A monitoring method and system for charging pile cable and charging pile cable

Abstract

The present invention relates to a monitoring method, system and charging pile cable for charging pile cables. The method: collect and preprocess the charging pile transmission cables to obtain a multi-source time series data set; predict the cable temperature through a hybrid model of a deep neural network and a recurrent neural network to obtain cable temperature prediction data; perform Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result; perform spatiotemporal variation feature analysis and three-dimensional space interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field; perform computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management plan; perform reinforcement learning and hierarchical decision processing on the cable airflow management plan to obtain a cable air cooling temperature control strategy. The implementation of the present invention realizes intelligent air cooling control of charging pile transmission cables and improves the safety of the charging process of charging pile transmission cables.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission cable cooling, and in particular to a monitoring method and system for a charging pile cable and a charging pile cable. Background Art

[0002] With the rapid growth of electric vehicle charging demand, the safe and reliable operation of charging pile transmission cables has become crucial. Although the traditional liquid cooling system has high heat dissipation efficiency, there are many problems in practical application. The liquid cooling system requires complex pipelines and pumping equipment, which not only increases the complexity and maintenance cost of the system, but also is prone to leakage risks, causing potential harm to the surrounding environment. In addition, the electrical insulation performance of the liquid medium is poor, and once a leak occurs, it may cause safety hazards such as short circuits.

[0003] Existing cable monitoring technology mainly relies on decentralized temperature sensors, which makes it difficult to achieve comprehensive monitoring of the entire cable. This point-based monitoring method is prone to blind spots and cannot detect abnormal conditions such as local overheating in a timely manner. At the same time, traditional monitoring systems lack intelligent analysis and prediction capabilities, making it difficult to accurately predict the temperature change trend of cables, which is not conducive to taking preventive maintenance measures in a timely manner. In addition, existing cable cooling control strategies often use simple switch control or fixed mode control, which is difficult to flexibly adjust according to actual load and environmental conditions. This extensive control method not only reduces cooling efficiency, but also may cause energy waste. Summary of the invention

[0004] The objective of the present invention is to provide a monitoring method and system for a charging pile cable and a charging pile cable, so as to realize intelligent air cooling control of the charging pile transmission cable and improve the safety of the charging process of the charging pile transmission cable.

[0005] To achieve the above object, the present invention provides a method for monitoring a charging pile cable, comprising the following steps:

[0006] The temperature data, environmental parameters and airflow information of the charging pile transmission cable are collected and preprocessed to obtain a multi-source time series data set;

[0007] Inputting the multi-source time series data set into a hybrid model of a deep neural network and a recurrent neural network to predict cable temperature, thereby obtaining cable temperature prediction data;

[0008] Performing Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result;

[0009] Performing temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field;

[0010] Performing computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management solution;

[0011] Reinforcement learning and hierarchical decision processing are performed on the cable airflow management scheme to obtain a cable air cooling temperature control strategy.

[0012] The present invention also provides a monitoring system for a charging pile cable, comprising:

[0013] The acquisition unit is used to collect and preprocess the temperature data, environmental parameters and airflow information of the charging pile transmission cable to obtain a multi-source time series data set;

[0014] A prediction unit, used for inputting the multi-source time series data set into a hybrid model of a deep neural network and a recurrent neural network to perform cable temperature prediction, thereby obtaining cable temperature prediction data;

[0015] An analysis unit, used for performing Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result;

[0016] An interpolation unit, used to perform temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field;

[0017] An optimization unit, used for performing computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management solution;

[0018] The decision unit is used to perform reinforcement learning and hierarchical decision processing on the cable airflow management scheme to obtain a cable air cooling temperature control strategy.

[0019] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0020] The present invention also provides a charging pile cable, which is used to implement the steps of any of the methods described above.

[0021] In summary, the technical solution provided by the present invention realizes the fusion of multi-source time series data by collecting and preprocessing the temperature data, environmental parameters and airflow information of the charging pile transmission cable, and adopts a hybrid model of deep neural network and recurrent neural network to predict the cable temperature, making full use of the spatial feature extraction ability of convolutional neural network and the temporal feature capture ability of long short-term memory network, thereby improving the accuracy and robustness of temperature prediction. Through Kalman filter fusion analysis, the optimized estimation of cable temperature distribution is realized, the influence of measurement noise and system error is effectively reduced, and the reliability of temperature monitoring is improved. By adopting spatiotemporal variation feature analysis and three-dimensional space interpolation technology, the refined reconstruction of the target cable temperature field is realized, and the cable airflow management scheme is obtained through computational fluid dynamics analysis and airflow channel optimization, which realizes the balanced optimization of cooling efficiency and energy consumption. By adopting reinforcement learning and hierarchical decision processing technology, the adaptive optimization of cable air cooling temperature control strategy is realized, and the control parameters can be dynamically adjusted according to real-time status and environmental changes, thereby improving the accuracy and flexibility of temperature control. The implementation of the present invention realizes the intelligent air cooling control of charging pile transmission cables and improves the charging process safety of charging pile transmission cables. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 1 is a schematic diagram of the steps of a monitoring method for a charging pile cable in one embodiment of the present invention;

[0023] Figure 2 It is a structural block diagram of a monitoring system for a charging pile cable in one embodiment of the present invention;

[0024] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0025] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0027] Reference Figure 1 , this embodiment provides a method for monitoring a charging pile cable, comprising the following steps:

[0028] S1, collect and preprocess the temperature data, environmental parameters and airflow information of the charging pile transmission cable to obtain a multi-source time series data set;

[0029] Among them, the surface infrared thermal imaging scanning of the charging pile transmission cable is carried out to obtain the original temperature image sequence, and the original temperature image sequence is corrected for non-uniformity and geometry to eliminate the temperature deviation and image distortion caused by equipment or environmental factors, and the corrected temperature image data is obtained. The adaptive threshold segmentation technology is used for the corrected temperature image data to accurately distinguish the high temperature area and other parts of the cable, and combined with morphological processing to optimize the edge and structural features of the image to obtain the cable temperature hot spot area. The local weighted average filtering technology is applied to the temperature hot spot area to effectively remove noise and subtle temperature fluctuations and generate noise-reduced temperature data. At the same time, multi-point sampling measurement is carried out on the environment around the charging pile transmission cable to obtain the original environmental data, including multiple environmental parameters such as temperature, humidity, and air pressure. By analyzing the original environmental data, the spatiotemporal distribution data of the environmental parameters are obtained, reflecting the changing laws of environmental factors at different time and space positions. Ultrasonic flow measurement is carried out on the air-cooling pipe inside the charging pile transmission cable to obtain the original data of air flow velocity and flow in the pipe. Based on these original data, the air flow distribution model in the air-cooling pipe is constructed through simulation analysis to describe the velocity and direction characteristics of the air flow in the pipe. The de-noised temperature data, the spatiotemporal distribution data of environmental parameters, and the airflow distribution data in the air-cooling duct are time-aligned and multivariate co-registered. Through time alignment, data from different data sources are matched on the same time axis, and the multivariate co-registration criterion ensures the spatial and temporal consistency between different physical quantities, integrating them into a unified multi-source time series data set.

[0030] S2, inputting the multi-source time series data set into the hybrid model of deep neural network and recurrent neural network to predict the cable temperature and obtain the cable temperature prediction data;

[0031] Among them, the multi-source time series data set is divided into time windows to ensure the temporal continuity and correlation of the data. By dividing the time windows, representative features are extracted and combined into an input feature matrix. The input feature matrix is ​​input into a hybrid model of a deep neural network and a recurrent neural network, and passes through three two-dimensional convolutional layers in the convolutional neural network layer. These convolutional layers capture the local receptive field of the input data through sliding windows and extract spatial features. Then, it passes through two maximum pooling layers to compress the size of the feature map to retain the main spatial information while reducing the amount of calculation to obtain a more compact spatial feature map. The spatial feature map is flattened to convert the high-dimensional feature map into a one-dimensional vector, which is then input into the fully connected layer to achieve further feature fusion. The spatial features are fused through the linear combination and nonlinear activation function of the fully connected layer to generate a fused feature vector. The fused feature vector is reshaped into a time series format and input into the long short-term memory network (LSTM) layer. Through the three-layer stacked LSTM structure, the time series features are gradually extracted. These time series features capture the dependencies in the time dimension and obtain a time series feature sequence. The self-attention mechanism is applied to the time series feature sequence to calculate the weight relationship between different time steps to identify which time steps have higher importance in the prediction process. The attention mechanism obtains a weighted time series feature sequence by assigning different weights to different time steps, and inputs it into the gated recurrent unit (GRU) network. Through the 2-layer bidirectional GRU structure, the context information of the time series is captured, and context-enhanced features are generated, which contain information from the past and the future, so that the model can understand the time series data more comprehensively. Multi-head self-attention processing is performed on the context-enhanced features. Through 8 attention heads, the correlation between different feature dimensions is calculated to generate multi-scale feature representations. This enables the model to understand the relationship between features from different perspectives and capture more diverse information. The multi-scale feature representation is input into the feedforward neural network, and multi-level fusion features are generated through nonlinear transformation. The fusion features are input into the output layer for final mapping to generate cable temperature prediction data.

[0032] S3, performing Kalman filter fusion analysis on the cable temperature prediction data to obtain the cable temperature distribution estimation result;

[0033] Among them, the cable temperature prediction data is decomposed into time series to extract multi-scale temperature features, so that the time characteristics of the data can be effectively captured at different time scales. Based on the multi-scale temperature features, a cable temperature dynamic model is constructed, which can describe the dynamic process of temperature change. The constructed cable temperature dynamic model is discretized to convert the continuous temperature dynamic model into a discrete state equation, so that the model can be processed in a digital computing environment. Through discretization, a discrete state equation is obtained to describe the temperature change law at discrete time points. Combined with actual measurement data, the discrete state equation is constructed into a linear random model, which can not only describe the change of system state, but also capture the random factors therein, making the model closer to the actual situation. The linear random model is linearized to simplify the calculation process of the model so that it can be implemented through matrix operations. Through linearization, a linearized state transfer matrix and an observation matrix are obtained. These two matrices are used to describe the evolution of the system state over time and the relationship between the system state and the observed data. Based on these two matrices, the prior estimate and the prior error covariance are calculated. The prior estimate is a prediction of the system state at the current moment, while the prior error covariance quantifies the accuracy of this prediction. The prior state estimation and the actual measurement data are subjected to residual analysis, and the difference between the two is calculated to obtain the measurement update gain matrix. The measurement update gain matrix is ​​used to correct the prior state estimation to make it more consistent with the actual situation. By correcting the prior state estimation, the optimized state estimation result is obtained. The optimized state estimation result is adaptively adjusted to obtain the cable temperature distribution estimation result. The adjustment process dynamically adjusts the model parameters according to the actual situation, making the final cable temperature distribution estimation result more accurate and robust.

[0034] S4, performing temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain the target cable temperature field;

[0035] Among them, the cable temperature distribution estimation results are decomposed to extract multi-scale time features and capture the dynamic characteristics of temperature changes at different time scales. Based on the multi-scale time features, the time autocorrelation function and partial autocorrelation function are calculated to construct a time dynamic feature model to effectively describe the evolution law of temperature distribution in the time dimension. The cable temperature distribution estimation results are subjected to spatial autocorrelation analysis to identify the correlation of temperature distribution in the spatial dimension. The spatial hot spot distribution map is obtained through analysis to determine the areas where temperature changes are concentrated or abnormal. Based on the spatial hot spot distribution map, a spatial weight matrix is ​​constructed to quantify the correlation between different spatial positions and generate a spatial variation feature model. This model can describe the variation characteristics of temperature distribution in the spatial dimension and reflect the temperature difference and spatial correlation of cables at different positions. The time dynamic feature model and the spatial variation feature model are tensor decomposed to reduce the high-dimensional spatiotemporal features to obtain a low-dimensional spatiotemporal feature representation. Based on the low-dimensional spatiotemporal feature representation, a nonlinear spatiotemporal feature embedding is generated. By mapping the spatiotemporal features to a new feature space, the nonlinear relationship hidden in the original data is revealed. A target spatiotemporal variation pattern is obtained by nonlinearly mapping the nonlinear spatiotemporal feature embedding. Based on the target spatiotemporal variation pattern, a smooth three-dimensional temperature field surface is constructed. The surface is refined by adaptive meshing to ensure the smoothness and continuity of the temperature field surface in space. The three-dimensional thin plate spline interpolation technology is applied to interpolate these grids to finally generate a continuous and smooth three-dimensional temperature field.

[0036] S5, performing computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management solution;

[0037] That is, the temperature field of the target cable is reconstructed in three dimensions. By converting the temperature field into a geometric model, the geometric representation of the computational domain is obtained, and the spatial boundary required for computational fluid dynamics analysis is clarified. Based on the geometric representation, a computational grid is generated, and the computational domain is divided into several small units so that subsequent numerical calculations can be performed on these discrete units. The generated computational grid is discretized and solved, and the fluid flow and heat transfer process are simulated by numerical methods to obtain the initial flow field distribution, showing the velocity and pressure distribution of the fluid in space. Based on the initial flow field distribution, the vortex field is calculated, and the target vortex area is identified. These vortex areas are often key areas of flow instability and heat accumulation. The target vortex area is orthogonally decomposed to obtain the reduced-dimensional flow field characteristics. And based on the reduced-dimensional flow field characteristics, an airflow performance predictor is constructed to predict the cooling effect and energy consumption performance under different airflow configurations. The airflow performance predictor is optimized for cooling efficiency and energy consumption, and a set of candidate airflow channel configurations are obtained by simulating different airflow channel configurations. These configurations represent the best balance between cooling effect and energy consumption. Based on the candidate airflow channel configuration, an accompanying sensitivity analysis is performed to identify the key design parameters that affect cooling efficiency and energy consumption. By analyzing these parameters, the local optimization direction is determined. Based on the determined local optimization direction, the iterative solution method of sequential quadratic programming is used to gradually optimize the airflow channel design parameters until the optimal design solution is obtained. The optimized design parameters represent the optimal balance between cooling effect and energy efficiency under the current conditions. The optimized airflow channel design parameters are verified through computational fluid dynamics verification simulation to ensure that they can achieve the expected effect in actual applications, and finally the cable airflow management solution is obtained.

[0038] S6, reinforcement learning and hierarchical decision processing are performed on the cable airflow management scheme to obtain the cable air cooling temperature control strategy.

[0039] Among them, the cable airflow management scheme is discretized to transform the continuous airflow management problem into a discrete state space representation. The state space representation defines all possible states in the reinforcement learning environment. Based on the state space representation, evaluation criteria for reinforcement learning are created. These criteria include multiple indicators such as cooling effect, energy efficiency and system stability, which are used to evaluate the pros and cons of different strategies. Based on the created evaluation criteria, a double Q network calculation is performed to obtain a preliminary Q value estimation function, which provides a value evaluation for each state-action pair. Using the preliminary Q value estimation function, a learning strategy is constructed, and the strategy is subjected to trust region constraints and entropy regularization. The trust region constraint ensures that the policy update does not deviate too far from the original strategy, while the entropy regularization promotes the diversity and exploration of the strategy, resulting in a more robust policy gradient update scheme. Based on the policy gradient update scheme, a deep reinforcement learning model is generated, which learns the optimized strategy by continuously interacting with the environment. In order to cope with complex airflow management problems, the generated deep reinforcement learning model is hierarchically abstracted to form a hierarchical option framework. The framework decomposes the entire decision-making process into multiple levels, each level corresponds to a different decision granularity, so that the model can handle multi-level decision problems more effectively. Based on the hierarchical option framework, an adaptive task decomposition strategy is generated. This strategy decomposes complex tasks into several subtasks, so that the model can gradually optimize the execution of each subtask and improve the effectiveness and efficiency of the overall strategy. Meta-learning is performed according to the adaptive task decomposition strategy, that is, through the process of learning how to learn, it gradually optimizes and adapts to different task requirements to obtain the optimal target control strategy. Based on the target control strategy, a decision tree integration model is constructed. Through this integration model, comprehensive decisions are made based on multiple decision trees, and the weights of each decision tree are dynamically adjusted to improve the flexibility and accuracy of the decision. Through dynamic weight adjustment, a cable air cooling temperature control strategy is generated. This control strategy can adjust the airflow management plan in real time in actual operation to achieve the best cooling effect and energy consumption balance, thereby ensuring the safe and efficient operation of transmission cables under different working conditions.

[0040] In one example, the temperature data, environmental parameters and airflow information of the charging pile transmission cable are collected and preprocessed to obtain a multi-source time series data set, including:

[0041] Perform infrared thermal imaging scans on the surface of the charging pile transmission cable to obtain an original temperature image sequence, and perform non-uniformity correction and geometric correction on the original temperature image sequence to obtain corrected temperature image data;

[0042] Adaptive threshold segmentation and morphological processing are performed on the corrected temperature image data to obtain the cable temperature hotspot area, and local weighted average filtering is performed on the cable temperature hotspot area to obtain the temperature data after noise reduction;

[0043] Perform multi-point sampling measurement on the charging pile transmission cable to obtain the original environmental data, and analyze the original environmental data to obtain the spatiotemporal distribution data of environmental parameters;

[0044] Ultrasonic flow measurement is performed on the air-cooling pipe in the charging pile transmission cable to obtain the original data of air flow velocity and flow in the pipe, and simulation is performed based on the original data of air flow velocity and flow in the pipe to obtain the air flow distribution model in the air-cooling pipe;

[0045] The denoised temperature data, the spatiotemporal distribution data of environmental parameters, and the airflow distribution data in the air-cooling duct are time aligned and multivariate co-registered to obtain a multi-source time series dataset.

[0046] In this example, the surface of the power transmission cable is scanned by infrared thermal imaging, and the temperature distribution of the cable surface is obtained by using high-precision infrared camera equipment. Infrared thermal imaging technology can capture the temperature changes on the cable surface and output them in the form of image sequences, which are the original temperature image sequences. The original temperature image sequence is corrected for non-uniformity to eliminate the temperature reading error caused by the non-uniformity of the sensor response and obtain more accurate temperature data. At the same time, geometric correction is performed to correct the image distortion caused by the camera angle, position change, etc., to ensure the accurate spatial correspondence of the temperature data and obtain the corrected temperature image data. The corrected temperature image data is adaptively segmented and morphologically processed to extract the key temperature information of the cable. The temperature information in the image is classified according to different temperature intervals by using adaptive threshold segmentation technology. The adaptive threshold segmentation can automatically adjust the threshold according to the actual temperature distribution in the image to ensure that the temperature abnormal area can be accurately identified. Morphological processing is performed to optimize the segmentation results. Through morphological operations such as expansion and corrosion, noise and isolated points in the image are removed, making the cable temperature hot spot area clearer and more coherent. The identified hotspot areas are subjected to local weighted average filtering. The filtering process effectively reduces the temperature fluctuations caused by random noise by weighted averaging the temperature values ​​in the hotspot areas, and obtains the temperature data after noise reduction. At the same time, multi-point sampling and measurement of environmental parameters around the charging pile are performed. By arranging environmental sensors at multiple key locations, the original data of the environment is obtained, which usually includes parameters such as ambient temperature, humidity, wind speed, and air pressure. The original environmental data is analyzed to generate the spatiotemporal distribution data of environmental parameters, reflecting the changes of these parameters in different time and space. For example, changes in wind speed may cause fluctuations in the surface temperature of the cable, while changes in humidity may affect the heat dissipation efficiency. Air-cooling ducts are usually set inside the transmission cable, and air cooling is carried out through these ducts. Ultrasonic flow measurement is performed on the air flow velocity and flow rate in the air-cooling duct to obtain the fluid movement in these ducts. The ultrasonic flowmeter can accurately measure the flow velocity and flow rate of the fluid and output the original data of the air flow velocity and flow rate in the duct. Based on these original data, fluid dynamics simulation is performed to establish an air flow distribution model in the air-cooling duct to describe the flow of air in the duct and identify possible areas of uneven air flow.

[0047] In one example, a multi-source time series dataset is input into a hybrid model of a deep neural network and a recurrent neural network for cable temperature prediction, and cable temperature prediction data is obtained, including:

[0048] Perform time window division and feature extraction on multi-source time series data sets to obtain input feature matrix;

[0049] The input feature matrix is ​​input into the convolutional neural network layer in the hybrid model of the deep neural network and the recurrent neural network, and spatial features are extracted through three two-dimensional convolutional layers and two maximum pooling layers to obtain a spatial feature map;

[0050] The spatial feature map is flattened and input into the fully connected layer for feature fusion to obtain a fused feature vector. The fused feature vector is reshaped into a time series format and input into the long short-term memory network layer. The time series features are extracted through a 3-layer stacked LSTM structure to obtain a time series feature sequence.

[0051] The time series feature sequence is weighted by the self-attention mechanism, and the weight relationship between different time steps is calculated to obtain the attention-weighted time series features. The attention-weighted time series features are input into the gated recurrent unit network, and the context information is captured through a 2-layer bidirectional GRU structure to obtain context-enhanced features.

[0052] Multi-head self-attention processing is performed on the context-enhanced features. The correlation between different feature dimensions is calculated through 8 attention heads to obtain multi-scale feature representation;

[0053] The multi-scale feature representation is input into the feedforward neural network for nonlinear transformation to obtain multi-level fusion features, and the multi-level fusion features are input into the output layer for mapping to obtain the cable temperature prediction data.

[0054] In this example, the multi-source time series dataset is divided into time windows to fully capture the time dynamic characteristics of the system. For example, if the dataset contains readings at minute intervals covering a time range of several hours, a 60-minute window is selected to capture the behavioral characteristics within each hour. In each time window, key features such as average temperature, temperature variance, airflow rate, etc. are extracted to generate the input feature matrix , where each row represents a specific time window and each column corresponds to a feature extracted from that window. The input feature matrix is ​​fed into a hybrid model that combines the strengths of deep neural networks (DNNs) and recurrent neural networks (RNNs). The first component of the model is a convolutional neural network (CNN) layer that excels at capturing spatial relationships in the data. The CNN architecture consists of three 2D convolutional layers, each followed by an activation function and with two max pooling layers in between to reduce dimensionality and highlight key features. For example, consider the operation of the first convolutional layer. For a given input feature map , the convolution operation applies a set of filters in the spatial dimension . Mathematically, the position The output feature map at It can be expressed as:

[0055] ;

[0056] in, represents the activation function (such as ReLU), It's location The filter weights at is the input value at the corresponding position, is the bias term. This operation emphasizes local spatial patterns in the data. In the convolutional layer, the data is subjected to two maximum pooling operations. Maximum pooling downsamples the feature map by selecting the maximum value in the local area, reducing its dimensionality while retaining the most significant features. After this series of convolutional and pooling layers, the data is converted into a spatial feature map that contains complex spatial interdependencies. The spatial feature map is flattened to convert the multi-dimensional tensor into a one-dimensional vector. The flattened representation is input to a fully connected (dense) layer, which facilitates the fusion of extracted spatial features. The fully connected layer captures the complex combinatorial relationships of the features through linear transformations and nonlinear activations. After feature fusion, the resulting vector is reshaped into a time series format. The reshaped data is input into a layer consisting of three stacked long short-term memory (LSTM) networks. LSTM is a specialized recurrent neural network designed to capture long-term dependencies in sequence data by alleviating problems such as gradient vanishing. Each LSTM unit maintains an internal state that allows it to remember information over long sequences. For example, at time step In the LSTM unit, the cell state and hidden state Based on input , the previous hidden state and the previous cell state Update. Its operations can be summarized as follows:

[0057]

[0058] in, , and They represent the forget gate, input gate, and output gate, respectively, regulating the flow of information; The term represents the weight matrix; The term is bias; and tanh are sigmoid and hyperbolic tangent activation functions respectively. As the data passes through the stacked LSTM layers, the model extracts temporal features that capture the dynamics and dependencies between different time steps. The output of the LSTM layer is a sequence of temporal features that contains these complex temporal patterns. The temporal feature sequence is weighted by a self-attention mechanism. This mechanism calculates the correlation between time steps and assigns higher weights to more influential time steps. The model is able to focus on key moments that have a significant impact on the prediction. Time step The attention weight It can be calculated as:

[0059] ;

[0060] in, is the time step The attention score is usually calculated as the similarity between the hidden state and the context vector. The attention-weighted temporal features are input into a layer consisting of two layers of bidirectional gated recurrent unit (GRU) networks. The bidirectional nature ensures that the model considers both past and future contexts, while the GRU unit is more efficient than LSTM in capturing dependencies through its gating mechanism. In the GRU layer model, a multi-head self-attention mechanism with eight attention heads is adopted. Each head allows the model to focus on a different representation subspace, capturing different aspects of the data. The mechanism calculates the association between different feature dimensions, enabling the model to understand multi-scale feature representations. The output aggregated from the attention heads is input into a feedforward neural network. The network performs nonlinear transformations, combining multi-level and multi-scale features to produce a fused feature representation. The integrated feature representation is input into the output layer, which maps it to the target output space to generate prediction data for cable temperature.

[0061] In one example, Kalman filter fusion analysis is performed on the cable temperature prediction data to obtain the cable temperature distribution estimation result, including:

[0062] The cable temperature prediction data is decomposed into time series to obtain multi-scale temperature features, and a cable temperature dynamic model is constructed based on the multi-scale temperature features;

[0063] Discretize the cable temperature dynamic model to obtain the discrete state equation, and build a linear random model based on the actual measurement data and the discrete state equation;

[0064] Linearize the linear random model to obtain a linearized state transfer matrix and an observation matrix, and calculate a priori estimates and a priori error covariance based on the linearized state transfer matrix and the observation matrix to obtain a priori state estimates;

[0065] Perform residual analysis on the prior state estimation and actual measurement data to obtain a measurement update gain matrix, and correct the prior state estimation according to the measurement update gain matrix to obtain an optimized state estimation result;

[0066] The optimized state estimation result is adaptively adjusted to obtain the cable temperature distribution estimation result.

[0067] In this example, the cable temperature prediction data is decomposed into a time series, and multi-scale temperature features are extracted to capture the dynamic behavior of temperature changes at different time scales. Time series decomposition usually uses methods such as wavelet transform or empirical mode decomposition to decompose complex temperature time series into several components of different frequencies. Each component corresponds to a temperature feature at a different time scale, which can reveal short-term fluctuations and long-term trends. For example, the temperature series is decomposed into multiple detail components (representing high-frequency information) and an approximate component (representing low-frequency information) through wavelet transform. A cable temperature dynamic model is constructed based on multi-scale temperature features. The model usually takes the form of a state space model, in which the state of the system represents the change of temperature over time. The dynamic model is discretized to convert the continuous time model into a discrete time model, so that calculations can be performed at discrete time points. The discretization process is usually achieved through differential equation approximation. Assume that the state equation under continuous time is ,in Indicates the system status. is the system matrix, is the input matrix, As the input signal, through discretization, the discrete time state equation is obtained: ;

[0068] in, is the time step The system status, and are the discretized system matrix and input matrix respectively. Based on the actual measurement data and the discrete state equation, a linear random model is constructed. The linear random model not only describes the evolution of the system state, but also takes into account the randomness or uncertainty in the system. Usually, a Gaussian white noise term is introduced To describe this randomness, the equation form is:

[0069] ;

[0070] in, is a matrix with zero mean and covariance Gaussian noise. This model can more realistically reflect the random changes in temperature dynamics. Linearize the linear random model and simplify the complex nonlinear system into a linear system to facilitate solving through matrix operations. After linearization, the linearized state transfer matrix is ​​obtained and the observation matrix . Among them, the state transfer matrix describes the change of the system state over time, and the observation matrix describes the relationship between the system state and the actual measurement data. Based on the linearized state transfer matrix and observation matrix, the prior state estimate and prior error covariance are calculated. The prior state estimate is a prediction of the system state at the current moment, which is calculated based on the state estimate and state transfer matrix at the previous moment. The prior error covariance quantifies the accuracy of this prediction, which is usually expressed as:

[0071] ;

[0072] in, is the time step The prior error covariance matrix of is the state estimation error covariance matrix at the previous moment, is the process noise covariance matrix. Perform residual analysis on the prior state estimate and the actual measurement data, and calculate the difference between the two to obtain the measurement update gain matrix (Kalman gain matrix) , this matrix is ​​used to correct the prior state estimate to make it closer to the actual situation. The corrected estimate is the posterior state estimate. The calculation formula of the measurement update gain matrix is:

[0073] ;

[0074] in, is the measurement noise covariance matrix. The prior state estimate is corrected by the gain matrix to obtain the optimized state estimation result:

[0075] ;

[0076] in, is the time step The measurement data, is the predicted observation value of the prior state. The optimized state estimation result is adjusted by adaptive factors. By introducing adaptive factors, the model parameters are dynamically adjusted to adapt to the changes of system state over time. Through adaptive adjustment, a more accurate cable temperature distribution estimation result is finally obtained.

[0077] In one example, the cable temperature distribution estimation result is subjected to temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation to obtain the target cable temperature field, including:

[0078] Decompose the cable temperature distribution estimation result to obtain multi-scale time characteristics, and calculate the time autocorrelation function and partial autocorrelation function based on the multi-scale time characteristics to obtain the time dynamic characteristic model;

[0079] Perform spatial autocorrelation analysis on the cable temperature distribution estimation results to obtain a spatial hotspot distribution map, and construct a spatial weight matrix based on the spatial hotspot distribution map to obtain a spatial variation feature model;

[0080] Perform tensor decomposition on the temporal dynamic feature model and the spatial variation feature model to obtain a low-dimensional spatiotemporal feature representation;

[0081] Generate nonlinear spatiotemporal feature embedding based on low-dimensional spatiotemporal feature representation, and perform nonlinear mapping on the nonlinear spatiotemporal feature embedding to obtain the target spatiotemporal change pattern;

[0082] A smooth three-dimensional temperature field surface is constructed according to the target spatiotemporal variation pattern, and the smooth three-dimensional temperature field surface is adaptively meshed and three-dimensional thin plate spline interpolated to obtain the target cable temperature field.

[0083] In this example, the cable temperature distribution estimation results are decomposed to extract multi-scale time features. The decomposition process usually uses methods such as wavelet transform or empirical mode decomposition to decompose complex time series into multiple components, each corresponding to a different time scale. Wavelet transform captures features in different frequency ranges by gradually refining the signal, while empirical mode decomposition reveals its intrinsic structure by decomposing the signal into a sequence of intrinsic mode functions (IMFs). Multi-scale time features reflect the variation of temperature at different time scales, from short-term fluctuations to long-term trends. The temporal autocorrelation function and partial autocorrelation function are calculated based on the multi-scale time features to reveal the correlation between different time points in the temperature series. Autocorrelation function It measures the time series at different lags The similarity degree under is defined as:

[0084] ;

[0085] in, It's time The temperature value, is the mean of the series, is the length of the time series. The partial autocorrelation function reveals the direct time dependence by controlling the influence of the intermediate lag. These calculation results are used to construct a time dynamic characteristic model, which describes the evolution of temperature over time, especially the periodicity and trend of temperature change. Spatial autocorrelation analysis is performed on the cable temperature distribution estimation results to identify the regularity and structure of temperature distribution in space. By calculating the spatial autocorrelation coefficient (such as Moran's index), the similarity of temperature between adjacent spatial regions is evaluated. Spatial autocorrelation coefficient Defined as:

[0086] ;

[0087] in, and The spatial position and The temperature value, It's location and The spatial weight between is the total number of spatial locations. Based on the calculation results, a spatial hot spot distribution map is drawn to identify areas with abnormally high temperatures. This information is used to construct a spatial weight matrix, which quantifies the correlation between different spatial locations, forms a spatial variation feature model, and describes the temperature variation characteristics in space. The temporal dynamic feature model and the spatial variation feature model are tensor decomposed. Tensor decomposition is a high-dimensional data dimensionality reduction technology that can decompose complex spatiotemporal data into several low-dimensional factors to obtain a low-dimensional spatiotemporal feature representation, which contains the core information of temperature distribution in time and space. Based on the low-dimensional spatiotemporal feature representation, a nonlinear spatiotemporal feature embedding is generated. This embedding transforms the low-dimensional feature representation into a new feature space through nonlinear mapping to reveal the potential nonlinear structure in the data. Nonlinear mapping is usually implemented through kernel functions or neural networks, which can capture complex spatiotemporal dependencies. Through this process, the target spatiotemporal variation pattern is obtained, revealing the main variation law of temperature distribution in time and space. A smooth three-dimensional temperature field surface is constructed according to the target spatiotemporal variation pattern to describe the temperature distribution in the entire space. Adaptive meshing technology is used to mesh the three-dimensional space, refine the areas with drastic temperature changes, and simplify the areas with gentle temperature changes. Adaptive meshing can effectively reduce the amount of calculation while maintaining high accuracy. The target cable temperature field is generated using three-dimensional thin plate spline interpolation. Thin plate spline interpolation is a generalized polynomial interpolation method that can generate smooth surfaces on irregular grids. The interpolation process minimizes the surface energy function To achieve this, it is defined as:

[0088] ;

[0089] in, is the surface function to be interpolated, and is the spatial coordinate. By minimizing this function, the resulting surface is not only smooth but also well adapted to the distribution of data points.

[0090] In one example, computational fluid dynamics analysis and airflow channel optimization are performed on the target cable temperature field to obtain a cable airflow management solution, including:

[0091] Perform three-dimensional geometric reconstruction on the target cable temperature field to obtain a computational domain geometric representation, and generate a computational grid based on the computational domain geometric representation;

[0092] Discretize and solve the computational grid to obtain the initial flow field distribution, and calculate the vortex field based on the initial flow field distribution to obtain the target vortex area;

[0093] Orthogonal decomposition is performed on the target vortex region to obtain the reduced-dimensional flow field characteristics, and an airflow performance predictor is constructed based on the reduced-dimensional flow field characteristics;

[0094] Optimize the cooling efficiency and energy consumption of the airflow performance predictor to obtain candidate airflow channel configurations, and perform accompanying sensitivity analysis based on the candidate airflow channel configurations to obtain local optimization directions;

[0095] The local optimization direction is iteratively solved by sequential quadratic programming to obtain the target airflow channel design parameters, and a computational fluid dynamics verification simulation is performed based on the target airflow channel design parameters to obtain the cable airflow management solution.

[0096] In this example, the target cable temperature field is reconstructed in 3D geometry, and a geometric model of the cable and its surroundings, i.e., the computational domain geometry representation, is constructed from the existing temperature distribution data. The physical form, surface characteristics, and surrounding air flow area of ​​the cable are accurately reproduced by using computer-aided design tools or other 3D modeling software. A computational mesh is generated based on the computational domain geometry representation, and the continuous geometric domain is discretized into a finite number of small units, such as triangles or tetrahedrons. These small units define the computational nodes in the computational fluid dynamics (CFD) simulation and determine the accuracy of the numerical solution. In order to maintain computational efficiency while ensuring the accuracy of the results, smaller mesh units are usually used in complex or critical areas (such as the cable surface or high-temperature areas), while larger mesh units are used in areas with more gradual changes. The computational mesh is discretized and solved to solve fluid dynamics equations, such as the Navier-Stokes equations, which describe the motion state and temperature distribution of the fluid. These equations are approximated on the discrete mesh using numerical methods, such as the finite volume method or the finite element method, to obtain the initial flow field distribution. The initial flow field distribution includes velocity field, pressure field and temperature field, which reflects the motion state of the fluid in the computational domain. The vorticity field is calculated based on the initial flow field distribution to identify the target vortex area. The vorticity field is a key concept in fluid dynamics, which describes the intensity and direction of fluid rotation. The calculation formula is:

[0097] ;

[0098] in, is the vorticity vector, is the velocity vector, Represents the curl operator. By calculating the vorticity field, the vortex structures in the flow field are identified. These structures are often the key areas of heat accumulation and flow instability. After determining the target vortex area, orthogonal decomposition is performed to simplify the complexity of the flow field. Orthogonal decomposition is a dimensionality reduction technique that extracts the most important flow patterns by decomposing the flow field data. These patterns can explain most of the energy and dynamic characteristics of the flow field. The complex flow field is simplified into a few main features through orthogonal decomposition. Each flow pattern corresponds to a reduced-dimensional flow field feature. Based on the reduced-dimensional flow field features, an airflow performance predictor is constructed. The airflow performance predictor is a mathematical model that can predict performance indicators such as cooling efficiency and energy consumption based on the flow field characteristics. The predictor is usually constructed using regression analysis or machine learning technology, and a model that can accurately predict the flow field behavior is obtained through training. The airflow performance predictor is optimized for cooling efficiency and energy consumption. While ensuring the cooling effect, energy consumption is minimized to obtain candidate airflow channel configurations, representing possible airflow management solutions. Each configuration is generated by changing design parameters such as the shape, size, or position of the airflow channel, and each configuration affects the trade-off between cooling efficiency and energy consumption. Candidate airflow channel configurations are evaluated using adjoint sensitivity analysis. Adjoint sensitivity analysis is an optimization tool that calculates the sensitivity of design parameter changes to an objective function, such as cooling efficiency or energy consumption. Through analysis, local optimization directions are determined, i.e., which design parameter adjustments can significantly improve cooling performance or reduce energy consumption. Based on the local optimization directions, sequential quadratic programming is used for iterative solution. Sequential quadratic programming is a commonly used optimization algorithm that optimizes nonlinear problems by constructing quadratic approximations and gradually approaching the optimal solution. In each iteration, the algorithm updates the design parameters to approach the optimal airflow channel configuration. The resulting design parameters are the target airflow channel design parameters, which represent the design solution that achieves the best cooling efficiency and lowest energy consumption under the current conditions. Based on the target airflow channel design parameters, computational fluid dynamics (CFD) verification simulations are performed to verify the effect of the optimized design parameters in actual operation to ensure that they can achieve the expected performance indicators. By applying the design parameters to the CFD model and performing simulation analysis, the cooling effect and energy consumption performance under different working conditions are evaluated.

[0099] In one example, reinforcement learning and hierarchical decision processing are performed on the cable airflow management scheme to obtain the cable air cooling temperature control strategy, including:

[0100] Discretize the cable airflow management scheme to obtain the state space representation of the reinforcement learning environment, and create the reinforcement learning evaluation criteria based on the state space representation;

[0101] Perform double Q network calculation on the evaluation criteria to obtain a preliminary Q value estimation function;

[0102] Construct a learning strategy based on the preliminary Q-value estimation function, and perform trust region constraints and entropy regularization on the learning strategy to obtain a policy gradient update scheme;

[0103] Generate a deep reinforcement learning model according to the policy gradient update scheme, and perform hierarchical abstraction on the deep reinforcement learning model to obtain a hierarchical option framework;

[0104] Generate an adaptive task decomposition strategy based on a hierarchical option framework, and perform meta-learning on the adaptive task decomposition strategy to obtain a target control strategy;

[0105] A decision tree integration model is constructed according to the target control strategy, and the cable air cooling temperature control strategy is generated through dynamic weight adjustment through the decision tree integration model.

[0106] In this example, the cable airflow management solution is discretized to convert the continuous control variables and state space into discrete representations, so that reinforcement learning algorithms can be applied on a finite set of states and actions. The state space representation is the core of the entire reinforcement learning environment and defines the possible state combinations of the system at each moment. For example, the state space may include the current temperature distribution of the cable, airflow velocity, environmental conditions, etc. Each state is represented as a vector ,in Represents a time step. The discretization of the state space simplifies complex physical phenomena into a finite number of states, allowing the reinforcement learning algorithm to effectively explore and optimize in this space. Create evaluation criteria for reinforcement learning. These criteria are usually defined as a reward function ,in is the current state of the system, is in state The reward can be based on multiple factors, such as cooling efficiency, energy consumption, system stability, etc. The goal is to maximize the cumulative reward at each time step, that is, the long-term performance. The evaluation criteria are calculated using a double Q network. The double Q network is an improved method of Q learning that reduces the common over-estimation problem in Q learning by introducing two independent Q networks. At each time step , the formula for updating the Q value is:

[0107] ;

[0108] in, is the learning rate, is the discount factor, is the next state given by the second Q network Q-value estimation. In this way, a more robust Q-value estimation function is obtained. Based on the preliminary Q-value estimation function, the corresponding learning strategy is constructed. This strategy is used to guide the reinforcement learning agent to select the optimal action under different states. In order to prevent the policy update from being too aggressive, resulting in insufficient exploration or falling into the local optimum, the learning strategy is subjected to trust region constraints and entropy regularization. The trust region constraint ensures that the amplitude of each policy update does not exceed the preset range, thereby maintaining the stability of the strategy; entropy regularization encourages the randomness of the strategy and promotes further exploration of unexplored states. The policy gradient update scheme is based on the following formula:

[0109] ;

[0110] in, is the strategy parameter, is the expected reward of the strategy, is the policy gradient, which is adjusted by the trust region constraint and entropy regularization to update the policy parameters . Based on the policy gradient update scheme, a deep reinforcement learning model is generated. The model approximates complex Q-value functions and policy functions through deep neural networks. In order to cope with complex task structures, the deep reinforcement learning model is hierarchically abstracted to form a hierarchical option framework. In this framework, tasks are decomposed into multiple levels, each responsible for different time scales or task granularity. For example, the top level is responsible for long-term policy planning, such as airflow distribution optimization; the bottom level is responsible for short-term operations, such as specific fan speed adjustment. Complex decision problems are decomposed into smaller and more controllable sub-problems, each of which can be optimized by the deep reinforcement learning model. Based on the hierarchical option framework, an adaptive task decomposition strategy is generated. This strategy automatically decomposes complex tasks into multiple subtasks by identifying and exploiting the intrinsic structure of the task. Through adaptive task decomposition, the reinforcement learning agent can flexibly adjust its strategy to adapt to the changing environment and task requirements. In order to improve the generalization ability and adaptability of the model, the adaptive task decomposition strategy is meta-learned. Meta-learning enables the model to quickly adapt to new tasks by training on multiple related tasks. The goal is to find a meta-strategy that can quickly optimize policy parameters with a small number of updates when encountering new tasks. A decision tree ensemble model is constructed based on the target control strategy obtained by meta-learning. The decision tree ensemble model can provide more stable and accurate decisions by combining the prediction results of multiple decision trees. Each decision tree is trained on different subtasks or sub-regions of the state space, and the contribution of different decision trees is optimized through dynamic weight adjustment. The purpose of dynamic weight adjustment is to adjust the weight of each decision tree in real time according to the current task requirements and environmental status, so as to make the best decision in different task stages and environmental conditions.

[0111] Reference Figure 2 , this embodiment provides a monitoring system for a charging pile cable, including:

[0112] The acquisition unit 1 is used to collect and preprocess the temperature data, environmental parameters and airflow information of the charging pile transmission cable to obtain a multi-source time series data set;

[0113] Prediction unit 2, used for inputting multi-source time series data sets into a hybrid model of deep neural network and recurrent neural network to predict cable temperature and obtain cable temperature prediction data;

[0114] Analysis unit 3, used for performing Kalman filter fusion analysis on the cable temperature prediction data to obtain the cable temperature distribution estimation result;

[0115] An interpolation unit 4 is used to perform temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field;

[0116] The optimization unit 5 is used to perform computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management solution;

[0117] The decision unit 6 is used to perform reinforcement learning and hierarchical decision processing on the cable airflow management scheme to obtain the cable air cooling temperature control strategy.

[0118] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0119] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input system, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0120] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0121] An embodiment of the present invention further provides a charging pile cable, and the charging pile cable is used to implement the above method.

[0122] In summary, a monitoring method for a charging pile cable is proposed in an embodiment of the present invention.

[0123] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0124] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, system, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, system, article or method including the element.

[0125] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for monitoring a charging pile cable, characterized in that: The following steps are involved: The temperature data, environmental parameters and airflow information of the charging pile transmission cable are collected and preprocessed to obtain a multi-source time series data set; Inputting the multi-source time series data set into a hybrid model of a deep neural network and a recurrent neural network to predict cable temperature, thereby obtaining cable temperature prediction data; Performing Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result; Performing temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field; The target cable temperature field is subjected to computational fluid dynamics analysis and airflow channel optimization to obtain a cable airflow management solution; specifically comprising: performing three-dimensional geometric reconstruction on the target cable temperature field to obtain a computational domain geometric representation, and generating a computational grid based on the computational domain geometric representation; discretizing and solving the computational grid to obtain an initial flow field distribution, and calculating the vortex field based on the initial flow field distribution to obtain a target vortex area; orthogonally decomposing the target vortex area to obtain reduced-dimensional flow field characteristics, and constructing an airflow performance predictor based on the reduced-dimensional flow field characteristics; optimizing the airflow performance predictor for cooling efficiency and energy consumption to obtain candidate airflow channel configurations, and performing accompanying sensitivity analysis based on the candidate airflow channel configurations to obtain local optimization directions; iteratively solving the local optimization directions using sequential quadratic programming to obtain target airflow channel design parameters, and performing computational fluid dynamics verification simulation based on the target airflow channel design parameters to obtain a cable airflow management solution; Reinforcement learning and hierarchical decision processing are performed on the cable airflow management scheme to obtain a cable air cooling temperature control strategy.

2. The method for monitoring the charging pile cable according to claim 1, characterized in that: The temperature data, environmental parameters and airflow information of the charging pile transmission cable are collected and preprocessed to obtain a multi-source time series data set, including: Perform infrared thermal imaging scanning on the surface of the charging pile transmission cable to obtain an original temperature image sequence, and perform non-uniformity correction and geometric correction on the original temperature image sequence to obtain corrected temperature image data; Performing adaptive threshold segmentation and morphological processing on the corrected temperature image data to obtain a cable temperature hotspot area, and performing local weighted average filtering on the cable temperature hotspot area to obtain temperature data after noise reduction; Performing multi-point sampling measurement on the charging pile transmission cable to obtain raw environmental data, and analyzing the raw environmental data to obtain spatiotemporal distribution data of environmental parameters; Perform ultrasonic flow measurement on the air cooling pipe in the charging pile transmission cable to obtain the original data of air flow velocity and flow in the pipe, and perform simulation based on the air flow velocity and flow original data in the pipe to obtain the air flow distribution data in the air cooling pipe; The de-noised temperature data, the spatiotemporal distribution data of the environmental parameters and the airflow distribution data in the air-cooling pipe are time-aligned and multivariate co-registered to obtain a multi-source time series data set.

3. The method for monitoring the charging pile cable according to claim 1, characterized in that: The step of inputting the multi-source time series data set into a hybrid model of a deep neural network and a recurrent neural network to perform cable temperature prediction to obtain cable temperature prediction data includes: Performing time window division and feature extraction on the multi-source time series data set to obtain an input feature matrix; Inputting the input feature matrix into the convolutional neural network layer in the hybrid model of the deep neural network and the recurrent neural network, performing spatial feature extraction through three two-dimensional convolutional layers and two maximum pooling layers to obtain a spatial feature map; The spatial feature map is flattened and input into a fully connected layer for feature fusion to obtain a fused feature vector, the fused feature vector is reshaped into a time series format, and input into a long short-term memory network layer, and time series features are extracted through a 3-layer stacked LSTM structure to obtain a time series feature sequence; The time series feature sequence is weighted by a self-attention mechanism, and the weight relationship between different time steps is calculated to obtain attention-weighted time series features, and the attention-weighted time series features are input into a gated recurrent unit network, and context information is captured through a 2-layer bidirectional GRU structure to obtain context-enhanced features; Performing multi-head self-attention processing on the context-enhanced features, calculating the correlation between different feature dimensions through 8 attention heads, and obtaining a multi-scale feature representation; The multi-scale feature representation is input into a feed-forward neural network for nonlinear transformation to obtain multi-level fusion features, and the multi-level fusion features are input into an output layer for mapping to obtain cable temperature prediction data.

4. The method for monitoring the charging pile cable according to claim 1, characterized in that: The performing Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result includes: Performing time series decomposition on the cable temperature prediction data to obtain multi-scale temperature features, and constructing a cable temperature dynamic model based on the multi-scale temperature features; Discretizing the cable temperature dynamic model to obtain a discrete state equation, and constructing a linear random model based on actual measurement data and the discrete state equation; Linearizing the linear stochastic model to obtain a linearized state transfer matrix and an observation matrix, and calculating a priori estimates and a priori error covariance based on the linearized state transfer matrix and the observation matrix to obtain a priori state estimates; Performing residual analysis on the priori state estimation and the actual measurement data to obtain a measurement update gain matrix, and correcting the priori state estimation according to the measurement update gain matrix to obtain an optimized state estimation result; The optimized state estimation result is adaptively adjusted to obtain a cable temperature distribution estimation result.

5. The method for monitoring the charging pile cable according to claim 1, characterized in that: The performing of spatiotemporal variation characteristic analysis and three-dimensional space interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field includes: Decomposing the cable temperature distribution estimation result to obtain multi-scale time characteristics, and calculating the time autocorrelation function and the partial autocorrelation function according to the multi-scale time characteristics to obtain a time dynamic characteristic model; Performing spatial autocorrelation analysis on the cable temperature distribution estimation result to obtain a spatial hot spot distribution map, and constructing a spatial weight matrix according to the spatial hot spot distribution map to obtain a spatial variation feature model; Performing tensor decomposition on the temporal dynamic feature model and the spatial variation feature model to obtain a low-dimensional spatiotemporal feature representation; Generating a nonlinear spatiotemporal feature embedding according to the low-dimensional spatiotemporal feature representation, and performing nonlinear mapping on the nonlinear spatiotemporal feature embedding to obtain a target spatiotemporal variation pattern; A smooth three-dimensional temperature field surface is constructed according to the target spatiotemporal variation pattern, and adaptive meshing and three-dimensional thin plate spline interpolation are performed on the smooth three-dimensional temperature field surface to obtain a target cable temperature field.

6. The method for monitoring the charging pile cable according to claim 1, characterized in that: The cable airflow management scheme is subjected to reinforcement learning and hierarchical decision processing to obtain a cable air cooling temperature control strategy, including: Discretizing the cable airflow management scheme to obtain a state space representation of a reinforcement learning environment, and creating an evaluation criterion for reinforcement learning based on the state space representation; Performing double Q network calculation on the evaluation criteria to obtain a preliminary Q value estimation function; Constructing a learning strategy according to the preliminary Q-value estimation function, and performing trust region constraints and entropy regularization on the learning strategy to obtain a policy gradient update scheme; Generate a deep reinforcement learning model according to the policy gradient update scheme, and perform hierarchical abstraction on the deep reinforcement learning model to obtain a hierarchical option framework; generating an adaptive task decomposition strategy according to the hierarchical option framework, and performing meta-learning on the adaptive task decomposition strategy to obtain a target control strategy; A decision tree integration model is constructed according to the target control strategy, and dynamic weight adjustment is performed through the decision tree integration model to generate a cable air cooling temperature control strategy.

7. A monitoring system for a charging pile cable, characterized in that: Used to perform the monitoring method of the charging pile cable according to any one of claims 1 to 6, the system comprising: The acquisition unit is used to collect and preprocess the temperature data, environmental parameters and airflow information of the charging pile transmission cable to obtain a multi-source time series data set; A prediction unit, used for inputting the multi-source time series data set into a hybrid model of a deep neural network and a recurrent neural network to perform cable temperature prediction, thereby obtaining cable temperature prediction data; An analysis unit, used for performing Kalman filter fusion analysis on the cable temperature prediction data to obtain a cable temperature distribution estimation result; An interpolation unit, used to perform temporal and spatial variation characteristic analysis and three-dimensional spatial interpolation on the cable temperature distribution estimation result to obtain a target cable temperature field; An optimization unit is used to perform computational fluid dynamics analysis and airflow channel optimization on the target cable temperature field to obtain a cable airflow management solution; specifically comprising: performing three-dimensional geometric reconstruction on the target cable temperature field to obtain a computational domain geometric representation, and generating a computational grid based on the computational domain geometric representation; discretizing and solving the computational grid to obtain an initial flow field distribution, and calculating the vortex field based on the initial flow field distribution to obtain a target vortex area; orthogonally decomposing the target vortex area to obtain reduced-dimensional flow field characteristics, and constructing an airflow performance predictor based on the reduced-dimensional flow field characteristics; optimizing the cooling efficiency and energy consumption of the airflow performance predictor to obtain a candidate airflow channel configuration, and performing accompanying sensitivity analysis based on the candidate airflow channel configuration to obtain a local optimization direction; iteratively solving the local optimization direction using sequential quadratic programming to obtain target airflow channel design parameters, and performing computational fluid dynamics verification simulation based on the target airflow channel design parameters to obtain a cable airflow management solution; The decision unit is used to perform reinforcement learning and hierarchical decision processing on the cable airflow management scheme to obtain a cable air cooling temperature control strategy.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A charging pile cable, characterized in that: The charging pile cable is used to implement the steps of any one of claims 1 to 6.

Images