Intelligent leakage monitoring and protection system for charging pile

Abstract

The invention relates to the technical field of charging pile protection, and discloses an intelligent leakage monitoring and protection system for a charging pile. The system comprises a multi-source leakage sensing module, a leakage feature fusion module, a leakage risk decision module and a self-adaptive protection execution module. The multi-source leakage sensing module synchronously captures gas concentration gradient distribution, sound wave abnormal frequency band energy, liquid seepage rate and other leakage related signals of the operating environment of the charging pile. The leakage feature fusion module performs cross-modal feature alignment on the multi-source signal to generate a fusion leakage feature tensor in space-time correlation; the leakage risk decision module generates a leakage type identifier, leakage source probability distribution and a protection level instruction based on the tensor; the adaptive protection execution module dynamically adjusts the leakage suppression strategy according to the protection level instruction. The system can improve the leakage monitoring accuracy and protection pertinence, and guarantees the safe operation of the charging pile.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of charging pile protection, in particular to a charging pile intelligent leakage monitoring and protection system. BACKGROUND

[0002] With the rapid development of new energy vehicle industry, as the core facility of energy supply, the installation scale and use frequency of charging piles continue to rise. During the operation of the charging pile, it needs to be exposed to the complex outdoor environment for a long time, facing high temperature, humidity, corrosion gas erosion and other multiple influences, while the internal electrical components aging, sealing structure damage and other problems will also gradually appear with the use time, which is easy to cause gas leakage, liquid leakage and other safety hazards. The current leakage monitoring scheme of charging pile on the market mostly uses single sensor for data collection, such as relying only on gas concentration sensor to detect flammable gas leakage, or only through humidity sensor to judge liquid leakage, which has obvious limitations. Single sensor cannot fully capture the multi-dimensional characteristics in the leakage process, and when there is trace leakage or complex environmental interference, it is easy to produce false alarm or miss report, which is difficult to realize the accurate identification and prediction of leakage risk.

[0003] In the aspect of protection execution, the existing charging pile mostly adopts single protection measure triggered by fixed threshold, such as directly cutting off the power supply once the gas concentration exceeds the standard, which lacks the dynamic adaptation ability to the leakage scene. Different types of leakage (such as electrolyte leakage, coolant leakage, flammable gas leakage) and different leakage degree have significant differences in the demand for protection strategy, and the fixed protection mode either overreacts in the early stage of leakage, affecting the normal charging service, or is insufficient in protection when the leakage intensifies, leading to the expansion of safety risk. In addition, there is lack of effective data interaction and collaborative linkage between the monitoring module and the protection module in the existing system, and the monitoring data cannot be timely and accurately converted into targeted protection instructions, so that the whole leakage prevention and control process presents passivity and lag, which is difficult to meet the high reliability demand of charging pile safe operation, and poses potential threat to equipment safety, user personal safety and surrounding environment. SUMMARY

[0004] The purpose of the present application is to provide a charging pile intelligent leakage monitoring and protection system to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides a charging pile intelligent leakage monitoring and protection system, which comprises:

[0006] A multi-source leakage perception module is used to synchronously capture the leakage associated signals of the charging pile operating environment, and the leakage associated signals include gas concentration gradient distribution, abnormal frequency band energy of sound wave and liquid seepage rate.

[0007] a leakage feature fusion module configured to perform cross-modal feature alignment on the leakage-related signals collected by the multi-source leakage perception module to generate a spatio-temporally correlated fusion leakage feature tensor;

[0008] a leakage risk decision module configured to generate a leakage type identification, a leakage source probability distribution, and a protection level instruction based on the fusion leakage feature tensor;

[0009] an adaptive protection execution module configured to dynamically adjust a leakage suppression strategy based on the protection level instruction.

[0010] Preferably, the multi-source leakage perception module specifically includes:

[0011] a distributed gas sensing unit arranged at the bottom of the charging pile and the cable interface area and configured to capture a gas concentration gradient distribution using a nano gas sensing material array;

[0012] a wideband acoustic wave acquisition unit embedded in the insulating shell of the charging pile and configured to extract acoustic wave abnormal frequency band energy through a piezoelectric sensor group;

[0013] a microfluidic detection unit integrated in the foundation structure layer of the charging pile and configured to quantify a liquid seepage rate based on a capillary pressure sensing technology.

[0014] Preferably, the leakage feature fusion module specifically includes:

[0015] a time series feature extraction submodule configured to perform sliding window segmentation on the gas concentration gradient distribution to extract a concentration mutation slope and a steady-state drift amount;

[0016] a frequency domain feature decomposition submodule configured to perform wavelet packet transform on the acoustic wave abnormal frequency band energy to separate high-frequency impact components and low-frequency resonance components;

[0017] a flow state feature modeling submodule configured to construct a seepage path topology graph based on the liquid seepage rate and calculate a node flow rate difference coefficient.

[0018] Preferably, the leakage feature fusion module further includes:

[0019] a cross-modal alignment submodule configured to map the concentration mutation slope, the high-frequency impact components, and the node flow rate difference coefficient to a unified timestamp coordinate system;

[0020] a feature tensor generation submodule configured to output a three-dimensional fusion leakage feature tensor by weighted splicing the steady-state drift amount, the low-frequency resonance components, and the seepage path topology graph through a multi-head attention mechanism.

[0021] Preferably, the leakage risk decision module specifically includes:

[0022] a leakage type classification unit, configured to process the fusion leakage feature tensor by using a time convolution network, and output a combustible gas leakage identifier, an electrolyte leakage identifier, or an insulating oil leakage identifier;

[0023] a leakage source positioning unit, configured to analyze spatial correlation features in the fusion leakage feature tensor based on a graph neural network, and generate a leakage source probability heat map;

[0024] a protection strategy decision unit, configured to match a preset protection level instruction according to the leakage type identifier and the leakage source probability heat map.

[0025] Preferably, the leakage source positioning unit specifically performs:

[0026] constructing a three-dimensional space grid model of a charging pile, and mapping the fusion leakage feature tensor to corresponding grid nodes;

[0027] capturing concentration gradient correlation of adjacent nodes by using a graph convolution layer;

[0028] outputting a set of grid nodes with labeled leakage probability values, and forming a leakage source probability heat map.

[0029] Preferably, the adaptive protection execution module specifically includes:

[0030] a gas barrier spraying assembly, configured to release an inert gas envelope layer to a high-probability area in the leakage source probability heat map when receiving the combustible gas leakage identifier;

[0031] a liquid-solid conversion assembly, configured to inject a high-molecular solidifying agent at a key node of a liquid seepage path when receiving the electrolyte leakage identifier;

[0032] an adsorption recovery assembly, configured to start a negative pressure adsorption device and locate a leakage source coordinate when receiving the insulating oil leakage identifier.

[0033] Preferably, the adaptive protection execution module further includes:

[0034] a dynamic strategy adjustment module, configured to correct a release concentration of the inert gas envelope layer, an injection flow of the high-molecular solidifying agent, or a power parameter of the negative pressure adsorption device in real time according to a gradient change rate of the leakage source probability heat map.

[0035] Preferably, the leakage risk decision module further includes:

[0036] a protection effectiveness feedback loop, configured to collect a concentration distribution of the inert gas envelope layer, a coverage area of the high-molecular solidifying agent, or a recovery efficiency of the negative pressure adsorption device, and update a gas concentration gradient distribution, an abnormal frequency band energy of a sound wave, and a liquid seepage rate in the fusion leakage feature tensor.

[0037] Preferably, the protection strategy decision unit is further configured to:

[0038] When the protection effectiveness feedback loop detects that the gas concentration gradient distribution has not decreased to a safety threshold, the protection level instruction level is raised and the dynamic strategy adjustment module is triggered to increase the inert gas release concentration;

[0039] When it is detected that the liquid seepage rate continues to rise, an injection instruction of the high molecular solidifying agent to the liquid-solid conversion assembly is sent.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] The intelligent leakage monitoring and protection system of the charging pile synchronously captures multi-dimensional leakage correlation signals such as gas concentration gradient distribution, abnormal frequency band energy of sound waves, and liquid seepage rate through a multi-source leakage perception module, breaking the limitations of traditional single-sensor monitoring, and being able to comprehensively capture characteristic information in the leakage process from different physical dimensions, effectively avoiding monitoring deviations caused by single signal misjudgment or omission, making leakage monitoring more comprehensive and accurate. Compared with the traditional monitoring method which can only obtain discrete data of a single dimension, the synchronous collection of multi-source signals can fully present the dynamic development process of leakage, providing more abundant and reliable data sources for subsequent leakage characteristic analysis and risk judgment, and helping the system to more accurately perceive the leakage state.

[0042] The leakage characteristic fusion module performs cross-modal feature alignment on the multi-source perception signals to generate a spatio-temporally correlated fusion leakage feature tensor, realizing deep integration and collaborative analysis of different types of monitoring data. In traditional systems, each monitoring data is independent of each other, and the internal correlation between data cannot be established, leading to fragmented cognition of the leakage scene. The cross-modal feature alignment technology can correlate and match feature data of different dimensions such as gas, sound waves, and liquids in time and space dimensions, excavate the leakage rules and characteristics hidden behind the data, make the generated fusion leakage feature tensor more truly and comprehensively reflect the essential properties of the leakage event, and provide a more scientific basis for subsequent leakage risk decision-making, avoiding decision-making bias caused by data isolation.

[0043] The leakage risk decision module generates a leakage type identifier, a leakage source probability distribution, and a protection level instruction based on the fusion leakage feature tensor, realizing accurate classification and quantitative evaluation of the leakage risk. Instead of relying on fixed thresholds for simple judgment, this module can accurately distinguish different types of leakage events through deep analysis of multi-dimensional fusion features, clearly identify the possible location and probability distribution of the leakage source, and divide the corresponding protection level according to the severity of the leakage, making the risk decision more targeted and refined. This decision-making method can effectively avoid the blindness of protection measures in traditional systems due to the inability to distinguish leakage types and locate leakage sources, making subsequent protection actions more targeted, and improving the control ability of leakage risk.

[0044] The adaptive protection execution module dynamically adjusts the leakage suppression strategy based on the protection level instruction, realizing the accurate adaptation of the protection measures and the leakage scene. The traditional fixed protection mode cannot be flexibly adjusted according to the actual leakage situation, while the module can automatically select and execute the matching protection strategy according to the protection level instruction output by the risk decision module, such as local plugging and early warning for trace electrolyte leakage, emergency power-off and ventilation for large amount of flammable gas leakage, etc. This kind of dynamically adaptive protection mode can minimize the impact on the normal operation of the charging pile under the premise of ensuring safety, avoid service interruption caused by excessive protection, and ensure that the protection measures can be upgraded in time when the leakage risk intensifies, effectively curb the spread of leakage and reduce the possibility of safety accidents, comprehensively improve the safety and reliability of the operation of the charging pile, and provide stronger safety protection for the equipment, users and surrounding environment. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The timing diagram of the charging pile intelligent leakage monitoring and protection system described in the application;

[0046] Figure 2 The working principle flow chart of the multi-source leakage perception module;

[0047] Figure 3 The working principle flow chart of the sub-module of the leakage feature fusion module;

[0048] Figure 4 The working principle flow chart of the leakage source positioning unit. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0050] Please refer to Figure 1 The application provides a charging pile intelligent leakage monitoring and protection system, which comprises:

[0051] The system realizes comprehensive management and control of the leakage risk of the operating environment through systematic data acquisition, processing and execution mechanism, and is composed of four core modules: multi-source leakage perception module, leakage feature fusion module, leakage risk decision module and adaptive protection execution module. These modules form a closed-loop control system through hardware integration and algorithm cooperation, realizing omnidirectional monitoring and intelligent protection of the operating state of the charging pile.

[0052] The multi-source leakage perception module is responsible for synchronously capturing various leakage-related signals in the charging pile operating environment as the system data acquisition front end. It adopts a distributed sensing network layout to cover the charging pile body and the surrounding key areas. The perception signals include three core parameters: gas concentration gradient distribution, abnormal frequency band energy of sound waves, and liquid seepage rate. The gas concentration gradient distribution is obtained by arranging a nano gas sensor array at the bottom of the charging pile and the cable interface area. The sensors are distributed in a grid topology, and the spacing density is optimized according to the leakage diffusion model. Real-time monitoring of the concentration spatial distribution characteristics of flammable gas and volatile organic compounds is achieved. The abnormal frequency band energy of sound waves is collected by a piezoelectric sensor group embedded in the charging pile insulation shell. The sensor is designed with a wide frequency band response, focusing on capturing acoustic signals in the 20kHz-100kHz frequency band, which contains high-frequency vibration components generated by typical leakage events. The liquid seepage rate monitoring is completed by a microfluidic detection unit integrated into the charging pile foundation structure layer. Based on the capillary pressure sensing principle, it quantifies the flow characteristics of the leakage liquid in real time. All sensing data are synchronously collected through an industrial bus with a time synchronization accuracy of milliseconds, ensuring the spatiotemporal consistency of multi-source data.

[0053] The leakage feature fusion module is responsible for processing and fusing multi-source heterogeneous data. It performs cross-modal feature alignment and fusion processing on the raw signals collected by the perception module. It uses a time-space dual alignment strategy to map gas concentration signals, sound wave signals, and liquid flow signals to a unified timestamp coordinate system. It solves the synchronization problem of different sampling rates data through interpolation algorithms. Then feature extraction and selection are performed: for gas concentration gradient distribution, sliding window segmentation processing is used to extract time domain features such as concentration mutation slope and steady-state drift; for abnormal frequency band energy of sound waves, wavelet packet transform is performed to separate high-frequency impact components and low-frequency resonance components; for liquid seepage rate data, flow state modeling is performed to construct a seepage path topology graph and calculate the node flow rate difference coefficient.

[0054] The multi-head attention mechanism is used to realize cross-modal feature weighted fusion, different physical meaning feature vectors are mapped to a unified feature space, and a three-dimensional fusion leakage feature tensor with spatio-temporal correlation is generated, which contains time dimension, space dimension and feature dimension information. The leakage risk decision module makes leakage risk judgment and decision based on the fusion feature tensor, adopts a deep learning algorithm architecture, and includes three functional units of leakage type identification, leakage source positioning and protection strategy generation. The leakage type classification unit uses a time convolution network to process the feature tensor, extracts time sequence features through multiple causal convolution and dilated convolution, and outputs three types of identification and their probability distribution of combustible gas leakage, electrolyte leakage and insulating oil leakage. The leakage source positioning unit is based on a graph neural network architecture, constructs a three-dimensional space grid model of the charging pile, maps the feature tensor to the grid nodes, captures the spatial correlation between nodes through the graph convolution layer, and generates a leakage source probability heat map to visually display the spatial distribution of the leakage probability. The protection strategy decision unit matches the preset protection level instructions according to the leakage type identification and the leakage source probability distribution, and the protection level is divided into a three-level response mechanism corresponding to different risk levels of the leakage scene.

[0055] The adaptive protection execution module dynamically adjusts and executes the leakage suppression strategy according to the protection instructions output by the decision module, including multiple special execution mechanisms: the gas barrier injection assembly for combustible gas leakage releases inert gas to form an envelope layer through the nozzle array arranged in the key area; the liquid-solid conversion assembly for electrolyte leakage injects a high-molecular solidifying agent at the key nodes of the seepage path through a precision injection system; the adsorption recovery assembly for insulating oil leakage uses a negative pressure adsorption device to cooperate with a mobile collection mechanism to realize the recovery and treatment of the leakage. All execution components are equipped with parameter-adjustable control interfaces, which can adjust the operation parameters in real time according to the changes in the leakage situation.

[0056] The system realizes continuous optimization operation through a closed-loop feedback mechanism. The protection effectiveness feedback loop collects execution effect data in real time, including gas concentration distribution changes, solidifying agent coverage area, adsorption recovery efficiency and other parameters. These data are fed back to the feature fusion module for updating the feature tensor, which in turn affects the subsequent decision-making process. When the protection effect is not as expected, the system automatically raises the protection level and adjusts the execution parameters, forming a complete control closed loop of monitoring-decision-execution-feedback. The entire system uses an industrial-grade hardware platform and a real-time operating system to ensure reliability and real-time performance in the complex operating environment of the charging pile, and realizes early detection, accurate positioning and effective suppression of leakage risks.

[0057] Example 1: see Figure 2The specific implementation of the distributed gas sensing unit involves the arrangement of nanometer gas sensitive material array and the data acquisition method. The sensor nodes made of metal oxide semiconductor material are uniformly distributed in the form of a grid inside the protective shell around the cable interface at the bottom of the charging pile bearing platform. Each sensor node contains a nanometer gas sensitive material sensitive element and a temperature compensation element. The sensitive element adopts a tin dioxide nanowire structure, and its resistance value will change with the change of the surrounding gas concentration. These sensor nodes are connected into a network through the I2C digital bus. The bus controller polls the node data at a frequency of 10 Hz. The calculation of the gas concentration gradient distribution is based on the concentration difference between adjacent nodes. By measuring the concentration difference of four adjacent nodes in the east-west-south-north direction, a vector field of concentration change is formed. In order to eliminate the measurement error caused by the change of environmental temperature, each node is equipped with a digital temperature sensor, and a polynomial fitting algorithm is used to compensate the original concentration data in real time. The power supply of the sensor array uses an independent DC power module to avoid electromagnetic interference with the main power supply of the charging pile. During the data acquisition process, the baseline of each node is calibrated, and the reference benchmark is established by monitoring the resistance value in clean air. Then the actual gas concentration value is calculated in a differential manner. The collected concentration data is transmitted to the central processing unit through the CAN bus. The transmission protocol uses a custom real-time data frame structure. Each frame of data contains node number, concentration value, temperature value and time stamp information.

[0058] The implementation of the broadband acoustic wave acquisition unit is based on the arrangement of a piezoelectric sensor group and a signal processing scheme. Four PZT-5A piezoelectric ceramic sensors are embedded in the four corners of the inner wall of the charging pile insulation shell, forming an acoustic monitoring array. The sensors are protected by epoxy resin packaging and directly coupled with the shell material to ensure the efficiency of sound wave propagation. Each sensor is connected to a high input impedance preamplifier with a gain setting of 100. The subsequent signal is digitized and sampled by a 24-bit analog-to-digital converter with a sampling rate of 200 kHz to meet the Nyquist sampling theorem requirements. The extraction of abnormal frequency band energy of acoustic wave uses a digital signal processing method. The original signal is band-pass filtered to retain the frequency components of 20 kHz to 100 kHz, and then the spectral energy distribution is calculated by short-time Fourier transform. The abnormal frequency band monitoring focuses on the energy changes in the range of 40 kHz to 60 kHz, which is usually related to the turbulent noise generated by gas leakage and the acoustic characteristics of liquid injection. The signal processing algorithm uses a sliding window method with a window length of 1024 sampling points and an overlap rate of 50%. The power spectral density is calculated for each window and the peak energy is detected. To distinguish between normal operation noise and leakage noise, the system establishes an acoustic feature database containing typical noise spectra generated by cooling fans, relay actions, and power electronic device switching during normal operation of the charging pile. The spectral subtraction algorithm is used to eliminate these background noise disturbances. The collected acoustic feature data is transmitted to the processing unit through the Ethernet interface, and the UDP protocol is used to ensure the real-time transmission.

[0059] The implementation of the microfluidic detection unit relies on a network of microfluidic chips integrated in the foundation structure layer of the charging pile. The chips are made of PDMS material using soft lithography technology and contain a series of micron-scale channels and pressure sensing chambers. The microchannel network covers the surface area of the foundation, with a channel width of 200 microns and a depth of 100 microns, forming a capillary-like liquid passage. MEMS pressure sensors are integrated at each intersection of the channels. The sensors work on the principle of piezoresistive effect and generate pressure changes when liquid flows into the channels. The quantification of liquid seepage rate is achieved by monitoring the rate of pressure change. The pressure gradient between adjacent sensing points is calculated using differential pressure measurement method. The system collects pressure data from each node at 1 second intervals, and converts the pressure difference into flow rate values through the Hagen-Poiseuille equation. To accommodate liquids of different viscosities, the system also integrates an online viscosity detection function, which estimates liquid viscosity by measuring the pressure decay time constant. A temperature compensation algorithm corrects for changes in liquid viscosity and sensor drift caused by temperature changes in real time. The microfluidic chip is connected to a data acquisition card through a flexible circuit. The acquisition card digitizes the pressure signal with 16-bit resolution and transmits it to the central processing unit through an RS485 bus. The entire microfluidic network is designed in a modular fashion, with each module covering an area of 0.25 square meters. Multiple modules are connected to form a complete monitoring network. To prevent channel blockage, the system also has a reverse flushing mechanism that periodically injects clean air into the channels to keep them unobstructed.

[0060] Example 2: see Figure 3The time-series feature extraction submodule adopts a sliding time window analysis method when processing the gas concentration gradient distribution data. The window length is set to 30 seconds, and there is a 15-second overlap between adjacent windows, which can ensure continuous monitoring and not miss any mutation signals. In each time window, the system calculates the change characteristics of the gas concentration. The concentration mutation slope is obtained by first-order derivation of the concentration-time curve, and this value reflects the degree of concentration change. The steady-state drift is obtained by calculating the root mean square deviation of the concentration data in the window from the baseline value, which is used to quantify the slow change trend of the concentration. These calculation processes rely on digital signal processing algorithms, which are implemented in C++ language and run on embedded processors. The processed feature data is time-stamped and stored in a ring buffer for subsequent fusion processing. The frequency-domain feature decomposition submodule uses wavelet packet transform when processing the sound wave signal. Wavelet packet transform can be understood as a more refined frequency band division method, which uses db4 wavelet basis function to decompose the sound wave signal into 16 different frequency bands. High-frequency impact components refer to the total energy of wavelet packet coefficients extracted from the frequency band above 60 kHz. These high-frequency components are usually related to sudden leakage events. Low-frequency resonance components are the total energy extracted from the frequency band from 20 kHz to 40 kHz, which are often related to continuous leakage. The entire decomposition process is accelerated by a digital signal processor, and the overlap reservation method is used to improve the calculation efficiency. The processed frequency-domain features are synchronized with the time-domain features. The flow state feature modeling submodule constructs a seepage path topology graph for liquid seepage data. The graph is composed of nodes and edges. Nodes correspond to each pressure sensor point in the microfluidic detection unit, and edges represent the connectivity between adjacent sensor points. The node flow velocity difference coefficient is obtained by calculating the Euclidean distance of the flow velocity vector between adjacent nodes. This coefficient reflects the non-uniformity of liquid flow, and the calculation formula is:

[0061] wherein: represents the node flow velocity difference coefficient, represents the flow velocity value of the th node, represents the number of adjacent nodes. This calculation process is executed in real time by a graph computing engine, which stores the topology relationship based on the adjacency list data structure and uses the breadth-first search algorithm to update the flow path.

[0062] The cross-modal alignment submodule is responsible for mapping features from different sources into a unified time coordinate system. Since the sampling rate of gas concentration data is 10 Hz, the sampling rate of acoustic feature data is 100 Hz, and the sampling rate of liquid flow rate data is 1 Hz, the system uses a linear interpolation algorithm to resample all data to a unified frequency of 100 Hz. The interpolation algorithm is based on a timestamp synchronization mechanism, which assigns a uniform time index to each data point. The concentration mutation slope, high-frequency impact component, and node flow rate difference coefficient are selected for alignment because they all reflect the transient characteristics of the leakage event. A sliding time window matching strategy is used during the alignment process, with a window length of 100 milliseconds, ensuring that the time deviation of different modal data does not exceed 10 milliseconds. The feature tensor generation submodule uses a multi-head attention mechanism to weight and fuse the steady-state drift, low-frequency resonance component, and seepage path topology graph. The attention mechanism can be understood as an intelligent weighting system that automatically assesses the importance of different features and assigns appropriate weights. This mechanism contains 4 independent attention heads, each responsible for handling different dimensional feature relationships: the first head handles feature associations in the time dimension, the second head handles feature distributions in the spatial dimension, the third head handles correlations between different physical quantities, and the fourth head handles cross-modal feature interactions. The weighted feature vectors output by each attention head are concatenated into a comprehensive feature vector, which is then reorganized into a 100x50x12 fusion leakage feature tensor through a three-dimensional reorganization operation, where 100 represents the time step, 50 represents the number of spatial grid points, and 12 represents the number of feature dimensions. The entire fusion process is accelerated on a graphics processing unit, using a CUDA parallel computing architecture to improve processing efficiency.

[0063] When certain sensor data is temporarily missing, the system employs Kalman filtering algorithm for data completion, predicting the current missing values based on historical data trends. For the dimension mismatch problem that occurs during feature fusion, the system uses zero padding and feature scaling methods to ensure data consistency. To protect data integrity, all feature data is subjected to error detection using cyclic redundancy check codes, and once data errors are detected, immediate retransmission is required. The entire processing flow is designed using a pipeline architecture, with each sub-module running in parallel, and data exchanged through shared memory areas to reduce data transmission delays. The system also has a real-time monitoring interface that can visually display the entire process of feature extraction and fusion, including time series curves, frequency spectrum graphs, and changes in topology graphs. All algorithm parameters can be dynamically adjusted through configuration files, such as sliding window size, wavelet basis function type, and number of attention heads, which allows the system to adapt to the monitoring needs of different types of charging piles. System status data is recorded in real-time to log files, including processing delays, memory usage, and computing load, providing a reference for subsequent system optimization. The implementation of the entire embodiment ensures that the leakage monitoring system can efficiently process multi-source heterogeneous data and provide accurate feature input for subsequent risk decision-making.

[0064] Embodiment 3: refer to Figure 4 The leakage type classification unit uses a time convolution network to process the fused leakage feature tensor. This network structure contains four causal convolution modules, each with 64 filters, a convolution kernel size of 3, and an expansion coefficient set to 1, 2, 4, and 8 for each level. Causal convolution ensures that the output only depends on the current and historical inputs, avoiding future information leakage. Batch normalization and activation operations are performed after each convolution, and the output is compressed in the time dimension by a global average pooling layer and mapped to a three-dimensional classification space by a fully connected layer. The function generates a probability distribution for the three types of leakage identifiers. A probability threshold of 0.7 is set, and when the probability of a certain type of leakage exceeds this threshold, the corresponding identifier is triggered. The network is trained using labeled historical leakage data, with the Adam algorithm as the optimizer and cross-entropy loss as the loss function. Early stopping is used during training to prevent overfitting. The leakage source positioning unit constructs a three-dimensional space grid model of the charging pile based on a graph neural network, mapping the fused leakage feature tensor to a 5-centimeter resolution spatial grid node, each containing a 12-dimensional feature vector. The graph neural network uses a graph attention mechanism, and its node update formula is represented as:

[0065] where: represents the updated feature representation of node i, represents the activation function, represents the neighbor node set of node , Represents a node With nodes Attention weights between them This represents a trainable weight matrix. This represents the original feature vector of neighboring node j. Attention weights are obtained by calculating the feature similarity between nodes, focusing on neighboring nodes with strong correlation in concentration gradients. After three layers of graph convolution operations, the output layer... The function calculates the leakage probability value for each node and generates a heatmap of leakage source probability. This heatmap is visualized using pseudo-color rendering, with red areas representing high-probability leakage sources and blue areas representing low-probability areas.

[0066] The protection strategy decision unit generates protection level instructions based on the matching results between the leak type identifier and the leak source probability heat map. The matching algorithm calculates the product of the leak type probability value and the highest probability value in the heat map. When the product value exceeds 0.5, a Level-3 protection level instruction is triggered. Level-1 instructions correspond to low-risk scenarios, issuing only a warning signal and initiating regular monitoring mode; Level-2 instructions correspond to medium-risk scenarios, initiating pre-protection measures and increasing monitoring frequency; Level-3 instructions correspond to high-risk scenarios, immediately initiating comprehensive protection measures and triggering the emergency response mechanism. The decision logic is implemented using a rule engine, supporting online updates of matching rules to adapt to different operating environments. During system operation, changes in the leak situation are continuously monitored. When the probability product value is detected to move from the low-risk range to the high-risk range, the protection level is automatically upgraded; when the situation eases, the protection level is gradually reduced. Detailed logs are recorded for all decision-making processes, including timestamps, input features, decision results, and confidence scores, providing data support for subsequent system optimization.

[0067] When multiple leakage sources occur simultaneously, the algorithm employs a non-maximum suppression strategy to handle multiple peak regions in the heatmap, ensuring accurate identification of the main leakage source. For uncertain identification results, the system introduces a confidence assessment mechanism; when the confidence level falls below a preset threshold, a multimodal data verification process is initiated. The graph neural network training uses a semi-supervised learning approach, leveraging a small amount of labeled data and a large amount of unlabeled data to improve the model's generalization ability. The real-time inference process is accelerated through model quantization technology, converting floating-point operations to fixed-point operations to improve computational efficiency while maintaining accuracy. The system also features online learning capabilities, automatically updating model parameters based on newly occurring leakage cases to gradually improve identification accuracy. All model versions undergo rigorous testing and verification to ensure no performance degradation occurs during updates. The system's real-time monitoring interface displays leakage type identification results, leakage source location heatmaps, and the protection level decision-making process. Operators can view the system's operating status at any time and access a manual intervention interface.

[0068] Example 4: The specific implementation of the adaptive protection execution module involves the coordinated control and parameter adjustment mechanism of multiple sets of execution mechanisms. The gas barrier injection assembly is composed of a high-pressure nitrogen gas storage tank, an electromagnetic valve array, and a nozzle group. The working pressure of the storage tank is 20 MPa, and the output pressure is stabilized at 0.8 MPa through a pressure reducing valve. The electromagnetic valve adopts a two-position three-way structure with a response time less than 50 milliseconds. The nozzle arrangement density is 4 per square meter, covering the entire area at the bottom of the charging pile. When receiving the combustible gas leakage identifier, the control system calculates the required nozzle combination to be opened according to the region coordinates with a probability value greater than 0.8 in the probability thermal distribution map of the leakage source. The initial release concentration is 10 L / min, and the nitrogen purity requirement is above 99.99%. An inert gas envelope layer is formed to dilute the combustible gas concentration below the lower explosive limit. The liquid-solid conversion assembly includes a high-molecular solidifying agent storage tank, a metering pump, and an injection nozzle. The solidifying agent uses acrylic fast-setting material with a setting time adjustable range of 5-30 seconds. The injection pump is driven by a stepper motor with a minimum flow control accuracy of 0.1 mL / s and an initial setting flow of 5 mL / s. The system determines the key injection position according to the nodes with a flow rate difference coefficient greater than 0.3 in the seepage path topology map, and the injection pressure is maintained within the range of 0.2-0.5 MPa. The adsorption recovery assembly is composed of a vacuum pump, an oil-water separator, and a six-degree-of-freedom mechanical arm. The vacuum pump has a limit vacuum degree of -95 kPa and an air pumping rate of 50 L / min. The mechanical arm has a positioning accuracy of ±1 mm, and the end effector is equipped with an infrared vision sensor for precise positioning of the leakage source. When receiving the insulating oil leakage identifier, the system preliminarily locates through the thermal map coordinates, performs secondary precise positioning using the vision sensor, starts the adsorption recovery operation, and sets the initial adsorption power to -50 kPa.

[0069] The dynamic strategy adjustment module monitors the gradient change rate of the leakage source probability thermal distribution map in real time. The rate is obtained by calculating the difference of the probability values of adjacent time frames with a sampling interval of 1 second. According to the rate change amplitude, the system automatically adjusts the operating parameters of each execution component. The adjustment logic is based on a pre-set parameter mapping table that establishes the correspondence between the gradient change rate and the execution parameters. For example, when the gradient change rate is detected to exceed 0.1 / s, the system proportionally increases the inert gas release concentration to 15 L / min, increases the high-molecular solidifying agent injection flow to 8 mL / s, and increases the power of the negative pressure adsorption device to -70 kPa. During the adjustment process, a PID control algorithm is used to achieve smooth transition and avoid system oscillation caused by parameter mutation. The running state of all execution components is fed back in real time through a sensor network, including gas concentration sensors, flow meters, and pressure sensors, forming a closed-loop control circuit. Refer to Table 1.

[0070] Table 1: Parameter adjustment logic table

[0071] When multiple leakage sources are detected to exist simultaneously, the system adopts a priority scheduling algorithm to determine the processing order according to the risk level and probability value of the leakage type. For the case of failure of the execution component, the system has a redundant switching function, which automatically enables the standby device when the main execution component fails. All parameter adjustment processes are recorded in detail, including time stamp, gradient change rate value, parameter value before and after adjustment, and adjustment result status. These data are used for subsequent system optimization and maintenance. The operation interface displays the working state of each execution component and the parameter adjustment process in real time, and provides a manual intervention interface for special case processing. The system regularly checks the performance indicators of each execution component, including nozzle blockage detection, pump efficiency test, and mechanical arm precision calibration, to ensure the reliability of the protection execution. The entire implementation process ensures that the leakage protection system can dynamically adjust the operating parameters according to real-time monitoring data, achieving precise and effective leakage suppression.

[0072] In embodiment 5, the effectiveness feedback loop is implemented. Real-time execution effect data is collected by a multi-source sensor network, including inert gas envelope layer concentration distribution, polymer curing agent coverage area, and negative pressure adsorption device recovery efficiency. The gas concentration distribution monitoring uses the original distributed gas sensing unit to continue monitoring the concentration changes of each node after protection execution, with a sampling frequency of 10 Hz. The curing agent coverage area is indirectly calculated by the pressure change of the microfluidic detection unit. After the curing agent is injected, the liquid flow rate changes, and the curing coverage range can be calculated according to the flow rate change curve. The adsorption recovery efficiency is calculated by the flow sensor and concentration sensor built-in the negative pressure adsorption device, which monitors the mass flow and component concentration of the adsorbed material in real time. All these effectiveness data are sent back to the central processing unit through the same data transmission path, and are associated with the original monitoring data using timestamp alignment. The system uses a sliding window average algorithm to update the feature tensor, and the newly collected effectiveness data is integrated into the historical data with dynamic weights, ensuring that the feature update reflects the latest state while maintaining data stability. The feature tensor update process is calculated using the following formula:

[0073]

[0074] Where: represents the updated feature tensor, represents the current feature tensor, is the historical data weight coefficient, ranging from 0 to 1, represents the data value of the th effectiveness monitoring point, represents the confidence weight of the corresponding data point, represents the number of monitoring points. The coefficient is dynamically adjusted according to the data timeliness, the weight of new data gradually increases over time, and the weight coefficient of historical data over 5 minutes is exponentially decayed to ensure the sensitivity of the system to the current state. The entire feedback data collection cycle is synchronized with the protection execution cycle to ensure that the corresponding performance evaluation data is obtained after each protection action.

[0075] The protection strategy decision unit dynamically adjusts the protection level instruction according to the monitoring results of the feedback loop. When the gas concentration sensor detects that the concentration value after protection has not decreased to the safety threshold, the safety threshold is set to 20% of the lower explosive limit concentration of combustible gas, and the system automatically upgrades the protection level instruction level. The upgrading process adopts a step-by-step upgrade strategy. When upgrading from Level-2 to Level-3, the inert gas release concentration is increased by 15%. If the concentration still does not decrease, the release concentration will continue to be increased, and the increase amplitude is calculated based on the proportion of the concentration difference. At the same time, the dynamic strategy adjustment module is triggered, which calculates the amount of inert gas release that needs to be increased based on the concentration decrease rate, and adjusts the electromagnetic valve opening degree and injection duration using a proportional-integral-derivative control algorithm. For liquid seepage scenarios, when the microfluidic detection unit monitors a continuous increase in flow rate, defined as a 10% increase in flow rate over three consecutive sampling periods, the system sends an increase injection instruction to the liquid-solid conversion component. The instruction generation is based on the functional relationship between the flow rate change rate and the current injection flow rate, and uses a feedforward control algorithm to pre-adjust the injection parameters to avoid response delay. The injection flow rate adjustment amplitude is determined by table lookup method, which matches the preset adjustment coefficient according to the flow rate growth rate, to ensure that the injection flow rate and seepage velocity maintain a proper proportion. All adjustment instructions are safety checked to ensure that parameter changes are within the device's allowed operating range, avoiding over-adjustment that can cause device damage or secondary pollution.

[0076] A perfect abnormality handling mechanism is established during system operation. When the feedback data deviates significantly from the expected results, the system starts the root cause analysis process, checking the sensor accuracy, actuator status, and communication link in turn. For persistent abnormal conditions, the system can automatically switch to a degraded mode, using conservative parameter settings to ensure basic protection functions. All adjustment processes record detailed operation logs, including initial state, adjustment instructions, execution results, and final state, which are used for subsequent system performance analysis and optimization. The system also has a learning function, gradually optimizing parameter adjustment strategies by analyzing historical adjustment records and effect data, improving protection efficiency. The operator can view the feedback loop running state through the human-machine interface, including real-time data curves, adjustment history records, and system performance indicators, and manually intervene when necessary. The entire feedback loop design ensures that the protection system can continuously optimize operating parameters based on actual effects, achieving increasingly accurate leakage protection. Regular system maintenance will calibrate and test the feedback loop to ensure that all sensors and actuators are working within the specified accuracy range, and maintenance records are included in the system archives for subsequent query and analysis. The weight coefficient κ in the formula is determined through experimental data optimization, and the confidence weight According to the sensor accuracy level distribution, the number of monitoring points n is dynamically adjusted according to the total number of sensors actually deployed, and these parameters together ensure the accuracy and reliability of feature tensor updates.

[0077] It should be noted that in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0078] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A charging pile intelligent leakage monitoring and protection system, characterized in that, The method comprises the following steps: A multi-source leakage perception module is used to synchronously capture leakage-related signals of the charging pile operating environment, wherein the leakage-related signals include gas concentration gradient distribution, abnormal frequency band energy of sound waves, and liquid seepage rate; A leakage feature fusion module is used to perform cross-modal feature alignment on the leakage-related signals collected by the multi-source leakage perception module, and generate a spatiotemporally correlated fusion leakage feature tensor; A leakage risk decision module is used to generate a leakage type identification, a leakage source probability distribution, and a protection level instruction according to the fusion leakage feature tensor; An adaptive protection execution module is used to dynamically adjust a leakage suppression strategy based on the protection level instruction. The leakage feature fusion module specifically comprises: A time series feature extraction submodule is used to perform sliding window segmentation on the gas concentration gradient distribution, and extract concentration mutation slope and steady-state drift; A frequency domain feature decomposition submodule is used to perform wavelet packet transform on the abnormal frequency band energy of sound waves, and separate high-frequency impact components and low-frequency resonance components; A flow state feature modeling submodule is used to construct a seepage path topology graph according to the liquid seepage rate, and calculate a node flow rate difference coefficient. The leakage feature fusion module further comprises: A cross-modal alignment submodule is used to map the concentration mutation slope, high-frequency impact components, and node flow rate difference coefficient to a unified timestamp coordinate system; A feature tensor generation submodule is used to output a three-dimensional fusion leakage feature tensor by weighting and splicing the steady-state drift, low-frequency resonance components, and seepage path topology graph through a multi-head attention mechanism.

2. The intelligent leakage monitoring and protection system for charging pile according to claim 1, characterized in that, The multi-source leakage perception module specifically comprises: A distributed gas sensing unit is arranged at the bottom of the charging pile and the cable interface area, and uses a nano gas-sensitive material array to capture the gas concentration gradient distribution; A wideband sound wave acquisition unit is embedded in the inside of the charging pile insulation shell, and extracts the abnormal frequency band energy of sound waves through a piezoelectric sensor group; A microfluidic detection unit is integrated in the charging pile foundation structure layer, and quantifies the liquid seepage rate based on capillary pressure sensing technology.

3. The intelligent leak monitoring and protection system for electric charging piles according to claim 1, characterized in that, The leakage risk decision module specifically comprises: A leakage type classification unit is used to process the fusion leakage feature tensor using a time convolution network, and output a flammable gas leakage identification, an electrolyte leakage identification, or an insulating oil leakage identification; A leakage source positioning unit is used to analyze the spatial correlation features in the fusion leakage feature tensor based on a graph neural network, and generate a leakage source probability heat distribution map; A protection strategy decision unit is used to match the leakage type identification and the leakage source probability heat distribution map with a preset protection level instruction.

4. The intelligent leak monitoring and protection system for charging piles according to claim 3, characterized in that, The leakage source positioning unit specifically performs the following steps: A three-dimensional space grid model of the charging pile is constructed, and the fusion leakage feature tensor is mapped to the corresponding grid nodes; The concentration gradient correlation of adjacent nodes is captured through a graph convolution layer; A set of grid nodes with labeled leakage probability values is output, forming a leakage source probability heat distribution map.

5. The intelligent leak monitoring and protection system for electric charging piles according to claim 4, characterized in that, The adaptive protection execution module specifically comprises: A gas barrier injection assembly is used to release an inert gas envelope layer to the high-probability area in the leakage source probability heat distribution map when receiving the flammable gas leakage identification; A liquid-solid conversion assembly is used to inject a high-molecular solidifying agent at the key nodes of the liquid seepage path when receiving the electrolyte leakage identification. The adsorption recovery assembly starts a negative pressure adsorption device and locates a leakage source coordinate when receiving an insulating oil leakage identifier.

6. The intelligent leak monitoring and protection system for electric charging piles according to claim 5, characterized in that, The adaptive protection execution module further comprises: The dynamic strategy adjustment module adjusts in real time a release concentration of the inert gas envelope layer, an injection flow of the polymer solidifying agent or a power parameter of the negative pressure adsorption device according to a gradient change rate of the leakage source probability thermal distribution map.

7. The intelligent leak monitoring and protection system for electric charging piles according to claim 6, characterized in that, The leakage risk decision module further comprises: The protection efficiency feedback loop collects a concentration distribution of the inert gas envelope layer, a coverage area of the polymer solidifying agent or a recovery efficiency of the negative pressure adsorption device, and updates a gas concentration gradient distribution, an abnormal frequency band energy of a sound wave and a liquid seepage rate in the fusion leakage feature tensor.

8. The intelligent leak monitoring and protection system for electric charging piles according to claim 7, characterized in that, The protection strategy decision unit is further configured to: When the protection efficiency feedback loop detects that the gas concentration gradient distribution does not decrease to a safety threshold, the protection level instruction level is raised and the dynamic strategy adjustment module is triggered to increase the release concentration of the inert gas; When it is detected that the liquid seepage rate continues to rise, an injection instruction of the polymer solidifying agent is sent to the liquid-solid conversion assembly.

Images