A method, medium and system for predicting overcurrent of a liquid-cooled supercharging pile

Abstract

The present invention provides an overcurrent prediction method, medium and system for a liquid-cooled supercharger, which belongs to the technical field of measuring electrical variables, including: by collecting the charging parameters and heat exchange parameters of the liquid-cooled supercharger, a time-sensitive HyperLogLog (TS-HLL) structure is designed to encode these parameters. The statistical characteristics of the charging parameters and the heat exchange parameters are extracted by using the time sliding window technology, and the correlation matrix between the two is calculated. Based on the correlation matrix, a relationship map between the charging current and other parameters is constructed, and a graph neural network is used to learn it, and the complex nonlinear relationship between the charging current and other parameters is extracted. Finally, the learned features are combined with historical charging data and input into a recurrent neural network to predict the charging current in the future, thereby solving the problem of rapid overcurrent prediction of liquid-cooled superchargers that is difficult to solve with the current existing technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of measuring electrical variables, and in particular, relates to an overcurrent prediction method, medium and system for a liquid-cooled supercharging pile. Background Art

[0002] With the rapid development of new energy vehicles, the demand for charging infrastructure is increasing. Among them, liquid-cooled super charging piles, as an emerging charging technology, have attracted widespread attention and application due to their ability to provide high power and fast charging. Liquid-cooled super charging piles effectively dissipate the large amount of heat generated during the charging process through a liquid cooling system, thereby avoiding safety hazards caused by excessive battery temperature.

[0003] At present, liquid-cooled supercharging piles have been widely deployed in charging stations and highway service areas in some cities. Users can use these charging piles to quickly charge electric vehicles, which greatly shortens the charging time and improves the convenience of using the car. However, in actual use, liquid-cooled supercharging piles also face some technical challenges. First, during high-power charging, the charging current may suddenly increase, causing system overcurrent, which in turn causes safety hazards. Overcurrent events will not only damage the charging pile itself, but may also endanger the user's property and personal safety. Therefore, how to effectively predict and prevent overcurrent phenomena has become a technical problem that needs to be solved urgently. The current existing technology is difficult to solve the problem of overcurrent prediction of fast liquid-cooled supercharging piles. Summary of the invention

[0004] In view of this, the present invention provides an overcurrent prediction method, medium and system for a liquid-cooled supercharging pile, which can solve the problem of rapid overcurrent prediction of a liquid-cooled supercharging pile that is difficult to solve with the current prior art.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides an overcurrent prediction method for a liquid-cooled supercharging charging pile, comprising the following steps:

[0007] S10, collecting charging parameters and heat exchange parameters during the charging process of the liquid-cooled supercharging pile;

[0008] S20, designing a TS-HLL structure according to charging parameters and heat exchange parameters; the TS-HLL is time-sensitive HyperLogLog; the design of the TS-HLL structure is to introduce the time dimension into the HyperLogLog algorithm, and create an independent HLL structure for each collection moment;

[0009] S30, encoding the charging parameters and heat exchange parameters at each collection moment using the designed TS-HLL structure to obtain a charging parameter code and a heat exchange parameter code;

[0010] S40, for the charging parameter code, using the time sliding window technology, calculate the similarity of the TS-HLL structure between adjacent time windows as an indicator of the parameter change trend, extract the statistical characteristics of the parameters in each time window, and record them as charging characteristics; use the same method to obtain the heat exchange characteristics;

[0011] S50, aligning the charging characteristics and the heat exchange characteristics according to the time window, and calculating the correlation matrix of the charging characteristics and the heat exchange characteristics of all time windows by using the Pearson correlation coefficient;

[0012] S60, performing matrix decomposition on the correlation matrix according to the positive correlation and negative correlation of the charging current by using a threshold segmentation method to obtain a positive correlation matrix and a negative correlation matrix;

[0013] S70, constructing a relationship map of the charging current and other parameters based on the positive correlation matrix and the negative correlation matrix, and using GNN to learn the relationship map to extract the complex nonlinear relationship between the charging current and other parameters; wherein GNN represents a graph neural network;

[0014] S80. Combining the features learned by the GNN with the historical charging data, inputting them into the RNN, and predicting the charging current in the future. If the predicted charging current exceeds a preset safety threshold, an overcurrent warning mechanism is triggered; wherein RNN represents a recurrent neural network.

[0015] On the basis of the above technical solution, the overcurrent prediction method of a liquid-cooled supercharging charging pile of the present invention can also be improved as follows:

[0016] The charging parameters include charging voltage, charging current, charging power, charging time, battery state of charge, and battery health status.

[0017] Among them, the heat exchange parameters include the coolant temperature, coolant flow, charging gun temperature, battery temperature, ambient temperature, cooling system power, and radiator efficiency of the liquid-cooled supercharging pile.

[0018] The collection time interval is 0.1~1 second.

[0019] The charging characteristics include mean, variance, maximum value and minimum value.

[0020] Among them, each element of the correlation matrix represents the correlation degree between a certain charging feature and a certain heat exchange feature in all time windows, and the value range is [-1, 1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation.

[0021] The length of the time window is 15 seconds to 120 seconds.

[0022] Among them, an independent HLL structure is created for each collection moment, specifically, the charging parameters, hot exchange parameters and collection time collected at each collection moment are merged into a vector, the vector is converted into Base64 code by Base64 serialization, and then the 64-bit hash value of the Base64 code is extracted.

[0023] The step of determining the safety threshold specifically includes:

[0024] Step 1: Collect historical charging data, including data on normal charging and overcurrent events.

[0025] Step 2: Perform statistical analysis on the data and calculate the mean and standard deviation of the normal charging current.

[0026] Step 3: Set the safety factor k (usually 2-3), and preliminarily set the safety threshold to: mean + k * standard deviation.

[0027] Step 4: Adjust the preliminary threshold based on equipment specifications and safety standards.

[0028] Step 5: Conduct experimental verification to ensure that the safety threshold can effectively warn of overcurrent without frequent false alarms.

[0029] Step 6: Regularly update the safety threshold based on new data to maintain its applicability.

[0030] The structure of the GNN is specifically:

[0031] Input layer: takes the node features in the relationship graph as input.

[0032] Multiple graph convolutional layers: Each layer contains message passing and feature aggregation operations.

[0033] Pooling layer: aggregates graph-level features.

[0034] Fully connected layer: performs the final feature transformation.

[0035] Output layer: Outputs the learned charging current feature representation.

[0036] The learning process of the GNN is specifically:

[0037] Step 101: Initialize the GNN network parameters.

[0038] Step 102, initializing the GWO population: representing each individual as a set of GNN network parameters, and taking the node and edge weights in the relationship graph as the optimization target of GWO; the GWO represents the gray wolf hunting algorithm;

[0039] Step 103, forward propagation: passing information through the graph structure and updating the node representation;

[0040] Step 104, calculate the loss: compare the predicted result with the true label to obtain the fitness;

[0041] Step 105, update alpha, beta, delta wolves: update the positions of alpha, beta, delta wolves in the population according to the GWO algorithm mechanism;

[0042] Step 106, update the population: based on the positions of alpha, beta, and delta wolves, update the entire GWO population;

[0043] Step 107, determine whether convergence: if the set number of iterations or convergence conditions are reached, jump to step 109; otherwise, repeat steps 103-106;

[0044] Step 108, output the optimal GNN model parameters: After iterative optimization of the GWO algorithm, the optimal GNN model parameters are obtained.

[0045] The structure of the RNN is specifically:

[0046] Input layer: receives the features and historical charging data output by GNN.

[0047] Multiple LSTM or GRU layers: process time series information.

[0048] Fully connected layer: performs feature transformation.

[0049] Output layer: predict future charging current.

[0050] The learning process of the RNN is specifically:

[0051] Step 201: Initialize network parameters.

[0052] Step 202: Input sequence data in chronological order.

[0053] Step 203. Forward propagation: Process the sequence through the loop structure.

[0054] Step 204. Calculate loss: compare the predicted value with the true value.

[0055] Step 205. Back Propagation Through Time (BPTT): Calculate gradients and update parameters.

[0056] Step 206. Iterative training: repeat steps 202-205 until convergence.

[0057] The use of TS-HLL to encode two sets of parameters has the following effects: significantly reducing data storage space and memory usage; speeding up data processing and improving system response efficiency; retaining time sensitivity and effectively capturing the trend of parameter changes over time; reducing computational complexity while ensuring accuracy; improving system scalability and adapting to large-scale data processing needs; providing efficient feature representation for GNN and RNN inputs, accelerating model training and reasoning processes. Furthermore, the use of the Gray Wolf algorithm to improve the efficiency of GNN allows charging piles to complete model learning at any time and apply the model to predict overcurrent conditions.

[0058] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned overcurrent prediction method for a liquid-cooled supercharging charging pile.

[0059] A third aspect of the present invention provides an overcurrent prediction system for a liquid-cooled supercharging pile, which includes the above-mentioned computer-readable storage medium.

[0060] Compared with the prior art, the present invention provides a method, medium and system for predicting overcurrent of a liquid-cooled supercharging pile, which have the following beneficial effects:

[0061] 1. The time-sensitive HyperLogLog (TS-HLL) data structure is used to efficiently encode charging parameters and hot swap parameters. Compared with traditional data storage methods, the TS-HLL structure not only greatly compresses the amount of data and reduces memory usage, but also retains the dynamic characteristics of parameters changing over time, laying the foundation for subsequent time series analysis; due to the reduction in memory usage and computing consumption, the overcurrent prediction of liquid-cooled supercharging charging piles can be quickly realized under wired computing resources. This design greatly improves the processing efficiency and scalability of the system and can adapt to the needs of large-scale data.

[0062] 2. Through correlation analysis and relationship graph modeling, the complex nonlinear relationship between charging current and other parameters is captured. The learning method based on graph neural network (GNN) can effectively extract the feature representation of this complex relationship, providing a powerful input for subsequent current prediction. Compared with the traditional linear model, the method of the present invention can more accurately characterize the relationship between parameters and improve the accuracy of overcurrent prediction.

[0063] 3. The recurrent neural network (RNN) is combined with the features learned by GNN to predict the charging current in the future. When the predicted value exceeds the preset safety threshold, the overcurrent warning mechanism can be triggered in time, providing strong guarantee for the safe operation of liquid-cooled supercharging charging piles. This data-driven prediction method can adapt to the changes in charging pile behavior under complex working conditions and is more flexible and reliable than traditional empirical rules.

[0064] 4. Innovative technologies such as time-sensitive parameter encoding, complex relationship modeling, and efficient data processing are widely used in the entire overcurrent prediction process. These technical means not only improve the response speed and scalability of the system, but also can adaptively capture the dynamic changes of parameters during the charging process, providing technical support for the intelligent management of liquid-cooled supercharging charging piles.

[0065] In summary, the present invention solves the problem of overcurrent prediction of a fast liquid-cooled supercharging charging pile which is difficult to solve in the current prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0068] like Figure 1 As shown, it is a flow chart of an overcurrent prediction method for a liquid-cooled supercharging charging pile provided by the present invention, and the method comprises the following steps:

[0069] S10. Collect charging parameters and heat exchange parameters during the charging process of the liquid-cooled supercharging pile, wherein the charging parameters include charging voltage, charging current, charging power, charging time, battery state of charge, and battery health status; the heat exchange parameters include the coolant temperature, coolant flow, charging gun temperature, battery temperature, ambient temperature, cooling system power, and radiator efficiency of the liquid-cooled supercharging pile; wherein the collection time interval is 0.1 to 1 second;

[0070] S20, designing a TS-HLL structure according to charging parameters and heat exchange parameters; the TS-HLL is time-sensitive HyperLogLog; the design of the TS-HLL structure is to introduce the time dimension into the HyperLogLog algorithm, and create an independent HLL structure for each collection moment;

[0071] S30, encoding the charging parameters and heat exchange parameters at each collection moment using the designed TS-HLL structure to obtain a charging parameter code and a heat exchange parameter code;

[0072] S40. For the charging parameter code, the time sliding window technology is used to calculate the similarity of the TS-HLL structure between adjacent time windows as an indicator of the parameter change trend, and the statistical characteristics of the parameters in each time window are extracted and recorded as charging characteristics, including mean, variance, maximum value, and minimum value; the heat exchange characteristics are obtained in the same way;

[0073] S50, aligning the charging characteristics and the heat exchange characteristics according to the time window, and using the Pearson correlation coefficient to calculate the correlation matrix of the charging characteristics and the heat exchange characteristics of all time windows, wherein each element of the correlation matrix represents the degree of correlation between a certain charging characteristic and a certain heat exchange characteristic in all time windows, and the value range is [-1, 1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation;

[0074] S60, performing matrix decomposition on the correlation matrix according to the positive correlation and negative correlation of the charging current by using a threshold segmentation method to obtain a positive correlation matrix and a negative correlation matrix;

[0075] S70, constructing a relationship map of the charging current and other parameters based on the positive correlation matrix and the negative correlation matrix, and using GNN to learn the relationship map to extract the complex nonlinear relationship between the charging current and other parameters; wherein GNN represents a graph neural network;

[0076] S80. Combining the features learned by the GNN with the historical charging data, inputting them into the RNN, and predicting the charging current in the future. If the predicted charging current exceeds a preset safety threshold, an overcurrent warning mechanism is triggered; wherein RNN represents a recurrent neural network.

[0077] The specific implementation methods of the above steps are described in detail below:

[0078] Step S10: Collect charging parameters and heat exchange parameters during the charging process of the liquid-cooled supercharging pile. The specific implementation is as follows:

[0079] 1. Through the sensors installed on the liquid-cooled supercharger, the charging parameters such as charging voltage, charging current, charging power, charging time, battery state of charge, battery health status, etc. are collected at intervals of 0.1 to 1 second. At the same time, heat exchange parameters such as coolant temperature, coolant flow, charging gun temperature, battery temperature, ambient temperature, cooling system power, and radiator efficiency also need to be collected.

[0080] 2. The collected charging parameters and heat exchange parameters are recorded in real time in a database or other data storage medium for subsequent analysis. These parameter data can fully reflect the working status and heat exchange of the liquid-cooled supercharging pile.

[0081] 3. The time interval for data collection is set to 0.1~1 second, which can fully capture the dynamic characteristics of parameters changing with time and lay the foundation for subsequent time-sensitive analysis.

[0082] Through this step, comprehensive parameter data of the liquid-cooled supercharger during the charging process were obtained, providing reliable input for subsequent overcurrent prediction. These parameter data reflect the working status and heat exchange of the charging pile, which is crucial for analyzing abnormal phenomena during the charging process.

[0083] Step S20: Design a time-sensitive HyperLogLog (TS-HLL) structure based on charging parameters and heat exchange parameters. The specific implementation is as follows:

[0084] 1.HyperLogLog (HLL) is a data structure based on probability statistics that can estimate the number of independent elements in a data stream within a limited memory space.

[0085] 2. In order to capture the characteristics of parameters changing over time, the time dimension is introduced into the HLL algorithm, and an independent HLL structure is created for each acquisition moment, that is, a time-sensitive HLL (TS-HLL) structure is designed.

[0086] 3. Each TS-HLL structure can encode the charging parameters and hot swap parameters at that moment and retain the time information. This not only greatly reduces the data storage space, but also can effectively reflect the trend of parameter changes over time.

[0087] 4. Compared with a single HLL, the TS-HLL structure can retain both time information and parameter encoding, providing a basis for subsequent time series analysis. This design greatly improves the scalability of the system and is conducive to meeting large-scale data processing needs.

[0088] Through this step, a novel TS-HLL data structure was designed, which greatly compressed the data while retaining time sensitivity, laying the foundation for subsequent parameter analysis and overcurrent prediction. This design greatly improved the response efficiency and scalability of the system.

[0089] Step S30: Encode the charging parameters and heat exchange parameters at each acquisition moment using the designed TS-HLL structure to obtain a charging parameter code and a heat exchange parameter code. The specific implementation is as follows:

[0090] 1. For each charging parameter at each acquisition moment, such as charging voltage, charging current, charging power, etc., encode it into the TS-HLL structure at the corresponding moment to obtain the charging parameter code.

[0091] 2. Similarly, for each heat exchange parameter at each acquisition moment, such as coolant temperature, coolant flow, battery temperature, etc., it is encoded into the TS-HLL structure at the corresponding moment to obtain the heat exchange parameter code.

[0092] 3. This encoding method makes full use of the time sensitivity and compressibility of the TS-HLL structure, greatly reducing the storage space while retaining the timing information of parameter changes.

[0093] 4. The obtained charging parameter code and hot swap parameter code can be used as input features for subsequent analysis and prediction, providing efficient feature representation for the training of GNN and RNN models.

[0094] Through this step, the original charging parameters and hot swap parameters are encoded into the TS-HLL structure, generating a more time-sensitive parameter code. This lays the foundation for subsequent feature extraction and model training, greatly improving the processing efficiency of the system.

[0095] Step S40: For the charging parameter code, the time sliding window technology is used to calculate the similarity of the TS-HLL structure between adjacent time windows as an indicator of parameter change trend, and the statistical characteristics of the parameters in each time window are extracted and recorded as charging characteristics; the heat exchange characteristics are obtained in the same way. The specific implementation method is as follows:

[0096] 1. For the charging parameter code, the time sliding window technology is used to calculate the similarity of the TS-HLL structure between adjacent time windows. Here, indicators such as Jaccard similarity or cosine similarity can be used to quantify the similarity of the TS-HLL structure.

[0097] 2. This similarity index can reflect the changing trend of parameters in adjacent time windows and provide valuable information for subsequent parameter analysis.

[0098] 3. At the same time, the statistical characteristics of the charging parameter code in each time window are extracted, including the mean, variance, maximum value, minimum value, etc., to form a charging feature vector.

[0099] 4. For the hot exchange parameter code, the same method as the charging parameter code is used to extract the similarity of adjacent time windows and the statistical features within each time window to form a hot exchange feature vector.

[0100] Through this step, two sets of important feature vectors are obtained: charging features and heat exchange features. These features not only reflect the changing trend of parameters over time, but also contain their own statistical characteristics, which provide a basis for subsequent correlation analysis and overcurrent prediction.

[0101] Step S50: Align the charging characteristics and heat exchange characteristics according to the time window, and use the Pearson correlation coefficient to calculate the correlation matrix of the charging characteristics and heat exchange characteristics of all time windows. The specific implementation is as follows:

[0102] 1. Since the charging features and hot swap features are extracted based on different time windows, their time series need to be aligned to ensure that the features of each time window correspond.

[0103] 2. After the alignment is completed, the Pearson correlation coefficient is used to calculate the correlation between the charging characteristics and the heat exchange characteristics in the entire time window.

[0104] 3. Pearson correlation coefficient The value range is ,in indicates a completely positive correlation, indicates a completely negative correlation, Indicates no correlation.

[0105] 4. Through calculation, a correlation matrix between charging characteristics and heat exchange characteristics is obtained. Each element of the matrix represents the degree of correlation between a charging characteristic and a heat exchange characteristic in all time windows.

[0106] The purpose of this step is to explore the correlation between various parameters in the charging process, laying the foundation for subsequent relationship graph modeling and feature learning. The correlation matrix contains the complex relationship between charging parameters and heat exchange parameters, providing important clues for overcurrent prediction.

[0107] Step S60: Decompose the correlation matrix by threshold segmentation for the positive correlation and negative correlation of the charging current to obtain a positive correlation matrix and a negative correlation matrix. The specific implementation is as follows:

[0108] 1. Set a threshold , judge each element in the correlation matrix:

[0109] if , then keep the element in the positive correlation matrix;

[0110] if , then keep the element in the negative correlation matrix;

[0111] Otherwise, the element is discarded.

[0112] 2. Through this threshold segmentation, two new matrices can be obtained: positive correlation matrix and negative correlation matrix. These two matrices reflect the positive correlation and negative correlation between charging current and other parameters respectively.

[0113] 3. Threshold The setting needs to be adjusted according to the actual situation. or Too large a threshold may lose some valuable correlation information, while too small a threshold may introduce more noise.

[0114] The purpose of this step is to extract the part that is most relevant to the charging current from the correlation between the charging parameters and the heat exchange parameters, and to provide targeted information for the subsequent relationship map construction and parameter prediction.

[0115] Step S70: Based on the positive correlation matrix and the negative correlation matrix, a relationship map between the charging current and other parameters is constructed, and a graph neural network (GNN) is used to learn the relationship map to extract the complex nonlinear relationship between the charging current and other parameters. The specific implementation is as follows:

[0116] 1. Represent the positive correlation matrix and the negative correlation matrix as a directed graph, where the nodes represent the charging parameters and the heat exchange parameters, and the edge weights represent the degree of correlation between them.

[0117] 2. The relationship diagram constructed in this way can intuitively show the complex relationship between charging current and other parameters.

[0118] 3. In order to learn this complex nonlinear relationship, graph neural network (GNN) is used to model the relationship graph.

[0119] 4. The GNN model includes an input layer, multiple graph convolutional layers, a pooling layer, and a fully connected layer, which can effectively capture the features in graph structured data.

[0120] 5. During the learning process of GNN, the Grey Wolf Algorithm (GWO) is used to optimize the network parameters to improve the learning efficiency of the model.

[0121] 6. Finally, the GNN model is able to extract the complex nonlinear relationship between charging current and other parameters, providing important feature representation for subsequent current prediction.

[0122] Through this step, a graph model reflecting the relationship between charging current and other parameters is constructed, and GNN is used to effectively learn it. This provides a powerful feature representation capability for subsequent overcurrent prediction and helps capture the complex relationship between parameters.

[0123] Step S80: Combine the features learned by GNN with historical charging data and input them into a recurrent neural network (RNN) to predict the charging current in the future. If the predicted charging current exceeds the preset safety threshold, the overcurrent warning mechanism is triggered. The specific implementation is as follows:

[0124] 1. The RNN model includes an input layer, multiple LSTM or GRU layers, a fully connected layer, and an output layer.

[0125] 2. Input the features learned by the GNN model and the historical charging data into the RNN model, and use the RNN's time series modeling capability to predict the charging current in the future.

[0126] 3. For the predicted charging current, a safety threshold needs to be set in advance. When the predicted value exceeds the threshold, the overcurrent warning mechanism will be triggered.

[0127] The steps to determine the safety threshold are as follows:

[0128] Step 1. Collect historical charging data, including data of normal charging and overcurrent events.

[0129] Step 2. Perform statistical analysis on the data and calculate the average value of normal charging current and standard deviation .

[0130] Step 3. Set the safety factor (Usually take ), the safety threshold is initially set as: .

[0131] Step 4. Adjust the preliminary thresholds based on equipment specifications and safety standards.

[0132] Step 5. Conduct experimental verification to ensure that the safety threshold can effectively warn of overcurrent without frequent false alarms.

[0133] Step 6. Regularly update the safety threshold based on new data to maintain its applicability.

[0134] Through this step, the RNN model is used to predict the future charging current and trigger the overcurrent warning according to the preset safety threshold. The determination of the safety threshold needs to be adjusted in combination with historical data, equipment specifications and actual usage to ensure the effectiveness and reliability of the warning mechanism.

[0135] In summary, the technical solution of this invention includes the following key steps:

[0136] 1. Collect the charging parameters and heat exchange parameters of the liquid-cooled supercharging pile at a time interval of 0.1 to 1 second.

[0137] 2. Design a time-sensitive HyperLogLog (TS-HLL) structure to efficiently encode the collected parameters.

[0138] 3. Extract the statistical characteristics of charging parameters and heat exchange parameters, and calculate the correlation matrix between them.

[0139] 4. Based on the correlation matrix, a relationship map between charging current and other parameters is constructed, and a graph neural network (GNN) is used to learn it to extract the complex nonlinear relationship between charging current and other parameters. During the GNN learning process, the Grey Wolf Algorithm (GWO) is used to optimize the network parameters to improve the learning efficiency of the model.

[0140] The specific implementation process of the GWO algorithm is as follows:

[0141] Step 101. Initialize the GNN network parameters.

[0142] Step 102. Initialize the GWO population: represent each individual as a set of GNN network parameters, and use the node and edge weights in the relationship graph as the optimization target of GWO.

[0143] Step 103. Forward propagation: pass information through the graph structure and update node representation.

[0144] Step 104. Calculate loss: Compare the predicted results with the true labels to get the fitness.

[0145] Step 105. Update alpha, beta, delta wolves: Update the positions of alpha, beta, delta wolves in the population according to the GWO algorithm mechanism.

[0146] Step 106. Update the population: Based on the positions of alpha, beta, and delta wolves, update the entire GWO population.

[0147] Step 107. Determine whether convergence occurs: if the set number of iterations or convergence condition is reached, jump to step 109; otherwise, repeat steps 103-106.

[0148] Step 108. Output the optimal GNN model parameters: After iterative optimization of the GWO algorithm, the optimal GNN model parameters are obtained.

[0149] The specific structure of the GNN model is as follows:

[0150] Input layer: takes the node features in the relationship graph as input.

[0151] Multiple graph convolutional layers: Each layer contains message passing and feature aggregation operations.

[0152] Pooling layer: aggregates graph-level features.

[0153] Fully connected layer: performs the final feature transformation.

[0154] Output layer: Outputs the learned charging current feature representation.

[0155] Through this step, a model based on graph neural network is constructed, which can effectively capture the complex nonlinear relationship between charging current and other parameters. The introduction of gray wolf algorithm improves the learning efficiency of GNN model and provides powerful feature representation capability for subsequent current prediction.

[0156] 4. Input the features learned by GNN together with the historical charging data into the RNN model, and use the time series modeling capability of RNN to predict the charging current in the future.

[0157] The specific structure of the RNN model is as follows:

[0158] Input layer: receives the features and historical charging data output by GNN.

[0159] Multiple LSTM or GRU layers: process time series information.

[0160] Fully connected layer: performs feature transformation.

[0161] Output layer: predict future charging current.

[0162] The learning process of the RNN model is as follows:

[0163] Step 201: Initialize network parameters.

[0164] Step 202: Input sequence data in chronological order.

[0165] Step 203. Forward propagation: Process the sequence through the loop structure.

[0166] Step 204. Calculate loss: compare the predicted value with the true value.

[0167] Step 205. Back Propagation Through Time (BPTT): Calculate gradients and update parameters.

[0168] Step 206. Iterative training: repeat steps 202-205 until convergence.

[0169] Through this step, the RNN model is used to predict the future charging current. The time series modeling capability of RNN, combined with the complex features learned by GNN, can effectively predict the future charging current change trend.

[0170] Finally, a safety threshold needs to be set. When the predicted charging current exceeds the threshold, the overcurrent warning mechanism will be triggered. The process of determining the safety threshold is as follows:

[0171] Step 1. Collect historical charging data, including data of normal charging and overcurrent events.

[0172] Step 2. Perform statistical analysis on the data and calculate the average value of normal charging current and standard deviation .

[0173] Step 3. Set the safety factor (Usually take ), the safety threshold is initially set as: .

[0174] Step 4. Adjust the preliminary thresholds based on equipment specifications and safety standards.

[0175] Step 5. Conduct experimental verification to ensure that the safety threshold can effectively warn of overcurrent without frequent false alarms.

[0176] Step 6. Regularly update the safety threshold based on new data to maintain its applicability.

[0177] The key to this step is to determine an appropriate safety threshold based on historical data and actual conditions, which can effectively warn of overcurrent without frequently triggering false alarms. At the same time, as new data accumulates, the safety threshold needs to be updated regularly to ensure its continued applicability.

[0178] In summary, the specific implementation of this invention includes the following steps:

[0179] 1. Collect the charging parameters and heat exchange parameters of the liquid-cooled supercharging pile.

[0180] 2. Design a time-sensitive HyperLogLog (TS-HLL) structure to encode the parameters.

[0181] 3. Extract the statistical characteristics of charging parameters and heat exchange parameters, and calculate the correlation matrix between them.

[0182] 4. Based on the correlation matrix, a relationship graph between charging current and other parameters is constructed, and the graph neural network (GNN) is used to learn it. During the learning process of GNN, the grey wolf algorithm (GWO) is used to optimize the network parameters.

[0183] 5. Combine the features learned by GNN with historical charging data and input them into the recurrent neural network (RNN) to predict the charging current in the future.

[0184] 6. Set a safety threshold. When the predicted charging current exceeds the threshold, the overcurrent warning mechanism is triggered.

[0185] This invention makes full use of time-sensitive parameter encoding, correlation analysis, graph neural network and recurrent neural network technologies to achieve overcurrent prediction during the charging process of liquid-cooled supercharging charging piles. The key technical innovations include:

[0186] 1. A time-sensitive TS-HLL data structure was designed, which greatly compressed the parameter data and improved the response efficiency and scalability of the system.

[0187] 2. Through correlation analysis and relationship map construction, the complex nonlinear relationship between charging current and other parameters is captured.

[0188] 3. The GNN and GWO algorithms are used to achieve effective learning of the relationship graph, providing powerful feature representation for subsequent current prediction.

[0189] 4. Combine the time series modeling capabilities of RNN to predict future charging current and set safety thresholds to trigger overcurrent warnings.

[0190] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned overcurrent prediction method for a liquid-cooled supercharging charging pile.

[0191] A third aspect of the present invention provides an overcurrent prediction system for a liquid-cooled supercharging pile, which includes the above-mentioned computer-readable storage medium.

[0192] Specifically, the principle of the present invention is to make full use of the parameter data collected during the charging process, and accurately describe the complex relationship between the charging current and other parameters through advanced data processing and modeling technology, thereby achieving effective prediction of future overcurrent events. The specific technical principles are as follows:

[0193] First, the present invention uses a time-sensitive HyperLogLog (TS-HLL) data structure to encode charging parameters and hot swap parameters. HyperLogLog is a data structure based on probability statistics, which can estimate the number of independent elements in a data stream within a limited memory space. By introducing the time dimension into the HLL algorithm and creating an independent TS-HLL structure for each acquisition moment, it can not only greatly compress data storage, but also retain the dynamic characteristics of parameters changing over time. This time-sensitive encoding method lays the foundation for subsequent parameter analysis and relationship modeling; at the same time, it greatly reduces the resource consumption in the storage and reading process of the computing unit during the calculation process, improves the computing efficiency under limited resources, and helps to achieve overcurrent prediction faster.

[0194] Secondly, based on the charging parameter code and hot swap parameter code obtained by TS-HLL encoding, the time sliding window technology is used to extract the statistical characteristics of each parameter in different time windows, including mean, variance, maximum and minimum values. At the same time, the similarity of TS-HLL structure between adjacent time windows is calculated as an indicator of parameter change trend. These features can not only reflect the dynamic changes of parameters over time, but also lay the foundation for exploring the correlation between charging parameters and hot swap parameters.

[0195] Then, the correlation matrix between the charging characteristics and the heat exchange characteristics was calculated using the Pearson correlation coefficient. This matrix contains the complex correlation between the various parameters in the charging process. In order to further extract the part most related to the charging current, the positive correlation matrix and negative correlation matrix of the charging current were obtained by threshold segmentation. This targeted relationship extraction provides targeted input features for subsequent model training and parameter prediction.

[0196] Based on the relationship matrix obtained from the above analysis, a graph model reflecting the relationship between charging current and other parameters is constructed. In order to learn this complex nonlinear relationship, a graph neural network (GNN) is used to model the relationship graph. GNN can effectively capture the features in graph structure data and transform them into useful feature representations. In the training process of GNN, the gray wolf algorithm (GWO) is introduced to optimize the network parameters to improve the learning efficiency of the model.

[0197] Finally, the features learned by GNN are input into the recurrent neural network (RNN) together with the historical charging data, and the charging current in the future is predicted using the time series modeling capability of RNN. When the predicted value exceeds the preset safety threshold, the overcurrent warning mechanism is triggered. This data-driven prediction method can adapt to the changes in the behavior of charging piles under complex working conditions and is more flexible and reliable than traditional empirical rules.

[0198] In summary, the technical solution of the present invention makes full use of innovative technologies such as time-sensitive parameter encoding, complex relationship modeling, and efficient data processing. This integrated innovative method can accurately capture the complex relationship between parameters in the charging process, and provides solid technical support for the effective prediction and prevention of overcurrent accidents. The core of this method is to start from the data and achieve in-depth understanding and accurate prediction of charging pile behavior through advanced modeling and analysis techniques.

[0199] In order to better understand and implement the present invention, a specific embodiment 1 of the method of the present invention is provided below. The specific steps of this embodiment 1 are described as follows: Step S10: Collect charging parameters and heat exchange parameters during the charging process of the liquid-cooled supercharging pile.

[0200] For the collection of charging parameters, set the collection time interval to Seconds. At each collection moment , record the following parameters:

[0201] Charging voltage , Charging current , Charging power , Charging time , Battery state of charge and battery health status .

[0202] For the acquisition of heat exchange parameters, record the following parameters:

[0203] Coolant temperature , Coolant flow , Charging gun temperature , Battery temperature , Ambient temperature , Cooling system power and radiator efficiency .

[0204] The collected charging parameters and heat exchange parameters are recorded in real time in the data storage medium to provide reliable input data for subsequent analysis.

[0205] Step S20: Designing a time-sensitive HyperLogLog (TS-HLL) structure according to charging parameters and hot swap parameters.

[0206] HyperLogLog is a data structure based on probability statistics, which can estimate the number of independent elements in a data stream within a limited memory space. Create an independent TS-HLL structure, and its encoding process is as follows:

[0207] For charging parameters , encode it into TS-HLL(t) to get the charging parameter code .

[0208] Similarly, for the heat exchange parameters , encode it into TS-HLL(t) to get the hot exchange parameter code .

[0209] This design not only greatly compresses the data storage space, but also retains the time information of each acquisition moment, laying the foundation for subsequent time series analysis.

[0210] Step S30: Encode the charging parameters and heat exchange parameters at each acquisition moment using the designed TS-HLL structure.

[0211] For charging parameters , encode it into TS-HLL(t) to get the charging parameter code , where the encoding process can be expressed as:

[0212] ;

[0213] ;

[0214] ;

[0215] ;

[0216] ;

[0217] ;

[0218] Similarly, for the heat exchange parameters , encode it into TS-HLL(t) to get the hot exchange parameter code , the encoding process is:

[0219] ;

[0220] ;

[0221] ;

[0222] ;

[0223] ;

[0224] ;

[0225] ;

[0226] This encoding method makes full use of the time sensitivity and compressibility of the TS-HLL structure, greatly reducing the storage space while retaining the timing information of parameter changes.

[0227] Step S40: For the charging parameter code, the time sliding window technology is used to calculate the similarity of the TS-HLL structure between adjacent time windows as an indicator of the parameter change trend, and the statistical characteristics of the parameters in each time window are extracted.

[0228] For charging parameter code ,The time sliding window technique is used to calculate the TS-HLL structural similarity of adjacent ,time windows as an indicator of the parameter change trend.

[0229] For time windows The charging parameter code in the , can calculate its statistical characteristics, including:

[0230] Mean ;

[0231] variance ;

[0232] Maximum ;

[0233] Minimum ;

[0234] The same method can also be used to obtain the hot exchange parameter code. The corresponding statistical characteristics.

[0235] Through this step, the statistical characteristics of charging parameters and heat exchange parameters within the time window are extracted, laying the foundation for subsequent correlation analysis and feature learning.

[0236] Step S50: aligning the charging characteristics and the heat exchange characteristics according to the time window, and using the Pearson correlation coefficient to calculate the correlation matrix of the charging characteristics and the heat exchange characteristics of all time windows.

[0237] Assume the charging characteristic vector is , the heat exchange characteristic vector is In order to calculate the correlation between them, it is necessary to first align the correspondence between the two sets of feature vectors in the time dimension.

[0238] For time windows , the charging characteristics can be obtained and heat exchange characteristics . Calculate the Pearson correlation coefficient matrix for all time windows ,in express and The correlation degree in all time windows ranges from .

[0239] ;

[0240] in is the time average of each feature, is the total number of time windows.

[0241] Through this step, the correlation matrix between charging characteristics and heat exchange characteristics is obtained, which provides an important basis for subsequent relationship analysis and feature learning.

[0242] Step S60: Decomposing the correlation matrix by threshold segmentation for the positive correlation and negative correlation of the charging current.

[0243] Setting the Threshold , for the correlation matrix For each element in the test:

[0244] if , then keep the element in the positive correlation matrix middle;

[0245] if , then keep the element in the negative correlation matrix middle;

[0246] Otherwise, the element is discarded.

[0247] Through this threshold segmentation, two new matrices can be obtained: positive correlation matrix And the negative correlation matrix The two matrices reflect the positive correlation and negative correlation between the charging current and other parameters respectively.

[0248] Threshold The setting needs to be adjusted according to the actual situation. or Too large a threshold may lose some valuable correlation information, while too small a threshold may introduce more noise.

[0249] The purpose of this step is to extract the part that is most relevant to the charging current from the correlation between the charging parameters and the heat exchange parameters, and to provide targeted information for the subsequent relationship map construction and parameter prediction.

[0250] Step S70: Based on the positive correlation matrix and the negative correlation matrix, a relationship map between the charging current and other parameters is constructed, and a graph neural network (GNN) is used to learn the relationship map to extract the complex nonlinear relationship between the charging current and other parameters.

[0251] The positive correlation matrix And the negative correlation matrix Represented as a directed graph , where the node set represents charging parameters and heat exchange parameters, edge set Indicates the degree of correlation between them.

[0252] For nodes , its eigenvector can be expressed as , which is composed of charging characteristics and heat exchange characteristics.

[0253] In order to learn this complex nonlinear relationship, a graph neural network (GNN) is used to map the relationship graph. Modeling. The GNN model includes the following layer structures:

[0254] Input layer: receiving node features as input.

[0255] Multiple graph convolutional layers: Each layer contains message passing and feature aggregation operations, which can effectively capture the features in graph structured data.

[0256] Pooling layer: aggregates graph-level features.

[0257] Fully connected layer: performs the final feature transformation.

[0258] Output layer: Outputs the learned charging current feature representation.

[0259] In the learning process of GNN, the gray wolf algorithm (GWO) is used to optimize the network parameters to improve the learning efficiency of the model. GWO is a meta-heuristic algorithm based on swarm intelligence, which optimizes the objective function by simulating the hunting behavior of gray wolves. The specific learning process is as follows:

[0260] Step 101. Initialize GNN network parameters .

[0261] Step 102. Initialize the GWO population: represent each individual as a set of GNN network parameters , and the relationship map The node and edge weights in are used as the optimization objectives of GWO.

[0262] Step 103. Forward propagation: pass information through the graph structure and update node representation .

[0263] Step 104. Calculate the loss: Compare the predicted result with the true label (here taking charging current as an example) to get the fitness .

[0264] Step 105. Update alpha, beta, and delta wolves: Update the positions of alpha, beta, and delta wolves in the population according to the GWO algorithm mechanism .

[0265] Step 106. Update the population: Update the entire GWO population based on the positions of alpha, beta, and delta wolves .

[0266] Step 107. Determine whether convergence occurs: if the set number of iterations or convergence condition is reached, jump to step 109; otherwise, repeat steps 103-106.

[0267] Step 108. Output the optimal GNN model parameters :After iterative optimization of the GWO algorithm, the optimal GNN model parameters are obtained.

[0268] Through this step, a model based on graph neural network is constructed, which can effectively capture the complex nonlinear relationship between charging current and other parameters. The introduction of gray wolf algorithm improves the learning efficiency of GNN model and provides powerful feature representation capability for subsequent current prediction.

[0269] Step S80: Combine the features learned by GNN with historical charging data and input them into a recurrent neural network (RNN) to predict the charging current in the future. If the predicted charging current exceeds the preset safety threshold, the overcurrent warning mechanism is triggered.

[0270] The features learned by the GNN model and historical charging data The RNN model is used to input the charging current in the future using the RNN's time series modeling capability. Make predictions, where is the number of prediction steps.

[0271] The specific structure of the RNN model is as follows:

[0272] Input layer: receives the features output by GNN and historical charging data .

[0273] Multiple LSTM or GRU layers: can effectively process time series information.

[0274] Fully connected layer: performs feature transformation.

[0275] Output layer: output the predicted value of future charging current .

[0276] For the predicted charging current , a safety threshold needs to be set in advance ,When the predicted value exceeds the threshold, the overcurrent warning mechanism will be triggered. The process of determining the safety threshold is as follows:

[0277] Step 1. Collect historical charging data, including data of normal charging and overcurrent events.

[0278] Step 2. Perform statistical analysis on the data and calculate the average value of normal charging current and standard deviation .

[0279] Step 3. Set the safety factor (Usually take ), the safety threshold is initially set as: .

[0280] Step 4. Adjust the preliminary thresholds based on equipment specifications and safety standards.

[0281] Step 5. Conduct experimental verification to ensure that the safety threshold can effectively warn of overcurrent without frequent false alarms.

[0282] Step 6. Regularly update the safety threshold based on new data to maintain its applicability.

[0283] Through this step, the RNN model is used to predict the future charging current and trigger the overcurrent warning according to the preset safety threshold. The determination of the safety threshold needs to be adjusted in combination with historical data, equipment specifications and actual usage to ensure the effectiveness and reliability of the warning mechanism.

[0284] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A bus company in a certain city recently introduced a batch of pure electric buses, and for this purpose, multiple liquid-cooled supercharging charging piles were built along the main bus routes. In order to ensure the safe and stable operation of the charging piles, the bus company decided to adopt the overcurrent prediction method proposed in the present invention to perform real-time monitoring and intelligent early warning of the charging process.

[0285] The basic parameters of the charging pile are shown in Table 1, including battery pack capacity, maximum charging power, and cooling system parameters.

[0286] Table 1 Basic parameters of liquid-cooled supercharger

[0287] Parameter name Parameter Value Battery capacity 400 kWh Maximum charging power 500 kW Coolant flow 30 L / min Coolant inlet temperature 25 °C Ambient temperature 35 °C

[0288] In order to collect key parameters during the charging process, multiple sensors are installed on the charging pile to collect data every 0.5 seconds. The specific parameters collected include: charging voltage , Charging current , Charging power , Charging time , battery state of charge SOC, battery state of health SOH, and coolant temperature , Coolant flow , Charging gun temperature , Battery temperature , Ambient temperature , Cooling system power and radiator efficiency These parameter data are recorded in the database in real time, providing a basis for subsequent analysis and prediction.

[0289] Step S10: Collect charging parameters and heat exchange parameters

[0290] Taking a typical charging process as an example, some parameter data collected within 10 minutes are shown in Table 2. It can be seen that with the passage of charging time, the values ​​of each parameter have changed to a certain extent, reflecting the dynamic characteristics of the charging process.

[0291] Table 2 Data collected from some charging parameters and heat exchange parameters

[0292]

[0293] Step S20: Design TS-HLL data structure

[0294] The charging parameters and heat exchange parameters collected in Table 2 are encoded into the TS-HLL structure respectively. For example, the encoding process is as follows:

[0295] ;

[0296] in, Indicates time Similarly, the encoding results of other parameters can be obtained, as shown in Table 3.

[0297] Table 3 TS-HLL encoding results of some parameters

[0298]

[0299] As can be seen from Table 3, TS-HLL coding greatly compresses the data storage space while retaining the time information of the parameters. For subsequent parameter analysis and model training, this coding method significantly improves the response efficiency and scalability of the system.

[0300] Step S40: Extracting charging characteristics and heat exchange characteristics

[0301] Next, the time sliding window technique is used to extract the statistical characteristics of each parameter in different time windows. and coolant temperature For example, the specific results are shown in Table 4 and Table 5.

[0302] Table 4 Charging current Statistical characteristics of

[0303]

[0304] Table 5 Coolant temperature Statistical characteristics of

[0305]

[0306] From the above characteristics, it can be seen that as the charging time goes by, the charging current continues to increase and the coolant temperature gradually increases. At the same time, the TS-HLL similarity between adjacent time windows is also gradually decreasing, reflecting the intensification of the parameter change trend. These characteristics provide an important basis for subsequent correlation analysis and overcurrent prediction.

[0307] Step S50: Calculate the correlation matrix

[0308] Next, the charging characteristics and heat exchange characteristics are aligned in the time dimension, and the correlation matrix between them is calculated using the Pearson correlation coefficient. , as shown in Table 6.

[0309] Table 6 Correlation matrix between charging characteristics and heat exchange characteristics

[0310]

[0311] It can be seen from Table 6 that the statistical characteristics of the charging current Statistical characteristics of coolant temperature There is a strong positive correlation, which shows that the change of charging current has a significant impact on the heat exchange of the system.

[0312] Step S60: Constructing a relationship graph

[0313] Based on the correlation matrix calculated in the previous step , decompose it according to positive correlation and negative correlation, and get the positive correlation matrix And the negative correlation matrix , as shown in Table 7 and Table 8.

[0314] Table 7 Positive correlation matrix of charging current

[0315]

[0316] Table 8 Negative correlation matrix of charging current

[0317]

[0318] From the above two matrices, we can see that there is a complex correlation between the statistical characteristics of the charging current and the statistical characteristics of the coolant temperature. In order to further characterize this relationship, we construct them into a directed graph ,The nodes represent charging parameters and heat exchange parameters, and the edge weights reflect the ,degree of correlation between them.,Through this visualization method, the complex relationship between charging ,current and other parameters can be understood more intuitively.

[0319] Step S70: Using the GNN model to learn directed graph features

[0320] In order to learn this complex nonlinear relationship, a graph neural network (GNN) is used to map the above relationship graph. Modeling is performed. The specific GNN model structure is as follows:

[0321] Input layer: receiving node features ,in and They represent charging characteristics and heat exchange characteristics respectively.

[0322] Multiple graph convolutional layers: Each layer contains message passing and feature aggregation operations, which can effectively capture the features in graph structured data.

[0323] Pooling layer: aggregates graph-level features.

[0324] Fully connected layer: performs the final feature transformation.

[0325] Output layer: Output the learned charging current feature representation .

[0326] In the training process of GNN, the gray wolf algorithm (GWO) is used to optimize the network parameters. GWO is a meta-heuristic algorithm based on swarm intelligence, which optimizes the objective function by simulating the hunting behavior of gray wolves. The specific optimization process is as follows:

[0327] Step 101. Initialize GNN network parameters .

[0328] Step 102. Initialize the GWO population: represent each individual as a set of GNN network parameters , and the relationship map The node and edge weights in are used as the optimization objectives of GWO.

[0329] Step 103. Forward propagation: pass information through the graph structure and update node representation .

[0330] Step 104. Calculate the loss: Compare the predicted result with the true label (here taking charging current as an example) to get the fitness .

[0331] Step 105. Update alpha, beta, and delta wolves: Update the positions of alpha, beta, and delta wolves in the population according to the GWO algorithm mechanism .

[0332] Step 106. Update the population: Update the entire GWO population based on the positions of alpha, beta, and delta wolves .

[0333] Step 107. Determine whether convergence occurs: if the set number of iterations or convergence condition is reached, jump to step 109; otherwise, repeat steps 103-106.

[0334] Step 108. Output the optimal GNN model parameters :After iterative optimization of the GWO algorithm, the optimal GNN model parameters are obtained.

[0335] Through this combination of GNN and GWO, the complex nonlinear relationship between charging current and other parameters can be effectively learned, providing powerful feature representation for subsequent overcurrent prediction.

[0336] Step S80: Using RNN model to predict overcurrent

[0337] Finally, the features learned by GNN and historical charging data The charging current in the future is input into the recurrent neural network (RNN) Make predictions.

[0338] The specific structure of the RNN model is as follows:

[0339] Input layer: receives the features output by GNN and historical charging data .

[0340] Multiple LSTM or GRU layers: can effectively process time series information.

[0341] Fully connected layer: performs feature transformation.

[0342] Output layer: output the predicted value of future charging current .

[0343] For the predicted charging current , a safety threshold needs to be set in advance ,When the predicted value exceeds the threshold, the overcurrent warning mechanism will be triggered. The process of determining the safety threshold is as follows:

[0344] Step 1. Collect historical charging data, including data of normal charging and overcurrent events.

[0345] Step 2. Perform statistical analysis on the data and calculate the average value of normal charging current A and standard deviation A.

[0346] Step 3. Set the safety factor , the safety threshold is initially set as: A.

[0347] Step 4. Based on the specifications and safety standards of the charging pile, adjust the initial threshold to 550A.

[0348] Step 5. Conduct experimental verification to ensure that the safety threshold can effectively warn of overcurrent without frequent false alarms.

[0349] Step 6. Regularly update the safety threshold based on new data to maintain its applicability.

[0350] Summary of implementation effects:

[0351] By adopting the overcurrent prediction method proposed in the present invention, the bus company has achieved remarkable results in the operation of liquid-cooled supercharging charging piles:

[0352] 1. The TS-HLL encoding technology is used to greatly compress the storage space of parameter data, improving the response efficiency and scalability of the system. While ensuring data accuracy, the memory usage is only about 10% of the original data.

[0353] 2. Through correlation analysis and graph modeling, the complex nonlinear relationship between charging current and other parameters is accurately captured. The application of the GNN model enables the effective learning of the feature representation of this relationship, providing a powerful input for subsequent current prediction.

[0354] 3. The RNN model combined with the GNN features accurately predicted the future charging current. The scientific setting of the warning threshold enables the system to warn in time and prevent overcurrent accidents, and the safety of charging piles has been significantly improved.

[0355] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting overcurrent of a liquid-cooled supercharging pile, characterized in that: The following steps are involved: S10, collecting charging parameters and heat exchange parameters during the charging process of the liquid-cooled supercharging pile; S20, designing a TS-HLL structure according to charging parameters and heat exchange parameters; the TS-HLL is time-sensitive HyperLogLog; the design of the TS-HLL structure is to introduce the time dimension into the HyperLogLog algorithm, and create an independent HLL structure for each collection moment; S30, encoding the charging parameters and heat exchange parameters at each collection moment using the designed TS-HLL structure to obtain a charging parameter code and a heat exchange parameter code; S40, for the charging parameter code, using the time sliding window technology, calculate the similarity of the TS-HLL structure between adjacent time windows as an indicator of the parameter change trend, extract the statistical features of the parameters in each time window, and record them as charging features; In the same way, the heat exchange characteristics are obtained; S50, aligning the charging characteristics and the heat exchange characteristics according to the time window, and calculating the correlation matrix of the charging characteristics and the heat exchange characteristics of all time windows by using the Pearson correlation coefficient; S60, performing matrix decomposition on the correlation matrix according to the positive correlation and negative correlation of the charging current by using a threshold segmentation method to obtain a positive correlation matrix and a negative correlation matrix; S70, constructing a relationship map of the charging current and other parameters based on the positive correlation matrix and the negative correlation matrix, and using GNN to learn the relationship map to extract the complex nonlinear relationship between the charging current and other parameters; wherein GNN represents a graph neural network; S80. Combining the features learned by the GNN with the historical charging data, inputting them into the RNN, and predicting the charging current in the future. If the predicted charging current exceeds a preset safety threshold, an overcurrent warning mechanism is triggered; wherein RNN represents a recurrent neural network.

2. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: The charging parameters include charging voltage, charging current, charging power, charging time, battery state of charge, and battery health status.

3. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: The heat exchange parameters include the coolant temperature, coolant flow, charging gun temperature, battery temperature, ambient temperature, cooling system power, and radiator efficiency of the liquid-cooled supercharging pile.

4. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: The collection time interval is 0.1~1 second.

5. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: The charging characteristics include mean, variance, maximum value, and minimum value.

6. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: Each element of the correlation matrix represents the correlation between a certain charging feature and a certain heat exchange feature in all time windows, and the value range is [-1, 1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation.

7. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: The length of the time window is 15 seconds to 120 seconds.

8. The overcurrent prediction method for a liquid-cooled supercharging pile according to claim 1, characterized in that: An independent HLL structure is created at each collection moment, specifically, the charging parameters, hot exchange parameters and collection moment collected at each collection moment are merged into a vector, the vector is converted into a Base64 code by Base64 serialization, and then the 64-bit hash value of the Base64 code is extracted.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the overcurrent prediction method for a liquid-cooled supercharging pile according to any one of claims 1 to 8.

10. An overcurrent prediction system for a liquid-cooled supercharging pile, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images