A charging pile health diagnosis method, device, equipment and medium

Abstract

The application discloses a charging pile health diagnosis method and device, equipment and medium, and relates to the technical field of charging piles. The method comprises the following steps: acquiring first monitoring data and simulation data corresponding to a target charging pile; wherein the first monitoring data comprises electrical monitoring data, temperature monitoring data, acoustic vibration monitoring data and environmental monitoring data; the simulation data comprises simulated electrical harmonic characteristics, simulated temperature field distribution characteristics and simulated vibration frequency spectrum characteristics, and is obtained by synchronous simulation based on a multi-modal data twin model and real-time working condition parameters of the target charging pile; the multi-modal data twin model simulates electrical behavior, thermodynamic behavior and mechanical vibration behavior of the target charging pile; feature extraction and feature fusion are performed on the first monitoring data to obtain first feature data; feature alignment and residual calculation are performed based on the first feature data and the simulation data to obtain a residual sequence; and time series analysis and degradation analysis are performed based on the residual sequence to obtain a health diagnosis result corresponding to the target charging pile. The method can realize accurate diagnosis of the target charging pile.

Description

Technical Field

[0001] This application relates to the field of charging pile technology, and in particular to a method, device, equipment and medium for health diagnosis of charging piles. Background Technology

[0002] With the rapid development of the new energy vehicle industry, charging piles, as supporting infrastructure, have seen continuous expansion in deployment scale and coverage, and are now widely used in diverse scenarios such as public transportation stations, residential communities, and commercial parks. Currently, the charging demand for new energy vehicles is showing a trend towards higher frequency and higher power. Charging piles operate under high loads and complex environments (such as high temperature, high humidity, and dust) for extended periods, significantly increasing the risk of failures such as aging electrical components, thermal management failures, and loosening of mechanical structures. Therefore, conducting accurate and efficient health diagnostics of charging piles has become one of the important requirements in the field of new energy infrastructure.

[0003] To ensure the safe operation of charging stations, various health diagnostic solutions for charging stations have emerged in existing technologies. These solutions mostly focus on monitoring the operational data of the charging stations, collecting electrical parameters by deploying voltage and current sensors, or collecting temperature data of key components by using temperature sensors, and then making health assessments. However, these existing charging station health diagnostic technologies lack in-depth analysis of time-series data, ultimately leading to insufficient accuracy in the diagnostic results.

[0004] Therefore, there is an urgent need for a method that can accurately diagnose the health of charging piles. Summary of the Invention

[0005] This application provides a method, device, equipment, and medium for health diagnosis of charging piles, which can achieve accurate diagnosis of target charging piles.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a method for diagnosing the health of a charging pile, including: The process involves acquiring first monitoring data and simulation data corresponding to the target charging pile. The first monitoring data includes electrical monitoring data, temperature monitoring data, acoustic and vibration monitoring data, and environmental monitoring data. The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics, obtained through synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical, thermodynamic, and mechanical vibration behaviors of the target charging pile. Feature extraction and feature fusion are performed on the first monitoring data to obtain first feature data. Based on the first feature data and simulation data, feature alignment and residual calculation are performed to obtain a residual sequence. Based on the residual sequence, time-series analysis and degradation analysis are performed to obtain the health diagnosis results corresponding to the target charging pile.

[0007] In some possible implementations, feature extraction and feature fusion are performed on the first monitoring data to obtain first feature data, including: Feature extraction is performed on temperature monitoring data to obtain thermal gradient distribution characteristics and local hot spot spatial migration characteristics; feature extraction is performed on acoustic and vibration monitoring data to obtain key frequency band energy ratio characteristics and time-frequency domain anomalous resonance peak characteristics; feature extraction is performed on electrical monitoring data to obtain transient harmonic distortion characteristics and load fluctuation trend characteristics; based on environmental monitoring data, operating condition normalization compensation is performed on thermal gradient distribution characteristics, local hot spot spatial migration characteristics, key frequency band energy ratio characteristics, time-frequency domain anomalous resonance peak characteristics, transient harmonic distortion characteristics, and load fluctuation trend characteristics to obtain the first feature data.

[0008] In some possible implementations, feature alignment and residual calculation are performed based on the first feature data and simulation data to obtain a residual sequence, including: The first feature data and the simulation data are timestamped to obtain an aligned feature pair; based on the aligned feature pair, the first physical residual is calculated; wherein, the first physical residual includes at least one of thermal distribution consistency residual, vibration energy spectrum deviation residual and current waveform distortion residual; for each physical residual, they are arranged in chronological order to obtain a residual sequence.

[0009] In some possible implementations, time-series analysis and degradation analysis are performed based on the residual sequence to obtain the health diagnosis results corresponding to the target charging pile, including: Sliding window time series decomposition is performed on the residual series to extract long-term trend components and short-term fluctuation components; Based on the long-term trend component, a degradation trajectory of the charging pile is constructed; based on the short-term fluctuation component and the degradation trajectory of the charging pile, a health status assessment is performed to obtain the health diagnosis result corresponding to the target charging pile; the health diagnosis result includes at least one of the degradation type, health index and remaining service life prediction value.

[0010] In some possible implementations, the calculation process for the thermal distribution consistency residual is as follows: Based on each temperature monitoring data and the corresponding simulated temperature field distribution characteristics, the Laplace operator residual corresponding to that monitoring point is calculated; where each temperature monitoring data corresponds to a unique monitoring point; based on the heat conduction equation and the Laplace operator residual, the abnormal heat source estimation term is obtained; spatial integration and statistical aggregation within the time window are performed on the abnormal heat source estimation terms corresponding to all monitoring points to obtain the heat distribution consistency residual.

[0011] In some possible implementations, the process of determining the degradation trajectory of charging piles includes: Based on recursive least squares support vector machine, dynamic system identification is performed on long-term trend components to obtain nonlinear functions; based on nonlinear functions, the current state is fitted to the long-term trend components to obtain the current fitting result; based on the current fitting result, recursive prediction is performed to obtain the degradation trajectory of charging piles.

[0012] In some possible implementations, a health status assessment is performed based on short-term fluctuation components and the degradation trajectory of the charging pile to obtain the health diagnosis results corresponding to the target charging pile, including: The time-frequency characteristics of short-term fluctuation components are matched with a pre-defined database of typical fault transient characteristics. If a match is successful, a health diagnosis result corresponding to the target charging pile is generated based on the matched fault type. Otherwise, the rate of curvature change and trajectory acceleration of the charging pile's degradation trajectory are calculated. Based on the rate of curvature change and trajectory acceleration, the comprehensive instability index corresponding to the target charging pile is determined. Based on the comprehensive instability index, the degradation type, health index, and predicted remaining service life of the target charging pile are determined. Based on the degradation type, health index, and predicted remaining service life of the target charging pile, the health diagnosis result corresponding to the target charging pile is obtained.

[0013] Secondly, this application provides a charging pile health diagnostic device, comprising: The acquisition module is used to acquire the first monitoring data and simulation data corresponding to the target charging pile. The first monitoring data includes electrical monitoring data, temperature monitoring data, acoustic and vibration monitoring data, and environmental monitoring data. The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics. These are obtained by synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical behavior, thermodynamic behavior, and mechanical vibration behavior of the target charging pile. The fusion module is used to extract and fuse features from the first monitoring data to obtain the first feature data. The calculation module is used to perform feature alignment and residual calculation based on the first feature data and simulation data to obtain the residual sequence; The diagnostic module is used to perform time series analysis and degradation analysis based on the residual sequence to obtain the health diagnosis results corresponding to the target charging pile.

[0014] Thirdly, this application provides a computing device, including a memory and a processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of the first aspects.

[0015] Fourthly, this application provides a computer-readable storage medium for storing a computer program for performing the method as described in any one of the first aspects.

[0016] Fifthly, this application provides a computer program product comprising one or more computer instructions, wherein when the computer instructions are executed by a computer, the computer performs the method as described in any one of the first aspects.

[0017] As can be seen from the above technical solution, this application has at least the following beneficial effects: In this application, first monitoring data and simulation data corresponding to the target charging pile are obtained. The first monitoring data includes electrical monitoring data, temperature monitoring data, acoustic and vibration monitoring data, and environmental monitoring data. The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics. These are obtained through synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical, thermodynamic, and mechanical vibration behaviors of the target charging pile, covering the full-dimensional operating status of the charging pile in terms of electrical, temperature, acoustic and vibration, and environmental aspects, thus solving the problems of traditional diagnostic methods. The previous method, relying solely on single electrical data, suffered from missed fault detections. By introducing simulation data synchronized with real-time operating parameters, it addresses the technical challenge of fixed simulation models failing to match the real-time operating status of equipment. Feature extraction and fusion are performed on the first monitoring data to obtain first feature data, transforming high-dimensional raw data into low-dimensional feature vectors. This overcomes the limitations of single-dimensional features, enabling complementary verification of multi-physics information. The first feature data accurately reflects the health status of the charging pile, significantly reducing the computational load of subsequent residual calculations and time-series analysis. This improves the real-time performance and universality of the diagnostic algorithm, making it adaptable to various applications. This study investigates charging pile diagnostic scenarios under the same geographical and climatic conditions. Based on the first feature data and simulation data, feature alignment and residual calculation are performed to obtain a residual sequence. This sequence fully preserves the dynamic trends of equipment operation deviations (e.g., a continuous increase in residuals corresponds to progressive aging of components, and a sudden change in residuals corresponds to instantaneous failures). This provides a data source for subsequent time-series analysis and degradation trajectory construction, achieving an upgrade from static anomaly identification to dynamic trend analysis. It solves the problems of "black box" output and poor interpretability in traditional machine learning diagnostic methods. Based on the residual sequence, time-series analysis and degradation analysis are performed to obtain the health diagnosis results corresponding to the target charging pile. Accurate diagnostic results avoid interference from short-term fluctuations in the assessment of degradation trends, precisely identify the gradual degradation process of charging piles, provide early warnings for preventative maintenance, help maintenance personnel formulate accurate maintenance plans, reduce downtime losses, and adopt a dual-path strategy of short-term fluctuation fault matching and degradation trajectory trend assessment: short-term fluctuation components can quickly identify sudden faults (such as component loosening, grid impact), and degradation trajectory can predict long-term degradation trends (such as power device aging), achieving comprehensive coverage of instantaneous anomalies and gradual degradation, improving the integrity and reliability of diagnostic methods, and ultimately achieving accurate diagnosis of target charging piles.

[0018] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0019] Figure 1 This application provides an illustration of the application environment for a charging pile health diagnosis method. Figure 2 A flowchart illustrating a health diagnosis method for a charging pile provided in an embodiment of this application; Figure 3 A structural diagram of a charging pile health diagnosis device provided in an embodiment of this application; Figure 4 This is a schematic diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," and "third," etc., used in this application specification and accompanying drawings are used to distinguish different objects, not to limit a specific order.

[0021] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the related technologies is given first: To ensure the safe operation of charging stations, various health diagnosis solutions for charging stations have emerged in existing technologies. Most of these solutions focus on monitoring the operational data of the charging stations, collecting electrical parameters through voltage and current sensors, or collecting temperature data of key components through temperature sensors, and making health judgments based on the collected data from a single dimension or a few categories. Some advanced solutions introduce simple data analysis models, such as setting fixed thresholds to determine whether parameters exceed limits, or using traditional machine learning algorithms to train on historical fault data to achieve preliminary identification of fault types and make health diagnosis results. Other solutions attempt to use simulation technology to assist in diagnosis, but these are mostly simulations based on a single physical field (such as electrical or temperature-only) and do not achieve dynamic matching with actual operating conditions.

[0023] Existing charging pile health diagnostic technologies still suffer from significant shortcomings in diagnostic accuracy, failing to meet actual operation and maintenance needs. On the one hand, existing solutions often rely on single-dimensional or limited monitoring data, failing to comprehensively cover the multi-dimensional operating status of charging piles, including electrical, thermodynamic, mechanical vibration, and external environmental factors. This can easily lead to missed faults due to data gaps. On the other hand, single-physics simulations have poor adaptability to actual dynamic operating conditions, making it impossible to construct accurate ideal operating benchmarks. This results in difficulties in quantifying and distinguishing between parameter fluctuations caused by environmental interference and anomalies caused by equipment degradation. Furthermore, traditional analysis methods lack in-depth analysis of data time-series trends, failing to accurately capture the gradual degradation process of equipment. They can only identify faults after they occur, making it difficult to predict health status and remaining service life in advance. Ultimately, this leads to insufficient accuracy in diagnostic results, failing to provide reliable support for operation and maintenance decisions.

[0024] In view of this, this application provides a method for health diagnosis of charging piles. To make the technical solution of this application clearer and easier to understand, the application scenarios of the technical solution of this application will be described below with reference to the accompanying drawings. Figure 1 As shown in the figure, this figure is a schematic diagram of an application scenario provided by an embodiment of this application.

[0025] In this application scenario, terminal 102 collects the first monitoring data (electrical, temperature, acoustic vibration, and environmental monitoring data) and real-time operating parameters of the target charging pile in real time and uploads them to server 104 via the communication network. Simultaneously, it can send commands such as diagnostic start-up and diagnostic parameter configuration to server 104. After receiving the data and commands, server 104 sequentially executes steps such as multimodal data twin model simulation, feature extraction and fusion, feature alignment residual calculation, time series analysis, and degradation analysis to generate health diagnostic results including degradation type, health index, and remaining service life. Finally, server 104 feeds back the diagnostic results and abnormal warning information to terminal 102 in real time, where terminal 102 visualizes the results to facilitate timely follow-up work by maintenance personnel.

[0026] To make the technical solution of this application clearer and easier to understand, the following describes a charging pile health diagnosis method provided by an embodiment of this application, in conjunction with the above application scenarios. Figure 2 As shown, this figure is a flowchart of a charging pile health diagnosis method provided in an embodiment of this application. The charging pile health diagnosis method includes: S201. Obtain the first monitoring data and simulation data corresponding to the target charging pile.

[0027] The first monitoring data includes electrical monitoring data, temperature monitoring data, sound and vibration monitoring data, and environmental monitoring data; The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics. It is obtained by synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical behavior, thermodynamic behavior, and mechanical vibration behavior of the target charging pile.

[0028] Target charging piles refer to specific charging pile equipment that requires health status assessment and fault hazard investigation. They can include mainstream types such as DC fast charging piles and AC slow charging piles, and can be adapted to different application scenarios such as public charging stations, residential communities, and commercial parks. They are the physical carriers of the diagnostic process.

[0029] The first monitoring data refers to the data set that is directly collected through various sensors, monitoring terminals and other hardware devices, reflecting the actual operating status of the target charging pile and its environment. It is the basis for subsequent feature extraction and health diagnosis.

[0030] Electrical monitoring data is collected by electrical testing equipment such as voltage transformers, current transformers, and power analyzers. It is related to the power conversion and transmission of the charging pile and can include input voltage, output voltage, charging current, power factor, harmonic content, current ripple coefficient, and power device switching status parameters, which directly reflect the operational stability of the charging pile's electrical system.

[0031] Temperature monitoring data is collected by devices such as thermocouples, thermistors, and infrared temperature sensors. The temperature information of key components of the charging pile and its surrounding environment, such as the temperature of the charging module, the temperature of power devices such as IGBTs (Insulated Gate Bipolar Transistors), the temperature of cable joints, and the temperature of the internal cavity of the charging pile, is important data for judging the risk of thermal failure of the equipment.

[0032] The sound and vibration monitoring data is collected by sound sensors (microphones) and vibration acceleration sensors. The sound waves and vibration signals generated during the operation of the charging pile include fan noise, power module vibration frequency, cabinet resonance signal, abnormal vibration sound waves caused by loose parts, etc., which can indirectly reflect the health status of mechanical structure and moving parts.

[0033] Environmental monitoring data is collected by sensors such as temperature and humidity sensors, dust concentration sensors, rainfall sensors, and atmospheric pressure sensors. These data include parameters of the external environment where the target charging pile is located, such as ambient temperature and humidity, atmospheric pressure, dust content, rain and snow conditions, and ultraviolet radiation intensity. This data is used to correct for the impact of different environmental conditions on the equipment's operating status.

[0034] Simulation data refers to virtual data generated synchronously based on a multimodal data twin model after inputting the real-time operating parameters of the target charging pile. This data is used to form a benchmark for comparing the ideal and actual states with the first monitoring data (actual data).

[0035] The simulated electrical harmonic characteristics are output from the multimodal data twin model simulation. They represent the electrical harmonic parameters of the target charging pile under the ideal operating conditions, including the amplitude, phase, and total harmonic distortion rate of each harmonic, corresponding to the actual electrical monitoring data.

[0036] The simulated temperature field distribution characteristics are three-dimensional temperature field distribution data inside the charging pile generated by the multimodal data twin model simulation, including the ideal temperature values ​​of each important component, the temperature gradient distribution law, the predicted location of hot spots, etc., corresponding to the actual temperature monitoring data.

[0037] The simulated vibration spectrum characteristics are output from the multimodal data twin model simulation. They are the vibration signal spectrum parameters under ideal operating conditions of the equipment, including the dominant vibration frequency, energy distribution of each frequency band, theoretical position of resonance peaks, etc., corresponding to the actual acoustic and vibration monitoring data.

[0038] Multimodal data twin model refers to a virtual mapping model built based on digital twin technology that corresponds one-to-one with the target charging pile entity. It integrates the simulation logic of three major physical fields: electrical, thermodynamic, and mechanical vibration. Its value lies in achieving accurate simulation of the multi-dimensional operation behavior of the charging pile. By inputting real-time operating parameters, it can output simulation data that is synchronized with the operating status of the physical equipment, providing an ideal benchmark for judging anomalies in actual monitoring data.

[0039] Real-time operating parameters refer to the dynamic working status parameters of the target charging pile during operation. They are important inputs for driving the multimodal data twin model to perform synchronous simulation. These parameters include real-time charging power, charging current / voltage setpoints, equipment load rate, grid voltage fluctuation amplitude, and current charging stage (constant current / constant voltage / trickle charge). They directly determine the adaptability of the model simulation to operating conditions and the accuracy of the data.

[0040] Electrical behavior refers to the electrical characteristics of charging piles during the process of power conversion, transmission and charging services, including the transformation law of voltage and current, the switching characteristics of power devices, the generation and propagation law of harmonics, and the electrical response characteristics under load changes. It is the simulation object of the electrical sub-model in the multimodal data twin model.

[0041] Thermodynamic behavior refers to the generation, conduction, diffusion, and heat dissipation characteristics of heat during the operation of a charging pile. It includes the heating mechanism of power devices, the heat transfer efficiency between components, the working effect of the heat dissipation system (fan / heat sink), the distribution and dynamic change of the internal temperature field, etc. It is the simulation object of the thermodynamic sub-model in the model.

[0042] Mechanical vibration behavior refers to the vibration characteristics generated during the operation of a charging pile due to component movement, electromagnetic excitation, structural resonance, etc., including the vibration law of rotating components such as fans / transformers, electromagnetic vibration characteristics of power devices, resonance response of cabinet structure, and vibration anomalies caused by loose connectors, etc. It is the simulation object of the mechanical vibration sub-model in the model.

[0043] For example, to construct a dual data system of actual operating data and ideal simulation data, providing a high-quality data foundation for subsequent operations, voltage transformers, current transformers, and high-precision power analyzers can be deployed at key electrical nodes such as the input and output sides of the charging pile and the input / output terminals of the charging modules. The sampling frequency is set to no less than 1kHz to ensure accurate capture of high-frequency electrical characteristics such as transient harmonic distortion, thereby collecting electrical monitoring data. Simultaneously, operating condition information such as charging stage and load rate is recorded to support subsequent data correlation analysis. A "point-to-surface" monitoring scheme is adopted, with thermocouples attached to the surfaces of important power devices such as IGBTs and rectifier bridges, thermistors deployed at heat-prone areas such as cable joints and busbars, and infrared thermometers installed inside the charging pile to achieve full-area temperature scanning and collect temperature monitoring data. The period is set to 100ms to ensure the capture of dynamic temperature changes; vibration acceleration sensors are deployed near vibration sources (fans, transformers, charging modules) and key structural components of the cabinet; sound sensors are installed at the air outlet and air inlet of the charging pile to collect sound and vibration monitoring data, with sampling frequencies covering 20Hz-20kHz (sound signal) and 10Hz-1kHz (vibration signal) to ensure complete acquisition of the frequency domain characteristics of sound and vibration signals; an integrated environmental sensor (including temperature, humidity, dust, rainfall, and atmospheric pressure detection functions) is installed in an unobstructed location above the charging pile to collect environmental monitoring data, with a sampling period set to 1min, while also recording qualitative information such as weather conditions (sunny / rainy / snowy).

[0044] The raw data collected may contain anomalies such as noise, missing data, and drift, requiring preprocessing to ensure data quality. Specific measures may include: Noise removal involves using wavelet threshold denoising algorithm to process power grid interference noise in electrical monitoring data; using adaptive filtering algorithm to remove environmental background noise in acoustic and vibration signals; and using moving average filtering to process random noise in temperature data. Missing value completion refers to the use of time-series-based interpolation algorithms to fill in missing data caused by sensor failure or transmission delay. For example, linear interpolation is used for short-term missing data (≤5 sampling points), and predictive interpolation based on LSTM neural networks is used for long-term missing data (>5 sampling points) to ensure data continuity. Data calibration involves calibrating electrical monitoring data according to the accuracy level of the transformer; correcting temperature monitoring data for deviations by combining the sensor's calibration curve; and eliminating measurement deviations caused by the sensor's installation location for environmental monitoring data (such as avoiding excessively high temperatures caused by direct sunlight).

[0045] The generation of simulation data needs to be synchronized in real time with the actual operating status of the target charging pile. The charging pile's BMS (Battery Management System) interface and charging pile controller (PLC / MCU) data interface can be used to collect real-time operating parameters such as charging power, current / voltage setpoints, load rate, and charging stage. 5G / Industrial Ethernet is used to achieve low-latency data transmission (transmission delay ≤100ms), ensuring the synchronization of operating parameters with the actual operating status of the equipment, thus completing the real-time acquisition and transmission of operating parameters. Furthermore, the collected real-time operating parameters can be input into a multimodal data twin model to initiate synchronous simulation calculations (such as calculating the ideal electrical harmonic characteristics under the current operating condition based on circuit topology theory and power device characteristics, including the amplitude, phase, and total harmonic distortion rate of each harmonic; calculating the ideal temperature field distribution characteristics inside the charging pile based on the heat conduction equation and heat dissipation system simulation logic, outputting the ideal temperature values ​​and temperature gradients of each key monitoring point; and calculating the vibration spectrum characteristics under ideal operating conditions based on dynamic theory and component vibration characteristics, determining the ideal energy distribution and resonance peak positions of each frequency band, ensuring that the simulation step size is consistent with the sampling period of the first monitoring data, and ensuring synchronization in the time dimension), to achieve collaborative simulation of the multimodal data twin model.

[0046] To ensure the validity of simulation data, validity verification can be performed. For example, before output, the validity verification can be performed by comparing the monitoring data under historical normal operating conditions with the simulation data under the corresponding operating conditions and calculating the root mean square error. If the root mean square error is ≤5%, the simulation data is considered valid and can be output directly. If the root mean square error is >5%, the model parameters are corrected using the Bayesian regularization algorithm, and the simulation calculation is repeated until the accuracy requirements are met.

[0047] There may be a slight time difference between the acquisition of the first monitoring data and the generation of the simulation data. Timestamp alignment can be performed. For example, the timestamp of the simulation data can be fine-tuned based on the acquisition timestamp of the first monitoring data to ensure that at the same time, each type of actual monitoring data (such as the temperature data of a certain monitoring point) has a unique corresponding simulation data (such as the simulation temperature value of the monitoring point), forming a one-to-one "actual-simulation" data pair.

[0048] S202. Perform feature extraction and feature fusion on the first monitoring data to obtain the first feature data.

[0049] One feasible approach involves extracting features from temperature monitoring data to obtain thermal gradient distribution features and local hotspot spatial migration features; extracting features from acoustic and vibration monitoring data to obtain key frequency band energy ratio features and time-frequency domain anomalous resonance peak features; extracting features from electrical monitoring data to obtain transient harmonic distortion features and load fluctuation trend features; and based on environmental monitoring data, performing operating condition normalization compensation on the thermal gradient distribution features, local hotspot spatial migration features, key frequency band energy ratio features, time-frequency domain anomalous resonance peak features, transient harmonic distortion features, and load fluctuation trend features to obtain the first feature data.

[0050] Feature extraction refers to the process of mining and refining key information (i.e., features) that can characterize the operating status of charging piles from the original first monitoring data using specific algorithms. Original monitoring data often contains redundancy, noise, and other interference. Feature extraction can reduce the dimensionality of the data and retain the core information closely related to the health status of the charging piles, laying the foundation for subsequent diagnostic analysis.

[0051] Feature fusion refers to the process of integrating multiple features extracted from different types of monitoring data (temperature, sound and vibration, electrical), such as multi-type feature aggregation and environmental condition normalization compensation, to ultimately form a unified dimension of first feature data that can be directly used for subsequent analysis, thereby achieving complementarity and enhancement of multi-dimensional information and improving the comprehensiveness and accuracy of diagnosis.

[0052] The first feature data refers to the final feature set obtained after feature extraction and environmental condition normalization compensation. It is standardized data that integrates core information from multiple dimensions such as temperature, sound and vibration, and electrical components, and eliminates environmental interference. It can be directly used for subsequent diagnostic steps such as alignment with simulation data and residual calculation.

[0053] Thermal gradient distribution characteristics refer to the features extracted from temperature monitoring data that characterize the spatial variation of temperature inside the charging pile and its key components. By calculating the temperature difference and temperature change rate between different monitoring points, the conduction and diffusion of heat inside the charging pile can be reflected. If there is an abnormal thermal gradient, it may indicate a failure of the heat dissipation system or abnormal heating of components.

[0054] The spatial migration characteristics of local hot spots refer to the dynamic changes in the spatial location of local high-temperature areas (hot spots) during the operation of charging piles. By tracking the location, range, and temperature peak changes of hot spots at different time points, the development trend of heat-generating faults can be determined. For example, aging of power devices may cause hot spots to gradually expand and migrate to surrounding components.

[0055] The key frequency band energy ratio feature is a feature extracted from the frequency domain analysis results of acoustic and vibration monitoring data. It refers to the ratio of vibration / sound wave energy within a specific key frequency bandwidth to the overall frequency band energy. The normal operation and fault status of different components (such as fans, transformers, and power modules) correspond to specific frequency band energy distributions. This feature can be used to quickly identify abnormal energy accumulation in acoustic and vibration signals and determine the mechanical status of components.

[0056] The time-frequency domain abnormal resonance peak feature refers to the resonance peak feature that deviates from the normal range in the time-frequency domain analysis of the acoustic vibration signal; when the charging pile is running normally, the position and amplitude of the resonance peak of the acoustic vibration signal are relatively stable; when the components are loose, worn, or have resonance faults, new resonance peaks or abnormal amplitude and position of the original resonance peaks will appear. This feature can accurately capture such mechanical vibration anomalies.

[0057] Transient harmonic distortion features refer to the characteristics extracted from electrical monitoring data that characterize the degree of instantaneous harmonic distortion of the electrical signal of the charging pile. Transient harmonics are generated in scenarios such as power device switching and load changes in the electrical system of the charging pile. If the degree of harmonic distortion exceeds the normal range, it may indicate electrical problems such as aging of power devices and failure of filter circuit. This feature focuses on capturing instantaneous harmonic anomalies.

[0058] Load fluctuation trend characteristics refer to the trend characteristics of load changes over time extracted from load-related parameters (such as charging current and power) in electrical monitoring data. By analyzing the amplitude, frequency, and stability of load fluctuations, the adaptability of the charging pile's electrical system to load changes can be reflected. Abnormal load fluctuations may be related to unstable power supply, charging module failure, etc.

[0059] Operating condition normalization compensation refers to the process of standardizing and correcting extracted temperature, sound and vibration, and electrical features based on environmental monitoring data. Different environmental conditions (such as high temperature, high humidity, and high dust) can interfere with the operating status of charging piles, causing "pseudo-anomalies" in the feature data. Normalization compensation can eliminate the influence of environmental interference and make the feature data more accurately reflect the health status of the charging pile itself.

[0060] For example, by accurately extracting features by category and integrating them with environmental adaptability normalization, important information in various dimensions is mined. Then, by using operating condition normalization to eliminate environmental interference, a unified first feature data is finally formed. That is, for the three types of operating data—temperature, sound and vibration, and electrical—specific features are extracted using algorithms adapted to their data characteristics to ensure that no key information in each dimension is missed. Then, based on environmental monitoring data, operating condition normalization compensation is applied to all extracted features to eliminate the interference of different environmental conditions (such as high temperature and high humidity) on the features. Finally, the data is integrated to form the first feature data that can truly reflect the health status of the charging pile itself.

[0061] For example, the value of temperature monitoring data lies in reflecting the thermal operating status of charging piles. The goal of extraction is to capture the temperature distribution pattern and dynamic changes of hot spots. The specific implementation is divided into two steps: First, thermal gradient distribution feature extraction. The temperature data of discrete monitoring points is transformed into a continuous temperature field distribution through spatial interpolation algorithms. Then, the temperature gradient (i.e., the rate of temperature change) between any two points in the temperature field is calculated, and parameters such as the maximum gradient, average gradient, and gradient distribution variance are statistically analyzed to form thermal gradient distribution features. If the gradient value of a certain area suddenly increases, it may indicate that the heat dissipation channel in that area is blocked or the component is abnormally overheating. Second, local hot spot spatial migration feature extraction. The hot spot area is determined at each time point using a threshold segmentation algorithm (usually, the area 10°C above the normal operating temperature threshold is set as the hot spot). By calculating parameters such as the center coordinate offset of the hot spot, the rate of change of the hot spot area, and the trend of the peak temperature change at different time stamps, local hot spot spatial migration features are formed. For example, if the hot spot migrates from the power module to the surrounding cables, it may indicate that the power module is aging more rapidly and affecting the surrounding components.

[0062] Acoustic and vibration monitoring data includes two types of signals: sound waves and vibrations. Both have obvious time-frequency characteristics, requiring feature extraction through time-frequency domain analysis. The goal is to capture abnormal energy distribution and resonance states. One key aspect is the extraction of energy ratio features in critical frequency bands. First, Fourier transform is used to convert the time-domain acoustic and vibration signals into frequency-domain signals. Then, combined with the inherent vibration frequencies of the charging pile's core components (fan, transformer, power module), corresponding critical frequency bands are defined (e.g., fan operation corresponds to the 20-50Hz band, transformer operation corresponds to the 100-200Hz band). Finally, the ratio of energy within each critical frequency band to the total energy of the frequency band is calculated. The first method involves identifying key frequency band energy ratio characteristics. A sudden increase in the energy ratio of a key frequency band may indicate a mechanical failure in the corresponding component (such as wear of the fan bearing). The second method involves extracting abnormal resonance peak features in the time-frequency domain. Wavelet transform algorithm is used to perform time-frequency joint analysis on the acoustic vibration signal to obtain a time-frequency matrix. Resonance peaks in the time-frequency matrix are identified by peak detection algorithm. By comparing the position and amplitude of resonance peaks under normal operating conditions, parameters such as amplitude, frequency, and duration of abnormal resonance peaks that deviate from the normal range are extracted to form abnormal resonance peak features in the time-frequency domain. For example, the appearance of a new resonance peak may indicate structural resonance caused by component loosening.

[0063] Electrical monitoring data directly reflects the power conversion and transmission status of charging piles. The extraction objective is to capture harmonic distortion and load stability. Specifically, the implementation involves: 1) Transient harmonic distortion feature extraction: For transient harmonics generated during charging, such as power device switching and load abrupt changes, Fast Fourier Transform is used to perform harmonic analysis on instantaneous electrical signals (voltage / current), calculating the amplitude and phase of each harmonic, as well as parameters such as total harmonic distortion rate and single harmonic distortion rate. The focus is on extracting the peak value and rate of change of harmonic distortion under transient conditions (such as charging start-up and load switching) to form transient harmonic distortion features. 2) Load fluctuation trend feature extraction: Based on time-series charging power and current data, a sliding window method is used to divide the data into segments. Parameters such as the load mean, fluctuation amplitude, and fluctuation frequency within each window are calculated. Then, through linear fitting and trend line slope calculation, the long-term trend of load variation is extracted to form load fluctuation trend features. If the load fluctuation amplitude suddenly increases and the trend line slope is negative, it may indicate that the charging module output is unstable.

[0064] Furthermore, features extracted from temperature, acoustic vibration, and electrical data are susceptible to false anomalies due to external environmental conditions (e.g., in high-temperature environments, the normal operating temperature characteristics of charging piles will be higher than in normal-temperature environments, which can easily be misjudged as a fault if directly compared). Therefore, it is necessary to perform operating condition normalization compensation based on environmental monitoring data to standardize the features and ultimately achieve fusion. For example, analyzing the correlation between environmental monitoring data (ambient temperature, humidity, dust concentration, atmospheric pressure, etc.) and each extracted feature can help identify important influencing factors: for example, ambient temperature mainly affects the thermal gradient distribution characteristics and the spatial migration characteristics of local hot spots; ambient humidity mainly affects electrical insulation performance, thereby indirectly affecting transient harmonic distortion characteristics; dust concentration indirectly affects temperature-related features and acoustic vibration-related features by affecting heat dissipation efficiency (dust accumulation leads to increased fan load and changes in acoustic vibration characteristics). By using correlation analysis (such as Pearson correlation coefficient calculation), the dominant environmental impact factors of each feature can be screened out. Then, based on the screened environmental impact factors, a normalized compensation model can be established: using standard environmental conditions (such as ambient temperature 25℃, humidity 50%, and no dust) as a benchmark, a feature correction coefficient library under different environmental factors can be established through experimental calibration or machine learning algorithms. For example, regarding the thermal gradient distribution characteristics, if the actual ambient temperature is 35℃ (higher than the standard 25℃), the corresponding compensation coefficient is queried from the correction coefficient library to linearly correct the extracted thermal gradient characteristics, eliminating the feature shift caused by the increase in ambient temperature. Similarly, for the other five types of characteristics, such as the spatial migration characteristics of local hotspots and the energy ratio characteristics of key frequency bands, corresponding corrections are made according to their dominant environmental impact factors. Finally, after completing the normalization compensation of all characteristics, the six types of standardized characteristics (thermal gradient distribution characteristics, spatial migration characteristics of local hotspots, energy ratio characteristics of key frequency bands, time-frequency domain abnormal resonance peak characteristics, transient harmonic distortion characteristics, and load fluctuation trend characteristics) are dimensionally integrated to form a unified feature vector. At the same time, redundancy is removed from the integrated features (such as removing features with too small variance and low information contribution through variance analysis), finally obtaining the first feature data with reasonable dimensions, complete information, and no environmental interference.

[0065] S203. Based on the first feature data and simulation data, feature alignment and residual calculation are performed to obtain the residual sequence.

[0066] One possible approach is to align the first feature data with the simulation data using timestamps to obtain aligned feature pairs; based on the aligned feature pairs, calculate the first physical residual; wherein the first physical residual includes at least one of thermal distribution consistency residual, vibration energy spectrum deviation residual, and current waveform distortion residual; for each physical residual, arrange them in chronological order to obtain a residual sequence.

[0067] Optionally, the calculation process for the thermal distribution consistency residual is as follows: Based on each temperature monitoring data and the corresponding simulated temperature field distribution characteristics, the Laplace operator residual corresponding to that monitoring point is calculated; where each temperature monitoring data corresponds to a unique monitoring point; based on the heat conduction equation and the Laplace operator residual, the abnormal heat source estimation term is obtained; spatial integration and statistical aggregation within the time window are performed on the abnormal heat source estimation terms corresponding to all monitoring points to obtain the heat distribution consistency residual.

[0068] Feature alignment refers to the process of matching and calibrating the first feature data (actual features) with the simulation data (ideal features) according to specific rules. The core of this step is timestamp alignment, which ensures that the actual features and ideal features with the same time node and the same physical meaning form an accurate correspondence, thereby eliminating the time dimension deviation interference for subsequent residual calculation.

[0069] Residual calculation refers to the process of quantifying the difference between the aligned first feature data (actual feature) and the simulation data (ideal feature). By calculating the deviation value between the two, the degree of deviation between the actual operating state and the ideal operating state of the charging pile is measured. The larger the deviation value, the worse the health status of the equipment is usually.

[0070] The residual sequence refers to the time-series data sequence formed by arranging the calculated physical residuals in order of their corresponding timestamps. This sequence completely preserves the trend of residual changes over time and serves as the data source for subsequent time-series analysis and degradation trajectory construction.

[0071] Timestamp alignment is a method of feature alignment. It refers to using the timestamp of one type of data (usually the acquisition timestamp of the first feature data) as a reference to fine-tune and calibrate the timestamp of the other type of data (simulation data) to ensure that at the same time, each actual feature has a unique corresponding ideal feature, forming an actual-ideal feature pair.

[0072] The aligned feature pairs are one-to-one feature combinations formed after timestamp alignment. Each feature pair contains an actual feature (from the first feature data) and an ideal feature with the same physical meaning and time node (from the simulation data), such as "actual thermal gradient distribution feature - simulated temperature field distribution feature" and "actual key frequency band energy ratio feature - simulated vibration spectrum feature".

[0073] The first physical residual refers to the deviation between the actual and ideal characteristics, which is quantitatively calculated from the physical mechanism level based on the aligned feature pairs. It is directly related to the electrical, thermodynamic, mechanical vibration and other physical behaviors of the charging pile. Specifically, it includes three types: heat distribution consistency residual, vibration energy spectrum deviation residual, and current waveform distortion residual, which can accurately reflect the abnormal operation of each physical field of the equipment.

[0074] Thermal distribution consistency residual is an important type of first physical residual. It is a residual parameter that specifically quantifies the deviation between the actual temperature distribution and the ideal temperature field distribution of the charging pile. It is calculated by combining the physical laws of heat conduction and can accurately identify temperature distribution inconsistencies caused by heat dissipation failures, abnormal component heating, etc.

[0075] Vibration energy spectrum deviation residual is a residual parameter that quantifies the deviation between the key frequency band energy characteristics extracted from actual acoustic and vibration monitoring data and the simulated vibration spectrum characteristics. It can reflect the deviation between the operating state of the charging pile's mechanical structure (such as fans and transformers) and the ideal state, and help identify faults such as mechanical wear and loosening.

[0076] Current waveform distortion residual is a residual parameter that quantifies the deviation between the transient harmonic distortion characteristics extracted from actual electrical monitoring data and the simulated electrical harmonic characteristics. It is directly related to the operational stability of the charging pile's electrical system and can accurately capture current waveform abnormalities caused by power device aging, filter circuit failures, etc.

[0077] The Laplace operator residual is an intermediate parameter for calculating the thermal distribution consistency residual. It refers to the temperature field curvature deviation value calculated by the Laplace operator based on the actual temperature data of a single monitoring point and the corresponding ideal simulated temperature field distribution characteristics. It can reflect the degree of anomaly in the local temperature distribution of the monitoring point.

[0078] The heat conduction equation is a physical equation describing the conduction of heat within a substance; its expression is: Where T is temperature and t is time. Where is the thermal diffusivity, For the Laplace operator, This represents the intensity of the internal heat source. In the calculation of heat distribution consistency residuals, it is used to infer anomalous heat source information by combining the Laplace operator residuals.

[0079] The abnormal heat source estimation term is derived based on the heat conduction equation and the Laplace operator residual. It is a parameter that characterizes the intensity and range of abnormal heat sources around the monitoring point and can quantitatively determine whether there is abnormal heating in the area that exceeds the ideal state.

[0080] Spatial integration and statistical aggregation within a time window are the final aggregation calculation methods for thermal distribution consistency residuals. Spatial integration refers to performing spatial dimension integration on the abnormal heat source estimates of all monitoring points to obtain the global abnormal heat source distribution of the entire charging pile. Statistical aggregation within a time window refers to performing statistical analysis (such as calculating the mean and peak value) on the spatial integration results within a set time window (such as 5 minutes) to finally obtain the thermal distribution consistency residual within that time window.

[0081] For example, timestamp alignment ensures spatiotemporal matching between actual and ideal features. Then, multi-dimensional residuals are calculated based on physical laws, and finally integrated into a residual sequence in chronological order. This involves first achieving spatiotemporal synchronization through timestamp alignment, then calculating residuals for aligned feature pairs from the perspectives of electrical, thermodynamic, and mechanical vibration mechanisms, and finally sorting the residuals by time to form a sequence, fully preserving the temporal trend of the deviation. For instance, the timestamp formats of the two types of data are first standardized. The timestamp of the first feature data comes from the acquisition time of the original monitoring data (synchronously generated by the sensor terminal), while the timestamp of the simulation data comes from the trigger time of the model simulation calculation (driven by the reception time of real-time operating parameters). Both need to be uniformly converted to a format with UTC time and millisecond-level precision to avoid alignment deviations caused by differences in time formats.

[0082] Furthermore, the timestamp of the first feature data can be used as a benchmark (because it is directly related to the actual operating status of the equipment, and its time accuracy is more valuable for reference). Time difference statistical analysis can be used to detect the time deviation between the two types of data. Specifically, a continuous time series (such as 10 minutes) is selected, and the timestamp difference between the simulation data and the first feature data under the same feature dimension (such as temperature-related features) is calculated. If the difference is ≤100ms (matching the data sampling period), the time deviation is considered negligible. If the difference is >100ms, the timestamp of the simulation data is fine-tuned and calibrated using linear interpolation. For example, if the timestamp of the simulation data lags by 200ms, the entire simulation data segment is shifted forward by 200ms to ensure that it matches the timestamp of the first feature data.

[0083] After timestamp alignment, feature dimension matching is performed based on the principle of physical consistency. This involves matching the thermal gradient distribution and local hotspot spatial migration features in the first feature data with the simulated temperature field distribution features in the simulation data; matching the key frequency band energy ratio features and time-frequency domain anomalous resonance peak features with the simulated vibration spectrum features; and matching the transient harmonic distortion features and load fluctuation trend features with the simulated electrical harmonic features. Each successfully matched "actual feature - ideal feature" pair forms an aligned feature pair, ultimately resulting in a set of multiple parallel feature pairs covering all dimensions of electrical, temperature, and acoustic vibration.

[0084] The core of residual calculation is based on physical mechanisms, quantifying the degree of deviation of each aligned feature pair, focusing on three core physical residuals: thermal distribution consistency residual, vibration energy spectrum deviation residual, and current waveform distortion residual. Specifically, the thermal distribution consistency residual focuses on the consistency of the temperature field distribution, achieving precise quantification by combining the physical laws of heat conduction, avoiding the limitations of simple temperature difference calculation. The specific steps are: First, calculate the Laplace operator residual for a single monitoring point. For each temperature monitoring point, extract its actual temperature data Treal and the ideal temperature Tsim at that location in the simulated temperature field distribution characteristics of the corresponding timestamp, and then calculate the residual using the Laplace operator. Calculate the curvature deviation of the temperature field, i.e., the Laplacian operator residual. ²(Treal-Tsim), the larger this value, the more significant the deviation between the local temperature distribution at the monitoring point and the ideal state; the second step is to derive the estimation term for the abnormal heat source. Substitute the Laplace operator residual into the heat conduction equation. The abnormal heat source intensity qest (i.e., the abnormal heat source estimation term) is obtained by back-reasoning through equation transformation. qest > 0 indicates the presence of an abnormal heat source in the area, while qest = 0 indicates that the temperature distribution conforms to the ideal state. The third step is to obtain the residual through global aggregation. Spatial integration is performed on the abnormal heat source estimation terms of all monitoring points to obtain the total abnormal heat source within the entire charging pile. Then, a fixed time window is set (e.g., 5 minutes, which can be adjusted according to diagnostic accuracy requirements), and the spatial integration results within the time window are statistically aggregated to calculate the mean or peak value of the total abnormal heat source within the time window, which serves as the heat distribution consistency residual corresponding to that time window.

[0085] The calculation of the vibration energy spectrum deviation residual can be approached by using an energy deviation quantification method for the feature pair of "critical frequency band energy ratio characteristics - simulated vibration spectrum characteristics". First, extract the actual critical frequency band energy ratio Ereal and the ideal critical frequency band energy ratio Esim from the feature pair. Then, calculate the relative deviation between the two, i.e., vibration energy spectrum deviation residual = |Ereal-Esim| / Esim×100%. Finally, take the weighted average of the deviation residuals of multiple critical frequency bands (such as the frequency bands corresponding to fans and transformers) within the same time window to obtain the final vibration energy spectrum deviation residual for that time window. The weights are set according to the degree of fault impact of the corresponding components in each frequency band (e.g., the frequency band corresponding to transformers has a higher weight than that of fans).

[0086] The calculation of current waveform distortion residuals can focus on the distortion deviation of the current waveform by targeting the "transient harmonic distortion characteristics - simulated electrical harmonic characteristics" feature pair. That is, extracting the actual transient harmonic distortion rate THDreal and the ideal transient harmonic distortion rate THDsim. The residual is calculated by combining absolute deviation and relative deviation, i.e., current waveform distortion residual = |THDreal - THDsim| + (|THDreal - THDsim| / THDsim) × 50%. This formula considers both the magnitude of the absolute deviation and the influence of the relative deviation, avoiding misjudgment of deviation due to the ideal distortion rate being too small. Finally, the maximum residual value within the time window is taken as the current waveform distortion residual of that time window, highlighting the influence of transient anomalies.

[0087] Finally, a residual sequence is generated. After calculating the first physical residual within each time window, the residuals are integrated and sorted according to the chronological order of the time windows to form a residual sequence. For example, each calculated residual (thermal distribution consistency residual, vibration energy spectrum deviation residual, and current waveform distortion residual) is first marked with the starting timestamp of the corresponding time window. Then, based on the timestamp, the three types of residuals are arranged in chronological order to form three parallel single-dimensional residual sequences (such as the thermal distribution consistency residual sequence and the vibration energy spectrum deviation residual sequence). If subsequent analysis needs to be simplified, the three single-dimensional residual sequences can be weighted and fused to obtain a comprehensive residual sequence (the weights are set according to the contribution of each residual to fault diagnosis). The final generated residual sequence fully preserves the changing trends of different dimensional deviations over time. For example, a continuous increase in the thermal distribution consistency residual sequence may indicate an accelerated degradation of the heat dissipation system.

[0088] S204. Based on the residual sequence, perform time series analysis and degradation analysis to obtain the health diagnosis results corresponding to the target charging pile.

[0089] One possible approach is to perform a sliding window time series decomposition on the residual sequence to extract long-term trend components and short-term fluctuation components; construct a charging pile degradation trajectory based on the long-term trend components; and conduct a health status assessment based on the short-term fluctuation components and the charging pile degradation trajectory to obtain the health diagnosis result corresponding to the target charging pile. The health diagnosis result includes at least one of degradation type, health index, and remaining service life prediction value.

[0090] Optionally, the process for determining the degradation trajectory of charging piles includes: Based on recursive least squares support vector machine, dynamic system identification is performed on long-term trend components to obtain nonlinear functions; based on nonlinear functions, the current state is fitted to the long-term trend components to obtain the current fitting result; based on the current fitting result, recursive prediction is performed to obtain the degradation trajectory of charging piles.

[0091] For example, the time-frequency characteristics of short-term fluctuation components are matched with a preset database of typical fault transient characteristics. If the match is successful, a health diagnosis result corresponding to the target charging pile is generated based on the matched fault type. Otherwise, the rate of curvature change and trajectory acceleration of the charging pile's degradation trajectory are calculated. Based on the rate of curvature change and trajectory acceleration, the comprehensive instability index corresponding to the target charging pile is determined. Based on the comprehensive instability index, the degradation type, health index, and predicted remaining service life of the target charging pile are determined. Based on the degradation type, health index, and predicted remaining service life of the target charging pile, the health diagnosis result corresponding to the target charging pile is obtained.

[0092] Among them, time series analysis refers to the process of decomposing, extracting and analyzing the time series data of residual sequence. The goal is to explore the changing patterns of residuals from the time dimension, distinguish between long-term trends and short-term fluctuations, and provide a basis for subsequent judgment of charging pile degradation trends and identification of instantaneous anomalies.

[0093] Degradation analysis refers to the process of judging the trend of equipment health deterioration (degradation trajectory), identifying degradation types, and quantitatively assessing health level and remaining service life based on time series analysis results, combined with the operating mechanism and fault patterns of charging piles. It is the analytical step for achieving health diagnosis.

[0094] Health diagnosis results refer to the final conclusions reflecting the health status of the target charging pile, obtained through time series analysis and degradation analysis. These conclusions include at least one of the following: degradation type, health index, and predicted remaining service life, providing a direct basis for operation and maintenance decisions.

[0095] Sliding window time series decomposition is an important technique in time series analysis. It involves setting a fixed-length sliding window to divide the residual sequence into multiple continuous subsequence segments. Signal decomposition algorithms (such as wavelet decomposition and empirical mode decomposition) are then used to decompose the subsequences within each window, ultimately extracting the long-term trend components and local short-term fluctuation components that run through the entire sequence.

[0096] The long-term trend component is one of the core components obtained after the residual sequence is decomposed by sliding window. It reflects the long-term and slow trend of the residual over time and is not affected by short-term random fluctuations. This component directly corresponds to the gradual degradation process of the charging pile's health status (such as component aging and performance degradation) and is an important basis for constructing the degradation trajectory.

[0097] The short-term fluctuation component is another core component obtained after decomposing the residual sequence. It reflects the random fluctuation or instantaneous change of the residual in the short term. It mainly corresponds to the instantaneous interference (such as grid voltage fluctuation and environmental change) or early fault transient signals during the operation of charging piles. It is an important basis for identifying sudden anomalies and early faults.

[0098] The charging pile degradation trajectory refers to the gradual change path of the charging pile's health status from normal to degradation and even failure, constructed based on long-term trend components. It is presented as a curve with time as the horizontal axis and quantitative indicators of health status (such as residual trend values) as the vertical axis. This trajectory clearly shows the speed, stage and trend of equipment degradation and is an important basis for assessing the remaining service life.

[0099] Recursive least squares support vector machine (RSM) is a machine learning algorithm that combines recursive learning and nonlinear fitting capabilities, making it an important and viable algorithm for constructing degenerate trajectories. By dynamically updating model parameters and systematically identifying long-term trend components, it can accurately fit the nonlinear changes in long-term trends, thereby improving the accuracy of degenerate trajectory construction.

[0100] Dynamic system identification refers to the process of analyzing the long-term trend component, the "input data," through recursive least squares support vector machines, to uncover the underlying nonlinear change patterns, and finally output a nonlinear function that can accurately describe the trend change, thereby realizing the mathematical modeling of the "dynamic system" of charging pile degradation.

[0101] Nonlinear functions are the output of dynamic system identification. They are mathematical functions (such as polynomial functions and Gaussian kernel functions) that can accurately fit the changes in long-term trend components. These functions quantify the nonlinear relationship between time and the long-term trend value of the residuals, providing mathematical support for fitting the current state and predicting future trends.

[0102] Current state fitting refers to the process of using a nonlinear function identified by a dynamic system to fit existing long-term trend component data, resulting in a fitting curve that closely matches the actual trend. This process can verify the fitting accuracy of the nonlinear function and provide a reliable basis for subsequent recursive prediction.

[0103] Recursive prediction refers to the process of progressively predicting the residual trend value of future time nodes based on the existing long-term trend component data and the nonlinear function after fitting and verification. Through continuous recursive calculation, the trend prediction results for a period of time in the future can be obtained, thereby constructing a complete charging pile degradation trajectory.

[0104] Time-frequency characteristics refer to feature parameters extracted from short-term fluctuation components that contain information in both time and frequency dimensions (such as instantaneous frequency, fluctuation amplitude, frequency band energy ratio, etc.). The time-frequency characteristics of short-term fluctuation components can accurately depict the changing patterns of instantaneous signals and are an important basis for matching transient characteristics of faults.

[0105] The typical fault transient feature library refers to a pre-built database containing the transient time-frequency features corresponding to various typical faults of charging piles (such as power module aging, fan bearing wear, filter capacitor failure, etc.). This database is built based on a large number of historical fault cases and experimental data, providing a reference benchmark for fault matching of short-term fluctuation components.

[0106] The rate of change of trajectory curvature describes the rate at which the curvature of the charging pile's degradation trajectory curve changes over time, and is an important indicator for quantifying the rate of degradation. The larger the rate of change of curvature, the faster the health of the charging pile deteriorates. If the rate of change of curvature suddenly increases, it may indicate that the equipment has entered a rapid degradation stage.

[0107] Trajectory acceleration refers to the rate of change of the slope of the degradation trajectory curve over time, i.e., the acceleration of the degradation rate. Positive acceleration indicates that the degradation rate is accelerating, negative acceleration indicates that the degradation rate is slowing down, and zero acceleration indicates that the degradation rate is stable. It can accurately reflect the dynamic characteristics of the degradation process.

[0108] The comprehensive instability index is a comprehensive quantitative index obtained by weighted fusion calculation based on the trajectory curvature change rate and trajectory acceleration. It is used to comprehensively assess the degree of instability of the health status of charging piles. This index integrates dual information of degradation speed and speed change trend, which can more accurately determine the degradation stage and risk level of the equipment.

[0109] Degradation type is an important part of the health diagnosis results. It refers to the specific type of deterioration in the health status of the charging pile as determined by the analysis results, such as electrical system degradation (power device aging), thermodynamic system degradation (decreased heat dissipation efficiency), mechanical system degradation (fan failure), etc., which provides direction for targeted operation and maintenance.

[0110] The health index is an indicator used to quantitatively assess the current health status of a charging pile. Its value range is usually 0-1 (0 represents complete failure, and 1 represents complete health). It is calculated by comprehensively considering parameters such as instability indicators and degradation trajectory locations, and intuitively reflects the health level of the equipment.

[0111] The remaining service life prediction value refers to the remaining operating time required for a charging pile to go from its current state to a complete failure state, which is obtained based on the degradation trajectory prediction. It is calculated by setting a failure threshold (such as residual limit value, health index critical value) and combining it with the predicted trend of the degradation trajectory, providing a time reference for the formulation of operation and maintenance plans.

[0112] It should be noted that the residual sequence includes two key pieces of information: first, long-term, slow changes reflecting the gradual deterioration of the charging pile's health status, such as the continuous increase in residuals due to component aging; and second, short-term, fluctuating information reflecting instantaneous interference or early faults, such as instantaneous peak values ​​of residuals caused by sudden grid changes. Direct analysis of the original residual sequence is susceptible to short-term fluctuations and cannot accurately capture the true degradation trend. Therefore, we can first separate the two types of information through time-series decomposition to obtain the long-term trend component and the short-term fluctuation component; then, based on the long-term trend component, we can construct a degradation trajectory to accurately characterize the long-term degradation pattern of the equipment; finally, through dual-path fusion analysis of short-term fluctuation fault matching and degradation trajectory trend assessment, we can achieve a comprehensive assessment of the health status, enabling rapid identification of sudden faults and accurate prediction of long-term degradation trends, ultimately outputting diagnostic results including degradation type, health index, and remaining service life.

[0113] For example, in order to distinguish between long-term degradation and short-term fluctuations, the time series pattern is decomposed; then, a degradation trajectory is constructed to characterize the long-term deterioration trend; finally, the health status is evaluated by fusion analysis and the diagnostic results are output. That is, by using a sliding window time series decomposition to separate the different change characteristics of the residuals, a degradation trajectory is constructed based on the long-term trend, and then the health status is comprehensively evaluated by combining short-term fluctuations and the degradation trajectory.

[0114] For example, based on the operating characteristics and diagnostic accuracy requirements of the charging pile, key parameters of the sliding window are set: window length and sliding step size. The window length needs to balance trend capture and real-time performance, and is typically set to cover 1-2 typical operating cycles (e.g., 30 minutes to 1 hour). The sliding step size is set to 1 / 5 to 1 / 3 of the window length (e.g., a step size of 6-10 minutes when the window length is 30 minutes), ensuring continuous coverage of the entire residual sequence without generating excessive redundant calculations. For example, for a residual sequence sampled every 100ms, the window length is set to 3600 data points (corresponding to 1 hour), and the sliding step size is set to 720 data points (corresponding to 12 minutes).

[0115] Following a defined sliding window, the entire residual sequence is divided into multiple consecutive overlapping subsequence segments. For each subsequence segment, either Empirical Mode Decomposition (EMD) or Wavelet Packet Decomposition (WPD) is used for temporal decomposition. EMD adaptively decomposes the subsequence into multiple intrinsic mode functions (EMFs) and a residual component, where the residual component represents the local long-term trend within the window, and each EEM component represents local short-term fluctuations. Wavelet Packet Decomposition, on the other hand, uses multi-scale decomposition to divide the subsequence into different frequency bands, with low-frequency components corresponding to long-term trends and high-frequency components corresponding to short-term fluctuations. Complementary verification of these two algorithms improves the accuracy of component extraction.

[0116] The local long-term trend components and short-term fluctuation components obtained from all sliding window decompositions are integrated: the local long-term trend components of each window are concatenated in chronological order, and abrupt changes at the window junctions are eliminated through smoothing (such as moving average) to obtain the global long-term trend component that runs through the entire residual sequence; similarly, the short-term fluctuation components of each window are concatenated and integrated to obtain the global short-term fluctuation component. Finally, two parallel component sequences are formed, which are used for subsequent degradation trajectory construction and fault transient identification, respectively.

[0117] Degradation analysis is divided into two stages: degradation trajectory construction and health status assessment. The degradation trajectory construction can be improved by using a recursive least squares support vector machine as an option. The health status assessment adopts a dual-path strategy of prioritizing fault matching and using trend assessment as a fallback. The specific implementation is as follows: (1) Construction of charging pile degradation trajectory Based on the long-term trend components of the residual sequence, the accurate fitting and future prediction of the nonlinear trend are achieved through Recursive Least Squares - Support Vector Machine (RLS-SVM), ultimately forming a complete path of health state degradation.

[0118] The construction of the degradation trajectory of charging piles includes three steps: dynamic system identification, current state fitting, recursive prediction and trajectory generation. In the dynamic system identification, the RLS-SVM algorithm is used to identify the long-term trend components, explore the nonlinear mapping relationship between time and trend value, and output a nonlinear function that can accurately describe the degradation trend.

[0119] Dynamic system identification includes an input data preprocessing process, specifically: Let the time series data of the long-term trend component be the dataset. ,in, The timestamp of the i-th data point; is the long-term trend component value corresponding to the i-th timestamp (i.e., the long-term trend value of the residual); N is the total amount of data for the long-term trend component.

[0120] To improve the model fitting accuracy, the input data is standardized. The timestamps and trend values ​​are normalized to the [0, 1] interval. The standardization formula is as follows:

[0121] in, The minimum value of the timestamp; The maximum value of the timestamp; It represents the minimum value of the long-term trend component; This represents the maximum value of the long-term trend component; The standardized timestamp of the i-th data point in the long-term trend component; Let be the standardized degradation trend value of the i-th data point.

[0122] In dynamic system identification, firstly, considering that the degradation trend of charging piles often exhibits nonlinear characteristics (such as accelerated degradation of power device aging and gradual performance degradation of the heat dissipation system), a Gaussian kernel function is selected as the kernel function of RLS-SVM (mapping linearly inseparable data in low-dimensional space to high-dimensional feature space, making it linearly separable in high-dimensional space, thus enabling support vector machines to handle nonlinear problems). Its expression is:

[0123] in, The output of the kernel function represents the similarity between two input vectors; The timestamp to be fitted / predicted (after standardization); The standardized timestamp of the i-th data point in the long-term trend component; The kernel width parameter is determined through 5-fold cross-validation, with a value ranging from 0.1 to 10 to ensure the model's adaptability to nonlinear trends.

[0124] To balance fitting accuracy and generalization ability, and to avoid overfitting or underfitting, the goal is to find an optimal set of model parameters. b and (Among them, the optimal w and b define the linear regression hyperplane in the high-dimensional feature space, which can accurately capture the nonlinear trend of charging pile degradation; the optimal This reflects the model's fitting error for each training sample and is used to evaluate the model's fitting quality. The objective function can be defined as:

[0125] in, For the constraint weight vector; is the slack variable (i.e., fitting error) for the i-th data point (i.e., the sample); C is the regularization parameter (determined through cross-validation, with a value range of 1 to 100; the larger C is, the more emphasis is placed on fitting accuracy, and the smaller C is, the more emphasis is placed on generalization ability); N is the total amount of data for the long-term trend component; b is the bias term. For regularization terms; This represents the fitting error term.

[0126] In solving for the optimal model parameters b and During the process, constraints must be met; these constraints are set based on the linear regression relationship in high-dimensional space, and can be specifically expressed as follows:

[0127] in, Let i be the standardized degenerate value of the i-th data point; Kernel function mapping maps the input temporal features to a high-dimensional feature space; This represents a linear regression relationship in a high-dimensional space.

[0128] To transform a constrained optimization problem into an unconstrained one, Lagrange multipliers can be introduced for each constraint. The objective Lagrangian function (which includes the weighted sum of the original objective function and all constraints) can be expressed as:

[0129] right Taking the partial derivatives and setting them equal to 0, we obtain the optimality conditions, which are respectively: , and Furthermore, Substituting into the linear prediction formula in high-dimensional space, we get:

[0130] Using the definition of Gaussian kernel function Replacing the dot product, we get:

[0131] Substituting the specific form of the Gaussian kernel, we finally obtain the nonlinear function:

[0132] Fitting the current state to the long-term trend component based on a nonlinear function is an example of using a nonlinear function. This involves fitting the current state to the long-term trend components. Specifically, it involves standardizing all historical time points. Substitute the nonlinear function into the value, and calculate the fitting degradation value at each time point. These fitted degradation values ​​and the true degradation values The resulting set is the current fitting result.

[0133] After obtaining the fitting results for the current state, a complete degradation trajectory of the charging pile is generated through recursive prediction. Let the historical time point be... The corresponding fitting degradation value is The trained nonlinear prediction function can be expressed as:

[0134] The core idea of ​​recursive prediction is to feed the prediction result of each step back to the model as a new observation, thereby continuously advancing the prediction of future moments. The specific process is as follows: First, take the current last historical moment. Calculate the time point of the next moment. ( (where is the time step), substituting it into the nonlinear function yields the predicted degradation value at that moment, which can be expressed as:

[0135] Then, the new time-degradation value pair Add historical datasets and update the model's input; then... Starting from, calculate And then call the nonlinear function again to obtain This process is repeated, and the prediction at step k is:

[0136] in, It is the first The updated Lagrange multipliers and bias terms are then used. By continuously feeding new predictions into the model, we can obtain degradation predictions for a series of future time points. These predicted values, when connected in chronological order, form a complete degradation trajectory of the charging pile, visually demonstrating the non-linear degradation trend of the equipment in the future.

[0137] By combining short-term fluctuation components and degradation trajectories, a comprehensive assessment is achieved through a strategy that prioritizes fault matching and uses trend assessment as a fallback. This is specifically divided into two scenarios: Scenario 1: Short-term fluctuation component fault matching. First, time-frequency analysis is performed on the short-term fluctuation component to extract time-frequency features such as instantaneous frequency, peak fluctuation amplitude, and energy proportion in a specific frequency band. The extracted time-frequency features are then matched with a pre-defined database of typical fault transient features (using cosine similarity or Euclidean distance to calculate the matching degree). If the matching degree is ≥85% (the threshold can be adjusted according to diagnostic accuracy), the match is considered successful. The degradation type is directly determined based on the matched fault type (e.g., if the "fan bearing wear" feature is matched, the degradation type is mechanical system degradation). Then, the health index is adjusted based on the influence weight corresponding to the fault type (e.g., if the normal health index is 1.0, this fault corresponds to a deduction of 0.3, and the current health index is 0.7). Based on the historical degradation rate data of the fault type, the remaining service life prediction is corrected (e.g., if the average remaining service life of this fault is 6 months, it is adjusted to 5 months based on the current degradation trajectory), finally generating a health diagnosis result.

[0138] Scenario 2: Degradation Trajectory Trend Assessment (When Fault Matching Fails). When the time-frequency characteristics of short-term fluctuation components match the fault feature database by less than 85%, it is determined that there is no clear sudden fault. The health status needs to be assessed through the trend characteristics of the degradation trajectory: First, calculate the trajectory curvature change rate and trajectory acceleration. Take the second derivative of the degradation trajectory curve to obtain the curvature change rate (characterizing the change in the curvature of the curve); take the first derivative of the slope of the trajectory curve to obtain the trajectory acceleration (characterizing the change in the degradation speed); Second, calculate the comprehensive instability index. According to the operating priority of the charging pile, assign weights to the curvature change rate and trajectory acceleration (e.g., the electrical system has a higher degradation risk, corresponding to a weight of 0.6; the mechanical system corresponds to 0.4). Calculate the comprehensive instability index by weighted summation. The formula is: Comprehensive Instability Index = ω1 × Curvature Change Rate + ω2 × Trajectory Acceleration (ω1 + ω2 = 1); Third, determine the diagnostic results. The system presets threshold values ​​for the comprehensive instability index (e.g., 0-0.3 for mild instability, 0.3-0.7 for moderate instability, and 0.7-1.0 for severe instability). It then determines the degradation type based on these threshold values ​​(e.g., mild instability corresponds to "slow aging degradation," and severe instability corresponds to "rapid failure degradation"). Based on the mapping relationship between the comprehensive instability index and the health index (e.g., mild instability corresponds to a health index of 0.7-1.0, moderate instability to 0.4-0.7, and severe instability to 0-0.4), it calculates the health index. Finally, it calculates the predicted remaining service life by finding the intersection of the predicted degradation trajectory value and the failure threshold (e.g., the trend value corresponding to a health index of 0.1). For example, if the comprehensive instability index is 0.6 (moderate instability), the degradation type is "moderate aging degradation," the health index is 0.5, and the predicted remaining service life is 8 months.

[0139] Based on the above, the charging pile health diagnosis method acquires the first monitoring data and simulation data corresponding to the target charging pile. The first monitoring data includes electrical monitoring data, temperature monitoring data, acoustic and vibration monitoring data, and environmental monitoring data. The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics, which are obtained through synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical, thermodynamic, and mechanical vibration behaviors of the target charging pile, covering the full-dimensional operation of the charging pile in terms of electrical, temperature, acoustic and vibration, and environmental aspects. This approach addresses the problem of missed fault detection caused by traditional diagnostic methods relying solely on single electrical data. By introducing simulation data generated synchronously with real-time operating parameters, it overcomes the technical challenge of fixed simulation models failing to match the real-time operating status of equipment. Feature extraction and fusion are performed on the first monitoring data to obtain first feature data, transforming high-dimensional raw data into low-dimensional feature vectors. This overcomes the limitations of single-dimensional features, enabling complementary verification of multi-physics information. The first feature data accurately reflects the health status of the charging pile, significantly reducing the computational load of subsequent residual calculations and time-series analysis, and improving the real-time performance and versatility of the diagnostic algorithm. This system is adaptable to charging pile diagnostic scenarios under different geographical and climatic conditions. Based on the first feature data and simulation data, feature alignment and residual calculation are performed to obtain a residual sequence, which fully preserves the dynamic trend of equipment operation deviations (e.g., a continuous increase in residuals corresponds to progressive aging of components, and a sudden change in residuals corresponds to instantaneous failures). This provides a data source for subsequent time series analysis and degradation trajectory construction, realizing an upgrade from static anomaly identification to dynamic trend analysis. It solves the problems of "black box" output and poor interpretability of traditional machine learning diagnostic methods. Based on the residual sequence, time series analysis and degradation analysis are performed to obtain the health diagnosis results corresponding to the target charging pile. This approach yields accurate diagnostic results, avoids interference from short-term fluctuations in assessing degradation trends, and precisely identifies the gradual degradation process of charging piles. It provides early warnings for preventative maintenance, helps maintenance personnel develop precise maintenance plans, reduces downtime losses, and employs a dual-path strategy of short-term fluctuation fault matching and degradation trajectory trend assessment: short-term fluctuation components can quickly identify sudden faults (such as component loosening or grid impact), while degradation trajectories can predict long-term degradation trends (such as power device aging). This achieves comprehensive coverage of instantaneous anomalies and gradual degradation, improves the completeness and reliability of diagnostic methods, and ultimately enables accurate diagnosis of target charging piles.

[0140] The above text combined Figures 1 to 2 The charging pile health diagnosis method provided in the embodiments of this application has been described in detail. The apparatus and equipment provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0141] This application also provides a charging pile health diagnosis device, such as... Figure 3 As shown in the figure, this is a schematic diagram of a charging pile health diagnosis device provided in an embodiment of this application. The device includes: The acquisition module 301 is used to acquire the first monitoring data and simulation data corresponding to the target charging pile. The first monitoring data includes electrical monitoring data, temperature monitoring data, sound and vibration monitoring data, and environmental monitoring data. The simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics. It is obtained by synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile. The multimodal data twin model simulates the electrical behavior, thermodynamic behavior, and mechanical vibration behavior of the target charging pile. The fusion module 302 is used to extract and fuse features from the first monitoring data to obtain the first feature data; The calculation module 303 is used to perform feature alignment and residual calculation based on the first feature data and simulation data to obtain the residual sequence; The diagnostic module 304 is used to perform time series analysis and degradation analysis based on the residual sequence to obtain the health diagnosis results corresponding to the target charging pile.

[0142] In some possible implementations, the fusion module 302 is specifically used for: Feature extraction is performed on temperature monitoring data to obtain thermal gradient distribution characteristics and local hot spot spatial migration characteristics; feature extraction is performed on acoustic and vibration monitoring data to obtain key frequency band energy ratio characteristics and time-frequency domain anomalous resonance peak characteristics; feature extraction is performed on electrical monitoring data to obtain transient harmonic distortion characteristics and load fluctuation trend characteristics; based on environmental monitoring data, operating condition normalization compensation is performed on thermal gradient distribution characteristics, local hot spot spatial migration characteristics, key frequency band energy ratio characteristics, time-frequency domain anomalous resonance peak characteristics, transient harmonic distortion characteristics, and load fluctuation trend characteristics to obtain the first feature data.

[0143] In some possible implementations, the calculation module 303 is specifically used for: The first feature data and the simulation data are timestamped to obtain an aligned feature pair; based on the aligned feature pair, the first physical residual is calculated; wherein, the first physical residual includes at least one of thermal distribution consistency residual, vibration energy spectrum deviation residual and current waveform distortion residual; for each physical residual, they are arranged in chronological order to obtain a residual sequence.

[0144] In some possible implementations, diagnostic module 304 is specifically used for: The residual sequence is decomposed by sliding window to extract long-term trend components and short-term fluctuation components. Based on the long-term trend components, the degradation trajectory of the charging pile is constructed. Based on the short-term fluctuation components and the degradation trajectory of the charging pile, a health status assessment is performed to obtain the health diagnosis result corresponding to the target charging pile. The health diagnosis result includes at least one of the following: degradation type, health index, and predicted remaining service life.

[0145] In some possible implementations, the calculation module 303 is specifically used for: Based on each temperature monitoring data and the corresponding simulated temperature field distribution characteristics, the Laplace operator residual corresponding to that monitoring point is calculated; where each temperature monitoring data corresponds to a unique monitoring point; based on the heat conduction equation and the Laplace operator residual, the abnormal heat source estimation term is obtained; spatial integration and statistical aggregation within the time window are performed on the abnormal heat source estimation terms corresponding to all monitoring points to obtain the heat distribution consistency residual.

[0146] In some possible implementations, diagnostic module 304 is specifically used for: Based on recursive least squares support vector machine, dynamic system identification is performed on long-term trend components to obtain nonlinear functions; based on nonlinear functions, the current state is fitted to the long-term trend components to obtain the current fitting result; based on the current fitting result, recursive prediction is performed to obtain the degradation trajectory of charging piles.

[0147] In some possible implementations, diagnostic module 304 is specifically used for: The time-frequency characteristics of short-term fluctuation components are matched with a pre-defined database of typical fault transient characteristics. If a match is successful, a health diagnosis result corresponding to the target charging pile is generated based on the matched fault type. Otherwise, the rate of curvature change and trajectory acceleration of the charging pile's degradation trajectory are calculated. Based on the rate of curvature change and trajectory acceleration, the comprehensive instability index corresponding to the target charging pile is determined. Based on the comprehensive instability index, the degradation type, health index, and predicted remaining service life of the target charging pile are determined. Based on the degradation type, health index, and predicted remaining service life of the target charging pile, the health diagnosis result corresponding to the target charging pile is obtained.

[0148] The charging pile health diagnosis device according to the embodiments of this application can correspondingly execute the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the charging pile health diagnosis device are respectively for implementing Figure 2 For the sake of brevity, the corresponding processes of each method in the illustrated embodiments will not be described in detail here.

[0149] This application also provides a computing device. For example... Figure 4As shown in the figure, this is a schematic diagram of a computing device provided in an embodiment of this application. The computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0150] Bus 401 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0151] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0152] Communication interface 403 is used for communication with external devices.

[0153] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0154] The memory 404 stores executable code, and the processor 402 executes the executable code to perform the aforementioned charging pile health diagnosis method.

[0155] Specifically, in achieving Figure 3 In the case of the illustrated embodiment, and Figure 3 When the modules or units of the charging pile health diagnosis device described in the embodiment are implemented through software, the following steps are performed: Figure 3The software or program code required for the functions of each module / unit can be partially or wholly stored in the memory 404. The processor 402 executes the program code corresponding to each unit stored in the memory 404 to perform the aforementioned charging pile health diagnosis method.

[0156] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the aforementioned charging pile health diagnosis method.

[0157] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0158] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0159] When the computer program product is executed by a computer, the computer performs any of the aforementioned methods of the charging pile health diagnosis method. The computer program product can be a software installation package; when any of the aforementioned methods of the charging pile health diagnosis method is required, the computer program product can be downloaded and executed on the computer.

[0160] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0161] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A method for health diagnosis of charging piles, characterized in that, The method includes: Acquire first monitoring data and simulation data corresponding to the target charging pile; wherein, the first monitoring data includes electrical monitoring data, temperature monitoring data, acoustic and vibration monitoring data, and environmental monitoring data; the simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics, which are obtained by synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile, wherein the multimodal data twin model simulates the electrical behavior, thermodynamic behavior, and mechanical vibration behavior of the target charging pile; The first monitoring data is subjected to feature extraction and feature fusion to obtain the first feature data; Based on the first feature data and simulation data, feature alignment and residual calculation are performed to obtain the residual sequence; Based on the residual sequence, time series analysis and degradation analysis are performed to obtain the health diagnosis results corresponding to the target charging pile.

2. The method according to claim 1, characterized in that, The step of extracting and fusing features from the first monitoring data to obtain first feature data includes: Feature extraction is performed on the temperature monitoring data to obtain thermal gradient distribution features and local hotspot spatial migration features; Feature extraction was performed on the acoustic and vibration monitoring data to obtain key frequency band energy ratio features and time-frequency domain anomalous resonance peak features; Feature extraction is performed on the electrical monitoring data to obtain transient harmonic distortion characteristics and load fluctuation trend characteristics; Based on the environmental monitoring data, the operating condition normalization compensation is performed on the thermal gradient distribution characteristics, local hot spot spatial migration characteristics, key frequency band energy ratio characteristics, time-frequency domain abnormal resonance peak characteristics, transient harmonic distortion characteristics, and load fluctuation trend characteristics to obtain the first feature data.

3. The method according to claim 1, characterized in that, The step of performing feature alignment and residual calculation based on the first feature data and simulation data to obtain a residual sequence includes: The first feature data is timestamped and aligned with the simulation data to obtain the aligned feature pair. Based on the aligned feature pairs, a first physical residual is calculated; wherein, the first physical residual includes at least one of thermal distribution consistency residual, vibration energy spectrum deviation residual, and current waveform distortion residual; For each physical residual, arrange them in chronological order to obtain the residual sequence.

4. The method according to claim 1, characterized in that, Based on the residual sequence, time-series analysis and degradation analysis are performed to obtain the health diagnosis results corresponding to the target charging pile, including: The residual sequence is subjected to sliding window time series decomposition to extract long-term trend components and short-term fluctuation components; Based on the aforementioned long-term trend components, a charging pile degradation trajectory is constructed; Based on the short-term fluctuation components and the degradation trajectory of the charging pile, a health status assessment is performed to obtain the health diagnosis result corresponding to the target charging pile; the health diagnosis result includes at least one of degradation type, health index and remaining service life prediction value.

5. The method according to claim 3, characterized in that, The calculation process for the heat distribution consistency residual is as follows: Based on each temperature monitoring data and the corresponding simulated temperature field distribution characteristics, the Laplace operator residual corresponding to that monitoring point is calculated; where each temperature monitoring data corresponds to a unique monitoring point. Based on the heat conduction equation and the Laplace operator residual, the estimation term for the abnormal heat source is obtained; Spatial integration and statistical aggregation within a time window are performed on the estimated abnormal heat sources corresponding to all monitoring points to obtain the heat distribution consistency residual.

6. The method according to claim 4, characterized in that, The process of determining the degradation trajectory of the charging pile includes: Based on recursive least squares support vector machine, dynamic system identification is performed on the long-term trend components to obtain nonlinear functions; Based on a nonlinear function, the current state is fitted to the long-term trend component to obtain the current fitting result. Based on the current fitting results, recursive prediction is performed to obtain the degradation trajectory of the charging pile.

7. The method according to claim 4, characterized in that, The health status assessment based on the short-term fluctuation components and the charging pile degradation trajectory is used to obtain the health diagnosis result corresponding to the target charging pile, including: The time-frequency characteristics of the short-term fluctuation components are matched with a preset library of typical fault transient characteristics. If a match is successful, a health diagnosis result corresponding to the target charging pile is generated based on the matched fault type. Otherwise, calculate the rate of curvature change and trajectory acceleration of the charging pile's degradation trajectory; Based on the curvature change rate and trajectory acceleration, the comprehensive instability index corresponding to the target charging pile is determined; Based on the comprehensive instability index, the degradation type, health index and predicted remaining service life of the target charging pile are determined. Based on the degradation type, health index, and predicted remaining service life of the target charging pile, the health diagnosis result of the target charging pile is obtained.

8. A health diagnostic device for charging piles, characterized in that, The device includes: The acquisition module is used to acquire the first monitoring data and simulation data corresponding to the target charging pile; wherein, the first monitoring data includes electrical monitoring data, temperature monitoring data, sound and vibration monitoring data, and environmental monitoring data; the simulation data includes simulated electrical harmonic characteristics, simulated temperature field distribution characteristics, and simulated vibration spectrum characteristics, which are obtained by synchronous simulation based on a multimodal data twin model and the real-time operating parameters of the target charging pile, wherein the multimodal data twin model simulates the electrical behavior, thermodynamic behavior, and mechanical vibration behavior of the target charging pile; The fusion module is used to extract and fuse features from the first monitoring data to obtain first feature data; The calculation module is used to perform feature alignment and residual calculation based on the first feature data and simulation data to obtain a residual sequence; The diagnostic module is used to perform time series analysis and degradation analysis based on the residual sequence to obtain the health diagnosis result corresponding to the target charging pile.

9. A computing device, characterized in that, Including memory and processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method as described in any one of claims 1 to 7.

Images