Early warning method for electric energy metering box failure based on time series data analysis

Abstract

The present application relates to a power metering box fault early warning method based on time series data analysis, aiming to solve the problem of low sensitivity, false alarm and inability to adapt to dynamic migration of working conditions of existing electrical equipment state judgment. Its core scheme is: through multi-dimensional operation parameter acquisition and standardization, extracting load fluctuation rate, phase offset trend and daily cycle similarity, forming a device state identification vector, combining state continuity analysis, dynamically and adaptively generating a double-ring hysteresis decision threshold with inertia constraint. The method further integrates historical working condition migration map and real-time parameter abnormal crossing depth to quantitatively evaluate the warning credibility. Through multi-level judgment and soft locking-smooth release mechanism, accurate grading warning and reliability evaluation of parameter fluctuation or abnormality are realized, false alarm and missed alarm are significantly reduced, and the practicality and intelligent level of power metering box operation and maintenance are improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault early warning and intelligent operation and maintenance technology, and in particular to an early warning method for power metering box faults based on time series data analysis. Background Technology

[0002] Existing fault early warning technologies for electricity metering boxes are widely used in smart grids and distribution systems. They typically rely on real-time monitoring of operating parameters such as voltage, current, temperature, and harmonics, comparing these parameters with set thresholds to automatically detect and categorize abnormal conditions. Mainstream industry solutions often employ methods like single-point threshold determination, periodic smoothing, and moving averages to control early warning level switching, enabling rapid response in scenarios with significant parameter fluctuations but stable trends. However, with increasing density of electricity metering boxes and increasingly complex operating environments, the dynamic overlay of multi-source operating data and the frequent occurrence of sudden fluctuations mean that traditional static threshold mechanisms lack effective modeling of the continuity of equipment operating states and the inertia of operating condition transitions. This can easily lead to frequent triggering of early warning level switching when parameters oscillate within critical ranges, resulting in decreased stability in operation and maintenance decisions.

[0003] Recent studies have introduced solutions such as adaptive threshold adjustment, machine learning-based operating condition classification, temporal anomaly detection, and reinforcement learning-based dynamic parameter tuning to improve early warning accuracy and scenario adaptability. However, these solutions generally suffer from high deployment complexity and high requirements for computing power and basic data, making them unsuitable for large-scale deployment on edge nodes with limited resources or communication conditions. Furthermore, strategies based on operating condition classification and multi-source data fusion often rely on complex model outputs, resulting in high debugging and maintenance barriers. They also lack robustness in the face of extreme operating conditions or sudden data anomalies, failing to fundamentally address the essential problem of level oscillations.

[0004] The main shortcomings of current technologies are: traditional static threshold mechanisms lack modeling of state continuity and transition inertia, easily leading to frequent switching of warning levels; some solutions based on external feedback closed loops or reinforcement learning are complex to deploy and unsuitable for edge deployment; existing strategies have limited intelligent constraint capabilities on state transition paths and state locking, and do not provide mature and efficient suppression strategies specifically for the "state oscillation" phenomenon during warning level switching. Therefore, the industry urgently needs an improved solution that can leverage the internal continuity and inertia characteristics of time-series data to achieve smooth switching of multi-level warnings, automatically adapt to changes in operating conditions, and possess high compatibility and low deployment costs, in order to improve the stability and user experience of the power metering box fault warning system. Summary of the Invention

[0005] This application provides an early warning method for power metering box faults based on time-series data analysis, aiming to solve one of the problems or issues of the existing technology mentioned in the background.

[0006] The early warning method for power metering box faults based on time-series data analysis provided in this application specifically includes: S1: Obtain the multi-dimensional operating parameters of the power metering box, calculate the state feature parameters based on the multi-dimensional operating parameters, and generate a state identification vector; S2: Based on the analysis results of the changes of the state identifier vector in multiple consecutive time windows, construct a double-loop hysteresis threshold structure and generate asymmetric threshold boundary pairs; S3: Generate a state transition graph library using the working condition transition sequence in the historical operation and maintenance records, and predict the state evolution direction based on the matching path of the state identifier vector in the state transition graph library, and generate a state evolution prediction factor that includes at least the slope direction adjustment amount. S4: Integrate the asymmetric threshold boundary pair with the state evolution prediction factor, and adjust the trigger sensitivity of the asymmetric threshold boundary pair according to the slope direction adjustment amount to generate a dynamic judgment threshold group; S5: Based on the dynamic judgment threshold group, perform hierarchical traversal judgment on the multidimensional operating parameters collected in real time, and generate a primary warning status signal with a time lock mark; S6: If the multi-dimensional operating parameters collected in real time are continuously within the new level's allowable range multiple times during the duration of the soft-lock timer, then the time lock flag is released and the smooth transition logic is executed to generate a stable early warning level instruction. S7: Based on the current state identifier vector, dynamic judgment threshold group and soft lock timer, generate a state confidence label and attach the state confidence label to the stable warning level instruction to form a complete warning data packet.

[0007] S8: Determine whether the state confidence label in the complete early warning data packet is higher than the preset baseline. If so, push a visual prompt to the operation and maintenance terminal. Otherwise, store the complete early warning data packet in the local cache and use it as historical data input for the next round of state identifier vector generation.

[0008] The early warning method for power metering box faults based on time-series data analysis provided in this application has the following beneficial effects: (1) To address the problem of frequent level oscillations caused by single-point threshold determination in traditional power metering box fault early warning, this invention constructs a hierarchical early warning decision framework with state evolution memory capability. By introducing lightweight state fingerprints such as load volatility, phase offset trend and daily cycle similarity, it provides an adaptive adjustment basis for the asymmetric hysteresis loop mechanism, maintains strong anti-interference capability under transient disturbances, and maintains sufficient response sensitivity under continuous deterioration conditions, effectively overcoming the contradiction between stability and sensitivity.

[0009] (2) This invention designs a condition-aware hierarchical locking mechanism and a path prediction-driven buffer dynamic adjustment strategy. After the early warning is activated, a soft locking timer determined by the stability level of the operating condition is started. It can only be downgraded if it meets the dual conditions of continuous sampling and time constraints. At the same time, the evolution path is predicted based on the state transition map library generated by historical operation and maintenance sequences. The slope of the hysteresis loop and the range of the transition buffer are adjusted in advance so that the early warning logic fits the actual degradation trajectory of the equipment. Confidence labels are added to filter low-confidence intermediate states, which significantly improves the time coherence and engineering credibility of the early warning evolution.

[0010] (3) This solution completely avoids dependence on external feedback loops or complex models. Starting from the inherent continuity and evolutionary inertia of the device's own time-series data, it constructs an endogenous, lightweight, and robust early warning mechanism that can be deployed at the edge. The core logic does not rely on large-scale labeled samples or offline training models, avoiding the risks of operating condition drift and cold start. Without additional hardware costs, it significantly improves the accuracy of early fault identification and the orderly operation and maintenance response of distributed power metering boxes, providing key technical support for building a highly reliable and self-consistent edge intelligent diagnostic system. Attached Figure Description

[0011] Figure 1 This is the main flowchart of an early warning method for power metering box faults based on time-series data analysis; Figure 2 This is a sub-flowchart of an early warning method for power metering box faults based on time-series data analysis; Figure 3 This is another sub-flowchart of the method for early warning of faults in electricity metering boxes based on time-series data analysis. Detailed Implementation

[0012] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0013] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0014] like Figure 1As shown, this application provides an early warning method for power metering box faults based on time-series data analysis, specifically including: S1: Obtain the multi-dimensional operating parameters of the power metering box, calculate the state feature parameters based on the multi-dimensional operating parameters, and generate a state identification vector; S2: Based on the analysis results of the changes of the state identifier vector in multiple consecutive time windows, construct a double-loop hysteresis threshold structure and generate asymmetric threshold boundary pairs; S3: Generate a state transition graph library using the working condition transition sequence in the historical operation and maintenance records, and predict the state evolution direction based on the matching path of the state identifier vector in the state transition graph library, and generate a state evolution prediction factor that includes at least the slope direction adjustment amount. S4: Integrate the asymmetric threshold boundary pair with the state evolution prediction factor, and adjust the trigger sensitivity of the asymmetric threshold boundary pair according to the slope direction adjustment amount to generate a dynamic judgment threshold group; S5: Based on the dynamic judgment threshold group, perform hierarchical traversal judgment on the multidimensional operating parameters collected in real time, and generate a primary warning status signal with a time lock mark; S6: If the multi-dimensional operating parameters collected in real time are continuously within the new level's allowable range multiple times during the duration of the soft-lock timer, then the time lock flag is released and the smooth transition logic is executed to generate a stable early warning level instruction. S7: Based on the current state identifier vector, dynamic judgment threshold group and soft lock timer, generate a state confidence label and attach the state confidence label to the stable warning level instruction to form a complete warning data packet.

[0015] S8: Determine whether the state confidence label in the complete early warning data packet is higher than the preset baseline. If so, push a visual prompt to the operation and maintenance terminal. Otherwise, store the complete early warning data packet in the local cache and use it as historical data input for the next round of state identifier vector generation.

[0016] Step S1: Obtain the multi-dimensional operating parameters of the electricity metering box, and calculate the state characteristic parameters based on the multi-dimensional operating parameters to generate a state identification vector. The multi-dimensional operating parameters include: voltage, current, temperature, and harmonic content; the state characteristic parameters include: load volatility, phase shift trend, and daily cycle similarity. Specifically, this includes: S1.1: Synchronize and align the voltage, current, temperature, and harmonic content data collected by the power metering box within a continuous time window, and perform noise filtering to eliminate sensor measurement errors and environmental interference, and generate a standardized multi-dimensional operating parameter sequence.

[0017] Sub-step S1.1 aims to perform spatiotemporal alignment and denoising preprocessing on the original multidimensional operating parameters, providing a high signal-to-noise ratio data foundation for subsequent feature extraction.

[0018] The voltage signal, current signal, temperature data and harmonic content data collected by the power metering box within a continuous time window are acquired to construct an initial multi-source heterogeneous dataset.

[0019] Timestamp verification is performed on the data streams of each sensor, and linear interpolation resampling is performed based on the minimum sampling interval to eliminate timing misalignment caused by communication delay and generate a time-synchronized multidimensional parameter matrix.

[0020] To address high-frequency random noise in voltage and current signals, a soft thresholding denoising process based on wavelet thresholding is adopted. The Daubechies4 wavelet basis is selected for three-level decomposition, and detail coefficients below the adaptive threshold are removed to reconstruct a smooth electrical waveform sequence.

[0021] The temperature data was processed using a sliding median filter with a window length of 5 sampling points to remove spike interference caused by sudden environmental changes and retain the true trend component of temperature change.

[0022] Outlier detection is performed on the harmonic content data. Outliers are identified and replaced using the three-standard-deviation criterion to ensure the statistical consistency of the harmonic spectrum data.

[0023] The filtered multidimensional data is subjected to Z-Score normalization transformation, and the normalized value is calculated using the following formula: Where x is the original measurement value, μ is the mean within the sliding window, and σ is the standard deviation. norm These are the standardized dimensionless parameters.

[0024] The standardized voltage, current, temperature, and harmonic data are spliced ​​together along the time axis to form a standardized multidimensional operating parameter sequence with a unified structure.

[0025] By using synchronous alignment and multi-level filtering, the original discrete data from the previous step is transformed into a standard multidimensional operating parameter sequence with time consistency and high signal-to-noise ratio, thereby achieving the expected technical effect of eliminating sensor errors and environmental interference.

[0026] S1.2: Based on the current signal amplitude variation characteristics in the standardized multidimensional operating parameter sequence, perform sliding variance statistical calculations to quantify the severity of load changes and generate a load volatility value that characterizes the dynamic characteristics of the load.

[0027] Receive the standardized multidimensional operating parameter sequence generated by S1.1, and extract the current signal amplitude component as the original data source for load volatility calculation.

[0028] Set the sliding time window length and step size, segment the current signal amplitude sequence, and obtain the local current subsequence of the current time t and the previous W-1 sampling points.

[0029] Calculate the arithmetic mean of the local current subsequence as a benchmark reference value characterizing the average load level within the current window.

[0030] The sliding variance of the current window is calculated using the following formula: Among them, I t-i Let be the current amplitude at time ti. Let W be the average current within the window, and W be the length of the sliding time window. Let be the sliding variance of the window at the current time t.

[0031] The calculated sliding variance value is normalized to eliminate the influence of dimensions and generate a dimensionless load volatility value.

[0032] By using the sliding variance statistical operation method, the standardized current sequence from the previous step is transformed into a load volatility value that quantifies the severity of load changes, thereby achieving accurate capture of the dynamic characteristics of the load and providing a key basis for state stability in the subsequent construction of an adaptive hysteresis threshold.

[0033] S1.3: Based on the zero-crossing time difference of the voltage and current signals in the standardized multidimensional operating parameter sequence, perform linear regression fitting analysis to capture the long-term drift direction of the phase angle and generate a phase offset trend coefficient characterizing the stability of the electrical connection.

[0034] Voltage and current signals are extracted from the standardized multidimensional operating parameter sequence to locate the zero-crossing point of each fundamental cycle. The time difference between the voltage and current zero-crossing points within the same cycle is calculated to construct a phase time difference sequence.

[0035] A sliding window is used to extract data from the phase time difference sequence, and the data from the most recent N periods are selected as the fitting sample set.

[0036] The least squares method is used to perform linear regression modeling on the sample set to fit the trend line of phase time difference changing with time.

[0037] The phase offset trend coefficient is calculated using the following formula: Where k is the phase offset trend coefficient, t i For the i-th sampling time, τ i Let i be the phase time difference of the i-th period. and denoted as the mean of the time and phase difference, respectively, and n equals the number of data samples from the most recent N periods captured by the sliding window.

[0038] Analyze the sign and absolute value of the phase shift trend coefficient to determine the drift direction and drift rate of the phase angle.

[0039] By using linear regression fitting, the discrete phase time difference is transformed into a phase offset trend coefficient that characterizes the stability of electrical connections, thus enabling the quantitative capture of early characteristics of progressive faults such as poor contact.

[0040] S1.4: Based on the waveform data of the standardized multidimensional operating parameter sequence in the current time window and the historical benchmark waveform data of the same period, perform Pearson correlation coefficient calculation to evaluate the periodicity of the operating mode and generate a daily periodic similarity index that characterizes the regularity of the operating conditions.

[0041] Voltage and current waveform data segments within the current time window are extracted from the standardized multidimensional operating parameter sequence as real-time operating condition samples to be evaluated. Historical benchmark waveform data with the same weekday attribute and hourly interval as the current time window are extracted from the historical database to construct a reference time series. Length-aligned interpolation is performed on the real-time operating condition sample and the reference time series to ensure strict consistency in the number of sampling points, eliminating phase misalignment errors caused by minor fluctuations in sampling frequency. The covariance matrix between the real-time operating condition sample and the reference time series is calculated to quantify their joint fluctuation characteristics in amplitude variation trends. The standard deviations of the real-time operating condition sample and the reference time series are calculated separately to characterize the dispersion and energy distribution range of their respective data. The linear correlation strength between the two series is calculated using the Pearson correlation coefficient formula. Where r is the daily cycle similarity index, representing the strength of the linear correlation between the current real-time operating condition and the historical benchmark operating condition in terms of waveform morphology; x i For real-time voltage or current sample values, y i This represents the historical baseline sample value for the same period, where n is the total number of sample points. This is the average value of the real-time operating condition sample sequence. The average value of the historical benchmark sequence is used. The calculated daily cycle similarity index undergoes absolute value mapping to eliminate semantic ambiguity caused by negative correlation and preserve the consistency of waveform morphology. The mapped values ​​are then normalized to the 0-1 range to generate a daily cycle similarity index characterizing the regularity of operating conditions. Through this processing method, the results of the previous step are transformed into a quantitative index reflecting the periodic repetition of equipment operating modes, achieving the expected technical effect of effectively filtering non-fault anomalies caused by random load fluctuations.

[0042] S1.5: Based on the generated load volatility value, phase offset trend coefficient and daily cycle similarity index, perform vector encapsulation and normalization processing to form a unified dimension of working condition description data structure and generate a status identifier vector representing the current comprehensive working condition of the equipment.

[0043] Step S2: Based on the analysis results of the changes in the state identifier vector over multiple consecutive time windows, construct a double-loop hysteresis threshold structure and generate asymmetric threshold boundary pairs. Preferably, the analysis results are based on the changes in the state identifier vector over three consecutive time windows. The double-loop hysteresis threshold structure has adaptive scaling characteristics. Constructing the double-loop hysteresis threshold structure and generating asymmetric threshold boundary pairs includes: defining the difference between the uplink trigger threshold and the downlink release threshold in the double-loop hysteresis threshold structure as the dynamic hysteresis loop width, and generating the asymmetric threshold boundary pairs based on the dynamic hysteresis loop width. Specifically, this includes: S2.1: Perform time-series difference operation on the state identifier vector sequences of the three most recent time windows to calculate the gradient of changes in load volatility, phase offset trend and daily cycle similarity between adjacent windows, and generate a state change gradient vector that characterizes the continuity of the working condition evolution.

[0044] Load volatility, phase offset trend, and daily cycle similarity components are extracted from the state identifier vector sequence of three consecutive time windows generated by S1.5, and a three-dimensional time series data matrix is ​​constructed as the input object for difference operation.

[0045] Perform element-wise subtraction on the state identifier vectors of adjacent time windows to calculate the numerical differences between the current window and the previous window, and between the previous window and the window before that, generating a first-order difference sequence.

[0046] The rate of change of each component is calculated based on the first-order difference sequence, and the acceleration of load volatility, the angular velocity of phase shift, and the decay rate of daily cycle similarity are quantified to form the original gradient data.

[0047] The original gradient data is normalized using the Euclidean norm to eliminate the influence of different physical dimensions on the gradient magnitude and ensure the comparability of multidimensional features in the vector space.

[0048] The normalized gradient components are recombined to construct a state change gradient vector containing both directional and amplitude information. This vector accurately represents the continuity and abrupt change tendency of the working condition evolution.

[0049] By using time-series difference and norm normalization, the discrete state identifier vector is transformed into a state change gradient vector that represents the continuity of the operating condition evolution. This enables a quantitative description of the dynamic migration characteristics of the equipment's operating state and provides a precise gradient basis for subsequent adaptive hysteresis loop width adjustment.

[0050] S2.2: Perform a stability level mapping operation based on the magnitude of the state change gradient vector to quantify the degree of jitter or stability trend of the current equipment operating state and generate an operating condition stability coefficient that characterizes the degree of operating condition stability.

[0051] Receive the state change gradient vector generated in the previous steps. This vector contains the differential components of load volatility, phase shift trend, and daily cycle similarity between adjacent time windows.

[0052] The energy distribution matrix is ​​constructed by squaring each component of the gradient vector of state change to eliminate directional differences and amplify the weights of significant changes.

[0053] The sum of the squared values ​​of each component is used to calculate the sum of squared Euclidean norms, which characterizes the severity of the evolution of the working condition: Where G is the state change gradient vector, g i Let be the gradient value of the change in the i-th feature dimension.

[0054] Perform a square root operation on the above sum of squares to obtain the magnitude of the gradient vector of state change. This value directly reflects the overall jitter amplitude of the current device operating state.

[0055] A preset stability mapping function is provided, which uses an inverse proportional decay characteristic to map the modulus value to a stability coefficient range between 0 and 1. The stability mapping function is as follows: Where S is the working condition stability coefficient and α is the sensitivity adjustment factor, used to control the rate at which the modulus affects stability.

[0056] By using the above processing method, the state change gradient of the previous step is transformed into a quantified operating condition stability coefficient, which realizes an accurate measurement of the smoothness of equipment operation and provides a benchmark for subsequent dynamic adjustment of the hysteresis loop width.

[0057] S2.3: The preset basic hysteresis loop width parameter is nonlinearly scaled and adjusted using the operating condition stability coefficient to dynamically adjust the tolerance of the threshold range according to the degree of operating condition jitter, thereby generating a dynamic hysteresis loop width value that adapts to the current operating environment.

[0058] The operating condition stability coefficient generated in the previous steps is obtained as an input variable for nonlinear scaling adjustment. This coefficient quantifies the degree of fluctuation or stability trend of the equipment's operating state within a continuous time window.

[0059] Read the preset basic hysteresis loop width parameter, which defines the default threshold tolerance of the system under standard steady-state conditions, and serves as a benchmark reference value for dynamic adjustment.

[0060] A nonlinear mapping function is constructed to map the operating condition stability coefficient to a scaling factor. An sigmoid function or an exponentially decaying function is used to ensure that the scaling factor increases rapidly in the low stability (high jitter) range, while keeping the scaling factor close to 1 in the high stability range.

[0061] A nonlinear scaling operation is performed, multiplying the base hysteresis loop width parameter by the calculated scaling factor. The dynamic hysteresis loop width is then calculated using the following formula: Where W is the dynamic hysteresis loop width, W0 is the basic hysteresis loop width parameter, k is the maximum scaling factor, α is the attenuation constant, and S is the operating condition stability coefficient.

[0062] Boundary constraint verification is performed on the calculated dynamic hysteresis loop width value to ensure that it is between the preset minimum safe width and the maximum allowable width, so as to prevent false alarms caused by excessively narrow thresholds due to extreme working conditions or false alarms caused by excessively wide thresholds.

[0063] The output is the final dynamic hysteresis loop width value adapted to the current operating environment. This value directly determines the distance between the subsequent uplink trigger threshold and the downlink release threshold.

[0064] By using a nonlinear scaling adjustment process, the operating condition stability coefficient from the previous step is converted into a dynamic hysteresis loop width value, achieving the expected technical effect of adaptively adjusting the threshold range tolerance according to the degree of operating condition jitter, effectively balancing early warning sensitivity and anti-interference capability.

[0065] S2.4: Based on the dynamic hysteresis loop width value and the real-time average of the current multidimensional operating parameters, perform asymmetric boundary offset calculation to determine the specific numerical positions of the uplink trigger threshold and the downlink release threshold, and generate an asymmetric threshold boundary pair containing the uplink trigger threshold and the downlink release threshold.

[0066] Obtain the dynamic hysteresis loop width value generated in the previous steps, and extract the real-time arithmetic mean of the multidimensional running parameter sequence within the current time window as the reference center point for asymmetric boundary offset calculation.

[0067] Based on the asymmetric characteristics of the fault evolution of the power metering box, the bias coefficient of the uplink trigger threshold is set to be greater than the bias coefficient of the downlink release threshold, so as to reflect the high sensitivity to the fault rise trend and the anti-disturbance of the recovery process.

[0068] The specific numerical location of the uplink trigger threshold is calculated using the following formula: Among them, Tup t The uplink trigger threshold is the current time, μM is the real-time average of the multidimensional operating parameters, WH is the dynamic hysteresis loop width, and α is the uplink bias adjustment factor, with a value range of 1.2 to 1.5.

[0069] The specific numerical location of the downlink release threshold is calculated using the following formula: Among them, Tdown t The current downlink reset threshold is set to β, which is the downlink bias adjustment factor with a value range of 0.8 to 1.0 to ensure that the downlink threshold is closer to the mean to accelerate state reset.

[0070] The calculated uplink trigger threshold and downlink release threshold are paired and encapsulated to form an asymmetric threshold boundary pair data structure containing upper and lower boundary values.

[0071] By using the above-mentioned asymmetric boundary offset calculation method, the dynamic hysteresis loop width and the real-time operating condition mean are transformed into asymmetric threshold boundary pairs with directional differences, thereby achieving the expected technical effect of differentiated configuration of early warning triggering and cancellation conditions.

[0072] S2.5: Encapsulate the asymmetric threshold boundary pair into a double-loop hysteresis threshold structure data object with adaptive scaling characteristics to form a dynamic judgment benchmark that can automatically adjust the sensitivity according to the operating condition migration characteristics, and generate the final double-loop hysteresis threshold structure.

[0073] like Figure 2 As shown, step S3: A state transition graph library is generated using the operating condition transition sequences from historical operation and maintenance records. Based on the matching path of the state identifier vector in the state transition graph library, the state evolution direction is predicted, and a state evolution prediction factor containing at least a slope direction adjustment is generated. In this embodiment, the state evolution prediction factor further includes: a transition buffer range. Specifically, it includes: S3.1 extracts and standardizes the actual operating condition transition sequences in historical operation and maintenance records, and constructs a state transition graph library containing the initial state node, the target state node, and the transition probability weights, which serves as the basic data entity for subsequent state evolution analysis.

[0074] Multi-dimensional time-series data segments containing voltage, current, temperature, and harmonic content were retrieved from the historical operation and maintenance database of the power metering boxes. Complete operating condition transition sequences accompanying warning level changes or fault records were selected. The selected raw data underwent timestamp alignment and missing value imputation to eliminate data breakpoints caused by communication interruptions, generating a continuous and complete historical operating condition trajectory dataset. The historical operating condition trajectory dataset was divided into fixed-length sliding windows, and the load volatility, phase offset trend, and daily cycle similarity within each window were calculated and mapped to discretized state identifier vectors. The K-means clustering algorithm was used to spatially partition the full set of state identifier vectors, dividing the feature space into several physically meaningful typical operating condition clusters. The center of each cluster was defined as a state node in the state transition graph. The jump frequency of state nodes between adjacent time windows was statistically analyzed, constructing a directed graph structure where nodes represent typical operating conditions and directed edges represent the direction of operating condition transitions. The transition probability weight of each directed edge was calculated, which is equal to the number of jumps from a specific starting node to the target node divided by the total number of jumps from all outgoing edges of that starting node. The transition probability weights are calculated using the following formula: Wherein, P(S) j |S i ) is from state node S i Transition to state node S j The probability, N(S) i →S j ) is historical data from S i Jump to S j The number of times, the denominator is from S i The sum of all jumps from the starting point. The generated directed graph is pruned, removing sparse edges with transition probabilities below a preset threshold, retaining high-frequency salient paths, and forming a simplified state transition graph library. Through this processing method, unstructured historical operation and maintenance data is transformed into a state transition graph library with topological structure and probabilistic attributes, realizing digital modeling of equipment operating condition evolution patterns and providing basic data support for subsequent prediction of state evolution directions.

[0075] S3.2: Based on the state identifier vector, perform topological matching retrieval in the state transition map library to identify the starting state node that best matches the current equipment operating condition and its associated high-frequency transition path set, and generate a state matching path set representing the potential evolution direction.

[0076] In this embodiment, the high-frequency transition path set refers to a set of several potential evolution paths with high transition probability weights, selected from the state transition graph library and associated with the starting state node that best matches the current equipment operating condition. The high-frequency transition path set is obtained as follows: based on the state identifier vector, a topological matching search is performed in the state transition graph library to identify the starting state node that best matches the current operating condition; all directed transition edges associated with this starting state node are retrieved; the associated paths are sorted in descending order according to their transition probability weights; the Top-N paths with a cumulative weight contribution rate reaching a preset threshold are selected; and invalid branches are removed after a temporal consistency check to form the high-frequency transition path set. Here, N in Top-N refers to the number of high-frequency paths to be retained to meet the cumulative weight contribution rate requirement. Its specific value can be adaptively adjusted according to the total number of edges associated with the starting node in the state transition graph library and the weight distribution characteristics. In this embodiment, N is preferably 3, that is, the top 3 paths with the highest probability weights are retained in the sorted path list.

[0077] The system acquires the state transition graph library constructed in S3.1 and the current state identifier vector generated in S1, mapping the state identifier vector to query nodes in the graph space. It calculates the Euclidean distance between the query node and all starting state nodes in the graph, selecting the K candidate starting nodes with the smallest distances as matching benchmarks. For each candidate starting node, it retrieves all its associated directed transition edges, extracting the target state node and corresponding historical transition probability weights for each edge. The associated paths are sorted in descending order based on their transition probability weights, and the top-N high-frequency transition paths with a cumulative weight contribution rate reaching a preset threshold are selected. A temporal consistency check is performed on each selected high-frequency transition path, eliminating invalid path branches with logical conflicts or probability anomalies. The retained valid high-frequency transition paths and their included starting nodes, intermediate transition nodes, target nodes, and transition probability sequences are encapsulated as structured data. Through topology matching retrieval and probability filtering, discrete historical operation and maintenance records are transformed into a set of state matching paths representing potential evolution directions, enabling forward-looking prediction for early warning decisions.

[0078] S3.3: Calculate the slope of the path evolution trend using the transition probability weights in the state matching path set to quantify the rate and direction of change of the equipment operating parameters within the future time window, and generate a state evolution trend descriptor containing the slope direction adjustment amount.

[0079] Obtain the transition probability weights of each potential evolution path in the state matching path set and the corresponding historical state node parameter sequence.

[0080] Perform a difference operation on the multidimensional running parameters of the starting state node and the target state node in each path to calculate the parameter change vector.

[0081] Dividing the parameter change vector by the corresponding time window interval yields the single-step evolution rate vector, which characterizes the state drift amplitude per unit time.

[0082] A weighted average method is used, with the transition probability weights as coefficients, to linearly combine the single-step evolution rate vectors of all candidate paths.

[0083] The slope vector of the overall evolutionary trend is calculated using the following formula: Where S is the slope vector of the comprehensive evolution trend, w i V represents the normalized transition probability weights for the i-th path. t+1 With V t Δt represents the parameter vectors of the target and initial state nodes, respectively, and Δt is the time interval.

[0084] The magnitude of the slope vector of the comprehensive evolution trend is calculated and the direction angle is extracted to separate the scalar of the rate of change and the indicator of the evolution direction.

[0085] The rate of change scalar is mapped to a sensitivity correction coefficient, the evolution direction identifier is converted into a threshold offset symbol, and a state evolution trend descriptor is generated by encapsulation.

[0086] By weighted aggregation of multi-path evolution rates and decoupling direction and magnitude, discrete path probabilities are transformed into continuous trend control quantities, enabling quantitative prediction of future operating condition migration directions.

[0087] S3.4: Evaluate the uncertainty range of the state transition based on the rate of change in the state evolution trend descriptor, in order to determine the size of the buffer region required to prevent misjudgment, and generate a transition buffer range that defines the safe transition interval.

[0088] Receive the state evolution trend descriptor generated in the previous steps, extract the change rate value and direction identifier contained therein, and use it as the basic input data for evaluating the uncertainty of state transition.

[0089] Based on the numerical value of the rate of change, the ratio of its standard deviation to the historical baseline rate distribution is calculated to quantify the dispersion of the current operating condition evolution and generate an uncertainty coefficient characterizing the reliability of the prediction.

[0090] The uncertainty coefficient is used to exponentially amplify the preset basic buffer width parameter. When the uncertainty coefficient is higher than the threshold, the buffer area is significantly increased to cover the potential range of severe fluctuations.

[0091] The system combines the direction indicator to determine whether it is currently in the upward triggering phase or the downward release phase. If it is in the upward phase, the upper limit of the buffer is expanded upward; if it is in the downward phase, the lower limit of the buffer is expanded downward, thus generating an initial buffer boundary with direction adaptability.

[0092] Smoothing filtering is applied to the initial buffer boundary to eliminate boundary jitter caused by instantaneous rate changes, ensuring the continuity and stability of the transition interval and generating a transition buffer range that defines the safe transition interval.

[0093] By using the above processing method, the result of the previous step is transformed into a spatial constraint index with anti-interference capability, thereby achieving the expected technical effect of preventing misjudgment and improving the stability of early warning switching.

[0094] For example, with an extraction rate of 0.8A / min and a historical baseline standard deviation of 0.2A / min, the calculated uncertainty coefficient is 4.0. The preset base buffer width is 0.5A, and an exponential amplification formula is used: Where W is the buffer width, W0 is the preset basic buffer width of 0.5A, k is the exponential amplification factor of 0.5, and α is the uncertainty coefficient of 4.0. The uncertainty coefficient is provided by an external calculation module and characterizes the degree of uncertainty of the current operating condition or signal fluctuation. The calculated buffer width is 3.69A. If the current stage is uplink, then 3.69A is added to the original uplink trigger threshold as the upper limit of the buffer, forming a wide safe transition range, effectively avoiding false triggering of secondary alarms due to short-term load spikes, and significantly improving the robustness of the system under dynamic operating conditions.

[0095] S3.5: Integrate the slope direction adjustment amount and the transition buffer range and encapsulate them to form a comprehensive control command that can directly drive the dynamic correction of the threshold, and generate the final state evolution prediction factor.

[0096] like Figure 3 As shown, step S4: The asymmetric threshold boundary pair and the state evolution prediction factor are fused, and the trigger sensitivity of the asymmetric threshold boundary pair is corrected according to the slope direction adjustment amount to generate a dynamic decision threshold group. The dynamic decision threshold group possesses inertial constraint characteristics. Specifically, it includes: S4.1: Obtain the asymmetric threshold boundary pairs and state evolution prediction factors generated in the previous steps, and extract the slope direction adjustment amount and transition buffer range from the state evolution prediction factors as the basic input data for correcting the asymmetric threshold boundary pairs.

[0097] Receive the asymmetric threshold boundary pair data object output from the preceding step S2. This object contains the initial settings for the uplink trigger threshold and downlink release threshold for the current warning level. Simultaneously, read the state evolution predictor data structure generated in the preceding step S3. This structure encapsulates the slope direction adjustment amount and transition buffer range derived from the historical operating condition migration map.

[0098] Field parsing is performed on the state evolution predictor to extract the slope direction adjustment value, which represents the trend of parameter change. This value reflects the rate and direction of the equipment's operating state evolution within a future time window. Simultaneously, the transition buffer range in the state evolution predictor is extracted; this parameter defines the width of the safety tolerance interval required to prevent misjudgments.

[0099] The extracted slope direction adjustment amount and the transition buffer range are mapped to a temporary correction variable space in memory, and a logical association index is established with each threshold component in the asymmetric threshold boundary pair. The data type and dimension of the extracted parameters are verified to ensure that the slope adjustment amount has directional sign characteristics and that the buffer range is a non-negative real number.

[0100] A joint input dataset containing the original threshold boundary, slope adjustment amount, and buffer range is constructed as the benchmark input source for subsequent dynamic threshold correction calculations. Through data decoupling and recombination, the state evolution prediction information from the previous step is transformed into control parameters that can directly drive the threshold boundary offset, achieving precise alignment between the early warning judgment benchmark and the equipment state evolution trend.

[0101] S4.2: Perform gradient offset processing on the uplink trigger threshold in the asymmetric threshold boundary pair based on the slope direction adjustment amount, so as to change the rise rate characteristics of the uplink trigger threshold and generate an uplink dynamic trigger threshold with directional inertia.

[0102] It receives the asymmetric threshold boundary pairs and state evolution prediction factors output from the previous steps, and extracts the slope direction adjustment amount as the core control parameter for uplink trigger threshold correction.

[0103] The sign and magnitude of the slope direction adjustment are analyzed to determine whether the current working condition is accelerating upward, decelerating and stabilizing, or falling back, and the corresponding threshold offset polarity is determined.

[0104] A gradient offset operator based on an exponentially decaying function is constructed to map the slope direction adjustment amount to a dynamic offset increment. This operator ensures that a large threshold inertial drag is provided when the state changes rapidly.

[0105] The offset increment of the uplink dynamic trigger threshold is calculated using the following formula: Where ΔT is the threshold offset increment, k is the sensitivity gain coefficient, S is the slope direction adjustment amount, λ is the attenuation factor, and t is the current time window index.

[0106] The calculated offset increment is then algebraically superimposed on the uplink trigger threshold based on the original asymmetric threshold boundary to generate a preliminary corrected candidate value for the uplink trigger threshold.

[0107] The monotonicity constraint of the preliminary revised uplink trigger threshold candidate value is checked to ensure that its value is always higher than the average value of the current real-time operating parameters, so as to prevent the warning from failing due to excessive offset.

[0108] By using gradient offset processing, the state evolution prediction results of the previous step are transformed into an uplink dynamic trigger threshold with directional inertia, which realizes effective filtering of instantaneous disturbances and adaptive adjustment of early warning sensitivity.

[0109] S4.3: The downlink release threshold in the asymmetric threshold boundary pair is widened by using the transition buffer range to increase the hysteresis margin of the downlink release threshold and generate a downlink dynamic release threshold with disturbance rejection stability.

[0110] Extract the transition buffer range from the state evolution predictor. This parameter represents the size of the safety margin interval required to prevent misjudgment under the current operating condition transition path.

[0111] Read the downlink release threshold reference value from the asymmetric threshold boundary pair generated in the previous step, and use it as the initial anchor point for interval widening processing.

[0112] The downlink hysteresis spread is calculated based on the transition buffer range, and the expanded threshold offset is determined using the following formula: Where ΔT is the downlink threshold spread, β is the anti-interference gain coefficient, R is the transition buffer range, and S unc The uncertainty weights for state transitions.

[0113] The calculated downlink threshold extension is algebraically subtracted from the downlink release threshold baseline to reduce the numerical level of the release threshold.

[0114] The reduced threshold value is subjected to boundary rationality verification to ensure that it is not lower than the minimum detection limit of the physical sensor and is lower than the uplink trigger threshold, thus forming a downlink dynamic release threshold with anti-disturbance stability.

[0115] By using the above-mentioned interval widening processing method, the predictive factors of the previous step are transformed into threshold boundary data with lag characteristics, so as to achieve the expected technical effect of maintaining the early warning status lock and avoiding frequent oscillations during the period of operating condition fluctuations.

[0116] S4.4: Perform vector fusion processing on the uplink dynamic trigger threshold and the downlink dynamic release threshold to construct a composite threshold structure that includes a sensitivity correction coefficient and a hysteresis constraint factor, forming an initial dynamic judgment threshold group with inertial constraint characteristics.

[0117] The system receives the uplink dynamic trigger threshold processed by gradient offset and the downlink dynamic release threshold processed by interval widening as input data objects to construct the composite threshold structure.

[0118] A vector space mapping is performed on the uplink dynamic trigger threshold and the downlink dynamic release threshold to transform the scalar threshold into a two-dimensional threshold vector containing sensitivity and hysteresis dimensions.

[0119] The slope direction adjustment is extracted from the state evolution predictor and normalized to become a sensitivity correction coefficient, which characterizes the system's response weight to the rate of parameter change.

[0120] The transition buffer range in the state evolution prediction factor is extracted and quantified as a hysteresis constraint factor, which characterizes the system's disturbance resistance margin during state transitions.

[0121] Construct a sensitivity correction matrix, and set the diagonal elements as sensitivity correction coefficients and hysteresis constraint factors respectively, to form a transformation operator for adjusting each component of the threshold vector.

[0122] By employing matrix multiplication, the two-dimensional threshold vector is multiplied on the left by the sensitivity correction matrix to achieve coordinated adjustment of the threshold boundary in two orthogonal dimensions: sensitivity and hysteresis. Specifically: Among them, Tinit vec Let T be the initial dynamic decision threshold vector, α be the sensitivity correction coefficient, β be the hysteresis constraint factor, and T be the latency constraint factor. up_dyn T is the uplink dynamic trigger threshold. down_dyn The threshold for dynamic release in the downlink direction.

[0123] The calculation results are decoupled to separate the corrected uplink threshold component and downlink threshold component, forming an initial dynamic judgment threshold group with inertial constraint characteristics.

[0124] By using vector fusion processing, the independent thresholds from the previous step are transformed into composite threshold structure data containing sensitivity correction coefficients and hysteresis constraint factors, thereby achieving the expected technical effect of adaptive matching of the early warning judgment benchmark to the inertia of the operating condition evolution.

[0125] S4.5: Perform boundary smoothing verification on the initial dynamic judgment threshold group to ensure that the uplink dynamic trigger threshold is always higher than the downlink dynamic release threshold and the difference meets the requirements of working condition transition continuity, and output the final dynamic judgment threshold group with inertial constraint characteristics for hierarchical crossing judgment.

[0126] Step S5: Based on the dynamic judgment threshold group, perform hierarchical traversal judgment on the real-time collected multi-dimensional operating parameters to generate a primary warning status signal with a time lock mark. Specifically, performing hierarchical traversal judgment on the real-time collected multi-dimensional operating parameters based on the dynamic judgment threshold group to generate a primary warning status signal with a time lock mark includes: If the multi-dimensional operating parameters collected in real time exceed the uplink trigger threshold, the corresponding warning level is activated and a soft-lock timer is started. The duration of the soft-lock timer is set according to the operating condition stability level to which the status identifier vector belongs, and a primary warning status signal with a time lock mark is generated. Specifically, this includes: S5.1: Perform numerical comparison processing on the multidimensional operating parameters and the uplink trigger threshold in the dynamic judgment threshold group to generate a hierarchical crossing judgment result representing the parameter limit-crossing state.

[0127] Receive a dynamic threshold set with inertial constraint characteristics from the output of step S4. This data structure includes the uplink trigger threshold and downlink release threshold corrected for the current operating condition. Synchronously acquire the multi-dimensional operating parameter sequence collected in real time by the power metering box at the current sampling moment, including voltage amplitude, current RMS value, box temperature, and total harmonic distortion rate.

[0128] The key monitoring indicators in the multidimensional operating parameters are decoupled dimensionally, and the core feature quantities directly related to the current warning level are extracted to form a set of real-time state scalars to be determined. According to the warning level definition, the real-time state scalars are matched one by one with the uplink trigger thresholds in the dynamic judgment threshold group.

[0129] Numerical comparison logic is used to calculate the difference between the real-time state scalar and the corresponding uplink trigger threshold, constructing a threshold crossing discrimination function. If the difference is greater than zero, it is determined to be a positive crossing event; if the difference is less than or equal to zero, it is determined to be a non-threshold crossing state. The specific formula is as follows: Where D is the threshold discrimination value, and V real T is a real-time state scalar. up This is the uplink trigger threshold.

[0130] Based on the limit-crossing judgment result, a Boolean-type hierarchical crossing judgment flag is generated to indicate whether a warning level transition has occurred at the current time. This judgment flag is then encapsulated with the corresponding warning level identifier and the crossing timestamp to form the hierarchical crossing judgment result.

[0131] Through the above numerical comparison and logical judgment processing, the dynamic threshold benchmark of the previous step is transformed into a hierarchical crossing judgment result that represents the parameter exceeding the limit state, realizing the accurate capture and status marking of instantaneous abnormal events, and providing a clear trigger basis for the subsequent activation of the soft locking mechanism.

[0132] S5.2: Based on the abnormal events that cross the uplink trigger threshold in the hierarchical traversal determination results, perform an activation operation on the current corresponding warning level and initialize the soft lock timer, generating a warning level identifier in an active state and a soft lock timer instance with a configurable duration.

[0133] Receive the hierarchical crossing determination result output by S5.1 and identify the abnormal event data records marked as exceeding the uplink trigger threshold.

[0134] Analyze the abnormal event data record to extract the warning level code corresponding to the current moment and a snapshot of the real-time multi-dimensional operating parameters that triggered the event.

[0135] Based on the extracted warning level code, an active warning level identifier object is instantiated in the system memory, and the status bit of the identifier is set to true.

[0136] Synchronously initialize a soft-lock timer instance, set the initial count value of the timer to zero, and mark its running state as pending configuration mode.

[0137] Logically associate the active warning level identifier with the soft lock timer instance with the duration to be configured to construct a primary warning status signal structure containing time lock attributes.

[0138] Through the above instantiation and association processing, the limit violation judgment result of the previous step is transformed into an active state early warning data structure with a time dimension, realizing the state transformation of the early warning response from instantaneous triggering to continuous locking, and providing an execution carrier for subsequent dynamic configuration of duration based on operating condition stability.

[0139] S5.3: Perform a mapping query process on the operating condition stability level to which the state identifier vector representing the current operating condition of the equipment belongs, and generate a quantitative value of the operating condition stability level that determines the time lock length.

[0140] The operating condition stability level is a metric used to quantitatively assess the stability or volatility of the current operating state of the electricity metering box. It is obtained as follows: based on the state identifier vector, the typical operating condition cluster to which the current operating condition belongs is determined through matching and positioning in the state transition map library; according to the dispersion of parameters such as load volatility, phase shift trend, and daily cycle similarity statistically obtained from historical data within the operating condition cluster, each typical operating condition cluster is pre-divided into different stability levels (e.g., high stability, medium stability, low stability, etc.); by querying a pre-set stability level mapping table corresponding to the operating condition cluster to which the state identifier vector belongs, the quantitative value of the operating condition stability level is generated. A higher value indicates a more stable current operating condition, and vice versa.

[0141] S5.4: Use the quantized value of the operating condition stability level to perform a duration setting operation on the soft lock timer instance to be configured, and generate a configured soft lock timer with a defined duration.

[0142] The system receives the quantitative value of the operational stability level generated in the previous steps. This value characterizes the degree of jitter or stabilization trend of the current equipment operating state and serves as the core basis for setting the duration. It retrieves a preset duration mapping configuration table, which establishes a non-linear correspondence between operational stability levels and basic lockout duration intervals, used to define the time constraint boundaries under different stability levels. An exponential decay function is used to dynamically correct the basic lockout duration, and the duration of the configured soft-lock timer is calculated using the following formula: Where T is the duration of the configured soft-lock timer, T base The preset baseline duration is given, k is the attenuation coefficient, and S is the quantized value of the operating condition stability level. Based on the calculated duration T, a parameter injection operation is performed on the initialized soft-lock timer instance, transforming the abstract time value into a specific counter threshold. The timer state after injection is verified to ensure it is in an initialized ready state, awaiting startup and not overflowing. Through the above processing method, the quantized value of the operating condition stability level from the previous step is transformed into a configured soft-lock timer with a defined duration, achieving adaptive matching between the warning lock time and the equipment operating condition fluctuation characteristics, effectively suppressing frequent oscillations in the warning level caused by instantaneous disturbances.

[0143] S5.5: Integrate the warning level identifier that is in an active state with the configured soft lock timer with a defined duration, associate and bind the two together, and generate a primary warning status signal with a time lock mark.

[0144] Step S6: If, within the duration of the soft-lock timer, the real-time collected multidimensional operating parameters are continuously within the new level's allowable range multiple times, then the time-lock flag is released and smooth transition logic is executed to generate a stable warning level instruction. The stable warning level instruction is used to eliminate the risk of state oscillation. In this embodiment, the preferred number of times the multidimensional operating parameters are continuously within the new level's allowable range is five times; however, this can be adjusted adaptively according to actual needs. Specifically, this includes: S6.1: Obtain the current active warning level identifier from the primary warning status signal with time lock mark, and retrieve the corresponding downlink release threshold boundary and new level allowable range from the preset threshold mapping table based on the current active warning level identifier, and generate a target compliance detection window containing upper and lower limit constraints.

[0145] The primary warning status signal with time lock marker is parsed, and the currently active warning level identifier encapsulated therein is extracted. This identifier clearly indicates the current abnormal level status of the device and serves as the unique index key for subsequent threshold retrieval.

[0146] Based on the current active warning level identifier, a precise matching retrieval operation is performed in the preset threshold mapping table to locate the downlink release threshold boundary parameter corresponding to the warning level. This boundary parameter defines the lower limit of the value at which the warning level falls back to a lower level or normal state.

[0147] Synchronously retrieve the upper and lower limit constraints related to the allowable range of the new level. This range is jointly defined by the downlink release threshold and the uplink trigger threshold of the lower level, and is used to determine whether the equipment operating parameters have stably returned to the safe or suboptimal operating condition domain.

[0148] The retrieved downlink release threshold boundary and the upper and lower limit constraints of the new level allowable interval are structurally encapsulated to construct a detection window object containing strict numerical boundaries, which clarifies the judgment criteria for parameter compliance.

[0149] Through the above retrieval and encapsulation process, the abstract warning level identifier is transformed into a specific and quantifiable target compliance detection window, providing a definite comparison benchmark for continuous compliance verification of real-time multi-dimensional operating parameters during the soft lockout period, and realizing boundary constraints for a smooth transition of the warning state.

[0150] S6.2: Using the target compliance detection window, the real-time multidimensional operating parameter sequence collected at a fixed frequency within the continuous duration of the soft-lock timer is traversed and compared point by point to identify whether the parameter value at each sampling moment falls within the allowable range of the new level, thereby generating a Boolean sampling judgment flag sequence representing the compliance status of a single sampling.

[0151] S6.3: Perform sliding window continuous counting statistical operation based on the Boolean sampling judgment flag sequence to calculate the number of truth flags that appear consecutively within a preset time sub-window, so as to quantitatively evaluate the continuous stability of the real-time multidimensional operating parameters in the new level allowable range, and then generate a continuous compliance count value that reflects the parameter residence time.

[0152] Receive the Boolean sampling decision flag sequence generated by S6.2. This sequence consists of 0 and 1, which respectively represent whether the single sampling parameter value falls into or falls into the new level's allowed range.

[0153] A sliding time window of length 5 is constructed, and the Boolean sampling decision flag sequence is filled into the window in chronological order to form the local state observation vector at the current moment.

[0154] Perform a summation operation on the five Boolean values ​​in the window, calculate the number of consecutive occurrences of the truth flag in the current window, and generate preliminary candidate values ​​for compliance count.

[0155] A continuity check logic is introduced, which determines that the continuity compliance condition is met only when the flag bits of the last 5 sampling points in the window are all 1; otherwise, the compliance count value is reset to 0.

[0156] If any zero value is detected within the window, the current counting accumulation process is immediately interrupted, and the sliding window is moved forward by one sampling step to extract the next set of local state observation vectors.

[0157] Repeat the above sliding and verification operation until the soft lock timer expires or a continuous compliance sequence that meets the conditions is detected, and output a continuous compliance count value that reflects the stability of parameter residence.

[0158] If, during the duration of the soft-lock timer, the real-time collected multidimensional operating parameters fail to fall within the new level's allowable range (i.e., a false value appears in the single sampling judgment flag sequence), the current operating state is considered unstable. The system will automatically reset the current continuous compliance count value to its initial zero value and interrupt the current continuous accumulation process. This mechanism ensures that state unlocking can only be triggered if the device's operating parameters remain consistently stable within the new level's allowable range, effectively avoiding false downgrades and frequent oscillations in the warning level caused by momentary interference or single data fluctuations.

[0159] By using continuous counting and statistical operations through a sliding window, discrete Boolean decision results are transformed into quantified continuous compliance count values, enabling accurate assessment of the stability of the early warning state transition and effectively avoiding mis-locking caused by instantaneous noise.

[0160] S6.4: The continuous compliance count value is logically compared with the preset five-sample threshold benchmark. If the continuous compliance count value reaches or exceeds the five-sample threshold benchmark, the state unlock enable signal is triggered to confirm that the device operating state has met the stability requirements for smooth transition, thereby generating a state transition permission instruction with the ability to rescind authorization.

[0161] Receive the continuous compliance count value generated in the preceding steps. This value represents the number of consecutive samples in which the real-time multidimensional operating parameters fall within the new level's allowable range during the soft-lock period. Set a preset five-sampling threshold as a fixed integer constant as the minimum time window length required to determine state stability.

[0162] A logical comparison operation is performed to compare the continuous compliance count with the five-sample threshold, generating a Boolean-type state unlock enable signal. If the continuous compliance count is less than the five-sample threshold, the state unlock enable signal remains false, maintaining the current warning level's soft lock state, and continuing to accumulate compliance data for subsequent sampling points.

[0163] If the continuous compliance count reaches or exceeds the five-sample threshold benchmark, the state unlock enable signal is set to true, triggering the logic for generating the state transition permission instruction. This logic confirms that the device operating parameters have been stable within the new level's allowable range for a sufficiently long time window, eliminating the risk of misjudgment caused by instantaneous noise or brief fluctuations.

[0164] Based on the condition that the state unlock enable signal is true, an object with a state transition permission instruction for deauthorization is instantiated. This instruction object contains an identifier of the current target stability level, a timestamp for unlocking the time lock, and a state smooth transition completion flag, which is used to drive subsequent state update operations.

[0165] By using logical comparison and threshold judgment, the continuous compliance count value from the previous step is transformed into a state transition permission instruction that can be deauthorized, thereby achieving the expected technical effect of stable early warning level switching to eliminate the risk of state oscillation.

[0166] For example, a five-sample threshold is set to 5. During the soft-lock timer, the system collects voltage and current data at 1-second intervals. If, in the five consecutive sampling points from the 10th to the 14th second, the voltage value is below the level 2 alarm release threshold but above the level 1 concern threshold, the continuous compliance count is accumulated to 5. At this point, the continuous compliance count of 5 equals the five-sample threshold of 5, the logical comparison result is true, and the state unlock enable signal is triggered. The system generates a state transition permission instruction, indicating that the warning level will smoothly transition from level 2 alarm to level 1 concern, and clears the soft-lock flag. If, after the first three samples are compliant, the fourth sample abnormally jumps out of the allowed range, the continuous compliance count is reset to 0, and five consecutive compliant samples must be accumulated again before unlocking can be triggered, effectively avoiding frequent changes in the warning level due to single data fluctuations.

[0167] S6.5: In response to the state transition permission instruction, perform time lock flag clearing operation and early warning level update writing process, forcibly switch the current system early warning state from primary early warning state signal to target stability level, so as to complete the oscillation-free transition from temporary locked state to steady state, and finally output a stable early warning level instruction to eliminate the risk of state oscillation.

[0168] Receive the state transition permission instruction generated in the previous steps. This instruction carries the currently active alert level identifier and target stability level information. Parse the state transition permission instruction and extract the time lock flag field to be cleared and the target alert level code to be written.

[0169] A memory address location operation is performed to retrieve the warning status storage unit corresponding to the current energy metering box device ID in the system's global status register. A logical reset operation is performed on the time lock flag bit in the warning status storage unit, forcibly setting the flag value from the locked state to zero, thereby releasing the hardware-level or software-level interlock constraint on the warning level change.

[0170] Read the target stability level code and convert it into the system's internal standard enumeration type data format. Perform an atomic write operation on the current warning level field in the warning status storage unit, overwriting and updating the original primary warning status signal with the target stability level code, ensuring the atomicity of the status switching process and data consistency.

[0171] A state transition completion confirmation signal is generated, which includes records of level changes before and after the transition and precise timestamp information. This state transition completion confirmation signal is encapsulated into a stable early warning level instruction to eliminate the risk of state oscillations, serving as the standard input source for the subsequent confidence assessment module.

[0172] By performing time-locked flag clearing and early warning level atomic update processing, the state transition permission of the previous step is transformed into a steady-state early warning instruction at the system level, realizing a smooth, oscillating transition from the temporary locked state to the target stable level, effectively eliminating the risk of frequent early warning level jumps caused by instantaneous disturbances.

[0173] Step S7: Based on the current state identifier vector, dynamic judgment threshold group, and soft-lock timer, generate a state confidence label and attach the state confidence label to the stable warning level instruction to form a complete warning data packet. Specifically, the state confidence label is calculated by fusing the fingerprint matching degree of the current state identifier vector, the traversal depth of the dynamic judgment threshold group, and the remaining duration of the soft-lock timer; the state confidence label is used to characterize the confidence level of the warning. This includes: S7.1: Use cosine similarity to calculate the vector space matching degree between the current state identifier vector and the standard fingerprint sequence in the historical operating condition template library, and output the fingerprint matching degree value that represents the degree of conformity of the device's operating mode.

[0174] The S7.1 sub-step aims to provide a fingerprint matching index for early warning confidence assessment by quantifying the similarity between the current operating conditions and historical standard patterns. This step follows the stable early warning level command generated in S6 and the state identifier vector generated in S1, serving as the basic input for subsequent multi-source confidence synthesis.

[0175] Retrieve a set of standard fingerprint sequences that match the current equipment type, rated capacity, and installation environment from the historical operating condition template library. The standard fingerprint sequence consists of load fluctuation rate, phase offset trend, and daily cycle similarity benchmark value under normal operating conditions.

[0176] The current state identifier vector is normalized using the L2 norm to eliminate the interference of dimensional differences on similarity calculation and generate a current working condition feature vector of unit length.

[0177] The retrieved standard fingerprint sequences are subjected to the same L2 norm normalization process to generate historical baseline feature vectors of unit length, ensuring that both are in the same vector space dimension.

[0178] The cosine similarity formula is used to calculate the cosine value of the angle between the current operating condition feature vector and the historical baseline feature vector. The fingerprint matching degree is then calculated using the following formula: Where Sim is the fingerprint matching score, and V is the fingerprint matching score. i T is the i-th component of the current working condition feature vector. i Let be the i-th component of the historical baseline feature vector, and n be the dimension of the feature vector.

[0179] The calculated fingerprint matching score is processed by interval mapping, which constrains it to a closed interval between 0 and 1. The closer the score is to 1, the higher the degree of consistency between the current operating mode and the historical normal operating conditions.

[0180] By using the cosine similarity vector space mapping calculation method, the state identifier vector of the previous step is transformed into a fingerprint matching degree value that represents the degree of consistency of the equipment operation mode. This enables a quantitative assessment of the consistency of the background conditions of the early warning event, providing a basis for distinguishing between real faults and environmental disturbances.

[0181] S7.2: Based on the real-time multidimensional operating parameters corresponding to the fingerprint matching degree value and the uplink trigger threshold or downlink release threshold in the dynamic judgment threshold group, perform difference normalization processing to extract the crossing depth index that reflects the severity of abnormal deviation.

[0182] Obtain the fingerprint matching score output from the previous steps and the real-time multi-dimensional operating parameter sequence, including voltage amplitude, current RMS value, chamber temperature, and total harmonic distortion rate. Retrieve the uplink trigger threshold boundary or downlink release threshold boundary corresponding to the current activated warning level from the dynamic judgment threshold group to determine the benchmark reference value used to calculate the degree of deviation.

[0183] For each component of the multidimensional operating parameters, the difference between the real-time measured value and the corresponding threshold boundary is calculated to construct an original deviation vector reflecting the instantaneous exceedance magnitude of each physical quantity. This process aims to quantify the absolute distance of the current operating state relative to the safety boundary, providing basic data support for subsequent normalization processing.

[0184] The rated range of each dimension parameter is introduced as a normalization factor to perform dimensionless processing on the original deviation vector. The crossing depth index of the i-th dimension parameter is calculated using the following formula: Among them, D i Let V be the traversal depth of the i-th parameter. real V is a real-time measured value. thresh For dynamic threshold boundaries, V max With V min These are the historical maximum and minimum allowable ranges for this parameter, respectively.

[0185] A weighted aggregation operation is performed on the crossing depth indicators of each dimension, with the weight coefficients pre-generated based on the sensitivity level of the parameter in that dimension during the fault evolution process. The comprehensive crossing depth scalar is obtained by weighted summation, which intuitively represents the severity and urgency of the current abnormal state of the equipment.

[0186] By using difference normalization and weighted aggregation, the fingerprint matching results from the previous step are transformed into a crossing depth index that reflects the severity of abnormal deviations, thereby achieving a quantitative assessment of the severity of the warning event and providing key input for subsequent confidence fusion.

[0187] S7.3: Read the current count value of the active soft lock timer and the preset total duration, use the time ratio conversion logic to calculate the remaining time ratio, and generate a lock remaining duration factor that characterizes the continuous stability of the warning state.

[0188] The system reads the active soft-lock timer instance and extracts the current cumulative running time count and the preset total lock duration parameter. Using the current cumulative running time count as the minuend and the preset total lock duration parameter as the subtrahend, it performs a subtraction operation to calculate the absolute value of the remaining lock time. The preset total lock duration parameter is used as the denominator to construct a time ratio conversion benchmark. A division operation logic is used to divide the absolute value of the remaining lock time by the preset total lock duration parameter, generating a normalized time ratio value. This time ratio value is then linearly mapped to convert it into a lock remaining duration factor characterizing the sustained stability of the warning state. Through the time ratio conversion logic, the physical time parameter of the timer is transformed into a dimensionless stability quantification index, achieving standardized input of the time dimension in the warning confidence assessment. For example, the preset total soft-lock duration for a level 2 alarm is 120 seconds. At a certain sampling moment, the soft-lock timer has run for 40 seconds. The remaining lock time is calculated as 120 minus 40, which equals 80 seconds. The time ratio is calculated as 80 divided by 120, resulting in approximately 0.67. This ratio is directly mapped to a remaining lockout duration factor of 0.67. If the equipment is in extremely unstable condition at this time, this factor will reduce the overall confidence score in subsequent confidence fusion; if the condition is stable and the lockout period is nearing its end, this factor will approach 0, indicating that the lockout is about to be lifted, and the system needs to carefully judge whether to maintain the current warning level in conjunction with other indicators.

[0189] S7.4: Using the fingerprint matching degree value, the traversal depth index, and the remaining lock duration factor as joint input variables, a weighted fusion evaluation strategy is applied to perform multi-source confidence synthesis processing to calculate the state confidence label characterizing the credibility of this early warning event.

[0190] The fingerprint matching score, traversal depth index, and remaining lock duration factor generated in the preceding steps are received as input variables for multi-source confidence synthesis.

[0191] Normalization preprocessing is performed on the three input variables to map them to the [0,1] interval, eliminating the influence of dimensional differences on the fusion result and generating a standardized feature vector.

[0192] Based on the business priority of the power metering box fault warning, weight coefficients are assigned to fingerprint matching degree, traversal depth and remaining lock duration to construct a weighted evaluation matrix.

[0193] A linear weighted summation process is used to perform a dot product operation between the standardized feature vector and the corresponding weight coefficient to calculate the initial confidence score.

[0194] A nonlinear correction function is introduced to smooth the initial confidence score, suppress the excessive interference of extreme values ​​on the final label, and generate the basic state confidence label.

[0195] Based on the hysteresis loop width of the dynamic judgment threshold group, the basic state confidence label is adaptively scaled to enhance the discrimination robustness in scenarios with drastic fluctuations in operating conditions.

[0196] By employing the aforementioned weighted fusion and nonlinear correction methods, multidimensional heterogeneous evaluation indicators are transformed into state confidence labels with unified dimensions, enabling a quantitative assessment of the reliability of early warnings and providing a precise basis for subsequent tiered push decision-making.

[0197] S7.5: Receive the previously generated stable early warning level instruction and the state confidence label, and perform a data encapsulation protocol to perform a structured assembly operation to form a complete early warning data packet with confidence assessment containing complete confidence assessment information.

[0198] Step S8: Determine whether the state confidence label in the complete early warning data packet is higher than a preset baseline. If so, push a visual prompt to the operation and maintenance terminal; otherwise, store the complete early warning data packet in the local cache and use it as historical data input for the next round of state identifier vector generation. Specifically, this includes: S8.1: Parse and process the complete early warning data packet with confidence assessment, extract the encapsulated state confidence label as the benchmark comparison object, and obtain the key confidence value for subsequent logical judgment.

[0199] S8.2: Perform a size comparison operation based on the preset baseline parameters and the extracted state confidence labels to generate a binary judgment result representing the credibility of the warning, so as to determine whether the current warning information meets the conditions for external release.

[0200] The system reads the state confidence label values ​​obtained from the previous steps and loads them into the comparison logic unit as variables to be determined. It then calls the system's preset warning release baseline parameters, which represent the minimum confidence threshold required for the maintenance terminal to receive visual prompts. A numerical comparison operation is performed to determine if the state confidence label is strictly greater than the preset baseline parameter. If the comparison result is true, a binary judgment flag of 1 representing high confidence is generated, indicating that the current warning information meets the conditions for external release. If the comparison result is false, a binary judgment flag of 0 representing low confidence is generated, indicating that the current warning information does not meet the conditions for external release. The generated binary judgment result is encapsulated into a Boolean control signal to drive the routing selection of subsequent branch logic. Through the above comparison and flag generation processes, the continuous confidence quantification value is transformed into a discrete release permission instruction, achieving precise control of the warning information distribution strategy.

[0201] S8.3: If the binarization judgment result indicates that the state confidence label is higher than the preset baseline, then the complete early warning data packet with confidence assessment is formatted and encapsulated to generate a visual prompt instruction that is compatible with the operation and maintenance terminal display protocol.

[0202] Extract the state confidence label, warning level identifier, and multi-dimensional operating parameter snapshot from the complete early warning data package with confidence assessment, and use them as the input source for generating visualization instructions.

[0203] The operation and maintenance terminal display protocol specification is analyzed, and a formatted encapsulation framework is constructed, which includes an alarm level color value mapping table, a text template library, and interactive control definitions.

[0204] Based on the warning level identifier, retrieve the alarm level color value mapping table to obtain the corresponding high-risk red, medium-risk orange, or low-risk yellow RGB color codes, and establish the visual warning tone.

[0205] Call the dynamic description template in the text template library that matches the warning level, fill the placeholder with the key out-of-limit values ​​in the multi-dimensional operating parameter snapshot, and generate the warning text with semantic information.

[0206] Read the status confidence label values, convert them into percentages, and map them to the confidence progress bar fill ratio to intuitively display the credibility of the warning results.

[0207] Calculate the remaining duration of the soft lock timer. If the remaining duration is greater than zero, render a countdown ring progress component in the visualization interface to prompt the maintenance personnel that the system is currently in a state lock protection period.

[0208] The color codes, alarm text, confidence progress bar data, and countdown component parameters are assembled into a standard JSON format visual prompt instruction object.

[0209] The generated JSON object is serialized and encoded, and a timestamp and device ID header information are added to form the final data packet that conforms to the communication protocol requirements.

[0210] By using a formatted encapsulation process, the results of the previous step are transformed into visual prompts that are compatible with the display protocol of the operation and maintenance terminal, enabling accurate delivery and intuitive presentation of high-confidence early warning information and improving operation and maintenance response efficiency.

[0211] S8.4: If the binarization judgment result indicates that the state confidence label is not higher than the preset baseline, then the complete early warning data packet with confidence assessment is written to the local cache storage area to generate a historical data input source with timestamp.

[0212] S8.5: Update the dataset of the continuous time window using the generated historical data input source, and re-trigger the calculation process of the state identifier vector to form a closed-loop data foundation to support the next round of working condition migration characteristic analysis.

[0213] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0214] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0215] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for early warning of faults in electricity metering boxes based on time-series data analysis, characterized in that, include: S1: Obtain the multi-dimensional operating parameters of the power metering box, calculate the state feature parameters based on the multi-dimensional operating parameters, and generate a state identification vector; S2: Based on the analysis results of the changes of the state identifier vector in multiple consecutive time windows, construct a double-loop hysteresis threshold structure and generate asymmetric threshold boundary pairs; S3: Generate a state transition graph library using the working condition transition sequence in the historical operation and maintenance records, and predict the state evolution direction based on the matching path of the state identifier vector in the state transition graph library, and generate a state evolution prediction factor that includes at least the slope direction adjustment amount. S4: Integrate the asymmetric threshold boundary pair with the state evolution prediction factor, and adjust the trigger sensitivity of the asymmetric threshold boundary pair according to the slope direction adjustment amount to generate a dynamic judgment threshold group; S5: Based on the dynamic judgment threshold group, perform hierarchical traversal judgment on the multidimensional operating parameters collected in real time, generate a primary warning status signal with a time lock mark, and start a soft lock timer. S6: If the multi-dimensional operating parameters collected in real time are continuously within the new level's allowable range multiple times during the duration of the soft-lock timer, then the time lock flag is released and the smooth transition logic is executed to generate a stable early warning level instruction. S7: Based on the current state identifier vector, dynamic judgment threshold group and soft lock timer, generate a state confidence label and attach the state confidence label to the stable warning level instruction to form a complete warning data packet.

2. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 1, characterized in that, After step S7, the following is included: S8: Determine whether the state confidence label in the complete early warning data packet is higher than the preset baseline. If so, push a visual prompt to the operation and maintenance terminal. Otherwise, store the complete early warning data packet in the local cache and use it as historical data input for the next round of state identifier vector generation.

3. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 1, characterized in that, The state evolution predictor also includes: a transition buffer range.

4. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 3, characterized in that, Step S3 specifically includes: Extract and standardize the actual operating condition transition sequences from historical operation and maintenance records to construct a state transition graph library containing initial state nodes, target state nodes, and transition probability weights. Based on the state identifier vector, a topology matching search is performed in the state transition graph library to identify the starting state node that best matches the current equipment operating condition and its associated set of high-frequency transfer paths, thereby generating a state matching path set. The slope of the path evolution trend is calculated using the transition probability weights in the state matching path set, the rate and direction of change of the equipment operating parameters in the future time window are quantified, and a state evolution trend descriptor containing the slope and direction adjustment amount is generated. The uncertainty range of state transitions is assessed based on the rate of change in the state evolution trend descriptor, the size of the buffer region required to prevent misjudgment is determined, and the transition buffer range defining the safe transition interval is generated. The slope direction adjustment amount and the transition buffer range are combined and encapsulated to form a comprehensive control command that can directly drive the dynamic correction of the threshold, thereby generating the state evolution prediction factor.

5. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 4, characterized in that, Constructing a dual-loop hysteresis threshold structure and generating asymmetric threshold boundary pairs includes: defining the difference between the uplink trigger threshold and the downlink release threshold in the dual-loop hysteresis threshold structure as the dynamic hysteresis loop width, and generating the asymmetric threshold boundary pairs based on the dynamic hysteresis loop width.

6. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 5, characterized in that, Step S4 specifically includes: Obtain the asymmetric threshold boundary pair and the state evolution prediction factor, and extract the slope direction adjustment amount and transition buffer range from the state evolution prediction factor as the basic input data for correcting the asymmetric threshold boundary pair; Based on the slope direction adjustment, gradient offset processing is performed on the uplink trigger threshold in the asymmetric threshold boundary pair to generate an uplink dynamic trigger threshold. The downlink release threshold in the asymmetric threshold boundary pair is widened using the transition buffer range to generate a downlink dynamic release threshold. The uplink dynamic trigger threshold and the downlink dynamic release threshold are vector-fused to construct a composite threshold structure and form an initial dynamic judgment threshold group. Perform boundary smoothing verification on the initial dynamic judgment threshold group and output the dynamic judgment threshold group.

7. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 5, characterized in that, Based on the dynamic judgment threshold group, hierarchical traversal judgment is performed on the multidimensional operating parameters collected in real time, generating a primary warning status signal with a time lock mark and starting a soft lock timer, including: If the multi-dimensional operating parameters collected in real time exceed the uplink trigger threshold, the corresponding warning level is activated and a soft lock timer is started. The duration of the soft lock timer is set according to the operating condition stability level to which the status identifier vector belongs, and the primary warning status signal with a time lock mark is generated. The operating condition stability level is obtained by querying a preset stability level mapping table according to the status identifier vector.

8. The method for early warning of faults in electricity metering boxes based on time-series data analysis according to claim 7, characterized in that, Step S5 specifically includes: The multidimensional operating parameters are numerically compared with the uplink trigger threshold in the dynamic determination threshold group to generate a hierarchical traversal determination result. Based on the abnormal events that cross the uplink trigger threshold in the hierarchical crossing determination results, an activation operation is performed on the current corresponding warning level and the soft lock timer is initialized, generating a warning level identifier in an active state and a soft lock timer instance with a configurable duration. A mapping query process is performed on the operating condition stability level to which the state identifier vector representing the current operating condition of the equipment belongs, generating a quantitative value of the operating condition stability level that determines the time lock length. The duration setting operation is performed on the soft lock timer instance to be configured using the quantized value of the operating condition stability level, thereby generating a configured soft lock timer with a definite duration. The warning level identifier that is in an active state is combined with the configured soft lock timer with a defined duration, and the two are associated and bound to generate the primary warning state signal with a time lock mark.

9. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 1, characterized in that, Based on the current state identifier vector, dynamic judgment threshold group, and soft-lock timer, a state confidence label is generated, including: The state confidence label is calculated and generated by integrating the fingerprint matching degree of the current state identifier vector, the traversal depth of the dynamic judgment threshold group, and the remaining duration of the soft lock timer.

10. The method for early warning of faults in power metering boxes based on time-series data analysis according to claim 1, characterized in that, The multidimensional operating parameters include: voltage, current, temperature, and harmonic content; the state characteristic parameters include: load volatility, phase shift trend, and daily cycle similarity.

Images