A liquid-cooled charging pile fault detection method and system

Abstract

The application relates to the technical field of new energy vehicles, and discloses a liquid-cooled charging pile fault detection method and system, the method comprises the following steps: collecting temperature and pressure data streams, performing multi-scale feature extraction and denoising, and obtaining a temperature fluctuation interference mode through correlation analysis; combining historical correlation data clustering to generate a dynamic correction coefficient, performing deviation compensation on the original pressure data stream to obtain a corrected pressure signal; extracting a high-frequency component of the corrected pressure signal, performing modal decomposition to obtain a leakage fluctuation feature description when the fluctuation level is out of limit; calculating a deviation vector mapping leakage risk probability value to obtain a preliminary risk level; generating state gain based on the thermal field conduction rate deviation and the risk level fusion, dynamically updating the probability value to obtain a refined risk assessment result and performing a graded alarm. The application effectively overcomes the interference of complex air temperature fluctuation and strong background noise, accurately strips non-fault pressure drift, and realizes high-precision extraction and dynamic early warning of early micro-leakage faults.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle technology, and in particular to a method and system for detecting faults in liquid-cooled charging piles. Background Technology

[0002] Currently, liquid-cooled charging piles are a crucial infrastructure for the manufacturing and widespread adoption of new energy vehicle-related facilities, making their operational stability and safety particularly critical. Especially in liquid-cooled charging piles, the health of the cooling system directly affects the equipment's efficient heat dissipation and long-term lifespan; even minor malfunctions can lead to significant safety hazards. With the high-density deployment of charging networks, accurately monitoring and promptly detecting potential problems in the cooling system has become a key area for breakthroughs in the industry. Therefore, intelligent fault detection technology based on sensor data has become a core means of ensuring equipment operational reliability in this field.

[0003] In one existing technology, a conventional pressure detection scheme based on single parameter monitoring is mainly adopted. This scheme typically collects the absolute pressure value inside the cooling pipes under a default stable operating environment, relies on the overall system pressure drop characteristics caused by a large fluid leakage, and identifies obvious defects such as pipe rupture or component failure by setting fixed upper and lower pressure safety thresholds.

[0004] This method, relying on the perception of a single physical quantity and judgment based on a basic threshold, has inherent limitations. Because this detection method often struggles to cope with complex operating environments, especially in scenarios with frequent temperature changes, it is easily affected by external interference, leading to inaccurate judgments of the cooling system's status. In particular, temperature fluctuations significantly impact coolant pressure. Since temperature changes cause slight expansion or contraction of coolant volume, this natural phenomenon means that the pressure data contains a large amount of interference information unrelated to the fault, making it difficult to directly reflect the true system status. The presence of this interference also makes it extremely difficult to detect early signs of minute leaks, as initial leaks often manifest as extremely slight pressure deviations, easily masked by temperature fluctuations. When dealing with subtle anomalies in dynamic environments, this method often fails to effectively distinguish between normal fluctuations and genuine problem signals, thus missing early warning opportunities and increasing the difficulty and risk of equipment maintenance.

[0005] Therefore, the core technical challenge of existing technologies lies in how to accurately extract pressure change signals related to faults and promptly identify early signs of minute leaks in environments with frequent temperature variations. This requires in-depth analysis of the multi-scale mapping relationship between temperature and pressure data to overcome the shortcomings of traditional single-feature monitoring, which is easily masked by thermodynamic phenomena. This enables high-precision extraction and dynamic early warning of potential leaks in liquid-cooled charging piles, addressing the low detection accuracy of traditional methods under complex temperature interference and high-frequency dynamic operating conditions. Summary of the Invention

[0006] This invention provides a method and system for detecting faults in liquid-cooled charging piles, which solves the problem of low detection accuracy of traditional methods under complex temperature interference and high-frequency dynamic operating conditions, and realizes high-precision extraction and dynamic early warning of potential leakage faults in liquid-cooled charging piles.

[0007] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a fault detection method for liquid-cooled charging piles, comprising: The temperature data stream and raw pressure data stream of the liquid-cooled charging pile cooling system are collected. Based on the temperature data stream and raw pressure data stream, temperature fluctuation interference analysis is performed to obtain the temperature fluctuation interference mode. Historical correlation data matching the temperature fluctuation interference pattern is acquired and multidimensional feature clustering is performed to obtain interference feature clusters. Dynamic correction coefficients are determined based on the interference feature clusters. Deviation compensation operation is performed on the original pressure data stream using the dynamic correction coefficients to obtain the corrected pressure signal. High-frequency components of pressure fluctuations are extracted from the corrected pressure signal, the fluctuation deviation of the high-frequency components of pressure fluctuations is calculated, and the real-time fluctuation level is determined based on the fluctuation deviation. When the real-time fluctuation level is determined to exceed the preset dynamic range threshold, feature extraction and mode decomposition are performed on the high-frequency components of pressure fluctuations to obtain a description of leakage fluctuation characteristics. The deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark is calculated. The deviation vector is mapped using a preset probability mapping function to obtain the leakage risk probability value. The leakage risk probability value is compared with a preset dynamic risk warning threshold to obtain a preliminary risk level. Thermodynamic field values ​​are constructed based on the temperature data stream and the absolute deviation of conduction rate is calculated. The absolute deviation of conduction rate is weighted and fused with the preliminary risk level to obtain the state gain. The state gain is used to update the leakage risk probability value to obtain the refining risk assessment result. When the refining risk assessment result is determined to be greater than the preset final warning threshold, a tiered alarm mechanism is triggered, and log data is generated and input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

[0008] Secondly, the present invention provides a liquid-cooled charging pile fault detection device, comprising: The interference mode construction module is used to collect the temperature data stream and the original pressure data stream of the liquid-cooled charging pile cooling system, and perform temperature fluctuation interference analysis based on the temperature data stream and the original pressure data stream to obtain the temperature fluctuation interference mode. The pressure signal correction module is used to acquire historical correlation data that matches the temperature fluctuation interference pattern and perform multi-dimensional feature clustering to obtain interference feature clusters. Based on the interference feature clusters, dynamic correction coefficients are determined, and deviation compensation operations are performed on the original pressure data stream using the dynamic correction coefficients to obtain the corrected pressure signal. The high-frequency feature extraction module is used to extract the high-frequency components of pressure fluctuation from the corrected pressure signal, calculate the fluctuation deviation of the high-frequency components of pressure fluctuation, determine the real-time fluctuation level based on the fluctuation deviation, and when it is determined that the real-time fluctuation level exceeds the preset dynamic range threshold, perform feature extraction and mode decomposition on the high-frequency components of pressure fluctuation to obtain a leakage fluctuation feature description. The preliminary risk assessment module is used to calculate the deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark, map the deviation vector using a preset probability mapping function to obtain the leakage risk probability value, and compare the leakage risk probability value with a preset dynamic risk warning threshold to obtain the preliminary risk level. The risk dynamic refining module is used to construct a thermodynamic field value based on the temperature data stream and calculate the absolute deviation of the conduction rate. The absolute deviation of the conduction rate is weighted and fused with the preliminary risk level to obtain a state gain. The state gain is used to update the leakage risk probability value to obtain the refining risk assessment result. The alarm and log recording module is used to trigger a graded alarm mechanism when the refining risk assessment result is determined to be greater than the preset final warning threshold, and to generate log data that is input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs an interference decoupling mechanism based on the correlation mapping between temperature multi-scale features and pressure signals to obtain temperature fluctuation interference patterns. Addressing the technical pain point that existing technologies mainly rely on sensing a single physical quantity, which easily leads to a large amount of interference information in the pressure data due to the thermal expansion and contraction of the coolant during frequent temperature changes, thus masking minute leakage signals, this invention decomposes and denoises the temperature and pressure data streams and performs correlation mapping calculations. This accurately removes temperature fluctuation interference unrelated to the fault, solving the problem of difficulty in extracting weak leakage signals under complex temperature interference, and significantly improving the system's detection accuracy for early minor defects.

[0010] (2) This invention utilizes historical correlation data to perform multi-dimensional feature clustering and generates dynamic correction coefficients to perform deviation compensation on the original pressure data stream. Addressing the technical shortcomings of traditional fault detection schemes that use fixed pressure safety thresholds and struggle to cope with baseline pressure drop drift caused by dynamic charging loads, this invention extracts historical features for clustering to determine dynamic correction coefficients and performs deviation compensation operations. This achieves accurate correction of baseline pressure deviations under dynamic operating conditions, solving the problem of false alarms or missed alarms easily caused by the fixed threshold method, and ensuring the baseline stability of subsequent pressure anomaly detection.

[0011] (3) This invention accurately extracts leakage fluctuation characteristics by performing frequency domain segmentation and mode decomposition on the corrected pressure signal. Addressing the technical shortcoming that fluid turbulence in cooling system pipelines is easily confused with the slight pressure attenuation caused by minor leaks, leading to false alarms, this invention separates the high-frequency components of pressure fluctuations and calculates the fluctuation deviation. When the deviation exceeds a threshold, mode decomposition is performed to extract the attenuation rate characteristics, effectively separating the real leakage signal from non-leakage disturbances, significantly improving the accuracy of leakage characteristic confirmation under complex operating conditions.

[0012] (4) This invention constructs a risk dynamic refinement system based on the spatiotemporal alignment of thermal fields and the fusion of multidimensional state gains, and generates diagnostic records by combining distributed storage. In view of the shortcomings of traditional systems, such as the lack of cross-validation of spatial dimensions and the tendency of single-node probability estimation to produce bias, this invention obtains the state gain by calculating the weighted fusion of the absolute deviation of conduction rate and the initial risk level, and dynamically updates and verifies the confidence level of leakage risk probability value, thereby realizing the iterative convergence of fault warning level and ensuring the extremely high reliability and traceability of the final evaluation results during long-term high-load continuous operation. Attached Figure Description

[0013] Figure 1 This is a schematic flowchart of a liquid-cooled charging pile fault detection method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a liquid-cooled charging pile fault detection system provided in the second embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Reference Figure 1 The first embodiment of the present invention provides a fault detection method for liquid-cooled charging piles, including the following steps: S11, Collect the temperature data stream and the original pressure data stream of the liquid-cooled charging pile cooling system, and perform temperature fluctuation interference analysis based on the temperature data stream and the original pressure data stream to obtain the temperature fluctuation interference mode. S12, acquire historical correlation data matching the temperature fluctuation interference pattern and perform multi-dimensional feature clustering to obtain interference feature clusters, determine dynamic correction coefficients based on the interference feature clusters, and use the dynamic correction coefficients to perform deviation compensation operation on the original pressure data stream to obtain a corrected pressure signal; S13, extract the high-frequency components of pressure fluctuation from the corrected pressure signal, calculate the fluctuation deviation of the high-frequency components of pressure fluctuation, determine the real-time fluctuation level based on the fluctuation deviation, and when it is determined that the real-time fluctuation level exceeds the preset dynamic range threshold, perform feature extraction and mode decomposition on the high-frequency components of pressure fluctuation to obtain a description of leakage fluctuation characteristics. S14, calculate the deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark, map the deviation vector using a preset probability mapping function to obtain the leakage risk probability value, and compare the leakage risk probability value with a preset dynamic risk warning threshold to obtain a preliminary risk level. S15, construct a thermal field value based on the temperature data stream and calculate the absolute deviation of the conduction rate. Calculate the state gain by weighting and fusing the absolute deviation of the conduction rate with the preliminary risk level. Update the leakage risk probability value using the state gain to obtain the refining risk assessment result. S16, when the refining risk assessment result is determined to be greater than the preset final warning threshold, a graded alarm mechanism is triggered, and log data is generated and input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

[0016] In step S11, it is necessary to collect the temperature data stream and the original pressure data stream of the liquid-cooled charging pile cooling system, and perform temperature fluctuation interference analysis based on the temperature data stream and the original pressure data stream to obtain the temperature fluctuation interference mode, including: Temperature data streams are acquired by high-precision temperature sensors deployed in the cooling system of liquid-cooled charging piles, and raw pressure data streams are acquired synchronously by pressure sensors. The temperature data stream is decomposed into multiple levels using a preset wavelet basis function to extract low-frequency temperature trend components and high-frequency temperature detail components as multi-scale temperature features. High-frequency random noise in the original pressure data stream is filtered out using a preset adaptive filter to obtain a preprocessed pressure signal; A sliding window cross-correlation calculation is performed on the preprocessed pressure signal and the low-frequency temperature trend component to determine the static temperature offset, and a pressure correction base is generated based on the static temperature offset. The pressure correction base is nonlinearly superimposed with the instantaneous thermal shock pulse identified based on the high-frequency detail components of the temperature to construct a temperature fluctuation interference mapping model. The temperature fluctuation interference mapping model is used to analyze the degree of distortion of the preprocessed pressure signal under different preset temperature gradients, and the temperature fluctuation interference mode is determined.

[0017] First, temperature data streams are acquired using high-precision temperature sensors deployed in the cooling system of the liquid-cooled charging pile, while raw pressure data streams are acquired synchronously using pressure sensors. In one embodiment, high-precision temperature and pressure sensors are simultaneously deployed at the inlet and outlet of the liquid-cooled circulation loop inside the charging pile, as well as at the heat dissipation nodes of the core power module. Real-time changes in the physical quantities of the fluid medium are acquired through a high-frequency sampling mode. It should be noted that the preset high-frequency sampling frequency is set based on the Nyquist sampling theorem of fluid dynamics and the nominal operating frequency of the liquid-cooled pump to ensure complete capture of high-frequency fluid pulsations. For example, for charging piles with a built-in liquid-cooled pump fundamental frequency of approximately 50Hz, the system presets the synchronous acquisition frequency to 200Hz to acquire temperature data streams and raw pressure data stream sequences containing complete dynamic temporal characteristics.

[0018] Secondly, the temperature data stream is decomposed into multiple levels using a preset wavelet basis function to extract low-frequency trend components and high-frequency detail components, which serve as multi-scale temperature features. In one embodiment, based on the non-stationary and abrupt changes in the temperature data stream during charging load changes, the system matches the well-supported Daubechies4 (Db4) as the preset wavelet basis function and performs a four-level discrete wavelet decomposition operation on the temperature data stream. After decomposition, the fourth-level approximate component is obtained as the low-frequency trend component, and the first to third-level detail components are obtained as the high-frequency detail components.

[0019] Subsequently, a preset adaptive filter is used to filter out high-frequency random noise in the original pressure data stream, resulting in a preprocessed pressure signal. In one embodiment, considering the strong electromagnetic interference environment inside the charging pile, the original pressure data stream collected by the sensor often contains electromagnetic radiation noise and high-frequency random white noise. The system uses a Kalman filter algorithm to construct a preset adaptive filter, dynamically updates the filter gain matrix, and performs smoothing and denoising by calculating the observation variance and process variance of the pressure signal in real time. It should be noted that the initial process noise covariance parameter of the preset adaptive filter is determined based on common knowledge in the field and the initial calibration data of the equipment. The system extracts the pure static hydraulic base noise of the charging pile in the no-load standby state to calculate this prior parameter. For example, the original pressure data stream contains high-frequency spikes with an amplitude of ±5kPa around the reference working pressure of 150kPa. After processing by the adaptive filter, random noise above 100Hz is completely eliminated, and a smooth, high signal-to-noise ratio preprocessed pressure signal is output.

[0020] Next, a sliding window cross-correlation calculation is performed on the preprocessed pressure signal and the low-frequency temperature trend component to determine the static temperature offset, and a pressure correction base is generated based on the static temperature offset. In one embodiment, the system sets a fixed-length sliding time window, synchronously extracts the low-frequency temperature trend component sequence and the preprocessed pressure signal sequence within the window, calculates the maximum value of their cross-correlation function, and determines the static temperature offset based on the time delay and amplitude mapping coefficient corresponding to this maximum value. Specifically, the system shifts the extracted low-frequency temperature trend component sequence frame by frame relative to the preprocessed pressure signal sequence on the time axis and calculates the sum of the data dot products corresponding to each shift. When the sum of the dot products reaches its maximum value, it indicates that the two sets of signals have achieved optimal alignment in terms of waveform evolution trends. The system extracts the corresponding number of shifted frames as the actual thermodynamic conduction time delay and calculates the ratio of the amplitude changes of the two sets of signals within this interval after alignment as the amplitude mapping coefficient, thereby accurately determining the static temperature offset.

[0021] It is worth noting that the width of this sliding time window is set based on common-sense empirical values ​​of thermodynamic conduction. Considering that the physical conduction delay from the temperature rise of the heating element to the pressure response of the pipeline fluid in the cooling system is usually between 2 and 5 seconds, the system sets the sliding time window width to a fixed preset of 3 seconds to ensure the time alignment accuracy of the cross-correlation calculation. For example, when the system calculates a static offset of 1.2 kPa corresponding to a 1°C increase in temperature within a certain window, it generates a pressure correction base sequence containing corresponding base pressure offset compensation based on this mapping ratio and the current temperature value. The system extracts the calibration temperature and rated base pressure of the cooling system under no-load and normal temperature conditions as the starting point for calculation. It subtracts the calibration temperature from the real-time low-frequency trend component of the temperature to obtain a relative temperature rise sequence. Then, it multiplies this relative temperature rise sequence by the mapping ratio of 1.2 kPa to obtain the theoretical pressure increment sequence. Finally, the system adds the theoretical pressure increment sequence to the rated base pressure point by point, splicing them together to form a continuous time series curve that dynamically and smoothly changes with temperature fluctuations, thus generating the pressure correction base sequence.

[0022] Furthermore, a temperature fluctuation disturbance mapping model is constructed by nonlinearly superimposing the pressure correction base with the transient thermal shock pulse identified based on the high-frequency detail components of temperature. In one implementation, the system first extracts the set of extreme points whose envelope amplitude exceeds a preset perturbation judgment threshold from the high-frequency detail components of temperature, and marks them as transient thermal shock pulses caused by a sudden surge in charging current. Subsequently, the system uses a second-order polynomial transfer function to perform nonlinear superposition fitting in the time domain dimension between the pressure correction base sequence representing steady-state thermal expansion and contraction and the transient thermal shock pulse sequence representing transient changes. It should be noted that the preset perturbation judgment threshold is determined by grid search on a historical normal operation dataset. The system performs feature matching within the candidate threshold range of 0.1℃ to 0.5℃, and finally selects 0.3℃, which minimizes the false alarm rate of the shock, as the fixed judgment threshold. For example, the system combines the basic offset model with the transient pulse shock features to output a complete temperature fluctuation disturbance mapping model, which can dynamically output the theoretically expected pressure value caused by temperature changes at any sampling time based on the real-time input temperature features.

[0023] Finally, the distortion degree of the preprocessed pressure signal under different preset temperature gradients is analyzed using a temperature fluctuation interference mapping model to determine the temperature fluctuation interference mode. In one embodiment, the system calculates the difference between the actual preprocessed pressure signal at the current moment and the theoretical expected pressure value output by the temperature fluctuation interference mapping model to obtain a distortion feature sequence. The time-domain envelope and frequency-domain energy distribution of this distortion feature sequence are mapped to the corresponding preset temperature gradient interval, and specific interference mode categories are classified by Euclidean distance comparison. It is worth noting that the division interval of the preset temperature gradient is calculated equidistantly based on the rated operating temperature range of the charging pile cooling equipment. For example, according to the equipment design and operation specifications, the temperature gradient is divided into 5°C increments from 0°C to 60°C to cover the full load condition. For example, in the temperature gradient interval of 35°C to 40°C, the system calculates that the actual pressure deviates from the theoretical model output by 2.5 kPa. Based on this distortion amplitude and occurrence gradient, the system indexes and classifies it as a specific mid-temperature step-type temperature fluctuation interference mode.

[0024] In step S12, historical correlation data matching the temperature fluctuation interference pattern needs to be acquired and multi-dimensional feature clustering is performed to obtain interference feature clusters. Dynamic correction coefficients are determined based on these interference feature clusters, and deviation compensation operations are performed on the original pressure data stream using these dynamic correction coefficients to obtain a corrected pressure signal, including: Temperature and pressure records that match the temperature fluctuation disturbance pattern are retrieved from a preset time series database and used as historical correlation data. Extract the temperature change rate feature and pressure fluctuation amplitude feature from the historical associated data, use the temperature change rate feature and pressure fluctuation amplitude feature to perform multidimensional feature clustering on the historical associated data to obtain interference feature clusters, and extract the distribution parameters of the interference feature clusters to calculate the dynamic correction coefficient. The original pressure data stream is subjected to nonlinear deviation compensation calculation using the dynamic correction coefficient to obtain a preliminary corrected pressure data stream. The mapping deviation value is calculated by comparing the initial corrected pressure data stream with the preset interference mode mapping relationship. Determine whether the mapping deviation value is greater than a preset secondary correction trigger threshold; If it is greater than, then a preset extreme pressure sample is extracted from the historical associated data for secondary mapping reconstruction. The secondary mapping reconstruction result of the preset extreme pressure sample is used to perform secondary deviation calibration on the preliminary corrected pressure data stream to obtain the corrected pressure signal. If it is not greater than, then the preliminary corrected pressure data stream is used as the corrected pressure signal.

[0025] First, temperature and pressure records matching the temperature fluctuation interference pattern are retrieved from a pre-defined time-series database and used as historical correlation data. In one implementation, the system connects to a local or cloud-based time-series database storing the entire lifecycle operation logs of the charging pile, and performs high-concurrency condition matching using the currently determined temperature fluctuation interference pattern category label as an index. It should be noted that the time-series database accumulates and stores massive amounts of high-frequency sampling node data of the device under different seasons and charging loads. For example, if the currently determined interference pattern is a mid-temperature step-type temperature fluctuation interference pattern, the system automatically retrieves all historical operating segments with this pattern label from the database over the past year, extracts 2000 high-dimensional data records containing synchronous temperature rise and accompanying pressure fluctuations, packages them as historical correlation data, and provides ample data prior support for subsequent algorithms.

[0026] Secondly, the temperature change rate characteristics and pressure fluctuation amplitude characteristics are extracted from the historical correlation data. These characteristics are then used to perform multi-dimensional feature clustering on the historical correlation data to obtain interference feature clusters. The distribution parameters of these interference feature clusters are extracted to calculate the dynamic correction coefficients. In one implementation, the system traverses each time-series slice of the historical correlation data, calculates the first derivative of temperature within a unit time window to obtain the temperature change rate characteristics, and statistically analyzes the pressure peak-to-valley difference within the corresponding time window to obtain the pressure fluctuation amplitude characteristics.

[0027] Subsequently, the system employs the density-based spatial clustering algorithm DBSCAN to perform unsupervised clustering of the extracted historical feature points. Before performing clustering, the system first calculates the K-distance curve of the historical feature point set and extracts the distance value corresponding to the maximum inflection point of the curve as a preset neighborhood radius parameter. Simultaneously, a preset minimum number of contained points is set based on the dimension of the feature space. Using the preset neighborhood radius parameter and the preset minimum number of contained points, the system traverses all feature points, connecting core points and their density-reachable feature points and grouping them into the same high-dimensional spatial region. At the same time, isolated feature points that are not density-reachable are marked as discrete abnormal noise points and removed, thereby accurately outputting interference feature clusters that characterize typical operating conditions.

[0028] It is worth noting that the system extracts the geometric center coordinates and covariance matrix of the interference feature cluster as distribution parameters, and substitutes them into a preset compensation factor calculation formula to generate dynamic correction coefficients. Specifically, the system analyzes the most representative temperature change rate center value and pressure fluctuation amplitude center value from the geometric center coordinates, calculates the absolute ratio of the pressure fluctuation amplitude center value to the temperature change rate center value, and obtains the environmental temperature sensitivity coefficient that quantifies the thermal expansion and contraction effect. The system uses the eigenvalues ​​of the covariance matrix to calculate the distribution dispersion as a confidence weight, and multiplies the confidence weight by the environmental temperature sensitivity coefficient, then substitutes it into a preset natural exponential decay compensation formula for calculation, outputting a dynamic correction coefficient between 0 and 1.

[0029] Next, nonlinear deviation compensation calculations are performed on the original pressure data stream using dynamic correction coefficients to obtain a preliminary corrected pressure data stream. The preliminary corrected pressure data stream is then compared with a preset interference mode mapping relationship to calculate the mapping deviation value. In one embodiment, the system constructs a temperature-pressure compensation polynomial function containing a first-order linear term and a second-order nonlinear term, and uses the dynamic correction coefficients obtained in the previous steps as the core nonlinear gain parameter in this compensation function. The system reads the real-time acquired original pressure data stream frame by frame, along with the synchronously acquired temperature rise amplitude value, and substitutes this temperature rise amplitude value into the temperature-pressure compensation polynomial function to calculate the theoretical pressure offset caused by the thermodynamic expansion or contraction of the coolant at the current sampling moment.

[0030] Subsequently, the system directly subtracts the theoretical pressure offset from the corresponding values ​​of the original pressure data stream, thereby accurately eliminating complex nonlinear pressure drift in the data stream and outputting a preliminary corrected pressure data stream with a stable baseline. Next, the system retrieves the standard pressure time-series curve generated during factory calibration under constant temperature and leak-free conditions, using it as a pre-set interference mode mapping relationship based on pure fluid dynamics. The system extracts the standard pressure time-series curve with the same time span as the preliminary corrected pressure data stream, calculates the sum of squares of the numerical differences between the two at all aligned time-series nodes, takes the mean of the sum of squares, and then performs a square root operation to obtain the root mean square error characterizing the overall difference between the two waveforms. This physical error is then directly quantified as a mapping deviation value.

[0031] Finally, it is determined whether the mapping deviation value is greater than the preset secondary correction trigger threshold, so as to adaptively switch the deviation calibration path. The preset secondary correction trigger threshold is determined based on the statistical distribution law of historical normal samples. Specifically, during the initial operation of the equipment or the periodic maintenance phase, the system collects a large set of mapping deviation values ​​after preliminary correction under a large number of leak-free and healthy conditions, calculates the mean and standard deviation of this set, and follows the confidence interval principle in statistics to set the preset secondary correction trigger threshold as the mean plus twice the standard deviation. This setting aims to establish a statistical boundary covering more than 95% of the fluctuations in normal operating conditions, ensuring that normal environmental interference will not excessively trigger complex computing power. In one embodiment, the preset secondary correction trigger threshold is fixed at 2.0 kPa. When the calculated mapping deviation value is greater than this threshold (for example, reaching 3.5 kPa), it indicates that the current cooling system has encountered extreme temperature changes or rare operating conditions that exceed the coverage of the conventional clustering model, and a single dynamic correction coefficient can no longer effectively suppress the interference. At this time, the system triggers a deep reconstruction mechanism to extract extreme pressure samples at the edge of the historical correlation data as a support vector set.

[0032] Specifically, the system employs a support vector regression algorithm with a Gaussian radial basis function kernel to construct a secondary calibration model. The system packages the current preliminary corrected pressure data stream and the synchronized temperature change rate as input feature vectors, implicitly mapping these input feature vectors to a high-dimensional nonlinear feature space using the Gaussian radial basis function kernel. Subsequently, the system calculates the inner product distance between the input feature vectors and the support vector set, constructs a nonlinear regression decision function by solving for Lagrange multipliers, and predicts the secondary deviation compensation sequence for the current extreme operating condition. The system subtracts the secondary deviation compensation sequence point by point from the preliminary corrected pressure data stream, performing secondary deviation calibration to forcibly remove abnormal drift and output a high-fidelity corrected pressure signal. Conversely, if the calculated mapping deviation value is less than or equal to the threshold (e.g., 1.5 kPa), it indicates that the current operating condition is within the typical feature cluster coverage range, and the first nonlinear compensation has sufficiently eliminated the rigid interference caused by temperature fluctuations. The system directly skips the computationally intensive secondary reconstruction step and outputs the preliminary corrected pressure data stream directly as the corrected pressure signal. This dual-track parallel adaptive correction mechanism balances the prediction accuracy under extreme conditions with the real-time efficiency under normal monitoring scenarios.

[0033] In step S13, it is necessary to extract the high-frequency components of pressure fluctuations from the corrected pressure signal, calculate the fluctuation deviation of the high-frequency components, determine the real-time fluctuation level based on the fluctuation deviation, and when it is determined that the real-time fluctuation level exceeds a preset dynamic range threshold, feature extraction and mode decomposition are performed on the high-frequency components of pressure fluctuations to obtain a leakage fluctuation feature description, including: A fast Fourier transform is performed on the corrected pressure signal to obtain pressure spectrum distribution data, and a preset bandpass filter is used to filter and reduce noise in the pressure spectrum distribution data to separate the high-frequency components of pressure fluctuations. Extract reference spectral features from the historical records of similar interference patterns, correlate and compare the high-frequency components of the pressure fluctuation with the reference spectral features, and calculate the fluctuation deviation. The fluctuation deviation is retrieved from the preset dynamic range probability distribution table, and the corresponding extreme value distribution probability is extracted. The extreme value distribution probability is used as the real-time fluctuation level. Determine whether the real-time fluctuation level is greater than a preset dynamic range threshold; If it is not greater than, then the preset no-leakage steady-state characteristic data is extracted and used as the leakage fluctuation characteristic description. If it is greater than that, then variational mode decomposition is performed on the high-frequency components of the pressure fluctuation to extract the intrinsic mode components; The intrinsic mode components are mapped to the time-frequency domain using the Hilbert transform to determine the mode evolution trajectory. The decay rate characteristics reflecting the physical properties of leakage are extracted based on the modal evolution trajectory. The decay rate feature is input into a preset leakage fluctuation feature library for similarity matching calculation. When the matching similarity reaches a preset confidence level, the successfully matched feature data is used as the leakage fluctuation feature description.

[0034] First, a Fast Fourier Transform (FFT) is performed on the corrected pressure signal to obtain pressure spectrum distribution data. Then, a preset bandpass filter is used to filter and reduce noise in the pressure spectrum distribution data, separating the high-frequency components of the pressure fluctuations. It should be noted that the cutoff frequency range of this preset bandpass filter is determined based on the turbulent acoustic frequency band caused by minute leaks in fluid dynamics. The high-frequency noise from fluid jets generated by minute leaks is mainly concentrated between 500Hz and 2kHz. Therefore, the system presets the lower cutoff frequency of the bandpass filter to 500Hz to shield the fundamental frequency of the liquid-cooled pump's mechanical vibration, and the upper cutoff frequency to 2kHz to eliminate high-frequency electromagnetic interference. For example, after a corrected pressure signal undergoes a FFT, its energy is widely distributed in the 0 to 5kHz range. After being filtered by the 500Hz to 2kHz bandpass filter, the system successfully extracts the high signal-to-noise ratio high-frequency components of the pressure fluctuations that contain only potential leak characteristics.

[0035] Secondly, reference spectral features are extracted from the preset historical records of similar interference patterns. The high-frequency components of pressure fluctuations are correlated and compared with the reference spectral features to calculate the fluctuation deviation. In one embodiment, the system retrieves the historical records of similar interference patterns under fault-free and healthy conditions from the local database by searching the interference pattern labels determined in the previous steps, and extracts the reference spectral features from the records. The system calculates the Pearson correlation coefficient between the amplitude vector of the high-frequency components of pressure fluctuations acquired in real time and the amplitude vector of the reference spectral features. Given that the Pearson correlation coefficient positively represents the linear similarity of the spectral waveform (the closer the value is to 1, the more consistent the waveform), and risk warning requires quantifying the difference features, the system constructs a preset difference conversion formula to calculate the fluctuation deviation, where the fluctuation deviation is equal to a constant 1 minus the real-time calculated Pearson correlation coefficient, thereby converting the similarity index into a feature index that quantifies abnormal differences.

[0036] Next, the system searches a preset dynamic range probability distribution table based on the fluctuation deviation, extracts the corresponding extreme value distribution probability, and uses this extreme value distribution probability as the real-time fluctuation level. In one implementation, the system substitutes the calculated fluctuation deviation value into the preset dynamic range probability distribution table for index lookup. It should be noted that this preset dynamic range probability distribution table is constructed by fitting kernel density estimation data to a large amount of historical pressure fluctuation data during normal charging cycles. Specifically, the kernel density estimation preferably uses a Gaussian kernel function to ensure the global continuity and smoothness of the probability density curve, and adaptively solves for the optimal smoothing bandwidth parameter using a maximum likelihood cross-validation algorithm, thereby strictly mapping the cumulative distribution probability of different deviation values ​​under leak-free conditions. For example, the system searches the distribution table with a fluctuation deviation of 0.18, extracts the extreme value distribution probability of this deviation under normal conditions as 5%, and directly uses this extreme value distribution probability as the current real-time fluctuation level.

[0037] It should be noted that the preset dynamic range threshold is set based on statistical anomaly detection standards. The system pre-extracts the 95th percentile of the fluctuation level in historical leak-free sample sets as a safety boundary and fixes the preset dynamic range threshold at this critical value. In one embodiment, when the real-time fluctuation level is determined to be greater than this threshold, it indicates a significant high-frequency anomaly inside the pipeline. The system immediately calls the variational mode decomposition algorithm to adaptively divide the high-frequency components of the pressure fluctuation into frequency bands and extract the intrinsic mode components with the most concentrated energy. Subsequently, the system performs a Hilbert transform on this intrinsic mode component to generate a modal evolution trajectory in the three-dimensional time-frequency domain space, and calculates the logarithmic slope of energy decay along the time axis of this trajectory to extract the decay rate feature. Further, the system sends the decay rate feature to a preset leakage fluctuation feature library for sequence similarity matching. If the matching result is determined to meet the preset confidence standard, the corresponding feature data in the library is extracted as a high-fidelity description of the leakage fluctuation feature.

[0038] If the value is not greater than the threshold, the system directly retrieves the baseline zero-bias feature vector from the local system, which is the preset no-leakage steady-state feature data as the description of leakage fluctuation features.

[0039] In step S14, it is necessary to calculate the deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark, map the deviation vector using a preset probability mapping function to obtain a leakage risk probability value, and compare the leakage risk probability value with a preset dynamic risk warning threshold to obtain a preliminary risk level, including: A preset steady-state operating benchmark is retrieved, and the leakage fluctuation characteristic description is compared with the preset steady-state operating benchmark by Euclidean distance calculation to obtain the deviation vector in the multi-dimensional feature space. The distribution position of the deviation vector in the multidimensional feature space is determined, and a nonlinear mapping operation is performed on the distribution position using a preset probability mapping function to calculate the leakage risk probability value that reflects the possibility of leakage. The leakage risk probability value is input into a preset real-time monitoring sequence, and the leakage risk probability value is compared with a preset dynamic risk warning threshold frame by frame to calculate the abnormal increment that exceeds the safe range. Determine whether the abnormal increment falls within a specific interval of a preset threshold trigger matrix; If it falls into the range, the corresponding early warning response level is matched by the index of the preset threshold trigger matrix, and the early warning response level is determined as a preliminary risk level that matches the leakage risk probability value. If it does not fall into the category, the current system is determined to be in a security monitoring state, and the preset minimum security level is extracted as the preliminary risk level.

[0040] First, a preset steady-state operating benchmark is retrieved. The leakage fluctuation characteristic description is then compared with the preset steady-state operating benchmark using Euclidean distance calculation to obtain a deviation vector in a multi-dimensional feature space. In one implementation, the system reads a preset steady-state operating benchmark containing multi-dimensional features such as standard pressure deviation, base decay rate, and steady-state frequency from the equipment's underlying benchmark parameter library. The system aligns the leakage fluctuation characteristic description output from the previous steps with this benchmark vector in a unified high-dimensional feature space, and performs a square root operation of the sum of squares of the differences in each dimension of the Euclidean distance to quantify the absolute geometric distance between the current fluctuation state and the healthy state, outputting the deviation vector. It should be noted that this preset steady-state operating benchmark is determined based on the average value of the equipment's factory acceptance test and the first month of fault-free operation data, representing the baseline of the cooling system under optimal operating conditions. For example, if the leakage fluctuation characteristics are described by a feature value of 2.5 kPa and an attenuation rate of 0.02 kPa / ms in a specific frequency band, the system calculates the Euclidean distance between it and the reference vector (1.0 kPa, 0.00 kPa / ms) to obtain a deviation vector pointing to a specific high-frequency leakage direction.

[0041] Secondly, the distribution position of the deviation vector in the multidimensional feature space is determined, and a preset probability mapping function is used to perform a nonlinear mapping operation on the distribution position to calculate the leakage risk probability value reflecting the likelihood of leakage. In one implementation, the system projects the coordinate system of the deviation vector onto a pre-constructed fault space model to determine its relative distribution position. Subsequently, the system calls a preset probability mapping function, such as the Sigmoid nonlinear normalization function, to smoothly transform the dimensionless position deviation distance to a mathematical range of 0 to 100%, and outputs the leakage risk probability value.

[0042] It is worth noting that the center point and scaling factor of the preset probability mapping function are determined by parameter fitting based on the statistical distribution of historical leakage accidents in the field. The system fits the relationship curve between the characteristic deviation of a large number of known leakage events and the actual physical leakage amount, setting that when the deviation distance reaches a certain statistical empirical boundary, the risk probability exponentially approaches 100%. For example, for a deviation vector with an extremely high deviation, whose distribution location is close to the high-risk leakage zone at the spatial edge, the system directly converts it into a leakage risk probability value of 87.5% through probability mapping function calculation.

[0043] Next, the leakage risk probability value is input into a preset real-time monitoring sequence. The leakage risk probability value is then compared frame-by-frame with a preset dynamic risk warning threshold to calculate the abnormal increment exceeding the safe range. In one implementation, the system fills the instantaneously calculated leakage risk probability value into the preset real-time monitoring sequence in timestamp order, forming a continuous probability timeline. The system slides along the timeline, directly subtracting the probability value of each frame from the preset dynamic risk warning threshold, and extracting the positive difference as the abnormal increment. It should be noted that the preset dynamic risk warning threshold is dynamically calculated and issued based on the current charging power load of the device.

[0044] Specifically, the system pre-conducts destructive micropore leakage pressure tests on the same model of liquid cooling system in a laboratory environment, statistically analyzes a large number of risk probability distribution curves before the occurrence of early micro-leakage, and extracts the inflection point value that optimally balances the false alarm rate and the missed alarm rate from these probability distribution curves as the safety benchmark value. In one embodiment, this safety benchmark value is strictly calibrated to 40%. Based on this, the system obtains the current charging power load ratio of the device in real time and substitutes this load ratio into a preset adaptive adjustment function to dynamically output the final warning threshold. During the full-load fast charging stage, due to the surge in internal heat generation, the safety tolerance space of the fluid circulation pressure system is significantly compressed. The system uses the adaptive adjustment function to dynamically lower the warning threshold from the benchmark value of 40% to 30% to improve the system's sensitivity to extremely weak anomalies. During the no-load standby stage, due to the stable pipeline pressure and extremely low physical risk, the system raises the threshold to 50% to filter out false fluctuations caused by environmental white noise. For example, when the system is in a high-power charging state and the dynamic risk warning threshold is calculated and set to 30%, if the leakage risk probability value of the current frame is 45%, the system calculates the difference to obtain an abnormal increment of 15%.

[0045] Finally, it is determined whether the abnormal increment falls into a specific range of the preset threshold trigger matrix; if it does, the corresponding warning response level is matched with the index of the preset threshold trigger matrix, and the warning response level is determined as the preliminary risk level that matches the leakage risk probability value; if it does not fall into the range, it is determined that the current system is in a safe monitoring state, and the preset minimum safety level is extracted as the preliminary risk level.

[0046] In one implementation, the system invokes a preset threshold trigger matrix containing multiple tiered anomaly intervals and mapping standards, using the calculated anomaly increment value as the query key to scan each specific interval. It's worth noting that the interval division of this preset threshold trigger matrix is ​​based on common sense in security engineering defense, strictly dividing it into a light observation zone of [0%, 10%], a moderate intervention zone of [10%, 25%], and an emergency shutdown zone of [25%, 100%]. For example, if the preceding anomaly increment is 15%, the system determines it falls within the [10%, 25%] interval, and then extracts the corresponding Level 2 yellow warning level through matrix indexing, outputting it as the preliminary risk level. Conversely, if the anomaly increment is 0% or negative, the system extracts Level 0 green normal, representing absolute safety, i.e., the preset minimum safety level, as the current preliminary risk level output.

[0047] In step S15, a thermodynamic field value needs to be constructed based on the temperature data stream, and the absolute deviation of the conduction rate needs to be calculated. The absolute deviation of the conduction rate is then weighted and fused with the preliminary risk level to obtain a state gain. The state gain is used to update the leakage risk probability value to obtain a refining risk assessment result, including: The real-time temperature data stream is processed by spatiotemporal alignment of multiple sensor nodes to generate a thermodynamic field value that reflects the flow state of the cooling medium. Extract the temperature change amplitude of the thermal field value within a preset time window, calculate the temperature conduction rate of the current medium based on the temperature change amplitude, and calculate the absolute value of the difference between the temperature conduction rate and the preset environmental conduction benchmark to obtain the absolute deviation of the conduction rate. Preset deviation fusion weights and preset level fusion weights are assigned to the absolute deviation of the conduction rate and the preliminary risk level, respectively. The state gain reflecting the impact of the current operating condition fluctuation is obtained by weighted summation using the preset deviation fusion weights and the preset level fusion weights. The state gain is input into a preset nonlinear mapping function for probability distribution mapping to generate a real-time calibration correction factor. The real-time calibration correction factor is then applied to the temporal evolution trend of the leakage risk probability value to update and adjust the state, thereby obtaining a dynamically adjusted risk probability value. The data rationality of the dynamically adjusted risk probability value is verified by using a preset multidimensional confidence verification model to determine whether the dynamically adjusted risk probability value passes the verification. If approved, the dynamically adjusted risk probability value will be used as the refined risk assessment result. If the test fails, a preset safety backup risk value is retrieved and used as the refining risk assessment result.

[0048] First, the real-time acquired temperature data stream undergoes spatiotemporal alignment processing across multiple sensor nodes to generate a thermodynamic field value reflecting the flow state of the cooling medium. The temperature change amplitude within a preset time window is extracted from the thermodynamic field value. Based on this amplitude, the current temperature conduction rate of the medium is calculated, and the difference between this temperature conduction rate and a preset environmental conduction benchmark is calculated, with the absolute value taken to obtain the absolute deviation of the conduction rate. In one implementation, the system resamples the asynchronous time series transmitted from multiple temperature sensor nodes deployed within the liquid-cooled charging pile using a cubic spline interpolation algorithm, unifying the timestamps to construct a multidimensional matrix of thermodynamic field values. Subsequently, the system extracts this thermodynamic field value sequence, calculates the extreme temperature difference within a specific time period to obtain the temperature change amplitude, and divides it by the time span to calculate the temperature conduction rate.

[0049] It should be noted that the preset time window width is determined based on the nominal circulation cycle of the charging pile's cooling pump. Typically, it takes approximately 5 seconds for the coolant to complete one full internal circulation cycle within the pipeline. Therefore, this time window is preset to 5 seconds to fully assess the heat exchange state of one fluid cycle. Simultaneously, the preset environmental conduction benchmark is the ideal heat dissipation rate calculated based on the coolant's factory-spatiotemporal specific heat capacity and rated flow rate. For example, the system constructs the current thermodynamic field matrix through spatiotemporal alignment, and within the 5-second window, a temperature rise of 2℃ is observed, calculating the actual temperature conduction rate to be 0.4℃ / s. The system retrieves the environmental conduction benchmark of 0.1℃ / s for this model of equipment, subtracts the two, and takes the absolute value to calculate an absolute deviation in conduction rate of 0.3℃ / s.

[0050] Secondly, preset deviation fusion weights and preset level fusion weights are assigned to the absolute deviation of the conduction rate and the preliminary risk level, respectively. The state gain, reflecting the impact of fluctuations in the current operating condition, is calculated by weighted summation using these preset deviation fusion weights and preset level fusion weights. In one implementation, the system normalizes and aligns the absolute deviation of the conduction rate and the preliminary risk level output from the previous step, then multiplies them by their corresponding weight parameters, and obtains the state gain through linear superposition. It is worth noting that the preset deviation fusion weights and preset level fusion weights are determined through a grid search on a historical dataset. The system performs leakage detection tests with candidate weight combinations ranging from 0.1 to 0.9, ultimately selecting the combination that minimizes the overall false alarm rate and is most sensitive to minor leaks. The preset deviation fusion weight is fixed at 0.4, and the preset level fusion weight is fixed at 0.6. For example, if the normalized absolute deviation of the conduction rate is 0.5 and the current preliminary risk level normalization score is 0.8, the system calculates 0.5 multiplied by 0.4 plus 0.8 multiplied by 0.6, and outputs a state gain of 0.68, which comprehensively considers the dual impact of transient thermodynamic fluctuations and steady-state characteristic warnings.

[0051] Next, the state gain is input into a preset nonlinear mapping function for probability distribution mapping processing to generate a real-time calibration correction factor. This real-time calibration correction factor is then applied to the temporal evolution trend of the leakage risk probability value to update the state, resulting in a dynamically adjusted risk probability value. In one implementation, the system calls a preset nonlinear mapping function of exponential decay or logistic type, inputs the state gain into it, and outputs a product coefficient that fluctuates around a baseline value of 1.0, which is the real-time calibration correction factor. Subsequently, the system directly multiplies this factor by the leakage risk probability value obtained from the initial assessment to complete the state update. It should be noted that the slope parameter of this preset nonlinear mapping function is set according to the security defense conservatism level to ensure that when the state gain increases sharply, the correction factor amplifies exponentially to expose the risk in advance. For example, for the aforementioned state gain of 0.68, the system calculates a real-time calibration correction factor of 1.15 using the nonlinear mapping function; the system obtains the original leakage risk probability value of 45%, multiplies it by 1.15, and obtains a dynamically adjusted risk probability value of 51.75%.

[0052] Finally, a preset multi-dimensional confidence verification model is used to verify the data rationality of the dynamically adjusted risk probability value, determining whether the value passes the verification. If it passes, the dynamically adjusted risk probability value is used as the refining risk assessment result; if it fails, a preset safety backup risk value is retrieved and used as the refining risk assessment result. In one implementation, the system calls the multi-dimensional confidence verification model to check whether the jump amplitude of the dynamically adjusted risk probability value per unit time conforms to physical laws. It is worth noting that the jump tolerance limit in this model is determined based on the deformation fracture mechanical limit of the cooling pipe material. Even in the most severe instantaneous pipe burst, the pressure and probability evolution sensed by the sensor have a physical lag, making it impossible for a jump of more than 50% without warning to occur within milliseconds. Therefore, the jump tolerance limit is set to ±30% per second. For example, if the system adjusts the risk probability from 45% to 51.75%, the jump is only 6.75%, far below the physical limit of 30%, and the system determines that it passes the verification and officially outputs 51.75% as the refined risk assessment result. Conversely, if the dynamic adjustment risk probability value soars to 99% (jumping by 54%) due to a momentary short circuit in the sensor, the system determines that it fails the verification, immediately blocks the transmission of this absurd data, and extracts the 45% probability that passed the verification in the previous second plus a conservative redundancy of 5%, i.e., 50%, as a preset safety backup risk value, and outputs it as the refined risk assessment result for this period, thereby effectively avoiding system crashes and malfunctions caused by hardware glitches.

[0053] In step S16, when the refining risk assessment result is determined to be greater than the preset final warning threshold, a tiered alarm mechanism is triggered, and log data is generated and input into the distributed storage system for time-series aggregation to obtain fault diagnosis records, including: Determine whether the refining risk assessment result is greater than the preset final warning threshold; If the risk level is greater than the threshold, a tiered alarm mechanism is triggered. Based on the refined risk assessment results, a preset push channel is matched to distribute alarm information containing risk level and location information to the corresponding receiving terminal. The real-time risk value and triggering conditions at the time of alarm triggering are encapsulated into raw log data. The raw log data is input into the distributed storage system, and the structured information with high-precision timestamps is written into the redundant shards of the preset storage cluster through the data nodes. By using a preset index association technique, historical risk values ​​within the storage cluster are aggregated in a time series to generate log data with time-series logic. A consistency check operation is performed on the log data. After the check passes, traceable data that supports multi-dimensional retrieval is constructed as a fault diagnosis record. If the value is not greater than the specified value, the normal monitoring status is maintained, the refining risk assessment result is encapsulated into a regular operation log, and the regular operation log is input into the distributed storage system as the fault diagnosis record.

[0054] First, the system determines whether the refining risk assessment result exceeds a preset final warning threshold. In one implementation, the system acquires the refining risk assessment result from the dynamic refining output of the preceding steps in real time and compares it numerically with the preset final warning threshold. It should be noted that this preset final warning threshold is determined statistically based on the safety boundary of the equipment's thermal runaway critical point and historical severe leakage failure samples. To avoid frequent false alarms caused by minor disturbances interfering with normal operation and maintenance, the system strictly sets it to 80%. For example, if the system calculates the current refining risk assessment result to be 85%, which is greater than the 80% warning threshold, the system immediately switches to the emergency response branch; conversely, if the result is only 25%, it enters the normal operation branch.

[0055] Secondly, when the refining risk assessment result is determined to be greater than the preset final warning threshold, a tiered alarm mechanism is triggered. Based on the refining risk assessment result, a preset push channel is matched, and alarm information containing risk level and location information is distributed to the corresponding receiving terminals. The real-time risk value and triggering conditions at the time of alarm triggering are encapsulated into raw log data. In one implementation, the system calls a built-in tiered response routing table, mapping 80% to 90% of the risk range to orange alerts and above 90% of the risk range to red alerts. The system automatically matches preset push channels according to different levels; for example, orange alerts are pushed to the handheld terminals of nearby maintenance personnel via a mobile application, while red alerts are sent directly to the central control and dispatch center via SMS and high-priority network signaling. It is worth noting that while distributing alarm information, the system dynamically retrieves the current refining risk assessment result, trigger threshold parameters, and the current geographical coordinates of the equipment from the underlying memory, and encapsulates them into raw log data containing a complete on-site snapshot through serialization operations.

[0056] Next, the raw log data is input into the distributed storage system, and the data nodes write the structured information with high-precision timestamps into the redundant shards of the preset storage cluster. In one implementation, the system uses an asynchronous network transmission protocol to push the raw log data to the cloud-based distributed storage system and calls the system clock to add a millisecond-level high-precision timestamp to the data stream. Subsequently, the system controls the underlying data nodes to perform data write operations, synchronously writing the structured information into multiple redundant shards of the preset storage cluster. It should be noted that the number of replicas of the redundant shards is set based on industrial-grade data disaster recovery standards. To ensure that data is not lost under catastrophic conditions such as extreme power outages or hardware failures, the system presets the number of replicas to 3, distributed across storage nodes in different physical racks. For example, a leak alarm structured information carrying a millisecond-level timestamp is simultaneously written by the system into independent hard disk sectors of nodes A, B, and C to ensure the absolute security of critical data.

[0057] Subsequently, the system uses a pre-defined index association technique to perform time-series aggregation of historical risk values ​​within the storage cluster, generating log data with time-series logic. In one implementation, the system uses the unique device identification code of the current alarm charging pile and the alarm time window as a combined primary key, and invokes a pre-defined B+ tree index association technique to reverse-retrieve all historical risk value records of the device in the distributed storage cluster over the past 24 hours. The system then performs time-series aggregation and smooth splicing of these discrete historical risk slices according to the chronological order of their timestamps, generating a complete log data that reflects the entire evolution of the device from a minor leak to the final triggering of a red-line alarm.

[0058] Furthermore, a consistency check is performed on the log data. After the check passes, traceable data supporting multi-dimensional retrieval is constructed as a fault diagnosis record. In one implementation, the system extracts the aggregated log data, calculates its secure hash value (such as the SHA-256 algorithm), and compares it with the hash values ​​of the original files stored in each redundant shard for consistency verification. After the check confirms that the aggregated data has not experienced block corruption or network transmission interruption, the system adds multiple dimensions of retrieval engine tags, such as time, spatial location, and fault characteristic type, to the log data, constructing traceable data supporting multi-dimensional retrieval. This data is then formally archived as a fault diagnosis record output for subsequent manual accident review or secondary iteration training of the diagnostic algorithm model.

[0059] Finally, if the value is not greater than the threshold, the system maintains normal monitoring, encapsulates the refining risk assessment result into a regular operation log, and inputs the regular operation log into the distributed storage system as a fault diagnosis record. In one implementation, if the system determines that the current refining risk assessment result (e.g., the aforementioned 25%) is below the preset final warning threshold, it indicates that the current liquid cooling circulation system is operating healthily and has no obvious signs of impending failure. The system maintains the current cooling pump speed and monitoring polling frequency, maintaining normal monitoring. Simultaneously, the system only extracts the 25% health risk value and basic operating condition parameters, packages them as heartbeat status data into a regular operation log, and inputs it into the distributed storage system using a low-priority channel. This is directly archived as a fault diagnosis record at that sampling moment, thereby significantly reducing the system's alarm redundancy load while ensuring the integrity of the monitoring chain.

[0060] In summary, this invention discloses a fault detection method for liquid-cooled charging piles, including collecting temperature and pressure data to decouple temperature fluctuation interference, using historical feature clustering for dynamic pressure compensation, extracting high-frequency components and performing mode decomposition to obtain leakage characteristics, mapping risk probabilities and fusing thermodynamic field state gain to refine the evaluation results, and finally triggering graded alarms and recording and archiving the data. This invention effectively overcomes the interference of complex temperature fluctuations and strong background noise through a joint mechanism of multi-scale feature association decoupling, dynamic deviation correction, and global thermodynamic state constraints, achieving high-precision extraction, adaptive early warning, and reliable backtracking for early-stage minor leakage faults.

[0061] Reference Figure 2 The second embodiment of the present invention provides a liquid-cooled charging pile fault detection system, comprising: The interference mode construction module is used to collect the temperature data stream and the original pressure data stream of the liquid-cooled charging pile cooling system, and perform temperature fluctuation interference analysis based on the temperature data stream and the original pressure data stream to obtain the temperature fluctuation interference mode. The pressure signal correction module is used to acquire historical correlation data that matches the temperature fluctuation interference pattern and perform multi-dimensional feature clustering to obtain interference feature clusters. Based on the interference feature clusters, dynamic correction coefficients are determined, and deviation compensation operations are performed on the original pressure data stream using the dynamic correction coefficients to obtain the corrected pressure signal. The high-frequency feature extraction module is used to extract the high-frequency components of pressure fluctuation from the corrected pressure signal, calculate the fluctuation deviation of the high-frequency components of pressure fluctuation, determine the real-time fluctuation level based on the fluctuation deviation, and when it is determined that the real-time fluctuation level exceeds the preset dynamic range threshold, perform feature extraction and mode decomposition on the high-frequency components of pressure fluctuation to obtain a leakage fluctuation feature description. The preliminary risk assessment module is used to calculate the deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark, map the deviation vector using a preset probability mapping function to obtain the leakage risk probability value, and compare the leakage risk probability value with a preset dynamic risk warning threshold to obtain the preliminary risk level. The risk dynamic refining module is used to construct a thermodynamic field value based on the temperature data stream and calculate the absolute deviation of the conduction rate. The absolute deviation of the conduction rate is weighted and fused with the preliminary risk level to obtain a state gain. The state gain is used to update the leakage risk probability value to obtain the refining risk assessment result. The alarm and log recording module is used to trigger a graded alarm mechanism when the refining risk assessment result is determined to be greater than the preset final warning threshold, and to generate log data that is input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

[0062] It should be noted that the liquid-cooled charging pile fault detection system provided in this embodiment of the invention is used to execute all the process steps of the liquid-cooled charging pile fault detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0063] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0064] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A fault detection method for liquid-cooled charging piles, characterized in that, include: The temperature data stream and raw pressure data stream of the liquid-cooled charging pile cooling system are collected. Based on the temperature data stream and raw pressure data stream, temperature fluctuation interference analysis is performed to obtain the temperature fluctuation interference mode. Historical correlation data matching the temperature fluctuation interference pattern is acquired and multidimensional feature clustering is performed to obtain interference feature clusters. Dynamic correction coefficients are determined based on the interference feature clusters. Deviation compensation operation is performed on the original pressure data stream using the dynamic correction coefficients to obtain the corrected pressure signal. High-frequency components of pressure fluctuations are extracted from the corrected pressure signal, the fluctuation deviation of the high-frequency components of pressure fluctuations is calculated, and the real-time fluctuation level is determined based on the fluctuation deviation. When the real-time fluctuation level is determined to exceed the preset dynamic range threshold, feature extraction and mode decomposition are performed on the high-frequency components of pressure fluctuations to obtain a description of leakage fluctuation characteristics. The deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark is calculated. The deviation vector is mapped using a preset probability mapping function to obtain the leakage risk probability value. The leakage risk probability value is compared with a preset dynamic risk warning threshold to obtain a preliminary risk level. Thermodynamic field values ​​are constructed based on the temperature data stream and the absolute deviation of conduction rate is calculated. The absolute deviation of conduction rate is weighted and fused with the preliminary risk level to obtain the state gain. The state gain is used to update the leakage risk probability value to obtain the refining risk assessment result. When the refining risk assessment result is determined to be greater than the preset final warning threshold, a tiered alarm mechanism is triggered, and log data is generated and input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

2. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, The temperature data stream and raw pressure data stream of the liquid-cooled charging pile cooling system are collected. Based on the temperature data stream and raw pressure data stream, temperature fluctuation interference analysis is performed to obtain temperature fluctuation interference modes, including: Temperature data streams are acquired by high-precision temperature sensors deployed in the cooling system of liquid-cooled charging piles, and raw pressure data streams are acquired synchronously by pressure sensors. The temperature data stream is decomposed into multiple levels using a preset wavelet basis function to extract low-frequency temperature trend components and high-frequency temperature detail components as multi-scale temperature features. High-frequency random noise in the original pressure data stream is filtered out using a preset adaptive filter to obtain a preprocessed pressure signal; A sliding window cross-correlation calculation is performed on the preprocessed pressure signal and the low-frequency temperature trend component to determine the static temperature offset, and a pressure correction base is generated based on the static temperature offset. The pressure correction base is nonlinearly superimposed with the instantaneous thermal shock pulse identified based on the high-frequency detail components of the temperature to construct a temperature fluctuation interference mapping model. The temperature fluctuation interference mapping model is used to analyze the degree of distortion of the preprocessed pressure signal under different preset temperature gradients, and the temperature fluctuation interference mode is determined.

3. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, The process of acquiring historical correlation data matching the temperature fluctuation interference pattern and performing multidimensional feature clustering to obtain interference feature clusters, determining dynamic correction coefficients based on the interference feature clusters, and using the dynamic correction coefficients to perform deviation compensation operations on the original pressure data stream to obtain a corrected pressure signal includes: Temperature and pressure records that match the temperature fluctuation disturbance pattern are retrieved from a preset time series database and used as historical correlation data. Extract the temperature change rate feature and pressure fluctuation amplitude feature from the historical associated data, use the temperature change rate feature and pressure fluctuation amplitude feature to perform multidimensional feature clustering on the historical associated data to obtain interference feature clusters, and extract the distribution parameters of the interference feature clusters to calculate the dynamic correction coefficient. The original pressure data stream is subjected to nonlinear deviation compensation calculation using the dynamic correction coefficient to obtain a preliminary corrected pressure data stream. The mapping deviation value is calculated by comparing the initial corrected pressure data stream with the preset interference mode mapping relationship. Determine whether the mapping deviation value is greater than a preset secondary correction trigger threshold; If it is greater than, then a preset extreme pressure sample is extracted from the historical associated data for secondary mapping reconstruction. The secondary mapping reconstruction result of the preset extreme pressure sample is used to perform secondary deviation calibration on the preliminary corrected pressure data stream to obtain the corrected pressure signal. If it is not greater than, then the preliminary corrected pressure data stream is used as the corrected pressure signal.

4. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, The process involves extracting high-frequency components of pressure fluctuations from the corrected pressure signal, calculating the fluctuation deviation of these high-frequency components, determining the real-time fluctuation level based on the fluctuation deviation, and performing feature extraction and mode decomposition on the high-frequency components of the pressure fluctuations when the real-time fluctuation level exceeds a preset dynamic range threshold. This yields a description of the leakage fluctuation characteristics, including: A fast Fourier transform is performed on the corrected pressure signal to obtain pressure spectrum distribution data, and a preset bandpass filter is used to filter and reduce noise in the pressure spectrum distribution data to separate the high-frequency components of pressure fluctuations. Extract reference spectral features from the historical records of similar interference patterns, correlate and compare the high-frequency components of the pressure fluctuation with the reference spectral features, and calculate the fluctuation deviation. The fluctuation deviation is retrieved from the preset dynamic range probability distribution table, and the corresponding extreme value distribution probability is extracted. The extreme value distribution probability is used as the real-time fluctuation level. Determine whether the real-time fluctuation level is greater than a preset dynamic range threshold; If it is not greater than, then the preset no-leakage steady-state characteristic data is extracted and used as the leakage fluctuation characteristic description. If it is greater than that, then variational mode decomposition is performed on the high-frequency components of the pressure fluctuation to extract the intrinsic mode components; The intrinsic mode components are mapped to the time-frequency domain using the Hilbert transform to determine the mode evolution trajectory. The decay rate characteristics reflecting the physical properties of leakage are extracted based on the modal evolution trajectory. The decay rate feature is input into a preset leakage fluctuation feature library for similarity matching calculation. When the matching similarity reaches a preset confidence level, the successfully matched feature data is used as the leakage fluctuation feature description.

5. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, The process involves calculating the deviation vector between the leakage fluctuation characteristic description and a preset steady-state operating benchmark, mapping the deviation vector using a preset probability mapping function to obtain a leakage risk probability value, and comparing the leakage risk probability value with a preset dynamic risk warning threshold to obtain a preliminary risk level, including: A preset steady-state operating benchmark is retrieved, and the leakage fluctuation characteristic description is compared with the preset steady-state operating benchmark by Euclidean distance calculation to obtain the deviation vector in the multi-dimensional feature space. The distribution position of the deviation vector in the multidimensional feature space is determined, and a nonlinear mapping operation is performed on the distribution position using a preset probability mapping function to calculate the leakage risk probability value that reflects the possibility of leakage. The leakage risk probability value is input into a preset real-time monitoring sequence, and the leakage risk probability value is compared with a preset dynamic risk warning threshold frame by frame to calculate the abnormal increment that exceeds the safe range. Determine whether the abnormal increment falls within a specific interval of a preset threshold trigger matrix; If it falls into the range, the corresponding early warning response level is matched by the index of the preset threshold trigger matrix, and the early warning response level is determined as a preliminary risk level that matches the leakage risk probability value. If it does not fall into the category, the current system is determined to be in a security monitoring state, and the preset minimum security level is extracted as the preliminary risk level.

6. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, The process of constructing a thermal field value based on the temperature data stream and calculating the absolute deviation of the conduction rate, weighting and fusing the absolute deviation of the conduction rate with the preliminary risk level to obtain a state gain, and using the state gain to update the leakage risk probability value to obtain a refining risk assessment result includes: The real-time temperature data stream is processed by spatiotemporal alignment of multiple sensor nodes to generate a thermodynamic field value that reflects the flow state of the cooling medium. Extract the temperature change amplitude of the thermal field value within a preset time window, calculate the temperature conduction rate of the current medium based on the temperature change amplitude, and calculate the absolute value of the difference between the temperature conduction rate and the preset environmental conduction benchmark to obtain the absolute deviation of the conduction rate. Preset deviation fusion weights and preset level fusion weights are assigned to the absolute deviation of the conduction rate and the preliminary risk level, respectively. The state gain reflecting the impact of the current operating condition fluctuation is obtained by weighted summation using the preset deviation fusion weights and the preset level fusion weights. The state gain is input into a preset nonlinear mapping function for probability distribution mapping to generate a real-time calibration correction factor. The real-time calibration correction factor is then applied to the temporal evolution trend of the leakage risk probability value to update and adjust the state, thereby obtaining a dynamically adjusted risk probability value. The data rationality of the dynamically adjusted risk probability value is verified by using a preset multidimensional confidence verification model to determine whether the dynamically adjusted risk probability value passes the verification. If approved, the dynamically adjusted risk probability value will be used as the refined risk assessment result. If the test fails, a preset safety backup risk value is retrieved and used as the refining risk assessment result.

7. The fault detection method for liquid-cooled charging piles according to claim 1, characterized in that, When the refining risk assessment result is determined to be greater than the preset final warning threshold, a tiered alarm mechanism is triggered, and log data is generated and input into the distributed storage system for time-series aggregation to obtain fault diagnosis records, including: Determine whether the refining risk assessment result is greater than the preset final warning threshold; If the risk level is greater than the threshold, a tiered alarm mechanism is triggered. Based on the refined risk assessment results, a preset push channel is matched to distribute alarm information containing risk level and location information to the corresponding receiving terminal. The real-time risk value and triggering conditions at the time of alarm triggering are encapsulated into raw log data. The raw log data is input into the distributed storage system, and the structured information with high-precision timestamps is written into the redundant shards of the preset storage cluster through the data nodes. By using a preset index association technique, historical risk values ​​within the storage cluster are aggregated in a time series to generate log data with time-series logic. A consistency check operation is performed on the log data. After the check passes, traceable data that supports multi-dimensional retrieval is constructed as a fault diagnosis record. If the value is not greater than the specified value, the normal monitoring status is maintained, the refining risk assessment result is encapsulated into a regular operation log, and the regular operation log is input into the distributed storage system as the fault diagnosis record.

8. A liquid-cooled charging pile fault detection system, characterized in that, include: The interference mode construction module is used to collect the temperature data stream and the original pressure data stream of the liquid-cooled charging pile cooling system, and perform temperature fluctuation interference analysis based on the temperature data stream and the original pressure data stream to obtain the temperature fluctuation interference mode. The pressure signal correction module is used to acquire historical correlation data that matches the temperature fluctuation interference pattern and perform multi-dimensional feature clustering to obtain interference feature clusters. Based on the interference feature clusters, dynamic correction coefficients are determined, and deviation compensation operations are performed on the original pressure data stream using the dynamic correction coefficients to obtain the corrected pressure signal. The high-frequency feature extraction module is used to extract the high-frequency components of pressure fluctuation from the corrected pressure signal, calculate the fluctuation deviation of the high-frequency components of pressure fluctuation, determine the real-time fluctuation level based on the fluctuation deviation, and when it is determined that the real-time fluctuation level exceeds the preset dynamic range threshold, perform feature extraction and mode decomposition on the high-frequency components of pressure fluctuation to obtain a leakage fluctuation feature description. The preliminary risk assessment module is used to calculate the deviation vector between the leakage fluctuation characteristic description and the preset steady-state operating benchmark, map the deviation vector using a preset probability mapping function to obtain the leakage risk probability value, and compare the leakage risk probability value with a preset dynamic risk warning threshold to obtain the preliminary risk level. The risk dynamic refining module is used to construct a thermodynamic field value based on the temperature data stream and calculate the absolute deviation of the conduction rate. The absolute deviation of the conduction rate is weighted and fused with the preliminary risk level to obtain a state gain. The state gain is used to update the leakage risk probability value to obtain the refining risk assessment result. The alarm and log recording module is used to trigger a graded alarm mechanism when the refining risk assessment result is determined to be greater than the preset final warning threshold, and to generate log data that is input into the distributed storage system for time-series aggregation to obtain fault diagnosis records.

Images