Charging pile metering abnormity identification system based on deep learning

Abstract

The invention discloses a charging pile metering anomaly identification system based on deep learning, and particularly relates to the field of charging pile metering anomaly identification. A power change rate sequence is constructed by collecting voltage, current and power signals, energy gradient stratified analysis and feature quantitative modeling are carried out, and an energy stratified weight coefficient is formed; and in combination with continuous period energy residual calculation and direction consistency judgment, generating a residual direction offset rate, and dynamically correcting an abnormal mark based on an offset difference to realize accurate identification and self-correction management of the charging pile metering abnormality. According to the method, a self-learning quantitative identification mechanism is constructed through multi-layer energy feature extraction and residual chain correlation judgment, the influence of power offset, voltage fluctuation and phase difference on metering abnormity can be reflected in real time, misjudgment and delay are reduced, the abnormity detection accuracy and stability are improved, dynamic self-correction and mark correction in the charging metering process are achieved, and the charging metering efficiency is improved. And reliable metering data and fair charging are ensured.

Description

Technical Field

[0001] This invention relates to the field of charging pile metering anomaly identification, and in particular to a charging pile metering anomaly identification system based on deep learning. Background Technology

[0002] With the increasing popularity of electric vehicles, charging piles, as an important component of charging infrastructure, play a crucial role in ensuring accurate electricity metering, avoiding metering errors, and enabling data monitoring and management. The field of charging pile metering technology also involves improving metering accuracy during the charging process, preventing metering anomalies, ensuring fair and equitable electricity pricing, and ultimately supporting the management and optimization of smart grids.

[0003] Among them, the deep learning-based charging pile metering anomaly identification system refers to the technology of identifying anomalies in charging pile metering data by applying deep learning algorithms. It proposes a data-driven solution to address potential metering anomalies that may occur during the actual use of charging piles. The system mainly analyzes the electricity data during charging pile operation, uses deep learning technology to learn data patterns, and automatically identifies anomalies. Specifically, the system collects and analyzes historical metering data, uses deep learning model training to establish classification criteria for normal and abnormal data, thereby achieving anomaly detection and early warning for the charging pile metering system.

[0004] Existing technologies rely on centralized learning and static classification of historical metering data on the operational side. In the online phase, they often employ a batch-entry-follow-judgment processing rhythm, leading to time misalignment between data collection and identification. Short-term abrupt changes are smoothed out after aggregation, and directional continuity and segmental structural information are weakened throughout the learning process. The coupling relationship between voltage fluctuations and current phase is difficult to stably represent under a single classification threshold. When on-site operating conditions change frequently, the classification boundary obtained through training is not entirely consistent with the energy path of new sessions. The identification results are not sensitive enough to slow drifts caused by station load switching, vehicle-end power limitations, and cable temperature rise. Labels rely on manual spot checks and periodic backtracking for a long time, resulting in labeling lag and inconsistencies. Anomalies within the billing cycle are often only discovered after settlement. Taking a suburban mixed substation as an example, during the evening peak load fluctuation, multiple phase misalignments and voltage drops occur within minutes. The classification process trained using historical aggregation provides a stable judgment within an hourly window, diluting short-term deviations in peak periods. This leads to cross-day complaints for the same vehicle session, requiring the station to conduct a second review and export the original waveform, consuming maintenance resources and extending processing time. Summary of the Invention

[0005] The main objective of this invention is to provide a deep learning-based charging pile metering anomaly identification system. By hierarchically acquiring, gradient analyzing, feature quantization modeling, and residual self-correction linkage calculation of charging pile voltage, current, and power signals, the system solves the problems of lagging anomaly detection, high misjudgment rate, and inability to dynamically correct markings during the charging metering process.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A deep learning-based charging pile metering anomaly identification system, the system comprising: The signal acquisition module acquires the current, voltage, and instantaneous power values ​​of the charging pile, arranges them in a time series according to the sampling time interval, multiplies the voltage and current values ​​to obtain the power time series, calculates the power change rate, and generates the energy path fluctuation rate value. The energy gradient analysis module, based on the energy path fluctuation rate value, selects adjacent sampling points to calculate the power change rate difference, counts the frequency of change direction in the gradient sequence and calculates the positive and negative change ratio, divides energy node segments according to the power change rate threshold, calculates the ratio of voltage change amplitude to current change rate in each segment, and generates gradient stratification concentration value. The feature quantization modeling module extracts the power change rate, voltage offset amplitude and current phase difference based on the gradient stratification concentration value, performs normalization operation and calculates the mean square error ratio to generate energy stratification weight coefficients. The residual chain determination module calls the energy stratification weight coefficient, calculates the energy residual sequence between power output and input, calculates the residual difference within a continuous period, performs a consistency judgment of the difference direction, and generates the residual direction offset rate value. The self-calibration management module for labeling calculates the ratio of sample power offset to voltage fluctuation amplitude for the residual direction offset rate value, compares it with the difference of residual direction offset rate, updates the labeling parameters, and generates anomaly labeling correction rate value.

[0007] Preferably, the energy path volatility value is specifically the current change rate, voltage change rate, and power change rate; the gradient layer concentration value is specifically the ratio of signal gradient directionality, change frequency, voltage change amplitude, and current change rate; the energy layer weight coefficient is specifically the power change weight, voltage offset weight, and current phase weight; the residual direction offset rate value is specifically the residual difference direction, residual change rate, and periodic residual difference; and the anomaly mark correction rate value is specifically the mark adjustment amplitude, power offset difference, and voltage fluctuation ratio.

[0008] Preferably, the signal acquisition module includes: The current acquisition submodule is used to acquire the output value of the current sensor in the charging pile, record the current sampling sequence at each time according to the sampling time interval, perform differential calculation on the current value of continuous sampling points to determine the current change rate per unit time, and generate the current change rate value. The voltage acquisition submodule is used to establish a voltage sampling sequence based on the voltage sampling terminal's acquired values, at the same time interval as the current sampling, perform differential calculations on the voltage values ​​of adjacent sampling points to obtain the voltage change rate per unit time, and generate the voltage change rate value. The power calculation submodule calls the current change rate value from the current acquisition submodule and the voltage change rate value from the voltage acquisition submodule. It performs a product operation on the current value and voltage value at each sampling time to obtain the power time series, performs differential calculation on the power series to determine the power change rate per unit time, and generates the power change rate value. The volatility generation submodule is used to combine the power change rate, current change rate, and voltage change rate values ​​according to the time series to establish the energy path volatility series, and perform average and variance calculations on the series to generate the energy path volatility value.

[0009] Preferably, the energy gradient analysis module includes: The interval construction submodule selects continuous sampling points to form a signal interval based on the energy path volatility value, performs difference calculation on the power change rate of adjacent sampling points to obtain the gradient sequence, and records the gradient sign sequence and amplitude sequence according to the sampling time interval to establish a dataset and obtain the gradient difference sequence. The direction statistics submodule counts the number of consecutive segments with the same symbol and accumulates the length based on the gradient difference sequence. It calculates the ratio of the number of positive symbols to the number of negative symbols, records the first and last indices and amplitude sum of each segment, and generates a direction continuity statistics table based on the power change rate threshold to obtain the direction continuity rate value. The segment quantization submodule calls the voltage change rate and current change rate in the direction continuity rate value and energy path volatility value, calculates the ratio of voltage change amplitude to current change rate within the same index segment to obtain the energy transfer ratio, and then averages it with the gradient difference sequence within the segment to form a numerical table, and establishes the segment quantization matrix. The concentration output submodule, for the segmented quantization matrix, calculates the variance of the numerical table based on the number of samples in each segment and merges them into the same time order. It combines the directional continuity rate value and the segment index to form a hierarchical mapping table, outputs the segment change concentration sequence according to the mapping table, and generates gradient hierarchical concentration values.

[0010] Preferably, the feature quantization modeling module includes: The extraction submodule extracts three parameters—power change rate, voltage offset amplitude, and current phase difference—from the energy segment based on the gradient stratification concentration value. The three parameters are then scaled and normalized according to the minimum and maximum values ​​based on the segment index. They are arranged into column vectors based on the same timestamp. Missing sampling points are filled with the mean of adjacent time points while maintaining the consistent segment order. The data are then concatenated row by row to form a sequence set, generating a normalized vector value for the three parameters. The ratio calculation submodule calls the three-parameter normalized vector value, calculates the difference between each pair of the three column vectors in each energy segment, squares them, and then takes the mean according to the number of samples to obtain three sets of root mean square errors. Then, it calculates the ratio of the three sets of root mean square errors by row and performs proportional normalization. Based on the proportional difference, it forms a three-dimensional array on the segment index to obtain the feature weight ratio value. The coefficient output submodule, based on the feature weight ratio value, performs cumulative summation on the corresponding ratio sequence for all energy segments in chronological order, separates the power ratio sequence, voltage ratio sequence and phase ratio sequence into three coefficient trajectories, and generates energy stratification weight coefficients by taking the segment cumulative results from the end of the trajectory and forming a column vector.

[0011] Preferably, the residual chain determination module includes: The energy difference submodule calls the energy layering weight coefficient, obtains the cumulative output and input of the power detection terminal within a continuous charging cycle, performs difference calculation on the output and input of the same cycle to form an energy residual sequence, records the residual amplitude and direction mark according to the cycle index, and generates the energy residual sequence value. The direction determination submodule, based on the energy residual sequence value, performs differential calculation on the residuals of adjacent periods to obtain the residual difference sequence, counts the positive and negative directions of the difference and calculates the number of consecutive directions, performs proportional judgment on the length of the consecutive direction interval, extracts the interval with strong direction consistency as the offset segment, and establishes the residual direction consistency rate value. The node calculation submodule calculates the rate of change of residual amplitude within the direction consistency interval based on the residual direction consistency rate value, performs a product operation on the rate and energy stratification weight coefficient to form a node vector, takes the average value of the node vector according to the time series and records the trend of change, and generates residual chain node values. The offset output submodule calculates the average rate of change of adjacent nodes in a continuous time series for the residual chain node values, determines the offset direction by combining the direction consistency sequence, and outputs the ratio of the difference magnitude to the rate of change during the cycle to generate the residual direction offset rate value.

[0012] Preferably, the marker self-correction management module includes: The sample matching submodule selects samples from the anomaly marker set for the residual direction offset rate value, performs a ratio calculation on the sample power offset value and voltage fluctuation amplitude and pairs them according to the time index, compares the difference between the ratio and the residual direction offset rate value according to the same index and records the absolute difference, summarizes the difference vector by segment, and generates the offset difference value. The difference determination submodule judges the difference vector and energy offset threshold of each segment according to the offset difference value, records the index sequence that exceeds the threshold and calculates the continuous length ratio, extracts the label level on the corresponding index according to the ratio and gives the adjustment step according to the number of exceeding the threshold, and obtains the label adjustment amplitude value. The correction output submodule calls the flag adjustment amplitude value, performs addition and subtraction operations on the original flag parameter set according to the index in the abnormal flag set, and records the change ratio of power offset before and after adjustment. The correction ratio sequence is obtained by accumulating the ratio by the segment and dividing it by the number of samples. The abnormal flag correction rate value is generated by taking the summation ratio at the end of the sequence.

[0013] Preferably, the charging pile metering anomaly identification system further includes an anomaly identification and marking management module, which, for the identified anomaly marking, first obtains the power offset value and voltage fluctuation amplitude of the anomaly sample, compares them with the difference of the corresponding residual direction offset rate value, adjusts the marking level based on the degree of deviation, performs a marking state reset operation, and generates marking correction results.

[0014] Preferably, the mark correction result specifically includes the mark adjustment range, power offset difference, and voltage fluctuation ratio.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a power change rate by performing time-series product and difference of voltage, current and power within the same sampling link, and introduces the gradient difference and direction sign of adjacent sampling points to form continuous statistics and positive and negative proportions. Then, based on the power change rate threshold, energy node segments are divided, and the ratio of voltage change amplitude to current change rate is calculated at the segment level to obtain a quantitative characterization for the transmission channel. Subsequently, the power offset, voltage offset and phase difference are normalized and measured by the mean square error ratio and accumulated as hierarchical weights. The input and output of continuous charging cycles are subtracted to obtain the residual sequence. The direction consistency and node mean change are superimposed to give the residual direction offset rate. Finally, the ratio of power offset to voltage fluctuation is indexed and paired with the offset rate, and the difference is adaptively labeled and adjusted to form a label correction closed loop. This processing path incorporates both instantaneous changes and segment structure, including signal logic based on directional continuity, channel logic based on energy transfer ratio, and settlement logic based on periodic residuals. These three types of dimensions are cross-validated within the same index system. Anomaly representation evolves from a single label to a multi-dimensional numerical chain. The label is adjusted synchronously with the intensity and directionality of the difference. False alarms and false misses are constrained by timing consistency and channel ratio. It has stronger discrimination and convergence capabilities in the face of power step, phase jitter, short-term voltage drop, and other operating conditions. Label maintenance no longer relies on offline batch processing but is directly driven by the difference quantity to reset online. The location of metering anomalies is closer to the generation stage, and deviations within the billing period can be reflected and quantified and corrected during the cycle. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the working module of a deep learning-based charging pile metering anomaly identification system according to the present invention. Figure 2 This is a schematic diagram of the working process of the signal acquisition module of the present invention; Figure 3 This is a schematic diagram of the workflow of the energy gradient analysis module of the present invention; Figure 4 This is a schematic diagram of the workflow of the feature quantization modeling module of the present invention; Figure 5 This is a schematic diagram of the workflow of the residual chain determination module of the present invention; Figure 6 This is a schematic diagram of the workflow of the self-correction management module for marking according to the present invention. Detailed Implementation

[0017] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the linguistic context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0018] It should be understood that the terms "system," "device," "unit," and / or "module" as used in this specification are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0019] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0021] The following is a detailed description of the deep learning-based charging pile metering anomaly identification system provided in the embodiments of this specification, with reference to the accompanying drawings.

[0022] Figure 1 This is an exemplary flowchart of a deep learning-based charging pile metering anomaly identification system according to some embodiments of this specification. In some embodiments, the deep learning-based charging pile metering anomaly identification system can be executed by processing logic, which may include hardware (e.g., circuits, dedicated logic, programmable logic, microcode, etc.), software (instructions running on a processing device to execute hardware simulations), and any combination thereof. In some embodiments, Figure 1 One or more operations in the flowchart of the deep learning-based charging pile metering anomaly identification system shown can be implemented by a processing device and / or a terminal device. For example, the deep learning-based charging pile metering anomaly identification system can be stored in a storage device in the form of computer programs and / or instructions, and invoked and / or executed by the processing device and / or the terminal device.

[0023] like Figure 1 As shown, the present invention discloses a deep learning-based charging pile metering anomaly identification system, including a signal acquisition module, an energy gradient analysis module, a feature quantization modeling module, a residual chain determination module, a label self-calibration management module, and an anomaly identification and label management module. The specific modules are as follows: The signal acquisition module acquires the current, voltage, and instantaneous power values ​​of the charging pile, arranges them in a time series according to the sampling time interval, multiplies the voltage and current values ​​to obtain the power time series, calculates the power change rate, and generates the energy path fluctuation rate value; the energy path fluctuation rate value specifically includes the current change rate, voltage change rate, and power change rate. The energy gradient analysis module, based on the energy path volatility value, selects adjacent sampling points to calculate the power change rate difference, counts the frequency of change direction in the gradient sequence and calculates the ratio of positive to negative changes, divides energy node segments according to the power change rate threshold, calculates the ratio of voltage change amplitude to current change rate in each segment, and generates a gradient stratification concentration value; the gradient stratification concentration value specifically includes the signal gradient directionality, change frequency, voltage change amplitude, and current change rate ratio. The feature quantization modeling module extracts the power change rate, voltage offset amplitude, and current phase difference based on the gradient stratification concentration value, performs normalization operation and calculates the mean square error ratio to generate energy stratification weight coefficients; the energy stratification weight coefficients are specifically power change weight, voltage offset weight, and current phase weight. The residual chain determination module calls the energy stratification weighting coefficient to calculate the energy residual sequence between the power output and the input, calculates the residual difference within consecutive periods, and performs a consistency judgment on the direction of the difference to generate a residual direction offset rate value. Specifically, the residual direction offset rate value includes the residual difference direction, the residual change rate, and the periodic residual difference. The self-calibration management module for labeling calculates the ratio of sample power offset to voltage fluctuation amplitude for the residual direction offset rate value, compares it with the difference of residual direction offset rate, updates labeling parameters, and generates anomaly labeling correction rate value; the anomaly labeling correction rate value specifically includes labeling adjustment amplitude, power offset difference, and voltage fluctuation ratio; The charging pile metering anomaly identification system also includes an anomaly identification and marking management module. For the identified anomaly markings, it first obtains the power offset value and voltage fluctuation amplitude of the anomaly sample, compares them with the difference of the corresponding residual direction offset rate value, adjusts the marking level based on the degree of deviation, performs the marking state reset operation, and generates marking correction results. The marking correction results are specifically the marking adjustment amplitude, power offset difference, and voltage fluctuation ratio.

[0024] like Figure 2 As shown, in some embodiments of the present invention, the signal acquisition module includes: The current acquisition submodule is used to acquire the output value of the current sensor in the charging pile, record the current sampling sequence at each time according to the sampling time interval, perform differential calculation on the current value of continuous sampling points to determine the current change rate per unit time, and generate the current change rate value. For example, firstly, the output value of the current sensor in the charging pile is obtained. This output is the current signal sampling result of the current sensor per unit time. For example, if the sampling frequency is 1kHz and the sampling time interval is 1ms, then 1000 sampling points can be obtained within 1 second. Each sampling point is denoted as It, where t is the sampling time index value. Then, according to the sampling time interval, the sampled current at each time is arranged in chronological order to form a current sampling sequence I1, I2, ..., In. On this sequence, continuous sampling point differential calculation is performed, that is, for any two adjacent points, the formula ΔIt=It+1−It / Δt is used, where Δt is the sampling interval (0.001s here). Thus, the current change rate sequence per unit time is calculated. In actual implementation, if I1=15.2A, I2=15.8A, then the current change rate ΔI1=15.8−15.2 / 0.001=600A / s. Similarly, the change rate values ​​of the remaining sampling points are calculated to form a continuous current change rate sequence. Subsequently, a moving average is calculated for the entire sequence to eliminate random noise fluctuations. For example, a 5-point averaging method is used to average five consecutive rates of change to form a smooth result. Then, the difference between the maximum and minimum values ​​is calculated. If the difference exceeds a set threshold (the current change threshold is set to 1000A / s, which is based on the upper limit of the charging pile design current of 200A and the sampling rate parameter), it is marked as a current mutation point. Finally, the average rate of change and its variance are calculated based on the smoothed rate of change sequence to generate the current change rate value. The voltage acquisition submodule is used to establish a voltage sampling sequence based on the voltage sampling terminal's acquired values, at the same time interval as the current sampling, perform differential calculations on the voltage values ​​of adjacent sampling points to obtain the voltage change rate per unit time, and generate the voltage change rate value. For example, based on the voltage sampling terminal, which also collects voltage signals at a frequency of 1kHz to form a time series U1, U2, ..., Un, a voltage sampling sequence is established according to the same time interval as the current sampling. For adjacent sampling points, differential calculation is performed to obtain the voltage change rate per unit time, using the formula ΔUt=Ut+1−Ut / Δt, where Δt=0.001s. If the voltage at the first point is 380.0V and the voltage at the second point is 380.4V, then the voltage change rate ΔU1=380.4−380.0 / 0.001=400V / s. After performing this operation on the entire sequence, a voltage change rate sequence is formed. Then, to reduce instantaneous spike interference, the moving median method is used to smooth the change rate of every 10 sampling points, thereby eliminating local anomalies caused by occasional electromagnetic interference. The average voltage change rate and its variance of each sampling segment are then calculated. If the variance in the data exceeds the set voltage change threshold (set to 800V / s, corresponding to a 5ms fluctuation range of 3.8V for an AC 380V system), the interval is recorded as an abnormal voltage fluctuation segment. Finally, the average change rate of all sampling segments is integrated to generate the voltage change rate value. The power calculation submodule calls the current change rate value from the current acquisition submodule and the voltage change rate value from the voltage acquisition submodule. It performs a product operation on the current value and voltage value at each sampling time to obtain the power time series, performs differential calculation on the power series to determine the power change rate per unit time, and generates the power change rate value. During implementation, this submodule calls upon the current change rate value obtained from the current acquisition submodule and the voltage change rate value from the voltage acquisition submodule. At each sampling moment, it performs a product operation on the current value It and the voltage value Ut to obtain the instantaneous power Pt = Ut × It, forming a power time series. For example, when Ut = 380.2V and It = 15.5A at a certain moment, the power Pt = 380.2 × 15.5 = 5893.1W. Subsequently, it performs differential calculation on the power time series to determine the power change rate per unit time, using the formula ΔPt = Pt + 1 − Pt / Δt. When Δt = 0.001s and the power difference between adjacent units is 12.4W, the power change rate is 12400W / s. The obtained sequence is smoothed to eliminate sampling errors and the average rate of change is taken. Then, the power change rate is linearly corrected according to the ratio between the voltage change rate and the current change rate. If the power change rate exceeds the set power change threshold (10kW / s, which is a 5% change reference under the rated power of 20kW), the time point is recorded as the power anomaly point. Finally, the average value and standard deviation of the power change rate sequence are generated and the power change rate value is output. The volatility generation submodule is used to combine the power change rate, current change rate, and voltage change rate values ​​according to the time series to establish the energy path volatility series, perform average and variance calculations on the series, and generate the energy path volatility value. Furthermore, the results from the aforementioned three sub-modules—power change rate, current change rate, and voltage change rate—are combined in chronological order of sampling time to form a three-dimensional energy sequence matrix. Each row corresponds to a data item ΔIt, ΔUt, or ΔPt at a given time point. Then, a row-by-row averaging calculation is performed on the matrix to obtain the comprehensive fluctuation index Et = ΔIt + ΔUt + ΔPt / 3. Finally, the mean μ and variance σ² of all Et values ​​are calculated, where μ represents the average fluctuation intensity and σ² represents the fluctuation dispersion. In practice, if 1000 consecutive points are calculated, μ = 3200 and σ² = 4.2 × 10⁻⁶. 6 This value serves as a reference for energy stability during the current period. Finally, μ and σ² are combined proportionally to obtain a comprehensive volatility index η, using the formula η=μ / (1+σ² / μ). This index is used to quantify the stability of the energy path during the charging process, ultimately generating the energy path volatility value.

[0025] like Figure 3 As shown, in another embodiment of the present invention, the energy gradient analysis module includes: The interval construction submodule selects continuous sampling points to form a signal interval based on the energy path volatility value, performs difference calculation on the power change rate of adjacent sampling points to obtain the gradient sequence, and records the gradient sign sequence and amplitude sequence according to the sampling time interval to establish a dataset and obtain the gradient difference sequence. In the specific implementation process, firstly, based on the energy path volatility value, continuous sampling points are selected to form a signal interval. The selection criteria for this signal interval are sampling points with large volatility values ​​and obvious trends within a certain time window. For example, if the energy path volatility value changes drastically within 1 second, this 1 second is selected as the signal interval, and the start and end times of the interval are determined. Then, the power change rate of this interval is calculated. Within this signal interval, the difference in power change rates between adjacent sampling points is calculated to obtain a gradient sequence, i.e., the difference in power change rates between every two points, expressed by the formula ΔPt=Pt+1−Pt, where Pt is the power value at a certain moment, t is the sampling time point, and the difference represents the power change per unit time. For example, assuming that the power values ​​between two adjacent points are 12W and 15W respectively, the difference is ΔPt=15−12=3W, which represents the power change per unit time. Then, the sign sequence and amplitude sequence of this gradient are recorded according to the sampling time interval. The sign sequence is used to indicate the direction of power change, and the amplitude sequence is used to indicate the intensity of power change. Finally, a gradient difference sequence is constructed, which facilitates subsequent analysis and quantification of signal fluctuations, generating the gradient difference sequence.

[0026] The direction statistics submodule counts the number of consecutive segments with the same symbol and accumulates the length based on the gradient difference sequence. It calculates the ratio of the number of positive symbols to the number of negative symbols, records the first and last indices and amplitude sum of each segment, and generates a direction continuity statistics table based on the power change rate threshold to obtain the direction continuity rate value. For example, for a gradient difference sequence, firstly, the segments with consecutive identical signs are counted and their lengths are accumulated. For instance, assuming the gradient difference sequence is 3,2,1,−1,−3,−2, the positive and negative sign segments can be recorded as positive segment 3,2,1 and negative segment −1,−3,−2, respectively, and the length of each segment is counted, such as a positive segment length of 3 and a negative segment length of 3. Next, the frequency and length ratio of the positive and negative sign segments are calculated. Assuming a positive sign segment appears twice with a total length of 6, and a negative sign segment also appears twice with a total length of 6, the positive-to-negative sign ratio is 1:1. Then, the start and end indices and amplitude sum of each segment are recorded. Assuming the amplitude sum of a positive segment is 6W and the amplitude sum of a negative segment is also 6W, this data can help determine the stability and abruptness of signal changes. Based on the power change rate threshold (for example, setting the threshold to 2W, indicating that fluctuations less than this value are within an acceptable range), all paragraphs with gradient values ​​greater than this threshold are counted as valid paragraphs, and finally a directional continuity statistics table is formed, recording the number and length of positive and negative signs, and generating directional continuity rate values. The segment quantization submodule calls the voltage change rate and current change rate in the direction continuity rate value and energy path volatility value, calculates the ratio of voltage change amplitude to current change rate within the same index segment to obtain the energy transfer ratio, and then averages it with the gradient difference sequence within the segment to form a numerical table, and establishes the segment quantization matrix. The voltage and current change rates from the directional continuity rate and energy path volatility values ​​are used for further analysis within the same index segment. The energy transfer ratio is calculated by measuring the ratio of voltage change amplitude to current change rate. For example, assuming a voltage change amplitude of 2V and a current change rate of 1A / s within a signal segment, the energy transfer ratio for that segment is 2V / 1A / s = 2V·s / A. This indicates the degree to which voltage and current changes affect energy transfer within that segment. Then, the gradient difference sequence for that segment is averaged. Assuming the gradient difference sequence is 3, 4, 5, 6, the average value is 3 + 4 + 5 + 6 = 4.5, which is used as the average gradient value for that segment. This value is combined with the energy transfer ratio to generate a piecewise quantization matrix, which is ultimately used to describe the comprehensive characteristics of power changes within each segment. Example of a piecewise quantization matrix: Assume that three segments are obtained by dividing the data based on the directional continuity rate and index: S1=(t100,t199), S2=(t200,t299), and S3=(t300,t399). Within each segment, first calculate the voltage change amplitude (the difference between the maximum and minimum voltage values ​​within the segment) and the average current change rate (the average absolute value of the current difference within the segment divided by the sampling interval). The energy transfer ratio R = voltage change amplitude / average current change rate. Then, take the average G of the power change rate difference sequence (gradient difference) for that segment. Based on this, the "piecewise quantization matrix" can be organized by columns as follows: The columns are, in order: segment identifier, start index, end index, number of points, energy transfer ratio R, and gradient mean G; The concentration output submodule, for the segmented quantization matrix, calculates the variance of the numerical table within each segment based on the number of samples in that segment and merges them into the same time order. It then combines the directional continuity rate value with the segment index to form a hierarchical mapping table, outputting the segment variation concentration sequence according to the mapping table, and generating gradient hierarchical concentration values. For the piecewise quantization matrix, the variance of the data table is first calculated within each segment. This variance reflects the dispersion of power changes within that segment. Assuming the data table for a segment is 3, 4, 5, 6, its variance is 3−4.52+4−4.52+5−4.52+6−4.524=1.25, representing the fluctuation intensity within that segment. Then, the variances of all segments are merged in chronological order to form a more accurate energy fluctuation trend. Combining the directional continuity rate value and segment index information, a hierarchical mapping table is formed. This table records the variation characteristics of each segment. Based on the mapping table, a sequence of segment variation concentration is output, representing the degree of energy variation concentration within each segment. Finally, a gradient hierarchical concentration value is generated, which can help further analyze the stability and abnormal changes in system operation.

[0027] Example of hierarchical mapping: Within each segment, calculate the variance σ² of the gradient difference sequence corresponding to the "segmented quantization matrix" (using the overall variance of the sequence within the segment), and then combine it with the directional continuity rate C (given by the directional statistics submodule, the value is the proportion of consecutive samples with the same sign to the total number of samples in the segment). Set a judgment threshold to perform hierarchical mapping on the segments. Example thresholds: continuity rate threshold C_thr = 0.60, variance threshold σ_thr² = 1.50. Mapping rules: C ≥ 0.60 and σ² ≤ 1.50 is recorded as level 1, C ≥ 0.60 and σ² > 1.50 is recorded as level 2, C < 0.60 and σ² ≤ 1.50 is recorded as level 3, and C < 0.60 and σ² > 1.50 is recorded as level 4. The variance of each segment can be directly calculated from the sample sequences: S1: Gradient sequence segments 12, 8, 10, 9, mean 9.75, σ1²=((12−9.75)²+(8−9.75)²+(10−9.75)²+(9−9.75)²) / 4=2.1875; S2: Gradient sequence segments 5, 4, 3, 5, mean 4.25, σ2²=0.6875; S3: Gradient sequence segments 9, 8, 12, 11, mean 10.00, σ3²=2.50; The directional continuity rate is set as follows: C1=0.72, C2=0.55, C3=0.81. The stratification results are shown in the table below: like Figure 4 As shown, in another embodiment of the present invention, the feature quantization modeling module includes: The extraction submodule extracts three parameters—power change rate, voltage offset amplitude, and current phase difference—from the energy segment based on the gradient stratification concentration value. The three parameters are then scaled and normalized according to the minimum and maximum values ​​based on the segment index. They are arranged into column vectors based on the same timestamp. Missing sampling points are filled with the mean of adjacent time points while maintaining the consistent segment order. The data are then concatenated row by row to form a sequence set, generating a normalized vector value for the three parameters. For example, based on the gradient stratification concentration value, the time index set to be extracted is located within the energy segment. Following a sampling frequency of 1kHz for the same charging session, three parameters—power change rate, voltage offset amplitude, and current phase difference—are read item by item from the power sequence, voltage sequence, and phase sequence. The power change rate is calculated within each segment by dividing the difference between adjacent power points by the sampling interval and recorded as a column vector within that segment. The voltage offset amplitude is recorded as a column vector by the difference between the maximum and minimum voltage values ​​within the segment. The current phase difference is recorded as a column vector by the difference between the voltage and current phase angles at the same timestamp. To address the dimensional differences between different segments, the minimum and maximum values ​​within each segment are scaled and normalized while maintaining the index. To avoid column vector misalignment due to missing samples, gaps at the same index are filled with the average of adjacent time points. Then, the three columns are... The quantities are concatenated row by row according to the same timestamp to form a sequence set. For example, in segment S1, the original value of the power change rate is in the range of 6500 to 12800, the voltage offset is in the range of 2.1 to 6.0, and the phase difference is in the range of −6 to 11. The minimum and maximum values ​​of the three columns are scaled according to the segment index. For example, at point t=205, the power change rate is scaled from 8200 to 0.27, the voltage offset is scaled from 3.5 to 0.32, and the phase difference is scaled from 5 to 0.73. If the phase difference is missing at t=207, it is filled with the average of the phase differences at t=206 and t=208 while maintaining the consistency of the segment order. Then, several triplet row vectors are obtained in S1. S2 and S3 are processed in the same way and merged to obtain a set of three columns of equal length aligned across segments. Finally, a three-parameter normalized vector value is generated. The ratio calculation submodule calls the three-parameter normalized vector value, calculates the difference between each pair of the three column vectors in each energy segment, squares them, and then takes the mean according to the number of samples to obtain three sets of root mean square errors. Then, it calculates the ratio of the three sets of root mean square errors by row and performs proportional normalization. Based on the proportional difference, it forms a three-dimensional array on the segment index to obtain the feature weight ratio value. During implementation, the normalized vector values ​​of the three parameters are used to calculate the pairwise differences and squares of the three vector columns within each energy segment. The average of these differences is then calculated based on the sample size to obtain three sets of mean square errors, denoted as the mean square error of power versus voltage, power versus phase, and voltage versus phase. Within each segment, these three mean square errors are then divided pairwise by row to obtain dimensionless ratios. Subsequently, the three ratios within the same segment are proportionally normalized to ensure they sum to 1 and recorded in the corresponding segment's three-dimensional array cell. Thresholds and benchmarks are not included in the segment's calculation; instead, the sample mean and sample size are used as the sole references. For example, in S1, with 100 samples of the normalized vector values ​​of the three parameters, the average of the squared differences between the power and voltage columns over 100 samples is 0.148, and the average of the power and phase column differences is 0.21. 2. The voltage and phase columns yield 0.175. After comparing the three terms row by row, they can be converted into unnormalized weights of 0.148, 0.212, and 0.175. Then, the sum of the three terms, 0.535, is used for proportional normalization to obtain 0.276, 0.396, and 0.327, which are then written into the three-dimensional array cell at position S1. If the root mean square error of the three terms in S2 is 0.091, 0.133, and 0.102, it is normalized to obtain 0.292, 0.426, and 0.382. If the root mean square error of S3 is 0.205, 0.187, and 0.243, it is normalized to obtain 0.321, 0.293, and 0.386. The mean of each segment is calculated using the number of samples in that segment as the denominator. Proportional normalization is performed row by row on the segment index to finally obtain the feature weight ratio value. The coefficient output submodule, based on the feature weight ratio value, performs cumulative summation on the corresponding ratio sequence for all energy segments in chronological order, separates the power ratio sequence, voltage ratio sequence and phase ratio sequence into three coefficient trajectories, and generates energy stratification weight coefficients by taking the segment cumulative results from the end of the trajectory and forming a column vector.

[0028] During implementation, based on the characteristic weight ratio values, the corresponding ratio sequences for all energy segments are cumulatively summed in chronological order. The power ratio sequence, voltage ratio sequence, and phase ratio sequence are accumulated independently, and the accumulated value is recorded in each segment to form three coefficient trajectories. Subsequently, the three trajectories are separated on the same time axis as the power coefficient trajectory, voltage coefficient trajectory, and phase coefficient trajectory. For example, if the power ratios along the segment sequence S1, S2, and S3 are 0.276, 0.292, and 0.321, the accumulated values ​​are 0.276, 0.568, and 0.889; if the voltage ratios are 0.396, 0.426, and 0.293, the accumulated value is 0.396. The values ​​are 0.822 and 1.115, and the phase ratios are 0.327, 0.382, and 0.386, resulting in cumulative values ​​of 0.327, 0.709, and 1.095. The cumulative results at the ends of the three trajectories are read under the same index and arranged into column vectors in the order of power, voltage, and phase. To facilitate parameter implementation within the device, the cumulative values ​​are limited to the non-negative range and no rescaling operation is performed. If a single-step update is required, the increment of each step can be recorded internally without changing the end value. In the example above, the ends of the three trajectories are 0.889, 1.115, and 1.095, respectively. These values ​​are arranged into column vectors and written into the parameter storage unit to finally generate the energy stratification weight coefficients.

[0029] like Figure 5 As shown, in another embodiment of the present invention, the residual chain determination module includes: The energy difference submodule calls the energy layering weight coefficient, obtains the cumulative output and input of the power detection terminal within a continuous charging cycle, performs difference calculation on the output and input of the same cycle to form an energy residual sequence, records the residual amplitude and direction mark according to the cycle index, and generates the energy residual sequence value. The specific process is as follows: Let the output of period t be Etout, and the input be Etin, both in Wh. Then the residual calculation formula is ΔEt = Etout − Etin. Taking five consecutive charging cycles as an example: The input and output quantities are Etin = (420.5, 438.2, 460.7, 472.1, 495.0) and Etout = (419.8, 437.9, 462.0, 473.3, 496.5), respectively. Performing the difference operation yields the residual sequence ΔEt = (-0.7, -0.3, +1.3, +1.2, +1.5)Wh. For each residual term, its amplitude and direction are recorded, where "positive" indicates the output is greater than the input, and "negative" indicates the opposite. To maintain consistency with the periodic index, index numbers t1 to t5 are used to record the residual amplitude and direction sequentially, forming a table structure. For example, t1 corresponds to a negative amplitude of 0.7, t2 to a negative amplitude of 0.3, t3 to a positive amplitude of 1.3, and so on. The residual sequence is then corrected by combining the energy stratification weight coefficient. Assuming the weight coefficient is 0.92, the corrected residual is the original value multiplied by the coefficient to obtain (-0.64, -0.28, +1.20, +1.10, +1.38) Wh, and the energy residual sequence value is established with time index as the output.

[0030] The direction determination submodule, based on the energy residual sequence value, performs differential calculation on the residuals of adjacent periods to obtain the residual difference sequence, counts the positive and negative directions of the difference and calculates the number of consecutive directions, performs proportional judgment on the length of the consecutive direction interval, extracts the interval with strong direction consistency as the offset segment, and establishes the residual direction consistency rate value. The residuals of adjacent periods are differentially calculated based on the energy residual sequence values ​​to identify the direction of residual change. The calculation formula is as follows: For example, from the aforementioned sequence (-0.64, -0.28, +1.20, +1.10, +1.38), we can obtain... =(+0.36,+1.48,-0.10,+0.28). Record the positive and negative signs in the positive and negative statistics tables respectively. In this example, the positive direction is counted 3 times and the negative direction is counted 1 time. The direction continuity rate is calculated using the formula... Calculation, where For the number of consecutive directions with the same number of times, The total number of periods is 5. If three of the four difference terms are consecutively positive, then... Next, the interval is judged based on the proportion of continuous length. A judgment threshold C_thr = 0.6 is set. If C ≥ C_thr, the interval is identified as a segment with consistent direction. According to this standard, the proportion of continuous direction in the sequence in the above example is 75%, which meets the threshold condition. Therefore, the period t1 to t4 is extracted as the interval with consistent direction, and the residual direction consistency rate value of 0.75 is output. The node calculation submodule calculates the rate of change of residual amplitude within the direction consistency interval based on the residual direction consistency rate value, performs a product operation on the rate and energy stratification weight coefficient to form a node vector, takes the average value of the node vector according to the time series and records the trend of change, and generates residual chain node values. During implementation, the rate of change of residual amplitude is calculated within the identified directional consistency interval based on the residual direction consistency rate value, and then multiplied with the energy stratification weighting coefficient to form a node vector. The formula for calculating the residual change rate is as follows: ,in The sampling interval is defined as ΔT = 1 hour. Taking this as an example, within the directional interval (t1 to t4), the residual changes are +0.36, +1.48, and -0.10, respectively. Taking their absolute values ​​and averaging them yields the average rate of change |S| = (0.36 + 1.48 + 0.10) / 3 = 0.65 Wh / h. This rate is then multiplied by the energy stratification weighting coefficient of 0.92 to obtain the node vector value of 0.598 Wh / h. The node values ​​for multiple periods are averaged chronologically, and the trend is recorded. Assuming the average values ​​for the following periods are 0.60, 0.64, and 0.63, the time series trend is slightly upward, recorded as the node change sequence (0.60, 0.64, 0.63). The calculated average rate of change is 0.623 Wh / h, and the residual chain node values ​​are output. The offset output submodule calculates the average rate of change of adjacent nodes in a continuous time series for the residual chain node values, determines the offset direction by combining the direction consistency sequence, and outputs the ratio of the difference magnitude to the rate of change during the cycle to generate the residual direction offset rate value.

[0031] The average rate of change of adjacent nodes in the residual chain node values ​​is calculated over a continuous time series, and the offset direction is determined by combining this with the directional consistency sequence. The calculation formula is as follows: ,in Let be the average rate of change of the nodes. Based on the previous results (0.60, 0.64, 0.63), the following is calculated: , The average rate of change is calculated as (0.0667-0.0156) / 2 = 0.0255. Then, combined with the directional consistency sequence (0.75), the overall offset direction is determined to be positive (since most changes are upward). Simultaneously, the ratio of the period difference amplitude to the rate of change is calculated. The difference amplitude is taken as the maximum residual change within the directional consistency segment, 1.48Wh, and the rate of change is taken as the average rate of change, 0.0255. Therefore, the ratio = 1.48 / 0.0255 = 58.0. This ratio is ultimately used to characterize the overall offset intensity of the energy residual within the period, generating the residual direction offset rate value.

[0032] like Figure 6 As shown, in some embodiments of the present invention, the marker self-correction management module includes: The sample matching submodule selects samples from the anomaly marker set for the residual direction offset rate value, performs a ratio calculation on the sample power offset value and voltage fluctuation amplitude and pairs them according to the time index, compares the difference between the ratio and the residual direction offset rate value according to the same index and records the absolute difference, summarizes the difference vector by segment, and generates the offset difference value. For example, for the residual direction offset rate value, samples corresponding to the time period are selected from the anomaly marker set. A ratio calculation is performed on the power offset value and voltage fluctuation amplitude of these samples, and they are matched one-to-one according to the time index to ensure a strict correspondence with the time period of the residual direction offset rate. In specific implementation, five anomaly time periods T1–T5 are selected from the continuously sampled charging pile energy data, each lasting 10 seconds, and the corresponding power offset values ​​are... The voltage fluctuation ranges are (0.8, 1.2, 1.0, 0.7, 1.5) kW respectively. The value is (4.0, 3.5, 3.8, 4.1, 3.9)V. Perform a ratio operation. The ratio sequence (0.20, 0.34, 0.26, 0.17, 0.38) kW / V was obtained. This sequence was then compared with the residual direction offset sequence at the same time index. Compare each item in the expression (0.24, 0.30, 0.21, 0.19, 0.36) and calculate the absolute value of the difference. The values ​​obtained are (0.04, 0.04, 0.05, 0.02, 0.02). To ensure statistical stability, three adjacent samples are grouped into segments and the segment mean is calculated. The mean of the first segment (T1–T3) is 0.043, and the mean of the second segment (T3–T5) is 0.03. The mean differences of each segment are used to construct a difference vector (0.043, 0.03), and combined with the time index, a shift difference value is generated. This value characterizes the difference strength between the energy shift of the outlier-labeled sample and the residual signal. The difference determination submodule judges the difference vector and energy offset threshold of each segment according to the offset difference value, records the index sequence that exceeds the threshold and calculates the continuous length ratio, extracts the label level on the corresponding index according to the ratio and gives the adjustment step according to the number of exceeding the threshold, and obtains the label adjustment amplitude value. The difference vector of each segment is evaluated against the energy offset threshold based on the offset difference value. The threshold is set based on the empirical value of the system's rated energy fluctuation. In this embodiment, the energy offset threshold is... =0.035 indicates that a difference between samples exceeding 0.035 is considered abnormal fluctuation. First, the offset difference vector (0.043, 0.03) is compared with the threshold. The first segment (0.043 > 0.035) is considered exceeding the threshold, while the second segment (0.03 < 0.035) is considered normal. The index of the exceeding-threshold segment is recorded, and the length ratio of consecutive exceeding the threshold is calculated. Assuming that consecutive exceeding the threshold accounts for 50% of the total number of segments, the ratio P = 0.5. According to the ratio, the label level is extracted on the corresponding index, and the label level L is set from 1 to 5. The ratio 0–0.2 corresponds to L1, 0.2–0.4 corresponds to L2, 0.4–0.6 corresponds to L3, 0.6–0.8 corresponds to L4, and 0.8–1.0 corresponds to L5. Therefore, the current P = 0.5 corresponds to L3. The number of exceeding the threshold within the segment is counted as 1, accounting for 50% of the total number of segments (2). An adjustment step ΔL = 1 is given based on the number of exceeding the threshold, and the corresponding label level is increased by one level. Finally, the adjustment range of each segment is summarized into a list ((segment 1, L3, ΔL=1), (segment 2, L1, ΔL=0)), and the adjustment range value of the segment is output. The correction output submodule calls the flag adjustment amplitude value, performs addition and subtraction operations on the original flag parameter set according to the index in the abnormal flag set, and records the change ratio of power offset before and after adjustment. The correction ratio sequence is obtained by accumulating the ratio by the segment and dividing it by the number of samples. The abnormal flag correction rate value is generated by taking the summary ratio at the end of the sequence. In practice, the adjustment magnitude value of the markers is called, and the original marker parameter set is incremented or decremented based on the index in the abnormal marker set. The power offset ratio of each sample in the original marker parameter set... and corresponding marking levels The values ​​are R = (0.20, 0.34, 0.26, 0.17, 0.38) and L = (2, 3, 2, 1, 3). For the segments identified as needing adjustment (e.g., segment 1 corresponding to samples 1–3), a level increment / decrement operation is performed sample by sample: sample 1's level increases by 1 to 3, sample 2 remains at 3, sample 3's level increases by 1 to 3, and samples 4–5 remain unchanged. The updated level L' = (3, 3, 3, 1, 3). Next, the percentage change in the power offset ratio before and after adjustment is calculated. If the adjusted ratio remains unchanged, the ratio is 0. If the power offset ratio of sample 3 is corrected from 0.26 to 0.28, then Q3 = |0.28 - 0.26| / 0.26 = 0.077. The average Q_t of the samples within each segment is calculated and divided by the number of samples to obtain the correction ratio sequence. For example, the average change of the three samples in segment 1 is (0 + 0 + 0.077) / 3 = 0.026, and in segment 2 it is 0, so the correction ratio sequence is (0.026, 0). The summation ratio of 0.026 at the end of this sequence is used as the overall parameter adjustment rate to generate the anomaly marker correction rate value.

[0033] The present invention also includes an anomaly identification and marking management module, which first obtains the power offset value and voltage fluctuation amplitude of the anomaly sample for the identified anomaly marking, compares the difference between them and the corresponding residual direction offset rate value, adjusts the marking level based on the degree of deviation, performs a marking state reset operation, and generates marking correction results; the marking correction results specifically include the marking adjustment amplitude, power offset difference, and voltage fluctuation ratio.

[0034] For example, the anomaly identification and labeling management module first reads the power offset value and voltage fluctuation amplitude according to the same charging session time index for labeled samples, and simultaneously retrieves the corresponding residual direction offset rate value. Then, it constructs a ternary record item for each sample time and performs pairing operations under the same index. For example, five samples T1 to T5 are taken, with power offset values ​​of 0.8 kW, 1.2 kW, 1.0 kW, 0.7 kW, and 1.5 kW, voltage fluctuation amplitudes of 4.0 V, 3.5 V, 3.8 V, 4.1 V, and 3.9 V, and residual direction... The offset values ​​were 0.24, 0.30, 0.21, 0.19, and 0.36. At each sample point, the voltage fluctuation ratio R was calculated as the ratio of the power offset value to the voltage fluctuation amplitude: R = Poff ÷ Uvar, yielding values ​​of 0.20 kW / V, 0.34 kW / V, 0.26 kW / V, 0.17 kW / V, and 0.38 kW / V. Then, the difference between R and the residual direction offset value was compared item by item according to the index, and the absolute value was taken to obtain the difference sequence: 0.04, 0.04, 0.05, 0.02, and 0.02. Subsequently, to obtain the power... The offset difference is calculated by multiplying the power offset value and the residual direction offset rate value by the voltage fluctuation amplitude, and then taking the absolute value of the difference to form a sequence of 0.16 kW, 0.15 kW, 0.202 kW, 0.079 kW, and 0.096 kW. The product represents the power derived from the residual direction offset rate value under the same voltage fluctuation conditions. Then, the original label level is adjusted based on the degree of deviation, and discrete step rules are given in the sample dimension. Differences less than 0.03 are considered unchanged with a step of 0; differences between 0.03 and 0.06 are considered constant. A step is classified as 1, and a difference of not less than 0.06 is classified as 2, which corresponds to two steps. In this example, the corresponding step counts are 1, 1, 1, 0, 0. Subsequently, the label state is reset in the index dimension, and the level changes before and after each sample update are recorded without modifying the time index arrangement. Finally, a quantitative summary of the three results is given in the session dimension. The label adjustment magnitude is taken as the average of the step counts of each sample to obtain a level of 0.6. The power offset difference is taken as the average of the above power difference sequence to obtain a power of 0.1374 kW. The voltage fluctuation ratio is taken as the average of the R values ​​of the five samples to obtain a voltage of 0.27 kW per volt.

[0035] In summary, this invention proposes a deep learning-based charging pile metering anomaly identification and self-calibration system, constructing a complete quantization chain from energy signal acquisition, feature extraction, gradient hierarchical analysis to residual chain determination and labeling self-calibration. The system is generally divided into five functional modules: signal acquisition module, energy gradient analysis module, feature quantization modeling module, residual chain determination module, and labeling self-calibration management module, ultimately extending to an anomaly identification and labeling management module to achieve accurate identification and labeling correction of charging metering anomalies. Through the fusion of multi-layer gradient analysis of energy signals and residual chain quantization calculation, a closed-loop processing flow from energy fluctuation detection and feature modeling to anomaly self-calibration is realized, forming an anomaly identification and labeling calibration mechanism that can be automatically executed and quantified in the charging pile metering system, providing a highly stable structured identification scheme for charging metering safety and data consistency.

[0036] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0037] Furthermore, this specification uses specific terms to describe embodiments thereof. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Moreover, certain features, structures, or characteristics in one or more embodiments of this specification can be appropriately combined.

[0038] Furthermore, those skilled in the art will understand that various aspects of this specification can be described and illustrated in several patentable ways or situations, including any new and useful combination of processes, machines, products, or substances, or any new and useful improvements thereof. Accordingly, various aspects of this specification can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. All of the above hardware or software may be referred to as a “data block,” “module,” “engine,” “unit,” “component,” or “system.” Furthermore, various aspects of this specification may be represented as a computer product located on one or more computer-readable media, including computer-readable program code.

[0039] Computer storage media may contain a propagated data signal containing computer program code, for example, on baseband or as part of a carrier wave. This propagated signal may take various forms, including electromagnetic, optical, and suitable combinations thereof. Computer storage media can be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, apparatus, or device to enable communication, propagation, or transmission of a program for use. The program code located on the computer storage medium can be propagated through any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0040] The computer program code required for the operation of each part of this manual can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc.; conventional procedural programming languages ​​such as C, Visual Basic, Fortran2003, Perl, COBOL2002, PHP, ABAP; dynamic programming languages ​​such as Python, Ruby, and Groovy; or other programming languages. This program code can run entirely on the user's computer, or as a standalone software package on the user's computer, or partially on the user's computer and partially on a remote computer, or entirely on a remote computer or processing device. In the latter case, the remote computer can be connected to the user's computer through any network, such as a local area network (LAN) or wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as Software as a Service (SaaS).

[0041] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented by hardware devices, they can also be implemented solely by software solutions, such as installing the described system on existing processing devices or mobile devices.

[0042] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0043] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A deep learning-based charging pile metering anomaly identification system, characterized in that, The system includes: The signal acquisition module acquires the current, voltage, and instantaneous power values ​​of the charging pile, arranges them in a time series according to the sampling time interval, multiplies the voltage and current values ​​to obtain the power time series, calculates the power change rate, and generates the energy path fluctuation rate value. The energy gradient analysis module, based on the energy path fluctuation rate value, selects adjacent sampling points to calculate the power change rate difference, counts the frequency of change direction in the gradient sequence and calculates the positive and negative change ratio, divides energy node segments according to the power change rate threshold, calculates the ratio of voltage change amplitude to current change rate in each segment, and generates gradient stratification concentration value. The feature quantization modeling module extracts the power change rate, voltage offset amplitude and current phase difference based on the gradient stratification concentration value, performs normalization operation and calculates the mean square error ratio to generate energy stratification weight coefficients. The residual chain determination module calls the energy stratification weight coefficient, calculates the energy residual sequence between power output and input, calculates the residual difference within a continuous period, performs a consistency judgment of the difference direction, and generates the residual direction offset rate value. The self-calibration management module for labeling calculates the ratio of sample power offset to voltage fluctuation amplitude for the residual direction offset rate value, compares it with the difference of residual direction offset rate, updates the labeling parameters, and generates anomaly labeling correction rate value.

2. The charging pile metering anomaly identification system according to claim 1, characterized in that, The energy path volatility value specifically refers to the current change rate, voltage change rate, and power change rate; the gradient layer concentration value specifically refers to the ratio of signal gradient directionality, change frequency, voltage change amplitude, and current change rate; the energy layer weight coefficient specifically refers to the power change weight, voltage offset weight, and current phase weight; the residual direction offset rate value specifically refers to the residual difference direction, residual change rate, and periodic residual difference; and the anomaly label correction rate value specifically refers to the label adjustment amplitude, power offset difference, and voltage fluctuation ratio.

3. The charging pile metering anomaly identification system according to claim 1, characterized in that, The signal acquisition module includes: The current acquisition submodule is used to acquire the output value of the current sensor in the charging pile, record the current sampling sequence at each time according to the sampling time interval, perform differential calculation on the current value of continuous sampling points to determine the current change rate per unit time, and generate the current change rate value. The voltage acquisition submodule is used to establish a voltage sampling sequence based on the voltage sampling terminal's acquired values, at the same time interval as the current sampling, perform differential calculations on the voltage values ​​of adjacent sampling points to obtain the voltage change rate per unit time, and generate the voltage change rate value. The power calculation submodule calls the current change rate value from the current acquisition submodule and the voltage change rate value from the voltage acquisition submodule. It performs a product operation on the current value and voltage value at each sampling time to obtain the power time series, performs differential calculation on the power series to determine the power change rate per unit time, and generates the power change rate value. The volatility generation submodule is used to combine the power change rate, current change rate, and voltage change rate values ​​according to the time series to establish the energy path volatility series, and perform average and variance calculations on the series to generate the energy path volatility value.

4. The charging pile metering anomaly identification system according to claim 1, characterized in that, The energy gradient analysis module includes: The interval construction submodule selects continuous sampling points to form a signal interval based on the energy path volatility value, performs difference calculation on the power change rate of adjacent sampling points to obtain the gradient sequence, and records the gradient sign sequence and amplitude sequence according to the sampling time interval to establish a dataset and obtain the gradient difference sequence. The direction statistics submodule counts the number of consecutive segments with the same symbol and accumulates the length based on the gradient difference sequence. It calculates the ratio of the number of positive symbols to the number of negative symbols, records the first and last indices and amplitude sum of each segment, and generates a direction continuity statistics table based on the power change rate threshold to obtain the direction continuity rate value. The segment quantization submodule calls the voltage change rate and current change rate in the direction continuity rate value and energy path volatility value, calculates the ratio of voltage change amplitude to current change rate within the same index segment to obtain the energy transfer ratio, and then averages it with the gradient difference sequence within the segment to form a numerical table, and establishes a segment quantization matrix. The concentration output submodule, for the segmented quantization matrix, calculates the variance of the numerical table based on the number of samples in each segment and merges them into the same time order. It combines the directional continuity rate value and the segment index to form a hierarchical mapping table, outputs the segment change concentration sequence according to the mapping table, and generates gradient hierarchical concentration values.

5. The charging pile metering anomaly identification system according to claim 1, characterized in that, The feature quantization modeling module includes: The extraction submodule extracts three parameters—power change rate, voltage offset amplitude, and current phase difference—from the energy segment based on the gradient stratification concentration value. The three parameters are then scaled and normalized according to the minimum and maximum values ​​based on the segment index. They are arranged into column vectors based on the same timestamp. Missing sampling points are filled with the mean of adjacent time points while maintaining the consistent segment order. The data are then concatenated row by row to form a sequence set, generating a normalized vector value for the three parameters. The ratio calculation submodule calls the three-parameter normalized vector value, calculates the difference between each pair of the three column vectors in each energy segment, squares them, and then takes the mean according to the number of samples to obtain three sets of root mean square errors. Then, it calculates the ratio of the three sets of root mean square errors by row and performs proportional normalization. Based on the proportional difference, it forms a three-dimensional array on the segment index to obtain the feature weight ratio value. The coefficient output submodule, based on the feature weight ratio value, performs cumulative summation on the corresponding ratio sequence for all energy segments in chronological order, separates the power ratio sequence, voltage ratio sequence and phase ratio sequence into three coefficient trajectories, and generates energy stratification weight coefficients by taking the segment cumulative results from the end of the trajectory and forming a column vector.

6. The charging pile metering anomaly identification system according to claim 1, characterized in that, The residual chain determination module includes: The energy difference submodule calls the energy layering weight coefficient, obtains the cumulative output and input of the power detection terminal within a continuous charging cycle, performs difference calculation on the output and input of the same cycle to form an energy residual sequence, records the residual amplitude and direction mark according to the cycle index, and generates the energy residual sequence value. The direction determination submodule, based on the energy residual sequence value, performs differential calculation on the residuals of adjacent periods to obtain the residual difference sequence, counts the positive and negative directions of the difference and calculates the number of consecutive directions, performs proportional judgment on the length of the consecutive direction interval, extracts the interval with strong direction consistency as the offset segment, and establishes the residual direction consistency rate value. The node calculation submodule calculates the rate of change of residual amplitude within the direction consistency interval based on the residual direction consistency rate value, performs a product operation on the rate and energy stratification weight coefficient to form a node vector, takes the average value of the node vector according to the time series and records the trend of change, and generates residual chain node values. The offset output submodule calculates the average rate of change of adjacent nodes in a continuous time series for the residual chain node values, determines the offset direction by combining the direction consistency sequence, and outputs the ratio of the difference magnitude to the rate of change during the cycle to generate the residual direction offset rate value.

7. The charging pile metering anomaly identification system according to claim 1, characterized in that, The marker self-correction management module includes: The sample matching submodule selects samples from the anomaly marker set for the residual direction offset rate value, performs a ratio calculation on the sample power offset value and voltage fluctuation amplitude and pairs them according to the time index, compares the difference between the ratio and the residual direction offset rate value according to the same index and records the absolute difference, summarizes the difference vector by segment, and generates the offset difference value. The difference determination submodule judges the difference vector and energy offset threshold of each segment according to the offset difference value, records the index sequence that exceeds the threshold and calculates the continuous length ratio, extracts the label level on the corresponding index according to the ratio and gives the adjustment step according to the number of exceeding the threshold, and obtains the label adjustment amplitude value. The correction output submodule calls the flag adjustment amplitude value, performs addition and subtraction operations on the original flag parameter set according to the index in the abnormal flag set, and records the change ratio of power offset before and after adjustment. The correction ratio sequence is obtained by accumulating the ratio by the segment and dividing it by the number of samples. The abnormal flag correction rate value is generated by taking the summation ratio at the end of the sequence.

8. The charging pile metering anomaly identification system according to claim 1, characterized in that, The charging pile metering anomaly identification system also includes an anomaly identification and marking management module. For the identified anomaly markings, it first obtains the power offset value and voltage fluctuation amplitude of the anomaly sample, compares them with the difference of the corresponding residual direction offset rate value, adjusts the marking level based on the degree of deviation, performs a marking state reset operation, and generates marking correction results.

9. The charging pile metering anomaly identification system according to claim 8, characterized in that, The specific results of the label correction are the label adjustment range, power offset difference, and voltage fluctuation ratio.

Images