Intelligent charging pile based on dynamic fault diagnosis and control method thereof

Abstract

The invention relates to the technical field of intelligent charging equipment control, and provides an intelligent charging pile based on dynamic fault diagnosis and a control method thereof. According to the method, parameter information is collected in real time through a built-in sensor A preset fault type is acquired, a fault prediction model is established by using a machine learning algorithm, the fault prediction model is used for identifying and classifying preset abnormal states based on the parameter information, and the abnormal states at least comprise overvoltage, undervoltage, overcurrent, electric leakage and relay adhesion; obtaining fault root cause information according to the identification and classification result of the abnormal state, and generating a stage processing scheme corresponding to the fault root cause information; when the intelligent charging pile is used for charging, the vehicle model and the battery parameters are analyzed through a charging interface communication protocol, the corresponding optimal charging current is obtained, and the charging curve is dynamically adjusted according to the optimal charging current, the real-time power grid voltage fluctuation and the grading processing scheme. Millimeter-level precision monitoring of the plug insertion state is achieved, and the charging safety level of the charging pile is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent charging equipment control, and in particular to an intelligent charging pile based on dynamic fault diagnosis and a control method thereof. BACKGROUND

[0002] The fault diagnosis and charging control method of the existing intelligent charging pile has the following technical bottlenecks: 1. Limited fault diagnosis capability: the traditional scheme relies on fixed threshold to judge abnormal state (such as overvoltage, overcurrent), which cannot effectively handle the time sequence correlation of parameter sequence and complex fault mode, resulting in high false alarm rate or ambiguous fault root cause positioning; 2. Rigid processing strategy: there is no grading response mechanism for faults of different severity, and the "one-size-fits-all" power-off processing is often used, which may affect the continuity of normal charging service and cannot meet the rapid disposal needs of emergency faults; 3. Coarse charging curve adjustment: the charging current setting is mainly based on the static parameters of the vehicle battery, without real-time combination of power grid voltage fluctuation and dynamic optimization of fault processing needs, which easily leads to low charging efficiency or equipment overload; 4. Insufficient safety design: the monitoring accuracy of the charging plug insertion state in standby state is insufficient, and the hardware interlocking mechanism mainly uses single contact detection, which has the safety hazard of false start charging when not fully inserted.

[0003] Therefore, a method is needed to solve at least one of the above problems. SUMMARY

[0004] The present application provides an intelligent charging pile based on dynamic fault diagnosis and a control method thereof, aiming to solve the problems of limited fault diagnosis capability, rigid processing strategy, coarse charging curve adjustment and insufficient safety design in the fault diagnosis and charging control method of the existing intelligent charging pile.

[0005] In a first aspect, the embodiments of the present application provide a control method based on dynamic fault diagnosis, which comprises: real-time acquisition of parameter information by built-in sensors, wherein the parameter information at least includes input voltage, output current, ground state and relay contact temperature parameter; obtaining a preset fault type, establishing a fault prediction model by using a machine learning algorithm, and the fault prediction model is used for identifying and classifying the preset abnormal state based on the parameter information, and the abnormal state at least includes overvoltage, undervoltage, overcurrent, electric leakage and relay sticking; obtaining fault root cause information according to the identification and classification results of the abnormal state, and generating a grading processing scheme corresponding to the fault root cause information; When charging at a smart charging station, the vehicle model and battery parameters are parsed through the charging interface communication protocol to obtain the corresponding optimal charging current. The charging curve is then dynamically adjusted based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme. In standby mode, a low-power sensor monitors the plug insertion action, and a hardware interlock design ensures that charging does not start if the plug is not fully inserted.

[0006] In some embodiments, the real-time acquisition of parameter information through built-in sensors includes: periodically and synchronously acquiring input voltage, output current, grounding resistance value and relay contact temperature through voltage sensor, current sensor, grounding status detection module and temperature sensor respectively; performing digital filtering preprocessing on the acquired raw data to remove high-frequency noise and abnormal jump values, generating a standardized parameter sequence and storing it in a real-time database.

[0007] In some embodiments, the fault types include overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; the step of obtaining the preset fault types and establishing a fault prediction model using machine learning algorithms includes: obtaining the feature vectors corresponding to overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; constructing a training dataset using historical fault data; extracting time-series feature information from time-series parameter information using a long short-term memory network algorithm; optimizing feature weights by combining an attention mechanism; training to form a multi-class fault prediction model; and ensuring the model's generalization ability through cross-validation.

[0008] In some embodiments, the fault prediction model is used to identify and classify preset abnormal states based on parameter information, including: inputting parameter information into the fault prediction model, generating a multi-dimensional feature vector through the feature extraction layer of the fault prediction model; performing probability calculation on the feature vector using the classification layer of the fault prediction model, and outputting the confidence value corresponding to the abnormal state; when the confidence value exceeds a preset threshold, determining that the corresponding abnormal state has occurred and outputting the classification result.

[0009] In some embodiments, obtaining fault root cause information based on the identification and classification results of abnormal states includes: establishing a knowledge base for mapping relationships between abnormal states and fault root causes, wherein the knowledge base contains potential fault points and triggering conditions corresponding to abnormal states; traversing the knowledge base through a rule engine to match corresponding fault root cause combinations based on the identified abnormal states; and combining fault root cause combinations, real-time parameter fluctuation trends, and historical fault records, using evidence theory to perform credibility fusion on multiple possible root causes to determine the fault root cause with the highest priority.

[0010] In some embodiments, the hierarchical processing scheme corresponding to the generation of fault root cause information includes: presetting a fault severity classification standard to divide the fault root cause into a first-level fault root cause, a second-level fault root cause, and a third-level fault root cause; generating a processing instruction to immediately cut off the charging circuit and report a fault code for a first-level fault root cause; generating a processing instruction to reduce the charging power and continuously monitor for a second-level fault root cause; and generating a processing instruction to provide an interface pop-up prompt and record a log for a third-level fault root cause.

[0011] In some embodiments, the step of parsing the vehicle model and battery parameters through the charging interface communication protocol to obtain the corresponding optimal charging current includes: receiving a charging communication protocol data frame sent by the vehicle; parsing the vehicle identification code, battery type, rated voltage, and remaining capacity in the charging communication protocol data frame; and querying a pre-stored charging parameter mapping table based on the vehicle model, battery type, rated voltage, and remaining capacity to obtain the corresponding optimal charging current.

[0012] In some embodiments, the dynamic adjustment of the charging curve based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme includes: using an adaptive PID control algorithm, taking the optimal charging current as a reference value, and collecting the effective value of the grid voltage in real time to calculate the fluctuation coefficient; when the fluctuation coefficient is detected to exceed the rated range, adjusting the charging current according to a preset proportional coefficient; if there is a fault graded processing instruction, superimposing and correcting the current charging current according to the instruction priority to generate a dynamically adjusted charging current sequence.

[0013] In some embodiments, the step of monitoring plug insertion action through a low-power sensor and ensuring that charging is not started when the plug is not fully inserted in the standby state includes: monitoring the insertion displacement signal of the charging plug in real time through a Hall sensor or micro switch; triggering the position detection module in the hardware interlock circuit when the plug insertion action is detected; determining whether the plug is fully inserted through the position detection contact of the mechanical locking mechanism; if no position signal is detected, controlling the charging circuit relay to remain in the open state and prohibiting the sending of the charging start signal.

[0014] Secondly, this application provides a smart charging pile, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method provided in any embodiment of this application when executing the computer program.

[0015] This application constructs a fault prediction model to deeply extract the temporal features of abnormal states such as overvoltage, undervoltage, overcurrent, leakage, and relay sticking. Compared with the traditional threshold method, the fault identification accuracy is improved, and potential fault trends can be predicted in advance. Based on the fault severity classification (emergency / warning / notification), differentiated handling schemes are generated, reducing the response time for emergency faults and dynamically adjusting the charging power to maintain service for non-emergency faults, thereby improving system robustness. By analyzing vehicle battery parameters to match the optimal charging current and combining it with grid voltage fluctuations and fault handling instructions, the charging curve is adjusted in real time, improving charging efficiency and reducing the risk of charging interruption caused by grid fluctuations. The combination of low-power sensors and hardware interlocking mechanisms (Hall sensor + mechanical locking position detection) achieves millimeter-level precision monitoring of plug insertion status, reducing the probability of relay malfunction in the charging circuit when not fully inserted, and significantly improving the charging safety level.

[0016] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart illustrating the steps of a control method based on dynamic fault diagnosis provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a smart charging pile provided in one embodiment of this application; Figure 3 This is a schematic block diagram of a control device based on dynamic fault diagnosis provided in an embodiment of this application; Figure 4 This is a schematic block diagram of the structure of a smart charging pile provided in one embodiment of this application.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0022] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0023] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0026] Existing fault diagnosis and charging control methods for smart charging piles suffer from the following technical bottlenecks: 1. Limitations in fault diagnosis capabilities: Traditional solutions rely on fixed thresholds to determine abnormal states (such as overvoltage and overcurrent), which cannot effectively handle the temporal correlation of parameter sequences and complex fault modes, resulting in a high false alarm rate or unclear fault root cause location. 2. Rigid handling strategy: There is a lack of graded response mechanism for faults of different severity levels. The "one-size-fits-all" power outage is often used, which may affect the continuity of normal charging services and cannot meet the needs of rapid handling of emergency faults. 3. Coarse adjustment of charging curve: The charging current setting is mostly based on the static parameters of the vehicle battery, without dynamic optimization in real time in combination with grid voltage fluctuations and fault handling needs, which can easily lead to low charging efficiency or equipment overload. 4. Inadequate safety design: The monitoring accuracy of the charging plug insertion status in standby mode is insufficient. The hardware interlocking mechanism mostly uses single contact detection, which poses a safety hazard of accidentally starting charging when the plug is not fully inserted.

[0027] Therefore, a method is urgently needed to solve at least one of the above problems.

[0028] To resolve the above issues, please refer to... Figure 1 This application provides a control method based on dynamic fault diagnosis, applicable to, for example... Figure 2 The smart charging pile shown is an example. It should also be noted that all information involved in the method provided in this application was extracted with the authorization of the relevant user and in accordance with relevant regulations, and will not infringe on user privacy.

[0029] The provided control method based on dynamic fault diagnosis includes steps S101 to S104. Details are as follows: Step S101. Collect parameter information in real time through built-in sensors. The parameter information includes at least the input voltage, output current, grounding status, and relay contact temperature parameters.

[0030] Specifically, the built-in sensors acquire key operational status parameters of the smart charging pile in real time, covering dimensions such as electrical performance, grounding safety, and hardware temperature, providing raw data support for subsequent fault diagnosis.

[0031] High-precision Hall effect sensors or shunts are connected in series or parallel to the charging circuit to monitor the AC input voltage (e.g., 220V / 380V) and DC output current in real time (accuracy up to ±0.5%). A grounding resistance detection module (e.g., bridge method or constant current source method) measures the grounding impedance in real time, with a threshold set to ≤4Ω to ensure the reliability of the grounding system. NTC thermistors or infrared temperature sensors are embedded near the relay contacts to monitor contact temperature rise (accuracy ±1℃) and prevent overheating faults caused by poor contact.

[0032] A high-speed ADC (such as a 16-bit Σ-Δ ADC) is used to sample analog signals at a sampling frequency ≥100Hz to ensure the capture of transient parameter changes. Integrated hardware filtering circuits (such as RC filters and digital low-pass filters) suppress electromagnetic interference (EMI) and noise, improving the data signal-to-noise ratio. Data is encapsulated in a time-series format (such as timestamped JSON / Protocol Buffers) and transmitted to the main control chip via an internal bus (such as SPI / I2C).

[0033] Step S102. Obtain the preset fault types and establish a fault prediction model using machine learning algorithms. The fault prediction model is used to identify and classify preset abnormal states based on parameter information. Abnormal states include at least overvoltage, undervoltage, overcurrent, leakage current, and relay adhesion.

[0034] Specifically, machine learning algorithms are used to train historical parameter sequences to build a fault prediction model that can identify and classify complex abnormal states, thus solving the problem that traditional fixed thresholds cannot capture temporal correlations.

[0035] Data annotation and feature engineering collects normal operation data and historical data on fault scenarios such as overvoltage, undervoltage, overcurrent, leakage, and relay sticking, and constructs a dataset through expert annotation.

[0036] Extract time-series features such as mean, variance, maximum / minimum values, rising / falling edge slopes within the sliding window, and frequency domain features after FFT transformation (used to identify harmonic anomalies).

[0037] LSTM (Long Short-Term Memory) or Transformer can be used to capture long-term temporal dependencies in the parameter sequence; for scenarios with high real-time requirements, it can be simplified to TCN (Temporal Convolutional Network).

[0038] The training process includes: dividing the dataset into 70% training set, 20% validation set, and 10% test set, using the cross-entropy loss function (for multi-class classification problems). The optimizer uses AdamW (with weight decay to prevent overfitting), and the learning rate is dynamically adjusted (e.g., using cosine annealing). Model lightweighting is achieved through pruning and quantization (e.g., FP16 / INT8) to compress the model size, adapting it to edge computing chips for charging stations (e.g., ARM Cortex-M7 or NXP i.MX series).

[0039] The main control chip receives the current parameter sequence in real time (such as sliding window data from the last 10 seconds), and the model outputs the probability value of each fault type. An alarm is triggered when the threshold is ≥0.9. The model is updated regularly via OTA (Over-The-Air) technology, continuously optimizing its generalization ability using new data from the cloud.

[0040] Step S103. Obtain the root cause information of the fault based on the identification and classification results of the abnormal state, and generate a graded processing scheme corresponding to the root cause information of the fault.

[0041] Specifically, based on the classification results of the fault prediction model, combined with the severity and scope of the fault, a differentiated graded response strategy is generated to avoid a "one-size-fits-all" power outage and to balance security and service continuity.

[0042] By establishing a mapping table between fault types and root causes (e.g., "overvoltage" may be caused by grid fluctuations, charging pile voltage regulator failure, or vehicle BMS anomaly), the range of root causes can be narrowed down through multi-parameter joint analysis (e.g., simultaneously detecting whether input voltage and output current are abnormal). A rule engine (e.g., Drools) is introduced to assist in root cause determination based on preset logic (e.g., "overcurrent + abnormal grounding impedance → leakage fault").

[0043] The tiered response strategy includes: Level 1 Fault (Emergency): Such as leakage or relay sticking, immediately trigger hardware power-off (cut off the main circuit within 10ms), and report to the operation and maintenance platform via 4G / NB-IoT, locking the charging pile and prohibiting its use. Level 2 Fault (Warning): Such as overvoltage or undervoltage, first adjust the charging current via PWM (current limit 50%), continuously monitor for 30 seconds, and if it does not recover, it is downgraded to Level 1 handling; at the same time, push an APP notification to the user "Charging power is limited, it is recommended to check the equipment". Level 3 Fault (Reminder): Such as slight overheating (contact temperature ≤70℃), log and report to the cloud, maintain normal charging, and trigger a regular maintenance reminder (such as "Please check the relay within 7 days").

[0044] Power-off / current-limiting operations are performed via solid-state relays (SSRs) or contactors, and the hardware status is visually displayed to the user using LED indicators (red / yellow / green). A processing log (including fault type, response time, and processing result) is recorded for subsequent fault reproduction analysis and strategy optimization.

[0045] Step S104. When charging at a smart charging station, the vehicle model and battery parameters are parsed through the charging interface communication protocol to obtain the corresponding optimal charging current, and the charging curve is dynamically adjusted according to the optimal charging current, real-time grid voltage fluctuations and graded processing scheme; wherein, in standby mode, the plug insertion action is monitored by a low-power sensor, and the hardware interlock design ensures that charging is not started if the plug is not fully inserted.

[0046] Specifically, the charging current is dynamically optimized by combining vehicle battery characteristics, real-time grid status, and fault handling requirements, and charging safety in standby mode is ensured through multiple hardware interlocking mechanisms.

[0047] The optimal charging current is calculated by parsing parameters such as vehicle model, battery capacity, SOC (state of charge), and SOH (state of health) through charging interface protocols (such as GB / T 2234.1-2015 and OCPP 2.0) and querying the built-in battery characteristic table (such as the charging curves of ternary lithium / lithium iron phosphate).

[0048] Segmented constant current control is adopted: when the SOC is low, a large current (such as 0.8C) is used for fast charging, and when it is close to full charge, it switches to a small current (such as 0.2C) trickle charging to avoid overcharging.

[0049] The grid voltage fluctuation compensation uses a phase-locked loop (PLL) to monitor the grid voltage frequency and phase in real time. When a sudden voltage drop is detected (e.g., ≤198V), the charging current is automatically reduced (e.g., from 60A to 40A) to prevent the charging pile from overloaded and tripping.

[0050] If a level 2 fault current limit is triggered, the charging curve will be adjusted to "safe mode" to limit the current to 50% of the rated value and extend the estimated charging time.

[0051] The plug insertion monitoring employs a dual hardware interlock design: mechanical contact detection utilizes two independent spring contacts built into the charging gun plug, which close simultaneously only when the plug is fully inserted (level changes are detected via GPIO pins). Low-power infrared LEDs are installed on both sides of the charging interface; when not inserted, the infrared light is blocked, and when inserted, the light path is open, outputting a level signal.

[0052] The software's anti-accidental touch logic requires that both "double contact closure" and "infrared light path conduction" be satisfied simultaneously for more than 200ms before it is considered "valid insertion," thus avoiding instantaneous contact that triggers charging.

[0053] In standby mode, the charging circuit relay remains open. Charging only begins when a valid insertion is detected and the user confirms (e.g., by swiping a card / launching an app).

[0054] In some embodiments, the real-time acquisition of parameter information through built-in sensors includes: periodically and synchronously acquiring input voltage, output current, grounding resistance value and relay contact temperature through voltage sensor, current sensor, grounding status detection module and temperature sensor respectively; performing digital filtering preprocessing on the acquired raw data to remove high-frequency noise and abnormal jump values, generating a standardized parameter sequence and storing it in a real-time database.

[0055] By using multiple types of sensors to achieve periodic synchronous acquisition of key parameters, and by using digital filtering preprocessing to improve data quality, a reliable data source is provided for subsequent fault diagnosis.

[0056] A high-precision voltage transformer is connected in parallel to the input side, and the input voltage is sampled every 10ms (AC 220V±15% range, accuracy ±0.2%). A closed-loop Hall current sensor is connected in series in the output circuit to synchronously sample the output current (DC 0-200A, accuracy ±1%).

[0057] By integrating a dedicated grounding resistance tester module, the grounding resistance is measured every 5 seconds, with a threshold set to ≤4Ω (national standard requirement). A DS18B20 digital temperature sensor is installed near the relay contacts, and the contact temperature is collected every 200ms (accuracy ±0.5℃). A synchronization mechanism triggers multi-sensor synchronous sampling via a timer interrupt (such as the STM32's TIMER) of the charging pile's main control chip, ensuring that the timestamp error of each parameter is <1ms.

[0058] Digital filtering applies a 5th-order FIR low-pass filter (cutoff frequency 50Hz) to the raw voltage / current data to remove high-frequency noise (such as switching power supply ripple); a moving average filter (window size 5) is used for temperature data to eliminate random jumps. Outlier handling uses the 3σ principle to detect outliers; if a parameter value exceeds [mean - 3σ, mean + 3σ], it is replaced by linear interpolation of data from the preceding and following time points. The processed data is normalized to the [-1, 1] interval (e.g., voltage normalization formula: (V - 220) / 33) and stored in a time series in a memory-based real-time database (such as Redis) for real-time access by the fault model.

[0059] In some embodiments, the fault types include overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; the step of obtaining the preset fault types and establishing a fault prediction model using machine learning algorithms includes: obtaining the feature vectors corresponding to overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; constructing a training dataset using historical fault data; extracting time-series feature information from time-series parameter information using a long short-term memory network algorithm; optimizing feature weights by combining an attention mechanism; training to form a multi-class fault prediction model; and ensuring the model's generalization ability through cross-validation.

[0060] For five typical fault types, a multi-class fault prediction model is constructed using LSTM + attention mechanism to solve the problem that traditional thresholding methods cannot capture temporal features.

[0061] For each sampling point, extract the current value, the mean of the previous 5 time points, and the trend slope (Δ value / Δt) of the next 5 time points to form a 15-dimensional feature vector (e.g., the input voltage corresponds to 3 dimensions: current value, mean, and slope).

[0062] Collect normal operation data (approximately 1 million records) and fault data (2,000 records each for overvoltage and undervoltage, and 1,500 records each for overcurrent, leakage current, and relay sticking) within 12 months, and label the fault type and the time of occurrence.

[0063] For a few types of faults (such as relay sticking), time series interpolation enhancement (such as Stretch DataAugmentation) is used to balance the category distribution.

[0064] The network structure includes: Input layer: receives feature vectors from 100 time steps (i.e., 100×15 dimensional data). Bidirectional LSTM layer: 2 layers, 128 neurons each, capturing forward and backward temporal dependencies. Attention layer: calculates attention weights for each time step, highlighting key fault features (such as voltage spikes during overvoltage). Fully connected layer: outputs the probabilities of 5 types of faults (overvoltage, undervoltage, overcurrent, leakage, and normal) through Softmax.

[0065] Training configuration includes: Loss function: cross-entropy loss; Optimizer: Adam (learning rate 1e-3); Cross-validation: 5-fold hierarchical cross-validation, 100 epochs per fold, with early stopping mechanism (stopping if the validation set loss does not decrease for 5 consecutive epochs); Model deployment: After training, the model is converted to a quantized model (INT8) via TensorFlow Lite and deployed to the local MCU of the charging station (e.g., NXP S32G), with a single inference time of <20ms.

[0066] In some embodiments, the fault prediction model is used to identify and classify preset abnormal states based on parameter information, including: inputting parameter information into the fault prediction model, generating a multi-dimensional feature vector through the feature extraction layer of the fault prediction model; performing probability calculation on the feature vector using the classification layer of the fault prediction model, and outputting the confidence value corresponding to the abnormal state; when the confidence value exceeds a preset threshold, determining that the corresponding abnormal state has occurred and outputting the classification result.

[0067] By using a fault prediction model to extract features and calculate probabilities from real-time parameter sequences, and determining abnormal states through confidence thresholds, accurate fault classification can be achieved.

[0068] The model inference process includes: Input preprocessing: The real-time acquired parameter sequence (e.g., the last 10 seconds, 500 sampling points) is processed according to the standardization method in Example 1, and divided into 5 sliding windows (each window has 100 time steps, with a step size of 20). Feature extraction layer: A bidirectional LSTM layer generates a hidden state vector (128-dimensional) for each time step, and the attention layer calculates the weight matrix for each window, outputting a global feature vector (128-dimensional). Classification layer calculation: The global feature vector is mapped to a 5-dimensional space through a fully connected layer, and the Softmax function outputs the confidence value of each fault category (e.g., overvoltage: 0.92, undervoltage: 0.05, others < 0.03).

[0069] The preset confidence threshold is 0.85. If the confidence level of a certain category is greater than or equal to the threshold, and the category is triggered in three consecutive sliding windows, then the corresponding abnormal state is determined to have occurred (to avoid false alarms due to random fluctuations). The output results include: fault type, confidence level, first trigger time, and current parameter values ​​(e.g., "overvoltage, confidence level 0.95, current voltage 255V"), which are transmitted to the charging pile controller via the CAN bus.

[0070] In some embodiments, obtaining fault root cause information based on the identification and classification results of abnormal states includes: establishing a knowledge base for mapping relationships between abnormal states and fault root causes, wherein the knowledge base contains potential fault points and triggering conditions corresponding to abnormal states; traversing the knowledge base through a rule engine to match corresponding fault root cause combinations based on the identified abnormal states; and combining fault root cause combinations, real-time parameter fluctuation trends, and historical fault records, using evidence theory to perform credibility fusion on multiple possible root causes to determine the fault root cause with the highest priority.

[0071] Based on a knowledge base that maps abnormal states to root causes of failures, and combined with a rule engine and evidence theory, we can achieve root cause localization of failures by fusing multi-source information.

[0072] The knowledge base is constructed using JSON format to define mapping relationships. When an "overvoltage" anomaly is detected, the rule engine automatically matches all relevant root causes in the knowledge base, extracts triggering conditions (such as input voltage and grounding resistance status), and queries real-time parameters to determine whether the conditions are met.

[0073] For each possible root cause, calculate the confidence level of the evidence: power grid fluctuation: if the input voltage is >242V and the grounding is normal, the confidence level m1 = 0.8; regulator failure: if the input is normal but the output is abnormal, the confidence level m2 = 0.7.

[0074] Multiple pieces of evidence are integrated and the joint credibility is calculated using Dempster's combination rule, such as m(grid fluctuation) = (0.8 × 0.3) / (1 - 0.8 × 0.2) = 0.857, and the root cause with the highest priority is determined to be "grid voltage fluctuation".

[0075] Generate a root cause report, which includes: abnormal status, the top 3 possible root causes (sorted by confidence level), and supporting evidence (such as "input voltage 250V, grounding resistance 2Ω"), to help maintenance personnel quickly locate the root cause.

[0076] In some embodiments, the hierarchical processing scheme corresponding to the generation of fault root cause information includes: presetting a fault severity classification standard to divide the fault root cause into a first-level fault root cause, a second-level fault root cause, and a third-level fault root cause; generating a processing instruction to immediately cut off the charging circuit and report a fault code for a first-level fault root cause; generating a processing instruction to reduce the charging power and continuously monitor for a second-level fault root cause; and generating a processing instruction to provide an interface pop-up prompt and record a log for a third-level fault root cause.

[0077] Based on the severity of the root causes of failures, differentiated handling strategies are developed to balance safety measures with service continuity.

[0078] Processing instruction generation includes: Level 1 Fault: Hardware Action: The main circuit relay is disconnected within 10ms via the driver chip (e.g., TC4427), illuminating the red alarm light. Communication Action: A fault code (e.g., F001) is sent to the cloud via the MQTT protocol, containing a timestamp and root cause information, preventing the user from initiating charging.

[0079] Level 2 Fault: Power Adjustment: The output current is limited to 60% of the rated value (e.g., from 100A to 60A) via the PWM module of the charging pile controller, and the yellow indicator light flashes. Continuous Monitoring: The fault model is rerun every 5 seconds. If the fault does not recover within 3 minutes, it is upgraded to Level 1 processing, and the APP pushes a "Charging Power Limited" notification.

[0080] Level 3 Fault: Interface Prompt: The charging pile screen displays "Contact temperature is too high, it is recommended to check within 72 hours," and the green indicator light remains on without affecting charging. Log Recording: Fault information is stored in a local SQLite database and uploaded to the cloud via OTA every morning, triggering maintenance work orders (such as "Relay Maintenance Reminder").

[0081] In some embodiments, the step of parsing the vehicle model and battery parameters through the charging interface communication protocol to obtain the corresponding optimal charging current includes: receiving a charging communication protocol data frame sent by the vehicle; parsing the vehicle identification code, battery type, rated voltage, and remaining capacity in the charging communication protocol data frame; and querying a pre-stored charging parameter mapping table based on the vehicle model, battery type, rated voltage, and remaining capacity to obtain the corresponding optimal charging current.

[0082] By parsing the charging interface communication protocol, vehicle and battery parameters are obtained, and the optimal charging current is determined by matching the pre-stored charging parameter mapping table.

[0083] By parsing the charging communication protocol data frames during the charging handshake phase, the following can be obtained: vehicle VIN code (17-bit string); battery type (e.g., LFP / ternary lithium), rated voltage (e.g., 350V), and remaining capacity (SOC, 0-100%). The mapping table structure is stored in an SQLite database. The vehicle brand and model are identified by the first 8 digits of the VIN code, such as "LFPB1234" → "Brand A, Model B". The range is determined by the SOC value, such as SOC=25% → "0-30%" range, and the max_current of the corresponding battery type is queried (e.g., for LFP batteries of Brand A, Model B, the 0-30% range is 120A).

[0084] If no matching record is found, the general charging curve is used: 0-80% SOC at 0.5C current (C = battery capacity / hour), and above 80% at 0.2C current.

[0085] In some embodiments, the dynamic adjustment of the charging curve based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme includes: using an adaptive PID control algorithm, taking the optimal charging current as a reference value, and collecting the effective value of the grid voltage in real time to calculate the fluctuation coefficient; when the fluctuation coefficient is detected to exceed the rated range, adjusting the charging current according to a preset proportional coefficient; if there is a fault graded processing instruction, superimposing and correcting the current charging current according to the instruction priority to generate a dynamically adjusted charging current sequence.

[0086] Based on the adaptive PID algorithm, combined with grid fluctuations and fault handling instructions, the charging current is dynamically adjusted to optimize charging efficiency and avoid equipment overload.

[0087] The power grid fluctuation coefficient is calculated by real-time acquisition of the effective value of the power grid voltage U_rms, and the fluctuation coefficient K = U_rms / U_nom (U_nom = 220V) is calculated. The normal range K ∈ [0.9, 1.1]. When K < 0.9 (undervoltage) or K > 1.1 (overvoltage), the current adjustment mechanism is activated: if K = 0.85 (voltage drop of 15%), then the current adjustment coefficient α = 0.85, and the target current I_target = optimal current × α (e.g., from 100A to 85A).

[0088] The control objective of adaptive PID control is to make the actual output current track I_target. An incremental PID algorithm is used, with the formula: Δu(k) = Kp(e(k)-e(k-1)) + Ki*e(k) + Kd(e(k)-2e(k-1)+e(k-2)); where e(k) = I_target - I_real, and the Kp / Ki / Kd parameters are automatically tuned using the Ziegler-Nichols method. If a secondary fault current limiting instruction exists (e.g., current limiting to 60%), then I_target = optimal current × 60% × α, and the fault handling logic is executed first.

[0089] The dynamic curve generation updates the current command with a period of 1 second, generating a sequence containing timestamps and current values ​​(such as [t0, 100A], [t1, 95A], …), which is sent to the DC / DC converter control module of the charging pile via the SPI interface.

[0090] In some embodiments, the step of monitoring plug insertion action through a low-power sensor and ensuring that charging is not started when the plug is not fully inserted in the standby state includes: monitoring the insertion displacement signal of the charging plug in real time through a Hall sensor or micro switch; triggering the position detection module in the hardware interlock circuit when the plug insertion action is detected; determining whether the plug is fully inserted through the position detection contact of the mechanical locking mechanism; if no position signal is detected, controlling the charging circuit relay to remain in the open state and prohibiting the sending of the charging start signal.

[0091] By monitoring the plug insertion displacement using Hall effect sensors or microswitches, and combining this with mechanical interlock contacts, a dual safety control mechanism is achieved: "charging will not start if the plug is not fully inserted."

[0092] Insertion displacement monitoring includes: Hall sensor solution: A Hall element is installed inside the charging gun socket, and a permanent magnet is built into the charging gun plug. When the plug is inserted to a depth of ≥50%, the Hall element detects the change in magnetic field and outputs a high-level signal (e.g., 3.3V). Microswitch solution: A limit switch is installed on the inner wall of the socket. When the plug is inserted, it presses against the switch contacts. When the displacement is ≥8mm, the contacts close and output a level signal.

[0093] The position detection module simultaneously acquires the Hall signal (S1) and the mechanical locking contact signal (S2). Only when S1 is high and S2 is closed (detected via a multimeter) is it considered "fully inserted." The charging circuit uses a relay coil connected in series with a logic AND gate (S1 AND S2). The relay coil is energized and closes only when both signals are valid, allowing charging to begin. After detecting that S1 and S2 are valid, the circuit must remain stable for 200ms (counted by the MCU timer) before triggering the charging process to avoid misjudgments due to jitter during insertion and removal.

[0094] If S2 is detected to be disconnected (plug loose) during charging, the hardware will immediately shut down and the charging station screen will display "Plug not plugged in properly, charging has stopped".

[0095] In some embodiments, an adaptive digital filtering framework based on self-supervised learning automatically optimizes filtering parameters using unlabeled data, thus solving the problem that traditional fixed filtering parameters cannot adapt to complex noisy environments.

[0096] Self-supervised task design forces the model to learn noise distribution characteristics by randomly masking the original signal (e.g., masking 20% ​​of the sampling points) and predicting missing values ​​through an encoder-decoder architecture (e.g., LSTM-Autoencoder). Sensor data is decomposed into low-frequency signal components (normal trend) and high-frequency noise components, and contrastive learning is used to make similar signal features close together in the embedding space, while noise features are far apart.

[0097] After training, the noise feature vector extracted by the encoder is input into the fully connected network, which outputs adaptive filtering parameters (such as FIR filter coefficients and sliding window size). The filtering parameters are dynamically adjusted every 10 minutes based on the current noise characteristics. For example, when the grid harmonic pollution is detected to be aggravated, the FIR filter cutoff frequency is automatically reduced to 20Hz.

[0098] Deploy a lightweight self-supervised model (model size <10MB) on the edge computing module of the charging pile (such as NVIDIA Jetson Nano), and use TensorRT to accelerate inference, with a single parameter update taking <50ms.

[0099] In some embodiments, a dynamic fault prediction model based on Transformer+reinforcement learning is used to optimize the prediction strategy through global attention modeling of temporal features and decision feedback.

[0100] The Transformer architecture improvements include: Input layer: Encoding multi-sensor data (voltage, current, temperature, etc.) into feature vectors with location embedding, with a time step of 100 (corresponding to 1 second of data). Encoder layer: Employing a multi-head attention mechanism (8 heads) to capture cross-correlation between different parameters (such as the correlation between voltage fluctuations and current abrupt changes). Prediction head: Outputting the fault probability distribution (including normal states) for the next 5 seconds, supporting advance warning.

[0101] The state is the current feature vector, the action is adjusting the model hyperparameters (such as the number of attention heads and the learning rate), and the reward function is the improvement in prediction accuracy. Using the Proximal Policy Optimization (PPO) algorithm, edge servers collect prediction error data hourly to update the policy network, enabling model self-evolution. Model quantization (FP32→INT8) and pruning techniques are employed to compress the Transformer model inference time to less than 15ms, meeting real-time requirements.

[0102] In some embodiments, a fault detection model based on generative adversarial networks (GANs) is used to address the difficulty in detecting rare faults (such as intermittent relay sticking) in historical data.

[0103] The GAN architecture design includes: Generator (G): Inputs a random noise vector and generates pseudo-samples that conform to the normal operating distribution (such as simulating normal voltage-current fluctuation trajectories). Discriminator (D): Simultaneously receives real data and generated data, and outputs a "real / abnormal" binary classification result (abnormal is defined as data that deviates from the normal manifold).

[0104] The GAN is trained using normal data, enabling the generator to realistically simulate normal operating conditions. For real-time data x, the reconstruction error in the generator's latent space is calculated as E = ||x - G(z_x)||. If E > 3σ (σ is the standard deviation of the reconstruction error of normal data), it is considered an anomaly.

[0105] When an anomaly is detected, the data is fed into a lightweight CNN (such as MobileNetV3) for fine classification. This CNN is pre-trained on anomaly samples generated by a GAN (such as injecting known fault features through a conditional GAN).

[0106] In some embodiments, an interpretable root cause reasoning model based on knowledge graph + graph neural network (GNN) solves the problem that traditional rule engines have difficulty handling complex correlation faults.

[0107] Knowledge graph construction: Entity types include sensors (such as voltage sensors), components (such as relays), fault types (such as overvoltage), and parameters (such as grounding resistance).

[0108] Components - contain - sensors (e.g., "charging module - contains - temperature sensor"); faults - associated - parameters (e.g., "overvoltage - associated - input voltage > 242V"); root causes - lead to - anomalies (e.g., "grid fluctuation - leads to - overvoltage"); using the Neo4j graph database to store millions of entities and their relationships.

[0109] The GraphSAGE algorithm is used, and node features include component status (such as relay contact temperature) and parameter threshold violations (such as grounding resistance > 4Ω).

[0110] Node embeddings are updated by aggregating features of neighboring nodes (such as average pooling) to capture multi-hop associations (such as "poor grounding → leakage → relay burnout").

[0111] The probability ranking of the root cause nodes is output (e.g., "grid fluctuation: 0.7, relay aging: 0.25"), and inference paths are generated through path search algorithms (e.g., "input voltage anomaly → knowledge graph related to root cause of grid fluctuation"), which are then displayed to maintenance personnel in the form of a visual graph.

[0112] In some embodiments, collaborative modeling of charging pile clusters based on federated learning addresses the issues of insufficient data per pile and privacy protection, and improves the ability to detect rare faults.

[0113] Each charging station acts as an edge node, while the cloud serves as the central server.

[0114] The communication protocol uses homomorphic encryption (such as Paillier) to transmit model gradients, ensuring that the original data does not leave stubs.

[0115] A global model (such as LSTM-ATTN) is distributed from the cloud to each charging station. Each charging station updates its model using local fault data (such as overcurrent events), and only uploads gradient increments.

[0116] The aggregation strategy uses the FedAvg algorithm to aggregate gradients and updates the global model every 10 iterations, focusing on optimizing the detection accuracy of rare faults (such as relay adhesion).

[0117] When multiple charging piles in a certain area simultaneously report "abnormal charging curves", the global model can quickly identify new fault modes (such as consistency defects in a batch of batteries) and synchronize them to all charging piles via OTA to achieve cluster defense.

[0118] In some embodiments, by constructing a digital twin of the charging pile, potential fault scenarios are simulated and prediction models are optimized by combining physical models with machine learning.

[0119] A thermal-electric coupling model of the charging pile was established using COMSOL Multiphysics to simulate physical processes such as relay contact temperature rise and cable aging, and to generate simulation fault data (such as voltage-current curves when contact resistance increases due to contact oxidation).

[0120] By mixing simulation data with real operation and maintenance data (such as disassembly and inspection records), a fault prediction model is trained to address the problem of scarce real fault samples. A "virtual sensor" mechanism is established: unmeasurable parameters (such as internal cable humidity) are predicted using digital twins and input into the physical model to enhance the prediction dimensionality.

[0121] The digital twin and the physical pegs are synchronized in real time (e.g., via the OPC UA protocol). When the virtual model predicts that the remaining lifespan of a component is less than 7 days, a level 3 warning is triggered (e.g., the interface prompts "The relay is expected to need to be replaced in 7 days").

[0122] Please see Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of a control device 200 based on dynamic fault diagnosis provided in an embodiment of this application. The control device 200 based on dynamic fault diagnosis is used to execute the steps of the control method based on dynamic fault diagnosis shown in the above embodiments. The control device 200 based on dynamic fault diagnosis can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0123] like Figure 3 As shown, the control device 200 based on dynamic fault diagnosis includes: Information acquisition unit 201 is used to acquire parameter information in real time through built-in sensors. The parameter information includes at least input voltage, output current, grounding status and relay contact temperature parameters. The type acquisition unit 202 is used to acquire preset fault types and establish a fault prediction model using machine learning algorithms. The fault prediction model is used to identify and classify preset abnormal states based on parameter information. Abnormal states include at least overvoltage, undervoltage, overcurrent, leakage current, and relay adhesion. The scheme generation unit 203 is used to obtain fault root cause information based on the identification and classification results of abnormal states, and generate a hierarchical processing scheme corresponding to the fault root cause information. The curve adjustment unit 204 is used to parse the vehicle model and battery parameters through the charging interface communication protocol when charging at the smart charging pile, obtain the corresponding optimal charging current, and dynamically adjust the charging curve according to the optimal charging current, real-time grid voltage fluctuations, and graded processing scheme. In standby mode, a low-power sensor monitors the plug insertion action, and a hardware interlock design ensures that charging does not start if the plug is not fully inserted.

[0124] In some embodiments, the real-time acquisition of parameter information through built-in sensors includes: periodically and synchronously acquiring input voltage, output current, grounding resistance value and relay contact temperature through voltage sensor, current sensor, grounding status detection module and temperature sensor respectively; performing digital filtering preprocessing on the acquired raw data to remove high-frequency noise and abnormal jump values, generating a standardized parameter sequence and storing it in a real-time database.

[0125] In some embodiments, the fault types include overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; the step of obtaining the preset fault types and establishing a fault prediction model using machine learning algorithms includes: obtaining the feature vectors corresponding to overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; constructing a training dataset using historical fault data; extracting time-series feature information from time-series parameter information using a long short-term memory network algorithm; optimizing feature weights by combining an attention mechanism; training to form a multi-class fault prediction model; and ensuring the model's generalization ability through cross-validation.

[0126] In some embodiments, the fault prediction model is used to identify and classify preset abnormal states based on parameter information, including: inputting parameter information into the fault prediction model, generating a multi-dimensional feature vector through the feature extraction layer of the fault prediction model; performing probability calculation on the feature vector using the classification layer of the fault prediction model, and outputting the confidence value corresponding to the abnormal state; when the confidence value exceeds a preset threshold, determining that the corresponding abnormal state has occurred and outputting the classification result.

[0127] In some embodiments, obtaining fault root cause information based on the identification and classification results of abnormal states includes: establishing a knowledge base for mapping relationships between abnormal states and fault root causes, wherein the knowledge base contains potential fault points and triggering conditions corresponding to abnormal states; traversing the knowledge base through a rule engine to match corresponding fault root cause combinations based on the identified abnormal states; and combining fault root cause combinations, real-time parameter fluctuation trends, and historical fault records, using evidence theory to perform credibility fusion on multiple possible root causes to determine the fault root cause with the highest priority.

[0128] In some embodiments, the hierarchical processing scheme corresponding to the generation of fault root cause information includes: presetting a fault severity classification standard to divide the fault root cause into a first-level fault root cause, a second-level fault root cause, and a third-level fault root cause; generating a processing instruction to immediately cut off the charging circuit and report a fault code for a first-level fault root cause; generating a processing instruction to reduce the charging power and continuously monitor for a second-level fault root cause; and generating a processing instruction to provide an interface pop-up prompt and record a log for a third-level fault root cause.

[0129] In some embodiments, the step of parsing the vehicle model and battery parameters through the charging interface communication protocol to obtain the corresponding optimal charging current includes: receiving a charging communication protocol data frame sent by the vehicle; parsing the vehicle identification code, battery type, rated voltage, and remaining capacity in the charging communication protocol data frame; and querying a pre-stored charging parameter mapping table based on the vehicle model, battery type, rated voltage, and remaining capacity to obtain the corresponding optimal charging current.

[0130] In some embodiments, the dynamic adjustment of the charging curve based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme includes: using an adaptive PID control algorithm, taking the optimal charging current as a reference value, and collecting the effective value of the grid voltage in real time to calculate the fluctuation coefficient; when the fluctuation coefficient is detected to exceed the rated range, adjusting the charging current according to a preset proportional coefficient; if there is a fault graded processing instruction, superimposing and correcting the current charging current according to the instruction priority to generate a dynamically adjusted charging current sequence.

[0131] In some embodiments, the step of monitoring plug insertion action through a low-power sensor and ensuring that charging is not started when the plug is not fully inserted in the standby state includes: monitoring the insertion displacement signal of the charging plug in real time through a Hall sensor or micro switch; triggering the position detection module in the hardware interlock circuit when the plug insertion action is detected; determining whether the plug is fully inserted through the position detection contact of the mechanical locking mechanism; if no position signal is detected, controlling the charging circuit relay to remain in the open state and prohibiting the sending of the charging start signal.

[0132] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the control device and its modules based on dynamic fault diagnosis described above can be found in the corresponding contents of the various embodiments of the control method based on dynamic fault diagnosis, and will not be repeated here.

[0133] The aforementioned control method based on dynamic fault diagnosis can be implemented as a computer program, which can be used in various ways, such as... Figure 3 It runs on the device shown.

[0134] Please see Figure 4 , Figure 4 This is a schematic block diagram of the structure of a smart charging pile provided in an embodiment of this application. The smart charging pile includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0135] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any control method based on dynamic fault diagnosis.

[0136] The processor provides computing and control capabilities to support the operation of the entire smart charging station.

[0137] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any control method based on dynamic fault diagnosis.

[0138] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the solution of this application and does not constitute a limitation on the terminal to which the solution of this application is applied. A specific smart charging pile may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0139] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0140] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: The built-in sensor collects parameter information in real time, including at least the input voltage, output current, grounding status, and relay contact temperature.

[0141] Obtain the preset fault types and use machine learning algorithms to build a fault prediction model. The fault prediction model is used to identify and classify the preset abnormal states based on parameter information. The abnormal states include at least overvoltage, undervoltage, overcurrent, leakage current and relay sticking.

[0142] Based on the identification and classification results of abnormal states, obtain the root cause information of the fault and generate a hierarchical processing scheme corresponding to the root cause information of the fault.

[0143] When charging at a smart charging station, the vehicle model and battery parameters are parsed through the charging interface communication protocol to obtain the corresponding optimal charging current. The charging curve is then dynamically adjusted based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme. In standby mode, a low-power sensor monitors the plug insertion action, and a hardware interlock design ensures that charging does not start if the plug is not fully inserted.

[0144] In some embodiments, the real-time acquisition of parameter information through built-in sensors includes: periodically and synchronously acquiring input voltage, output current, grounding resistance value and relay contact temperature through voltage sensor, current sensor, grounding status detection module and temperature sensor respectively; performing digital filtering preprocessing on the acquired raw data to remove high-frequency noise and abnormal jump values, generating a standardized parameter sequence and storing it in a real-time database.

[0145] In some embodiments, the fault types include overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; the step of obtaining the preset fault types and establishing a fault prediction model using machine learning algorithms includes: obtaining the feature vectors corresponding to overvoltage, undervoltage, overcurrent, leakage current, and relay sticking; constructing a training dataset using historical fault data; extracting time-series feature information from time-series parameter information using a long short-term memory network algorithm; optimizing feature weights by combining an attention mechanism; training to form a multi-class fault prediction model; and ensuring the model's generalization ability through cross-validation.

[0146] In some embodiments, the fault prediction model is used to identify and classify preset abnormal states based on parameter information, including: inputting parameter information into the fault prediction model, generating a multi-dimensional feature vector through the feature extraction layer of the fault prediction model; performing probability calculation on the feature vector using the classification layer of the fault prediction model, and outputting the confidence value corresponding to the abnormal state; when the confidence value exceeds a preset threshold, determining that the corresponding abnormal state has occurred and outputting the classification result.

[0147] In some embodiments, obtaining fault root cause information based on the identification and classification results of abnormal states includes: establishing a knowledge base for mapping relationships between abnormal states and fault root causes, wherein the knowledge base contains potential fault points and triggering conditions corresponding to abnormal states; traversing the knowledge base through a rule engine to match corresponding fault root cause combinations based on the identified abnormal states; and combining fault root cause combinations, real-time parameter fluctuation trends, and historical fault records, using evidence theory to perform credibility fusion on multiple possible root causes to determine the fault root cause with the highest priority.

[0148] In some embodiments, the hierarchical processing scheme corresponding to the generation of fault root cause information includes: presetting a fault severity classification standard to divide the fault root cause into a first-level fault root cause, a second-level fault root cause, and a third-level fault root cause; generating a processing instruction to immediately cut off the charging circuit and report a fault code for a first-level fault root cause; generating a processing instruction to reduce the charging power and continuously monitor for a second-level fault root cause; and generating a processing instruction to provide an interface pop-up prompt and record a log for a third-level fault root cause.

[0149] In some embodiments, the step of parsing the vehicle model and battery parameters through the charging interface communication protocol to obtain the corresponding optimal charging current includes: receiving a charging communication protocol data frame sent by the vehicle; parsing the vehicle identification code, battery type, rated voltage, and remaining capacity in the charging communication protocol data frame; and querying a pre-stored charging parameter mapping table based on the vehicle model, battery type, rated voltage, and remaining capacity to obtain the corresponding optimal charging current.

[0150] In some embodiments, the dynamic adjustment of the charging curve based on the optimal charging current, real-time grid voltage fluctuations, and a graded processing scheme includes: using an adaptive PID control algorithm, taking the optimal charging current as a reference value, and collecting the effective value of the grid voltage in real time to calculate the fluctuation coefficient; when the fluctuation coefficient is detected to exceed the rated range, adjusting the charging current according to a preset proportional coefficient; if there is a fault graded processing instruction, superimposing and correcting the current charging current according to the instruction priority to generate a dynamically adjusted charging current sequence.

[0151] In some embodiments, the step of monitoring plug insertion action through a low-power sensor and ensuring that charging is not started when the plug is not fully inserted in the standby state includes: monitoring the insertion displacement signal of the charging plug in real time through a Hall sensor or micro switch; triggering the position detection module in the hardware interlock circuit when the plug insertion action is detected; determining whether the plug is fully inserted through the position detection contact of the mechanical locking mechanism; if no position signal is detected, controlling the charging circuit relay to remain in the open state and prohibiting the sending of the charging start signal.

[0152] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the control method based on dynamic fault diagnosis provided in any embodiment of this application.

[0153] The computer-readable storage medium can be an internal storage unit of the smart charging pile described in the foregoing embodiments, such as the hard drive or memory of the smart charging pile. Alternatively, the computer-readable storage medium can be an external storage device of the smart charging pile, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the smart charging pile.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A control method based on dynamic fault diagnosis, characterized by, The application is applied to an intelligent charging pile and has the characteristics that it comprises: Real-time acquisition of parameter information through an internal sensor, wherein the parameter information at least includes input voltage, output current, grounding state and relay contact temperature parameter; Obtaining a preset fault type, establishing a fault prediction model by using a machine learning algorithm, and the fault prediction model is used for identifying and classifying the preset abnormal state based on the parameter information, and the abnormal state at least includes overvoltage, undervoltage, overcurrent, electric leakage and relay sticking; Obtaining fault root cause information according to the identification and classification results of the abnormal state, and generating a hierarchical processing scheme corresponding to the fault root cause information; When the intelligent charging pile is charging, the vehicle model and battery parameters are parsed through the charging interface communication protocol, the corresponding optimal charging current is obtained, and the charging curve is dynamically adjusted according to the optimal charging current, real-time power grid voltage fluctuation and hierarchical processing scheme; wherein in the standby state, the plug insertion action is monitored through a low-power sensor, and the hardware interlocking design is used to ensure that the charging is not started when it is not completely inserted.

2. The control method according to claim 1, characterized by, The real-time acquisition of parameter information through the internal sensor comprises: Periodically and synchronously collecting input voltage, output current, grounding resistance value and relay contact temperature through a voltage sensor, a current sensor, a grounding state detection module and a temperature sensor; The collected raw data is subjected to digital filtering pretreatment to remove high-frequency noise and abnormal jump values, standard parameter sequences are generated and stored in a real-time database.

3. The control method according to claim 1, characterized by, The fault types include overvoltage, undervoltage, overcurrent, electric leakage and relay sticking; the obtaining of the preset fault type and the establishment of the fault prediction model by using the machine learning algorithm comprise: Obtaining the feature vectors corresponding to overvoltage, undervoltage, overcurrent, electric leakage and relay sticking; constructing a training data set by using historical fault data, extracting time sequence features of time sequence parameter information by using a long short-term memory network algorithm, optimizing feature weights by using an attention mechanism, training to form a multi-classification fault prediction model, and ensuring the model generalization ability by cross-validation.

4. The control method according to claim 1, characterized by, The fault prediction model is used for identifying and classifying the preset abnormal state based on the parameter information, and comprises: Inputting the parameter information into the fault prediction model, generating a multi-dimensional feature vector through the feature extraction layer of the fault prediction model; Using the classification layer of the fault prediction model to calculate the probability of the feature vector, and outputting the confidence value corresponding to the abnormal state; When the confidence value exceeds a preset threshold, it is determined that the corresponding abnormal state occurs and the classification result is output.

5. The control method according to claim 1, characterized by, The fault root cause information is obtained according to the identification and classification results of the abnormal state, and comprises: Establishing a mapping relationship knowledge base of abnormal state and fault root cause, wherein the mapping relationship knowledge base contains potential fault points and trigger conditions corresponding to the abnormal state; According to the identified abnormal state, the knowledge base is traversed through a rule engine to match the corresponding fault root cause combination; Combining the fault root cause combination, the real-time parameter fluctuation trend and the historical fault record, the evidence theory is used to fuse the credibility of multiple possible root causes to determine the fault root cause with the highest priority.

6. The control method according to claim 1, characterized by The hierarchical processing scheme corresponding to the fault root cause information is generated, and comprises: A preset fault severity grading standard is used to divide fault causes into first-level fault causes, second-level fault causes and third-level fault causes; For the first-level fault causes, a processing instruction of immediately cutting off the charging circuit and reporting a fault code is generated; for the second-level fault causes, a processing instruction of reducing the charging power and continuously monitoring is generated; and for the third-level fault causes, a processing instruction of interface pop-up prompt and log recording is generated.

7. The control method according to claim 1, characterized by, The vehicle model and battery parameters are parsed through the charging interface communication protocol, and the corresponding optimal charging current is obtained, including: Receiving a charging communication protocol data frame sent by a vehicle; Parsing the vehicle identification code and battery type, rated voltage and remaining capacity in the charging communication protocol data frame; According to the vehicle model, battery type, rated voltage and remaining capacity, a pre-stored charging parameter mapping table is queried to match and obtain the corresponding optimal charging current.

8. The control method according to claim 1, characterized by, The charging curve is dynamically adjusted according to the optimal charging current, real-time power grid voltage fluctuation and hierarchical processing scheme, including: An adaptive PID control algorithm is used to take the optimal charging current as a reference value, and the power grid voltage effective value is collected in real time to calculate the fluctuation coefficient; When it is detected that the fluctuation coefficient exceeds the rated range, the charging current is adjusted according to the preset proportion coefficient; If there is a fault hierarchical processing instruction, the current charging current is superimposed and corrected according to the instruction priority to generate a dynamically adjusted charging current sequence.

9. The control method according to claim 1, characterized by, In the standby state, the plug insertion action is monitored by a low-power sensor, and the hardware interlocking design is used to ensure that the charging is not started when it is not completely inserted, including: A Hall sensor or micro switch is used to monitor the insertion displacement signal of the charging plug in real time, and when the plug insertion action is detected, a position detection module in the hardware interlocking circuit is triggered; Whether the plug is completely inserted is determined by the in-place detection contact of the mechanical locking mechanism, and if the in-place signal is not detected, the charging circuit relay is kept in an open state, and the charging start signal is prohibited from being sent.

10. An intelligent charging pile, characterized in that, The intelligent charging pile includes a memory and a processor; The memory is used to store a computer program; The processor is used to execute the computer program and implement the method of any one of claims 1 to 9 when executing the computer program.

Images