A liquid-cooled charging cable cooling loop risk early warning method and system

Abstract

The present application relates to the technical field of industrial system state monitoring and intelligent early warning, and specifically discloses a liquid-cooled charging cable cooling loop risk early warning method and system, which acquires multi-dimensional time sequence data composed of liquid path-electricity-environment parameters and carries out pretreatment, identifies five types of working conditions, namely standby, low power, high power fast charging, peak overcharging and reduced power operation, based on the combination of K-means clustering and rule determination, carries out time sequence prediction according to the working condition by loading Kalman filter-ARX model or TCN-Transformer fusion model, extracts residual cumulative value, growth rate, multivariate consistency and confidence interval deviation degree four types of trend characteristics based on the predicted residual sequence, dynamically weights according to the working condition and combines the confidence coefficient of abnormal duration to obtain the final risk total score, finally divides the early warning level and outputs the graded charging power linkage control instruction.

Description

Technical Field

[0001] This application relates to the field of industrial system condition monitoring and intelligent early warning technology, and specifically discloses a risk early warning method and system for the cooling circuit of a liquid-cooled charging cable. Background Technology

[0002] With the continuous increase in the number of electric vehicles and the ever-increasing demands of users for charging efficiency, high-power liquid-cooled charging technology has become the mainstream development direction in the industry. Compared with traditional naturally cooled or air-cooled charging cables, liquid-cooled charging cables utilize a coolant circulation loop built inside the cable to remove the large amount of Joule heat generated by the conductor during charging. This allows for several times the current-carrying capacity of conventional cables with the same conductor cross-sectional area, supporting continuous overcharging at current levels of 500A to 1000A and even higher. However, if the cooling loop experiences blockages, leaks, pump performance degradation, or coolant deterioration, the cable conductor temperature will rise sharply. This can trigger charging power reduction protection, affecting user experience, or even accelerate thermal aging or burn out the cable insulation, leading to serious safety accidents. Therefore, accurate, proactive, and reliable risk warnings for the operating status of the liquid-cooled charging cable cooling loop are a key technological prerequisite for ensuring the safety and reliability of high-power charging.

[0003] Currently, the following methods are mainly used for monitoring and early warning of the cooling circuit of liquid-cooled charging cables:

[0004] The threshold comparison method involves pre-setting upper and lower limits for key parameters such as coolant temperature, supply flow rate, and loop pressure difference. An alarm is triggered when the real-time parameters collected by the sensors exceed the preset thresholds. This method is simple to implement and provides a direct response, making it suitable for early low-power charging scenarios.

[0005] The fault diagnosis method based on physical models establishes a lumped parameter thermodynamic model of cable heating and cooling, estimates the expected temperature of the cable conductor based on the real-time charging current, and compares it with the actual measured temperature. If the deviation exceeds the set tolerance, it is judged as an abnormality in the cooling circuit.

[0006] The single-parameter trend analysis method monitors the rate of change of coolant temperature or flow rate in real time, and issues an early warning signal when the slope of change exceeds the empirically set safety range.

[0007] However, the existing methods described above have the following significant drawbacks in practical engineering applications:

[0008] First, existing methods generally adopt a uniform static threshold or fixed judgment logic for all operating conditions, without considering the essential differences in the fluctuation characteristics of cooling circuit parameters under different operating conditions such as standby, low-power charging, high-power fast charging, peak overcharging, and derating operation.

[0009] Secondly, the risk assessment of existing methods is based on the measured values ​​of parameters at the current moment or their instantaneous rate of change. They do not have the ability to predict the state of parameters at future moments and cannot identify the gradual risk that the parameters are not currently out of bounds but the residuals show a continuous accumulation trend.

[0010] Third, the early warning logic of existing methods only focuses on a single liquid circuit parameter and does not integrate the liquid circuit parameter with electrical parameters (charging current, charging power) and environmental parameters (ambient temperature, humidity) for multi-dimensional analysis.

[0011] In summary, the key technical problems that urgently need to be solved in this field are how to achieve adaptive high-precision risk identification across the entire charging operating range, how to capture the progressive degradation trend in advance before the parameters exceed the limit, how to integrate liquid circuit, electrical parameters, and environmental parameters to construct a multi-dimensional risk assessment system, and how to form a patented technical solution with data processing and intelligent algorithms as the core that meets specific classification orientations.

[0012] This invention provides a risk warning method and system for the cooling circuit of a liquid-cooled charging cable to solve the above-mentioned problems. Summary of the Invention

[0013] The purpose of this invention is to provide a risk warning method and system for the cooling circuit of liquid-cooled charging cables, which enables adaptive high-precision risk identification under different operating conditions, proactively captures progressive degradation trends before monitoring parameters reach preset thresholds, integrates multi-dimensional risk assessment with thermodynamic parameters, load parameters and environmental parameters, and implements closed-loop linkage control based on risk quantification results.

[0014] To achieve the above objectives, the basic solution of the present invention provides a risk warning method for the cooling circuit of a liquid-cooled charging cable, comprising the following steps:

[0015] Step A1: Collect multi-dimensional time-series data consisting of liquid circuit parameters, electrical parameters and environmental parameters of the cooling circuit of the liquid-cooled charging cable. Perform noise reduction and normalization processing on the collected multi-dimensional time-series data in sequence, and construct a sliding time-series sample window based on the processed data.

[0016] Step A2: Extract the operating condition sensitive feature vector based on the sliding time sequence sample window, and use a combination of clustering and rule judgment based on power threshold and current stability to identify the operating condition category of the current charging process and obtain the operating condition label;

[0017] Step A3: Load the corresponding pre-trained time series prediction model according to the working condition label. The loaded pre-trained time series prediction model receives the sliding time series sample window and outputs the core parameter prediction value matrix for the future predetermined step size.

[0018] Step A4: Compare the predicted values ​​of the core parameters with the synchronously collected and processed measured values ​​item by item, and calculate the absolute residual and relative residual of each core parameter;

[0019] Step A5: Based on the relative residual sequence at several consecutive sampling times, extract multi-dimensional trend residual features, including cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation.

[0020] Step A6: Dynamically configure the weight coefficients of each trend-type residual feature according to the working condition label, normalize each trend-type residual feature and sum them by weight to obtain the basic risk score, and combine it with the warning confidence coefficient based on the duration of abnormality to calculate the final total risk score;

[0021] Step A7: Based on the final total risk score, perform early warning classification and output charging power linkage control command corresponding to the early warning level.

[0022] Furthermore, in step A1, the length of the sliding time-series sample window Dynamically adjust based on the current charging conditions:

[0023] Take in standby mode Take under low power charging conditions High-power fast charging, peak overcharging, and derating operation conditions ;

[0024] The step size of the sliding time-series sample window is set to 1 sampling point.

[0025] Furthermore, step A2 includes the following sub-steps:

[0026] Step A201: Extract the condition-sensitive feature vector from the sliding time sequence sample window, including the average charging power, current fluctuation rate, average coolant flow rate, and average inlet and outlet temperature difference.

[0027] Step A202: Use a pre-trained K-means clustering model to classify the operating condition sensitive feature vectors to obtain preliminary operating condition labels with values ​​from 1 to 5, which correspond to standby operating condition, low power charging operating condition, high power fast charging operating condition, peak overcharging operating condition and derating operating condition, respectively.

[0028] Step A203: Based on the real-time charging power, current fluctuation rate, and charging management system derating instruction flag, the preliminary operating condition label is modified a second time to obtain the final operating condition label.

[0029] Furthermore, in step A3, Kalman filtering and ARX models are loaded for standby and low-power charging conditions respectively. The Kalman filtering and ARX models are based on the linear time-varying state-space expression of the core parameters. The ARX model describes the linear influence of exogenous input on the state. The Kalman filter in the Kalman filtering and ARX models recursively corrects the state estimate based on the measured values.

[0030] The high-power fast charging condition, peak supercharging condition and derating operation condition are respectively loaded with TCN-Transformer fusion deep network model. The TCN-Transformer fusion deep network model is set with a temporal convolutional network and a Transformer encoder in sequence. The temporal convolutional network extracts temporal local features through multi-scale dilated convolution, and the Transformer encoder models temporal long-range dependencies based on a multi-head self-attention mechanism.

[0031] The future predetermined step size The labels are dynamically adjusted based on the current operating conditions.

[0032] Take in standby mode Seconds, under low-power charging conditions Seconds, under reduced operating conditions Seconds, high-power fast charging and peak supercharging conditions Second.

[0033] Furthermore, in step A5, the cumulative residual value is the first... One core parameter in the past The sum of relative residuals within each sampling time point;

[0034] The residual growth rate is the past The slope of the linear regression of the relative residual sequence within each sampling time point based on the least squares method;

[0035] The multivariate residual consistency refers to the consistency of any two core parameters in the past... The arithmetic mean of the Pearson correlation coefficients between the relative residual sequences within each sampling time point;

[0036] The deviation of the residual confidence interval is calculated using the following formula:

[0037] ;

[0038] In the formula, Indicates the first The core parameter in the first Deviation of the residual confidence interval at time 10:00 For the first One core parameter in the past The mean of the relative residuals at each time point. , Indicates the first The core parameter in the first Relative residuals at time, and In the current working condition The mean and standard deviation of the residuals are pre-calibrated.

[0039] Furthermore, step A6 further includes:

[0040] Step A601: Query the pre-stored weight table based on the current operating condition label to obtain the corresponding feature weight vector:

[0041] ;

[0042] The feature weight vector satisfies ,in, to The weighting coefficients correspond to the cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation, respectively.

[0043] Step A602: Perform Sigmoid compression normalization on each trend-type residual feature to obtain normalized features of the cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation located in the [0,1] interval. , , , ;

[0044] Step A603: Summate the normalized features according to the feature weight vector using the following formula to obtain the basic risk score:

[0045] ;

[0046] Step A604: Calculate the early warning confidence coefficient using the following formula:

[0047] ;

[0048] In the formula, The attenuation coefficient is... Indicates the cumulative duration of persistent anomalies in residuals;

[0049] Step A605: Based on Calculate the final total risk score.

[0050] Furthermore, in step A7, the early warning classification rule is as follows:

[0051] when At the first-level warning, a continuous monitoring command is issued;

[0052] when When the warning level is set to Level 2, an instruction will be issued to reduce the charging power by 20% to 30%, with the reduction percentage based on... The score is determined by linear interpolation within the interval [30, 60);

[0053] when The system is in Level 3 warning mode, and a shutdown protection command is issued.

[0054] When the warning level drops from high to low, the level can only be downgraded if the final total risk score is continuously lower than the next level boundary value minus the preset hysteresis bandwidth for a preset duration.

[0055] The present invention also provides a risk warning system for the cooling circuit of a liquid-cooled charging cable, configured to execute the above-described risk warning method for the cooling circuit of a liquid-cooled charging cable, comprising:

[0056] The data acquisition and preprocessing module is used to perform step A1;

[0057] The working condition identification module is communicatively connected to the data acquisition and preprocessing module and is used to execute step A2.

[0058] The prediction model loading module is communicatively connected to the working condition identification module and has a built-in Kalman filter-ARX model library and TCN-Transformer fusion model library for executing step A3;

[0059] The residual calculation module is communicatively connected to the prediction model loading module and is used to execute step A4.

[0060] The trend feature extraction module is communicatively connected to the residual calculation module and is used to execute step A5.

[0061] The risk scoring module is communicatively connected to the trend feature extraction module and the working condition identification module, and is used to execute step A6;

[0062] The graded early warning and linkage control module is communicatively connected to the risk scoring module and to an external charging controller, and is used to execute step A7.

[0063] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method for risk warning of cooling circuit of liquid-cooled charging cable.

[0064] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for risk warning of cooling circuit of liquid-cooled charging cable.

[0065] The principle and effect of this solution are as follows:

[0066] 1. Compared with existing technologies, this invention constructs a multi-dimensional working condition hierarchical identification mechanism, which finely divides the charging process into five core working conditions: standby, low power, high power fast charging, peak overcharging, and derating operation. Differentiated time-series prediction models are configured for the parameter fluctuation characteristics of different working conditions: computationally efficient Kalman filtering and ARX models are used for standby and low power working conditions, while TCN-Transformer fusion deep network model with nonlinear long-range dependency capture capability is used for high power and peak overcharging working conditions. This solves the contradiction between false alarms and false alarms caused by the unified judgment logic for all working conditions.

[0067] 2. Compared with existing technologies, this invention breaks through the limitation of traditional solutions that rely solely on current parameter values. It innovatively constructs a multi-dimensional trend-based residual feature system that includes cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation. By performing residual analysis between the predicted future parameter values ​​output by the time-series prediction model and the measured values, it can identify the cumulative trend and accelerated deviation signs of the residual sequence before key parameters such as coolant temperature and flow rate reach conventional alarm thresholds.

[0068] 3. Compared with existing technologies, this invention constructs a three-dimensional risk scoring model integrating liquid circuit parameters, electrical parameters, and environmental parameters. On the one hand, by introducing electrical parameters such as charging current and charging power, and environmental parameters such as ambient temperature as external input variables for the time-series prediction model, the prediction results can adapt to changes in external operating conditions. On the other hand, in the multivariate residual consistency characteristics, the correlation between liquid circuit parameter residuals and electrical parameter residuals is analyzed simultaneously, thereby effectively distinguishing between the overall drift of liquid circuit parameters caused by sudden changes in ambient temperature and the independent anomalies of liquid circuit parameters caused by internal faults in the cooling circuit, reducing false alarms triggered by external disturbances.

[0069] 4. Compared with existing technologies, this invention directly links risk scoring results with charging power control commands, constructing a graded early warning and response closed loop: Level 1: continuous monitoring; Level 2: power limiting and proactive derating; Level 3: shutdown and emergency protection. Unlike traditional solutions that only output alarm signals and rely on manual intervention, this invention can automatically reduce the charging power when the risk score reaches the Level 2 warning range, maximizing the continuity of the charging process while ensuring safety. Shutdown protection is only triggered when the risk rapidly deteriorates to the Level 3 warning range. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 A flowchart of a risk warning method for the cooling circuit of a liquid-cooled charging cable according to an embodiment of this application is shown. Detailed Implementation

[0072] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0073] This embodiment provides a risk warning method for the cooling circuit of a liquid-cooled charging cable, the process of which is as follows: Figure 1 As shown. This method uses liquid-cooled cables applied to high-power charging facilities for electric vehicles as an example. These cables have an internal coolant circulation loop to conduct Joule heat generated in the conductors during charging and can carry continuous currents of 500A to 1000A. It should be understood that the multi-condition time-series prediction and residual trend fusion evaluation mechanism disclosed in this application is not limited to the aforementioned liquid-cooled charging scenario. Its early warning logic is also applicable to industrial heat exchange systems or fluid circulation monitoring scenarios with similar cooling loop dynamic characteristics.

[0074] This embodiment uses the loop monitoring data of a 1000A liquid-cooled supercharging pile as an example to illustrate the implementation steps of this method as follows:

[0075] Step A1: Collect multi-dimensional time-series data consisting of liquid circuit parameters, electrical parameters, and environmental parameters of the liquid-cooled charging cable cooling circuit. Perform denoising and normalization processing on the collected multi-dimensional time-series data sequentially, and construct a sliding time-series sample window based on the processed data. This provides a standardized input basis for subsequent operating condition identification and time-series prediction. This step includes the following sub-steps:

[0076] Step A101: Determine the set of multi-dimensional acquisition parameters. In this embodiment, the acquired parameters cover three categories: fluid flow, electrical parameters, and environment. The raw acquired data is expressed in vector form as follows:

[0077] ;

[0078] In the formula, Indicates the first Each sampling time point has a sampling frequency of 1Hz. This represents the total number of parameters collected, in this embodiment. , to The numbers represent the order of the numbers. The data includes the cable inlet coolant temperature, outlet coolant temperature, supply flow rate, circuit differential pressure, pump speed, pump current, coolant conductivity, leakage detection signal, connection temperature, conductor current, charging voltage, charging power, and ambient temperature at all times.

[0079] Among them, the liquid circuit parameters to Directly reflects the operating status of the cooling circuit, electrical parameters to Used to characterize charging load levels, environmental parameters These three factors—used to compensate for the impact of external conditions on the fluid circuit status—together constitute the information foundation for multi-dimensional risk assessment.

[0080] Step A102: Denoising and Normalization of Raw Data. To avoid interference from sensor transients and environmental noise on the accuracy of subsequent time series predictions, a moving average filter with a window width of 5 sampling points is first applied to the raw data to smooth high-frequency glitches. The parameter values ​​after the moving average filter are obtained by the following formula:

[0081] ;

[0082] In the formula, Indicates the first The parameter in the first... The value after moving average filtering at time , Indicates the first The parameter in the first... The value at any given moment.

[0083] Furthermore, the parameters are subjected to minimum and maximum normalization to map parameters with different dimensions to the [0,1] interval. The normalization formula is as follows:

[0084] ;

[0085] In the formula, Indicates the first The parameter in the first... The normalized value at time , and They represent the first The minimum and maximum values ​​of these parameters recorded during the historical healthy operation phase. The historical healthy operation phase refers to the full-condition operation period of the liquid-cooled charging cable during factory testing or initial field commissioning, after manual confirmation of a fault-free state, and the corresponding minimum values. and maximum value It is pre-stored in the non-volatile memory of the charging controller for online normalization calls.

[0086] Step A103: Construction of the sliding time-series sample window. Specifically, to capture the dynamic evolution of cooling loop parameters over time, the normalized continuous sampling points need to be organized into a time-series sample window of fixed length. Time series sample window at time The definition is as follows:

[0087] ;

[0088] In the formula, for A 3D matrix Indicates the first Time normalized dimensional parameter vector, , This indicates the length of the timing window, expressed in the number of sampling points.

[0089] In this embodiment, the sliding window step size is fixed at 1 sampling point, that is, the window slides forward by one step size every time a new sampling point is generated.

[0090] Furthermore, the timing window length The value is not fixed, but is dynamically adjusted according to the current charging conditions:

[0091] In standby mode, the cooling circuit parameters change extremely slowly. To capture subtle trends over long periods;

[0092] Under low-power charging conditions, take To balance response speed and statistical stability;

[0093] Under high-power fast charging, peak overcharging, and derating operation conditions, parameters change drastically and risks develop rapidly. To improve the timeliness of forecasts.

[0094] This dynamic window design ensures that subsequent time-series prediction models can obtain a suitable length of historical information support under different operating conditions. Meanwhile, when the system is initially powered on and the number of historical sampling points is insufficient to fill the window, a cold-start window is constructed using a combination of zero-value filling and copying of the initial sampled value. Once the actual sampled data is sufficient to fill the window, it switches to normal sliding mode, thereby avoiding prediction failure during the cold-start phase.

[0095] Step A2: Extract the operating condition sensitive feature vector based on the sliding time-series sample window. Use a combination of clustering and rule-based judgment based on power thresholds and current stability to identify the operating condition category of the current charging process, obtaining the operating condition label. This label provides the operating condition label for subsequent loading of the differentiated prediction model and configuration of risk weights. This step includes the following sub-steps:

[0096] Step A201: Extraction of operating condition-sensitive feature vectors. Specifically, from the time-series sample window... The operating condition-sensitive feature vector is extracted from the data. In this embodiment, four features are selected: average charging power, current fluctuation rate, average coolant flow rate, and average inlet and outlet temperature difference, which constitute the feature vector.

[0097] ;

[0098] In the formula, The arithmetic mean of the charging power within the time-series sample window, and the current fluctuation rate. Defined as the ratio of the standard deviation of the current to its mean, i.e. ,in The charging current sequence within the time-series sample window. This represents the arithmetic mean of the fluid supply flow rate within the time-series sample window. It is the arithmetic mean of the difference between the outlet coolant temperature and the inlet coolant temperature within the time-series sample window.

[0099] The above four characteristics can comprehensively depict the essential characteristics of the current charging condition from four dimensions: power level, load stability, coolant flow state, and heat dissipation intensity.

[0100] Step A202: Preliminary work condition classification based on k-means clustering. Specifically, the K-means clustering model trained in the offline phase is used, with the number of clusters... For the feature vector After classification, the preliminary working condition labels are obtained as shown in the following formula:

[0101] ;

[0102] In the formula, This is a preliminary operating condition label, with values ​​ranging from 1 to 5, corresponding to standby, low-power charging, high-power fast charging, peak overcharging, and derating operation conditions, respectively.

[0103] The coordinates of the five cluster centers of the above K-means model were obtained offline using full-condition data from historical healthy operation phases and have been stored in the read-only memory of the charging controller.

[0104] Step A203: Operating condition label correction based on power threshold and current stability. Specifically, since K-means clustering relies solely on data distribution characteristics, label jitter or misjudgment may occur during the transition phase of certain operating condition switching. Therefore, a rule-based label correction mechanism is further introduced. This is based on real-time charging power... Current fluctuation rate and the derating instruction flag of the charging management system As a basis for judgment, the initial working condition labels were reconfirmed and revised to obtain the final working condition labels. The revised rules are as follows:

[0105] ;

[0106] In the formula, Indicates the first The final operating condition label at any given moment. This indicates that the charging management system has issued a derating instruction or has entered a fault derating state.

[0107] The above rules are judged in order of priority from top to bottom:

[0108] If a derating instruction exists, it will be determined as operating condition 5 first; otherwise, it will be determined sequentially according to the power range.

[0109] in, Used to filter out instantaneous zero-power disturbances at the moment of gun insertion. This is used to distinguish between stable peak overcharging conditions and instantaneous power spikes caused by battery response fluctuations. After the above dual determination, the accuracy and stability of the operating condition label are effectively guaranteed.

[0110] Step A3: Load the corresponding pre-trained time series prediction model according to the working condition label. The loaded pre-trained time series prediction model receives the sliding time series sample window and outputs the core parameter prediction matrix for the future predetermined step size.

[0111] Specifically, based on the final operating condition label obtained in step A2 Load the pre-trained time-series prediction model corresponding to this operating condition to achieve accurate prediction of core parameters under different charging conditions. This step includes the following sub-steps:

[0112] Step A301: Working Condition and Model Layer Configuration. Specifically, the model layer configuration rules are as follows:

[0113] ;

[0114] In the formula, Indicates the first Prediction models corresponding to different working conditions.

[0115] Furthermore, the underlying logic of the above-mentioned hierarchical configuration lies in:

[0116] standby mode Low power charging conditions Under these conditions, the cooling circuit parameters fluctuate smoothly, mainly affected by ambient temperature drift and pump speed fine-tuning, exhibiting typical linear time-invariant characteristics. Satisfactory prediction accuracy can be obtained by using Kalman filtering combined with an autoregressive model ARX with exogenous variables. At the same time, the model has a very small computational load, making it suitable for low-computing-power environments for vehicle-mounted or pile-mounted controllers.

[0117] In high-power fast charging conditions Peak overcharging conditions and reduced operating conditions Under these conditions, there is a strong nonlinear coupling relationship between coolant temperature, flow rate and pressure difference, and the parameter timing exhibits significant nonstationary and long-range dependent characteristics due to the rapid adjustment of charging current. Therefore, a deep neural network model that integrates temporal convolutional network (TCN) and Transformer encoder is adopted to accurately capture the above complex dynamic characteristics.

[0118] Step A302: Configuration of the Kalman Filter-ARX Model. Specifically, for standby and low-power operating conditions, the Kalman Filter-ARX model used is based on the linear time-varying state-space expression of the core parameters, and the state equation is:

[0119] ;

[0120] The observation equation is:

[0121] ;

[0122] In the formula, The state vector consists of the current values ​​of the core prediction parameters and their first-order differences. The exogenous input vector consists of charging power, conductor current, and ambient temperature. , , These are the state transition matrix, exogenous input matrix, and observation matrix, respectively, obtained through least-squares identification of offline health data. and These are the process noise and the observation noise, respectively, set as zero-mean Gaussian white noise.

[0123] In this embodiment, the ARX model order is set to (3,2), meaning the autoregressive term is order 3 and the exogenous input term is order 2. After configuring the model as described above, the Kalman filter continuously corrects the state estimate based on the measured values ​​during the online inference phase and recursively outputs the future... Predicted values ​​of core parameters at each time point.

[0124] Step A303: Configuration of the TCN-Transformer fusion model. Specifically, the network structure parameters of the TCN-Transformer fusion model are set as follows:

[0125] The TCN has 4 layers, with an inflation coefficient sequence of [1,2,4,8], a kernel size of 3, and the number of channels in each layer is 32, 64, 64, and 32 respectively.

[0126] The Transformer encoder has 2 layers, the attention heads have 4, and the feedforward network has 128 dimensions.

[0127] The input layer receives the output of step A1. 3D time series sample window ;

[0128] The output layer is mapped through a fully connected layer and outputs... The core parameter prediction matrix of dimension, where Pre-determine step size for the future. The number of core prediction parameters.

[0129] Meanwhile, the training phase utilizes historical health datasets. Offline training was performed using time-series samples under operating conditions. The Adam optimizer was selected, with an initial learning rate of 0.001 and a cosine annealing strategy for decay. The loss function was... Using Huber Loss:

[0130] ;

[0131] In the formula, This represents the measured value. This represents the predicted value of the corresponding model. The threshold for the loss function is set in this embodiment. This is to enhance robustness to outlier sample points.

[0132] After training is complete, the model parameters are saved as model files indexed by working condition labels for loading during the online inference phase.

[0133] Step A304: Online Inference and Predicted Value Output. Specifically, during the online execution phase, the loaded model receives the time-series sample window output in step A1. Output the future using the formula below. Core parameter prediction matrix at each time point:

[0134] ;

[0135] In the formula, for A matrix of predicted values ​​in dimensional form.

[0136] In this embodiment, the core prediction parameters include: outlet coolant temperature, inlet and outlet temperature difference, supply flow rate, loop pressure difference, and connection end temperature rise rate, totaling [number missing]. item.

[0137] Furthermore, the planned step size in the future It is also dynamically adjusted according to the operating conditions:

[0138] standby mode Seconds, low power operation condition Seconds, derated operating conditions Seconds, high-power fast charging and peak supercharging conditions Seconds. The differentiated setting of prediction duration balances the lead time for risk warning with prediction accuracy: the more severe the operating conditions, the shorter the prediction duration to ensure prediction reliability.

[0139] Step A4: Compare the predicted values ​​of the core parameters with the measured values ​​that were collected and processed simultaneously, and calculate the absolute residuals and relative residuals of each core parameter.

[0140] Specifically, the future obtained in step A3 The predicted values ​​of the core parameters at each time point are compared with the measured values ​​collected and preprocessed by the sensors at the corresponding time points. The absolute and relative residuals of each core parameter are calculated to quantify the degree of deviation between the prediction model and the actual operating state.

[0141] No. The core parameter in the first absolute residual of time Defined as the absolute value of the difference between the predicted value and the measured value:

[0142] ;

[0143] In the formula, For the prediction step index, , Indicates the first The core parameter in the first The measured value after normalization of time. This represents the corresponding model prediction value.

[0144] Furthermore, to eliminate the influence of differences in the dimensions and numerical ranges of different parameters on subsequent feature extraction, the relative residual is calculated after the absolute residual calculation. The core parameter in the first Relative residuals at time The calculation formula is as follows:

[0145] ;

[0146] In the formula, Indicates the first The core parameters under current operating conditions Typical amplitude benchmarks for the next historical healthy operation phase, To preset a minimum threshold, this embodiment takes... This is used to avoid division by zero exceptions.

[0147] The relative residual measures the deviation from the predicted value, making the residuals of parameters with different dimensions such as temperature (°C), flow rate (L / min), and pressure difference (kPa) comparable and compatible.

[0148] Step A5: Based on the relative residual sequence at several consecutive sampling times, extract multi-dimensional trend residual features, including cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation.

[0149] Unlike traditional solutions that only focus on whether the parameter exceeds a fixed threshold at the current moment, this invention is based on continuous... The relative residual sequence at each sampling time point is further used to extract multi-dimensional trend-type residual features to capture the gradual evolution of the cooling circuit state, enabling early identification of abnormal trends even when parameters are within limits. In this embodiment, the residual statistical window length... Dynamic settings based on operating conditions:

[0150] standby mode Low power operating conditions High-power, peak overcharging, and derating operation conditions A longer window results in more stable trend characteristics but a slower response; a shorter window provides higher sensitivity but is more susceptible to transient noise. This step includes the following sub-steps:

[0151] Step A501: Cumulative Residual Value The calculation. Specifically, the cumulative residual value is defined as the first... One core parameter in the past The continuous summation of the relative residuals within each sampling time point is used to measure the persistent cumulative effect of the deviation:

[0152] ;

[0153] In the formula, Indicates the first The core parameter in the first The cumulative residual value at time step 1. Indicates the first The core parameter in the first The relative residual at any given moment.

[0154] When the cooling circuit is in good condition, the predicted residual fluctuates randomly around zero, and the cumulative value remains at a low level. Once a progressive failure occurs, such as slow decline in pump performance, gradual clogging of the filter, or minor coolant leakage, the relative residual will show a unidirectional deviation trend, and the cumulative value will continue to rise, thus issuing a warning signal before the parameters reach the alarm threshold.

[0155] Step A502: Residual Growth Rate The calculation is as follows. Specifically, the residual growth rate characterizes the growth trend of the relative residual sequence per unit time, reflecting the degree of intensification of the deviation from the trend. To avoid the excessive sensitivity of the simple difference form to instantaneous noise, this embodiment uses the slope of the linear regression based on the least squares method as the growth rate:

[0156] ;

[0157] In the formula, Indicates the first The core parameter in the first The residual growth rate at time t, its physical meaning is the past The linear regression slope of the relative residual sequence within each sampling time point Indicates the first The core parameter in the first The relative residual at any given moment.

[0158] This characteristic is particularly sensitive to precursors of sudden failures, namely, the accelerated growth of residuals. For example, when a rapidly developing localized blockage occurs in the cooling circuit, the flow residual not only rises horizontally, but its growth rate also increases dramatically. It will also increase synchronously, providing a basis for distinguishing between slow performance degradation and sudden failure.

[0159] Step A503: Consistency of Multivariate Residuals The calculation involves several aspects. Specifically, the residual of a single parameter may often be caused by sensor noise or local disturbances, while a real cooling circuit failure often leads to the synchronous deviation of the residuals of multiple related parameters. Therefore, the consistency of multivariate residuals is crucial. The degree of synergy among multi-parameter deviations is quantified by calculating the mean of the Pearson correlation coefficient between the residual sequences of any two core parameters.

[0160] ;

[0161] In the formula, The number of core parameters, Indicates the first One core parameter in the past The relative residual sequence at each time step. , This represents the Pearson correlation coefficient.

[0162] Multivariate residual consistency The value range is [-1, 1]. The closer it is to 1, the stronger the positive correlation of the multi-parameter residuals. That is, if multiple liquid circuit parameters deviate from the predicted value at the same time, the probability of a real failure in the cooling circuit is higher. Conversely, if only a single parameter residual is abnormal while the multi-variable residuals are consistent, the probability of a real failure in the cooling circuit is higher. A lower false alarm rate is more likely to be caused by occasional sensor disturbances or local measurement errors, thus effectively reducing the false alarm rate.

[0163] Step A504: Deviation of Residual Confidence Interval The calculation involves using the statistical distribution of relative residuals from historical healthy operating phases as a benchmark to measure the degree to which the current residual deviates from the normal fluctuation range. First, during the offline phase, historical health data is used to calculate the mean of the relative residuals for each core parameter under various operating conditions. with standard deviation And store it as a statistical parameter library.

[0164] Furthermore, to avoid the interference of transient disturbances on this feature, the online phase is based on a window... The mean of the internal relative residuals is standardized and calculated. The core parameter in the first Deviation of residual confidence interval at time The calculation formula is as follows:

[0165] ;

[0166] In the formula, For the first One core parameter in the past The mean of the relative residuals at each time point. , and In the current working condition The mean and standard deviation of the residuals are pre-calibrated.

[0167] residual confidence interval deviation Essentially, this is a standardization process for the mean of the current window residuals; the larger the absolute value, the more significantly the current residuals deviate from the normal fluctuation range under healthy conditions. According to the 3σ principle in statistics, when... When this occurs, the current residual can be determined to be significantly abnormal. This feature provides a probabilistic benchmark for risk scoring.

[0168] Step A6: Dynamically configure the weight coefficients of each trend-type residual feature according to the working condition label, normalize each trend-type residual feature and sum them by weight to obtain the basic risk score, and combine it with the warning confidence coefficient based on the duration of the abnormality to calculate the final total risk score.

[0169] Specifically, based on the four trend-type residual features extracted in step A5, combined with the current operating condition label determined in step A2... The process involves dynamically configuring the weighting coefficients of each feature, calculating the fusion-based basic risk score and early warning confidence coefficient, and finally generating the total risk score. This step includes the following sub-steps:

[0170] Step A601: Adaptive Weight Configuration for Different Charging Conditions. Specifically, the main risk types and sensitive parameters faced by the cooling circuit differ significantly under different charging conditions:

[0171] Under peak overcharge conditions, the cable generates a great deal of heat. The temperature difference between the inlet and outlet and the liquid supply flow rate are the key parameters that determine the cooling effect. Abnormal residual difference directly indicates the risk of overheating.

[0172] In standby or low-power operation, the circuit is in a low-flow-rate state for a long time, and the pump's operating status and changes in coolant conductivity are more likely to reflect potential faults.

[0173] Therefore, this embodiment constructs the following adaptive feature weight vector. :

[0174] ;

[0175] In the formula, to The corresponding values ​​are, in order, the cumulative residual value, the residual growth rate, the consistency of multivariate residuals, and the deviation of the residual confidence interval under the current operating conditions. The weighting coefficients are set below the normalization constraint. .

[0176] Furthermore, the weight configurations for the five operating conditions in this embodiment are shown in Table 1 below:

[0177] Table 1 Weighting of Five Operating Conditions

[0178]

[0179] The above weighting configuration emphasizes the cumulative residual value and growth rate to a total of 60% under peak overcharging conditions, in order to respond to the rapid development of deviation trends of key parameters such as temperature difference and flow rate. Under standby conditions, the weight of confidence interval deviation is increased to 40% to use statistical significance to identify pump or sensor anomalies at low flow rates. The weights for other conditions are between the two, forming a smooth transition.

[0180] Step A602: Normalization of trend characteristics. Specifically, due to the cumulative residual value... residual growth rate Consistency of multivariate residuals Deviation of residual confidence interval The four features have vastly different dimensions and numerical ranges. Directly weighting and summing them would cause a feature with a larger magnitude to dominate the final score, weakening the contribution of the other features. Therefore, this embodiment performs Sigmoid compression and normalization on each feature before weighting and summing, so that each feature is mapped to the [0,1] interval.

[0181] First, calculate the average value of each feature across all core parameters using the following formula:

[0182] ;

[0183] ;

[0184] ;

[0185] In the formula, This represents the average of the cumulative residuals. This represents the average residual growth rate. This represents the average deviation of the residual confidence interval.

[0186] Furthermore, the features, after Sigmoid compression and normalization, are expressed as follows:

[0187] ;

[0188] ;

[0189] ;

[0190] ;

[0191] In the formula, This represents the expression of the cumulative residual value after Sigmoid compression and normalization. This represents the residual growth rate after Sigmoid compression and normalization. This represents the expression of multivariate residual consistency after Sigmoid compression and normalization. This represents the deviation of the residual confidence interval after Sigmoid compression and normalization. , , To compress the steepness parameter, , , As the normalization center point, all are pre-calibrated through the statistical distribution of offline health data, so that each health state has a normalization center point. The value fluctuates below 0.1, while it rises above 0.7 under typical fault conditions.

[0192] Map the consistency of multivariate residuals to the [0,1] interval, and The bigger, The smaller the value, the higher the risk, which aligns with the intuitive logic that multiple parameters deviate together.

[0193] Step A603: Basic Risk Score The calculation is as follows: Specifically, the base risk score is obtained by weighted summation of the expressions of each feature after Sigmoid compression and normalization.

[0194] .

[0195] Due to each and ,therefore This provides a unified and comparable standard for subsequent risk classification.

[0196] Step A604: Early Warning Confidence Coefficient The calculation is as follows. Specifically, the risk score at an isolated moment is susceptible to transient disturbances; therefore, this embodiment introduces a confidence coefficient based on the duration of the anomaly. To enhance the robustness of the scoring:

[0197] ;

[0198] In the formula, For the attenuation coefficient, this embodiment takes... , This indicates the cumulative duration of persistent anomalies in residuals, expressed in seconds.

[0199] Cumulative duration of persistent anomalies in residuals The update rules are as follows:

[0200] If the first Basic risk score at any time Exceeding the preset attention threshold Then update Otherwise execute That is, the cumulative increase during the abnormal period, the decay rate at twice the rate during the recovery period until it returns to zero, and the initial value... .

[0201] Then calculate the confidence coefficient. Finally, proceed to step A605. This involves the confidence coefficient in the form of an exponential function. It has saturation characteristics and the abnormal duration is relatively short. The smaller size has a suppressive effect on the overall risk score, preventing transient spikes from triggering false alarms. When the anomaly persists for more than approximately 30 seconds... Approaching 1, the risk score is mainly determined by the basic risk score, ensuring that real faults are not missed.

[0202] Step A605: Final Risk Score The calculation involves multiplying the base risk score by the confidence coefficient and scaling it to the 0-100 range to obtain the final total risk score.

[0203] .

[0204] because and The result of the above formula It integrates information from five dimensions: residual amplitude, growth trend, multi-parameter consistency, statistical significance, and duration of anomalies, providing a single, intuitive, and comparable quantitative indicator for subsequent graded early warning.

[0205] Step A7: Based on the final total risk score, perform early warning classification and output charging power linkage control command corresponding to the early warning level.

[0206] Specifically, the final total risk score is calculated based on step A6. The system classifies warning levels and outputs corresponding warning information and charging power linkage control commands to achieve a closed loop of warning and response. This step includes the following sub-steps:

[0207] Step A701: Warning Level Classification. Specifically, the rules for classifying warning levels are as follows:

[0208] ;

[0209] In the formula, Indicates the first The level of warning at any given moment.

[0210] Step A702: Output of tiered linkage response instructions. Specifically, the linkage response instructions corresponding to each level are as follows:

[0211] ;

[0212] In the formula, Indicates the first Real-time coordinated response instructions This indicates the adjusted charging power command value. This indicates the charging power command value before adjustment.

[0213] Power reduction ratio during Level II warning The specific score within the interval [30, 60) is determined by linear interpolation. The higher the score, the greater the reduction. The reduction range is 20% to 30%. When a level 3 warning occurs, the charging output is immediately cut off and the shutdown protection procedure is executed. At the same time, the fault location information is pushed to the operation and maintenance platform.

[0214] Step A703: Generation of Structured Early Warning Details. Specifically, while outputting hierarchical control instructions, the system also generates a structured early warning details report, which includes: risk type, expected over-limit time, fault location suggestions, and handling measures already taken.

[0215] Furthermore, the risk type is determined based on the relative contribution of the residual characteristics of each core parameter:

[0216] ;

[0217] In the formula, The system identifies the core parameter that contributes the most to the residual and outputs the corresponding risk type description according to a preset mapping table. The mapping relationship is shown in Table 2 below:

[0218] Table 2 Mapping Relationship Table

[0219]

[0220] Furthermore, the estimated time of exceeding the limit. Based on the current residual growth rate Extrapolation yields the following formula for calculation:

[0221] ;

[0222] In the formula, The relative residual threshold for this parameter is taken as the threshold value in this embodiment. , The unit is seconds.

[0223] In this embodiment, only when ,and Only then will the calculation and output be performed. Otherwise, output a stable trend indicator. To set a lower limit for the preset growth rate, this embodiment takes... .

[0224] Step A704: Warning Reset Hysteresis Mechanism. Specifically, to avoid frequent jumps in warning levels near critical values, this embodiment introduces a hysteresis reset mechanism:

[0225] When the warning level drops from high to low, the following conditions must be met: Continuously below the next level boundary value minus hysteresis bandwidth For more than 10 seconds, this embodiment takes Only after this condition is met can the risk level be downgraded. This mechanism effectively avoids frequent adjustments to charging power caused by the total risk score fluctuating around the threshold.

[0226] Based on the above steps, the simulated fault injection test conducted on a 1000A liquid-cooled supercharging pile in this embodiment shows that:

[0227] Before the coolant flow rate drops below the safety threshold and traditional solutions cannot detect it, this invention identifies the progressive leakage risk approximately 25 seconds in advance. By actively reducing power, it reduces the heat generated by the cable by 30%, effectively avoiding overheating damage to the cable due to insufficient cooling. At the same time, it maintains the continuity of the charging process and avoids the impact on user experience caused by direct shutdown.

[0228] This embodiment also provides a risk warning system for the cooling circuit of a liquid-cooled charging cable, used to execute the risk warning method described above, including the following modules:

[0229] The data acquisition and preprocessing module is used to execute step A1 of Embodiment 1. It is configured to acquire multi-dimensional time-series data composed of liquid circuit parameters, electrical parameters and environmental parameters of the cooling circuit of the liquid-cooled charging cable, perform moving average filtering to remove noise and minimum and maximum normalization processing on the raw data, and construct a moving time-series sample window.

[0230] The operating condition identification module, used to execute step A2 of Embodiment 1, is communicatively connected to the data acquisition and preprocessing module. It is configured to extract operating condition-sensitive feature vectors based on a sliding time-series sample window, perform preliminary operating condition classification using K-means clustering, and correct operating condition labels based on power thresholds, current stability, and derating instruction flags, ultimately outputting the final operating condition labels. .

[0231] The prediction model loading module, used to execute step A3 of Example 1, communicates with the working condition identification module, and has a built-in Kalman filter-ARX model library and TCN-Transformer fusion model library. It is configured to load the model based on the final working condition label. Load the pre-trained time series prediction model corresponding to this working condition, and output the core parameter prediction matrix for the future predetermined step size.

[0232] The residual calculation module is used to execute step A4 of Example 1. It is connected in communication with the prediction model loading module and is configured to compare the predicted value matrix of core parameters with the synchronously collected measured values ​​item by item to calculate the absolute residual and relative residual of each core parameter.

[0233] The trend feature extraction module, used to execute step A5 of Example 1, is communicatively connected to the residual calculation module and is configured to be based on continuous The relative residual sequence at each sampling time point is used to extract multi-dimensional trend residual features, including cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation.

[0234] The risk scoring module, used to execute step A6 of Example 1, is communicatively connected to the trend feature extraction module and the working condition identification module, and is configured to base its scoring on the final working condition label. The weighting coefficients of each trend-type residual feature are dynamically configured. After Sigmoid compression and normalization of each feature, a weighted sum is obtained to obtain the basic risk score. Combined with the warning confidence coefficient based on the duration of the anomaly, the final total risk score is output. .

[0235] The tiered early warning and linkage control module is used to execute step A7 of Embodiment 1. It is communicatively connected to the risk scoring module, as well as to the external charging controller and the operation and maintenance platform. It is configured to base its actions on the final total risk score. The system classifies warning levels, outputs corresponding charging power linkage control commands and structured warning detail reports, and executes a hysteresis reset mechanism.

[0236] In this embodiment, the above modules can be implemented as software program modules and integrated into the charging controller of the liquid-cooled charging pile, thereby completing the system deployment through software upgrade without changing the existing hardware structure.

[0237] This embodiment also provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor is configured to execute the computer program to implement the liquid-cooled charging cable cooling circuit risk warning method as described above.

[0238] In this embodiment, the electronic device is a charging controller for a charging pile, an integrated industrial computer, an embedded gateway, or an edge computing node; the memory is a non-volatile memory, including flash memory, electrically erasable programmable read-only memory, or a solid-state drive; and the processor is an ARM-based embedded processor, a digital signal processor, or an application-specific integrated circuit.

[0239] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the risk warning method for the cooling circuit of the liquid-cooled charging cable as described above.

[0240] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any indirect modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A risk warning method for the cooling circuit of a liquid-cooled charging cable, characterized in that, Includes the following steps: Step A1: Collect multi-dimensional time-series data consisting of liquid circuit parameters, electrical parameters and environmental parameters of the cooling circuit of the liquid-cooled charging cable. Perform noise reduction and normalization processing on the collected multi-dimensional time-series data in sequence, and construct a sliding time-series sample window based on the processed data. The fluid parameters include cable inlet coolant temperature, outlet coolant temperature, supply flow rate, circuit differential pressure, pump speed, pump current, coolant conductivity, leakage detection signal, and connection end temperature; the electrical parameters include conductor current, charging voltage, and charging power; and the environmental parameters include ambient temperature. Step A2: Extract the operating condition sensitive feature vector based on the sliding time sequence sample window, and use a combination of clustering and rule judgment based on power threshold and current stability to identify the operating condition category of the current charging process and obtain the operating condition label; Step A2 includes the following sub-steps: Step A201: Extract the condition-sensitive feature vector from the sliding time sequence sample window, including the average charging power, current fluctuation rate, average coolant flow rate, and average inlet and outlet temperature difference. Step A202: Use a pre-trained K-means clustering model to classify the operating condition sensitive feature vectors to obtain preliminary operating condition labels with values ​​from 1 to 5, which correspond to standby operating condition, low power charging operating condition, high power fast charging operating condition, peak overcharging operating condition and derating operating condition, respectively. Step A203: Based on the real-time charging power, current fluctuation rate and charging management system derating instruction flag, the preliminary operating condition label is corrected a second time to obtain the final operating condition label; Step A3: Load the corresponding pre-trained time series prediction model according to the working condition label. The loaded pre-trained time series prediction model receives the sliding time series sample window and outputs the core parameter prediction value matrix for the future predetermined step size. Among them, the standby mode and the low power charging mode are respectively loaded with Kalman filter and ARX model. The Kalman filter and ARX model are based on the linear time-varying state space expression of the core parameters. The ARX model describes the linear influence of exogenous input on the state. The Kalman filter in the Kalman filter and ARX model recursively corrects the state estimate based on the measured value. The high-power fast charging condition, peak supercharging condition and derating operation condition are respectively loaded with TCN-Transformer fusion deep network model. The TCN-Transformer fusion deep network model is set with a temporal convolutional network and a Transformer encoder in sequence. The temporal convolutional network extracts temporal local features through multi-scale dilated convolution, and the Transformer encoder models temporal long-range dependencies based on a multi-head self-attention mechanism. The core parameters include outlet coolant temperature, inlet and outlet temperature difference, supply flow rate, circuit pressure difference, and connection end temperature rise rate. Step A4: Compare the predicted values ​​of the core parameters with the synchronously collected and processed measured values ​​item by item, and calculate the absolute residual and relative residual of each core parameter; Step A5: Based on the relative residual sequence at several consecutive sampling times, extract multi-dimensional trend residual features, including cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation. In step A5, the cumulative residual value is the first... One core parameter in the past The sum of relative residuals within each sampling time point; The residual growth rate is the past The slope of the linear regression of the relative residual sequence within each sampling time point based on the least squares method; The multivariate residual consistency refers to the consistency of any two core parameters in the past... The arithmetic mean of the Pearson correlation coefficients between the relative residual sequences within each sampling time point; The deviation of the residual confidence interval is calculated using the following formula: ； In the formula, Indicates the first The core parameter is in the first Deviation of the residual confidence interval at time 10:00 For the first One core parameter in the past The mean of the relative residuals at each time point. , Indicates the first The core parameter is in the first Relative residuals at time, and In the current working condition The pre-calibrated mean and standard deviation of the residuals; Step A6: Dynamically configure the weight coefficients of each trend-type residual feature according to the working condition label, normalize each trend-type residual feature and sum them by weight to obtain the basic risk score, and combine it with the warning confidence coefficient based on the duration of abnormality to calculate the final total risk score; Step A7: Based on the final total risk score, perform early warning classification and output charging power linkage control command corresponding to the early warning level.

2. The risk warning method for the cooling circuit of a liquid-cooled charging cable according to claim 1, characterized in that, In step A1, the length of the sliding timing sample window Dynamically adjust based on the current charging conditions: Take in standby mode Take under low power charging conditions High-power fast charging, peak overcharging, and derating operation conditions ; The step size of the sliding time-series sample window is set to 1 sampling point.

3. The risk warning method for the cooling circuit of a liquid-cooled charging cable according to claim 1, characterized in that, In step A3, the future predetermined step size The labels are dynamically adjusted based on the current operating conditions. Take in standby mode Seconds, under low-power charging conditions Seconds, under reduced operating conditions Seconds, high-power fast charging and peak supercharging conditions Second.

4. The risk warning method for the cooling circuit of a liquid-cooled charging cable according to claim 1, characterized in that, Step A6 further includes: Step A601: Query the pre-stored weight table based on the current operating condition label to obtain the corresponding feature weight vector: ； The feature weight vector satisfies ,in, to The weighting coefficients correspond to the cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation, respectively. Step A602: Perform Sigmoid compression normalization on each trend-type residual feature to obtain normalized features of the cumulative residual value, residual growth rate, multivariate residual consistency, and residual confidence interval deviation located in the [0,1] interval. , , , ; Step A603: Summate the normalized features according to the feature weight vector using the following formula to obtain the basic risk score: ； Step A604: Calculate the early warning confidence coefficient using the following formula: ； In the formula, The attenuation coefficient is... Indicates the cumulative duration of persistent anomalies in residuals; Step A605: Based on Calculate the final total risk score.

5. The risk warning method for the cooling circuit of a liquid-cooled charging cable according to claim 1, characterized in that, In step A7, the early warning classification rule is as follows: when At the first-level warning, a continuous monitoring command is issued; when When the warning level is set to Level 2, an instruction will be issued to reduce the charging power by 20% to 30%, with the reduction percentage based on... The score is determined by linear interpolation within the interval [30, 60); when The system is in Level 3 warning mode, and a shutdown protection command is issued. When the warning level drops from high to low, the level can only be downgraded if the final total risk score is continuously lower than the next level boundary value minus the preset hysteresis bandwidth for a preset duration.

6. A risk early warning system for the cooling circuit of a liquid-cooled charging cable, characterized in that, The method configured to perform a risk warning method for a liquid-cooled charging cable cooling circuit according to any one of claims 1 to 5 includes: The data acquisition and preprocessing module is used to perform step A1; The working condition identification module is communicatively connected to the data acquisition and preprocessing module and is used to execute step A2. The prediction model loading module is communicatively connected to the working condition identification module, and has a built-in Kalman filter-ARX model library and TCN-Transformer fusion model library, used to execute step A3; The residual calculation module is communicatively connected to the prediction model loading module and is used to execute step A4. The trend feature extraction module is communicatively connected to the residual calculation module and is used to execute step A5. The risk scoring module is communicatively connected to the trend feature extraction module and the working condition identification module, and is used to execute step A6. The graded early warning and linkage control module is communicatively connected to the risk scoring module and to an external charging controller, and is used to execute step A7.

7. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a risk warning method for the cooling circuit of a liquid-cooled charging cable according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a risk warning method for the cooling circuit of a liquid-cooled charging cable according to any one of claims 1 to 5.

Images