event trigger point; Construct a multi-timescale window based on the target event trigger point; Perform cross-scale consistency processing on the multi-timescale windows; Based on the processed multi-timescale window, multi-scale features corresponding to the electrical quantity time series data are extracted; The multi-scale features are fused and encoded to obtain a unified coding result; the unified coding result is used for the operation status analysis of power equipment, and the operation status analysis includes at least one of equipment identification, status perception, anomaly detection and operation inference.

2. The method according to claim 1, characterized in that, The electrical quantity time series data includes at least current time series data, active power time series data, and harmonic component time series data; The step of detecting the target event trigger point based on the electrical quantity timing data includes: Based on the current time series data, determine the first-order difference component of the current at the current sampling point, and the maximum absolute value of the first-order difference component of the current at all sampling points. Based on the active power time series data, determine the first-order power difference component of the current sampling point and the maximum absolute value of the first-order power difference component of all sampling points. Based on the harmonic component time series data, determine the harmonic disturbance intensity at the current sampling point and the maximum value of the harmonic disturbance intensity at all sampling points; Based on the comprehensive mutation rate scoring function, the comprehensive mutation rate score of the current sampling point is determined according to the first-order difference component of current, the first-order difference component of power, the harmonic disturbance intensity of the current sampling point, and the maximum absolute value of the first-order difference component of current, the maximum absolute value of the first-order difference component of power, and the maximum harmonic disturbance intensity of all sampling points. If the overall mutation score of the current sampling point is greater than a preset threshold, the current sampling point will be identified as a candidate event trigger point. If the time interval between the candidate event trigger point and the adjacent candidate event trigger point is less than the preset trigger interval, the candidate event trigger point with the larger comprehensive mutation score among the candidate event trigger point and the adjacent candidate event trigger points shall be determined as the target event trigger point. If the time interval between the candidate event trigger point and the adjacent candidate event trigger point is not less than the preset trigger interval, the candidate event trigger point is determined as the target event trigger point.

3. The method according to claim 1, characterized in that, The cross-scale consistency processing of the multi-timescale windows includes: Phase alignment, amplitude normalization, and timing registration are performed on the multi-timescale windows respectively.

4. The method according to claim 1, characterized in that, The multi-timescale windows include millisecond-level windows, second-level windows, and minute-level windows.

5. The method according to claim 4, characterized in that, The step of extracting multi-scale features corresponding to the electrical quantity time series data based on the processed multi-time-scale window includes: Based on the processed millisecond-level window, peak impulse, transient energy, and spectral perturbation characteristics are extracted from the electrical quantity time-series data; Based on the processed second-level window, the change slope, fluctuation intensity, and recovery time are extracted from the electrical quantity time series data; Based on the processed minute-level window, the steady-state mean, steady-state offset, and steady-state volatility are extracted from the electrical quantity time-series data.

6. The method according to any one of claims 1 to 5, characterized in that, The fusion of the multi-scale features includes: The multi-scale features are mapped based on the linear mapping functions corresponding to the multi-timescale windows, and the resulting multi-scale mapping values ​​are concatenated to obtain the fused features; or, The multi-scale features are mapped based on the linear mapping functions corresponding to the multi-timescale windows, and the resulting multi-scale mapping values ​​are weighted and fused to obtain fused features; or, The multi-scale features are mapped using a feature rearrangement mapping function, and a two-dimensional feature map is constructed based on the multi-scale mapping values ​​obtained from the mapping to obtain the fused features.

7. A power data processing device based on unified coding, characterized in that, The device includes: The data acquisition module is used to collect multi-dimensional electrical quantity time-series data from the user side; The event detection module is used to detect the target event trigger point based on the electrical quantity timing data; The feature extraction module is used to construct a multi-timescale window based on the target event trigger point; perform cross-scale consistency processing on the multi-timescale window; and extract multi-scale features corresponding to the electrical quantity time series data based on the processed multi-timescale window. The encoding module is used to fuse and encode the multi-scale features to obtain a unified encoding result; the unified encoding result is used for the operation status analysis of power equipment, and the operation status analysis includes at least one of equipment identification, status perception, anomaly detection and operation simulation.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images