Charging pile thermal runaway intelligent protection method and system based on edge calculation

Abstract

The invention discloses a charging pile thermal runaway intelligent protection method and system based on edge calculation, and relates to the technical field of charging pile thermal runaway protection. Comprising the following steps: S1, collecting thermal runaway multi-mode sensing data in real time, and carrying out data preprocessing; a charging pile thermal runaway multi-mode abnormal state is judged, and an abnormal feature data packet is generated; s2, multi-modal abnormal feature vectors are constructed, the thermal runaway risk probability is evaluated, and thermal runaway risk grading protection of the charging pile is carried out; s3, the response effect of thermal runaway risk grading protection is quantified, and thermal runaway risk grading protection is adjusted; and S4, monitoring and feeding back the thermal runaway risk, and optimizing algorithm parameters and a thermal runaway risk grading protection strategy. The problems that an existing charging pile thermal runaway protection system is difficult to adapt to complex working conditions, multiple in false alarm and missing alarm, lagging in response and insufficient in protocol compatibility and data stability, and consequently the thermal runaway risk is difficult to recognize and protect timely and accurately are solved.

Description

Technical Field

[0001] This invention relates to the field of thermal runaway protection technology for charging piles, specifically to an intelligent protection method and system for thermal runaway of charging piles based on edge computing. Background Technology

[0002] In recent years, with the rapid development of the new energy vehicle industry and the popularization of urban electric mobility, charging piles, as a crucial infrastructure supporting energy replenishment and green travel, have become a focus of social attention due to their safe operation and reliable service. Especially under complex operating conditions such as high-density, long-duration, and high-power charging, charging piles may face thermal runaway, fault propagation, and related safety risks. Enhancing the identification and response to thermal runaway in charging piles while ensuring charging efficiency has become an important task for promoting the healthy development of new energy transportation infrastructure, and is of great significance for improving user safety experience and promoting the construction of a new energy ecosystem.

[0003] For example, the invention patent with announcement number CN118013427B discloses a liquid-cooled car charging pile with leakage warning and a warning method. The method includes: acquiring historical leakage fault data of the liquid-cooled car charging pile and constructing a leakage fault evolution chain of the liquid-cooled car charging pile based on the historical leakage fault data; monitoring the liquid-cooled system status data of the liquid-cooled car charging pile to obtain an initial liquid-cooled system status dataset, and preprocessing the initial liquid-cooled system status dataset using a subspace recognition algorithm to obtain a target liquid-cooled system status dataset; extracting fault features and constructing a feature matrix of the target liquid-cooled system status dataset based on the leakage fault evolution chain to obtain a liquid-cooled system fault feature matrix; and inputting the liquid-cooled system fault feature matrix into a preset multi-classifier fusion decision model for leakage fault diagnosis to obtain leakage fault diagnosis results. This application improves the leakage warning accuracy of liquid-cooled car charging piles.

[0004] For example, the invention patent with publication number CN120180913A discloses an intelligent heat dissipation method, equipment, and medium for charging piles based on optimized heat conduction lines. This invention belongs to the field of charging pile heat dissipation technology. The technical problem this invention aims to solve is how to avoid the defects of high thermal resistance, slow response, and high energy consumption in traditional charging pile heat dissipation solutions, thereby improving the heat dissipation efficiency and operational reliability of charging piles. The technical solution adopted is as follows: optimizing the heat conduction structure; deploying thin-film thermocouples and constructing a three-dimensional temperature field model through infrared thermal imaging; establishing a thermal resistance-flow-power transfer function model using a hybrid algorithm of fuzzy PID and model predictive control; and analyzing historical charging data based on LSTM neural networks to predict the power peak 300ms in advance and pre-start the heat dissipation equipment.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] Existing thermal runaway protection systems for new energy vehicle charging piles are ill-suited to diverse operating conditions, suffer from high false alarm and false alarm rates, delayed response and blind spots, and lack protocol compatibility and data link stability. Consequently, the overall safety protection is unable to respond to thermal runaway risks in real-world operations in a timely and accurate manner.

[0007] Therefore, in order to address the above problems, there is an urgent need for an intelligent protection method and system for thermal runaway of charging piles based on edge computing. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides an intelligent protection method and system for thermal runaway of charging piles based on edge computing. This solves the problems of existing thermal runaway protection systems for charging piles being unable to adapt to complex operating conditions, having many false alarms and missed alarms, delayed response, insufficient protocol compatibility and data stability, which makes it difficult to identify and protect against thermal runaway risks in a timely and accurate manner.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: an intelligent protection method for thermal runaway of charging piles based on edge computing, comprising the following steps: S1, real-time acquisition of multimodal thermal runaway sensing data, and preprocessing of the multimodal thermal runaway sensing data; real-time identification of the multimodal abnormal state of thermal runaway of the charging pile based on the preprocessed multimodal thermal runaway sensing data, and generation of abnormal feature data packets; S2, receiving the abnormal feature data packets to construct a multimodal abnormal feature vector, real-time assessment of the probability of thermal runaway risk based on the multimodal abnormal feature vector, and construction of a digital twin space, while performing graded protection of thermal runaway risk of the charging pile based on the probability of thermal runaway risk; S3, comprehensive analysis of multimodal thermal runaway sensing data before and after graded protection of thermal runaway risk, quantification of the response effect of graded protection of thermal runaway risk, and adjustment of graded protection of thermal runaway risk based on the response effect; S4, through real-time linkage and data analysis of multiple platforms, visualization monitoring and feedback of the entire process of thermal runaway risk, and optimization of various algorithm parameters and graded protection strategies for thermal runaway risk.

[0012] Furthermore, the real-time acquisition of multimodal thermal runaway sensing data and the specific process of data preprocessing for this data are as follows: Multimodal thermal runaway sensing data is acquired in real-time using infrared and visible light dual-camera modules, anti-fog lenses, piezoelectric accelerometers, vibration sensors, a five-in-one electrochemical gas sensor, and a temperature and humidity sensor deployed at key locations on the charging pile. This multimodal thermal runaway sensing data includes: infrared radiation signals from the charging pile, images of the equipment surface, vibration signals, concentration values ​​of gases characteristic of thermal runaway, and ambient temperature and humidity data. The multimodal thermal runaway sensing data is aligned using a unified RTC timestamp. Using a blackbody calibration algorithm and temperature mapping relationship, the infrared radiation signals from the charging pile are converted into physical temperature values ​​in real-time to obtain the infrared temperature value of the charging pile. The equipment surface image is downsampled to reduce pixels; low-frequency interference in the vibration signal is filtered out using a high-pass filter and instrument amplification technology, while retaining high-frequency vibration information. The vibration signal is then converted into a digital signal using an ADC, and the short-time root mean square (RMS) value of the high-frequency vibration signal is extracted. The thermal runaway characterization gas concentration value is smoothed using a moving average algorithm to eliminate noise interference, and the thermal runaway characterization gas concentration value is corrected using a support vector regression algorithm based on ambient temperature and humidity. The thermal runaway multimodal sensing data is normalized. A charging pile thermal runaway protection database is established, storing the original and preprocessed thermal runaway multimodal sensing data, the charging pile infrared temperature value, and the high-frequency vibration signal RMS value in the charging pile thermal runaway protection database.

[0013] Furthermore, the specific process for real-time identification of the multimodal abnormal state of thermal runaway in charging piles based on preprocessed multimodal thermal runaway sensing data is as follows: Based on a sliding time window, the mean and standard deviation of the charging pile's infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterization gas concentration value are calculated respectively, yielding the mean and standard deviation of the charging pile's infrared temperature, the mean and standard deviation of the high-frequency vibration signal RMS value, the mean and standard deviation of the thermal runaway characterization gas concentration; the absolute value of the difference between the current charging pile's infrared temperature value and the mean of the charging pile's infrared temperature is then used. The infrared temperature anomaly value is obtained by dividing the value by the standard deviation of the infrared temperature of the charging pile; the vibration signal anomaly value is obtained by dividing the absolute value of the difference between the current high-frequency vibration signal RMS value and the mean RMS value of the high-frequency vibration signal by the standard deviation of the RMS value of the high-frequency vibration signal; the thermal runaway gas anomaly value is obtained by dividing the absolute value of the difference between the current thermal runaway gas concentration value and the mean thermal runaway gas concentration by the standard deviation of the thermal runaway gas concentration; the infrared temperature anomaly value, the vibration signal anomaly value, the thermal runaway gas anomaly value are summed with a constant, and the summation result is subjected to natural logarithm operation to obtain the multimodal anomaly detection value.

[0014] Furthermore, the specific process of generating the abnormal feature data packet is as follows: The multi-modal abnormal detection value is stored in the charging pile thermal runaway protection database in real time and compared with the abnormal threshold; when the multi-modal abnormal detection value is less than the abnormal threshold, it is determined that the thermal runaway multi-modal perception data is within the normal fluctuation range, and only the routine monitoring and storage of the multi-modal abnormal detection value are carried out; when the multi-modal abnormal detection value is greater than or equal to the abnormal threshold, the thermal runaway multi-modal perception data, infrared temperature abnormal value, vibration signal abnormal value, thermal runaway characterization gas abnormal value, and multi-modal abnormal detection value of the current period are marked, and an abnormal feature data packet is constructed and input to the edge computing layer.

[0015] Furthermore, the specific process of receiving the abnormal feature data packet to construct the multi-modal abnormal feature vector and evaluating the thermal runaway risk probability in real time according to the multi-modal abnormal feature vector is as follows: The edge computing layer receives the abnormal feature data packet, aligns the timestamps of all thermal runaway multi-modal perception data, infrared temperature abnormal value, vibration signal abnormal value, thermal runaway characterization gas abnormal value, and multi-modal abnormal detection value sequences, and fuses them into a multi-modal abnormal feature vector; based on the multi-modal abnormal feature vector, in the sliding time window, the infrared temperature abnormal value, vibration signal abnormal value, and thermal runaway characterization gas abnormal value corresponding to the previous moment and the current moment in three channels, including the infrared temperature channel, vibration signal channel, and thermal runaway characterization gas concentration channel, are respectively extracted; the change amount of the abnormal value corresponding to each channel is calculated by using the difference between the abnormal value at the current moment and the abnormal value at the previous moment, and then divided by the length of the sliding time window to obtain the feature change rate, and the square operation is performed on the feature change rate to obtain the abnormal rate square term; the abnormal rate square terms of the three channels are accumulated and averaged, and the negative value is taken as the exponential power, and the natural exponential operation is performed to obtain the abnormal suppression exponential term, and the multi-modal thermal runaway risk probability value is obtained by subtracting the abnormal suppression exponential term from the constant one.

[0016] Furthermore, the specific process of constructing the digital twin space and performing hierarchical thermal runaway protection of the charging pile according to the thermal runaway risk probability is as follows: The digital twin space is constructed by using the thermal runaway multi-modal perception data and the three-dimensional structure and attributes of the charging pile, and the edge computing layer maps the multi-modal thermal runaway risk probability value to the digital twin space in real time to perform holographic synchronization of the charging pile state; at the same time, the multi-modal thermal runaway risk probability value is compared in real time with the multi-level risk thresholds P1 and P2, and hierarchical thermal runaway protection measures are implemented; when < P1, only the routine monitoring and storage of the multi-modal thermal runaway risk probability value are carried out; when ≥ P1 and When P2, reduce the charging power of the charging pile, start the fan and liquid cooling pump to enhance heat dissipation, and trigger intermittent audible and visual alarms; share multi-modal perception data of thermal runaway and multi-modal thermal runaway risk probability values among multiple charging piles based on wireless ad hoc network, and identify group risks and abnormal diffusion trends; virtually reproduce the current anomaly in the digital twin space, evaluate the optimal combination of each response measure in the early warning scenario and push suggestions; when ≥P2, the dual-redundancy relay cuts off the charging circuit and isolates it; adjust the fresh air and gas discharge circuit; according to the infrared radiation signal of the charging pile and the surface image of the equipment, use the edge detection algorithm to identify the morphological changes of the battery compartment shell, synchronously start continuous audible and visual alarms and link the fire protection system; synchronously visualize the full process of response in the digital twin space.

[0017] Furthermore, the specific process of comprehensively analyzing the multi-modal perception data of thermal runaway before and after the hierarchical protection of thermal runaway risk and quantifying the response effect of the hierarchical protection of thermal runaway risk is as follows: after the implementation of the hierarchical thermal runaway protection measures, based on a sliding time window, count the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal, and the concentration value of the thermal runaway characterization gas within the current time window, and respectively calculate the change amount of the infrared temperature value of the charging pile after response, the change amount of the RMS value of the high-frequency vibration signal after response, and the change amount of the concentration value of the thermal runaway characterization gas after response through difference calculation; at the same time, extract the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal, and the concentration value of the thermal runaway characterization gas before the implementation of the hierarchical thermal runaway protection measures from the charging pile thermal runaway protection database, and respectively calculate the change amount of the infrared temperature value of the charging pile before response, the change amount of the RMS value of the high-frequency vibration signal before response, and the change amount of the concentration value of the thermal runaway characterization gas before response based on the sliding time window; square and sum the change amount of the infrared temperature value of the charging pile after response, the change amount of the RMS value of the high-frequency vibration signal after response, and the change amount of the concentration value of the thermal runaway characterization gas after response, and then take the square root to obtain the comprehensive fluctuation intensity value after response; square and sum the change amount of the infrared temperature value of the charging pile before response, the change amount of the RMS value of the high-frequency vibration signal before response, and the change amount of the concentration value of the thermal runaway characterization gas before response, and then take the square root to obtain the comprehensive fluctuation intensity value before response; subtract the ratio of the comprehensive fluctuation intensity value after response to the comprehensive fluctuation intensity value before response from the constant one to obtain the thermal runaway self-healing feedback value.

[0018] Furthermore, the specific process for adjusting the thermal runaway risk classification protection based on the response effect is as follows: After each graded thermal runaway protection measure is implemented, the thermal runaway self-healing feedback value is continuously monitored, written into the charging pile thermal runaway protection database, and compared with the self-healing threshold; when the thermal runaway self-healing feedback value is lower than the self-healing threshold and there is no upward trend within the sliding time window, the strength of each thermal runaway protection measure is increased, and each thermal runaway protection measure is assigned a priority; each adjustment of the thermal runaway protection measure and the corresponding thermal runaway self-healing feedback value are recorded; when the duration of the thermal runaway self-healing feedback value being lower than the self-healing threshold exceeds the maximum allowable threshold, a thermal runaway self-healing failure alarm is pushed, and a detailed response process and thermal runaway self-healing feedback value trend are generated and pushed to the fire protection system; when the thermal runaway self-healing feedback value is greater than or equal to the self-healing threshold, it is determined that the thermal runaway protection measure response has achieved substantial effect, the alarm state is lifted, and the equipment is gradually restored to normal operating mode.

[0019] Furthermore, through real-time linkage and data analysis across multiple platforms, the entire process of thermal runaway risk is visualized for monitoring and feedback. The specific process for optimizing algorithm parameters and thermal runaway risk tiered protection strategies is as follows: The edge computing layer is integrated with the BMS, power plant operation and maintenance, and fire protection systems to monitor multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values ​​in real time, enabling remote data sharing and big data analysis. Historical data from the charging pile thermal runaway protection database is periodically used to adjust the parameters of multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values, as well as anomaly thresholds, multi-level risk thresholds, and self-healing thresholds. New anomaly patterns are learned by integrating actual thermal runaway self-healing effects to optimize thermal runaway protection measures. Thermal runaway monitoring and feedback reports are periodically output, and the entire closed-loop response process of multimodal thermal runaway perception data and tiered thermal runaway protection measures is displayed in a digital twin space.

[0020] A second aspect of this invention provides an intelligent protection system for thermal runaway of charging piles based on edge computing, comprising: a multimodal perception and anomaly discrimination module, used to collect multimodal thermal runaway perception data in real time, preprocess the multimodal thermal runaway perception data, and based on the preprocessed multimodal thermal runaway perception data, to judge the multimodal abnormal state of thermal runaway of the charging pile in real time and generate anomaly feature data packets; and an edge risk assessment and early warning module, used to receive the anomaly feature data packets to construct a multimodal anomaly feature vector, and based on the multimodal anomaly feature vector, to assess the probability of thermal runaway risk in real time and construct... A digital twin space is built, and thermal runaway risk classification protection for charging piles is implemented based on the probability of thermal runaway risk. A response execution feedback module is used to comprehensively analyze multi-modal thermal runaway perception data before and after thermal runaway risk classification protection, quantify the response effect of thermal runaway risk classification protection, and adjust thermal runaway risk classification protection based on the response effect. A collaboration and operation and maintenance linkage module is used to visualize and monitor the entire process of thermal runaway risk through real-time linkage and data analysis of multiple platforms, and optimize the parameters of each algorithm and thermal runaway risk classification protection strategy.

[0021] Beneficial effects

[0022] The present invention has the following beneficial effects:

[0023] (1) This invention achieves comprehensive and real-time monitoring of the thermal runaway state of charging piles by multimodal sensing fusion of infrared, vibration, gas, image and environmental parameters, which significantly improves the sensitivity and accuracy of anomaly detection and effectively reduces the risk of false alarms and missed alarms in a single channel.

[0024] (2) This invention constructs a thermal runaway risk probability algorithm that dynamically fuses multimodal features, quantifies thermal runaway risk in real time at the edge computing layer, and drives graded protection response with risk probability as the core, ensuring that abnormal events can be quickly graded and efficiently responded to, thereby improving the intelligence and targeting of protection.

[0025] (3) This invention constructs a thermal runaway self-healing feedback value by dynamically comparing multimodal data before and after the implementation of protective measures, thereby realizing the quantitative evaluation and continuous optimization of the effectiveness of protective measures, ensuring that each response is guided by actual results, and promoting the formation of an intelligent closed loop of monitoring, response, feedback, and re-optimization.

[0026] (4) This invention, through intelligent linkage and data sharing with multiple platforms such as BMS, power plant operation and maintenance, and fire protection system, combined with the self-learning mechanism of Bayesian optimization and genetic algorithm, supports dynamic parameter optimization and new mode adaptation, and promotes the continuous improvement of intelligent operation and maintenance and safety capabilities throughout the entire life cycle.

[0027] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0028] Figure 1 This is a flowchart of an intelligent protection method for thermal runaway of charging piles based on edge computing.

[0029] Figure 2 This is a structural diagram of a smart protection system for thermal runaway of charging piles based on edge computing.

[0030] Figure 3 This is a flowchart of multimodal data acquisition and preprocessing.

[0031] Figure 4 This is a graph showing the changing trends of multi-channel characteristic change rates and multi-modal thermal runaway risk probability.

[0032] Figure 5 This is a flowchart of the edge computing layer workflow. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figures 1-5 This invention provides a technical solution: a method and system for intelligent protection against thermal runaway in charging piles based on edge computing, such as... Figure 1 As shown, the process includes the following steps: S1, real-time acquisition of thermal runaway multimodal sensing data, and data preprocessing of the thermal runaway multimodal sensing data; real-time identification of the thermal runaway multimodal abnormal state of the charging pile based on the preprocessed thermal runaway multimodal sensing data, and generation of abnormal feature data packets; S2, receiving the abnormal feature data packets to construct a multimodal abnormal feature vector, real-time assessment of the thermal runaway risk probability based on the multimodal abnormal feature vector, and construction of a digital twin space, while performing thermal runaway risk classification protection for the charging pile based on the thermal runaway risk probability; S3, comprehensive analysis of thermal runaway multimodal sensing data before and after thermal runaway risk classification protection, quantification of the response effect of thermal runaway risk classification protection, and adjustment of thermal runaway risk classification protection based on the response effect; S4, through real-time linkage and data analysis of multiple platforms, visualization monitoring and feedback of the entire thermal runaway risk process, and optimization of various algorithm parameters and thermal runaway risk classification protection strategies.

[0035] Specifically, the real-time acquisition of multimodal thermal runaway sensing data and the data preprocessing process are as follows: Multimodal thermal runaway sensing data is acquired in real-time using infrared and visible light dual-camera modules, anti-fog lenses, piezoelectric accelerometer vibration sensors, a five-in-one electrochemical gas sensor, and a temperature and humidity sensor deployed at key locations on the charging pile. This multimodal thermal runaway sensing data includes: infrared radiation signals from the charging pile, images of the equipment surface, vibration signals, concentration values ​​of thermal runaway characteristic gases, and ambient temperature and humidity data. The infrared and visible light dual-camera modules are used to simultaneously acquire the surface temperature distribution and visible image information of the charging pile, improving the spatial resolution of anomaly detection. The anti-fog lens effectively reduces interference from the external environment on image quality, ensuring data reliability. The piezoelectric accelerometer vibration sensor is responsible for highly sensitively acquiring minute mechanical vibration changes during equipment operation, promptly reflecting potential fault hazards. The five-in-one electrochemical gas sensor can simultaneously monitor multiple typical thermal runaway-related gases, including CO, etc. HF and Combined with temperature and humidity sensors, environmental compensation for gas concentration detection is achieved. Thermal runaway multimodal sensing data is aligned with a unified RTC timestamp to ensure accurate synchronization of data from all different sources, providing a time reference for subsequent fusion analysis and multimodal feature extraction. Utilizing a blackbody calibration algorithm and temperature mapping relationship, the accuracy of infrared temperature measurement is significantly improved. By comparing with a standard blackbody source, the response deviation of the infrared temperature measurement system is corrected, achieving precise mapping between physical temperature and infrared signals. The charging pile's infrared radiation signal is converted into physical temperature values ​​in real time to obtain the charging pile's infrared temperature value. Downsampling of equipment surface images reduces pixels. Downsampling not only reduces data volume and backend computational pressure but also enables rapid extraction of image features and coarse screening of abnormal contours, facilitating subsequent intelligent analysis. High-pass filters and instrument amplification processing technology are used to filter out low-frequency interference in vibration signals, retaining high-frequency vibration information. The vibration signal is converted into a digital signal via ADC conversion, and the short-time root mean square (RMS) value of the high-frequency vibration signal is extracted. The high-pass filter effectively suppresses low-frequency noise such as power frequency and temperature drift, improving the signal-to-noise ratio of high-frequency components in the vibration signal. Instrument amplification enhances the amplitude of weak vibration signals, ensuring the accuracy and dynamic range of subsequent A / D conversion. The short-time root mean square (RMS) value reflects the energy of the vibration signal within a specific time window, suitable for early micro-fault identification. A moving average algorithm is used to smooth fluctuations in the thermal runaway characterization gas concentration values, eliminating noise interference. Support vector regression is then used to correct the thermal runaway characterization gas concentration values ​​based on ambient temperature and humidity. The thermal runaway multimodal sensing data is normalized; the support vector regression algorithm, a machine learning regression method, establishes a multivariate mapping relationship between gas concentration and ambient temperature and humidity, dynamically compensating for the drift effect of temperature and humidity on the gas sensor output, ensuring data reliability. A charging pile thermal runaway protection database is established, storing the original and pre-processed thermal runaway multimodal sensing data, charging pile infrared temperature values, and high-frequency vibration signal RMS values.

[0036] like Figure 3 The diagram shows the multimodal data acquisition and preprocessing flowchart. It illustrates the main process of thermal runaway multimodal sensing data acquisition and preprocessing. By deploying infrared, vibration, gas, and temperature / humidity multimodal sensors at key locations in the charging pile, real-time thermal runaway multimodal sensing data, including infrared radiation, equipment images, vibration signals, gas concentration, and environmental parameters, is acquired. Intelligent preprocessing, including unified timestamp alignment, infrared signal physical temperature calibration, image denoising, vibration filtering and RMS extraction, gas concentration moving average denoising, and temperature / humidity compensation, achieves high precision, standardization, and environmental adaptive correction of the multimodal sensing signals. Simultaneously, thermal runaway anomalies are identified based on the preprocessed thermal runaway multimodal sensing data, which is then simultaneously stored in the database and input into the edge computing layer, laying a solid data foundation for subsequent intelligent thermal runaway identification and graded response.

[0037] In this implementation scheme, multi-source sensors are deployed at multiple points to collect multi-modal data in real time, including infrared, image, vibration, gas, temperature, and humidity data. Combined with preprocessing techniques such as time alignment, intelligent filtering, physical mapping, and algorithm correction, the accuracy of abnormal signal identification and the stability of data processing are effectively improved. Finally, a standardized database is established, providing high-quality data support for subsequent risk assessment and tiered protection, enhancing the sensitivity and reliability of charging pile thermal runaway monitoring.

[0038] Specifically, the process of real-time identification of multimodal thermal runaway anomalies in charging piles based on preprocessed multimodal thermal runaway sensing data is as follows: Using a sliding time window, the mean and standard deviation of the charging pile's infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterizing gas concentration value are calculated respectively. This yields the mean and standard deviation of the charging pile's infrared temperature, the mean and standard deviation of the high-frequency vibration signal RMS value, and the mean and standard deviation of the thermal runaway characterizing gas concentration. The sliding time window can be flexibly set according to actual monitoring needs to dynamically capture data change trends during the charging pile's operation, improving the timeliness and sensitivity of anomaly detection. The mean reflects the average level of each characteristic over a period of time, while the standard deviation measures the data fluctuation amplitude and is an important reference indicator for anomaly identification. The infrared temperature anomaly value is obtained by dividing the absolute value of the difference between the current infrared temperature value of the charging pile and the average infrared temperature value of the charging pile by the standard deviation of the infrared temperature value of the charging pile; the vibration signal anomaly value is obtained by dividing the absolute value of the difference between the current RMS value of the high-frequency vibration signal and the average RMS value of the high-frequency vibration signal by the standard deviation of the RMS value of the high-frequency vibration signal; the thermal runaway gas anomaly value is obtained by dividing the absolute value of the difference between the current thermal runaway gas concentration value and the average thermal runaway gas concentration by the standard deviation of the thermal runaway gas concentration. The degree of deviation of the characteristic value from the normal state at each moment is judged by the standardized residuals, which can effectively adapt to multi-source data under different operating conditions and dimensions, and reduce the risk of false alarms and false negatives caused by a single absolute threshold. The infrared temperature anomaly value, vibration signal anomaly value, thermal runaway gas anomaly value are summed with a constant, and the summation result is subjected to natural logarithmic operation to obtain the multimodal anomaly detection value. The introduction of the constant ensures the mathematical rationality of the subsequent logarithmic operation and avoids taking the logarithm of zero, while the natural logarithm can effectively compress the influence of extreme anomalies, making the multimodal anomaly detection value smoother and more suitable for threshold determination. The multimodal anomaly detection value integrates the anomaly characteristics of multiple physical quantities, reflecting the overall anomaly intensity of the charging pile in terms of thermal runaway risk dimension.

[0039] The specific formula for the multimodal anomaly detection value is as follows:

[0040] ;

[0041] In the formula, This multimodal anomaly detection value is used for real-time quantification and discrimination of multimodal anomalies in the thermal runaway state of charging piles. By normalizing the real-time monitoring data of three key channels—infrared temperature, equipment vibration, and the sum of typical thermal runaway characteristic gases—a multimodal anomaly detection value is calculated that can comprehensively reflect the current degree of anomaly. The higher the multimodal anomaly detection value, the more abnormal fluctuations have occurred simultaneously in multiple channels, and the greater the risk of thermal runaway of the charging pile. It serves as an important basis for early warning and response. The natural logarithm operation can smooth and dynamically compress anomaly values, enhance the sensitivity to small anomalies, and suppress the explosiveness of large anomalies, thereby improving the overall reliability, interpretability, and engineering practicality of the discrimination. This indicates the current infrared temperature value of the charging station; This represents the average infrared temperature of the charging pile. This indicates the standard deviation of the infrared temperature of the charging pile; It represents an abnormal value of infrared temperature, measures the degree of deviation of the current infrared temperature from the recent average level, and reflects the sensitivity of the temperature signal to sudden changes after being divided by the standard deviation. It is used to identify overheating and signs of thermal runaway in key parts. This indicates the current RMS value of the high-frequency vibration signal; This represents the RMS mean value of a high-frequency vibration signal; This represents the RMS standard deviation of the high-frequency vibration signal; It represents abnormal values ​​of vibration signals, measures the normalized deviation of the current equipment vibration RMS from the sliding window mean, reflects high-frequency vibration fluctuations under abnormal operating conditions of the equipment, and is used to identify precursors of thermal runaway. This indicates the current concentration of the gas characteristic of thermal runaway; This represents the average concentration of the gas characteristic of thermal runaway; This represents the standard deviation of the gas concentration characterizing thermal runaway; This indicates anomalies in the characteristic gases of thermal runaway, measuring the normalized deviation of the total concentration of all typical thermal runaway characteristic gases from the historical mean. It reflects the sudden release of gases caused by battery failure, decomposition, and abnormal reactions, including thermal runaway decomposition gases such as CO and HF. , and It has extremely high sensitivity.

[0042] In this implementation plan, dynamic statistical analysis of infrared temperature, vibration signal RMS, and gas concentration multimodal signals is performed within a sliding time window. Standardization processing is used to achieve dimensionless normalization comparison of different abnormal characteristics. The natural logarithm is used to comprehensively quantify multi-source anomalies, which improves the sensitivity and accuracy of identifying abnormal states of thermal runaway in charging piles and lays a solid data foundation for subsequent risk assessment and early warning.

[0043] Specifically, the process of generating anomaly feature data packets is as follows: Multimodal anomaly detection values ​​are stored in real-time in the charging pile thermal runaway protection database and compared with anomaly thresholds. When the multimodal anomaly detection value is less than the anomaly threshold, the thermal runaway multimodal sensing data is determined to be within the normal fluctuation range, and only routine monitoring and storage of the multimodal anomaly detection values ​​are performed. The normal fluctuation range refers to the normal range of various sensing data under natural operating conditions. Relevant data is recorded and archived for subsequent trend analysis and anomaly backtracking. When the multimodal anomaly detection value is greater than or equal to the anomaly threshold, the current period's thermal runaway multimodal sensing data, infrared temperature anomalies, vibration signal anomalies, thermal runaway characterizing gas anomalies, and multimodal anomaly detection values ​​are marked, anomaly feature data packets are constructed, and input to the edge computing layer. The anomaly feature data packets contain thermal runaway multimodal sensing data, infrared temperature anomalies, vibration signal anomalies, thermal runaway characterizing gas anomalies, multimodal anomaly detection values, and time tags, achieving a structured description of the entire abnormal event process, facilitating subsequent intelligent analysis, edge early warning, and closed-loop control. Edge computing refers to hardware platforms with local processing capabilities that can analyze and respond to abnormal data packets in real time, reducing data transmission latency and improving response speed and reliability.

[0044] In this implementation scheme, by comparing multimodal anomaly detection values ​​with anomaly thresholds in real time, normal fluctuations and abnormal states can be distinguished. Data from abnormal periods is accurately labeled and structurally encapsulated, ensuring that anomaly information is captured and uploaded to the edge computing layer in a timely manner. This not only automates and improves anomaly monitoring but also provides detailed and traceable data support for subsequent intelligent risk assessment and response.

[0045] Specifically, the process of receiving abnormal feature data packets to construct a multimodal abnormal feature vector and assessing the probability of thermal runaway risk in real time based on the multimodal abnormal feature vector is as follows: The edge computing layer receives abnormal feature data packets, timestamps all thermal runaway multimodal sensing data, infrared temperature anomalies, vibration signal anomalies, thermal runaway characterization gas anomalies, and multimodal abnormal detection value sequences, and fuses them into a multimodal abnormal feature vector. The multimodal abnormal feature vector is formed by concatenating the abnormal features of various channels to form a high-dimensional feature set that can be used for risk assessment and classification. Based on the multimodal abnormal feature vector, within a sliding time window, the infrared temperature anomaly, vibration signal anomaly, and thermal runaway characterization gas concentration anomalies corresponding to the previous and current times for the three channels, including the infrared temperature channel, vibration signal channel, and thermal runaway characterization gas concentration channel, are extracted respectively. The system controls the characterization of gas anomalies. It calculates the change in anomalies for each channel by using the difference between the current and previous anomaly values, and divides this by the sliding time window length to obtain the characteristic change rate. The characteristic change rate reflects the rate of change of an anomaly characteristic in a given channel per unit time, serving as a crucial indicator for dynamic risk assessment. The characteristic change rate is squared to obtain the anomaly rate square term. The squared anomaly rate terms for the three channels are summed and averaged, with the negative value used as the exponent. A natural exponential operation is then performed to obtain the anomaly suppression exponent term. Squaring the characteristic change rate highlights the impact of abrupt anomaly signals on overall risk. This exponential normalization maps the dynamic fluctuation intensity of various anomaly signals to probability values ​​between 0 and 1. Subtracting the anomaly suppression exponent term from a constant yields the probability value of multimodal thermal runaway risk.

[0046] The specific formula for the probability value of multimodal thermal runaway risk is as follows:

[0047] ;

[0048] In the formula, It represents the probability value of multimodal thermal runaway risk, which is used to quantify the probability of thermal runaway risk at the current moment in real time. It dynamically integrates the dramatic change trend of all channel outliers over time, maps the short-term mutation activity to a risk probability value between 0 and 1, sensitively captures abnormal mutations, and reduces random noise interference. This indicates the number of channels, with a value of 3, including the infrared temperature channel, the vibration signal channel, and the thermal runaway characterization gas concentration channel; The outlier values ​​of the i-th channel include infrared temperature outliers, vibration signal outliers, and thermal runaway characterization gas outliers, reflecting the normalized deviation of each physical quantity from its historical mean. This represents the change in outliers in the i-th channel, capturing the change in outliers within a short period of time. Indicates the length of the sliding time window; This represents the characteristic change rate of the i-th channel, which measures whether the anomaly is changing rapidly and highlights dynamic sensitivity. The larger the value, the more violent the channel fluctuation. The square operation is performed to enhance the mutation, eliminate positive and negative cancellation, and focus on violent anomalies. This represents the anomaly suppression exponential term, which exponentially suppresses extreme anomalies. It is mapped to a probability range of 0 to 1, ensuring smooth output and controllable engineering thresholds.

[0049] In this embodiment, Table 1 is a data table of multimodal thermal runaway risk probability values. This indicates the amount of change in abnormal infrared temperature values. This indicates the change in the gas anomaly value characteristic of thermal runaway. The table represents the change in abnormal vibration signal values, with the sliding time window length set to 1 minute. It details the characteristic change rate and multimodal thermal runaway risk probability value for each channel at five different time points. Specifically, at time point t1, the characteristic change rate for the infrared temperature channel is 0.5, the characteristic change rate for the thermal runaway characterization gas concentration channel is 0.7, the characteristic change rate for the vibration signal channel is 0.6, and the multimodal thermal runaway risk probability value is 0.306; at time point t2, the characteristic change rate for the infrared temperature channel is 1.0, the characteristic change rate for the thermal runaway characterization gas concentration channel is 1.2, the characteristic change rate for the vibration signal channel is 1.1, and the multimodal thermal runaway risk probability value is 0.703; at time point t3, the characteristic change rate for the infrared temperature channel is 2.0, and the characteristic change rate for the thermal runaway characterization gas concentration channel is... The change rate of the infrared temperature channel at time point t4 was 1.8, the change rate of the vibration signal channel characteristics was 2.3, and the probability value of multimodal thermal runaway risk was 0.984. The change rate of the infrared temperature channel characteristics at time point t4 was 3.5, the change rate of the thermal runaway characterization gas concentration channel characteristics was 3.2, the change rate of the vibration signal channel characteristics was 2.8, and the probability value of multimodal thermal runaway risk was 0.999. The change rate of the infrared temperature channel characteristics at time point t5 was 4.2, the change rate of the thermal runaway characterization gas concentration channel characteristics was 4.0, the change rate of the vibration signal channel characteristics was 4.5, and the probability value of multimodal thermal runaway risk was 1.000.

[0050] Table 1. Data on the probability values ​​of multimodal thermal runaway risk

[0051]

[0052] like Figure 4 As shown, this is a graph illustrating the changing trends of multi-channel characteristic change rates and multi-modal thermal runaway risk probability. It displays the characteristic change rates of the infrared temperature channel, thermal runaway characterization gas concentration channel, and vibration signal channel monitored by the charging pile at five consecutive time points, as well as the changing trend of the multi-modal thermal runaway risk probability value calculated based on these three channels over time. (Based on Table 1 and...) Figure 4It can be seen that as time goes by, the characteristic change rates of all three types of channels show an obvious increasing trend, especially with a relatively large increase during the stage from t3 to t5. As the characteristic change rates of each channel gradually increase, the multi-modal thermal runaway risk probability value shows a rapid upward trend and approaches 1 after t3. When the characteristic change rates of all three channels reach a relatively high level, the multi-modal thermal runaway risk probability value quickly jumps from 0.7 to nearly 1, indicating a sensitive response of the thermal runaway risk probability to drastic changes in characteristics.

[0053] In this implementation scheme, by uniformly aligning and fusing multi-modal abnormal feature data at the edge computing layer, dynamically extracting the abnormal feature change rates of each channel, and calculating the multi-modal thermal runaway risk probability based on the change trend, the sensitive capture and real-time quantification of risk mutations are achieved. This process can effectively improve the real-time performance and accuracy of thermal runaway risk assessment, providing a scientific decision-making basis for hierarchical response and intelligent protection.

[0054] Specifically, the specific process of constructing a digital twin space and performing hierarchical thermal runaway protection for charging piles according to the multi-modal thermal runaway risk probability is as follows: Use the multi-modal thermal runaway perception data, three-dimensional structure and attributes of the charging pile to construct a digital twin space, that is, based on the structural parameters, working conditions attributes of the physical charging pile and the multi-modal thermal runaway perception data, a virtual three-dimensional simulation environment is constructed and maintained in real time, realizing a one-to-one mapping and synchronous display of the states between the physical device and the virtual model, providing support for intelligent monitoring and remote collaboration. Use the edge computing layer to map the multi-modal thermal runaway risk probability value to the digital twin space in real time for holographic synchronization of the charging pile state; at the same time, compare the multi-modal thermal runaway risk probability value with the multi-level risk thresholds P1 and P2 in real time, and implement hierarchical thermal runaway protection measures; when < P1, only perform routine monitoring and storage of the multi-modal thermal runaway risk probability value, and do not trigger active intervention to ensure efficient resource allocation; when ≥ P1 and < P2, reduce the charging power of the charging pile, start the fan and liquid cooling pump to enhance heat dissipation, and trigger intermittent audible and visual alarms; share the multi-modal thermal runaway perception data and multi-modal thermal runaway risk probability value among multiple charging piles based on wireless ad hoc network to identify group risks and abnormal diffusion trends; virtually reproduce the current anomaly in the digital twin space, evaluate the optimal combination of each response measure in the early warning scenario and push suggestions; among them, trigger the data interconnection and group joint defense mechanism of multiple charging piles through the linkage of local and wireless ad hoc network. The wireless ad hoc network ensures redundant communication under network failures and single-point failures, and maximally avoids abnormal signal islands and collaborative failures. The intermittent audible and visual alarm is started intermittently through a relay to remind on-site operation and maintenance personnel to pay attention. The digital twin space realizes the visual reproduction of the current abnormal working condition and pushes the optimal response suggestions based on the algorithm. When When the voltage level is ≥P2, the dual-redundant relays disconnect the charging circuit and isolate it; adjust the fresh air and gas exhaust circuits; based on the infrared radiation signal of the charging pile and the image of the equipment surface, use edge detection algorithms to identify changes in the shape of the battery compartment shell, simultaneously activate continuous audible and visual alarms and link with the fire protection system; and synchronously visualize the entire process in the digital twin space. The dual-redundant relays are high-security hardware protection units with hardware-level interlocking and physical circuit breaking functions, ensuring that if any relay fails, the backup relay can still forcibly disconnect the circuit. The relay actions and the fresh air and exhaust systems are uniformly scheduled by the edge computing layer to ensure sequential order and safety interlocking: the charging circuit disconnection action comes first, followed by the exhaust and fresh air activation action; all hardware actions require feedback closure signals for confirmation before proceeding to the next step. The continuous audible and visual alarms and the linked fire protection signals are interlocked with the fire control panel and the automatic sprinkler system to achieve integrated emergency linkage.

[0055] like Figure 5 The diagram shows the workflow of the edge computing layer. It illustrates that the edge computing layer first acquires real-time thermal runaway multimodal sensing data via multimodal sensors, including infrared, vibration, gas, and temperature / humidity sensors. The AI ​​chip controls the FPGA through a high-speed interface, offloading real-time tasks such as image downsampling and vibration signal filtering to the FPGA. The AI ​​chip focuses on feature fusion and decision-making. After standardization, filtering, and compensation of the raw thermal runaway multimodal sensing data, a sliding window is used for normalized anomaly detection and feature extraction, dynamically outputting anomaly feature data packets. The edge computing layer archives the data and performs real-time anomaly detection, constructing anomaly feature vectors. These vectors are used for thermal runaway risk probability assessment and drive multi-level protective response measures based on multi-level risk thresholds. After each response, the thermal runaway multimodal sensing data before and after is compared, quantifying the self-healing feedback value to optimize and dynamically adjust measures. This forms an intelligent closed-loop management process for charging pile thermal runaway risk, from monitoring, detection, response to feedback optimization.

[0056] This implementation plan constructs a digital twin space for charging piles, achieving real-time synchronization and holographic visualization of multimodal risk probability values ​​and three-dimensional device status. It also incorporates tiered protection measures—dynamically triggering charging power adjustment, enhanced heat dissipation, audible and visual alarms, and power outage isolation—based on multi-level risk thresholds. This not only enables precise tiering and coordinated protection of the risk status of single and multiple charging piles but also allows for the virtual reenactment of abnormal scenarios within the digital twin environment. Simultaneously, it ensures the correct timing of risk responses and the security of hardware interlocks, optimizing response strategies and decision support. This significantly improves the intelligence, collaborative capabilities, and traceability of the response loop in thermal runaway protection.

[0057] Specifically, the process of quantifying the response effect of graded thermal runaway risk protection by comprehensively analyzing multimodal sensing data of thermal runaway risk before and after graded thermal runaway risk protection is as follows: After the graded thermal runaway protection measures are implemented, based on a sliding time window, the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal, and the concentration value of the thermal runaway characterizing gas within the current time window are statistically analyzed. The changes in the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal, and the concentration value of the thermal runaway characterizing gas after the response are calculated by the difference. These represent the core characteristics of the multidimensional anomalies of the charging pile in terms of heat, vibration, and gas release, covering the main physical quantities in the thermal runaway response. Simultaneously, the infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterization gas concentration value of the charging pile before the implementation of graded thermal runaway protection measures were extracted from the charging pile thermal runaway protection database. Based on a sliding time window, the changes in the infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterization gas concentration value before the response were calculated respectively. After the response, the changes in the infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterization gas concentration value were squared, summed, and the square root was taken to obtain the comprehensive fluctuation intensity value after the response. The comprehensive fluctuation intensity value before the response was obtained by squared, summing, and taking the square root. The comprehensive fluctuation intensity value after the response comprehensively measures the total abnormal fluctuation intensity of all channels within the current window period, highlighting the multi-channel synergistic improvement effect. The comprehensive fluctuation intensity value before the response serves as a benchmark before implementation to compare whether the self-healing effect is truly significant. By mapping the fluctuations from different channels to the same index system, the effective fusion and comprehensive quantification of multi-source characteristics are achieved. The thermal runaway self-healing feedback value is obtained by subtracting the ratio of the comprehensive fluctuation intensity value after the response to the comprehensive fluctuation intensity value before the response from a constant. This value is used to quantitatively measure the effect of graded protection measures on improving thermal runaway risk indicators. The larger the value, the more significantly the abnormal fluctuations are suppressed after the response. Conversely, if the thermal runaway self-healing feedback value remains low, it indicates that the measures are ineffective.

[0058] The specific formula for the thermal runaway self-healing feedback value is as follows:

[0059] ;

[0060] In the formula, This represents the thermal runaway self-healing feedback value, used to quantify the overall multimodal improvement effect after the implementation of graded thermal runaway protection measures. It reflects in real time whether the response action has effectively reduced the fluctuation level of multimodal abnormal signals such as temperature, gas leakage, and vibration. The improvement degree is expressed by subtracting the ratio from the constant. The closer to 1, the more significantly the abnormality is suppressed and the better the response measures are; below 0 or close to 0, it indicates poor effect and the need for optimization and upgrading of measures. This indicates the change in the infrared temperature value of the charging pile after the response, reflecting the improvement of the infrared temperature channel within the window period after the response is executed; It represents the change in the concentration of the thermal runaway characterizing gas after the response, assesses the suppression level after the response of gas leakage and decomposition risk gas signals, and directly reflects the airtightness and self-healing ability of abnormal reactions. It represents the change in the RMS value of the high-frequency vibration signal after the response, and monitors whether the physical fluctuations of mechanical structure, battery pack interior and thermal stress are significantly improved after the response. This indicates the change in the infrared temperature value of the charging pile before the response. This indicates the change in the concentration of the gas characterizing thermal runaway before the response; This indicates the change in the RMS value of the high-frequency vibration signal before the response.

[0061] In this implementation plan, the dynamic comparison of multimodal sensing data before and after the implementation of graded protection measures, using quantitative calculations of comprehensive fluctuation intensity and self-healing feedback values, can accurately assess the actual effectiveness of the protection measures and intuitively reflect the improvement after risk response. This not only provides a scientific basis for subsequent response adjustments and adaptive optimization, but also ensures that every thermal runaway protection action is data-driven and effect-oriented, continuously improving the closed-loop capability and intelligence level of thermal runaway risk management.

[0062] Specifically, the process of adjusting the thermal runaway risk classification protection based on the response effect is as follows: After each graded thermal runaway protection measure is implemented, the thermal runaway self-healing feedback value is continuously monitored, written into the charging pile thermal runaway protection database, and compared with the self-healing threshold; when the thermal runaway self-healing feedback value is lower than the self-healing threshold and there is no upward trend within the sliding time window, the strength of each thermal runaway protection measure is increased, and priority is assigned to each thermal runaway protection measure; for example, the air-cooled and liquid-cooled pump speed can be gradually increased, the audible and visual alarm time can be extended, and the response level of the circuit breaker isolation action can be improved; the priority ranking can be dynamically adjusted in combination with the distribution of abnormal values ​​in each channel, the equipment operating status, and historical thermal runaway self-healing feedback values ​​to ensure that the most effective protection measures are implemented first. Record each adjustment of thermal runaway protection measures and the corresponding thermal runaway self-healing feedback value to form a complete closed loop of measure adjustment and effect feedback, achieving process traceability and intelligent optimization. When the duration of the thermal runaway self-healing feedback value being lower than the self-healing threshold exceeds the maximum allowable threshold, a thermal runaway self-healing failure alarm is pushed, and a detailed response process and thermal runaway self-healing feedback value trend are generated and pushed to the fire protection system. The maximum allowable threshold is a protection parameter to prevent safety hazards caused by the failure to effectively suppress abnormalities for a long time. Once it is continuously ineffective, an alarm is automatically triggered, and the fire protection system is activated to ensure multiple protections. When the thermal runaway self-healing feedback value is greater than or equal to the self-healing threshold, it is determined that the thermal runaway protection measures have achieved substantial effect, the alarm state is lifted, and the equipment is gradually restored to the normal operating mode, ensuring a safe and orderly return to normal working state after the risk is eliminated, preventing false recovery and secondary failures.

[0063] This implementation plan continuously monitors and dynamically compares the self-healing feedback value, enabling the strengthening of measures and priority adjustments when the protection effect is insufficient. It also promptly pushes alarms and triggers the fire suppression system when the protection is ineffective for extended periods, ensuring closed-loop adaptive optimization of risk response. Once the self-healing feedback value reaches the self-healing threshold, the alarm is deactivated, and normal operation resumes. This achieves intelligent optimization of the entire thermal runaway protection process and timely closed-loop handling of anomalies, significantly improving the reliability and safety assurance capabilities of charging pile thermal runaway risk management.

[0064] Specifically, the process of visually monitoring and providing feedback on the entire process of thermal runaway risk through real-time linkage and data analysis across multiple platforms, and optimizing various algorithm parameters and thermal runaway risk classification and protection strategies, involves: connecting the edge computing layer with the BMS, power plant operation and maintenance, and fire protection systems to monitor multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values ​​in real time, and conducting remote data sharing and big data analysis; where the BMS is the battery management system, enabling interconnection and interoperability with the charging pile's real-time operating conditions, status, and fault information. Through multi-platform interface protocols, real-time synchronization and complementarity of monitoring information, alarm events, and historical data are achieved, improving the comprehensiveness and timeliness of anomaly detection and collaborative response. The parameters of multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values, as well as anomaly thresholds, multi-level risk thresholds, and self-healing thresholds, are periodically adjusted using historical data from the charging pile thermal runaway protection database, employing Bayesian optimization and genetic algorithms. Specifically, with a weekly update cycle, historical operating condition data from the recent period is periodically extracted from the thermal runaway protection database, including the actual performance of multimodal anomaly detection values, thermal runaway risk probability values, thermal runaway self-healing feedback values, and corresponding graded threshold parameters. The edge computing layer calls the integrated Bayesian optimization and genetic algorithm joint module to perform global sampling and intelligent search of the parameter space. An optimization objective function is constructed based on the false alarm rate, response latency, and handling success rate, where response latency is the average response time of the system's graded handling, and the handling success rate is the ratio of the actual elimination of anomalies by graded protection measures. Finally, the recent trend of thermal runaway self-healing feedback values ​​is integrated to jointly optimize and output the anomaly threshold, multi-level risk threshold, and self-healing threshold parameters. The new parameters output by the optimization algorithm are first simulated and backtested in the digital twin space. Cross-validation and virtual reenactment of abnormal scenarios using historical data are conducted to ensure that the optimal parameter combination outperforms existing settings under different behavioral modes. If validation is successful, the new parameters are distributed to the edge computing layer of each charging pile through a secure authorization mechanism. Old configurations and optimization process metadata are automatically backed up during parameter switching, facilitating anomaly tracing and rapid rollback. Subsequently, all newly collected data and response results during the operation of all charging piles are fed back to the charging pile thermal runaway protection database in real time and included in the next optimization cycle, achieving continuous learning and adaptive upgrades. The system also integrates actual thermal runaway self-healing effects to learn new abnormal patterns and optimize thermal runaway protection measures. Thermal runaway monitoring and feedback reports are regularly output, and the entire closed-loop response process of multimodal thermal runaway perception data and graded thermal runaway protection measures is displayed in the digital twin space. This not only supports static reporting and real-time monitoring visualization but also allows for dynamic replay of thermal runaway anomaly events and graded response actions.

[0065] This implementation plan achieves real-time linkage and data sharing with multiple platforms including BMS, power station operation and maintenance, and fire protection systems. It enables full-process, visualized monitoring of multimodal anomalies, thermal runaway risk probability, and self-healing feedback of charging piles. Based on big data analysis and intelligent optimization algorithms, it dynamically adjusts key parameters and protection strategies, continuously adapting to and improving the thermal runaway risk management model. Regular monitoring reports are generated, and the entire response closed-loop process is simultaneously displayed in a digital twin space, significantly enhancing the intelligence, collaboration, and continuous self-evolution capabilities of charging pile thermal runaway protection.

[0066] Reference Figure 2 As shown, the second aspect of the present invention provides an intelligent protection system for thermal runaway of charging piles based on edge computing, applied to the aforementioned intelligent protection method for thermal runaway of charging piles based on edge computing, comprising: a multimodal perception and anomaly discrimination module, used to collect multimodal perception data of thermal runaway in real time, and preprocess the multimodal perception data of thermal runaway; based on the preprocessed multimodal perception data of thermal runaway, to judge the multimodal abnormal state of thermal runaway of the charging pile in real time, and generate anomaly feature data packets; and an edge risk assessment and early warning module, used to receive the anomaly feature data packets to construct a multimodal anomaly feature vector, and to determine the multimodal anomaly feature vector based on the multimodal anomaly feature vector. The system performs real-time assessment of the probability of thermal runaway risk and constructs a digital twin space. It also implements graded protection for thermal runaway risk of charging piles based on this probability. A response execution feedback module comprehensively analyzes multi-modal thermal runaway perception data before and after graded protection, quantifies the response effect of graded protection, and adjusts the protection measures based on the response effect. A collaboration and operation linkage module visualizes and monitors the entire process of thermal runaway risk through real-time linkage and data analysis across multiple platforms, and optimizes algorithm parameters and graded protection strategies.

[0067] This implementation plan achieves efficient, real-time, and intelligent closed-loop management of charging pile thermal runaway risk through the collaborative work of multiple modules, including multimodal perception and anomaly detection, edge risk assessment and early warning, response execution feedback, and coordination and operation and maintenance linkage. It not only sensitively acquires and identifies multi-source abnormal signals and dynamically quantifies and hierarchically manages thermal runaway risk, but also continuously and adaptively optimizes protective measures based on response results. Furthermore, relying on multi-platform data sharing and intelligent analysis, it achieves visualized monitoring and strategy evolution throughout the entire process, significantly improving the safety, reliability, and intelligent operation and maintenance level of charging piles.

[0068] It should be noted that, in this document, 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. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A smart protection method for thermal runaway of charging piles based on edge computing, characterized in that, Includes the following steps: S1: Real-time acquisition of thermal runaway multimodal sensing data; preprocessing of thermal runaway multimodal sensing data; real-time identification of abnormal thermal runaway multimodal states of charging piles based on preprocessed thermal runaway multimodal sensing data; and generation of abnormal feature data packets. S2, receive abnormal feature data packets to construct multimodal abnormal feature vectors, assess the probability of thermal runaway risk in real time based on the multimodal abnormal feature vectors, construct a digital twin space, and perform graded protection of thermal runaway risk of charging piles based on the probability of thermal runaway risk. S3, comprehensively analyze the multimodal thermal runaway sensing data before and after thermal runaway risk classification protection, quantify the response effect of thermal runaway risk classification protection, and adjust the thermal runaway risk classification protection according to the response effect; S4 enables visualized monitoring and feedback of the entire process of thermal runaway risk through real-time linkage and data analysis across multiple platforms, and optimizes various algorithm parameters and thermal runaway risk classification and protection strategies.

2. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process for real-time acquisition of thermal runaway multimodal sensing data and data preprocessing of the thermal runaway multimodal sensing data is as follows: By deploying infrared and visible light dual-camera modules, anti-fog lenses, piezoelectric acceleration vibration sensors, five-in-one electrochemical gas sensors, and temperature and humidity sensors in key parts of the charging pile, multi-modal thermal runaway sensing data is collected in real time. The multi-modal thermal runaway sensing data includes: infrared radiation signals of the charging pile, images of the equipment surface, vibration signals, concentration values ​​of thermal runaway characterizing gases, and ambient temperature and humidity data. The thermal runaway multimodal sensing data is aligned using a unified RTC timestamp; the charging pile infrared radiation signal is converted into physical temperature value in real time using a blackbody calibration algorithm and temperature mapping relationship to obtain the charging pile infrared temperature value; the equipment surface image is downsampled to reduce pixels; low-frequency interference in the vibration signal is filtered out using high-pass filter and instrument amplification technology to retain high-frequency vibration information, and the vibration signal is converted into a digital signal through ADC conversion, and the short-time root mean square value of the high-frequency vibration signal is extracted; the thermal runaway characterization gas concentration value is smoothed by a moving average algorithm to eliminate noise interference, and the thermal runaway characterization gas concentration value is corrected by a support vector regression algorithm according to the ambient temperature and humidity; the thermal runaway multimodal sensing data is normalized; a charging pile thermal runaway protection database is established, and the original and preprocessed thermal runaway multimodal sensing data, charging pile infrared temperature value and high-frequency vibration signal RMS value are stored in the charging pile thermal runaway protection database.

3. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process for real-time determination of the multimodal abnormal state of thermal runaway in charging piles based on preprocessed multimodal thermal runaway sensing data is as follows: Based on the sliding time window, the mean and standard deviation of the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal, and the concentration value of the thermal runaway characterizing gas are calculated respectively, so as to obtain the mean value of the infrared temperature of the charging pile, the standard deviation of the infrared temperature of the charging pile, the mean value of the RMS value of the high-frequency vibration signal, the standard deviation of the RMS value of the high-frequency vibration signal, the mean value of the thermal runaway characterizing gas concentration, and the standard deviation of the thermal runaway characterizing gas concentration. The infrared temperature anomaly value is obtained by dividing the absolute value of the difference between the current infrared temperature value of the charging pile and the average infrared temperature value of the charging pile by the standard deviation of the infrared temperature value of the charging pile; the vibration signal anomaly value is obtained by dividing the absolute value of the difference between the current RMS value of the high-frequency vibration signal and the average RMS value of the high-frequency vibration signal by the standard deviation of the RMS value of the high-frequency vibration signal; the thermal runaway gas anomaly value is obtained by dividing the absolute value of the difference between the current thermal runaway gas concentration value and the average thermal runaway gas concentration by the standard deviation of the thermal runaway gas concentration. The multimodal anomaly detection values ​​are obtained by summing the infrared temperature anomaly values, vibration signal anomaly values, thermal runaway characterization gas anomaly values ​​with a constant, and then performing a natural logarithmic operation on the summation result.

4. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process for generating the abnormal feature data packet is as follows: Multimodal anomaly detection values ​​are stored in the charging pile thermal runaway protection database in real time and compared with anomaly thresholds. When the multimodal anomaly detection value is less than the anomaly threshold, the thermal runaway multimodal sensing data is determined to be in the normal fluctuation range, and only routine monitoring and storage of the multimodal anomaly detection value are performed. When the multimodal anomaly detection value is greater than or equal to the anomaly threshold, the current period's thermal runaway multimodal sensing data, infrared temperature anomaly value, vibration signal anomaly value, thermal runaway characterization gas anomaly value, and multimodal anomaly detection value are marked, anomaly feature data package is constructed, and input to the edge computing layer.

5. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process of constructing a multimodal anomaly feature vector from the received anomaly feature data packet, and then assessing the probability of thermal runaway risk in real time based on the multimodal anomaly feature vector, is as follows: The edge computing layer receives abnormal feature data packets, timestamps all thermal runaway multimodal sensing data, infrared temperature anomalies, vibration signal anomalies, thermal runaway characterization gas anomalies, and multimodal anomaly detection value sequences, and fuses them into a multimodal anomaly feature vector; Based on the multimodal anomaly feature vector, the infrared temperature anomaly value, vibration signal anomaly value, and thermal runaway characterization gas anomaly value are extracted from three channels (infrared temperature channel, vibration signal channel, and thermal runaway characterization gas concentration channel) in the sliding time window, respectively, at the previous and current times. The change in the outlier value for each channel is calculated by using the difference between the outlier value at the current time and the outlier value at the previous time. The change is then divided by the length of the sliding time window to obtain the characteristic change rate. The characteristic change rate is squared to obtain the squared term of the outlier rate. The squared terms of the outlier rates for the three channels are summed and averaged. The negative value is taken as the exponent and the natural exponent is used to obtain the outlier suppression exponent. The outlier suppression exponent is subtracted from the constant to obtain the probability value of the multimodal thermal runaway risk.

6. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process of constructing a digital twin space and simultaneously implementing graded protection against thermal runaway risk of charging piles based on the probability of thermal runaway risk is as follows: A digital twin space is constructed by using multimodal thermal runaway sensing data and the three-dimensional structure and attributes of the charging pile. The multimodal thermal runaway risk probability value is mapped to the digital twin space in real time using the edge computing layer to achieve holographic synchronization of the charging pile status. Simultaneously compare the probability values ​​of multimodal thermal runaway risk in real time. With multi-level risk thresholds P1 and P2, and implement graded thermal runaway protection measures; When <When P1, only the multi-modal thermal runaway risk probability value is routinely monitored and stored; When ≥ P1 and < P2, reduce the charging power of the charging pile, start the fan and liquid cooling pump to enhance heat dissipation, and trigger intermittent audible and visual alarms; share multi-modal perception data of thermal runaway and multi-modal thermal runaway risk probability values based on wireless ad hoc network among multiple charging piles to identify group risks and abnormal diffusion trends; The current anomaly is virtually reenacted in the digital twin space to assess the optimal combination of response measures under the early warning scenario and to push recommendations. when When the value is ≥P2, the dual redundant relays cut off the charging circuit and isolate it; adjust the fresh air and gas exhaust circuits; and use the edge detection algorithm to identify the shape changes of the battery compartment shell based on the infrared radiation signal of the charging pile and the image of the equipment surface, and simultaneously activate the continuous audible and visual alarm and link the fire protection system. Synchronously visualize the entire response process in the digital twin space.

7. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process for comprehensively analyzing multimodal thermal runaway sensing data before and after thermal runaway risk classification protection, and quantifying the response effect of thermal runaway risk classification protection, is as follows: After the graded thermal runaway protection measures are implemented, based on the sliding time window, the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal and the concentration value of the thermal runaway characterizing gas within the current time window are statistically analyzed, and the changes in the infrared temperature value of the charging pile, the RMS value of the high-frequency vibration signal and the concentration value of the thermal runaway characterizing gas after the response are calculated by the difference. Simultaneously, the infrared temperature value, high-frequency vibration signal RMS value, and thermal runaway characterization gas concentration value of the charging pile before the implementation of graded thermal runaway protection measures are extracted from the charging pile thermal runaway protection database. Based on the sliding time window, the changes in the infrared temperature value of the charging pile before the response, the changes in the high-frequency vibration signal RMS value before the response, and the changes in the thermal runaway characterization gas concentration value before the response are calculated respectively. The changes in the infrared temperature of the charging pile after the response, the changes in the RMS value of the high-frequency vibration signal after the response, and the changes in the concentration of the gas indicating thermal runaway after the response are squared, summed, and then the square root is taken to obtain the comprehensive fluctuation intensity value after the response; the changes in the infrared temperature of the charging pile before the response, the changes in the RMS value of the high-frequency vibration signal before the response, and the changes in the concentration of the gas indicating thermal runaway before the response are squared, summed, and then the square root is taken to obtain the comprehensive fluctuation intensity value before the response. The thermal runaway self-healing feedback value is obtained by subtracting the ratio of the comprehensive fluctuation intensity value after the response to the comprehensive fluctuation intensity value before the response from a constant.

8. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process for adjusting the thermal runaway risk classification and protection based on the response effect is as follows: After each graded thermal runaway protection measure is implemented, the thermal runaway self-healing feedback value is continuously monitored, written into the charging pile thermal runaway protection database, and compared with the self-healing threshold. When the thermal runaway self-healing feedback value is lower than the self-healing threshold and shows no upward trend within the sliding time window, the strength of each thermal runaway protection measure is increased, and priority is assigned to each thermal runaway protection measure; each adjustment of the thermal runaway protection measure and the corresponding thermal runaway self-healing feedback value are recorded; when the duration of the thermal runaway self-healing feedback value being lower than the self-healing threshold exceeds the maximum allowable threshold, a thermal runaway self-healing failure alarm is pushed, and a detailed response process and thermal runaway self-healing feedback value trend are generated and pushed to the fire protection system in conjunction with the response. When the thermal runaway self-healing feedback value is greater than or equal to the self-healing threshold, it is determined that the thermal runaway protection measures have achieved substantial effect, the alarm status is lifted, and the equipment is gradually restored to normal operating mode.

9. The intelligent protection method for thermal runaway of charging piles based on edge computing according to claim 1, characterized in that, The specific process of visually monitoring and providing feedback on the entire process of thermal runaway risk through real-time linkage and data analysis across multiple platforms, and optimizing various algorithm parameters and thermal runaway risk classification and protection strategies, is as follows: The edge computing layer is integrated with the BMS, power plant operation and maintenance, and fire protection systems to monitor multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values ​​in real time, enabling remote data sharing and big data analysis. Historical data from the charging pile thermal runaway protection database is periodically used to adjust the parameters of multimodal anomaly detection values, multimodal thermal runaway risk probability values, and thermal runaway self-healing feedback values, as well as anomaly thresholds, multi-level risk thresholds, and self-healing thresholds. New anomaly patterns are learned by integrating actual thermal runaway self-healing effects to optimize thermal runaway protection measures. Thermal runaway monitoring and feedback reports are regularly output, and the entire closed-loop response process of multimodal thermal runaway perception data and graded thermal runaway protection measures is displayed in a digital twin space.

10. A smart protection system for thermal runaway of charging piles based on edge computing, characterized in that, include: The multimodal sensing and anomaly detection module is used to collect thermal runaway multimodal sensing data in real time and perform data preprocessing on the thermal runaway multimodal sensing data; Based on the preprocessed thermal runaway multimodal sensing data, the abnormal state of the charging pile thermal runaway multimodal is identified in real time, and an abnormal feature data packet is generated. The edge risk assessment and early warning module is used to receive abnormal feature data packets to construct multimodal abnormal feature vectors, assess the probability of thermal runaway risk in real time based on the multimodal abnormal feature vectors, construct a digital twin space, and perform graded protection of thermal runaway risk of charging piles based on the probability of thermal runaway risk. The response execution feedback module is used to comprehensively analyze the multimodal thermal runaway perception data before and after thermal runaway risk classification protection, quantify the response effect of thermal runaway risk classification protection, and adjust the thermal runaway risk classification protection according to the response effect. The collaboration and operation and maintenance linkage module is used to visualize and monitor the entire process of thermal runaway risk through real-time linkage and data analysis across multiple platforms, and to optimize the parameters of various algorithms and the thermal runaway risk classification and protection strategy.

Images