A method and system for edge computing processing of charger quality inspection data

By identifying quality inspection events through edge nodes and calculating parameters using cloud features, the problem of feature extraction distortion in traditional quality inspection methods is solved, achieving lightweight and high-precision charger quality inspection and improving the accuracy and efficiency of charger quality inspection.

CN123114764APending Publication Date: 2026-11-17JIANGXI LIANDI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611373510.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional quality inspection methods cannot accurately distinguish between normal foldbacks and abnormal fluctuations in the dynamic performance of chargers, and the limited computing resources of edge nodes lead to problems such as feature extraction distortion and high bandwidth consumption.

Method used

By identifying candidate points for quality inspection events through edge nodes, calculating the response duration and boundary competition state, forming event fragments, and utilizing cloud-based aggregated feature calculation parameters, the charger quality inspection data is adaptively processed to achieve lightweight computation and high-precision feature extraction.

Benefits of technology

It effectively solves the problem of feature extraction distortion in traditional quality inspection methods, realizes the organic unity of lightweight edge computing and cloud optimization, and improves the accuracy and efficiency of charger quality inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114764A_ABST
    Figure CN123114764A_ABST
Patent Text Reader

Abstract

This invention relates to the field of edge computing technology, specifically to a method and system for edge computing processing of charger quality inspection data. The method includes: acquiring charger quality inspection data collected by edge nodes, forming a continuous quality inspection data sequence according to the sampling time, and identifying candidate points for quality inspection events; generating response start positions based on the candidate points, calculating the response duration based on the response start positions, determining the response boundary contention state based on the response duration and the response occupancy interval between adjacent candidate points, and identifying the event end positions using the response boundary contention state to divide the corresponding continuous quality inspection data into quality inspection event segments; and calculating voltage, current, and power characteristics based on the changes in quality inspection data within the quality inspection event segments. In this invention, edge nodes only need to call the corresponding parameters according to the current event load state and response category to complete feature calculation, eliminating the need for complex local backtracking identification and significantly reducing the real-time computing burden.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, specifically to a method and system for edge computing processing of charger quality inspection data. Background Technology

[0002] As charger products evolve towards high power density and multi-protocol fast charging, frequent load state switching has become commonplace in the quality inspection process. After load switching, the charger's output voltage and current undergo a transient response process, and its output power often exhibits a power foldback phenomenon, characterized by initial overshoot followed by correction (or initial undershoot followed by recovery). Accurately extracting event-level quality inspection characteristics such as the magnitude of voltage, current, and power changes and response time during this process is crucial for determining whether the charger's dynamic performance is up to standard. In traditional quality inspection methods, edge nodes typically use a fixed calculation method to extract the overall change magnitude and response time, applying the same rule regardless of whether power response foldback occurs. When power foldback occurs, the direction of change before and after the foldback is reversed. The overall magnitude calculation underestimates the actual fluctuation magnitude due to the cancellation of positive and negative values, and the response time determination deviates from the true steady-state time due to interference from the foldback point. This results in the true dynamic characteristics of the foldback event being smoothed out or severely distorted, making it impossible to accurately distinguish between normal foldback and abnormal fluctuations. Meanwhile, the charger quality inspection production line generates massive amounts of time-series data every day. Uploading all the raw data to the cloud for processing would result in problems such as high bandwidth consumption and large transmission latency. If complex event type identification and parameter adaptive adjustment are performed entirely at the edge, they would be limited by the computing resources and storage capacity of the edge nodes. Summary of the Invention

[0003] The purpose of this invention is to address the problems existing in the background technology by proposing a method and system for edge computing processing of charger quality inspection data.

[0004] The technical solution of the present invention: a method for edge computing processing of charger quality inspection data, comprising: S1. Acquire charger quality inspection data collected by edge nodes, form a continuous quality inspection data sequence according to the sampling time, and identify candidate points for quality inspection events; S2. Generate the response start position based on the candidate points of quality inspection events through edge nodes, calculate the response duration based on the response start position, determine the response boundary competition state based on the response duration and the response occupancy degree between adjacent candidate points of quality inspection events, and use the response boundary competition state to identify the event end position, which is used to divide the corresponding continuous quality inspection data into quality inspection event segments. S3. Calculate voltage, current and power characteristics based on changes in quality inspection data within the quality inspection event segment through edge nodes to form event-level quality inspection characteristics; S4. Receive and summarize event-level quality inspection features and quality inspection event fragments under various load conditions through the cloud, classify response categories according to the degree of power back-off and the difference in power change before and after back-off, and generate corresponding feature calculation parameters based on the response categories. S5. Process the newly collected charger quality inspection data by using feature calculation parameters through edge nodes to obtain the corresponding new event-level quality inspection features and output the charger quality inspection feature results.

[0005] As a further improvement to this technical solution, step S1, which involves forming a continuous quality inspection data sequence according to the sampling time and identifying candidate points for quality inspection events, includes the following steps: Charger quality inspection data generated by each detection channel is collected in real time through edge nodes, including at least the charger output voltage, output current and load status, and a sampling timestamp is attached to each sampled data. The charger quality inspection data is sorted in ascending order according to the sampling timestamp to generate a continuous quality inspection data sequence. The load status of adjacent sampling times is compared sequentially according to the continuous quality inspection data sequence, and candidate points of quality inspection events are identified.

[0006] As a further improvement to this technical solution, in step S2, the response duration is calculated based on the response start position, the response boundary contention state is determined based on the response duration and the response occupancy interval between adjacent quality inspection event candidate points, and the event end position is identified using the response boundary contention state, which is used to divide the corresponding continuous quality inspection data into quality inspection event segments, including the following steps: S2.1 For each candidate point of a quality inspection event in the continuous quality inspection data sequence, the sampling position corresponding to the candidate point of the quality inspection event is taken as the response start position of the current response process. The continuous changes of output voltage and output current are detected, the response end position is identified based on the continuous changes, and the response duration is obtained according to the number of sampling points between the current candidate point of the quality inspection event and the corresponding response end position. S2.2 Calculate the response occupancy based on the sampling point interval between two adjacent quality inspection event candidate points and the response duration corresponding to the previous quality inspection event candidate point; S2.3. Based on the response occupancy, use a preset occupancy threshold to determine whether the current quality inspection event is in a response boundary contention state; S2.4. Take the sampling position corresponding to the current quality inspection event candidate point as the event start position, and identify the event end position according to the response boundary competition state; S2.5. Based on the start and end positions of the event, extract the quality inspection data corresponding to each continuous sampling point between the start and end positions of the event from the continuous quality inspection data sequence to form the quality inspection event segment corresponding to the current quality inspection event.

[0007] As a further improvement to this technical solution, in step S2.3, based on the response occupancy rate, a preset occupancy rate threshold is used to determine whether the current quality inspection event is in a response boundary contention state, including the following steps: When the response occupancy is lower than the preset occupancy threshold, the current quality inspection event is determined to be in a state where there is no response boundary contention. When the response occupancy is higher than the preset occupancy threshold, the remaining sampling point interval between the current response end position and the candidate point of the next quality inspection event is calculated. If the remaining sampling point interval is less than the preset boundary interval, the current quality inspection event is determined to be in a state where there is a response boundary contention. Otherwise, it is determined to be in a state where there is no response boundary contention.

[0008] As a further improvement to this technical solution, step S2.4, identifying the end position of the event based on the response boundary contention state, includes the following steps: The sampling position corresponding to the current quality inspection event candidate point is taken as the event start position. When there is no response boundary contention, the response end position is taken as the event end position. When there is a response boundary contention, the next quality inspection event candidate point is taken as the back boundary reference. It is determined whether the response end position is already after the next quality inspection event candidate point. If so, the response end position is adjusted to the sampling point before the next candidate point. The sampling point interval between the adjusted end position and the next candidate point is calculated. If the sampling number corresponding to the adjusted end position is less than the sampling number corresponding to the event start position, the sampling point interval is regarded as 0. If the sampling point interval is still less than the preset boundary interval, the response end position is moved in the opposite direction of time to the position where the remaining sampling point interval between the next quality inspection event candidate point and the event start position reaches the preset boundary interval. At the same time, if the moved end position is before the event start position, the event end position is set as the event start position. If the moved end position is not earlier than the event start position, the moved position is taken as the event end position.

[0009] As a further improvement to this technical solution, in step S3, forming event-level quality inspection features includes the following steps: S3.1 For each quality inspection event segment, read the output voltage and output current corresponding to each sampling point in the quality inspection event segment according to the sampling time sequence, and calculate the voltage change and current change between adjacent sampling points to form a voltage change sequence and a current change sequence. S3.2. Based on the voltage change sequence and the current change sequence, with the output voltage value and output current value corresponding to the starting position of the quality inspection event as references, calculate the change amplitude of output voltage and output current within the event segment, and detect the position where the continuous change of the two first enters the corresponding stable range. Calculate the voltage response time and current response time based on the sampling point interval between this position and the starting position of the quality inspection event to obtain the voltage change characteristics and current change characteristics. S3.3 Calculate the corresponding instantaneous output power based on the output voltage and output current of each sampling point within the quality inspection event segment, form an event power sequence, and identify the power change amplitude and power response time based on the event power sequence to obtain power change characteristics; S3.4 Associate the voltage change characteristics, current change characteristics, and power change characteristics with the load state, event start position, and event end position corresponding to the current quality inspection event to form the event-level quality inspection characteristics corresponding to the current quality inspection event.

[0010] As a further improvement to this technical solution, in step S4, the response categories are divided according to the degree of power foldback and the difference in power change before and after foldback, and corresponding feature calculation parameters are generated based on the response categories, including the following steps: S4.1. Receive the event-level quality inspection characteristics, quality inspection event fragments and corresponding load status of each quality inspection event through the cloud, group the quality inspection events according to the load status, and form a corresponding historical quality inspection event set for quality inspection events belonging to the same load status. S4.2 For the set of historical quality inspection events, the cloud takes the starting position of each quality inspection event as the starting point, extracts the corresponding output power in the order of sampling time, calculates the power difference between adjacent sampling points, identifies the power change direction based on the sign change of adjacent power difference, and identifies the power return position based on the power change direction. S4.3. Divide the power change segment before the turnaround and the power change segment after the turnaround according to the power turnaround position of each quality inspection event, calculate the power change amplitude before and after the turnaround respectively, and generate the power turnaround degree by combining the degree of opposite change direction before and after the turnaround. S4.4. According to the degree of power back-off, quality inspection events under the same load condition are divided into no back-off, weak back-off and strong back-off response categories, and the distribution characteristics of voltage, current and power change amplitude in each response category are statistically analyzed. S4.5 Compare the differences between the power foldback degree and the distribution characteristics of voltage, current and power change amplitudes for different response categories, and determine the characteristic calculation method for each response category by combining the power foldback degree and power change process corresponding to each response category. S4.6. Based on the characteristic calculation method corresponding to each load state and power foldback response category, identify the corresponding characteristic calculation parameters and send the characteristic calculation parameters to the edge nodes.

[0011] As a further improvement to this technical solution, in S4.3, the power change segment before and after the power reversal is divided according to the power reversal position of each quality inspection event. The power change amplitude before and after the reversal is calculated respectively, and the power reversal degree is generated by combining the degree of oppositeity between the before and after changes. This includes the following steps: S4.31. Using the power reversal position as the dividing point of the power change segment, the continuous sampling points between the sampling point corresponding to the event start position and the sampling point corresponding to the power reversal position are divided into the power change segment before reversal; the continuous sampling points between the first sampling point after the power reversal position and the sampling point corresponding to the event end position are divided into the power change segment after reversal. S4.32. Based on the power change segment before and after the turnaround, extract the power values ​​of each sampling point in the segment before and after the turnaround, respectively, and calculate the power difference between adjacent sampling points in the two segments to form the power difference sequence before and after the turnaround. S4.33. Using the power value corresponding to the starting position of the event as the reference power for the change segment before the turnback, using the power value corresponding to the power turnback position as the reference power for the change segment after the turnback, and using the power value corresponding to the ending position of the event as the termination power, generate the power change amplitude before the turnback and the power change amplitude after the turnback. S4.34. Identify the direction of power change before the turnaround based on the cumulative sum of the power differences between adjacent sampling points in the section before the turnaround; identify the direction of power change after the turnaround based on the cumulative sum of the power differences between adjacent sampling points in the section after the turnaround. S4.35. Calculate the amplitude balance based on the power change amplitude before and after the turnaround, generate the degree of directional reversal based on the direction of change, and generate the power turnaround degree based on the amplitude balance and the degree of directional reversal using a weighted fusion method.

[0012] As a further improvement to this technical solution, step S5, outputting the charger quality inspection characteristic results, includes the following steps: S5.1 The edge node collects new charger quality inspection data in the manner of step S1, and forms a continuous quality inspection data sequence according to the sampling time, and identifies new quality inspection event candidate points based on the load state change; S5.2 For new quality inspection event candidate points, the edge nodes identify the corresponding event start position and event end position according to step S2, and extract new continuous quality inspection data to form a new quality inspection event segment corresponding to the current candidate point. S5.3 Calculate the output power corresponding to each sampling point based on the output voltage and output current in the new quality inspection event segment, and determine the power return response category of the current new quality inspection event based on the power sequence; S5.4 Based on the load status and power back-off response category of the current new quality inspection event, identify the corresponding feature calculation parameters from the feature calculation parameters sent from the cloud and form new event-level quality inspection features; S5.5. Associate the load status, event-level quality inspection characteristics, and corresponding time location of the new quality inspection event to form and output the charger quality inspection characteristic results.

[0013] On the other hand, the present invention provides a charger quality inspection data edge computing processing system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned charger quality inspection data edge computing processing method.

[0014] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: By collaborating between edge nodes and the cloud, the cloud aggregates historical event-level quality inspection features, automatically identifies power backtracking locations, and classifies responses into no backtracking, weak backtracking, and strong backtracking categories based on the degree of backtracking and the difference in amplitude before and after, generating differentiated feature calculation parameters and distributing them to edge nodes; edge nodes only need to call the corresponding parameters according to the current event load status and response category to complete feature calculation, without the need for complex local backtracking identification, greatly reducing the real-time computing burden, and at the same time enabling feature extraction rules to adapt to the dynamic form of power response, effectively solving the problem of backtracking event feature distortion under the traditional fixed calculation method, and realizing the organic unity of lightweight edge computing and intelligent cloud optimization. Attached Figure Description

[0015] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please refer to Figure 1 As shown, this embodiment provides a method for edge computing processing of charger quality inspection data, including the following steps: S1. Acquire charger quality inspection data collected by edge nodes, form a continuous quality inspection data sequence according to the sampling time, and identify candidate points for quality inspection events; In this embodiment, identifying candidate points for quality inspection events includes the following steps: During the charger quality inspection process, edge nodes collect charger quality inspection data generated by each detection channel in real time, including at least the charger output voltage, output current, and load status, and attach a sampling timestamp to each sampled data. The charger quality inspection data is sorted in ascending order according to the sampling timestamp to generate a continuous quality inspection data sequence. The load status of adjacent sampling times is compared sequentially according to the continuous quality inspection data sequence, and candidate points of quality inspection events are identified. If the load status corresponding to the current sampling time is inconsistent with the load status corresponding to the previous sampling time, the load status of three consecutive sampling points is continuously detected from the current sampling time. When the load status corresponding to the three consecutive sampling points is consistent with the load status of the current sampling time, the load status switching position corresponding to the current sampling time is determined as a candidate point of quality inspection event, and the load status before switching, the load status after switching, and the time position of the candidate point are recorded. A quality inspection event refers to a quality inspection process in which the output voltage and output current continuously change after the charger load status switches. This invention is executed collaboratively by edge nodes and the cloud.

[0018] S2. Generate the response start position based on the candidate points of quality inspection events through edge nodes, calculate the response duration based on the response start position, determine the response boundary competition state based on the response duration and the response occupancy degree between adjacent candidate points of quality inspection events, and use the response boundary competition state to identify the event end position, which is used to divide the corresponding continuous quality inspection data into quality inspection event segments. In this embodiment, the corresponding continuous quality inspection data is divided into quality inspection event segments, including the following steps: S2.1 For each candidate point of a quality inspection event in the continuous quality inspection data sequence, the sampling position corresponding to the candidate point is taken as the response start position of the current response process. Continuous changes in output voltage and output current are detected, and the response end position is identified based on these continuous changes. The response duration is obtained according to the number of sampling points between the current candidate point and the corresponding response end position. Specifically, continuously detecting changes in output voltage and output current from the response start position involves: starting from the sampling point corresponding to the current candidate point, sequentially calculating the output voltage change and output current change between adjacent sampling points, and comparing the output voltage change with a voltage stability threshold and the output current change with a current stability threshold, respectively. When continuous changes occur... If the output voltage change at each of the three consecutive sampling points is no greater than the voltage stability change threshold and the output current change is no greater than the current stability change threshold, the current output voltage and output current are determined to have entered a stable state. The sampling position that first meets the above conditions is determined as the response end position corresponding to the current quality inspection event candidate point. The response duration is obtained by subtracting the sampling point number corresponding to the response start position from the sampling point number corresponding to the response end position. For the first quality inspection event candidate point in the continuous quality inspection data sequence, the candidate point is directly used as the response start position, and the response end position is detected sequentially using the same method. The voltage stability change threshold is ±2% to ±5% of the rated output voltage, and the current stability change threshold is ±2% to ±5% of the rated output current. S2.2 Calculate the response occupancy based on the sampling interval between two adjacent quality inspection event candidate points and the response duration corresponding to the previous quality inspection event candidate point. That is, divide the response duration corresponding to the previous quality inspection event candidate point by the sampling interval between two adjacent quality inspection event candidate points. This is used to characterize the degree to which the current response process occupies the interval between adjacent quality inspection actions. S2.3. Based on the response occupancy, use a preset occupancy threshold to determine whether the current quality inspection event is in a response boundary contention state; Determining whether a response boundary race condition exists in the current quality inspection event includes the following steps: When the response occupancy is lower than or equal to the preset occupancy threshold, it indicates that the current response process has sufficient boundary buffer space between adjacent quality inspection events, and the current quality inspection event is determined to be in a state of no response boundary competition. When the response occupancy is higher than the preset occupancy threshold, the remaining sampling point interval between the current response end position and the candidate point of the next quality inspection event is calculated. If the remaining sampling point interval is less than the preset boundary interval, it indicates that the effective boundary space between the current response end position and the next quality inspection action is insufficient, and the current quality inspection event is determined to be in a state of response boundary competition; otherwise, it is determined to be in a state of no response boundary competition. The preset occupancy threshold ranges from 0.70 to 0.90, determined through engineering experience, and the preset boundary interval ranges from 3 to 10 sampling points, set in conjunction with the system's sampling frequency. S2.4. The sampling position corresponding to the current quality inspection event candidate point is taken as the event start position. The event start position is the response start position determined in step S2.1. Both are the same sampling point. The event end position is identified according to the response boundary competition state. The identification of the event end position according to the response boundary competition state is aimed at the charger quality inspection process. When the load state changes rapidly and continuously, the voltage or current dynamic response transient process of the previous event has not been completely calmed down before the subsequent event is triggered. This causes the response waveforms of adjacent events to have blurred boundaries, overlap, or insufficient effective buffer space on the time axis. As a result, traditional fixed time windows or simple threshold methods are difficult to accurately segment independent event segments and are prone to incorrectly classifying the transient tail of the previous event into the subsequent event. (or vice versa), which seriously distorts the specific engineering challenges of voltage, current or power feature extraction results and subsequent response category determination; This invention introduces a dual dynamic determination criterion of response occupancy and remaining boundary interval. By adaptively sensing the occupancy density of the current response process on the interval between adjacent events, it actively identifies the competition state and performs forward migration and dynamic correction on the event end boundary when competition occurs. This ensures that the segmented quality inspection event fragments are strictly independent in electrical response logic, effectively avoiding feature distortion and misjudgment caused by transient tail contamination, and maximizing the preservation of complete and effective response information of each event. It significantly improves the robustness, adaptability and accuracy of event segmentation at the edge end in scenarios with strong dynamic coupling and high load switching frequency. Furthermore, identifying the event end location based on the response boundary contention state includes the following steps: The sampling position corresponding to the current quality inspection event candidate point is taken as the event start position. When there is no response boundary contention, the response end position is directly taken as the event end position. When there is a response boundary contention, the next quality inspection event candidate point is used as the subsequent boundary reference. First, it is determined whether the response end position is already after the next quality inspection event candidate point, that is, the sampling sequence number corresponding to the response end position is greater than the sequence number of the next candidate point. If so, the response end position is adjusted to the sampling point before the next candidate point. Then, the sampling point interval between the adjusted end position and the next candidate point is calculated. If the sampling point interval corresponding to the adjusted end position is... If the sampling sequence number is less than the sampling sequence number corresponding to the event start position, the sampling point interval is considered to be 0. If the sampling point interval is still less than the preset boundary interval, the response end position is moved in the reverse direction of time to a position where the remaining sampling point interval between the candidate point of the next quality inspection event reaches the preset boundary interval. At the same time, if the moved end position is before the event start position, the event end position is set as the event start position. If the moved end position is not earlier than the event start position, the moved position is taken as the event end position, thereby avoiding the boundary overlap between the response tail of the current quality inspection event and the next load switching event. S2.5. Based on the start and end positions of the event, extract the quality inspection data corresponding to each continuous sampling point between the start and end positions of the event from the continuous quality inspection data sequence, and retain the corresponding load status, output voltage and output current data to form the quality inspection event segment corresponding to the current quality inspection event.

[0019] S3. Calculate voltage, current and power characteristics based on changes in quality inspection data within the quality inspection event segment through edge nodes to form event-level quality inspection characteristics; In this embodiment, the formation of event-level quality inspection features includes the following steps: S3.1 For each quality inspection event segment, read the output voltage and output current corresponding to each sampling point in the quality inspection event segment according to the sampling time sequence, and calculate the voltage change and current change between adjacent sampling points to form a voltage change sequence and a current change sequence. S3.2. Based on the voltage change sequence and current change sequence, using the output voltage and output current values ​​corresponding to the starting position of the quality inspection event as references, calculate the change amplitude of output voltage and output current within the event segment, and detect the position where the continuous change of the two first enters the corresponding stable range. Calculate the voltage response time and current response time based on the sampling point interval between this position and the starting position of the quality inspection event to obtain voltage change characteristics and current change characteristics. The continuous change of the two refers to the output voltage change and output current change between adjacent sampling points. The position where the first entry into the corresponding stable range refers to the detection starting from the starting position of the event. When the output voltage change and output current change of three consecutive sampling points are simultaneously less than their respective stable change thresholds, the sampling time or sampling point number corresponding to the first of these three consecutive sampling points, and the corresponding stable change thresholds, are the voltage stable change threshold and the current stable change threshold. S3.3 Calculate the instantaneous output power corresponding to the output voltage and output current of each sampling point in the quality inspection event segment to form an event power sequence. Identify the power change amplitude and power response time based on the event power sequence. That is, take the output power corresponding to the event start position as a reference, calculate the maximum absolute difference between each sampling point in the event power sequence and the reference power as the power change amplitude, and take the sampling time interval between the position where the power change continuously enters the stable range and the event start position as the power response time to obtain the power change characteristics. S3.4 Associate the voltage change characteristics, current change characteristics, and power change characteristics with the load status, event start position, and event end position corresponding to the current quality inspection event. That is, adopt the data association and field binding method based on the quality inspection event identifier and sampling time position to bind various change characteristics with the corresponding load status and event boundary information to the same quality inspection event record, forming the event-level quality inspection characteristics corresponding to the current quality inspection event. The event boundary information includes the event start position and event end position.

[0020] S4. Receive and aggregate event-level quality inspection features and event fragments under various load conditions through the cloud. Classify response categories based on the degree of power foldback and the difference in power change amplitude before and after foldback. Generate corresponding feature calculation parameters based on the response categories and send the feature calculation parameters to edge nodes. Among them, after the charger load switches, the output power response will show a power foldback phenomenon (such as overshoot or oscillation) that rises first and then falls (or vice versa). Traditional edge quality inspection methods use a fixed method to extract the overall change amplitude and response time regardless of whether foldback occurs, which causes the true dynamic features of the foldback event to be smoothed out or severely distorted, making it impossible to accurately distinguish between normal foldback and abnormal fluctuations, thus affecting the reliability of the quality inspection criteria. In this invention, the cloud automatically identifies the power back-off location and quantifies the back-off degree through historical events. Based on the difference in amplitude before and after the back-off, the events are subdivided into no back-off, weak back-off, and strong back-off response categories. Segmented feature calculation parameters are customized for events with strong back-off and significant differences (calculated before and after the back-off), while overall calculation is retained for other events. This adaptively integrates the power response pattern into the feature extraction rules, enabling edge nodes to obtain voltage, current, and power characteristics that are closest to the actual response process with minimal computational cost during subsequent processing. This not only improves the precision and accuracy of dynamic response evaluation, but also avoids the computational overhead of complex real-time classification at the edge by distributing parameters from the cloud, achieving a balance between lightweight and high precision. In this embodiment, generating corresponding feature calculation parameters based on the response category includes the following steps: S4.1 Receive the event-level quality inspection features, quality inspection event fragments and corresponding load status corresponding to each quality inspection event through the cloud, group the quality inspection events according to the load status, and form a corresponding historical quality inspection event set for quality inspection events belonging to the same load status. The data processed in step S4 is the completed quality inspection events that have been divided into events and calculated in S3 and uploaded to the cloud by the edge nodes. Step S4 uses the quality inspection events as historical quality inspection events. S4.2 For the set of historical quality inspection events, the cloud takes the starting position of each quality inspection event as the starting point, extracts the corresponding output power according to the sampling time sequence, and calculates the power difference between adjacent sampling points. Based on the sign change of adjacent power difference, the power change direction is identified, and the power reversal position is identified based on the power change direction. That is, the change direction corresponding to adjacent power difference is compared in turn. When the change direction of multiple consecutive power difference changes from positive to negative or from negative to positive, the sampling position where the direction reversal occurs is determined as a candidate reversal position. Further checks are made on whether the power change after the direction reversal continues to maintain the opposite direction. If the number of sampling points that continue to maintain the opposite direction after the direction reversal is less than 3, and the power change direction is subsequently restored to the direction before the direction reversal, the direction reversal is judged as an instantaneous direction fluctuation and is removed. Otherwise, the corresponding candidate reversal position is determined as the power reversal position, providing a dividing point for the subsequent division of the two power change segments before and after the reversal. S4.3. Based on the power back-off location of each quality inspection event, divide the power change segment before back-off and the power change segment after back-off, calculate the power change amplitude before and after back-off respectively, and generate the power back-off degree by combining the degree of oppositeity of the change direction before and after back-off, as a quantitative criterion for classifying response categories of no back-off, weak back-off and strong back-off. The degree of power foldback includes the following steps: S4.31. Using the power turnaround position as the dividing point of the power change segment, the continuous sampling points (including the sampling points corresponding to the power start position and the power turnaround position) between the sampling point corresponding to the power start position and the sampling point corresponding to the power turnaround position are divided into the power change segment before the turnaround; the continuous sampling points (excluding the power turnaround position but including the event end position) between the first sampling point after the power turnaround position and the sampling point corresponding to the event end position are divided into the power change segment after the turnaround. S4.32. Based on the power change segment before and after the turnaround, extract the power values ​​of each sampling point in the segment before and after the turnaround, respectively, and calculate the power difference between adjacent sampling points in each segment to form a power difference sequence before and after the turnaround. The power difference sequence before the turnaround is only used to describe the power change trend before the turnaround, and the power difference sequence after the turnaround is only used to describe the power change trend after the turnaround. The power difference is not calculated across the power turnaround position between the two segments to ensure that the turnaround point data is not reused. If the power change segment after the turnaround does not contain any sampling points (i.e., the power turnaround position coincides with the event end position, or the power turnaround position is located after the event end position), then the power change amplitude after the turnaround of the event is directly recorded as zero, and the segment after the turnaround is marked as empty. S4.33, Power value corresponding to the starting position of the event The reference power is used as the base power for the change segment before the power reversal, and the power value corresponding to the power reversal position is used. The reference power for the change segment after the turnback is the power value corresponding to the end position of the event. As the termination power, the magnitude of the power change before the turnback is generated. and the magnitude of power change after turnback Both are absolute values, representing the overall magnitude of power changes before and after the turnback, respectively; S4.34. Based on the cumulative sum of the power differences between adjacent sampling points within the segment before the turnaround (i.e., the power difference sequence before the turnaround), which is equivalent to the power at the end of the segment minus the power at the beginning, the sum is... Identify the direction of power change before the turnaround: if the cumulative sum is positive, the direction is upward; if it is negative, the direction is downward; if it is zero, the direction is no direction; based on the cumulative sum of the power differences between adjacent sampling points in the turnaround section (i.e., ... Identify the direction of power change after the turnaround; if the cumulative sum is positive, it is increasing; if it is negative, it is decreasing; if it is zero, it is directionless; if the segment after the turnaround is empty, mark its power change direction as directionless. S4.35, Based on the power change amplitude before the turnaround The magnitude of power change after the turnaround Calculate amplitude equalization The degree of opposite direction is generated based on the direction of change. If the two segments are in opposite directions (one ascending and one descending), then If the two segments have the same direction or one of the segments has no direction, then And based on amplitude equalization opposite to the direction Power foldback degree generated using weighted fusion method In the formula, For the magnitude balance weight, Weights are assigned to the degree of opposite direction. It is determined based on statistical analysis of historical data; S4.4. Based on the degree of power back-off, quality inspection events under the same load condition are divided into three response categories: no back-off, weak back-off, and strong back-off. The distribution characteristics of voltage, current, and power changes in each response category are statistically analyzed. Specifically, for each quality inspection event under the same load condition, the power back-off degree corresponding to each event is compared with the no-back-off threshold (0.15) and the strong back-off threshold (0.6). When the power back-off degree does not exceed the no-back-off threshold, the corresponding quality inspection event is classified as a no-back-off response. When the power back-off degree is greater than the no-back-off threshold but does not exceed the strong back-off threshold, the corresponding quality inspection event is classified as a weak back-off response. When the power back-off degree is greater than the strong back-off threshold, the corresponding quality inspection event is classified as a strong back-off response. Events were categorized into strong return response categories. For each response category, the voltage, current, and power change amplitudes of each quality inspection event were extracted and sorted according to their numerical values. Based on the sorting results, the range and concentration of the three change amplitudes under each response category were statistically analyzed to obtain the corresponding distribution characteristics, thus obtaining the distribution results of voltage, current, and power change amplitudes under different response categories. The concentration was described using the median and interquartile range (IQR) of each change amplitude value. The median represents the center position, and the IQR represents the dispersion, thus obtaining the distribution results of voltage, current, and power change amplitudes under different response categories. S4.5. Compare the differences between the power foldback degree and the distribution characteristics of voltage, current, and power change amplitudes for different response categories, and determine the feature calculation method for each response category based on the power foldback degree and power change process. Specifically: for no foldback, weak foldback, and strong foldback response categories, analyze the power change direction and amplitude differences before and after foldback for each quality inspection event within each category. For the no-foldback category, its power change maintains a single direction, so the feature calculation is based on the overall change process from the event start position to the event end position. For the strong foldback category, its power change exhibits a significant reversal of direction, so it is first necessary to determine whether the power change amplitude before and after foldback is significantly different. The determination method is as follows: based on the power change amplitude before foldback. and the magnitude of power change after turnback Calculate the ratio Simultaneously calculate the difference ratio If and only if satisfying or If the difference is significant, it is considered significant; otherwise, it is considered insignificant. If the difference is significant, the changes before and after the reversal are extracted separately for feature calculation, and the calculation range of each change process is determined according to the corresponding power reversal position. If the difference is insignificant, it indicates that the actual characteristics of the strong reversal event are closer to a large unidirectional change. To maintain the stability of feature extraction, the overall change process is downgraded for feature calculation (i.e., the same calculation method as the no-reversal category). If... In extreme cases (i.e., no power change before and after the turnback), the difference is directly determined to be insignificant to avoid the denominator being zero. For the weak turnback category, the difference in amplitude before and after the turnback is first determined according to the above internal event determination rules. If the difference is insignificant, it indicates that the power change of the weak turnback event is close to a single-direction change, and the overall change process from the event start position to the event end position is still used for feature calculation. If the difference is significant, it indicates that the power change before and after the turnback in the weak turnback event has significantly different characteristics. At this time, it is processed according to the same segmented calculation method as the strong turnback category, that is, the change segment before the turnback (from the event start position to the power turnback position) and the change segment after the turnback (from the power turnback position to the event end position) are extracted separately for feature calculation. Among them, the method of determining the power turnback position is completely consistent with S4.2 to ensure the uniformity of the determination criteria between different response categories. The feature calculation methods determined for each response category are respectively associated with the calculation of voltage, current and power change amplitude to form the feature calculation methods for the corresponding response categories. S4.6. Based on the feature calculation methods corresponding to each load state and power foldback response category, identify the corresponding feature calculation parameters and send them to the edge nodes. Include the category determination thresholds corresponding to each response category as part of the feature calculation parameters, also sending them to the edge nodes. This allows the edge nodes to call the corresponding feature calculation parameters to perform event-level quality inspection feature calculations based on the load state and power foldback response category of the new quality inspection event. Specifically, for the no-foldback, weak-foldback, and strong-foldback response categories under each load state, determine the voltage change amplitude, current change amplitude, and power change amplitude according to the determined feature calculation methods. The parameters required for the degree calculation; when using the overall power change process for calculation, determine the event start position and event end position as the characteristic calculation range, and determine the corresponding change amplitude calculation method and response time determination method; when using the segmented power change process before and after the turnaround for calculation, determine the power turnaround position, as well as the event start position and event end position as the characteristic calculation range, and determine the change amplitude calculation method and response time determination method before and after the turnaround respectively; use the calculation range, change amplitude calculation method and response time determination method determined under each load state and corresponding power turnaround response category as the corresponding characteristic calculation parameters.

[0021] S5. Process the newly collected charger quality inspection data by using feature calculation parameters through edge nodes to obtain the corresponding new event-level quality inspection features and output the charger quality inspection feature results. In this embodiment, the output of charger quality inspection characteristics results includes the following steps: S5.1 The edge node collects new charger quality inspection data in the manner of step S1, and forms a continuous quality inspection data sequence according to the sampling time, and identifies new quality inspection event candidate points based on the load state change; S5.2 For new quality inspection event candidate points, the edge nodes identify the corresponding event start position and event end position according to step S2, and extract new continuous quality inspection data between the two positions to form a new quality inspection event segment corresponding to the current candidate point. S5.3 Calculate the output power corresponding to each sampling point based on the output voltage and output current within the new quality inspection event segment, and determine the power back-off response category (no back-off, weak back-off, or strong back-off) of the current new quality inspection event based on the power sequence. Specifically, identify the power back-off position based on the output power, calculate the power change amplitude and direction before and after the back-off, obtain the power back-off degree of the current new quality inspection event, and then determine the response category based on the category determination threshold (no back-off threshold and strong back-off threshold) used to classify each category in S4.4. S5.4 Based on the load status of the current new quality inspection event and the determined power foldback response category, identify the corresponding feature calculation parameters from the feature calculation parameters sent from the cloud, and use only the feature calculation parameters to calculate the voltage change amplitude, current change amplitude, power change amplitude and corresponding response time of the current new quality inspection event to form a new event-level quality inspection feature. S5.5. Associate the load status, event-level quality inspection characteristics, and corresponding time location of the new quality inspection event to form and output the charger quality inspection characteristic results.

[0022] Example 2: This example provides a charger quality inspection data edge computing processing system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the charger quality inspection data edge computing processing method described in Example 1 above.

[0023] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for edge computing processing of charger quality inspection data, characterized in that, include: S1. Acquire charger quality inspection data collected by edge nodes, form a continuous quality inspection data sequence according to the sampling time, and identify candidate points for quality inspection events; S2. Generate the response start position based on the candidate points of quality inspection events through edge nodes, calculate the response duration based on the response start position, determine the response boundary competition state based on the response duration and the response occupancy degree between adjacent candidate points of quality inspection events, and use the response boundary competition state to identify the event end position, which is used to divide the corresponding continuous quality inspection data into quality inspection event segments. S3. Calculate voltage, current and power characteristics based on changes in quality inspection data within the quality inspection event segment through edge nodes to form event-level quality inspection characteristics; S4. Receive and summarize event-level quality inspection features and quality inspection event fragments under various load conditions through the cloud, classify response categories according to the degree of power back-off and the difference in power change before and after back-off, and generate corresponding feature calculation parameters based on the response categories. S5. Process the newly collected charger quality inspection data by using feature calculation parameters through edge nodes to obtain the corresponding new event-level quality inspection features and output the charger quality inspection feature results.

2. The method for edge computing processing of charger quality inspection data according to claim 1, characterized in that, In step S1, forming a continuous quality inspection data sequence according to the sampling time and identifying candidate points for quality inspection events includes the following steps: Charger quality inspection data generated by each detection channel is collected in real time through edge nodes, including at least the charger output voltage, output current and load status, and a sampling timestamp is attached to each sampled data. The charger quality inspection data is sorted in ascending order according to the sampling timestamp to generate a continuous quality inspection data sequence. The load status of adjacent sampling times is compared sequentially according to the continuous quality inspection data sequence, and candidate points of quality inspection events are identified.

3. The method for edge computing processing of charger quality inspection data according to claim 1, characterized in that, In step S2, the response duration is calculated based on the response start position. The response boundary contention state is determined based on the response duration and the response occupancy interval between adjacent quality inspection event candidate points. The event end position is then identified using the response boundary contention state. This process is used to divide the corresponding continuous quality inspection data into quality inspection event segments, including the following steps: S2.1 For each candidate point of a quality inspection event in the continuous quality inspection data sequence, the sampling position corresponding to the candidate point of the quality inspection event is taken as the response start position of the current response process. The continuous changes of output voltage and output current are detected, the response end position is identified based on the continuous changes, and the response duration is obtained according to the number of sampling points between the current candidate point of the quality inspection event and the corresponding response end position. S2.2 Calculate the response occupancy based on the sampling point interval between two adjacent quality inspection event candidate points and the response duration corresponding to the previous quality inspection event candidate point; S2.

3. Based on the response occupancy, use a preset occupancy threshold to determine whether the current quality inspection event is in a response boundary contention state; S2.

4. Take the sampling position corresponding to the current quality inspection event candidate point as the event start position, and identify the event end position according to the response boundary competition state; S2.

5. Based on the start and end positions of the event, extract the quality inspection data corresponding to each continuous sampling point between the start and end positions of the event from the continuous quality inspection data sequence to form the quality inspection event segment corresponding to the current quality inspection event.

4. The charger quality inspection data edge computing processing method according to claim 3, characterized in that, In step S2.3, based on the response occupancy rate, a preset occupancy rate threshold is used to determine whether the current quality inspection event is in a response boundary contention state, including the following steps: When the response occupancy is lower than the preset occupancy threshold, the current quality inspection event is determined to be in a state where there is no response boundary contention. When the response occupancy is higher than the preset occupancy threshold, the remaining sampling point interval between the current response end position and the candidate point of the next quality inspection event is calculated. If the remaining sampling point interval is less than the preset boundary interval, the current quality inspection event is determined to be in a state where there is a response boundary contention. Otherwise, it is determined to be in a state where there is no response boundary contention.

5. The charger quality inspection data edge computing processing method according to claim 3, characterized in that, In step S2.4, identifying the end position of the event based on the response boundary contention state includes the following steps: The sampling position corresponding to the current quality inspection event candidate point is taken as the event start position. When there is no response boundary contention, the response end position is taken as the event end position. When there is a response boundary contention, the next quality inspection event candidate point is taken as the back boundary reference. It is determined whether the response end position is already after the next quality inspection event candidate point. If so, the response end position is adjusted to the sampling point before the next candidate point. The sampling point interval between the adjusted end position and the next candidate point is calculated. If the sampling number corresponding to the adjusted end position is less than the sampling number corresponding to the event start position, the sampling point interval is regarded as 0. If the sampling point interval is still less than the preset boundary interval, the response end position is moved in the opposite direction of time to the position where the remaining sampling point interval between the next quality inspection event candidate point and the event start position reaches the preset boundary interval. At the same time, if the moved end position is before the event start position, the event end position is set as the event start position. If the moved end position is not earlier than the event start position, the moved position is taken as the event end position.

6. The method for edge computing processing of charger quality inspection data according to claim 1, characterized in that, In S3, the formation of event-level quality inspection features includes the following steps: S3.1 For each quality inspection event segment, read the output voltage and output current corresponding to each sampling point in the quality inspection event segment according to the sampling time sequence, and calculate the voltage change and current change between adjacent sampling points to form a voltage change sequence and a current change sequence. S3.

2. Based on the voltage change sequence and the current change sequence, with the output voltage value and output current value corresponding to the starting position of the quality inspection event as references, calculate the change amplitude of output voltage and output current within the event segment, and detect the position where the continuous change of the two first enters the corresponding stable range. Calculate the voltage response time and current response time based on the sampling point interval between this position and the starting position of the quality inspection event to obtain the voltage change characteristics and current change characteristics. S3.3 Calculate the corresponding instantaneous output power based on the output voltage and output current of each sampling point within the quality inspection event segment, form an event power sequence, and identify the power change amplitude and power response time based on the event power sequence to obtain power change characteristics; S3.4 Associate the voltage change characteristics, current change characteristics, and power change characteristics with the load state, event start position, and event end position corresponding to the current quality inspection event to form the event-level quality inspection characteristics corresponding to the current quality inspection event.

7. The method for edge computing processing of charger quality inspection data according to claim 1, characterized in that, In step S4, response categories are divided based on the degree of power foldback and the difference in power change before and after foldback. Corresponding feature calculation parameters are generated based on these response categories, including the following steps: S4.

1. Receive the event-level quality inspection characteristics, quality inspection event fragments and corresponding load status of each quality inspection event through the cloud, group the quality inspection events according to the load status, and form a corresponding historical quality inspection event set for quality inspection events belonging to the same load status. S4.2 For the set of historical quality inspection events, the cloud takes the starting position of each quality inspection event as the starting point, extracts the corresponding output power in the order of sampling time, calculates the power difference between adjacent sampling points, identifies the power change direction based on the sign change of adjacent power difference, and identifies the power return position based on the power change direction. S4.

3. Divide the power change segment before the turnaround and the power change segment after the turnaround according to the power turnaround position of each quality inspection event, calculate the power change amplitude before and after the turnaround respectively, and generate the power turnaround degree by combining the degree of opposite change direction before and after the turnaround. S4.

4. According to the degree of power back-off, quality inspection events under the same load condition are divided into no back-off, weak back-off and strong back-off response categories, and the distribution characteristics of voltage, current and power change amplitude in each response category are statistically analyzed. S4.5 Compare the differences between the power foldback degree and the distribution characteristics of voltage, current and power change amplitudes for different response categories, and determine the characteristic calculation method for each response category by combining the power foldback degree and power change process corresponding to each response category. S4.

6. Based on the characteristic calculation method corresponding to each load state and power foldback response category, identify the corresponding characteristic calculation parameters and send the characteristic calculation parameters to the edge nodes.

8. The charger quality inspection data edge computing processing method according to claim 7, characterized in that, In S4.3, the power change segment before and after the power reversal is divided according to the power reversal position of each quality inspection event. The power change amplitude before and after the reversal is calculated respectively, and the power reversal degree is generated by combining the degree of oppositeity between the before and after the change direction. This includes the following steps: S4.

31. Using the power reversal position as the dividing point of the power change segment, the continuous sampling points between the sampling point corresponding to the event start position and the sampling point corresponding to the power reversal position are divided into the power change segment before reversal; the continuous sampling points between the first sampling point after the power reversal position and the sampling point corresponding to the event end position are divided into the power change segment after reversal. S4.

32. Based on the power change segment before and after the turnaround, extract the power values ​​of each sampling point in the segment before and after the turnaround, respectively, and calculate the power difference between adjacent sampling points in the two segments to form the power difference sequence before and after the turnaround. S4.

33. Using the power value corresponding to the starting position of the event as the reference power for the change segment before the turnback, using the power value corresponding to the power turnback position as the reference power for the change segment after the turnback, and using the power value corresponding to the ending position of the event as the termination power, generate the power change amplitude before the turnback and the power change amplitude after the turnback. S4.

34. Identify the direction of power change before the turnaround based on the cumulative sum of the power differences between adjacent sampling points in the section before the turnaround; identify the direction of power change after the turnaround based on the cumulative sum of the power differences between adjacent sampling points in the section after the turnaround. S4.

35. Calculate the amplitude balance based on the power change amplitude before and after the turnaround, generate the degree of directional reversal based on the direction of change, and generate the power turnaround degree based on the amplitude balance and the degree of directional reversal using a weighted fusion method.

9. The method for edge computing processing of charger quality inspection data according to claim 1, characterized in that, In step S5, the output of the charger quality inspection characteristics includes the following steps: S5.1 The edge node collects new charger quality inspection data in the manner of step S1, and forms a continuous quality inspection data sequence according to the sampling time, and identifies new quality inspection event candidate points based on the load state change; S5.2 For new quality inspection event candidate points, the edge nodes identify the corresponding event start position and event end position according to step S2, and extract new continuous quality inspection data to form a new quality inspection event segment corresponding to the current candidate point. S5.3 Calculate the output power corresponding to each sampling point based on the output voltage and output current in the new quality inspection event segment, and determine the power return response category of the current new quality inspection event based on the power sequence; S5.4 Based on the load status and power back-off response category of the current new quality inspection event, identify the corresponding feature calculation parameters from the feature calculation parameters sent from the cloud and form new event-level quality inspection features; S5.

5. Associate the load status, event-level quality inspection characteristics, and corresponding time location of the new quality inspection event to form and output the charger quality inspection characteristic results.

10. A charger quality inspection data edge computing processing system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the charger quality inspection data edge computing processing method as described in any one of claims 1-9.