Image recognition-based charging pile state monitoring and safety warning method and system

Abstract

The present application relates to the technical field of charging pile state monitoring and safety warning, and particularly relates to a charging pile state monitoring and safety warning method and system based on image recognition. The method comprises the following steps: S1: obtaining the running interaction state information of the charging pile, dividing the running interaction state information according to the core driving source and the interaction structure of the interaction chain, and obtaining the first running interaction state, the second running interaction state and the third running interaction state; S2: taking the average temperature rise rate of the charging interface in the constant charging current stage as the cross-device uniform fading feature, determining the fixed reference section and the random event section based on the cross-device uniform fading feature, and obtaining the charging pile state monitoring result based on the fixed reference section and the random event section. The present application constructs a multi-dimensional perception framework of pile-vehicle-environment, and realizes the accurate differentiation of device degradation and temporary abnormality by quantifying the temperature rise rate and the feature deviation degree.

Description

Technical Field

[0001] This invention relates to the field of charging pile status monitoring and safety early warning technology, and in particular to a method and system for charging pile status monitoring and safety early warning based on image recognition. Background Technology

[0002] Image recognition technology refers to the analysis of image or video data using computer vision algorithms to identify, locate, and understand key targets and features. In the field of electric vehicle charging infrastructure, charging pile status monitoring refers to the continuous assessment of its operating parameters, equipment integrity, and connection reliability; safety early warning refers to the early identification and alarm of abnormal operating conditions that may cause risks during the charging process. Existing image recognition-based monitoring methods, by deploying cameras to acquire information about the appearance of charging equipment and its surrounding environment, effectively supplement the perception blind spots of traditional electrical sensors. They demonstrate significant advantages in identifying physical damage, encroachment behavior, and obvious smoke and fire anomalies, improving the initiative and intuitiveness of operation and maintenance.

[0003] However, existing technologies still have fundamental limitations when dealing with complex interactive scenarios. The charging process is essentially a dynamic system involving deep interaction between the charging pile, the electric vehicle, and the external environment. Current methods mostly focus on visual detection of single devices or static scenes, lacking the ability to collaboratively analyze and trace the chain interaction mechanisms among the charging pile, vehicle, and environment. This makes it difficult for the system to distinguish whether the root cause of the fault is due to performance degradation of the charging pile itself, vehicle malfunction, or an external accidental event when anomalies are caused by cross-device interaction or external environmental interference. This easily leads to false alarms and incorrect attribution, failing to meet the urgent need for accurate status assessment and responsibility determination in high-reliability charging operations.

[0004] Therefore, there is an urgent need to develop an intelligent monitoring method that can deeply integrate multi-dimensional interactive states and achieve precise collaborative analysis and traceability, so as to fundamentally improve the accuracy of charging pile status assessment and the reliability of safety early warning. Summary of the Invention

[0005] To overcome the shortcomings of existing charging pile monitoring methods in distinguishing the root causes of anomalies in cross-system interaction scenarios, this invention provides a charging pile status monitoring and safety early warning method and system based on image recognition.

[0006] The technical implementation scheme of the present invention is: a charging pile status monitoring and safety early warning method based on image recognition, comprising the following steps: S1: Obtain the operation interaction status information of the charging pile, divide the operation interaction status information according to the core driving source and interaction structure of the interaction chain, and obtain the first operation interaction status, the second operation interaction status and the third operation interaction status. S2: The average temperature rise rate of the charging interface during the constant charging current stage is used as a unified gradual change feature across devices. Based on the unified gradual change feature across devices, a fixed reference segment and a random event segment are determined. Based on the fixed reference segment and the random event segment, the charging pile status monitoring results are obtained. S3: The self-protection action information with time-marked characteristics generated by the charging pile body protection system and the vehicle body protection system during the charging process is used as the fourth operating interaction state. Based on the fourth operating interaction state, combined with the second and third operating interaction states, a safety warning and source tracing conclusion are generated.

[0007] Preferably, the step of obtaining the operating interaction status information of the charging pile, dividing the operating interaction status information according to the core driving source and interaction structure of the interaction chain to obtain a first operating interaction status, a second operating interaction status, and a third operating interaction status, includes: The first operational interaction state refers to the state information independently generated and output by the charging pile itself or the internal system of the vehicle itself. The second operational interaction state refers to the two-way information exchange and collaborative control state between the charging pile and the vehicle through direct electrical connection and communication protocol. The third operational interaction state refers to the chain-like state transmission process that is triggered by external environmental physical factors and continues to affect the charging pile or vehicle, ultimately requiring human intervention.

[0008] Preferably, the step of using the average temperature rise rate of the charging interface during the constant charging current phase as a unified gradual change characteristic across devices, and determining the fixed reference segment and the random event segment based on the unified gradual change characteristic across devices, includes: Based on the unified gradual change characteristics across devices, a long-term temperature rise rate sequence of charging piles is constructed to characterize the long-term evolution of the charging pile's performance, and a single temperature rise rate sequence of vehicles is constructed to characterize the characteristics of a single charging event. Based on the long-term temperature rise rate sequence of the charging pile, a continuous evolution subsequence from the normal state to the abnormal state of the charging pile is extracted from the first operation interaction state as a fixed reference segment. Based on the vehicle's single temperature rise rate sequence, a complete data segment from the vehicle's normal charging state to its abnormal charging state is extracted from the first operating interaction state as a random event segment.

[0009] Preferably, obtaining the charging pile status monitoring results based on a fixed reference segment and a random event segment includes: Align the fixed reference segment and the random event segment with the real-time interaction power curve recorded in the second running interaction state in the time dimension, and calculate the feature deviation between the fixed reference segment and the random event segment under the same interaction power. The characteristic deviation is calculated using a weighted root mean square error method based on a dynamic power box. Specifically, the real-time interactive power curve is divided into continuous power intervals; the mean and standard deviation of the temperature rise rate of the fixed reference segment in each continuous power interval are calculated as the baseline; the mean of the temperature rise rate of the random event segment in each continuous power interval is calculated; and the weighted normalized root mean square error of the mean of the random event segment relative to the baseline of the fixed reference segment is calculated as the characteristic deviation. If the feature deviation is lower than the preset deviation threshold, the charging pile is initially determined to be in an abnormal state. If the feature deviation is higher than the preset deviation threshold, it is preliminarily determined that the anomaly originates from this specific charging event; If the anomaly is initially determined to originate from this specific charging event, then the third operational interaction state analysis will be initiated.

[0010] Preferably, if it is initially determined that the anomaly originates from this specific charging event, then the third operational interaction state analysis is initiated, including: Data synchronization and preprocessing: acquire infrared thermal imaging video stream data and visible light video stream data of the area where the charging gun head and the vehicle socket are connected, which are synchronized with the time of this charging event, and align the two types of video stream data with the real-time interactive power curve of the second running interactive state on the time axis. Thermal feature extraction: From the aligned infrared thermal imaging video stream data, the temperature matrix of the area where the charging gun head and the vehicle socket meet is extracted frame by frame to form a temperature matrix time series, and the average temperature curve of the area where the charging gun head and the vehicle socket meet during the entire charging process is calculated. Local anomaly hotspot identification: The average temperature curve obtained from the thermal feature extraction step is used as the benchmark for the overall temperature rise trend of the area where the charging gun head and the vehicle socket meet. Each frame of the temperature matrix in the time series of the temperature matrix is ​​divided into M preset grid analysis units. For each grid analysis unit, in N consecutive frames of the temperature matrix, if the temperature value of the grid analysis unit continuously exceeds the average temperature curve value at the corresponding time by at least the threshold ΔT, and the temperature rise rate of the grid analysis unit is higher than the temperature rise rate of the surrounding grid analysis units, then the grid analysis unit is identified as a local high temperature point. External physical interference trace identification: Based on aligned visible light video stream data and image texture features extracted from infrared thermal imaging video stream data, identify water droplets, stains, frost or foreign object obstruction traces on the charging interface surface, as well as traces of charging gun not being fully inserted, mechanical deformation or displacement of cable connectors. The final charging pile status monitoring result is determined based on the preliminary judgment result of the fusion feature deviation and the third operation interaction status identification result.

[0011] Preferably, the step of determining the final charging pile status monitoring result based on the preliminary judgment result of the fusion feature deviation and the third operation interaction state identification result includes: If the feature deviation is higher than the preset deviation threshold, and the local abnormal hotspot identification step confirms the existence of a local abnormal hotspot or the external physical interference trace identification step confirms the existence of an external physical interference trace, then this abnormality will be attributed to a specific vehicle or this connection event. If the feature deviation is lower than the preset deviation threshold, and neither the local abnormal hotspot identification step nor the external physical interference trace identification step confirms the local abnormal hotspot or the external physical interference trace, then this abnormality is attributed to the gradual degradation of the charging pile itself.

[0012] Preferably, the fourth operational interaction state is defined as the self-protection action information with time-series markers generated by the charging pile protection system and the vehicle protection system during the charging process. Based on this fourth operational interaction state, combined with the second and third operational interaction states, a safety warning and tracing conclusion are generated, including: Based on the fourth operational interaction state, the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence are extracted respectively. Traverse the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence, identify and extract the time points when both sequences trigger alarms at the same time, define the corresponding time points as common warning points, and form a set of common warning points; The second anomaly correlation analysis result is obtained by performing an anomaly correlation analysis between each common early warning point in the common early warning point set and the second operational interaction state anomaly; The third anomaly correlation analysis result is obtained by analyzing the correlation between each common early warning point in the common early warning point set and the third operational interaction state anomaly. By combining the results of the second and third anomaly correlation analyses and common early warning points, a fusion decision is made to generate a safety early warning and source tracing conclusion.

[0013] Preferably, the step of obtaining the second anomaly correlation analysis result by performing anomaly correlation analysis between each common early warning point in the common early warning point set and the second operational interaction state includes: Based on the time of the common early warning point, the real-time interactive power curve and communication data of the corresponding time and the time periods before and after are extracted from the second operation interaction state; Sliding window statistics are performed on the real-time interactive power curve to calculate the standard deviation and abrupt change of voltage and current fluctuations. Fluctuations or abrupt changes exceeding the first preset threshold are judged as abnormal voltage fluctuations or current jumps. The communication data is parsed, and the message loss rate and error codes are statistically analyzed. If the loss rate exceeds the second preset threshold or a critical error code appears, it is determined as a communication interruption or a protocol-level error.

[0014] Preferably, the step of obtaining the third anomaly correlation analysis result by performing anomaly correlation analysis between each common early warning point in the common early warning point set and the third operational interaction state includes: Based on the time of the common early warning point, infrared thermal imaging video stream data and visible light video stream data of the corresponding time and the time periods before and after are extracted from the third operation interaction state. Threshold segmentation and connected component analysis are performed on the infrared image sequence formed by infrared thermal imaging video stream data. Regions whose temperature continuously exceeds the average temperature ΔT of the entire area and whose area is larger than the minimum area are identified as local abnormal hot spots. Visual detection algorithms based on inter-frame difference and feature matching are applied to visible light image sequences and infrared image sequences to identify and determine visible electric arcs, smoke, liquid leaks, or physical structure displacements.

[0015] Preferably, the image recognition-based charging pile status monitoring and safety early warning system includes: The operation status perception module is used to acquire and classify the first, second and third operation interaction status information, and extract the long-term and single temperature rise rate sequences of the charging pile and the vehicle to determine the fixed reference segment and the random event segment. The feature deviation analysis module is used to align the real-time interactive power curve and calculate the feature deviation degree, and preliminarily determine the abnormality of the charging pile itself or specific charging events based on the deviation degree threshold. The visual evidence collaboration module is used to initiate infrared and visible light video stream analysis, identify local abnormal hot spots and traces of external physical interference, and fuse visual evidence to complete the attribution of charging pile status. The collaborative safety early warning module is used to extract common early warning points in the fourth operational interaction state, correlate and analyze anomalies in the second and third states, and generate safety early warnings and source tracing conclusions.

[0016] Beneficial Effects: This invention establishes a multi-dimensional collaborative perception framework for the complex coupled system of charging pile, vehicle, and environment by constructing a four-dimensional operational interaction state encompassing the charging pile itself, the vehicle, the interaction process, and the external environment. This framework enables precise decoupling and independent analysis of three key information categories: the intrinsic state of the equipment, the bidirectional interaction process, and external chain effects. The solution innovatively introduces the average temperature rise rate of the charging interface as a unified, gradually changing characteristic across devices and designs a quantifiable index for feature deviation based on dynamic power boxes and weighted root mean square error. This solves the problem of accurately distinguishing between the gradual degradation of the charging pile itself and anomalies caused by a single vehicle or connection event. Furthermore, by defining common early warning points and implementing cross-state temporal correlation analysis, the collaborative actions of the protection system are causally anchored to real-time electrical interaction anomalies and intuitive physical visual evidence, constructing a complete evidence chain and tracing path from fault phenomena to root causes. Ultimately, while effectively eliminating false alarms and incorrect attributions, this significantly improves the accuracy of state assessment, the reliability of safety early warnings, and the intelligence level of operation and maintenance decisions for charging facilities in complex dynamic interaction scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart of the charging pile status monitoring and safety early warning method based on image recognition according to the present invention; Figure 2 This is a structural diagram of the charging pile status monitoring and safety early warning system based on image recognition of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0019] Example 1: A charging pile status monitoring and safety early warning method based on image recognition, such as Figure 1 As shown, it includes the following steps: S1: Obtain the operational interaction status information of the charging pile, divide the operational interaction status information according to the core driving source and interaction structure of the interaction chain, and obtain the first operational interaction status, the second operational interaction status, and the third operational interaction status, including: The first operational interaction state refers to the state information independently generated and output by the charging pile itself or the internal system of the vehicle itself. The second operational interaction state refers to the two-way information exchange and collaborative control state between the charging pile and the vehicle through direct electrical connection and communication protocol. The third operational interaction state refers to the chain-like state transmission process that is triggered by external environmental physical factors and continues to affect the charging pile or vehicle, ultimately requiring human intervention.

[0020] It should be noted that, addressing the shortcomings of existing charging pile monitoring solutions in distinguishing the root causes of complex interactions between the charging pile, vehicle, and environment, this step systematically reconstructs and categorizes the operational interaction status information. This information includes not only traditional sensor data from the charging pile and vehicle interior, such as battery temperature and module voltage, but also the physical state of the charging interface and environmental images collected by infrared and visible light vision systems. This information collectively constitutes an interactive chain reflecting the dynamic charging process. Its core driving forces include changes in the equipment's own performance, bidirectional energy control commands, and external physical interference. The interaction structure manifests in three basic modes: independent equipment operation, bidirectional communication between the charging pile and the vehicle, and environmentally triggered chain reactions.

[0021] Based on the above analysis, the operational interaction status information is divided into three categories, corresponding to different physical sources and interaction modes. The first operational interaction status directly originates from the self-test reports of the charging pile or vehicle controller, such as temperature alarms from the battery management system. The second operational interaction status captures the real-time dialogue process between the charging pile and the vehicle, such as the success or failure status of message exchanges during the charging handshake phase, and the deviation between the actual transmitted power and the requested power. The third operational interaction status records chain events triggered by external factors, such as rain or wind causing water to splash onto the charging port, leading to insulation monitoring anomalies and ultimately charging interruptions. This classification method constructs a multi-dimensional collaborative perception framework, providing a structured data foundation for subsequently distinguishing between equipment degradation, abnormal interaction processes, and external accidental events, effectively solving the industry pain point of difficulty in tracing cross-system faults.

[0022] S2-1: Using the average temperature rise rate of the charging interface during the constant charging current phase as a unified gradual change characteristic across devices, a fixed reference segment and a random event segment are determined based on the unified gradual change characteristic across devices, including: Based on the unified gradual change characteristics across devices, a long-term temperature rise rate sequence of charging piles is constructed to characterize the long-term evolution of the charging pile's performance, and a single temperature rise rate sequence of vehicles is constructed to characterize the characteristics of a single charging event. Based on the long-term temperature rise rate sequence of the charging pile, a continuous evolution subsequence from the normal state to the abnormal state of the charging pile is extracted from the first operation interaction state as a fixed reference segment. Based on the vehicle's single temperature rise rate sequence, a complete data segment from the vehicle's normal charging state to its abnormal charging state is extracted from the first operating interaction state as a random event segment.

[0023] It should be noted that, to overcome the technical challenge of distinguishing between the origin of an anomaly from the charging pile itself or a single event in interactive scenarios, this step of the analysis begins with the constant current phase of the charging process. In this phase, the charging current remains stable; for example, the current in a DC fast charging system is maintained at a certain set value. This provides a stable operating condition for observing thermal behavior. The average temperature rise rate of the charging interface under this condition, as a core observation indicator, physically represents the efficiency of converting electrical energy into heat energy. An increase in this rate usually indicates increased contact resistance, caused by pin oxidation or wear; conversely, a decrease in the rate suggests improved connection status or enhanced heat dissipation efficiency.

[0024] This rate was established as a unified, slowly varying characteristic across devices, its core function being to provide a state observation dimension that both the charging station and the electric vehicle can perceive and rely on. For example, carbon buildup in connectors due to long-term use can simultaneously lead to accelerated temperature rise detected by the charging station and abnormal port temperature detected by the vehicle's battery management system. By constructing long-term performance evolution sequences and single-charge event characteristic sequences for the charging station, this method achieves signal separation between device aging and instantaneous external interference. The long-term sequence depicts a slow temperature rise trend across multiple charge cycles, such as that caused by contact material fatigue; while the single-charge sequence records sudden temperature changes caused by poor connection during the charging process of a specific vehicle.

[0025] The fixed reference segment extracted in this way is essentially a quantitative description of the inherent performance degradation mode of the charging pile; the random event segment encapsulates all the specific perturbations that occur in a single charging interaction. The fundamental principle of this separate processing is to provide a comparable data foundation for accurately distinguishing between the inherent performance degradation of the device and the anomalies of a single interaction. Traditional methods often confuse these two distinct anomaly modes, while this step, through the innovation of the aforementioned feature engineering, lays a crucial foundation for achieving accurate fault tracing.

[0026] S2-2: Obtain charging pile status monitoring results based on fixed reference segments and random event segments, including: Align the fixed reference segment and the random event segment with the real-time interaction power curve recorded in the second running interaction state in the time dimension, and calculate the feature deviation between the fixed reference segment and the random event segment under the same interaction power. The characteristic deviation is calculated using a weighted root mean square error method based on a dynamic power box. Specifically, the real-time interactive power curve is divided into continuous power intervals; the mean and standard deviation of the temperature rise rate of the fixed reference segment in each continuous power interval are calculated as the baseline; the mean of the temperature rise rate of the random event segment in each continuous power interval is calculated; and the weighted normalized root mean square error of the mean of the random event segment relative to the baseline of the fixed reference segment is calculated as the characteristic deviation. If the feature deviation is lower than the preset deviation threshold, the charging pile is initially determined to be in an abnormal state. If the feature deviation is higher than the preset deviation threshold, it is preliminarily determined that the anomaly originates from this specific charging event; If the anomaly is initially determined to originate from this specific charging event, then the third operational interaction state analysis will be initiated.

[0027] It should be noted that, to overcome the limitations of traditional monitoring in distinguishing between the aging of the charging pile itself and instantaneous external interference, this method designs a quantitative diagnostic strategy. This strategy is based on power normalization and statistical comparison. The fixed reference segment and the random event segment must be time-aligned with the real-time interactive power curve in the second operating interactive state, because this curve completely records the energy transfer sequence during the charging process and is a key reference system for separating the influence of load changes and ensuring the comparability of temperature rise data at different times.

[0028] The characteristic deviation calculation based on this approach does not involve a simple comparison of overall data, but rather introduces a refined analysis method using dynamic power boxes. This method divides the charging process into continuous intervals based on real-time power values. Within each independent interval, a historical statistical baseline is established using data from a fixed reference segment, consisting of the mean temperature rise rate and its standard deviation. This baseline defines the expected normal thermal behavior and reasonable fluctuation boundaries of the charging pile under a specific power load. Subsequently, the mean temperature rise rate of the random event segment within the corresponding power interval is calculated, and the overall deviation of the event data from the historical baseline across multiple power dimensions is comprehensively evaluated using a weighted normalized root mean square error formula. The weighting method is set according to engineering needs, for example, weighting by the number of data points within each power interval or by the average power value of the interval, to ensure the reasonable contribution of different load segments to the overall deviation.

[0029] The resulting feature deviation has a clear physical orientation. If the feature deviation is below a preset deviation threshold, it indicates that the observed temperature rise pattern of the current charging event closely matches the inherent aging trajectory of the charging pile due to long-term use. This consistency constitutes a direct basis for determining the abnormal state of the charging pile itself. Conversely, if the feature deviation is above the preset deviation threshold, it indicates that the current temperature rise significantly exceeds the normal behavior range described by the historical baseline. This significant deviation strongly suggests that the root cause of the anomaly is not the aging of the equipment itself, but rather external specific factors such as vehicle interface contamination or poor physical connection during this charging process. This threshold-based judgment logic, through rigorous multi-dimensional statistical comparison, not only effectively replaces the fuzzy judgment of traditional single thresholds, but also lays a reliable decision-making foundation for accurate attribution and subsequent targeted analysis.

[0030] If the anomaly is initially determined to originate from this specific charging event, then the third operational interaction state analysis is initiated, including: Data synchronization and preprocessing: acquire infrared thermal imaging video stream data and visible light video stream data of the area where the charging gun head and the vehicle socket are connected, which are synchronized with the time of this charging event, and align the two types of video stream data with the real-time interactive power curve of the second running interactive state on the time axis. Thermal feature extraction: From the aligned infrared thermal imaging video stream data, the temperature matrix of the area where the charging gun head and the vehicle socket meet is extracted frame by frame to form a temperature matrix time series, and the average temperature curve of the area where the charging gun head and the vehicle socket meet during the entire charging process is calculated. Local anomaly hotspot identification: The average temperature curve obtained from the thermal feature extraction step is used as the benchmark for the overall temperature rise trend of the area where the charging gun head and the vehicle socket meet. Each frame of the temperature matrix in the time series of the temperature matrix is ​​divided into M preset grid analysis units. For each grid analysis unit, in N consecutive frames of the temperature matrix, if the temperature value of the grid analysis unit continuously exceeds the average temperature curve value at the corresponding time by at least the threshold ΔT, and the temperature rise rate of the grid analysis unit is higher than the temperature rise rate of the surrounding grid analysis units, then the grid analysis unit is identified as a local high temperature point. External physical interference trace identification: Based on aligned visible light video stream data and image texture features extracted from infrared thermal imaging video stream data, identify water droplets, stains, frost or foreign object obstruction traces on the charging interface surface, as well as traces of charging gun not being fully inserted, mechanical deformation or displacement of cable connectors. The final charging pile status monitoring result is determined based on the preliminary judgment result of the fusion feature deviation and the third operation interaction status identification result.

[0031] It should be noted that when the characteristic deviation analysis initially determines that the anomaly originates from a specific charging event, it means that the fault mode significantly deviates from the charging pile's own aging baseline and is caused by external instantaneous factors. To confirm the specific cause, a third-party operational interaction state analysis must be initiated to obtain direct visual evidence. However, when the anomaly is determined to be a charging pile status anomaly, since it is consistent with the inherent aging trend, no additional visual verification is required.

[0032] The area where the charging gun head meets the vehicle's charging socket is the core interface where the charging plug and the vehicle's charging port physically contact and transmit electrical energy. Poor contact in this area can easily lead to overheating. Simultaneous acquisition of infrared thermal imaging video stream data and visible light video stream data in this area is typically achieved by deploying dual-spectrum cameras at or near the charging station. Infrared data reflects the surface temperature distribution of an object, quantifying thermal effects, such as identifying abnormally hot spots caused by increased contact resistance. Visible light data records the object's appearance and color texture, used to identify its physical state, such as detecting water stains or frost at the charging gun connector.

[0033] Strict time alignment of the two types of visual data streams with the real-time interactive power curves of the second operational interaction state is crucial. Without alignment, it becomes impossible to correlate observed thermal anomalies or physical disturbances with specific electrical events, such as power drops, leading to a broken causal chain and making it difficult to identify the root cause of the anomaly. After alignment, the temperature matrix extracted frame-by-frame from the infrared video stream is essentially a two-dimensional array of temperature values ​​for each pixel in the junction region within each frame, directly representing the spatial temperature field at that moment. Arranging the temperature matrices of consecutive frames in chronological order forms a temperature matrix time series, thus comprehensively recording the dynamic evolution history of the temperature field in the entire junction region during charging. The average temperature curve calculated based on this series throughout the charging process describes the average trend of the overall temperature rise in the junction region over time, providing a global background benchmark for identifying local anomalies.

[0034] In the identification of localized hotspots, the average temperature curve is used as a benchmark to distinguish between overall uniform temperature rise and localized concentrated overheating. Each frame of the temperature matrix is ​​divided into M preset grid analysis units, such as using a uniform grid, to discretize the spatially continuous temperature field into sub-regions that can be analyzed independently. In N consecutive frames, a grid unit is identified as a localized high-temperature point only when its temperature value consistently exceeds the overall average temperature ΔT at the corresponding moment, and its temperature rise rate consistently exceeds that of its surrounding units. Here, M and N are preset integers based on the infrared camera resolution, frame rate, and required detection sensitivity, determined by those skilled in the art through conventional experiments. The surrounding grid analysis units refer to other grid units within a certain neighborhood of the current grid unit, such as an eight-neighborhood (i.e., eight adjacent units in the top, bottom, left, right, and four diagonal directions), or all grid analysis units within a larger window, such as a 3×3 or 5×5 rectangular window. This dual criterion of persistence and relativity aims to eliminate interference from transient thermal noise or light reflection, ensuring that the identified overheating fault point is a stable, diffuse, and genuine one, typically indicating contact point ablation or severe contamination. The threshold ΔT is set based on the statistical distribution of the difference between the grid cell temperature and the corresponding average temperature curve during historical normal charging processes. Specifically, a large amount of infrared thermal imaging data from normal charging events is collected, and the difference between each grid cell and the average temperature curve at each moment is calculated to construct a difference distribution model. ΔT is taken as the high percentile of this distribution, such as 99% or 99.5%, to ensure that temperature fluctuations caused by thermal noise, changes in illumination, or transient reflections are not misjudged as abnormal under normal operating conditions. Simultaneously, ΔT is fine-tuned in conjunction with the thermal tolerance standards and actual safety margins of the charging interface material to ensure that it is both statistically significant and meets engineering safety requirements.

[0035] External physical interference trace identification integrates morphological information from visible light video and texture features from infrared video. Edge detection and color segmentation algorithms in visible light images identify water droplets, stains, or foreign object obstructions on the interface surface. Simultaneously, texture changes in the infrared image sequence, such as the texture of abnormally low-temperature areas caused by localized cooling due to water stains, are analyzed in conjunction with structural consistency to determine if the charging gun is not fully inserted or if mechanical displacement exists. Finally, the localized high-temperature points or physical interference traces discovered through visual analysis are fused with the preliminary judgment results of feature deviation: if both are true, it strongly confirms that the anomaly originates from a specific vehicle or connection event; otherwise, it supports the determination of progressive degradation of the charging pile itself. This process, by introducing an objective visual evidence chain, solves the misjudgment problem caused by traditional methods relying on indirect electrical signals for inference in complex interaction scenarios, achieving accurate and visual source tracing of faults.

[0036] The final charging pile status monitoring results are determined based on the preliminary judgment results of the fusion feature deviation and the third operational interaction status identification results, including: If the feature deviation is higher than the preset deviation threshold, and the local abnormal hotspot identification step confirms the existence of a local abnormal hotspot or the external physical interference trace identification step confirms the existence of an external physical interference trace, then this abnormality will be attributed to a specific vehicle or this connection event. If the feature deviation is lower than the preset deviation threshold, and neither the local abnormal hotspot identification step nor the external physical interference trace identification step confirms the local abnormal hotspot or the external physical interference trace, then this abnormality is attributed to the gradual degradation of the charging pile itself.

[0037] It should be noted that the final determination of the charging pile status monitoring results relies on a set of preset deviation thresholds and a decision-making logic that integrates multi-source evidence. The specific value of the preset deviation threshold is not arbitrarily set, but rather calibrated by statistically analyzing the distribution of characteristic deviations calculated from massive amounts of historical normal operation data. For example, statistical methods are used to calculate the upper limit of the indicator within the normal fluctuation range, which serves as the quantitative boundary for defining the degree of deviation between the current event and historical patterns.

[0038] When the deviation of the features obtained from the previous analysis exceeds the preset threshold, it means that the thermal behavior pattern of this charging process can no longer be explained by the inherent aging process of the charging pile itself, suggesting that the anomaly originated from external factors. At this point, if infrared and visible light visual analysis further discovers physical traces such as localized high-temperature concentration points or the presence of water droplets or foreign objects obstructing the charging interface area, these visual evidences provide direct physical evidence to support the aforementioned externality inference. The pattern anomaly indicated by the high deviation and the specific physical defects revealed by the visual evidence both point to the same conclusion: this anomaly should be attributed to the specific vehicle currently connected or the specific plug-in operation performed.

[0039] Conversely, when the characteristic deviation is below the preset deviation threshold, it indicates that the observed temperature rise curve basically matches the trajectory of long-term performance degradation of the charging pile. Based on this, if the visual analysis step fails to detect any local hot spots or traces of external interference, the possibility of sudden external factors causing this anomaly is ruled out. The lack of evidence of external causes, coupled with the fact that the electrical characteristics conform to an internal aging pattern, makes attributing the anomaly to the gradual degradation of the charging pile itself the only reasonable explanation.

[0040] This fusion decision-making mechanism, which cross-validates quantitative pattern difference analysis with visual physical survey results, is essential because it effectively overcomes the limitations of relying on a single judgment criterion. The reliability of the analysis conclusions will be significantly compromised if any step is missing. For example, ignoring high deviations and concluding that a charging pile is normal solely based on the absence of obvious physical damage could lead to underreporting potential risks caused by vehicles; conversely, concluding a charging pile malfunction solely based on low deviations risks misjudging extremely subtle external contamination, which happens to resemble aging patterns, as internal degradation. Even if preliminary electrical analysis indicates an abnormal charging pile condition, this conclusion must be confirmed through cross-validation with visual evidence. This process design greatly improves the accuracy and reliability of condition monitoring and root cause tracing, directly addressing the core pain point of current charging facility operation and maintenance where it is difficult to accurately assign responsibility for complex interactive faults.

[0041] S3: Using the self-protection action information with time-marked characteristics generated by the charging pile's protection system and the vehicle's protection system during the charging process as the fourth operational interaction state, and combining the fourth operational interaction state with the second and third operational interaction states, a safety warning and source tracing conclusion are generated, including: Based on the fourth operational interaction state, the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence are extracted respectively. Traverse the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence, identify and extract the time points when both sequences trigger alarms at the same time, define the corresponding time points as common warning points, and form a set of common warning points; The second anomaly correlation analysis result is obtained by performing an anomaly correlation analysis between each common early warning point in the common early warning point set and the second operational interaction state anomaly; The third anomaly correlation analysis result is obtained by analyzing the correlation between each common early warning point in the common early warning point set and the third operational interaction state anomaly. By combining the results of the second and third anomaly correlation analyses and common early warning points, a fusion decision is made to generate a safety early warning and source tracing conclusion.

[0042] It should be noted that the core purpose of constructing the fourth operational interaction state in this step is to examine the action information of the charging pile and the electric vehicle's internal protection subsystem within a unified temporal framework. The charging pile's own protection system executes protection based on electrical parameters, such as actively cutting off power supply when it detects that the output current exceeds a safety limit; the vehicle's own protection system makes decisions based on battery status, such as requesting to stop charging when the battery pack temperature reaches an internally preset temperature protection threshold. This action information with temporal markers is integrated into the fourth state, providing a crucial data foundation for analyzing the interaction between the two independent systems during fault events.

[0043] The self-protection monitoring data sequences of the charging pile and the vehicle are formed by sorting the protection action events recorded by their respective systems during historical and current charging processes according to timestamps. By traversing and comparing these two sequences, the moments when both generate alarm signals at the same time are selected, i.e., common warning points, and a corresponding set is formed. Common warning points identify the instant when the protection mechanisms of both the charging pile and the vehicle react synchronously to the same potential danger. This is fundamentally different from warnings triggered by only one side. The former often points to a connection interface failure or a common external threat, such as both entering a protection state simultaneously due to an interface arc.

[0044] Each identified common warning point will trigger a multi-dimensional retrospective analysis. Anomaly correlation analysis with the second operational interaction state aims to examine the charging interaction dynamics just before the protection action occurs. By analyzing disturbance patterns in real-time power and communication data streams, it aims to determine whether the fault originates from abnormal energy or information exchange between the charging station and the vehicle. Anomaly correlation analysis with the third operational interaction state is used to find corroborating evidence at the physical level. Using synchronized infrared thermal imaging and visible light video data, it detects abnormal heating, discharging phenomena, or physical structural abnormalities at the charging interface at the warning point.

[0045] The final fusion decision-making logic is based on a comprehensive judgment of the above retrospective analysis results. If the analysis shows that there is a significant abnormality in the interaction process in the second state and no external physical interference is found in the third state, it is inferred that the protection coordination was triggered by a fault in the charging interaction process itself. Conversely, if the third state clearly captures traces of external physical interference, then regardless of the performance of the second state, the root cause of the event should be directly attributed to this external factor. If no clear leading anomaly is found in either analysis, the possibility of false alarms in the protection system or the existence of hidden compatibility faults needs to be considered. This analytical paradigm effectively solves the technical problem of accurately tracing the initial cause of a safety event in a complex coupled system by constructing a complete evidence chain from "synchronization of protection actions to retrospective analysis of the interaction process to investigation of physical evidence".

[0046] The second anomaly correlation analysis results are obtained by correlating each common early warning point in the common early warning point set with the second operational interaction state anomaly, including: Based on the time of the common early warning point, the real-time interactive power curve and communication data of the corresponding time and the time periods before and after are extracted from the second operation interaction state; Sliding window statistics are performed on the real-time interactive power curve to calculate the standard deviation and abrupt change of voltage and current fluctuations. Fluctuations or abrupt changes exceeding the first preset threshold are judged as abnormal voltage fluctuations or current jumps. The communication data is parsed, and the message loss rate and error codes are statistically analyzed. If the loss rate exceeds the second preset threshold or a critical error code appears, it is determined as a communication interruption or a protocol-level error.

[0047] It should be noted that the analysis in this section focuses on the instantaneous electrical and communication environment before and after the occurrence of the common warning point, aiming to trace the direct procedural causes that led to the synchronous triggering of the charging pile and vehicle protection mechanisms. The common warning point was selected as the time reference, and real-time interactive power curves and communication protocol data streams within an appropriate range before and after it formed the basis of the analysis. This design is based on a fundamental judgment: the coordinated actions of the protection system will inevitably leave traceable anomalies at the electrical or information interaction level. The real-time interactive power curve records the dynamic stability of energy transmission, while the communication data reflects the quality of command coordination and state synchronization between the charging pile and the vehicle battery management system.

[0048] In terms of specific analysis methods, using a sliding time window for statistical processing of the power curve is an effective approach. By setting a fixed-duration time window and sliding it along the curve, the standard deviation of voltage and current series fluctuations and abrupt changes are calculated segment by segment. The physical meaning of the standard deviation of fluctuation lies in characterizing the degree of dispersion of the signal around the mean; an abnormal increase in this value usually indicates a deterioration in power quality or severe load fluctuations, while a decrease indicates that the operation is becoming more stable. The detection of abrupt changes is specifically used to capture signals that foreshadow faults, such as sharp jumps in instantaneous short circuits. The basis for these statistical quantities constituting anomalies lies in the judgment of preset thresholds. The setting of thresholds is not subjective; for example, the first preset threshold is determined by statistical analysis of massive amounts of historical normal charging data, taking a specific high percentile (e.g., 99%) of the standard deviation distribution as the boundary. The second preset threshold is set based on the analysis of historical normal communication data, by statistically analyzing the distribution of message loss rates during normal charging, selecting a specific high percentile (e.g., 99.5%) as the loss rate threshold; at the same time, according to the charging communication protocol specification, specific codes that indicate serious fault states and require immediate protective actions are defined as critical error codes.

[0049] The parsing of communication data follows protocol specifications, decoding and verifying message sequences. By statistically analyzing the message loss rate and the frequency of critical error codes within a specific time window, the reliability of the communication link can be objectively assessed. When the loss rate consistently exceeds a second preset threshold based on historical normal communication quality, or when critical codes indicating serious protocol errors are parsed, a communication interruption or protocol-level error is determined to have occurred. Traditional methods, when faced with simultaneous actions of two protection systems, often stop at recording phenomena without in-depth analysis of the interaction process. This step, through implementing this refined retrospective correlation analysis, links the result of a common early warning point with the specific electrical disturbance or communication failure process, thus providing an indispensable technical path for deciphering the initial causes of complex protection coordination events and significantly improving the accuracy of security event tracing.

[0050] The third anomaly correlation analysis results are obtained by correlating each common early warning point in the common early warning point set with the third operational interaction state anomaly, including: Based on the time of the common early warning point, infrared thermal imaging video stream data and visible light video stream data of the corresponding time and the time periods before and after are extracted from the third operation interaction state. Threshold segmentation and connected component analysis are performed on the infrared image sequence formed by infrared thermal imaging video stream data. Regions whose temperature continuously exceeds the average temperature ΔT of the entire area and whose area is larger than the minimum area are identified as local abnormal hot spots. Visual detection algorithms based on inter-frame difference and feature matching are applied to visible light image sequences and infrared image sequences to identify and determine visible electric arcs, smoke, liquid leaks, or physical structure displacements.

[0051] It should be noted that the analysis in this section aims to find visual physical causes for the synchronized actions of charging piles and vehicle protection systems. A common warning point is selected as the time origin, and an appropriate time window is extracted before and after it. Infrared thermal imaging and visible light video data within this time period are retrieved simultaneously, forming the basis for visual evidence collection. The infrared data stream carries the temperature field information of the charging interface area, and its core function is to reveal abnormal heating caused by electrical faults; the visible light data stream records the appearance of the interface, used to capture physical damage, foreign object attachment, or abnormal connector posture.

[0052] Thresholding segmentation of the image sequence composed of infrared data essentially categorizes pixels in each frame into two main categories: suspected high-temperature areas and background, based on a set temperature threshold, thus initially extracting potential overheated regions. Subsequent connected component analysis clusters and labels these binarized regions, calculating the geometric properties of each independent region, such as its area. The logic for identifying localized anomalous hotspots—requiring the region's temperature to consistently exceed the overall average temperature ΔT and its area to be greater than a minimum threshold—aims to eliminate interference from random thermal noise and edge pixels, ensuring that the captured data represents genuine potential hazards with significant thermal energy and spatial scale. These hazards typically indicate contact point ablation or severe increases in contact resistance. The minimum area threshold is set based on two aspects: First, based on the spatial resolution of the infrared camera, such as pixel size, and the mapping relationship between it and the actual physical size of the charging interface junction area, the minimum physical area of ​​typical abnormal hotspots is converted into the number of pixels as a reference, such as contact point ablation and dirt accumulation. Second, by performing connected component analysis on historical normal charging infrared image sequences, statistical analysis of non-fault factors, such as thermal noise and image registration errors, and the resulting connected component area distribution, the high percentile of this distribution, such as 99.5%, is taken as the lower limit of the minimum area threshold, thereby effectively filtering out false hotspots and ensuring that the identified area has the real physical scale and engineering significance of the fault.

[0053] Applying inter-frame difference technology to visible and infrared dual sequences can effectively capture transient and drastic changes in pixel grayscale or temperature caused by events such as electric arc flashes, smoke generation, or liquid splashes in image sequences. Feature matching algorithms are used to track the spatial positions of key components of the charging gun head between different frames, determining incomplete insertion or mechanical displacement issues by comparing positional consistency. For example, liquid leakage appears as a diffused low-temperature region in the infrared sequence and is accompanied by specific reflective textures in the visible light sequence. This multimodal visual correlation analysis method overcomes the limitation of traditional monitoring, which relies solely on electrical logs for indirect inference after a protection action occurs. The aforementioned inter-frame difference and feature matching are mature computer vision methods. Those skilled in the art use optical flow and conventional ORB algorithms for displacement detection, and threshold segmentation and background subtraction methods for identifying electric arcs and smoke. By anchoring the common warning moment with direct physical image evidence, this step provides irrefutable on-site evidence to distinguish whether an event is caused by external physical impact, internal gradual degradation, or a false alarm from the protection system, greatly improving the certainty of safety event diagnosis and the credibility of the conclusions.

[0054] Example 2: Based on Example 1, a charging pile status monitoring and safety early warning system based on image recognition, such as... Figure 2 As shown, it includes: The operation status perception module is used to acquire and classify the first, second and third operation interaction status information, and extract the long-term and single temperature rise rate sequences of the charging pile and the vehicle to determine the fixed reference segment and the random event segment. The feature deviation analysis module is used to align the real-time interactive power curve and calculate the feature deviation degree, and preliminarily determine the abnormality of the charging pile itself or specific charging events based on the deviation degree threshold. The visual evidence collaboration module is used to initiate infrared and visible light video stream analysis, identify local abnormal hot spots and traces of external physical interference, and fuse visual evidence to complete the attribution of charging pile status. The collaborative safety early warning module is used to extract common early warning points in the fourth operational interaction state, correlate and analyze anomalies in the second and third states, and generate safety early warnings and source tracing conclusions.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A charging pile status monitoring and safety early warning method based on image recognition, characterized in that, Includes the following steps: S1: Obtain the operation interaction status information of the charging pile, divide the operation interaction status information according to the core driving source and interaction structure of the interaction chain, and obtain the first operation interaction status, the second operation interaction status and the third operation interaction status. S2: The average temperature rise rate of the charging interface during the constant charging current stage is used as a unified gradual change feature across devices. Based on the unified gradual change feature across devices, a fixed reference segment and a random event segment are determined. Based on the fixed reference segment and the random event segment, the charging pile status monitoring results are obtained. S3: The self-protection action information with time-marked characteristics generated by the charging pile body protection system and the vehicle body protection system during the charging process is used as the fourth operating interaction state. Based on the fourth operating interaction state, combined with the second and third operating interaction states, a safety warning and source tracing conclusion are generated.

2. The charging pile status monitoring and safety early warning method based on image recognition according to claim 1, characterized in that, The process of acquiring the charging pile's operational interaction status information involves dividing the information according to the core driving source and interaction structure of the interaction chain to obtain a first operational interaction status, a second operational interaction status, and a third operational interaction status, including: The first operational interaction state refers to the state information independently generated and output by the charging pile itself or the internal system of the vehicle itself. The second operational interaction state refers to the two-way information exchange and collaborative control state between the charging pile and the vehicle through direct electrical connection and communication protocol. The third operational interaction state refers to the chain-like state transmission process that is triggered by external environmental physical factors and continues to affect the charging pile or vehicle, ultimately requiring human intervention.

3. The charging pile status monitoring and safety early warning method based on image recognition according to claim 1, characterized in that, The method of using the average temperature rise rate of the charging interface during the constant charging current phase as a unified gradual change characteristic across devices, and determining the fixed reference segment and random event segment based on the unified gradual change characteristic across devices, includes: Based on the unified gradual change characteristics across devices, a long-term temperature rise rate sequence of charging piles is constructed to characterize the long-term evolution of the charging pile's performance, and a single temperature rise rate sequence of vehicles is constructed to characterize the characteristics of a single charging event. Based on the long-term temperature rise rate sequence of the charging pile, a continuous evolution subsequence from the normal state to the abnormal state of the charging pile is extracted from the first operation interaction state as a fixed reference segment. Based on the vehicle's single temperature rise rate sequence, a complete data segment from the vehicle's normal charging state to its abnormal charging state is extracted from the first operating interaction state as a random event segment.

4. The charging pile status monitoring and safety early warning method based on image recognition according to claim 1, characterized in that, The method for obtaining charging pile status monitoring results based on fixed reference segments and random event segments includes: Align the fixed reference segment and the random event segment with the real-time interaction power curve recorded in the second running interaction state in the time dimension, and calculate the feature deviation between the fixed reference segment and the random event segment under the same interaction power. The characteristic deviation is calculated using a weighted root mean square error method based on a dynamic power box. Specifically, the real-time interactive power curve is divided into continuous power intervals; the mean and standard deviation of the temperature rise rate of the fixed reference segment in each continuous power interval are calculated as the baseline; the mean of the temperature rise rate of the random event segment in each continuous power interval is calculated; and the weighted normalized root mean square error of the mean of the random event segment relative to the baseline of the fixed reference segment is calculated as the characteristic deviation. If the feature deviation is lower than the preset deviation threshold, the charging pile is initially determined to be in an abnormal state. If the feature deviation is higher than the preset deviation threshold, it is preliminarily determined that the anomaly originates from this specific charging event; If the anomaly is initially determined to originate from this specific charging event, then the third operational interaction state analysis will be initiated.

5. The charging pile status monitoring and safety early warning method based on image recognition according to claim 4, characterized in that, If it is initially determined that the anomaly originates from this specific charging event, then the third operational interaction state analysis is initiated, including: Data synchronization and preprocessing: acquire infrared thermal imaging video stream data and visible light video stream data of the area where the charging gun head and the vehicle socket are connected, which are synchronized with the time of this charging event, and align the two types of video stream data with the real-time interactive power curve of the second running interactive state on the time axis. Thermal feature extraction: From the aligned infrared thermal imaging video stream data, the temperature matrix of the area where the charging gun head and the vehicle socket meet is extracted frame by frame to form a temperature matrix time series, and the average temperature curve of the area where the charging gun head and the vehicle socket meet during the entire charging process is calculated. Local anomaly hotspot identification: The average temperature curve obtained from the thermal feature extraction step is used as the benchmark for the overall temperature rise trend of the area where the charging gun head and the vehicle socket meet. Each frame of the temperature matrix in the time series of the temperature matrix is ​​divided into M preset grid analysis units. For each grid analysis unit, in N consecutive frames of the temperature matrix, if the temperature value of the grid analysis unit continuously exceeds the average temperature curve value at the corresponding time by at least the threshold ΔT, and the temperature rise rate of the grid analysis unit is higher than the temperature rise rate of the surrounding grid analysis units, then the grid analysis unit is identified as a local high temperature point. External physical interference trace identification: Based on aligned visible light video stream data and image texture features extracted from infrared thermal imaging video stream data, identify water droplets, stains, frost or foreign object obstruction traces on the charging interface surface, as well as traces of charging gun not being fully inserted, mechanical deformation or displacement of cable connectors. The final charging pile status monitoring result is determined based on the preliminary judgment result of the fusion feature deviation and the third operation interaction status identification result.

6. The charging pile status monitoring and safety early warning method based on image recognition according to claim 5, characterized in that, The determination of the final charging pile status monitoring result based on the preliminary judgment result of the fusion feature deviation and the third operation interaction status identification result includes: If the feature deviation is higher than the preset deviation threshold, and the local abnormal hotspot identification step confirms the existence of a local abnormal hotspot or the external physical interference trace identification step confirms the existence of an external physical interference trace, then this abnormality will be attributed to a specific vehicle or this connection event. If the feature deviation is lower than the preset deviation threshold, and neither the local abnormal hotspot identification step nor the external physical interference trace identification step confirms the local abnormal hotspot or the external physical interference trace, then this abnormality is attributed to the gradual degradation of the charging pile itself.

7. The charging pile status monitoring and safety early warning method based on image recognition according to claim 1, characterized in that, The fourth operational interaction state is defined as the self-protection action information with time-series markers generated by the charging pile's protection system and the vehicle's protection system during the charging process. Based on this fourth operational interaction state, combined with the second and third operational interaction states, safety warnings and tracing conclusions are generated, including: Based on the fourth operational interaction state, the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence are extracted respectively. Traverse the charging pile self-protection monitoring data sequence and the vehicle self-protection monitoring data sequence, identify and extract the time points when both sequences trigger alarms at the same time, define the corresponding time points as common warning points, and form a set of common warning points; The second anomaly correlation analysis result is obtained by performing an anomaly correlation analysis between each common early warning point in the common early warning point set and the second operational interaction state anomaly. The third anomaly correlation analysis result is obtained by analyzing the correlation between each common early warning point in the common early warning point set and the third operational interaction state anomaly. By combining the results of the second and third anomaly correlation analyses and common early warning points, a fusion decision is made to generate safety early warnings and source tracing conclusions.

8. The charging pile status monitoring and safety early warning method based on image recognition according to claim 7, characterized in that, The step of obtaining the second anomaly correlation analysis result by associating each common early warning point in the common early warning point set with the second operational interaction state anomaly includes: Based on the time of the common early warning point, the real-time interactive power curve and communication data of the corresponding time and the time periods before and after are extracted from the second operation interaction state; Sliding window statistics are performed on the real-time interactive power curve to calculate the standard deviation and abrupt change of voltage and current fluctuations. Fluctuations or abrupt changes exceeding the first preset threshold are judged as abnormal voltage fluctuations or current jumps. The communication data is parsed, and the message loss rate and error codes are statistically analyzed. The loss rate exceeding the second preset threshold or the occurrence of critical error codes are judged as communication interruption or protocol-level error.

9. The charging pile status monitoring and safety early warning method based on image recognition according to claim 7, characterized in that, The step of obtaining the third anomaly correlation analysis result by associating each common early warning point in the common early warning point set with the third operational interaction state anomaly includes: Based on the time of the common early warning point, infrared thermal imaging video stream data and visible light video stream data of the corresponding time and the time periods before and after are extracted from the third operation interaction state. Threshold segmentation and connected component analysis are performed on the infrared image sequence formed by infrared thermal imaging video stream data. Regions whose temperature continuously exceeds the average temperature ΔT of the whole area and whose area is larger than the minimum area are identified as local abnormal hot spots. Visual detection algorithms based on inter-frame difference and feature matching are applied to visible light image sequences and infrared image sequences to identify and determine visible electric arcs, smoke, liquid leaks, or physical structure displacements.

10. A charging pile status monitoring and safety early warning system based on image recognition, used to implement the charging pile status monitoring and safety early warning method based on image recognition as described in any one of claims 1-9, characterized in that, include: The operation status perception module is used to acquire and classify the first, second and third operation interaction status information, and extract the long-term and single temperature rise rate sequences of the charging pile and the vehicle to determine the fixed reference segment and the random event segment. The feature deviation analysis module is used to align the real-time interactive power curve and calculate the feature deviation degree, and preliminarily determine the abnormality of the charging pile itself or specific charging events based on the deviation degree threshold. The visual evidence collaboration module is used to initiate infrared and visible light video stream analysis, identify local abnormal hot spots and traces of external physical interference, and fuse visual evidence to complete the attribution of charging pile status. The collaborative safety early warning module is used to extract common early warning points in the fourth operational interaction state, correlate and analyze anomalies in the second and third states, and generate safety early warnings and source tracing conclusions.

Images